| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * linux/mm/swapfile.c |
| 4 | * |
| 5 | * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds |
| 6 | * Swap reorganised 29.12.95, Stephen Tweedie |
| 7 | */ |
| 8 | |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/sched/mm.h> |
| 11 | #include <linux/sched/task.h> |
| 12 | #include <linux/hugetlb.h> |
| 13 | #include <linux/mman.h> |
| 14 | #include <linux/slab.h> |
| 15 | #include <linux/kernel_stat.h> |
| 16 | #include <linux/swap.h> |
| 17 | #include <linux/vmalloc.h> |
| 18 | #include <linux/pagemap.h> |
| 19 | #include <linux/namei.h> |
| 20 | #include <linux/shmem_fs.h> |
| 21 | #include <linux/blk-cgroup.h> |
| 22 | #include <linux/random.h> |
| 23 | #include <linux/writeback.h> |
| 24 | #include <linux/proc_fs.h> |
| 25 | #include <linux/seq_file.h> |
| 26 | #include <linux/init.h> |
| 27 | #include <linux/ksm.h> |
| 28 | #include <linux/rmap.h> |
| 29 | #include <linux/security.h> |
| 30 | #include <linux/backing-dev.h> |
| 31 | #include <linux/mutex.h> |
| 32 | #include <linux/capability.h> |
| 33 | #include <linux/syscalls.h> |
| 34 | #include <linux/memcontrol.h> |
| 35 | #include <linux/poll.h> |
| 36 | #include <linux/oom.h> |
| 37 | #include <linux/frontswap.h> |
| 38 | #include <linux/swapfile.h> |
| 39 | #include <linux/export.h> |
| 40 | #include <linux/swap_slots.h> |
| 41 | #include <linux/sort.h> |
| 42 | #include <linux/completion.h> |
| 43 | |
| 44 | #include <asm/tlbflush.h> |
| 45 | #include <linux/swapops.h> |
| 46 | #include <linux/swap_cgroup.h> |
| 47 | |
| 48 | static bool swap_count_continued(struct swap_info_struct *, pgoff_t, |
| 49 | unsigned char); |
| 50 | static void free_swap_count_continuations(struct swap_info_struct *); |
| 51 | |
| 52 | static DEFINE_SPINLOCK(swap_lock); |
| 53 | static unsigned int nr_swapfiles; |
| 54 | atomic_long_t nr_swap_pages; |
| 55 | /* |
| 56 | * Some modules use swappable objects and may try to swap them out under |
| 57 | * memory pressure (via the shrinker). Before doing so, they may wish to |
| 58 | * check to see if any swap space is available. |
| 59 | */ |
| 60 | EXPORT_SYMBOL_GPL(nr_swap_pages); |
| 61 | /* protected with swap_lock. reading in vm_swap_full() doesn't need lock */ |
| 62 | long total_swap_pages; |
| 63 | static int least_priority = -1; |
| 64 | |
| 65 | static const char Bad_file[] = "Bad swap file entry "; |
| 66 | static const char Unused_file[] = "Unused swap file entry "; |
| 67 | static const char Bad_offset[] = "Bad swap offset entry "; |
| 68 | static const char Unused_offset[] = "Unused swap offset entry "; |
| 69 | |
| 70 | /* |
| 71 | * all active swap_info_structs |
| 72 | * protected with swap_lock, and ordered by priority. |
| 73 | */ |
| 74 | static PLIST_HEAD(swap_active_head); |
| 75 | |
| 76 | /* |
| 77 | * all available (active, not full) swap_info_structs |
| 78 | * protected with swap_avail_lock, ordered by priority. |
| 79 | * This is used by get_swap_page() instead of swap_active_head |
| 80 | * because swap_active_head includes all swap_info_structs, |
| 81 | * but get_swap_page() doesn't need to look at full ones. |
| 82 | * This uses its own lock instead of swap_lock because when a |
| 83 | * swap_info_struct changes between not-full/full, it needs to |
| 84 | * add/remove itself to/from this list, but the swap_info_struct->lock |
| 85 | * is held and the locking order requires swap_lock to be taken |
| 86 | * before any swap_info_struct->lock. |
| 87 | */ |
| 88 | static struct plist_head *swap_avail_heads; |
| 89 | static DEFINE_SPINLOCK(swap_avail_lock); |
| 90 | |
| 91 | struct swap_info_struct *swap_info[MAX_SWAPFILES]; |
| 92 | |
| 93 | static DEFINE_MUTEX(swapon_mutex); |
| 94 | |
| 95 | static DECLARE_WAIT_QUEUE_HEAD(proc_poll_wait); |
| 96 | /* Activity counter to indicate that a swapon or swapoff has occurred */ |
| 97 | static atomic_t proc_poll_event = ATOMIC_INIT(0); |
| 98 | |
| 99 | atomic_t nr_rotate_swap = ATOMIC_INIT(0); |
| 100 | |
| 101 | static struct swap_info_struct *swap_type_to_swap_info(int type) |
| 102 | { |
| 103 | if (type >= MAX_SWAPFILES) |
| 104 | return NULL; |
| 105 | |
| 106 | return READ_ONCE(swap_info[type]); /* rcu_dereference() */ |
| 107 | } |
| 108 | |
| 109 | static inline unsigned char swap_count(unsigned char ent) |
| 110 | { |
| 111 | return ent & ~SWAP_HAS_CACHE; /* may include COUNT_CONTINUED flag */ |
| 112 | } |
| 113 | |
| 114 | /* Reclaim the swap entry anyway if possible */ |
| 115 | #define TTRS_ANYWAY 0x1 |
| 116 | /* |
| 117 | * Reclaim the swap entry if there are no more mappings of the |
| 118 | * corresponding page |
| 119 | */ |
| 120 | #define TTRS_UNMAPPED 0x2 |
| 121 | /* Reclaim the swap entry if swap is getting full*/ |
| 122 | #define TTRS_FULL 0x4 |
| 123 | |
| 124 | /* returns 1 if swap entry is freed */ |
| 125 | static int __try_to_reclaim_swap(struct swap_info_struct *si, |
| 126 | unsigned long offset, unsigned long flags) |
| 127 | { |
| 128 | swp_entry_t entry = swp_entry(si->type, offset); |
| 129 | struct page *page; |
| 130 | int ret = 0; |
| 131 | |
| 132 | page = find_get_page(swap_address_space(entry), offset); |
| 133 | if (!page) |
| 134 | return 0; |
| 135 | /* |
| 136 | * When this function is called from scan_swap_map_slots() and it's |
| 137 | * called by vmscan.c at reclaiming pages. So, we hold a lock on a page, |
| 138 | * here. We have to use trylock for avoiding deadlock. This is a special |
| 139 | * case and you should use try_to_free_swap() with explicit lock_page() |
| 140 | * in usual operations. |
| 141 | */ |
| 142 | if (trylock_page(page)) { |
| 143 | if ((flags & TTRS_ANYWAY) || |
| 144 | ((flags & TTRS_UNMAPPED) && !page_mapped(page)) || |
| 145 | ((flags & TTRS_FULL) && mem_cgroup_swap_full(page))) |
| 146 | ret = try_to_free_swap(page); |
| 147 | unlock_page(page); |
| 148 | } |
| 149 | put_page(page); |
| 150 | return ret; |
| 151 | } |
| 152 | |
| 153 | static inline struct swap_extent *first_se(struct swap_info_struct *sis) |
| 154 | { |
| 155 | struct rb_node *rb = rb_first(&sis->swap_extent_root); |
| 156 | return rb_entry(rb, struct swap_extent, rb_node); |
| 157 | } |
| 158 | |
| 159 | static inline struct swap_extent *next_se(struct swap_extent *se) |
| 160 | { |
| 161 | struct rb_node *rb = rb_next(&se->rb_node); |
| 162 | return rb ? rb_entry(rb, struct swap_extent, rb_node) : NULL; |
| 163 | } |
| 164 | |
| 165 | /* |
| 166 | * swapon tell device that all the old swap contents can be discarded, |
| 167 | * to allow the swap device to optimize its wear-levelling. |
| 168 | */ |
| 169 | static int discard_swap(struct swap_info_struct *si) |
| 170 | { |
| 171 | struct swap_extent *se; |
| 172 | sector_t start_block; |
| 173 | sector_t nr_blocks; |
| 174 | int err = 0; |
| 175 | |
| 176 | /* Do not discard the swap header page! */ |
| 177 | se = first_se(si); |
| 178 | start_block = (se->start_block + 1) << (PAGE_SHIFT - 9); |
| 179 | nr_blocks = ((sector_t)se->nr_pages - 1) << (PAGE_SHIFT - 9); |
| 180 | if (nr_blocks) { |
| 181 | err = blkdev_issue_discard(si->bdev, start_block, |
| 182 | nr_blocks, GFP_KERNEL, 0); |
| 183 | if (err) |
| 184 | return err; |
| 185 | cond_resched(); |
| 186 | } |
| 187 | |
| 188 | for (se = next_se(se); se; se = next_se(se)) { |
| 189 | start_block = se->start_block << (PAGE_SHIFT - 9); |
| 190 | nr_blocks = (sector_t)se->nr_pages << (PAGE_SHIFT - 9); |
| 191 | |
| 192 | err = blkdev_issue_discard(si->bdev, start_block, |
| 193 | nr_blocks, GFP_KERNEL, 0); |
| 194 | if (err) |
| 195 | break; |
| 196 | |
| 197 | cond_resched(); |
| 198 | } |
| 199 | return err; /* That will often be -EOPNOTSUPP */ |
| 200 | } |
| 201 | |
| 202 | static struct swap_extent * |
| 203 | offset_to_swap_extent(struct swap_info_struct *sis, unsigned long offset) |
| 204 | { |
| 205 | struct swap_extent *se; |
| 206 | struct rb_node *rb; |
| 207 | |
| 208 | rb = sis->swap_extent_root.rb_node; |
| 209 | while (rb) { |
| 210 | se = rb_entry(rb, struct swap_extent, rb_node); |
| 211 | if (offset < se->start_page) |
| 212 | rb = rb->rb_left; |
| 213 | else if (offset >= se->start_page + se->nr_pages) |
| 214 | rb = rb->rb_right; |
| 215 | else |
| 216 | return se; |
| 217 | } |
| 218 | /* It *must* be present */ |
| 219 | BUG(); |
| 220 | } |
| 221 | |
| 222 | sector_t swap_page_sector(struct page *page) |
| 223 | { |
| 224 | struct swap_info_struct *sis = page_swap_info(page); |
| 225 | struct swap_extent *se; |
| 226 | sector_t sector; |
| 227 | pgoff_t offset; |
| 228 | |
| 229 | offset = __page_file_index(page); |
| 230 | se = offset_to_swap_extent(sis, offset); |
| 231 | sector = se->start_block + (offset - se->start_page); |
| 232 | return sector << (PAGE_SHIFT - 9); |
| 233 | } |
| 234 | |
| 235 | /* |
| 236 | * swap allocation tell device that a cluster of swap can now be discarded, |
| 237 | * to allow the swap device to optimize its wear-levelling. |
| 238 | */ |
| 239 | static void discard_swap_cluster(struct swap_info_struct *si, |
| 240 | pgoff_t start_page, pgoff_t nr_pages) |
| 241 | { |
| 242 | struct swap_extent *se = offset_to_swap_extent(si, start_page); |
| 243 | |
| 244 | while (nr_pages) { |
| 245 | pgoff_t offset = start_page - se->start_page; |
| 246 | sector_t start_block = se->start_block + offset; |
| 247 | sector_t nr_blocks = se->nr_pages - offset; |
| 248 | |
| 249 | if (nr_blocks > nr_pages) |
| 250 | nr_blocks = nr_pages; |
| 251 | start_page += nr_blocks; |
| 252 | nr_pages -= nr_blocks; |
| 253 | |
| 254 | start_block <<= PAGE_SHIFT - 9; |
| 255 | nr_blocks <<= PAGE_SHIFT - 9; |
| 256 | if (blkdev_issue_discard(si->bdev, start_block, |
| 257 | nr_blocks, GFP_NOIO, 0)) |
| 258 | break; |
| 259 | |
| 260 | se = next_se(se); |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | #ifdef CONFIG_THP_SWAP |
| 265 | #define SWAPFILE_CLUSTER HPAGE_PMD_NR |
| 266 | |
| 267 | #define swap_entry_size(size) (size) |
| 268 | #else |
| 269 | #define SWAPFILE_CLUSTER 256 |
| 270 | |
| 271 | /* |
| 272 | * Define swap_entry_size() as constant to let compiler to optimize |
| 273 | * out some code if !CONFIG_THP_SWAP |
| 274 | */ |
| 275 | #define swap_entry_size(size) 1 |
| 276 | #endif |
| 277 | #define LATENCY_LIMIT 256 |
| 278 | |
| 279 | static inline void cluster_set_flag(struct swap_cluster_info *info, |
| 280 | unsigned int flag) |
| 281 | { |
| 282 | info->flags = flag; |
| 283 | } |
| 284 | |
| 285 | static inline unsigned int cluster_count(struct swap_cluster_info *info) |
| 286 | { |
| 287 | return info->data; |
| 288 | } |
| 289 | |
| 290 | static inline void cluster_set_count(struct swap_cluster_info *info, |
| 291 | unsigned int c) |
| 292 | { |
| 293 | info->data = c; |
| 294 | } |
| 295 | |
| 296 | static inline void cluster_set_count_flag(struct swap_cluster_info *info, |
| 297 | unsigned int c, unsigned int f) |
| 298 | { |
| 299 | info->flags = f; |
| 300 | info->data = c; |
| 301 | } |
| 302 | |
| 303 | static inline unsigned int cluster_next(struct swap_cluster_info *info) |
| 304 | { |
| 305 | return info->data; |
| 306 | } |
| 307 | |
| 308 | static inline void cluster_set_next(struct swap_cluster_info *info, |
| 309 | unsigned int n) |
| 310 | { |
| 311 | info->data = n; |
| 312 | } |
| 313 | |
| 314 | static inline void cluster_set_next_flag(struct swap_cluster_info *info, |
| 315 | unsigned int n, unsigned int f) |
| 316 | { |
| 317 | info->flags = f; |
| 318 | info->data = n; |
| 319 | } |
| 320 | |
| 321 | static inline bool cluster_is_free(struct swap_cluster_info *info) |
| 322 | { |
| 323 | return info->flags & CLUSTER_FLAG_FREE; |
| 324 | } |
| 325 | |
| 326 | static inline bool cluster_is_null(struct swap_cluster_info *info) |
| 327 | { |
| 328 | return info->flags & CLUSTER_FLAG_NEXT_NULL; |
| 329 | } |
| 330 | |
| 331 | static inline void cluster_set_null(struct swap_cluster_info *info) |
| 332 | { |
| 333 | info->flags = CLUSTER_FLAG_NEXT_NULL; |
| 334 | info->data = 0; |
| 335 | } |
| 336 | |
| 337 | static inline bool cluster_is_huge(struct swap_cluster_info *info) |
| 338 | { |
| 339 | if (IS_ENABLED(CONFIG_THP_SWAP)) |
| 340 | return info->flags & CLUSTER_FLAG_HUGE; |
| 341 | return false; |
| 342 | } |
| 343 | |
| 344 | static inline void cluster_clear_huge(struct swap_cluster_info *info) |
| 345 | { |
| 346 | info->flags &= ~CLUSTER_FLAG_HUGE; |
| 347 | } |
| 348 | |
| 349 | static inline struct swap_cluster_info *lock_cluster(struct swap_info_struct *si, |
| 350 | unsigned long offset) |
| 351 | { |
| 352 | struct swap_cluster_info *ci; |
| 353 | |
| 354 | ci = si->cluster_info; |
| 355 | if (ci) { |
| 356 | ci += offset / SWAPFILE_CLUSTER; |
| 357 | spin_lock(&ci->lock); |
| 358 | } |
| 359 | return ci; |
| 360 | } |
| 361 | |
| 362 | static inline void unlock_cluster(struct swap_cluster_info *ci) |
| 363 | { |
| 364 | if (ci) |
| 365 | spin_unlock(&ci->lock); |
| 366 | } |
| 367 | |
| 368 | /* |
| 369 | * Determine the locking method in use for this device. Return |
| 370 | * swap_cluster_info if SSD-style cluster-based locking is in place. |
| 371 | */ |
| 372 | static inline struct swap_cluster_info *lock_cluster_or_swap_info( |
| 373 | struct swap_info_struct *si, unsigned long offset) |
| 374 | { |
| 375 | struct swap_cluster_info *ci; |
| 376 | |
| 377 | /* Try to use fine-grained SSD-style locking if available: */ |
| 378 | ci = lock_cluster(si, offset); |
| 379 | /* Otherwise, fall back to traditional, coarse locking: */ |
| 380 | if (!ci) |
| 381 | spin_lock(&si->lock); |
| 382 | |
| 383 | return ci; |
| 384 | } |
| 385 | |
| 386 | static inline void unlock_cluster_or_swap_info(struct swap_info_struct *si, |
| 387 | struct swap_cluster_info *ci) |
| 388 | { |
| 389 | if (ci) |
| 390 | unlock_cluster(ci); |
| 391 | else |
| 392 | spin_unlock(&si->lock); |
| 393 | } |
| 394 | |
| 395 | static inline bool cluster_list_empty(struct swap_cluster_list *list) |
| 396 | { |
| 397 | return cluster_is_null(&list->head); |
| 398 | } |
| 399 | |
| 400 | static inline unsigned int cluster_list_first(struct swap_cluster_list *list) |
| 401 | { |
| 402 | return cluster_next(&list->head); |
| 403 | } |
| 404 | |
| 405 | static void cluster_list_init(struct swap_cluster_list *list) |
| 406 | { |
| 407 | cluster_set_null(&list->head); |
| 408 | cluster_set_null(&list->tail); |
| 409 | } |
| 410 | |
| 411 | static void cluster_list_add_tail(struct swap_cluster_list *list, |
| 412 | struct swap_cluster_info *ci, |
| 413 | unsigned int idx) |
| 414 | { |
| 415 | if (cluster_list_empty(list)) { |
| 416 | cluster_set_next_flag(&list->head, idx, 0); |
| 417 | cluster_set_next_flag(&list->tail, idx, 0); |
| 418 | } else { |
| 419 | struct swap_cluster_info *ci_tail; |
| 420 | unsigned int tail = cluster_next(&list->tail); |
| 421 | |
| 422 | /* |
| 423 | * Nested cluster lock, but both cluster locks are |
| 424 | * only acquired when we held swap_info_struct->lock |
| 425 | */ |
| 426 | ci_tail = ci + tail; |
| 427 | spin_lock_nested(&ci_tail->lock, SINGLE_DEPTH_NESTING); |
| 428 | cluster_set_next(ci_tail, idx); |
| 429 | spin_unlock(&ci_tail->lock); |
| 430 | cluster_set_next_flag(&list->tail, idx, 0); |
| 431 | } |
| 432 | } |
| 433 | |
| 434 | static unsigned int cluster_list_del_first(struct swap_cluster_list *list, |
| 435 | struct swap_cluster_info *ci) |
| 436 | { |
| 437 | unsigned int idx; |
| 438 | |
| 439 | idx = cluster_next(&list->head); |
| 440 | if (cluster_next(&list->tail) == idx) { |
| 441 | cluster_set_null(&list->head); |
| 442 | cluster_set_null(&list->tail); |
| 443 | } else |
| 444 | cluster_set_next_flag(&list->head, |
| 445 | cluster_next(&ci[idx]), 0); |
| 446 | |
| 447 | return idx; |
| 448 | } |
| 449 | |
| 450 | /* Add a cluster to discard list and schedule it to do discard */ |
| 451 | static void swap_cluster_schedule_discard(struct swap_info_struct *si, |
| 452 | unsigned int idx) |
| 453 | { |
| 454 | /* |
| 455 | * If scan_swap_map_slots() can't find a free cluster, it will check |
| 456 | * si->swap_map directly. To make sure the discarding cluster isn't |
| 457 | * taken by scan_swap_map_slots(), mark the swap entries bad (occupied). |
| 458 | * It will be cleared after discard |
| 459 | */ |
| 460 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
| 461 | SWAP_MAP_BAD, SWAPFILE_CLUSTER); |
| 462 | |
| 463 | cluster_list_add_tail(&si->discard_clusters, si->cluster_info, idx); |
| 464 | |
| 465 | schedule_work(&si->discard_work); |
| 466 | } |
| 467 | |
| 468 | static void __free_cluster(struct swap_info_struct *si, unsigned long idx) |
| 469 | { |
| 470 | struct swap_cluster_info *ci = si->cluster_info; |
| 471 | |
| 472 | cluster_set_flag(ci + idx, CLUSTER_FLAG_FREE); |
| 473 | cluster_list_add_tail(&si->free_clusters, ci, idx); |
| 474 | } |
| 475 | |
| 476 | /* |
| 477 | * Doing discard actually. After a cluster discard is finished, the cluster |
| 478 | * will be added to free cluster list. caller should hold si->lock. |
| 479 | */ |
| 480 | static void swap_do_scheduled_discard(struct swap_info_struct *si) |
| 481 | { |
| 482 | struct swap_cluster_info *info, *ci; |
| 483 | unsigned int idx; |
| 484 | |
| 485 | info = si->cluster_info; |
| 486 | |
| 487 | while (!cluster_list_empty(&si->discard_clusters)) { |
| 488 | idx = cluster_list_del_first(&si->discard_clusters, info); |
| 489 | spin_unlock(&si->lock); |
| 490 | |
| 491 | discard_swap_cluster(si, idx * SWAPFILE_CLUSTER, |
| 492 | SWAPFILE_CLUSTER); |
| 493 | |
| 494 | spin_lock(&si->lock); |
| 495 | ci = lock_cluster(si, idx * SWAPFILE_CLUSTER); |
| 496 | __free_cluster(si, idx); |
| 497 | memset(si->swap_map + idx * SWAPFILE_CLUSTER, |
| 498 | 0, SWAPFILE_CLUSTER); |
| 499 | unlock_cluster(ci); |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | static void swap_discard_work(struct work_struct *work) |
| 504 | { |
| 505 | struct swap_info_struct *si; |
| 506 | |
| 507 | si = container_of(work, struct swap_info_struct, discard_work); |
| 508 | |
| 509 | spin_lock(&si->lock); |
| 510 | swap_do_scheduled_discard(si); |
| 511 | spin_unlock(&si->lock); |
| 512 | } |
| 513 | |
| 514 | static void swap_users_ref_free(struct percpu_ref *ref) |
| 515 | { |
| 516 | struct swap_info_struct *si; |
| 517 | |
| 518 | si = container_of(ref, struct swap_info_struct, users); |
| 519 | complete(&si->comp); |
| 520 | } |
| 521 | |
| 522 | static void alloc_cluster(struct swap_info_struct *si, unsigned long idx) |
| 523 | { |
| 524 | struct swap_cluster_info *ci = si->cluster_info; |
| 525 | |
| 526 | VM_BUG_ON(cluster_list_first(&si->free_clusters) != idx); |
| 527 | cluster_list_del_first(&si->free_clusters, ci); |
| 528 | cluster_set_count_flag(ci + idx, 0, 0); |
| 529 | } |
| 530 | |
| 531 | static void free_cluster(struct swap_info_struct *si, unsigned long idx) |
| 532 | { |
| 533 | struct swap_cluster_info *ci = si->cluster_info + idx; |
| 534 | |
| 535 | VM_BUG_ON(cluster_count(ci) != 0); |
| 536 | /* |
| 537 | * If the swap is discardable, prepare discard the cluster |
| 538 | * instead of free it immediately. The cluster will be freed |
| 539 | * after discard. |
| 540 | */ |
| 541 | if ((si->flags & (SWP_WRITEOK | SWP_PAGE_DISCARD)) == |
| 542 | (SWP_WRITEOK | SWP_PAGE_DISCARD)) { |
| 543 | swap_cluster_schedule_discard(si, idx); |
| 544 | return; |
| 545 | } |
| 546 | |
| 547 | __free_cluster(si, idx); |
| 548 | } |
| 549 | |
| 550 | /* |
| 551 | * The cluster corresponding to page_nr will be used. The cluster will be |
| 552 | * removed from free cluster list and its usage counter will be increased. |
| 553 | */ |
| 554 | static void inc_cluster_info_page(struct swap_info_struct *p, |
| 555 | struct swap_cluster_info *cluster_info, unsigned long page_nr) |
| 556 | { |
| 557 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; |
| 558 | |
| 559 | if (!cluster_info) |
| 560 | return; |
| 561 | if (cluster_is_free(&cluster_info[idx])) |
| 562 | alloc_cluster(p, idx); |
| 563 | |
| 564 | VM_BUG_ON(cluster_count(&cluster_info[idx]) >= SWAPFILE_CLUSTER); |
| 565 | cluster_set_count(&cluster_info[idx], |
| 566 | cluster_count(&cluster_info[idx]) + 1); |
| 567 | } |
| 568 | |
| 569 | /* |
| 570 | * The cluster corresponding to page_nr decreases one usage. If the usage |
| 571 | * counter becomes 0, which means no page in the cluster is in using, we can |
| 572 | * optionally discard the cluster and add it to free cluster list. |
| 573 | */ |
| 574 | static void dec_cluster_info_page(struct swap_info_struct *p, |
| 575 | struct swap_cluster_info *cluster_info, unsigned long page_nr) |
| 576 | { |
| 577 | unsigned long idx = page_nr / SWAPFILE_CLUSTER; |
| 578 | |
| 579 | if (!cluster_info) |
| 580 | return; |
| 581 | |
| 582 | VM_BUG_ON(cluster_count(&cluster_info[idx]) == 0); |
| 583 | cluster_set_count(&cluster_info[idx], |
| 584 | cluster_count(&cluster_info[idx]) - 1); |
| 585 | |
| 586 | if (cluster_count(&cluster_info[idx]) == 0) |
| 587 | free_cluster(p, idx); |
| 588 | } |
| 589 | |
| 590 | /* |
| 591 | * It's possible scan_swap_map_slots() uses a free cluster in the middle of free |
| 592 | * cluster list. Avoiding such abuse to avoid list corruption. |
| 593 | */ |
| 594 | static bool |
| 595 | scan_swap_map_ssd_cluster_conflict(struct swap_info_struct *si, |
| 596 | unsigned long offset) |
| 597 | { |
| 598 | struct percpu_cluster *percpu_cluster; |
| 599 | bool conflict; |
| 600 | |
| 601 | offset /= SWAPFILE_CLUSTER; |
| 602 | conflict = !cluster_list_empty(&si->free_clusters) && |
| 603 | offset != cluster_list_first(&si->free_clusters) && |
| 604 | cluster_is_free(&si->cluster_info[offset]); |
| 605 | |
| 606 | if (!conflict) |
| 607 | return false; |
| 608 | |
| 609 | percpu_cluster = this_cpu_ptr(si->percpu_cluster); |
| 610 | cluster_set_null(&percpu_cluster->index); |
| 611 | return true; |
| 612 | } |
| 613 | |
| 614 | /* |
| 615 | * Try to get a swap entry from current cpu's swap entry pool (a cluster). This |
| 616 | * might involve allocating a new cluster for current CPU too. |
| 617 | */ |
| 618 | static bool scan_swap_map_try_ssd_cluster(struct swap_info_struct *si, |
| 619 | unsigned long *offset, unsigned long *scan_base) |
| 620 | { |
| 621 | struct percpu_cluster *cluster; |
| 622 | struct swap_cluster_info *ci; |
| 623 | unsigned long tmp, max; |
| 624 | |
| 625 | new_cluster: |
| 626 | cluster = this_cpu_ptr(si->percpu_cluster); |
| 627 | if (cluster_is_null(&cluster->index)) { |
| 628 | if (!cluster_list_empty(&si->free_clusters)) { |
| 629 | cluster->index = si->free_clusters.head; |
| 630 | cluster->next = cluster_next(&cluster->index) * |
| 631 | SWAPFILE_CLUSTER; |
| 632 | } else if (!cluster_list_empty(&si->discard_clusters)) { |
| 633 | /* |
| 634 | * we don't have free cluster but have some clusters in |
| 635 | * discarding, do discard now and reclaim them, then |
| 636 | * reread cluster_next_cpu since we dropped si->lock |
| 637 | */ |
| 638 | swap_do_scheduled_discard(si); |
| 639 | *scan_base = this_cpu_read(*si->cluster_next_cpu); |
| 640 | *offset = *scan_base; |
| 641 | goto new_cluster; |
| 642 | } else |
| 643 | return false; |
| 644 | } |
| 645 | |
| 646 | /* |
| 647 | * Other CPUs can use our cluster if they can't find a free cluster, |
| 648 | * check if there is still free entry in the cluster |
| 649 | */ |
| 650 | tmp = cluster->next; |
| 651 | max = min_t(unsigned long, si->max, |
| 652 | (cluster_next(&cluster->index) + 1) * SWAPFILE_CLUSTER); |
| 653 | if (tmp < max) { |
| 654 | ci = lock_cluster(si, tmp); |
| 655 | while (tmp < max) { |
| 656 | if (!si->swap_map[tmp]) |
| 657 | break; |
| 658 | tmp++; |
| 659 | } |
| 660 | unlock_cluster(ci); |
| 661 | } |
| 662 | if (tmp >= max) { |
| 663 | cluster_set_null(&cluster->index); |
| 664 | goto new_cluster; |
| 665 | } |
| 666 | cluster->next = tmp + 1; |
| 667 | *offset = tmp; |
| 668 | *scan_base = tmp; |
| 669 | return true; |
| 670 | } |
| 671 | |
| 672 | static void __del_from_avail_list(struct swap_info_struct *p) |
| 673 | { |
| 674 | int nid; |
| 675 | |
| 676 | for_each_node(nid) |
| 677 | plist_del(&p->avail_lists[nid], &swap_avail_heads[nid]); |
| 678 | } |
| 679 | |
| 680 | static void del_from_avail_list(struct swap_info_struct *p) |
| 681 | { |
| 682 | spin_lock(&swap_avail_lock); |
| 683 | __del_from_avail_list(p); |
| 684 | spin_unlock(&swap_avail_lock); |
| 685 | } |
| 686 | |
| 687 | static void swap_range_alloc(struct swap_info_struct *si, unsigned long offset, |
| 688 | unsigned int nr_entries) |
| 689 | { |
| 690 | unsigned int end = offset + nr_entries - 1; |
| 691 | |
| 692 | if (offset == si->lowest_bit) |
| 693 | si->lowest_bit += nr_entries; |
| 694 | if (end == si->highest_bit) |
| 695 | WRITE_ONCE(si->highest_bit, si->highest_bit - nr_entries); |
| 696 | si->inuse_pages += nr_entries; |
| 697 | if (si->inuse_pages == si->pages) { |
| 698 | si->lowest_bit = si->max; |
| 699 | si->highest_bit = 0; |
| 700 | del_from_avail_list(si); |
| 701 | } |
| 702 | } |
| 703 | |
| 704 | static void add_to_avail_list(struct swap_info_struct *p) |
| 705 | { |
| 706 | int nid; |
| 707 | |
| 708 | spin_lock(&swap_avail_lock); |
| 709 | for_each_node(nid) { |
| 710 | WARN_ON(!plist_node_empty(&p->avail_lists[nid])); |
| 711 | plist_add(&p->avail_lists[nid], &swap_avail_heads[nid]); |
| 712 | } |
| 713 | spin_unlock(&swap_avail_lock); |
| 714 | } |
| 715 | |
| 716 | static void swap_range_free(struct swap_info_struct *si, unsigned long offset, |
| 717 | unsigned int nr_entries) |
| 718 | { |
| 719 | unsigned long begin = offset; |
| 720 | unsigned long end = offset + nr_entries - 1; |
| 721 | void (*swap_slot_free_notify)(struct block_device *, unsigned long); |
| 722 | |
| 723 | if (offset < si->lowest_bit) |
| 724 | si->lowest_bit = offset; |
| 725 | if (end > si->highest_bit) { |
| 726 | bool was_full = !si->highest_bit; |
| 727 | |
| 728 | WRITE_ONCE(si->highest_bit, end); |
| 729 | if (was_full && (si->flags & SWP_WRITEOK)) |
| 730 | add_to_avail_list(si); |
| 731 | } |
| 732 | atomic_long_add(nr_entries, &nr_swap_pages); |
| 733 | si->inuse_pages -= nr_entries; |
| 734 | if (si->flags & SWP_BLKDEV) |
| 735 | swap_slot_free_notify = |
| 736 | si->bdev->bd_disk->fops->swap_slot_free_notify; |
| 737 | else |
| 738 | swap_slot_free_notify = NULL; |
| 739 | while (offset <= end) { |
| 740 | arch_swap_invalidate_page(si->type, offset); |
| 741 | frontswap_invalidate_page(si->type, offset); |
| 742 | if (swap_slot_free_notify) |
| 743 | swap_slot_free_notify(si->bdev, offset); |
| 744 | offset++; |
| 745 | } |
| 746 | clear_shadow_from_swap_cache(si->type, begin, end); |
| 747 | } |
| 748 | |
| 749 | static void set_cluster_next(struct swap_info_struct *si, unsigned long next) |
| 750 | { |
| 751 | unsigned long prev; |
| 752 | |
| 753 | if (!(si->flags & SWP_SOLIDSTATE)) { |
| 754 | si->cluster_next = next; |
| 755 | return; |
| 756 | } |
| 757 | |
| 758 | prev = this_cpu_read(*si->cluster_next_cpu); |
| 759 | /* |
| 760 | * Cross the swap address space size aligned trunk, choose |
| 761 | * another trunk randomly to avoid lock contention on swap |
| 762 | * address space if possible. |
| 763 | */ |
| 764 | if ((prev >> SWAP_ADDRESS_SPACE_SHIFT) != |
| 765 | (next >> SWAP_ADDRESS_SPACE_SHIFT)) { |
| 766 | /* No free swap slots available */ |
| 767 | if (si->highest_bit <= si->lowest_bit) |
| 768 | return; |
| 769 | next = si->lowest_bit + |
| 770 | prandom_u32_max(si->highest_bit - si->lowest_bit + 1); |
| 771 | next = ALIGN_DOWN(next, SWAP_ADDRESS_SPACE_PAGES); |
| 772 | next = max_t(unsigned int, next, si->lowest_bit); |
| 773 | } |
| 774 | this_cpu_write(*si->cluster_next_cpu, next); |
| 775 | } |
| 776 | |
| 777 | static int scan_swap_map_slots(struct swap_info_struct *si, |
| 778 | unsigned char usage, int nr, |
| 779 | swp_entry_t slots[]) |
| 780 | { |
| 781 | struct swap_cluster_info *ci; |
| 782 | unsigned long offset; |
| 783 | unsigned long scan_base; |
| 784 | unsigned long last_in_cluster = 0; |
| 785 | int latency_ration = LATENCY_LIMIT; |
| 786 | int n_ret = 0; |
| 787 | bool scanned_many = false; |
| 788 | |
| 789 | /* |
| 790 | * We try to cluster swap pages by allocating them sequentially |
| 791 | * in swap. Once we've allocated SWAPFILE_CLUSTER pages this |
| 792 | * way, however, we resort to first-free allocation, starting |
| 793 | * a new cluster. This prevents us from scattering swap pages |
| 794 | * all over the entire swap partition, so that we reduce |
| 795 | * overall disk seek times between swap pages. -- sct |
| 796 | * But we do now try to find an empty cluster. -Andrea |
| 797 | * And we let swap pages go all over an SSD partition. Hugh |
| 798 | */ |
| 799 | |
| 800 | si->flags += SWP_SCANNING; |
| 801 | /* |
| 802 | * Use percpu scan base for SSD to reduce lock contention on |
| 803 | * cluster and swap cache. For HDD, sequential access is more |
| 804 | * important. |
| 805 | */ |
| 806 | if (si->flags & SWP_SOLIDSTATE) |
| 807 | scan_base = this_cpu_read(*si->cluster_next_cpu); |
| 808 | else |
| 809 | scan_base = si->cluster_next; |
| 810 | offset = scan_base; |
| 811 | |
| 812 | /* SSD algorithm */ |
| 813 | if (si->cluster_info) { |
| 814 | if (!scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
| 815 | goto scan; |
| 816 | } else if (unlikely(!si->cluster_nr--)) { |
| 817 | if (si->pages - si->inuse_pages < SWAPFILE_CLUSTER) { |
| 818 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
| 819 | goto checks; |
| 820 | } |
| 821 | |
| 822 | spin_unlock(&si->lock); |
| 823 | |
| 824 | /* |
| 825 | * If seek is expensive, start searching for new cluster from |
| 826 | * start of partition, to minimize the span of allocated swap. |
| 827 | * If seek is cheap, that is the SWP_SOLIDSTATE si->cluster_info |
| 828 | * case, just handled by scan_swap_map_try_ssd_cluster() above. |
| 829 | */ |
| 830 | scan_base = offset = si->lowest_bit; |
| 831 | last_in_cluster = offset + SWAPFILE_CLUSTER - 1; |
| 832 | |
| 833 | /* Locate the first empty (unaligned) cluster */ |
| 834 | for (; last_in_cluster <= si->highest_bit; offset++) { |
| 835 | if (si->swap_map[offset]) |
| 836 | last_in_cluster = offset + SWAPFILE_CLUSTER; |
| 837 | else if (offset == last_in_cluster) { |
| 838 | spin_lock(&si->lock); |
| 839 | offset -= SWAPFILE_CLUSTER - 1; |
| 840 | si->cluster_next = offset; |
| 841 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
| 842 | goto checks; |
| 843 | } |
| 844 | if (unlikely(--latency_ration < 0)) { |
| 845 | cond_resched(); |
| 846 | latency_ration = LATENCY_LIMIT; |
| 847 | } |
| 848 | } |
| 849 | |
| 850 | offset = scan_base; |
| 851 | spin_lock(&si->lock); |
| 852 | si->cluster_nr = SWAPFILE_CLUSTER - 1; |
| 853 | } |
| 854 | |
| 855 | checks: |
| 856 | if (si->cluster_info) { |
| 857 | while (scan_swap_map_ssd_cluster_conflict(si, offset)) { |
| 858 | /* take a break if we already got some slots */ |
| 859 | if (n_ret) |
| 860 | goto done; |
| 861 | if (!scan_swap_map_try_ssd_cluster(si, &offset, |
| 862 | &scan_base)) |
| 863 | goto scan; |
| 864 | } |
| 865 | } |
| 866 | if (!(si->flags & SWP_WRITEOK)) |
| 867 | goto no_page; |
| 868 | if (!si->highest_bit) |
| 869 | goto no_page; |
| 870 | if (offset > si->highest_bit) |
| 871 | scan_base = offset = si->lowest_bit; |
| 872 | |
| 873 | ci = lock_cluster(si, offset); |
| 874 | /* reuse swap entry of cache-only swap if not busy. */ |
| 875 | if (vm_swap_full() && si->swap_map[offset] == SWAP_HAS_CACHE) { |
| 876 | int swap_was_freed; |
| 877 | unlock_cluster(ci); |
| 878 | spin_unlock(&si->lock); |
| 879 | swap_was_freed = __try_to_reclaim_swap(si, offset, TTRS_ANYWAY); |
| 880 | spin_lock(&si->lock); |
| 881 | /* entry was freed successfully, try to use this again */ |
| 882 | if (swap_was_freed) |
| 883 | goto checks; |
| 884 | goto scan; /* check next one */ |
| 885 | } |
| 886 | |
| 887 | if (si->swap_map[offset]) { |
| 888 | unlock_cluster(ci); |
| 889 | if (!n_ret) |
| 890 | goto scan; |
| 891 | else |
| 892 | goto done; |
| 893 | } |
| 894 | WRITE_ONCE(si->swap_map[offset], usage); |
| 895 | inc_cluster_info_page(si, si->cluster_info, offset); |
| 896 | unlock_cluster(ci); |
| 897 | |
| 898 | swap_range_alloc(si, offset, 1); |
| 899 | slots[n_ret++] = swp_entry(si->type, offset); |
| 900 | |
| 901 | /* got enough slots or reach max slots? */ |
| 902 | if ((n_ret == nr) || (offset >= si->highest_bit)) |
| 903 | goto done; |
| 904 | |
| 905 | /* search for next available slot */ |
| 906 | |
| 907 | /* time to take a break? */ |
| 908 | if (unlikely(--latency_ration < 0)) { |
| 909 | if (n_ret) |
| 910 | goto done; |
| 911 | spin_unlock(&si->lock); |
| 912 | cond_resched(); |
| 913 | spin_lock(&si->lock); |
| 914 | latency_ration = LATENCY_LIMIT; |
| 915 | } |
| 916 | |
| 917 | /* try to get more slots in cluster */ |
| 918 | if (si->cluster_info) { |
| 919 | if (scan_swap_map_try_ssd_cluster(si, &offset, &scan_base)) |
| 920 | goto checks; |
| 921 | } else if (si->cluster_nr && !si->swap_map[++offset]) { |
| 922 | /* non-ssd case, still more slots in cluster? */ |
| 923 | --si->cluster_nr; |
| 924 | goto checks; |
| 925 | } |
| 926 | |
| 927 | /* |
| 928 | * Even if there's no free clusters available (fragmented), |
| 929 | * try to scan a little more quickly with lock held unless we |
| 930 | * have scanned too many slots already. |
| 931 | */ |
| 932 | if (!scanned_many) { |
| 933 | unsigned long scan_limit; |
| 934 | |
| 935 | if (offset < scan_base) |
| 936 | scan_limit = scan_base; |
| 937 | else |
| 938 | scan_limit = si->highest_bit; |
| 939 | for (; offset <= scan_limit && --latency_ration > 0; |
| 940 | offset++) { |
| 941 | if (!si->swap_map[offset]) |
| 942 | goto checks; |
| 943 | } |
| 944 | } |
| 945 | |
| 946 | done: |
| 947 | set_cluster_next(si, offset + 1); |
| 948 | si->flags -= SWP_SCANNING; |
| 949 | return n_ret; |
| 950 | |
| 951 | scan: |
| 952 | spin_unlock(&si->lock); |
| 953 | while (++offset <= READ_ONCE(si->highest_bit)) { |
| 954 | if (data_race(!si->swap_map[offset])) { |
| 955 | spin_lock(&si->lock); |
| 956 | goto checks; |
| 957 | } |
| 958 | if (vm_swap_full() && |
| 959 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { |
| 960 | spin_lock(&si->lock); |
| 961 | goto checks; |
| 962 | } |
| 963 | if (unlikely(--latency_ration < 0)) { |
| 964 | cond_resched(); |
| 965 | latency_ration = LATENCY_LIMIT; |
| 966 | scanned_many = true; |
| 967 | } |
| 968 | } |
| 969 | offset = si->lowest_bit; |
| 970 | while (offset < scan_base) { |
| 971 | if (data_race(!si->swap_map[offset])) { |
| 972 | spin_lock(&si->lock); |
| 973 | goto checks; |
| 974 | } |
| 975 | if (vm_swap_full() && |
| 976 | READ_ONCE(si->swap_map[offset]) == SWAP_HAS_CACHE) { |
| 977 | spin_lock(&si->lock); |
| 978 | goto checks; |
| 979 | } |
| 980 | if (unlikely(--latency_ration < 0)) { |
| 981 | cond_resched(); |
| 982 | latency_ration = LATENCY_LIMIT; |
| 983 | scanned_many = true; |
| 984 | } |
| 985 | offset++; |
| 986 | } |
| 987 | spin_lock(&si->lock); |
| 988 | |
| 989 | no_page: |
| 990 | si->flags -= SWP_SCANNING; |
| 991 | return n_ret; |
| 992 | } |
| 993 | |
| 994 | static int swap_alloc_cluster(struct swap_info_struct *si, swp_entry_t *slot) |
| 995 | { |
| 996 | unsigned long idx; |
| 997 | struct swap_cluster_info *ci; |
| 998 | unsigned long offset; |
| 999 | |
| 1000 | /* |
| 1001 | * Should not even be attempting cluster allocations when huge |
| 1002 | * page swap is disabled. Warn and fail the allocation. |
| 1003 | */ |
| 1004 | if (!IS_ENABLED(CONFIG_THP_SWAP)) { |
| 1005 | VM_WARN_ON_ONCE(1); |
| 1006 | return 0; |
| 1007 | } |
| 1008 | |
| 1009 | if (cluster_list_empty(&si->free_clusters)) |
| 1010 | return 0; |
| 1011 | |
| 1012 | idx = cluster_list_first(&si->free_clusters); |
| 1013 | offset = idx * SWAPFILE_CLUSTER; |
| 1014 | ci = lock_cluster(si, offset); |
| 1015 | alloc_cluster(si, idx); |
| 1016 | cluster_set_count_flag(ci, SWAPFILE_CLUSTER, CLUSTER_FLAG_HUGE); |
| 1017 | |
| 1018 | memset(si->swap_map + offset, SWAP_HAS_CACHE, SWAPFILE_CLUSTER); |
| 1019 | unlock_cluster(ci); |
| 1020 | swap_range_alloc(si, offset, SWAPFILE_CLUSTER); |
| 1021 | *slot = swp_entry(si->type, offset); |
| 1022 | |
| 1023 | return 1; |
| 1024 | } |
| 1025 | |
| 1026 | static void swap_free_cluster(struct swap_info_struct *si, unsigned long idx) |
| 1027 | { |
| 1028 | unsigned long offset = idx * SWAPFILE_CLUSTER; |
| 1029 | struct swap_cluster_info *ci; |
| 1030 | |
| 1031 | ci = lock_cluster(si, offset); |
| 1032 | memset(si->swap_map + offset, 0, SWAPFILE_CLUSTER); |
| 1033 | cluster_set_count_flag(ci, 0, 0); |
| 1034 | free_cluster(si, idx); |
| 1035 | unlock_cluster(ci); |
| 1036 | swap_range_free(si, offset, SWAPFILE_CLUSTER); |
| 1037 | } |
| 1038 | |
| 1039 | int get_swap_pages(int n_goal, swp_entry_t swp_entries[], int entry_size) |
| 1040 | { |
| 1041 | unsigned long size = swap_entry_size(entry_size); |
| 1042 | struct swap_info_struct *si, *next; |
| 1043 | long avail_pgs; |
| 1044 | int n_ret = 0; |
| 1045 | int node; |
| 1046 | |
| 1047 | /* Only single cluster request supported */ |
| 1048 | WARN_ON_ONCE(n_goal > 1 && size == SWAPFILE_CLUSTER); |
| 1049 | |
| 1050 | spin_lock(&swap_avail_lock); |
| 1051 | |
| 1052 | avail_pgs = atomic_long_read(&nr_swap_pages) / size; |
| 1053 | if (avail_pgs <= 0) { |
| 1054 | spin_unlock(&swap_avail_lock); |
| 1055 | goto noswap; |
| 1056 | } |
| 1057 | |
| 1058 | n_goal = min3((long)n_goal, (long)SWAP_BATCH, avail_pgs); |
| 1059 | |
| 1060 | atomic_long_sub(n_goal * size, &nr_swap_pages); |
| 1061 | |
| 1062 | start_over: |
| 1063 | node = numa_node_id(); |
| 1064 | plist_for_each_entry_safe(si, next, &swap_avail_heads[node], avail_lists[node]) { |
| 1065 | /* requeue si to after same-priority siblings */ |
| 1066 | plist_requeue(&si->avail_lists[node], &swap_avail_heads[node]); |
| 1067 | spin_unlock(&swap_avail_lock); |
| 1068 | spin_lock(&si->lock); |
| 1069 | if (!si->highest_bit || !(si->flags & SWP_WRITEOK)) { |
| 1070 | spin_lock(&swap_avail_lock); |
| 1071 | if (plist_node_empty(&si->avail_lists[node])) { |
| 1072 | spin_unlock(&si->lock); |
| 1073 | goto nextsi; |
| 1074 | } |
| 1075 | WARN(!si->highest_bit, |
| 1076 | "swap_info %d in list but !highest_bit\n", |
| 1077 | si->type); |
| 1078 | WARN(!(si->flags & SWP_WRITEOK), |
| 1079 | "swap_info %d in list but !SWP_WRITEOK\n", |
| 1080 | si->type); |
| 1081 | __del_from_avail_list(si); |
| 1082 | spin_unlock(&si->lock); |
| 1083 | goto nextsi; |
| 1084 | } |
| 1085 | if (size == SWAPFILE_CLUSTER) { |
| 1086 | if (si->flags & SWP_BLKDEV) |
| 1087 | n_ret = swap_alloc_cluster(si, swp_entries); |
| 1088 | } else |
| 1089 | n_ret = scan_swap_map_slots(si, SWAP_HAS_CACHE, |
| 1090 | n_goal, swp_entries); |
| 1091 | spin_unlock(&si->lock); |
| 1092 | if (n_ret || size == SWAPFILE_CLUSTER) |
| 1093 | goto check_out; |
| 1094 | pr_debug("scan_swap_map of si %d failed to find offset\n", |
| 1095 | si->type); |
| 1096 | |
| 1097 | spin_lock(&swap_avail_lock); |
| 1098 | nextsi: |
| 1099 | /* |
| 1100 | * if we got here, it's likely that si was almost full before, |
| 1101 | * and since scan_swap_map_slots() can drop the si->lock, |
| 1102 | * multiple callers probably all tried to get a page from the |
| 1103 | * same si and it filled up before we could get one; or, the si |
| 1104 | * filled up between us dropping swap_avail_lock and taking |
| 1105 | * si->lock. Since we dropped the swap_avail_lock, the |
| 1106 | * swap_avail_head list may have been modified; so if next is |
| 1107 | * still in the swap_avail_head list then try it, otherwise |
| 1108 | * start over if we have not gotten any slots. |
| 1109 | */ |
| 1110 | if (plist_node_empty(&next->avail_lists[node])) |
| 1111 | goto start_over; |
| 1112 | } |
| 1113 | |
| 1114 | spin_unlock(&swap_avail_lock); |
| 1115 | |
| 1116 | check_out: |
| 1117 | if (n_ret < n_goal) |
| 1118 | atomic_long_add((long)(n_goal - n_ret) * size, |
| 1119 | &nr_swap_pages); |
| 1120 | noswap: |
| 1121 | return n_ret; |
| 1122 | } |
| 1123 | |
| 1124 | static struct swap_info_struct *__swap_info_get(swp_entry_t entry) |
| 1125 | { |
| 1126 | struct swap_info_struct *p; |
| 1127 | unsigned long offset; |
| 1128 | |
| 1129 | if (!entry.val) |
| 1130 | goto out; |
| 1131 | p = swp_swap_info(entry); |
| 1132 | if (!p) |
| 1133 | goto bad_nofile; |
| 1134 | if (data_race(!(p->flags & SWP_USED))) |
| 1135 | goto bad_device; |
| 1136 | offset = swp_offset(entry); |
| 1137 | if (offset >= p->max) |
| 1138 | goto bad_offset; |
| 1139 | return p; |
| 1140 | |
| 1141 | bad_offset: |
| 1142 | pr_err("%s: %s%08lx\n", __func__, Bad_offset, entry.val); |
| 1143 | goto out; |
| 1144 | bad_device: |
| 1145 | pr_err("%s: %s%08lx\n", __func__, Unused_file, entry.val); |
| 1146 | goto out; |
| 1147 | bad_nofile: |
| 1148 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
| 1149 | out: |
| 1150 | return NULL; |
| 1151 | } |
| 1152 | |
| 1153 | static struct swap_info_struct *_swap_info_get(swp_entry_t entry) |
| 1154 | { |
| 1155 | struct swap_info_struct *p; |
| 1156 | |
| 1157 | p = __swap_info_get(entry); |
| 1158 | if (!p) |
| 1159 | goto out; |
| 1160 | if (data_race(!p->swap_map[swp_offset(entry)])) |
| 1161 | goto bad_free; |
| 1162 | return p; |
| 1163 | |
| 1164 | bad_free: |
| 1165 | pr_err("%s: %s%08lx\n", __func__, Unused_offset, entry.val); |
| 1166 | out: |
| 1167 | return NULL; |
| 1168 | } |
| 1169 | |
| 1170 | static struct swap_info_struct *swap_info_get_cont(swp_entry_t entry, |
| 1171 | struct swap_info_struct *q) |
| 1172 | { |
| 1173 | struct swap_info_struct *p; |
| 1174 | |
| 1175 | p = _swap_info_get(entry); |
| 1176 | |
| 1177 | if (p != q) { |
| 1178 | if (q != NULL) |
| 1179 | spin_unlock(&q->lock); |
| 1180 | if (p != NULL) |
| 1181 | spin_lock(&p->lock); |
| 1182 | } |
| 1183 | return p; |
| 1184 | } |
| 1185 | |
| 1186 | static unsigned char __swap_entry_free_locked(struct swap_info_struct *p, |
| 1187 | unsigned long offset, |
| 1188 | unsigned char usage) |
| 1189 | { |
| 1190 | unsigned char count; |
| 1191 | unsigned char has_cache; |
| 1192 | |
| 1193 | count = p->swap_map[offset]; |
| 1194 | |
| 1195 | has_cache = count & SWAP_HAS_CACHE; |
| 1196 | count &= ~SWAP_HAS_CACHE; |
| 1197 | |
| 1198 | if (usage == SWAP_HAS_CACHE) { |
| 1199 | VM_BUG_ON(!has_cache); |
| 1200 | has_cache = 0; |
| 1201 | } else if (count == SWAP_MAP_SHMEM) { |
| 1202 | /* |
| 1203 | * Or we could insist on shmem.c using a special |
| 1204 | * swap_shmem_free() and free_shmem_swap_and_cache()... |
| 1205 | */ |
| 1206 | count = 0; |
| 1207 | } else if ((count & ~COUNT_CONTINUED) <= SWAP_MAP_MAX) { |
| 1208 | if (count == COUNT_CONTINUED) { |
| 1209 | if (swap_count_continued(p, offset, count)) |
| 1210 | count = SWAP_MAP_MAX | COUNT_CONTINUED; |
| 1211 | else |
| 1212 | count = SWAP_MAP_MAX; |
| 1213 | } else |
| 1214 | count--; |
| 1215 | } |
| 1216 | |
| 1217 | usage = count | has_cache; |
| 1218 | if (usage) |
| 1219 | WRITE_ONCE(p->swap_map[offset], usage); |
| 1220 | else |
| 1221 | WRITE_ONCE(p->swap_map[offset], SWAP_HAS_CACHE); |
| 1222 | |
| 1223 | return usage; |
| 1224 | } |
| 1225 | |
| 1226 | /* |
| 1227 | * Check whether swap entry is valid in the swap device. If so, |
| 1228 | * return pointer to swap_info_struct, and keep the swap entry valid |
| 1229 | * via preventing the swap device from being swapoff, until |
| 1230 | * put_swap_device() is called. Otherwise return NULL. |
| 1231 | * |
| 1232 | * Notice that swapoff or swapoff+swapon can still happen before the |
| 1233 | * percpu_ref_tryget_live() in get_swap_device() or after the |
| 1234 | * percpu_ref_put() in put_swap_device() if there isn't any other way |
| 1235 | * to prevent swapoff, such as page lock, page table lock, etc. The |
| 1236 | * caller must be prepared for that. For example, the following |
| 1237 | * situation is possible. |
| 1238 | * |
| 1239 | * CPU1 CPU2 |
| 1240 | * do_swap_page() |
| 1241 | * ... swapoff+swapon |
| 1242 | * __read_swap_cache_async() |
| 1243 | * swapcache_prepare() |
| 1244 | * __swap_duplicate() |
| 1245 | * // check swap_map |
| 1246 | * // verify PTE not changed |
| 1247 | * |
| 1248 | * In __swap_duplicate(), the swap_map need to be checked before |
| 1249 | * changing partly because the specified swap entry may be for another |
| 1250 | * swap device which has been swapoff. And in do_swap_page(), after |
| 1251 | * the page is read from the swap device, the PTE is verified not |
| 1252 | * changed with the page table locked to check whether the swap device |
| 1253 | * has been swapoff or swapoff+swapon. |
| 1254 | */ |
| 1255 | struct swap_info_struct *get_swap_device(swp_entry_t entry) |
| 1256 | { |
| 1257 | struct swap_info_struct *si; |
| 1258 | unsigned long offset; |
| 1259 | |
| 1260 | if (!entry.val) |
| 1261 | goto out; |
| 1262 | si = swp_swap_info(entry); |
| 1263 | if (!si) |
| 1264 | goto bad_nofile; |
| 1265 | if (!percpu_ref_tryget_live(&si->users)) |
| 1266 | goto out; |
| 1267 | /* |
| 1268 | * Guarantee the si->users are checked before accessing other |
| 1269 | * fields of swap_info_struct. |
| 1270 | * |
| 1271 | * Paired with the spin_unlock() after setup_swap_info() in |
| 1272 | * enable_swap_info(). |
| 1273 | */ |
| 1274 | smp_rmb(); |
| 1275 | offset = swp_offset(entry); |
| 1276 | if (offset >= si->max) |
| 1277 | goto put_out; |
| 1278 | |
| 1279 | return si; |
| 1280 | bad_nofile: |
| 1281 | pr_err("%s: %s%08lx\n", __func__, Bad_file, entry.val); |
| 1282 | out: |
| 1283 | return NULL; |
| 1284 | put_out: |
| 1285 | percpu_ref_put(&si->users); |
| 1286 | return NULL; |
| 1287 | } |
| 1288 | |
| 1289 | static unsigned char __swap_entry_free(struct swap_info_struct *p, |
| 1290 | swp_entry_t entry) |
| 1291 | { |
| 1292 | struct swap_cluster_info *ci; |
| 1293 | unsigned long offset = swp_offset(entry); |
| 1294 | unsigned char usage; |
| 1295 | |
| 1296 | ci = lock_cluster_or_swap_info(p, offset); |
| 1297 | usage = __swap_entry_free_locked(p, offset, 1); |
| 1298 | unlock_cluster_or_swap_info(p, ci); |
| 1299 | if (!usage) |
| 1300 | free_swap_slot(entry); |
| 1301 | |
| 1302 | return usage; |
| 1303 | } |
| 1304 | |
| 1305 | static void swap_entry_free(struct swap_info_struct *p, swp_entry_t entry) |
| 1306 | { |
| 1307 | struct swap_cluster_info *ci; |
| 1308 | unsigned long offset = swp_offset(entry); |
| 1309 | unsigned char count; |
| 1310 | |
| 1311 | ci = lock_cluster(p, offset); |
| 1312 | count = p->swap_map[offset]; |
| 1313 | VM_BUG_ON(count != SWAP_HAS_CACHE); |
| 1314 | p->swap_map[offset] = 0; |
| 1315 | dec_cluster_info_page(p, p->cluster_info, offset); |
| 1316 | unlock_cluster(ci); |
| 1317 | |
| 1318 | mem_cgroup_uncharge_swap(entry, 1); |
| 1319 | swap_range_free(p, offset, 1); |
| 1320 | } |
| 1321 | |
| 1322 | /* |
| 1323 | * Caller has made sure that the swap device corresponding to entry |
| 1324 | * is still around or has not been recycled. |
| 1325 | */ |
| 1326 | void swap_free(swp_entry_t entry) |
| 1327 | { |
| 1328 | struct swap_info_struct *p; |
| 1329 | |
| 1330 | p = _swap_info_get(entry); |
| 1331 | if (p) |
| 1332 | __swap_entry_free(p, entry); |
| 1333 | } |
| 1334 | |
| 1335 | /* |
| 1336 | * Called after dropping swapcache to decrease refcnt to swap entries. |
| 1337 | */ |
| 1338 | void put_swap_page(struct page *page, swp_entry_t entry) |
| 1339 | { |
| 1340 | unsigned long offset = swp_offset(entry); |
| 1341 | unsigned long idx = offset / SWAPFILE_CLUSTER; |
| 1342 | struct swap_cluster_info *ci; |
| 1343 | struct swap_info_struct *si; |
| 1344 | unsigned char *map; |
| 1345 | unsigned int i, free_entries = 0; |
| 1346 | unsigned char val; |
| 1347 | int size = swap_entry_size(thp_nr_pages(page)); |
| 1348 | |
| 1349 | si = _swap_info_get(entry); |
| 1350 | if (!si) |
| 1351 | return; |
| 1352 | |
| 1353 | ci = lock_cluster_or_swap_info(si, offset); |
| 1354 | if (size == SWAPFILE_CLUSTER) { |
| 1355 | VM_BUG_ON(!cluster_is_huge(ci)); |
| 1356 | map = si->swap_map + offset; |
| 1357 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { |
| 1358 | val = map[i]; |
| 1359 | VM_BUG_ON(!(val & SWAP_HAS_CACHE)); |
| 1360 | if (val == SWAP_HAS_CACHE) |
| 1361 | free_entries++; |
| 1362 | } |
| 1363 | cluster_clear_huge(ci); |
| 1364 | if (free_entries == SWAPFILE_CLUSTER) { |
| 1365 | unlock_cluster_or_swap_info(si, ci); |
| 1366 | spin_lock(&si->lock); |
| 1367 | mem_cgroup_uncharge_swap(entry, SWAPFILE_CLUSTER); |
| 1368 | swap_free_cluster(si, idx); |
| 1369 | spin_unlock(&si->lock); |
| 1370 | return; |
| 1371 | } |
| 1372 | } |
| 1373 | for (i = 0; i < size; i++, entry.val++) { |
| 1374 | if (!__swap_entry_free_locked(si, offset + i, SWAP_HAS_CACHE)) { |
| 1375 | unlock_cluster_or_swap_info(si, ci); |
| 1376 | free_swap_slot(entry); |
| 1377 | if (i == size - 1) |
| 1378 | return; |
| 1379 | lock_cluster_or_swap_info(si, offset); |
| 1380 | } |
| 1381 | } |
| 1382 | unlock_cluster_or_swap_info(si, ci); |
| 1383 | } |
| 1384 | |
| 1385 | #ifdef CONFIG_THP_SWAP |
| 1386 | int split_swap_cluster(swp_entry_t entry) |
| 1387 | { |
| 1388 | struct swap_info_struct *si; |
| 1389 | struct swap_cluster_info *ci; |
| 1390 | unsigned long offset = swp_offset(entry); |
| 1391 | |
| 1392 | si = _swap_info_get(entry); |
| 1393 | if (!si) |
| 1394 | return -EBUSY; |
| 1395 | ci = lock_cluster(si, offset); |
| 1396 | cluster_clear_huge(ci); |
| 1397 | unlock_cluster(ci); |
| 1398 | return 0; |
| 1399 | } |
| 1400 | #endif |
| 1401 | |
| 1402 | static int swp_entry_cmp(const void *ent1, const void *ent2) |
| 1403 | { |
| 1404 | const swp_entry_t *e1 = ent1, *e2 = ent2; |
| 1405 | |
| 1406 | return (int)swp_type(*e1) - (int)swp_type(*e2); |
| 1407 | } |
| 1408 | |
| 1409 | void swapcache_free_entries(swp_entry_t *entries, int n) |
| 1410 | { |
| 1411 | struct swap_info_struct *p, *prev; |
| 1412 | int i; |
| 1413 | |
| 1414 | if (n <= 0) |
| 1415 | return; |
| 1416 | |
| 1417 | prev = NULL; |
| 1418 | p = NULL; |
| 1419 | |
| 1420 | /* |
| 1421 | * Sort swap entries by swap device, so each lock is only taken once. |
| 1422 | * nr_swapfiles isn't absolutely correct, but the overhead of sort() is |
| 1423 | * so low that it isn't necessary to optimize further. |
| 1424 | */ |
| 1425 | if (nr_swapfiles > 1) |
| 1426 | sort(entries, n, sizeof(entries[0]), swp_entry_cmp, NULL); |
| 1427 | for (i = 0; i < n; ++i) { |
| 1428 | p = swap_info_get_cont(entries[i], prev); |
| 1429 | if (p) |
| 1430 | swap_entry_free(p, entries[i]); |
| 1431 | prev = p; |
| 1432 | } |
| 1433 | if (p) |
| 1434 | spin_unlock(&p->lock); |
| 1435 | } |
| 1436 | |
| 1437 | /* |
| 1438 | * How many references to page are currently swapped out? |
| 1439 | * This does not give an exact answer when swap count is continued, |
| 1440 | * but does include the high COUNT_CONTINUED flag to allow for that. |
| 1441 | */ |
| 1442 | int page_swapcount(struct page *page) |
| 1443 | { |
| 1444 | int count = 0; |
| 1445 | struct swap_info_struct *p; |
| 1446 | struct swap_cluster_info *ci; |
| 1447 | swp_entry_t entry; |
| 1448 | unsigned long offset; |
| 1449 | |
| 1450 | entry.val = page_private(page); |
| 1451 | p = _swap_info_get(entry); |
| 1452 | if (p) { |
| 1453 | offset = swp_offset(entry); |
| 1454 | ci = lock_cluster_or_swap_info(p, offset); |
| 1455 | count = swap_count(p->swap_map[offset]); |
| 1456 | unlock_cluster_or_swap_info(p, ci); |
| 1457 | } |
| 1458 | return count; |
| 1459 | } |
| 1460 | |
| 1461 | int __swap_count(swp_entry_t entry) |
| 1462 | { |
| 1463 | struct swap_info_struct *si; |
| 1464 | pgoff_t offset = swp_offset(entry); |
| 1465 | int count = 0; |
| 1466 | |
| 1467 | si = get_swap_device(entry); |
| 1468 | if (si) { |
| 1469 | count = swap_count(si->swap_map[offset]); |
| 1470 | put_swap_device(si); |
| 1471 | } |
| 1472 | return count; |
| 1473 | } |
| 1474 | |
| 1475 | static int swap_swapcount(struct swap_info_struct *si, swp_entry_t entry) |
| 1476 | { |
| 1477 | int count = 0; |
| 1478 | pgoff_t offset = swp_offset(entry); |
| 1479 | struct swap_cluster_info *ci; |
| 1480 | |
| 1481 | ci = lock_cluster_or_swap_info(si, offset); |
| 1482 | count = swap_count(si->swap_map[offset]); |
| 1483 | unlock_cluster_or_swap_info(si, ci); |
| 1484 | return count; |
| 1485 | } |
| 1486 | |
| 1487 | /* |
| 1488 | * How many references to @entry are currently swapped out? |
| 1489 | * This does not give an exact answer when swap count is continued, |
| 1490 | * but does include the high COUNT_CONTINUED flag to allow for that. |
| 1491 | */ |
| 1492 | int __swp_swapcount(swp_entry_t entry) |
| 1493 | { |
| 1494 | int count = 0; |
| 1495 | struct swap_info_struct *si; |
| 1496 | |
| 1497 | si = get_swap_device(entry); |
| 1498 | if (si) { |
| 1499 | count = swap_swapcount(si, entry); |
| 1500 | put_swap_device(si); |
| 1501 | } |
| 1502 | return count; |
| 1503 | } |
| 1504 | |
| 1505 | /* |
| 1506 | * How many references to @entry are currently swapped out? |
| 1507 | * This considers COUNT_CONTINUED so it returns exact answer. |
| 1508 | */ |
| 1509 | int swp_swapcount(swp_entry_t entry) |
| 1510 | { |
| 1511 | int count, tmp_count, n; |
| 1512 | struct swap_info_struct *p; |
| 1513 | struct swap_cluster_info *ci; |
| 1514 | struct page *page; |
| 1515 | pgoff_t offset; |
| 1516 | unsigned char *map; |
| 1517 | |
| 1518 | p = _swap_info_get(entry); |
| 1519 | if (!p) |
| 1520 | return 0; |
| 1521 | |
| 1522 | offset = swp_offset(entry); |
| 1523 | |
| 1524 | ci = lock_cluster_or_swap_info(p, offset); |
| 1525 | |
| 1526 | count = swap_count(p->swap_map[offset]); |
| 1527 | if (!(count & COUNT_CONTINUED)) |
| 1528 | goto out; |
| 1529 | |
| 1530 | count &= ~COUNT_CONTINUED; |
| 1531 | n = SWAP_MAP_MAX + 1; |
| 1532 | |
| 1533 | page = vmalloc_to_page(p->swap_map + offset); |
| 1534 | offset &= ~PAGE_MASK; |
| 1535 | VM_BUG_ON(page_private(page) != SWP_CONTINUED); |
| 1536 | |
| 1537 | do { |
| 1538 | page = list_next_entry(page, lru); |
| 1539 | map = kmap_atomic(page); |
| 1540 | tmp_count = map[offset]; |
| 1541 | kunmap_atomic(map); |
| 1542 | |
| 1543 | count += (tmp_count & ~COUNT_CONTINUED) * n; |
| 1544 | n *= (SWAP_CONT_MAX + 1); |
| 1545 | } while (tmp_count & COUNT_CONTINUED); |
| 1546 | out: |
| 1547 | unlock_cluster_or_swap_info(p, ci); |
| 1548 | return count; |
| 1549 | } |
| 1550 | |
| 1551 | static bool swap_page_trans_huge_swapped(struct swap_info_struct *si, |
| 1552 | swp_entry_t entry) |
| 1553 | { |
| 1554 | struct swap_cluster_info *ci; |
| 1555 | unsigned char *map = si->swap_map; |
| 1556 | unsigned long roffset = swp_offset(entry); |
| 1557 | unsigned long offset = round_down(roffset, SWAPFILE_CLUSTER); |
| 1558 | int i; |
| 1559 | bool ret = false; |
| 1560 | |
| 1561 | ci = lock_cluster_or_swap_info(si, offset); |
| 1562 | if (!ci || !cluster_is_huge(ci)) { |
| 1563 | if (swap_count(map[roffset])) |
| 1564 | ret = true; |
| 1565 | goto unlock_out; |
| 1566 | } |
| 1567 | for (i = 0; i < SWAPFILE_CLUSTER; i++) { |
| 1568 | if (swap_count(map[offset + i])) { |
| 1569 | ret = true; |
| 1570 | break; |
| 1571 | } |
| 1572 | } |
| 1573 | unlock_out: |
| 1574 | unlock_cluster_or_swap_info(si, ci); |
| 1575 | return ret; |
| 1576 | } |
| 1577 | |
| 1578 | static bool page_swapped(struct page *page) |
| 1579 | { |
| 1580 | swp_entry_t entry; |
| 1581 | struct swap_info_struct *si; |
| 1582 | |
| 1583 | if (!IS_ENABLED(CONFIG_THP_SWAP) || likely(!PageTransCompound(page))) |
| 1584 | return page_swapcount(page) != 0; |
| 1585 | |
| 1586 | page = compound_head(page); |
| 1587 | entry.val = page_private(page); |
| 1588 | si = _swap_info_get(entry); |
| 1589 | if (si) |
| 1590 | return swap_page_trans_huge_swapped(si, entry); |
| 1591 | return false; |
| 1592 | } |
| 1593 | |
| 1594 | /* |
| 1595 | * If swap is getting full, or if there are no more mappings of this page, |
| 1596 | * then try_to_free_swap is called to free its swap space. |
| 1597 | */ |
| 1598 | int try_to_free_swap(struct page *page) |
| 1599 | { |
| 1600 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 1601 | |
| 1602 | if (!PageSwapCache(page)) |
| 1603 | return 0; |
| 1604 | if (PageWriteback(page)) |
| 1605 | return 0; |
| 1606 | if (page_swapped(page)) |
| 1607 | return 0; |
| 1608 | |
| 1609 | /* |
| 1610 | * Once hibernation has begun to create its image of memory, |
| 1611 | * there's a danger that one of the calls to try_to_free_swap() |
| 1612 | * - most probably a call from __try_to_reclaim_swap() while |
| 1613 | * hibernation is allocating its own swap pages for the image, |
| 1614 | * but conceivably even a call from memory reclaim - will free |
| 1615 | * the swap from a page which has already been recorded in the |
| 1616 | * image as a clean swapcache page, and then reuse its swap for |
| 1617 | * another page of the image. On waking from hibernation, the |
| 1618 | * original page might be freed under memory pressure, then |
| 1619 | * later read back in from swap, now with the wrong data. |
| 1620 | * |
| 1621 | * Hibernation suspends storage while it is writing the image |
| 1622 | * to disk so check that here. |
| 1623 | */ |
| 1624 | if (pm_suspended_storage()) |
| 1625 | return 0; |
| 1626 | |
| 1627 | page = compound_head(page); |
| 1628 | delete_from_swap_cache(page); |
| 1629 | SetPageDirty(page); |
| 1630 | return 1; |
| 1631 | } |
| 1632 | |
| 1633 | /* |
| 1634 | * Free the swap entry like above, but also try to |
| 1635 | * free the page cache entry if it is the last user. |
| 1636 | */ |
| 1637 | int free_swap_and_cache(swp_entry_t entry) |
| 1638 | { |
| 1639 | struct swap_info_struct *p; |
| 1640 | unsigned char count; |
| 1641 | |
| 1642 | if (non_swap_entry(entry)) |
| 1643 | return 1; |
| 1644 | |
| 1645 | p = _swap_info_get(entry); |
| 1646 | if (p) { |
| 1647 | count = __swap_entry_free(p, entry); |
| 1648 | if (count == SWAP_HAS_CACHE && |
| 1649 | !swap_page_trans_huge_swapped(p, entry)) |
| 1650 | __try_to_reclaim_swap(p, swp_offset(entry), |
| 1651 | TTRS_UNMAPPED | TTRS_FULL); |
| 1652 | } |
| 1653 | return p != NULL; |
| 1654 | } |
| 1655 | |
| 1656 | #ifdef CONFIG_HIBERNATION |
| 1657 | |
| 1658 | swp_entry_t get_swap_page_of_type(int type) |
| 1659 | { |
| 1660 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
| 1661 | swp_entry_t entry = {0}; |
| 1662 | |
| 1663 | if (!si) |
| 1664 | goto fail; |
| 1665 | |
| 1666 | /* This is called for allocating swap entry, not cache */ |
| 1667 | spin_lock(&si->lock); |
| 1668 | if ((si->flags & SWP_WRITEOK) && scan_swap_map_slots(si, 1, 1, &entry)) |
| 1669 | atomic_long_dec(&nr_swap_pages); |
| 1670 | spin_unlock(&si->lock); |
| 1671 | fail: |
| 1672 | return entry; |
| 1673 | } |
| 1674 | |
| 1675 | /* |
| 1676 | * Find the swap type that corresponds to given device (if any). |
| 1677 | * |
| 1678 | * @offset - number of the PAGE_SIZE-sized block of the device, starting |
| 1679 | * from 0, in which the swap header is expected to be located. |
| 1680 | * |
| 1681 | * This is needed for the suspend to disk (aka swsusp). |
| 1682 | */ |
| 1683 | int swap_type_of(dev_t device, sector_t offset) |
| 1684 | { |
| 1685 | int type; |
| 1686 | |
| 1687 | if (!device) |
| 1688 | return -1; |
| 1689 | |
| 1690 | spin_lock(&swap_lock); |
| 1691 | for (type = 0; type < nr_swapfiles; type++) { |
| 1692 | struct swap_info_struct *sis = swap_info[type]; |
| 1693 | |
| 1694 | if (!(sis->flags & SWP_WRITEOK)) |
| 1695 | continue; |
| 1696 | |
| 1697 | if (device == sis->bdev->bd_dev) { |
| 1698 | struct swap_extent *se = first_se(sis); |
| 1699 | |
| 1700 | if (se->start_block == offset) { |
| 1701 | spin_unlock(&swap_lock); |
| 1702 | return type; |
| 1703 | } |
| 1704 | } |
| 1705 | } |
| 1706 | spin_unlock(&swap_lock); |
| 1707 | return -ENODEV; |
| 1708 | } |
| 1709 | |
| 1710 | int find_first_swap(dev_t *device) |
| 1711 | { |
| 1712 | int type; |
| 1713 | |
| 1714 | spin_lock(&swap_lock); |
| 1715 | for (type = 0; type < nr_swapfiles; type++) { |
| 1716 | struct swap_info_struct *sis = swap_info[type]; |
| 1717 | |
| 1718 | if (!(sis->flags & SWP_WRITEOK)) |
| 1719 | continue; |
| 1720 | *device = sis->bdev->bd_dev; |
| 1721 | spin_unlock(&swap_lock); |
| 1722 | return type; |
| 1723 | } |
| 1724 | spin_unlock(&swap_lock); |
| 1725 | return -ENODEV; |
| 1726 | } |
| 1727 | |
| 1728 | /* |
| 1729 | * Get the (PAGE_SIZE) block corresponding to given offset on the swapdev |
| 1730 | * corresponding to given index in swap_info (swap type). |
| 1731 | */ |
| 1732 | sector_t swapdev_block(int type, pgoff_t offset) |
| 1733 | { |
| 1734 | struct swap_info_struct *si = swap_type_to_swap_info(type); |
| 1735 | struct swap_extent *se; |
| 1736 | |
| 1737 | if (!si || !(si->flags & SWP_WRITEOK)) |
| 1738 | return 0; |
| 1739 | se = offset_to_swap_extent(si, offset); |
| 1740 | return se->start_block + (offset - se->start_page); |
| 1741 | } |
| 1742 | |
| 1743 | /* |
| 1744 | * Return either the total number of swap pages of given type, or the number |
| 1745 | * of free pages of that type (depending on @free) |
| 1746 | * |
| 1747 | * This is needed for software suspend |
| 1748 | */ |
| 1749 | unsigned int count_swap_pages(int type, int free) |
| 1750 | { |
| 1751 | unsigned int n = 0; |
| 1752 | |
| 1753 | spin_lock(&swap_lock); |
| 1754 | if ((unsigned int)type < nr_swapfiles) { |
| 1755 | struct swap_info_struct *sis = swap_info[type]; |
| 1756 | |
| 1757 | spin_lock(&sis->lock); |
| 1758 | if (sis->flags & SWP_WRITEOK) { |
| 1759 | n = sis->pages; |
| 1760 | if (free) |
| 1761 | n -= sis->inuse_pages; |
| 1762 | } |
| 1763 | spin_unlock(&sis->lock); |
| 1764 | } |
| 1765 | spin_unlock(&swap_lock); |
| 1766 | return n; |
| 1767 | } |
| 1768 | #endif /* CONFIG_HIBERNATION */ |
| 1769 | |
| 1770 | static inline int pte_same_as_swp(pte_t pte, pte_t swp_pte) |
| 1771 | { |
| 1772 | return pte_same(pte_swp_clear_flags(pte), swp_pte); |
| 1773 | } |
| 1774 | |
| 1775 | /* |
| 1776 | * No need to decide whether this PTE shares the swap entry with others, |
| 1777 | * just let do_wp_page work it out if a write is requested later - to |
| 1778 | * force COW, vm_page_prot omits write permission from any private vma. |
| 1779 | */ |
| 1780 | static int unuse_pte(struct vm_area_struct *vma, pmd_t *pmd, |
| 1781 | unsigned long addr, swp_entry_t entry, struct page *page) |
| 1782 | { |
| 1783 | struct page *swapcache; |
| 1784 | spinlock_t *ptl; |
| 1785 | pte_t *pte; |
| 1786 | int ret = 1; |
| 1787 | |
| 1788 | swapcache = page; |
| 1789 | page = ksm_might_need_to_copy(page, vma, addr); |
| 1790 | if (unlikely(!page)) |
| 1791 | return -ENOMEM; |
| 1792 | |
| 1793 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 1794 | if (unlikely(!pte_same_as_swp(*pte, swp_entry_to_pte(entry)))) { |
| 1795 | ret = 0; |
| 1796 | goto out; |
| 1797 | } |
| 1798 | |
| 1799 | dec_mm_counter(vma->vm_mm, MM_SWAPENTS); |
| 1800 | inc_mm_counter(vma->vm_mm, MM_ANONPAGES); |
| 1801 | get_page(page); |
| 1802 | if (page == swapcache) { |
| 1803 | page_add_anon_rmap(page, vma, addr, RMAP_NONE); |
| 1804 | } else { /* ksm created a completely new copy */ |
| 1805 | page_add_new_anon_rmap(page, vma, addr, false); |
| 1806 | lru_cache_add_inactive_or_unevictable(page, vma); |
| 1807 | } |
| 1808 | set_pte_at(vma->vm_mm, addr, pte, |
| 1809 | pte_mkold(mk_pte(page, vma->vm_page_prot))); |
| 1810 | swap_free(entry); |
| 1811 | out: |
| 1812 | pte_unmap_unlock(pte, ptl); |
| 1813 | if (page != swapcache) { |
| 1814 | unlock_page(page); |
| 1815 | put_page(page); |
| 1816 | } |
| 1817 | return ret; |
| 1818 | } |
| 1819 | |
| 1820 | static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd, |
| 1821 | unsigned long addr, unsigned long end, |
| 1822 | unsigned int type) |
| 1823 | { |
| 1824 | struct page *page; |
| 1825 | swp_entry_t entry; |
| 1826 | pte_t *pte; |
| 1827 | struct swap_info_struct *si; |
| 1828 | unsigned long offset; |
| 1829 | int ret = 0; |
| 1830 | volatile unsigned char *swap_map; |
| 1831 | |
| 1832 | si = swap_info[type]; |
| 1833 | pte = pte_offset_map(pmd, addr); |
| 1834 | do { |
| 1835 | if (!is_swap_pte(*pte)) |
| 1836 | continue; |
| 1837 | |
| 1838 | entry = pte_to_swp_entry(*pte); |
| 1839 | if (swp_type(entry) != type) |
| 1840 | continue; |
| 1841 | |
| 1842 | offset = swp_offset(entry); |
| 1843 | pte_unmap(pte); |
| 1844 | swap_map = &si->swap_map[offset]; |
| 1845 | page = lookup_swap_cache(entry, vma, addr); |
| 1846 | if (!page) { |
| 1847 | struct vm_fault vmf = { |
| 1848 | .vma = vma, |
| 1849 | .address = addr, |
| 1850 | .real_address = addr, |
| 1851 | .pmd = pmd, |
| 1852 | }; |
| 1853 | |
| 1854 | page = swapin_readahead(entry, GFP_HIGHUSER_MOVABLE, |
| 1855 | &vmf); |
| 1856 | } |
| 1857 | if (!page) { |
| 1858 | if (*swap_map == 0 || *swap_map == SWAP_MAP_BAD) |
| 1859 | goto try_next; |
| 1860 | return -ENOMEM; |
| 1861 | } |
| 1862 | |
| 1863 | lock_page(page); |
| 1864 | wait_on_page_writeback(page); |
| 1865 | ret = unuse_pte(vma, pmd, addr, entry, page); |
| 1866 | if (ret < 0) { |
| 1867 | unlock_page(page); |
| 1868 | put_page(page); |
| 1869 | goto out; |
| 1870 | } |
| 1871 | |
| 1872 | try_to_free_swap(page); |
| 1873 | unlock_page(page); |
| 1874 | put_page(page); |
| 1875 | try_next: |
| 1876 | pte = pte_offset_map(pmd, addr); |
| 1877 | } while (pte++, addr += PAGE_SIZE, addr != end); |
| 1878 | pte_unmap(pte - 1); |
| 1879 | |
| 1880 | ret = 0; |
| 1881 | out: |
| 1882 | return ret; |
| 1883 | } |
| 1884 | |
| 1885 | static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud, |
| 1886 | unsigned long addr, unsigned long end, |
| 1887 | unsigned int type) |
| 1888 | { |
| 1889 | pmd_t *pmd; |
| 1890 | unsigned long next; |
| 1891 | int ret; |
| 1892 | |
| 1893 | pmd = pmd_offset(pud, addr); |
| 1894 | do { |
| 1895 | cond_resched(); |
| 1896 | next = pmd_addr_end(addr, end); |
| 1897 | if (pmd_none_or_trans_huge_or_clear_bad(pmd)) |
| 1898 | continue; |
| 1899 | ret = unuse_pte_range(vma, pmd, addr, next, type); |
| 1900 | if (ret) |
| 1901 | return ret; |
| 1902 | } while (pmd++, addr = next, addr != end); |
| 1903 | return 0; |
| 1904 | } |
| 1905 | |
| 1906 | static inline int unuse_pud_range(struct vm_area_struct *vma, p4d_t *p4d, |
| 1907 | unsigned long addr, unsigned long end, |
| 1908 | unsigned int type) |
| 1909 | { |
| 1910 | pud_t *pud; |
| 1911 | unsigned long next; |
| 1912 | int ret; |
| 1913 | |
| 1914 | pud = pud_offset(p4d, addr); |
| 1915 | do { |
| 1916 | next = pud_addr_end(addr, end); |
| 1917 | if (pud_none_or_clear_bad(pud)) |
| 1918 | continue; |
| 1919 | ret = unuse_pmd_range(vma, pud, addr, next, type); |
| 1920 | if (ret) |
| 1921 | return ret; |
| 1922 | } while (pud++, addr = next, addr != end); |
| 1923 | return 0; |
| 1924 | } |
| 1925 | |
| 1926 | static inline int unuse_p4d_range(struct vm_area_struct *vma, pgd_t *pgd, |
| 1927 | unsigned long addr, unsigned long end, |
| 1928 | unsigned int type) |
| 1929 | { |
| 1930 | p4d_t *p4d; |
| 1931 | unsigned long next; |
| 1932 | int ret; |
| 1933 | |
| 1934 | p4d = p4d_offset(pgd, addr); |
| 1935 | do { |
| 1936 | next = p4d_addr_end(addr, end); |
| 1937 | if (p4d_none_or_clear_bad(p4d)) |
| 1938 | continue; |
| 1939 | ret = unuse_pud_range(vma, p4d, addr, next, type); |
| 1940 | if (ret) |
| 1941 | return ret; |
| 1942 | } while (p4d++, addr = next, addr != end); |
| 1943 | return 0; |
| 1944 | } |
| 1945 | |
| 1946 | static int unuse_vma(struct vm_area_struct *vma, unsigned int type) |
| 1947 | { |
| 1948 | pgd_t *pgd; |
| 1949 | unsigned long addr, end, next; |
| 1950 | int ret; |
| 1951 | |
| 1952 | addr = vma->vm_start; |
| 1953 | end = vma->vm_end; |
| 1954 | |
| 1955 | pgd = pgd_offset(vma->vm_mm, addr); |
| 1956 | do { |
| 1957 | next = pgd_addr_end(addr, end); |
| 1958 | if (pgd_none_or_clear_bad(pgd)) |
| 1959 | continue; |
| 1960 | ret = unuse_p4d_range(vma, pgd, addr, next, type); |
| 1961 | if (ret) |
| 1962 | return ret; |
| 1963 | } while (pgd++, addr = next, addr != end); |
| 1964 | return 0; |
| 1965 | } |
| 1966 | |
| 1967 | static int unuse_mm(struct mm_struct *mm, unsigned int type) |
| 1968 | { |
| 1969 | struct vm_area_struct *vma; |
| 1970 | int ret = 0; |
| 1971 | |
| 1972 | mmap_read_lock(mm); |
| 1973 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 1974 | if (vma->anon_vma) { |
| 1975 | ret = unuse_vma(vma, type); |
| 1976 | if (ret) |
| 1977 | break; |
| 1978 | } |
| 1979 | cond_resched(); |
| 1980 | } |
| 1981 | mmap_read_unlock(mm); |
| 1982 | return ret; |
| 1983 | } |
| 1984 | |
| 1985 | /* |
| 1986 | * Scan swap_map (or frontswap_map if frontswap parameter is true) |
| 1987 | * from current position to next entry still in use. Return 0 |
| 1988 | * if there are no inuse entries after prev till end of the map. |
| 1989 | */ |
| 1990 | static unsigned int find_next_to_unuse(struct swap_info_struct *si, |
| 1991 | unsigned int prev) |
| 1992 | { |
| 1993 | unsigned int i; |
| 1994 | unsigned char count; |
| 1995 | |
| 1996 | /* |
| 1997 | * No need for swap_lock here: we're just looking |
| 1998 | * for whether an entry is in use, not modifying it; false |
| 1999 | * hits are okay, and sys_swapoff() has already prevented new |
| 2000 | * allocations from this area (while holding swap_lock). |
| 2001 | */ |
| 2002 | for (i = prev + 1; i < si->max; i++) { |
| 2003 | count = READ_ONCE(si->swap_map[i]); |
| 2004 | if (count && swap_count(count) != SWAP_MAP_BAD) |
| 2005 | break; |
| 2006 | if ((i % LATENCY_LIMIT) == 0) |
| 2007 | cond_resched(); |
| 2008 | } |
| 2009 | |
| 2010 | if (i == si->max) |
| 2011 | i = 0; |
| 2012 | |
| 2013 | return i; |
| 2014 | } |
| 2015 | |
| 2016 | static int try_to_unuse(unsigned int type) |
| 2017 | { |
| 2018 | struct mm_struct *prev_mm; |
| 2019 | struct mm_struct *mm; |
| 2020 | struct list_head *p; |
| 2021 | int retval = 0; |
| 2022 | struct swap_info_struct *si = swap_info[type]; |
| 2023 | struct page *page; |
| 2024 | swp_entry_t entry; |
| 2025 | unsigned int i; |
| 2026 | |
| 2027 | if (!READ_ONCE(si->inuse_pages)) |
| 2028 | return 0; |
| 2029 | |
| 2030 | retry: |
| 2031 | retval = shmem_unuse(type); |
| 2032 | if (retval) |
| 2033 | return retval; |
| 2034 | |
| 2035 | prev_mm = &init_mm; |
| 2036 | mmget(prev_mm); |
| 2037 | |
| 2038 | spin_lock(&mmlist_lock); |
| 2039 | p = &init_mm.mmlist; |
| 2040 | while (READ_ONCE(si->inuse_pages) && |
| 2041 | !signal_pending(current) && |
| 2042 | (p = p->next) != &init_mm.mmlist) { |
| 2043 | |
| 2044 | mm = list_entry(p, struct mm_struct, mmlist); |
| 2045 | if (!mmget_not_zero(mm)) |
| 2046 | continue; |
| 2047 | spin_unlock(&mmlist_lock); |
| 2048 | mmput(prev_mm); |
| 2049 | prev_mm = mm; |
| 2050 | retval = unuse_mm(mm, type); |
| 2051 | if (retval) { |
| 2052 | mmput(prev_mm); |
| 2053 | return retval; |
| 2054 | } |
| 2055 | |
| 2056 | /* |
| 2057 | * Make sure that we aren't completely killing |
| 2058 | * interactive performance. |
| 2059 | */ |
| 2060 | cond_resched(); |
| 2061 | spin_lock(&mmlist_lock); |
| 2062 | } |
| 2063 | spin_unlock(&mmlist_lock); |
| 2064 | |
| 2065 | mmput(prev_mm); |
| 2066 | |
| 2067 | i = 0; |
| 2068 | while (READ_ONCE(si->inuse_pages) && |
| 2069 | !signal_pending(current) && |
| 2070 | (i = find_next_to_unuse(si, i)) != 0) { |
| 2071 | |
| 2072 | entry = swp_entry(type, i); |
| 2073 | page = find_get_page(swap_address_space(entry), i); |
| 2074 | if (!page) |
| 2075 | continue; |
| 2076 | |
| 2077 | /* |
| 2078 | * It is conceivable that a racing task removed this page from |
| 2079 | * swap cache just before we acquired the page lock. The page |
| 2080 | * might even be back in swap cache on another swap area. But |
| 2081 | * that is okay, try_to_free_swap() only removes stale pages. |
| 2082 | */ |
| 2083 | lock_page(page); |
| 2084 | wait_on_page_writeback(page); |
| 2085 | try_to_free_swap(page); |
| 2086 | unlock_page(page); |
| 2087 | put_page(page); |
| 2088 | } |
| 2089 | |
| 2090 | /* |
| 2091 | * Lets check again to see if there are still swap entries in the map. |
| 2092 | * If yes, we would need to do retry the unuse logic again. |
| 2093 | * Under global memory pressure, swap entries can be reinserted back |
| 2094 | * into process space after the mmlist loop above passes over them. |
| 2095 | * |
| 2096 | * Limit the number of retries? No: when mmget_not_zero() above fails, |
| 2097 | * that mm is likely to be freeing swap from exit_mmap(), which proceeds |
| 2098 | * at its own independent pace; and even shmem_writepage() could have |
| 2099 | * been preempted after get_swap_page(), temporarily hiding that swap. |
| 2100 | * It's easy and robust (though cpu-intensive) just to keep retrying. |
| 2101 | */ |
| 2102 | if (READ_ONCE(si->inuse_pages)) { |
| 2103 | if (!signal_pending(current)) |
| 2104 | goto retry; |
| 2105 | return -EINTR; |
| 2106 | } |
| 2107 | |
| 2108 | return 0; |
| 2109 | } |
| 2110 | |
| 2111 | /* |
| 2112 | * After a successful try_to_unuse, if no swap is now in use, we know |
| 2113 | * we can empty the mmlist. swap_lock must be held on entry and exit. |
| 2114 | * Note that mmlist_lock nests inside swap_lock, and an mm must be |
| 2115 | * added to the mmlist just after page_duplicate - before would be racy. |
| 2116 | */ |
| 2117 | static void drain_mmlist(void) |
| 2118 | { |
| 2119 | struct list_head *p, *next; |
| 2120 | unsigned int type; |
| 2121 | |
| 2122 | for (type = 0; type < nr_swapfiles; type++) |
| 2123 | if (swap_info[type]->inuse_pages) |
| 2124 | return; |
| 2125 | spin_lock(&mmlist_lock); |
| 2126 | list_for_each_safe(p, next, &init_mm.mmlist) |
| 2127 | list_del_init(p); |
| 2128 | spin_unlock(&mmlist_lock); |
| 2129 | } |
| 2130 | |
| 2131 | /* |
| 2132 | * Free all of a swapdev's extent information |
| 2133 | */ |
| 2134 | static void destroy_swap_extents(struct swap_info_struct *sis) |
| 2135 | { |
| 2136 | while (!RB_EMPTY_ROOT(&sis->swap_extent_root)) { |
| 2137 | struct rb_node *rb = sis->swap_extent_root.rb_node; |
| 2138 | struct swap_extent *se = rb_entry(rb, struct swap_extent, rb_node); |
| 2139 | |
| 2140 | rb_erase(rb, &sis->swap_extent_root); |
| 2141 | kfree(se); |
| 2142 | } |
| 2143 | |
| 2144 | if (sis->flags & SWP_ACTIVATED) { |
| 2145 | struct file *swap_file = sis->swap_file; |
| 2146 | struct address_space *mapping = swap_file->f_mapping; |
| 2147 | |
| 2148 | sis->flags &= ~SWP_ACTIVATED; |
| 2149 | if (mapping->a_ops->swap_deactivate) |
| 2150 | mapping->a_ops->swap_deactivate(swap_file); |
| 2151 | } |
| 2152 | } |
| 2153 | |
| 2154 | /* |
| 2155 | * Add a block range (and the corresponding page range) into this swapdev's |
| 2156 | * extent tree. |
| 2157 | * |
| 2158 | * This function rather assumes that it is called in ascending page order. |
| 2159 | */ |
| 2160 | int |
| 2161 | add_swap_extent(struct swap_info_struct *sis, unsigned long start_page, |
| 2162 | unsigned long nr_pages, sector_t start_block) |
| 2163 | { |
| 2164 | struct rb_node **link = &sis->swap_extent_root.rb_node, *parent = NULL; |
| 2165 | struct swap_extent *se; |
| 2166 | struct swap_extent *new_se; |
| 2167 | |
| 2168 | /* |
| 2169 | * place the new node at the right most since the |
| 2170 | * function is called in ascending page order. |
| 2171 | */ |
| 2172 | while (*link) { |
| 2173 | parent = *link; |
| 2174 | link = &parent->rb_right; |
| 2175 | } |
| 2176 | |
| 2177 | if (parent) { |
| 2178 | se = rb_entry(parent, struct swap_extent, rb_node); |
| 2179 | BUG_ON(se->start_page + se->nr_pages != start_page); |
| 2180 | if (se->start_block + se->nr_pages == start_block) { |
| 2181 | /* Merge it */ |
| 2182 | se->nr_pages += nr_pages; |
| 2183 | return 0; |
| 2184 | } |
| 2185 | } |
| 2186 | |
| 2187 | /* No merge, insert a new extent. */ |
| 2188 | new_se = kmalloc(sizeof(*se), GFP_KERNEL); |
| 2189 | if (new_se == NULL) |
| 2190 | return -ENOMEM; |
| 2191 | new_se->start_page = start_page; |
| 2192 | new_se->nr_pages = nr_pages; |
| 2193 | new_se->start_block = start_block; |
| 2194 | |
| 2195 | rb_link_node(&new_se->rb_node, parent, link); |
| 2196 | rb_insert_color(&new_se->rb_node, &sis->swap_extent_root); |
| 2197 | return 1; |
| 2198 | } |
| 2199 | EXPORT_SYMBOL_GPL(add_swap_extent); |
| 2200 | |
| 2201 | /* |
| 2202 | * A `swap extent' is a simple thing which maps a contiguous range of pages |
| 2203 | * onto a contiguous range of disk blocks. An ordered list of swap extents |
| 2204 | * is built at swapon time and is then used at swap_writepage/swap_readpage |
| 2205 | * time for locating where on disk a page belongs. |
| 2206 | * |
| 2207 | * If the swapfile is an S_ISBLK block device, a single extent is installed. |
| 2208 | * This is done so that the main operating code can treat S_ISBLK and S_ISREG |
| 2209 | * swap files identically. |
| 2210 | * |
| 2211 | * Whether the swapdev is an S_ISREG file or an S_ISBLK blockdev, the swap |
| 2212 | * extent list operates in PAGE_SIZE disk blocks. Both S_ISREG and S_ISBLK |
| 2213 | * swapfiles are handled *identically* after swapon time. |
| 2214 | * |
| 2215 | * For S_ISREG swapfiles, setup_swap_extents() will walk all the file's blocks |
| 2216 | * and will parse them into an ordered extent list, in PAGE_SIZE chunks. If |
| 2217 | * some stray blocks are found which do not fall within the PAGE_SIZE alignment |
| 2218 | * requirements, they are simply tossed out - we will never use those blocks |
| 2219 | * for swapping. |
| 2220 | * |
| 2221 | * For all swap devices we set S_SWAPFILE across the life of the swapon. This |
| 2222 | * prevents users from writing to the swap device, which will corrupt memory. |
| 2223 | * |
| 2224 | * The amount of disk space which a single swap extent represents varies. |
| 2225 | * Typically it is in the 1-4 megabyte range. So we can have hundreds of |
| 2226 | * extents in the list. To avoid much list walking, we cache the previous |
| 2227 | * search location in `curr_swap_extent', and start new searches from there. |
| 2228 | * This is extremely effective. The average number of iterations in |
| 2229 | * map_swap_page() has been measured at about 0.3 per page. - akpm. |
| 2230 | */ |
| 2231 | static int setup_swap_extents(struct swap_info_struct *sis, sector_t *span) |
| 2232 | { |
| 2233 | struct file *swap_file = sis->swap_file; |
| 2234 | struct address_space *mapping = swap_file->f_mapping; |
| 2235 | struct inode *inode = mapping->host; |
| 2236 | int ret; |
| 2237 | |
| 2238 | if (S_ISBLK(inode->i_mode)) { |
| 2239 | ret = add_swap_extent(sis, 0, sis->max, 0); |
| 2240 | *span = sis->pages; |
| 2241 | return ret; |
| 2242 | } |
| 2243 | |
| 2244 | if (mapping->a_ops->swap_activate) { |
| 2245 | ret = mapping->a_ops->swap_activate(sis, swap_file, span); |
| 2246 | if (ret >= 0) |
| 2247 | sis->flags |= SWP_ACTIVATED; |
| 2248 | if (!ret) { |
| 2249 | sis->flags |= SWP_FS_OPS; |
| 2250 | ret = add_swap_extent(sis, 0, sis->max, 0); |
| 2251 | *span = sis->pages; |
| 2252 | } |
| 2253 | return ret; |
| 2254 | } |
| 2255 | |
| 2256 | return generic_swapfile_activate(sis, swap_file, span); |
| 2257 | } |
| 2258 | |
| 2259 | static int swap_node(struct swap_info_struct *p) |
| 2260 | { |
| 2261 | struct block_device *bdev; |
| 2262 | |
| 2263 | if (p->bdev) |
| 2264 | bdev = p->bdev; |
| 2265 | else |
| 2266 | bdev = p->swap_file->f_inode->i_sb->s_bdev; |
| 2267 | |
| 2268 | return bdev ? bdev->bd_disk->node_id : NUMA_NO_NODE; |
| 2269 | } |
| 2270 | |
| 2271 | static void setup_swap_info(struct swap_info_struct *p, int prio, |
| 2272 | unsigned char *swap_map, |
| 2273 | struct swap_cluster_info *cluster_info) |
| 2274 | { |
| 2275 | int i; |
| 2276 | |
| 2277 | if (prio >= 0) |
| 2278 | p->prio = prio; |
| 2279 | else |
| 2280 | p->prio = --least_priority; |
| 2281 | /* |
| 2282 | * the plist prio is negated because plist ordering is |
| 2283 | * low-to-high, while swap ordering is high-to-low |
| 2284 | */ |
| 2285 | p->list.prio = -p->prio; |
| 2286 | for_each_node(i) { |
| 2287 | if (p->prio >= 0) |
| 2288 | p->avail_lists[i].prio = -p->prio; |
| 2289 | else { |
| 2290 | if (swap_node(p) == i) |
| 2291 | p->avail_lists[i].prio = 1; |
| 2292 | else |
| 2293 | p->avail_lists[i].prio = -p->prio; |
| 2294 | } |
| 2295 | } |
| 2296 | p->swap_map = swap_map; |
| 2297 | p->cluster_info = cluster_info; |
| 2298 | } |
| 2299 | |
| 2300 | static void _enable_swap_info(struct swap_info_struct *p) |
| 2301 | { |
| 2302 | p->flags |= SWP_WRITEOK; |
| 2303 | atomic_long_add(p->pages, &nr_swap_pages); |
| 2304 | total_swap_pages += p->pages; |
| 2305 | |
| 2306 | assert_spin_locked(&swap_lock); |
| 2307 | /* |
| 2308 | * both lists are plists, and thus priority ordered. |
| 2309 | * swap_active_head needs to be priority ordered for swapoff(), |
| 2310 | * which on removal of any swap_info_struct with an auto-assigned |
| 2311 | * (i.e. negative) priority increments the auto-assigned priority |
| 2312 | * of any lower-priority swap_info_structs. |
| 2313 | * swap_avail_head needs to be priority ordered for get_swap_page(), |
| 2314 | * which allocates swap pages from the highest available priority |
| 2315 | * swap_info_struct. |
| 2316 | */ |
| 2317 | plist_add(&p->list, &swap_active_head); |
| 2318 | add_to_avail_list(p); |
| 2319 | } |
| 2320 | |
| 2321 | static void enable_swap_info(struct swap_info_struct *p, int prio, |
| 2322 | unsigned char *swap_map, |
| 2323 | struct swap_cluster_info *cluster_info, |
| 2324 | unsigned long *frontswap_map) |
| 2325 | { |
| 2326 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
| 2327 | frontswap_init(p->type, frontswap_map); |
| 2328 | spin_lock(&swap_lock); |
| 2329 | spin_lock(&p->lock); |
| 2330 | setup_swap_info(p, prio, swap_map, cluster_info); |
| 2331 | spin_unlock(&p->lock); |
| 2332 | spin_unlock(&swap_lock); |
| 2333 | /* |
| 2334 | * Finished initializing swap device, now it's safe to reference it. |
| 2335 | */ |
| 2336 | percpu_ref_resurrect(&p->users); |
| 2337 | spin_lock(&swap_lock); |
| 2338 | spin_lock(&p->lock); |
| 2339 | _enable_swap_info(p); |
| 2340 | spin_unlock(&p->lock); |
| 2341 | spin_unlock(&swap_lock); |
| 2342 | } |
| 2343 | |
| 2344 | static void reinsert_swap_info(struct swap_info_struct *p) |
| 2345 | { |
| 2346 | spin_lock(&swap_lock); |
| 2347 | spin_lock(&p->lock); |
| 2348 | setup_swap_info(p, p->prio, p->swap_map, p->cluster_info); |
| 2349 | _enable_swap_info(p); |
| 2350 | spin_unlock(&p->lock); |
| 2351 | spin_unlock(&swap_lock); |
| 2352 | } |
| 2353 | |
| 2354 | bool has_usable_swap(void) |
| 2355 | { |
| 2356 | bool ret = true; |
| 2357 | |
| 2358 | spin_lock(&swap_lock); |
| 2359 | if (plist_head_empty(&swap_active_head)) |
| 2360 | ret = false; |
| 2361 | spin_unlock(&swap_lock); |
| 2362 | return ret; |
| 2363 | } |
| 2364 | |
| 2365 | SYSCALL_DEFINE1(swapoff, const char __user *, specialfile) |
| 2366 | { |
| 2367 | struct swap_info_struct *p = NULL; |
| 2368 | unsigned char *swap_map; |
| 2369 | struct swap_cluster_info *cluster_info; |
| 2370 | unsigned long *frontswap_map; |
| 2371 | struct file *swap_file, *victim; |
| 2372 | struct address_space *mapping; |
| 2373 | struct inode *inode; |
| 2374 | struct filename *pathname; |
| 2375 | int err, found = 0; |
| 2376 | unsigned int old_block_size; |
| 2377 | |
| 2378 | if (!capable(CAP_SYS_ADMIN)) |
| 2379 | return -EPERM; |
| 2380 | |
| 2381 | BUG_ON(!current->mm); |
| 2382 | |
| 2383 | pathname = getname(specialfile); |
| 2384 | if (IS_ERR(pathname)) |
| 2385 | return PTR_ERR(pathname); |
| 2386 | |
| 2387 | victim = file_open_name(pathname, O_RDWR|O_LARGEFILE, 0); |
| 2388 | err = PTR_ERR(victim); |
| 2389 | if (IS_ERR(victim)) |
| 2390 | goto out; |
| 2391 | |
| 2392 | mapping = victim->f_mapping; |
| 2393 | spin_lock(&swap_lock); |
| 2394 | plist_for_each_entry(p, &swap_active_head, list) { |
| 2395 | if (p->flags & SWP_WRITEOK) { |
| 2396 | if (p->swap_file->f_mapping == mapping) { |
| 2397 | found = 1; |
| 2398 | break; |
| 2399 | } |
| 2400 | } |
| 2401 | } |
| 2402 | if (!found) { |
| 2403 | err = -EINVAL; |
| 2404 | spin_unlock(&swap_lock); |
| 2405 | goto out_dput; |
| 2406 | } |
| 2407 | if (!security_vm_enough_memory_mm(current->mm, p->pages)) |
| 2408 | vm_unacct_memory(p->pages); |
| 2409 | else { |
| 2410 | err = -ENOMEM; |
| 2411 | spin_unlock(&swap_lock); |
| 2412 | goto out_dput; |
| 2413 | } |
| 2414 | del_from_avail_list(p); |
| 2415 | spin_lock(&p->lock); |
| 2416 | if (p->prio < 0) { |
| 2417 | struct swap_info_struct *si = p; |
| 2418 | int nid; |
| 2419 | |
| 2420 | plist_for_each_entry_continue(si, &swap_active_head, list) { |
| 2421 | si->prio++; |
| 2422 | si->list.prio--; |
| 2423 | for_each_node(nid) { |
| 2424 | if (si->avail_lists[nid].prio != 1) |
| 2425 | si->avail_lists[nid].prio--; |
| 2426 | } |
| 2427 | } |
| 2428 | least_priority++; |
| 2429 | } |
| 2430 | plist_del(&p->list, &swap_active_head); |
| 2431 | atomic_long_sub(p->pages, &nr_swap_pages); |
| 2432 | total_swap_pages -= p->pages; |
| 2433 | p->flags &= ~SWP_WRITEOK; |
| 2434 | spin_unlock(&p->lock); |
| 2435 | spin_unlock(&swap_lock); |
| 2436 | |
| 2437 | disable_swap_slots_cache_lock(); |
| 2438 | |
| 2439 | set_current_oom_origin(); |
| 2440 | err = try_to_unuse(p->type); |
| 2441 | clear_current_oom_origin(); |
| 2442 | |
| 2443 | if (err) { |
| 2444 | /* re-insert swap space back into swap_list */ |
| 2445 | reinsert_swap_info(p); |
| 2446 | reenable_swap_slots_cache_unlock(); |
| 2447 | goto out_dput; |
| 2448 | } |
| 2449 | |
| 2450 | reenable_swap_slots_cache_unlock(); |
| 2451 | |
| 2452 | /* |
| 2453 | * Wait for swap operations protected by get/put_swap_device() |
| 2454 | * to complete. |
| 2455 | * |
| 2456 | * We need synchronize_rcu() here to protect the accessing to |
| 2457 | * the swap cache data structure. |
| 2458 | */ |
| 2459 | percpu_ref_kill(&p->users); |
| 2460 | synchronize_rcu(); |
| 2461 | wait_for_completion(&p->comp); |
| 2462 | |
| 2463 | flush_work(&p->discard_work); |
| 2464 | |
| 2465 | destroy_swap_extents(p); |
| 2466 | if (p->flags & SWP_CONTINUED) |
| 2467 | free_swap_count_continuations(p); |
| 2468 | |
| 2469 | if (!p->bdev || !blk_queue_nonrot(bdev_get_queue(p->bdev))) |
| 2470 | atomic_dec(&nr_rotate_swap); |
| 2471 | |
| 2472 | mutex_lock(&swapon_mutex); |
| 2473 | spin_lock(&swap_lock); |
| 2474 | spin_lock(&p->lock); |
| 2475 | drain_mmlist(); |
| 2476 | |
| 2477 | /* wait for anyone still in scan_swap_map_slots */ |
| 2478 | p->highest_bit = 0; /* cuts scans short */ |
| 2479 | while (p->flags >= SWP_SCANNING) { |
| 2480 | spin_unlock(&p->lock); |
| 2481 | spin_unlock(&swap_lock); |
| 2482 | schedule_timeout_uninterruptible(1); |
| 2483 | spin_lock(&swap_lock); |
| 2484 | spin_lock(&p->lock); |
| 2485 | } |
| 2486 | |
| 2487 | swap_file = p->swap_file; |
| 2488 | old_block_size = p->old_block_size; |
| 2489 | p->swap_file = NULL; |
| 2490 | p->max = 0; |
| 2491 | swap_map = p->swap_map; |
| 2492 | p->swap_map = NULL; |
| 2493 | cluster_info = p->cluster_info; |
| 2494 | p->cluster_info = NULL; |
| 2495 | frontswap_map = frontswap_map_get(p); |
| 2496 | spin_unlock(&p->lock); |
| 2497 | spin_unlock(&swap_lock); |
| 2498 | arch_swap_invalidate_area(p->type); |
| 2499 | frontswap_invalidate_area(p->type); |
| 2500 | frontswap_map_set(p, NULL); |
| 2501 | mutex_unlock(&swapon_mutex); |
| 2502 | free_percpu(p->percpu_cluster); |
| 2503 | p->percpu_cluster = NULL; |
| 2504 | free_percpu(p->cluster_next_cpu); |
| 2505 | p->cluster_next_cpu = NULL; |
| 2506 | vfree(swap_map); |
| 2507 | kvfree(cluster_info); |
| 2508 | kvfree(frontswap_map); |
| 2509 | /* Destroy swap account information */ |
| 2510 | swap_cgroup_swapoff(p->type); |
| 2511 | exit_swap_address_space(p->type); |
| 2512 | |
| 2513 | inode = mapping->host; |
| 2514 | if (S_ISBLK(inode->i_mode)) { |
| 2515 | struct block_device *bdev = I_BDEV(inode); |
| 2516 | |
| 2517 | set_blocksize(bdev, old_block_size); |
| 2518 | blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 2519 | } |
| 2520 | |
| 2521 | inode_lock(inode); |
| 2522 | inode->i_flags &= ~S_SWAPFILE; |
| 2523 | inode_unlock(inode); |
| 2524 | filp_close(swap_file, NULL); |
| 2525 | |
| 2526 | /* |
| 2527 | * Clear the SWP_USED flag after all resources are freed so that swapon |
| 2528 | * can reuse this swap_info in alloc_swap_info() safely. It is ok to |
| 2529 | * not hold p->lock after we cleared its SWP_WRITEOK. |
| 2530 | */ |
| 2531 | spin_lock(&swap_lock); |
| 2532 | p->flags = 0; |
| 2533 | spin_unlock(&swap_lock); |
| 2534 | |
| 2535 | err = 0; |
| 2536 | atomic_inc(&proc_poll_event); |
| 2537 | wake_up_interruptible(&proc_poll_wait); |
| 2538 | |
| 2539 | out_dput: |
| 2540 | filp_close(victim, NULL); |
| 2541 | out: |
| 2542 | putname(pathname); |
| 2543 | return err; |
| 2544 | } |
| 2545 | |
| 2546 | #ifdef CONFIG_PROC_FS |
| 2547 | static __poll_t swaps_poll(struct file *file, poll_table *wait) |
| 2548 | { |
| 2549 | struct seq_file *seq = file->private_data; |
| 2550 | |
| 2551 | poll_wait(file, &proc_poll_wait, wait); |
| 2552 | |
| 2553 | if (seq->poll_event != atomic_read(&proc_poll_event)) { |
| 2554 | seq->poll_event = atomic_read(&proc_poll_event); |
| 2555 | return EPOLLIN | EPOLLRDNORM | EPOLLERR | EPOLLPRI; |
| 2556 | } |
| 2557 | |
| 2558 | return EPOLLIN | EPOLLRDNORM; |
| 2559 | } |
| 2560 | |
| 2561 | /* iterator */ |
| 2562 | static void *swap_start(struct seq_file *swap, loff_t *pos) |
| 2563 | { |
| 2564 | struct swap_info_struct *si; |
| 2565 | int type; |
| 2566 | loff_t l = *pos; |
| 2567 | |
| 2568 | mutex_lock(&swapon_mutex); |
| 2569 | |
| 2570 | if (!l) |
| 2571 | return SEQ_START_TOKEN; |
| 2572 | |
| 2573 | for (type = 0; (si = swap_type_to_swap_info(type)); type++) { |
| 2574 | if (!(si->flags & SWP_USED) || !si->swap_map) |
| 2575 | continue; |
| 2576 | if (!--l) |
| 2577 | return si; |
| 2578 | } |
| 2579 | |
| 2580 | return NULL; |
| 2581 | } |
| 2582 | |
| 2583 | static void *swap_next(struct seq_file *swap, void *v, loff_t *pos) |
| 2584 | { |
| 2585 | struct swap_info_struct *si = v; |
| 2586 | int type; |
| 2587 | |
| 2588 | if (v == SEQ_START_TOKEN) |
| 2589 | type = 0; |
| 2590 | else |
| 2591 | type = si->type + 1; |
| 2592 | |
| 2593 | ++(*pos); |
| 2594 | for (; (si = swap_type_to_swap_info(type)); type++) { |
| 2595 | if (!(si->flags & SWP_USED) || !si->swap_map) |
| 2596 | continue; |
| 2597 | return si; |
| 2598 | } |
| 2599 | |
| 2600 | return NULL; |
| 2601 | } |
| 2602 | |
| 2603 | static void swap_stop(struct seq_file *swap, void *v) |
| 2604 | { |
| 2605 | mutex_unlock(&swapon_mutex); |
| 2606 | } |
| 2607 | |
| 2608 | static int swap_show(struct seq_file *swap, void *v) |
| 2609 | { |
| 2610 | struct swap_info_struct *si = v; |
| 2611 | struct file *file; |
| 2612 | int len; |
| 2613 | unsigned long bytes, inuse; |
| 2614 | |
| 2615 | if (si == SEQ_START_TOKEN) { |
| 2616 | seq_puts(swap, "Filename\t\t\t\tType\t\tSize\t\tUsed\t\tPriority\n"); |
| 2617 | return 0; |
| 2618 | } |
| 2619 | |
| 2620 | bytes = si->pages << (PAGE_SHIFT - 10); |
| 2621 | inuse = si->inuse_pages << (PAGE_SHIFT - 10); |
| 2622 | |
| 2623 | file = si->swap_file; |
| 2624 | len = seq_file_path(swap, file, " \t\n\\"); |
| 2625 | seq_printf(swap, "%*s%s\t%lu\t%s%lu\t%s%d\n", |
| 2626 | len < 40 ? 40 - len : 1, " ", |
| 2627 | S_ISBLK(file_inode(file)->i_mode) ? |
| 2628 | "partition" : "file\t", |
| 2629 | bytes, bytes < 10000000 ? "\t" : "", |
| 2630 | inuse, inuse < 10000000 ? "\t" : "", |
| 2631 | si->prio); |
| 2632 | return 0; |
| 2633 | } |
| 2634 | |
| 2635 | static const struct seq_operations swaps_op = { |
| 2636 | .start = swap_start, |
| 2637 | .next = swap_next, |
| 2638 | .stop = swap_stop, |
| 2639 | .show = swap_show |
| 2640 | }; |
| 2641 | |
| 2642 | static int swaps_open(struct inode *inode, struct file *file) |
| 2643 | { |
| 2644 | struct seq_file *seq; |
| 2645 | int ret; |
| 2646 | |
| 2647 | ret = seq_open(file, &swaps_op); |
| 2648 | if (ret) |
| 2649 | return ret; |
| 2650 | |
| 2651 | seq = file->private_data; |
| 2652 | seq->poll_event = atomic_read(&proc_poll_event); |
| 2653 | return 0; |
| 2654 | } |
| 2655 | |
| 2656 | static const struct proc_ops swaps_proc_ops = { |
| 2657 | .proc_flags = PROC_ENTRY_PERMANENT, |
| 2658 | .proc_open = swaps_open, |
| 2659 | .proc_read = seq_read, |
| 2660 | .proc_lseek = seq_lseek, |
| 2661 | .proc_release = seq_release, |
| 2662 | .proc_poll = swaps_poll, |
| 2663 | }; |
| 2664 | |
| 2665 | static int __init procswaps_init(void) |
| 2666 | { |
| 2667 | proc_create("swaps", 0, NULL, &swaps_proc_ops); |
| 2668 | return 0; |
| 2669 | } |
| 2670 | __initcall(procswaps_init); |
| 2671 | #endif /* CONFIG_PROC_FS */ |
| 2672 | |
| 2673 | #ifdef MAX_SWAPFILES_CHECK |
| 2674 | static int __init max_swapfiles_check(void) |
| 2675 | { |
| 2676 | MAX_SWAPFILES_CHECK(); |
| 2677 | return 0; |
| 2678 | } |
| 2679 | late_initcall(max_swapfiles_check); |
| 2680 | #endif |
| 2681 | |
| 2682 | static struct swap_info_struct *alloc_swap_info(void) |
| 2683 | { |
| 2684 | struct swap_info_struct *p; |
| 2685 | struct swap_info_struct *defer = NULL; |
| 2686 | unsigned int type; |
| 2687 | int i; |
| 2688 | |
| 2689 | p = kvzalloc(struct_size(p, avail_lists, nr_node_ids), GFP_KERNEL); |
| 2690 | if (!p) |
| 2691 | return ERR_PTR(-ENOMEM); |
| 2692 | |
| 2693 | if (percpu_ref_init(&p->users, swap_users_ref_free, |
| 2694 | PERCPU_REF_INIT_DEAD, GFP_KERNEL)) { |
| 2695 | kvfree(p); |
| 2696 | return ERR_PTR(-ENOMEM); |
| 2697 | } |
| 2698 | |
| 2699 | spin_lock(&swap_lock); |
| 2700 | for (type = 0; type < nr_swapfiles; type++) { |
| 2701 | if (!(swap_info[type]->flags & SWP_USED)) |
| 2702 | break; |
| 2703 | } |
| 2704 | if (type >= MAX_SWAPFILES) { |
| 2705 | spin_unlock(&swap_lock); |
| 2706 | percpu_ref_exit(&p->users); |
| 2707 | kvfree(p); |
| 2708 | return ERR_PTR(-EPERM); |
| 2709 | } |
| 2710 | if (type >= nr_swapfiles) { |
| 2711 | p->type = type; |
| 2712 | /* |
| 2713 | * Publish the swap_info_struct after initializing it. |
| 2714 | * Note that kvzalloc() above zeroes all its fields. |
| 2715 | */ |
| 2716 | smp_store_release(&swap_info[type], p); /* rcu_assign_pointer() */ |
| 2717 | nr_swapfiles++; |
| 2718 | } else { |
| 2719 | defer = p; |
| 2720 | p = swap_info[type]; |
| 2721 | /* |
| 2722 | * Do not memset this entry: a racing procfs swap_next() |
| 2723 | * would be relying on p->type to remain valid. |
| 2724 | */ |
| 2725 | } |
| 2726 | p->swap_extent_root = RB_ROOT; |
| 2727 | plist_node_init(&p->list, 0); |
| 2728 | for_each_node(i) |
| 2729 | plist_node_init(&p->avail_lists[i], 0); |
| 2730 | p->flags = SWP_USED; |
| 2731 | spin_unlock(&swap_lock); |
| 2732 | if (defer) { |
| 2733 | percpu_ref_exit(&defer->users); |
| 2734 | kvfree(defer); |
| 2735 | } |
| 2736 | spin_lock_init(&p->lock); |
| 2737 | spin_lock_init(&p->cont_lock); |
| 2738 | init_completion(&p->comp); |
| 2739 | |
| 2740 | return p; |
| 2741 | } |
| 2742 | |
| 2743 | static int claim_swapfile(struct swap_info_struct *p, struct inode *inode) |
| 2744 | { |
| 2745 | int error; |
| 2746 | |
| 2747 | if (S_ISBLK(inode->i_mode)) { |
| 2748 | p->bdev = blkdev_get_by_dev(inode->i_rdev, |
| 2749 | FMODE_READ | FMODE_WRITE | FMODE_EXCL, p); |
| 2750 | if (IS_ERR(p->bdev)) { |
| 2751 | error = PTR_ERR(p->bdev); |
| 2752 | p->bdev = NULL; |
| 2753 | return error; |
| 2754 | } |
| 2755 | p->old_block_size = block_size(p->bdev); |
| 2756 | error = set_blocksize(p->bdev, PAGE_SIZE); |
| 2757 | if (error < 0) |
| 2758 | return error; |
| 2759 | /* |
| 2760 | * Zoned block devices contain zones that have a sequential |
| 2761 | * write only restriction. Hence zoned block devices are not |
| 2762 | * suitable for swapping. Disallow them here. |
| 2763 | */ |
| 2764 | if (blk_queue_is_zoned(p->bdev->bd_disk->queue)) |
| 2765 | return -EINVAL; |
| 2766 | p->flags |= SWP_BLKDEV; |
| 2767 | } else if (S_ISREG(inode->i_mode)) { |
| 2768 | p->bdev = inode->i_sb->s_bdev; |
| 2769 | } |
| 2770 | |
| 2771 | return 0; |
| 2772 | } |
| 2773 | |
| 2774 | |
| 2775 | /* |
| 2776 | * Find out how many pages are allowed for a single swap device. There |
| 2777 | * are two limiting factors: |
| 2778 | * 1) the number of bits for the swap offset in the swp_entry_t type, and |
| 2779 | * 2) the number of bits in the swap pte, as defined by the different |
| 2780 | * architectures. |
| 2781 | * |
| 2782 | * In order to find the largest possible bit mask, a swap entry with |
| 2783 | * swap type 0 and swap offset ~0UL is created, encoded to a swap pte, |
| 2784 | * decoded to a swp_entry_t again, and finally the swap offset is |
| 2785 | * extracted. |
| 2786 | * |
| 2787 | * This will mask all the bits from the initial ~0UL mask that can't |
| 2788 | * be encoded in either the swp_entry_t or the architecture definition |
| 2789 | * of a swap pte. |
| 2790 | */ |
| 2791 | unsigned long generic_max_swapfile_size(void) |
| 2792 | { |
| 2793 | return swp_offset(pte_to_swp_entry( |
| 2794 | swp_entry_to_pte(swp_entry(0, ~0UL)))) + 1; |
| 2795 | } |
| 2796 | |
| 2797 | /* Can be overridden by an architecture for additional checks. */ |
| 2798 | __weak unsigned long max_swapfile_size(void) |
| 2799 | { |
| 2800 | return generic_max_swapfile_size(); |
| 2801 | } |
| 2802 | |
| 2803 | static unsigned long read_swap_header(struct swap_info_struct *p, |
| 2804 | union swap_header *swap_header, |
| 2805 | struct inode *inode) |
| 2806 | { |
| 2807 | int i; |
| 2808 | unsigned long maxpages; |
| 2809 | unsigned long swapfilepages; |
| 2810 | unsigned long last_page; |
| 2811 | |
| 2812 | if (memcmp("SWAPSPACE2", swap_header->magic.magic, 10)) { |
| 2813 | pr_err("Unable to find swap-space signature\n"); |
| 2814 | return 0; |
| 2815 | } |
| 2816 | |
| 2817 | /* swap partition endianness hack... */ |
| 2818 | if (swab32(swap_header->info.version) == 1) { |
| 2819 | swab32s(&swap_header->info.version); |
| 2820 | swab32s(&swap_header->info.last_page); |
| 2821 | swab32s(&swap_header->info.nr_badpages); |
| 2822 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
| 2823 | return 0; |
| 2824 | for (i = 0; i < swap_header->info.nr_badpages; i++) |
| 2825 | swab32s(&swap_header->info.badpages[i]); |
| 2826 | } |
| 2827 | /* Check the swap header's sub-version */ |
| 2828 | if (swap_header->info.version != 1) { |
| 2829 | pr_warn("Unable to handle swap header version %d\n", |
| 2830 | swap_header->info.version); |
| 2831 | return 0; |
| 2832 | } |
| 2833 | |
| 2834 | p->lowest_bit = 1; |
| 2835 | p->cluster_next = 1; |
| 2836 | p->cluster_nr = 0; |
| 2837 | |
| 2838 | maxpages = max_swapfile_size(); |
| 2839 | last_page = swap_header->info.last_page; |
| 2840 | if (!last_page) { |
| 2841 | pr_warn("Empty swap-file\n"); |
| 2842 | return 0; |
| 2843 | } |
| 2844 | if (last_page > maxpages) { |
| 2845 | pr_warn("Truncating oversized swap area, only using %luk out of %luk\n", |
| 2846 | maxpages << (PAGE_SHIFT - 10), |
| 2847 | last_page << (PAGE_SHIFT - 10)); |
| 2848 | } |
| 2849 | if (maxpages > last_page) { |
| 2850 | maxpages = last_page + 1; |
| 2851 | /* p->max is an unsigned int: don't overflow it */ |
| 2852 | if ((unsigned int)maxpages == 0) |
| 2853 | maxpages = UINT_MAX; |
| 2854 | } |
| 2855 | p->highest_bit = maxpages - 1; |
| 2856 | |
| 2857 | if (!maxpages) |
| 2858 | return 0; |
| 2859 | swapfilepages = i_size_read(inode) >> PAGE_SHIFT; |
| 2860 | if (swapfilepages && maxpages > swapfilepages) { |
| 2861 | pr_warn("Swap area shorter than signature indicates\n"); |
| 2862 | return 0; |
| 2863 | } |
| 2864 | if (swap_header->info.nr_badpages && S_ISREG(inode->i_mode)) |
| 2865 | return 0; |
| 2866 | if (swap_header->info.nr_badpages > MAX_SWAP_BADPAGES) |
| 2867 | return 0; |
| 2868 | |
| 2869 | return maxpages; |
| 2870 | } |
| 2871 | |
| 2872 | #define SWAP_CLUSTER_INFO_COLS \ |
| 2873 | DIV_ROUND_UP(L1_CACHE_BYTES, sizeof(struct swap_cluster_info)) |
| 2874 | #define SWAP_CLUSTER_SPACE_COLS \ |
| 2875 | DIV_ROUND_UP(SWAP_ADDRESS_SPACE_PAGES, SWAPFILE_CLUSTER) |
| 2876 | #define SWAP_CLUSTER_COLS \ |
| 2877 | max_t(unsigned int, SWAP_CLUSTER_INFO_COLS, SWAP_CLUSTER_SPACE_COLS) |
| 2878 | |
| 2879 | static int setup_swap_map_and_extents(struct swap_info_struct *p, |
| 2880 | union swap_header *swap_header, |
| 2881 | unsigned char *swap_map, |
| 2882 | struct swap_cluster_info *cluster_info, |
| 2883 | unsigned long maxpages, |
| 2884 | sector_t *span) |
| 2885 | { |
| 2886 | unsigned int j, k; |
| 2887 | unsigned int nr_good_pages; |
| 2888 | int nr_extents; |
| 2889 | unsigned long nr_clusters = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
| 2890 | unsigned long col = p->cluster_next / SWAPFILE_CLUSTER % SWAP_CLUSTER_COLS; |
| 2891 | unsigned long i, idx; |
| 2892 | |
| 2893 | nr_good_pages = maxpages - 1; /* omit header page */ |
| 2894 | |
| 2895 | cluster_list_init(&p->free_clusters); |
| 2896 | cluster_list_init(&p->discard_clusters); |
| 2897 | |
| 2898 | for (i = 0; i < swap_header->info.nr_badpages; i++) { |
| 2899 | unsigned int page_nr = swap_header->info.badpages[i]; |
| 2900 | if (page_nr == 0 || page_nr > swap_header->info.last_page) |
| 2901 | return -EINVAL; |
| 2902 | if (page_nr < maxpages) { |
| 2903 | swap_map[page_nr] = SWAP_MAP_BAD; |
| 2904 | nr_good_pages--; |
| 2905 | /* |
| 2906 | * Haven't marked the cluster free yet, no list |
| 2907 | * operation involved |
| 2908 | */ |
| 2909 | inc_cluster_info_page(p, cluster_info, page_nr); |
| 2910 | } |
| 2911 | } |
| 2912 | |
| 2913 | /* Haven't marked the cluster free yet, no list operation involved */ |
| 2914 | for (i = maxpages; i < round_up(maxpages, SWAPFILE_CLUSTER); i++) |
| 2915 | inc_cluster_info_page(p, cluster_info, i); |
| 2916 | |
| 2917 | if (nr_good_pages) { |
| 2918 | swap_map[0] = SWAP_MAP_BAD; |
| 2919 | /* |
| 2920 | * Not mark the cluster free yet, no list |
| 2921 | * operation involved |
| 2922 | */ |
| 2923 | inc_cluster_info_page(p, cluster_info, 0); |
| 2924 | p->max = maxpages; |
| 2925 | p->pages = nr_good_pages; |
| 2926 | nr_extents = setup_swap_extents(p, span); |
| 2927 | if (nr_extents < 0) |
| 2928 | return nr_extents; |
| 2929 | nr_good_pages = p->pages; |
| 2930 | } |
| 2931 | if (!nr_good_pages) { |
| 2932 | pr_warn("Empty swap-file\n"); |
| 2933 | return -EINVAL; |
| 2934 | } |
| 2935 | |
| 2936 | if (!cluster_info) |
| 2937 | return nr_extents; |
| 2938 | |
| 2939 | |
| 2940 | /* |
| 2941 | * Reduce false cache line sharing between cluster_info and |
| 2942 | * sharing same address space. |
| 2943 | */ |
| 2944 | for (k = 0; k < SWAP_CLUSTER_COLS; k++) { |
| 2945 | j = (k + col) % SWAP_CLUSTER_COLS; |
| 2946 | for (i = 0; i < DIV_ROUND_UP(nr_clusters, SWAP_CLUSTER_COLS); i++) { |
| 2947 | idx = i * SWAP_CLUSTER_COLS + j; |
| 2948 | if (idx >= nr_clusters) |
| 2949 | continue; |
| 2950 | if (cluster_count(&cluster_info[idx])) |
| 2951 | continue; |
| 2952 | cluster_set_flag(&cluster_info[idx], CLUSTER_FLAG_FREE); |
| 2953 | cluster_list_add_tail(&p->free_clusters, cluster_info, |
| 2954 | idx); |
| 2955 | } |
| 2956 | } |
| 2957 | return nr_extents; |
| 2958 | } |
| 2959 | |
| 2960 | /* |
| 2961 | * Helper to sys_swapon determining if a given swap |
| 2962 | * backing device queue supports DISCARD operations. |
| 2963 | */ |
| 2964 | static bool swap_discardable(struct swap_info_struct *si) |
| 2965 | { |
| 2966 | struct request_queue *q = bdev_get_queue(si->bdev); |
| 2967 | |
| 2968 | if (!blk_queue_discard(q)) |
| 2969 | return false; |
| 2970 | |
| 2971 | return true; |
| 2972 | } |
| 2973 | |
| 2974 | SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags) |
| 2975 | { |
| 2976 | struct swap_info_struct *p; |
| 2977 | struct filename *name; |
| 2978 | struct file *swap_file = NULL; |
| 2979 | struct address_space *mapping; |
| 2980 | struct dentry *dentry; |
| 2981 | int prio; |
| 2982 | int error; |
| 2983 | union swap_header *swap_header; |
| 2984 | int nr_extents; |
| 2985 | sector_t span; |
| 2986 | unsigned long maxpages; |
| 2987 | unsigned char *swap_map = NULL; |
| 2988 | struct swap_cluster_info *cluster_info = NULL; |
| 2989 | unsigned long *frontswap_map = NULL; |
| 2990 | struct page *page = NULL; |
| 2991 | struct inode *inode = NULL; |
| 2992 | bool inced_nr_rotate_swap = false; |
| 2993 | |
| 2994 | if (swap_flags & ~SWAP_FLAGS_VALID) |
| 2995 | return -EINVAL; |
| 2996 | |
| 2997 | if (!capable(CAP_SYS_ADMIN)) |
| 2998 | return -EPERM; |
| 2999 | |
| 3000 | if (!swap_avail_heads) |
| 3001 | return -ENOMEM; |
| 3002 | |
| 3003 | p = alloc_swap_info(); |
| 3004 | if (IS_ERR(p)) |
| 3005 | return PTR_ERR(p); |
| 3006 | |
| 3007 | INIT_WORK(&p->discard_work, swap_discard_work); |
| 3008 | |
| 3009 | name = getname(specialfile); |
| 3010 | if (IS_ERR(name)) { |
| 3011 | error = PTR_ERR(name); |
| 3012 | name = NULL; |
| 3013 | goto bad_swap; |
| 3014 | } |
| 3015 | swap_file = file_open_name(name, O_RDWR|O_LARGEFILE, 0); |
| 3016 | if (IS_ERR(swap_file)) { |
| 3017 | error = PTR_ERR(swap_file); |
| 3018 | swap_file = NULL; |
| 3019 | goto bad_swap; |
| 3020 | } |
| 3021 | |
| 3022 | p->swap_file = swap_file; |
| 3023 | mapping = swap_file->f_mapping; |
| 3024 | dentry = swap_file->f_path.dentry; |
| 3025 | inode = mapping->host; |
| 3026 | |
| 3027 | error = claim_swapfile(p, inode); |
| 3028 | if (unlikely(error)) |
| 3029 | goto bad_swap; |
| 3030 | |
| 3031 | inode_lock(inode); |
| 3032 | if (d_unlinked(dentry) || cant_mount(dentry)) { |
| 3033 | error = -ENOENT; |
| 3034 | goto bad_swap_unlock_inode; |
| 3035 | } |
| 3036 | if (IS_SWAPFILE(inode)) { |
| 3037 | error = -EBUSY; |
| 3038 | goto bad_swap_unlock_inode; |
| 3039 | } |
| 3040 | |
| 3041 | /* |
| 3042 | * Read the swap header. |
| 3043 | */ |
| 3044 | if (!mapping->a_ops->readpage) { |
| 3045 | error = -EINVAL; |
| 3046 | goto bad_swap_unlock_inode; |
| 3047 | } |
| 3048 | page = read_mapping_page(mapping, 0, swap_file); |
| 3049 | if (IS_ERR(page)) { |
| 3050 | error = PTR_ERR(page); |
| 3051 | goto bad_swap_unlock_inode; |
| 3052 | } |
| 3053 | swap_header = kmap(page); |
| 3054 | |
| 3055 | maxpages = read_swap_header(p, swap_header, inode); |
| 3056 | if (unlikely(!maxpages)) { |
| 3057 | error = -EINVAL; |
| 3058 | goto bad_swap_unlock_inode; |
| 3059 | } |
| 3060 | |
| 3061 | /* OK, set up the swap map and apply the bad block list */ |
| 3062 | swap_map = vzalloc(maxpages); |
| 3063 | if (!swap_map) { |
| 3064 | error = -ENOMEM; |
| 3065 | goto bad_swap_unlock_inode; |
| 3066 | } |
| 3067 | |
| 3068 | if (p->bdev && blk_queue_stable_writes(p->bdev->bd_disk->queue)) |
| 3069 | p->flags |= SWP_STABLE_WRITES; |
| 3070 | |
| 3071 | if (p->bdev && p->bdev->bd_disk->fops->rw_page) |
| 3072 | p->flags |= SWP_SYNCHRONOUS_IO; |
| 3073 | |
| 3074 | if (p->bdev && blk_queue_nonrot(bdev_get_queue(p->bdev))) { |
| 3075 | int cpu; |
| 3076 | unsigned long ci, nr_cluster; |
| 3077 | |
| 3078 | p->flags |= SWP_SOLIDSTATE; |
| 3079 | p->cluster_next_cpu = alloc_percpu(unsigned int); |
| 3080 | if (!p->cluster_next_cpu) { |
| 3081 | error = -ENOMEM; |
| 3082 | goto bad_swap_unlock_inode; |
| 3083 | } |
| 3084 | /* |
| 3085 | * select a random position to start with to help wear leveling |
| 3086 | * SSD |
| 3087 | */ |
| 3088 | for_each_possible_cpu(cpu) { |
| 3089 | per_cpu(*p->cluster_next_cpu, cpu) = |
| 3090 | 1 + prandom_u32_max(p->highest_bit); |
| 3091 | } |
| 3092 | nr_cluster = DIV_ROUND_UP(maxpages, SWAPFILE_CLUSTER); |
| 3093 | |
| 3094 | cluster_info = kvcalloc(nr_cluster, sizeof(*cluster_info), |
| 3095 | GFP_KERNEL); |
| 3096 | if (!cluster_info) { |
| 3097 | error = -ENOMEM; |
| 3098 | goto bad_swap_unlock_inode; |
| 3099 | } |
| 3100 | |
| 3101 | for (ci = 0; ci < nr_cluster; ci++) |
| 3102 | spin_lock_init(&((cluster_info + ci)->lock)); |
| 3103 | |
| 3104 | p->percpu_cluster = alloc_percpu(struct percpu_cluster); |
| 3105 | if (!p->percpu_cluster) { |
| 3106 | error = -ENOMEM; |
| 3107 | goto bad_swap_unlock_inode; |
| 3108 | } |
| 3109 | for_each_possible_cpu(cpu) { |
| 3110 | struct percpu_cluster *cluster; |
| 3111 | cluster = per_cpu_ptr(p->percpu_cluster, cpu); |
| 3112 | cluster_set_null(&cluster->index); |
| 3113 | } |
| 3114 | } else { |
| 3115 | atomic_inc(&nr_rotate_swap); |
| 3116 | inced_nr_rotate_swap = true; |
| 3117 | } |
| 3118 | |
| 3119 | error = swap_cgroup_swapon(p->type, maxpages); |
| 3120 | if (error) |
| 3121 | goto bad_swap_unlock_inode; |
| 3122 | |
| 3123 | nr_extents = setup_swap_map_and_extents(p, swap_header, swap_map, |
| 3124 | cluster_info, maxpages, &span); |
| 3125 | if (unlikely(nr_extents < 0)) { |
| 3126 | error = nr_extents; |
| 3127 | goto bad_swap_unlock_inode; |
| 3128 | } |
| 3129 | /* frontswap enabled? set up bit-per-page map for frontswap */ |
| 3130 | if (IS_ENABLED(CONFIG_FRONTSWAP)) |
| 3131 | frontswap_map = kvcalloc(BITS_TO_LONGS(maxpages), |
| 3132 | sizeof(long), |
| 3133 | GFP_KERNEL); |
| 3134 | |
| 3135 | if (p->bdev && (swap_flags & SWAP_FLAG_DISCARD) && swap_discardable(p)) { |
| 3136 | /* |
| 3137 | * When discard is enabled for swap with no particular |
| 3138 | * policy flagged, we set all swap discard flags here in |
| 3139 | * order to sustain backward compatibility with older |
| 3140 | * swapon(8) releases. |
| 3141 | */ |
| 3142 | p->flags |= (SWP_DISCARDABLE | SWP_AREA_DISCARD | |
| 3143 | SWP_PAGE_DISCARD); |
| 3144 | |
| 3145 | /* |
| 3146 | * By flagging sys_swapon, a sysadmin can tell us to |
| 3147 | * either do single-time area discards only, or to just |
| 3148 | * perform discards for released swap page-clusters. |
| 3149 | * Now it's time to adjust the p->flags accordingly. |
| 3150 | */ |
| 3151 | if (swap_flags & SWAP_FLAG_DISCARD_ONCE) |
| 3152 | p->flags &= ~SWP_PAGE_DISCARD; |
| 3153 | else if (swap_flags & SWAP_FLAG_DISCARD_PAGES) |
| 3154 | p->flags &= ~SWP_AREA_DISCARD; |
| 3155 | |
| 3156 | /* issue a swapon-time discard if it's still required */ |
| 3157 | if (p->flags & SWP_AREA_DISCARD) { |
| 3158 | int err = discard_swap(p); |
| 3159 | if (unlikely(err)) |
| 3160 | pr_err("swapon: discard_swap(%p): %d\n", |
| 3161 | p, err); |
| 3162 | } |
| 3163 | } |
| 3164 | |
| 3165 | error = init_swap_address_space(p->type, maxpages); |
| 3166 | if (error) |
| 3167 | goto bad_swap_unlock_inode; |
| 3168 | |
| 3169 | /* |
| 3170 | * Flush any pending IO and dirty mappings before we start using this |
| 3171 | * swap device. |
| 3172 | */ |
| 3173 | inode->i_flags |= S_SWAPFILE; |
| 3174 | error = inode_drain_writes(inode); |
| 3175 | if (error) { |
| 3176 | inode->i_flags &= ~S_SWAPFILE; |
| 3177 | goto free_swap_address_space; |
| 3178 | } |
| 3179 | |
| 3180 | mutex_lock(&swapon_mutex); |
| 3181 | prio = -1; |
| 3182 | if (swap_flags & SWAP_FLAG_PREFER) |
| 3183 | prio = |
| 3184 | (swap_flags & SWAP_FLAG_PRIO_MASK) >> SWAP_FLAG_PRIO_SHIFT; |
| 3185 | enable_swap_info(p, prio, swap_map, cluster_info, frontswap_map); |
| 3186 | |
| 3187 | pr_info("Adding %uk swap on %s. Priority:%d extents:%d across:%lluk %s%s%s%s%s\n", |
| 3188 | p->pages<<(PAGE_SHIFT-10), name->name, p->prio, |
| 3189 | nr_extents, (unsigned long long)span<<(PAGE_SHIFT-10), |
| 3190 | (p->flags & SWP_SOLIDSTATE) ? "SS" : "", |
| 3191 | (p->flags & SWP_DISCARDABLE) ? "D" : "", |
| 3192 | (p->flags & SWP_AREA_DISCARD) ? "s" : "", |
| 3193 | (p->flags & SWP_PAGE_DISCARD) ? "c" : "", |
| 3194 | (frontswap_map) ? "FS" : ""); |
| 3195 | |
| 3196 | mutex_unlock(&swapon_mutex); |
| 3197 | atomic_inc(&proc_poll_event); |
| 3198 | wake_up_interruptible(&proc_poll_wait); |
| 3199 | |
| 3200 | error = 0; |
| 3201 | goto out; |
| 3202 | free_swap_address_space: |
| 3203 | exit_swap_address_space(p->type); |
| 3204 | bad_swap_unlock_inode: |
| 3205 | inode_unlock(inode); |
| 3206 | bad_swap: |
| 3207 | free_percpu(p->percpu_cluster); |
| 3208 | p->percpu_cluster = NULL; |
| 3209 | free_percpu(p->cluster_next_cpu); |
| 3210 | p->cluster_next_cpu = NULL; |
| 3211 | if (inode && S_ISBLK(inode->i_mode) && p->bdev) { |
| 3212 | set_blocksize(p->bdev, p->old_block_size); |
| 3213 | blkdev_put(p->bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL); |
| 3214 | } |
| 3215 | inode = NULL; |
| 3216 | destroy_swap_extents(p); |
| 3217 | swap_cgroup_swapoff(p->type); |
| 3218 | spin_lock(&swap_lock); |
| 3219 | p->swap_file = NULL; |
| 3220 | p->flags = 0; |
| 3221 | spin_unlock(&swap_lock); |
| 3222 | vfree(swap_map); |
| 3223 | kvfree(cluster_info); |
| 3224 | kvfree(frontswap_map); |
| 3225 | if (inced_nr_rotate_swap) |
| 3226 | atomic_dec(&nr_rotate_swap); |
| 3227 | if (swap_file) |
| 3228 | filp_close(swap_file, NULL); |
| 3229 | out: |
| 3230 | if (page && !IS_ERR(page)) { |
| 3231 | kunmap(page); |
| 3232 | put_page(page); |
| 3233 | } |
| 3234 | if (name) |
| 3235 | putname(name); |
| 3236 | if (inode) |
| 3237 | inode_unlock(inode); |
| 3238 | if (!error) |
| 3239 | enable_swap_slots_cache(); |
| 3240 | return error; |
| 3241 | } |
| 3242 | |
| 3243 | void si_swapinfo(struct sysinfo *val) |
| 3244 | { |
| 3245 | unsigned int type; |
| 3246 | unsigned long nr_to_be_unused = 0; |
| 3247 | |
| 3248 | spin_lock(&swap_lock); |
| 3249 | for (type = 0; type < nr_swapfiles; type++) { |
| 3250 | struct swap_info_struct *si = swap_info[type]; |
| 3251 | |
| 3252 | if ((si->flags & SWP_USED) && !(si->flags & SWP_WRITEOK)) |
| 3253 | nr_to_be_unused += si->inuse_pages; |
| 3254 | } |
| 3255 | val->freeswap = atomic_long_read(&nr_swap_pages) + nr_to_be_unused; |
| 3256 | val->totalswap = total_swap_pages + nr_to_be_unused; |
| 3257 | spin_unlock(&swap_lock); |
| 3258 | } |
| 3259 | |
| 3260 | /* |
| 3261 | * Verify that a swap entry is valid and increment its swap map count. |
| 3262 | * |
| 3263 | * Returns error code in following case. |
| 3264 | * - success -> 0 |
| 3265 | * - swp_entry is invalid -> EINVAL |
| 3266 | * - swp_entry is migration entry -> EINVAL |
| 3267 | * - swap-cache reference is requested but there is already one. -> EEXIST |
| 3268 | * - swap-cache reference is requested but the entry is not used. -> ENOENT |
| 3269 | * - swap-mapped reference requested but needs continued swap count. -> ENOMEM |
| 3270 | */ |
| 3271 | static int __swap_duplicate(swp_entry_t entry, unsigned char usage) |
| 3272 | { |
| 3273 | struct swap_info_struct *p; |
| 3274 | struct swap_cluster_info *ci; |
| 3275 | unsigned long offset; |
| 3276 | unsigned char count; |
| 3277 | unsigned char has_cache; |
| 3278 | int err; |
| 3279 | |
| 3280 | p = get_swap_device(entry); |
| 3281 | if (!p) |
| 3282 | return -EINVAL; |
| 3283 | |
| 3284 | offset = swp_offset(entry); |
| 3285 | ci = lock_cluster_or_swap_info(p, offset); |
| 3286 | |
| 3287 | count = p->swap_map[offset]; |
| 3288 | |
| 3289 | /* |
| 3290 | * swapin_readahead() doesn't check if a swap entry is valid, so the |
| 3291 | * swap entry could be SWAP_MAP_BAD. Check here with lock held. |
| 3292 | */ |
| 3293 | if (unlikely(swap_count(count) == SWAP_MAP_BAD)) { |
| 3294 | err = -ENOENT; |
| 3295 | goto unlock_out; |
| 3296 | } |
| 3297 | |
| 3298 | has_cache = count & SWAP_HAS_CACHE; |
| 3299 | count &= ~SWAP_HAS_CACHE; |
| 3300 | err = 0; |
| 3301 | |
| 3302 | if (usage == SWAP_HAS_CACHE) { |
| 3303 | |
| 3304 | /* set SWAP_HAS_CACHE if there is no cache and entry is used */ |
| 3305 | if (!has_cache && count) |
| 3306 | has_cache = SWAP_HAS_CACHE; |
| 3307 | else if (has_cache) /* someone else added cache */ |
| 3308 | err = -EEXIST; |
| 3309 | else /* no users remaining */ |
| 3310 | err = -ENOENT; |
| 3311 | |
| 3312 | } else if (count || has_cache) { |
| 3313 | |
| 3314 | if ((count & ~COUNT_CONTINUED) < SWAP_MAP_MAX) |
| 3315 | count += usage; |
| 3316 | else if ((count & ~COUNT_CONTINUED) > SWAP_MAP_MAX) |
| 3317 | err = -EINVAL; |
| 3318 | else if (swap_count_continued(p, offset, count)) |
| 3319 | count = COUNT_CONTINUED; |
| 3320 | else |
| 3321 | err = -ENOMEM; |
| 3322 | } else |
| 3323 | err = -ENOENT; /* unused swap entry */ |
| 3324 | |
| 3325 | WRITE_ONCE(p->swap_map[offset], count | has_cache); |
| 3326 | |
| 3327 | unlock_out: |
| 3328 | unlock_cluster_or_swap_info(p, ci); |
| 3329 | if (p) |
| 3330 | put_swap_device(p); |
| 3331 | return err; |
| 3332 | } |
| 3333 | |
| 3334 | /* |
| 3335 | * Help swapoff by noting that swap entry belongs to shmem/tmpfs |
| 3336 | * (in which case its reference count is never incremented). |
| 3337 | */ |
| 3338 | void swap_shmem_alloc(swp_entry_t entry) |
| 3339 | { |
| 3340 | __swap_duplicate(entry, SWAP_MAP_SHMEM); |
| 3341 | } |
| 3342 | |
| 3343 | /* |
| 3344 | * Increase reference count of swap entry by 1. |
| 3345 | * Returns 0 for success, or -ENOMEM if a swap_count_continuation is required |
| 3346 | * but could not be atomically allocated. Returns 0, just as if it succeeded, |
| 3347 | * if __swap_duplicate() fails for another reason (-EINVAL or -ENOENT), which |
| 3348 | * might occur if a page table entry has got corrupted. |
| 3349 | */ |
| 3350 | int swap_duplicate(swp_entry_t entry) |
| 3351 | { |
| 3352 | int err = 0; |
| 3353 | |
| 3354 | while (!err && __swap_duplicate(entry, 1) == -ENOMEM) |
| 3355 | err = add_swap_count_continuation(entry, GFP_ATOMIC); |
| 3356 | return err; |
| 3357 | } |
| 3358 | |
| 3359 | /* |
| 3360 | * @entry: swap entry for which we allocate swap cache. |
| 3361 | * |
| 3362 | * Called when allocating swap cache for existing swap entry, |
| 3363 | * This can return error codes. Returns 0 at success. |
| 3364 | * -EEXIST means there is a swap cache. |
| 3365 | * Note: return code is different from swap_duplicate(). |
| 3366 | */ |
| 3367 | int swapcache_prepare(swp_entry_t entry) |
| 3368 | { |
| 3369 | return __swap_duplicate(entry, SWAP_HAS_CACHE); |
| 3370 | } |
| 3371 | |
| 3372 | struct swap_info_struct *swp_swap_info(swp_entry_t entry) |
| 3373 | { |
| 3374 | return swap_type_to_swap_info(swp_type(entry)); |
| 3375 | } |
| 3376 | |
| 3377 | struct swap_info_struct *page_swap_info(struct page *page) |
| 3378 | { |
| 3379 | swp_entry_t entry = { .val = page_private(page) }; |
| 3380 | return swp_swap_info(entry); |
| 3381 | } |
| 3382 | |
| 3383 | /* |
| 3384 | * out-of-line methods to avoid include hell. |
| 3385 | */ |
| 3386 | struct address_space *swapcache_mapping(struct folio *folio) |
| 3387 | { |
| 3388 | return page_swap_info(&folio->page)->swap_file->f_mapping; |
| 3389 | } |
| 3390 | EXPORT_SYMBOL_GPL(swapcache_mapping); |
| 3391 | |
| 3392 | pgoff_t __page_file_index(struct page *page) |
| 3393 | { |
| 3394 | swp_entry_t swap = { .val = page_private(page) }; |
| 3395 | return swp_offset(swap); |
| 3396 | } |
| 3397 | EXPORT_SYMBOL_GPL(__page_file_index); |
| 3398 | |
| 3399 | /* |
| 3400 | * add_swap_count_continuation - called when a swap count is duplicated |
| 3401 | * beyond SWAP_MAP_MAX, it allocates a new page and links that to the entry's |
| 3402 | * page of the original vmalloc'ed swap_map, to hold the continuation count |
| 3403 | * (for that entry and for its neighbouring PAGE_SIZE swap entries). Called |
| 3404 | * again when count is duplicated beyond SWAP_MAP_MAX * SWAP_CONT_MAX, etc. |
| 3405 | * |
| 3406 | * These continuation pages are seldom referenced: the common paths all work |
| 3407 | * on the original swap_map, only referring to a continuation page when the |
| 3408 | * low "digit" of a count is incremented or decremented through SWAP_MAP_MAX. |
| 3409 | * |
| 3410 | * add_swap_count_continuation(, GFP_ATOMIC) can be called while holding |
| 3411 | * page table locks; if it fails, add_swap_count_continuation(, GFP_KERNEL) |
| 3412 | * can be called after dropping locks. |
| 3413 | */ |
| 3414 | int add_swap_count_continuation(swp_entry_t entry, gfp_t gfp_mask) |
| 3415 | { |
| 3416 | struct swap_info_struct *si; |
| 3417 | struct swap_cluster_info *ci; |
| 3418 | struct page *head; |
| 3419 | struct page *page; |
| 3420 | struct page *list_page; |
| 3421 | pgoff_t offset; |
| 3422 | unsigned char count; |
| 3423 | int ret = 0; |
| 3424 | |
| 3425 | /* |
| 3426 | * When debugging, it's easier to use __GFP_ZERO here; but it's better |
| 3427 | * for latency not to zero a page while GFP_ATOMIC and holding locks. |
| 3428 | */ |
| 3429 | page = alloc_page(gfp_mask | __GFP_HIGHMEM); |
| 3430 | |
| 3431 | si = get_swap_device(entry); |
| 3432 | if (!si) { |
| 3433 | /* |
| 3434 | * An acceptable race has occurred since the failing |
| 3435 | * __swap_duplicate(): the swap device may be swapoff |
| 3436 | */ |
| 3437 | goto outer; |
| 3438 | } |
| 3439 | spin_lock(&si->lock); |
| 3440 | |
| 3441 | offset = swp_offset(entry); |
| 3442 | |
| 3443 | ci = lock_cluster(si, offset); |
| 3444 | |
| 3445 | count = swap_count(si->swap_map[offset]); |
| 3446 | |
| 3447 | if ((count & ~COUNT_CONTINUED) != SWAP_MAP_MAX) { |
| 3448 | /* |
| 3449 | * The higher the swap count, the more likely it is that tasks |
| 3450 | * will race to add swap count continuation: we need to avoid |
| 3451 | * over-provisioning. |
| 3452 | */ |
| 3453 | goto out; |
| 3454 | } |
| 3455 | |
| 3456 | if (!page) { |
| 3457 | ret = -ENOMEM; |
| 3458 | goto out; |
| 3459 | } |
| 3460 | |
| 3461 | /* |
| 3462 | * We are fortunate that although vmalloc_to_page uses pte_offset_map, |
| 3463 | * no architecture is using highmem pages for kernel page tables: so it |
| 3464 | * will not corrupt the GFP_ATOMIC caller's atomic page table kmaps. |
| 3465 | */ |
| 3466 | head = vmalloc_to_page(si->swap_map + offset); |
| 3467 | offset &= ~PAGE_MASK; |
| 3468 | |
| 3469 | spin_lock(&si->cont_lock); |
| 3470 | /* |
| 3471 | * Page allocation does not initialize the page's lru field, |
| 3472 | * but it does always reset its private field. |
| 3473 | */ |
| 3474 | if (!page_private(head)) { |
| 3475 | BUG_ON(count & COUNT_CONTINUED); |
| 3476 | INIT_LIST_HEAD(&head->lru); |
| 3477 | set_page_private(head, SWP_CONTINUED); |
| 3478 | si->flags |= SWP_CONTINUED; |
| 3479 | } |
| 3480 | |
| 3481 | list_for_each_entry(list_page, &head->lru, lru) { |
| 3482 | unsigned char *map; |
| 3483 | |
| 3484 | /* |
| 3485 | * If the previous map said no continuation, but we've found |
| 3486 | * a continuation page, free our allocation and use this one. |
| 3487 | */ |
| 3488 | if (!(count & COUNT_CONTINUED)) |
| 3489 | goto out_unlock_cont; |
| 3490 | |
| 3491 | map = kmap_atomic(list_page) + offset; |
| 3492 | count = *map; |
| 3493 | kunmap_atomic(map); |
| 3494 | |
| 3495 | /* |
| 3496 | * If this continuation count now has some space in it, |
| 3497 | * free our allocation and use this one. |
| 3498 | */ |
| 3499 | if ((count & ~COUNT_CONTINUED) != SWAP_CONT_MAX) |
| 3500 | goto out_unlock_cont; |
| 3501 | } |
| 3502 | |
| 3503 | list_add_tail(&page->lru, &head->lru); |
| 3504 | page = NULL; /* now it's attached, don't free it */ |
| 3505 | out_unlock_cont: |
| 3506 | spin_unlock(&si->cont_lock); |
| 3507 | out: |
| 3508 | unlock_cluster(ci); |
| 3509 | spin_unlock(&si->lock); |
| 3510 | put_swap_device(si); |
| 3511 | outer: |
| 3512 | if (page) |
| 3513 | __free_page(page); |
| 3514 | return ret; |
| 3515 | } |
| 3516 | |
| 3517 | /* |
| 3518 | * swap_count_continued - when the original swap_map count is incremented |
| 3519 | * from SWAP_MAP_MAX, check if there is already a continuation page to carry |
| 3520 | * into, carry if so, or else fail until a new continuation page is allocated; |
| 3521 | * when the original swap_map count is decremented from 0 with continuation, |
| 3522 | * borrow from the continuation and report whether it still holds more. |
| 3523 | * Called while __swap_duplicate() or swap_entry_free() holds swap or cluster |
| 3524 | * lock. |
| 3525 | */ |
| 3526 | static bool swap_count_continued(struct swap_info_struct *si, |
| 3527 | pgoff_t offset, unsigned char count) |
| 3528 | { |
| 3529 | struct page *head; |
| 3530 | struct page *page; |
| 3531 | unsigned char *map; |
| 3532 | bool ret; |
| 3533 | |
| 3534 | head = vmalloc_to_page(si->swap_map + offset); |
| 3535 | if (page_private(head) != SWP_CONTINUED) { |
| 3536 | BUG_ON(count & COUNT_CONTINUED); |
| 3537 | return false; /* need to add count continuation */ |
| 3538 | } |
| 3539 | |
| 3540 | spin_lock(&si->cont_lock); |
| 3541 | offset &= ~PAGE_MASK; |
| 3542 | page = list_next_entry(head, lru); |
| 3543 | map = kmap_atomic(page) + offset; |
| 3544 | |
| 3545 | if (count == SWAP_MAP_MAX) /* initial increment from swap_map */ |
| 3546 | goto init_map; /* jump over SWAP_CONT_MAX checks */ |
| 3547 | |
| 3548 | if (count == (SWAP_MAP_MAX | COUNT_CONTINUED)) { /* incrementing */ |
| 3549 | /* |
| 3550 | * Think of how you add 1 to 999 |
| 3551 | */ |
| 3552 | while (*map == (SWAP_CONT_MAX | COUNT_CONTINUED)) { |
| 3553 | kunmap_atomic(map); |
| 3554 | page = list_next_entry(page, lru); |
| 3555 | BUG_ON(page == head); |
| 3556 | map = kmap_atomic(page) + offset; |
| 3557 | } |
| 3558 | if (*map == SWAP_CONT_MAX) { |
| 3559 | kunmap_atomic(map); |
| 3560 | page = list_next_entry(page, lru); |
| 3561 | if (page == head) { |
| 3562 | ret = false; /* add count continuation */ |
| 3563 | goto out; |
| 3564 | } |
| 3565 | map = kmap_atomic(page) + offset; |
| 3566 | init_map: *map = 0; /* we didn't zero the page */ |
| 3567 | } |
| 3568 | *map += 1; |
| 3569 | kunmap_atomic(map); |
| 3570 | while ((page = list_prev_entry(page, lru)) != head) { |
| 3571 | map = kmap_atomic(page) + offset; |
| 3572 | *map = COUNT_CONTINUED; |
| 3573 | kunmap_atomic(map); |
| 3574 | } |
| 3575 | ret = true; /* incremented */ |
| 3576 | |
| 3577 | } else { /* decrementing */ |
| 3578 | /* |
| 3579 | * Think of how you subtract 1 from 1000 |
| 3580 | */ |
| 3581 | BUG_ON(count != COUNT_CONTINUED); |
| 3582 | while (*map == COUNT_CONTINUED) { |
| 3583 | kunmap_atomic(map); |
| 3584 | page = list_next_entry(page, lru); |
| 3585 | BUG_ON(page == head); |
| 3586 | map = kmap_atomic(page) + offset; |
| 3587 | } |
| 3588 | BUG_ON(*map == 0); |
| 3589 | *map -= 1; |
| 3590 | if (*map == 0) |
| 3591 | count = 0; |
| 3592 | kunmap_atomic(map); |
| 3593 | while ((page = list_prev_entry(page, lru)) != head) { |
| 3594 | map = kmap_atomic(page) + offset; |
| 3595 | *map = SWAP_CONT_MAX | count; |
| 3596 | count = COUNT_CONTINUED; |
| 3597 | kunmap_atomic(map); |
| 3598 | } |
| 3599 | ret = count == COUNT_CONTINUED; |
| 3600 | } |
| 3601 | out: |
| 3602 | spin_unlock(&si->cont_lock); |
| 3603 | return ret; |
| 3604 | } |
| 3605 | |
| 3606 | /* |
| 3607 | * free_swap_count_continuations - swapoff free all the continuation pages |
| 3608 | * appended to the swap_map, after swap_map is quiesced, before vfree'ing it. |
| 3609 | */ |
| 3610 | static void free_swap_count_continuations(struct swap_info_struct *si) |
| 3611 | { |
| 3612 | pgoff_t offset; |
| 3613 | |
| 3614 | for (offset = 0; offset < si->max; offset += PAGE_SIZE) { |
| 3615 | struct page *head; |
| 3616 | head = vmalloc_to_page(si->swap_map + offset); |
| 3617 | if (page_private(head)) { |
| 3618 | struct page *page, *next; |
| 3619 | |
| 3620 | list_for_each_entry_safe(page, next, &head->lru, lru) { |
| 3621 | list_del(&page->lru); |
| 3622 | __free_page(page); |
| 3623 | } |
| 3624 | } |
| 3625 | } |
| 3626 | } |
| 3627 | |
| 3628 | #if defined(CONFIG_MEMCG) && defined(CONFIG_BLK_CGROUP) |
| 3629 | void __cgroup_throttle_swaprate(struct page *page, gfp_t gfp_mask) |
| 3630 | { |
| 3631 | struct swap_info_struct *si, *next; |
| 3632 | int nid = page_to_nid(page); |
| 3633 | |
| 3634 | if (!(gfp_mask & __GFP_IO)) |
| 3635 | return; |
| 3636 | |
| 3637 | if (!blk_cgroup_congested()) |
| 3638 | return; |
| 3639 | |
| 3640 | /* |
| 3641 | * We've already scheduled a throttle, avoid taking the global swap |
| 3642 | * lock. |
| 3643 | */ |
| 3644 | if (current->throttle_queue) |
| 3645 | return; |
| 3646 | |
| 3647 | spin_lock(&swap_avail_lock); |
| 3648 | plist_for_each_entry_safe(si, next, &swap_avail_heads[nid], |
| 3649 | avail_lists[nid]) { |
| 3650 | if (si->bdev) { |
| 3651 | blkcg_schedule_throttle(bdev_get_queue(si->bdev), true); |
| 3652 | break; |
| 3653 | } |
| 3654 | } |
| 3655 | spin_unlock(&swap_avail_lock); |
| 3656 | } |
| 3657 | #endif |
| 3658 | |
| 3659 | static int __init swapfile_init(void) |
| 3660 | { |
| 3661 | int nid; |
| 3662 | |
| 3663 | swap_avail_heads = kmalloc_array(nr_node_ids, sizeof(struct plist_head), |
| 3664 | GFP_KERNEL); |
| 3665 | if (!swap_avail_heads) { |
| 3666 | pr_emerg("Not enough memory for swap heads, swap is disabled\n"); |
| 3667 | return -ENOMEM; |
| 3668 | } |
| 3669 | |
| 3670 | for_each_node(nid) |
| 3671 | plist_head_init(&swap_avail_heads[nid]); |
| 3672 | |
| 3673 | return 0; |
| 3674 | } |
| 3675 | subsys_initcall(swapfile_init); |