• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * zswap.c - zswap driver file
3  *
4  * zswap is a backend for frontswap that takes pages that are in the process
5  * of being swapped out and attempts to compress and store them in a
6  * RAM-based memory pool.  This can result in a significant I/O reduction on
7  * the swap device and, in the case where decompressing from RAM is faster
8  * than reading from the swap device, can also improve workload performance.
9  *
10  * Copyright (C) 2012  Seth Jennings <sjenning@linux.vnet.ibm.com>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public License
14  * as published by the Free Software Foundation; either version 2
15  * of the License, or (at your option) any later version.
16  *
17  * This program is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20  * GNU General Public License for more details.
21 */
22 
23 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
24 
25 #include <linux/module.h>
26 #include <linux/cpu.h>
27 #include <linux/highmem.h>
28 #include <linux/slab.h>
29 #include <linux/spinlock.h>
30 #include <linux/types.h>
31 #include <linux/atomic.h>
32 #include <linux/frontswap.h>
33 #include <linux/rbtree.h>
34 #include <linux/swap.h>
35 #include <linux/crypto.h>
36 #include <linux/mempool.h>
37 #include <linux/zpool.h>
38 
39 #include <linux/mm_types.h>
40 #include <linux/page-flags.h>
41 #include <linux/swapops.h>
42 #include <linux/writeback.h>
43 #include <linux/pagemap.h>
44 
45 /*********************************
46 * statistics
47 **********************************/
48 /* Total bytes used by the compressed storage */
49 static u64 zswap_pool_total_size;
50 /* The number of compressed pages currently stored in zswap */
51 static atomic_t zswap_stored_pages = ATOMIC_INIT(0);
52 
53 /*
54  * The statistics below are not protected from concurrent access for
55  * performance reasons so they may not be a 100% accurate.  However,
56  * they do provide useful information on roughly how many times a
57  * certain event is occurring.
58 */
59 
60 /* Pool limit was hit (see zswap_max_pool_percent) */
61 static u64 zswap_pool_limit_hit;
62 /* Pages written back when pool limit was reached */
63 static u64 zswap_written_back_pages;
64 /* Store failed due to a reclaim failure after pool limit was reached */
65 static u64 zswap_reject_reclaim_fail;
66 /* Compressed page was too big for the allocator to (optimally) store */
67 static u64 zswap_reject_compress_poor;
68 /* Store failed because underlying allocator could not get memory */
69 static u64 zswap_reject_alloc_fail;
70 /* Store failed because the entry metadata could not be allocated (rare) */
71 static u64 zswap_reject_kmemcache_fail;
72 /* Duplicate store was encountered (rare) */
73 static u64 zswap_duplicate_entry;
74 
75 /*********************************
76 * tunables
77 **********************************/
78 /* Enable/disable zswap (disabled by default, fixed at boot for now) */
79 static bool zswap_enabled __read_mostly;
80 module_param_named(enabled, zswap_enabled, bool, 0444);
81 
82 /* Compressor to be used by zswap (fixed at boot for now) */
83 #define ZSWAP_COMPRESSOR_DEFAULT "lzo"
84 static char *zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
85 module_param_named(compressor, zswap_compressor, charp, 0444);
86 
87 /* The maximum percentage of memory that the compressed pool can occupy */
88 static unsigned int zswap_max_pool_percent = 20;
89 module_param_named(max_pool_percent,
90 			zswap_max_pool_percent, uint, 0644);
91 
92 /* Compressed storage to use */
93 #define ZSWAP_ZPOOL_DEFAULT "zbud"
94 static char *zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
95 module_param_named(zpool, zswap_zpool_type, charp, 0444);
96 
97 /* zpool is shared by all of zswap backend  */
98 static struct zpool *zswap_pool;
99 
100 /*********************************
101 * compression functions
102 **********************************/
103 /* per-cpu compression transforms */
104 static struct crypto_comp * __percpu *zswap_comp_pcpu_tfms;
105 
106 enum comp_op {
107 	ZSWAP_COMPOP_COMPRESS,
108 	ZSWAP_COMPOP_DECOMPRESS
109 };
110 
zswap_comp_op(enum comp_op op,const u8 * src,unsigned int slen,u8 * dst,unsigned int * dlen)111 static int zswap_comp_op(enum comp_op op, const u8 *src, unsigned int slen,
112 				u8 *dst, unsigned int *dlen)
113 {
114 	struct crypto_comp *tfm;
115 	int ret;
116 
117 	tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, get_cpu());
118 	switch (op) {
119 	case ZSWAP_COMPOP_COMPRESS:
120 		ret = crypto_comp_compress(tfm, src, slen, dst, dlen);
121 		break;
122 	case ZSWAP_COMPOP_DECOMPRESS:
123 		ret = crypto_comp_decompress(tfm, src, slen, dst, dlen);
124 		break;
125 	default:
126 		ret = -EINVAL;
127 	}
128 
129 	put_cpu();
130 	return ret;
131 }
132 
zswap_comp_init(void)133 static int __init zswap_comp_init(void)
134 {
135 	if (!crypto_has_comp(zswap_compressor, 0, 0)) {
136 		pr_info("%s compressor not available\n", zswap_compressor);
137 		/* fall back to default compressor */
138 		zswap_compressor = ZSWAP_COMPRESSOR_DEFAULT;
139 		if (!crypto_has_comp(zswap_compressor, 0, 0))
140 			/* can't even load the default compressor */
141 			return -ENODEV;
142 	}
143 	pr_info("using %s compressor\n", zswap_compressor);
144 
145 	/* alloc percpu transforms */
146 	zswap_comp_pcpu_tfms = alloc_percpu(struct crypto_comp *);
147 	if (!zswap_comp_pcpu_tfms)
148 		return -ENOMEM;
149 	return 0;
150 }
151 
zswap_comp_exit(void)152 static void zswap_comp_exit(void)
153 {
154 	/* free percpu transforms */
155 	if (zswap_comp_pcpu_tfms)
156 		free_percpu(zswap_comp_pcpu_tfms);
157 }
158 
159 /*********************************
160 * data structures
161 **********************************/
162 /*
163  * struct zswap_entry
164  *
165  * This structure contains the metadata for tracking a single compressed
166  * page within zswap.
167  *
168  * rbnode - links the entry into red-black tree for the appropriate swap type
169  * refcount - the number of outstanding reference to the entry. This is needed
170  *            to protect against premature freeing of the entry by code
171  *            concurrent calls to load, invalidate, and writeback.  The lock
172  *            for the zswap_tree structure that contains the entry must
173  *            be held while changing the refcount.  Since the lock must
174  *            be held, there is no reason to also make refcount atomic.
175  * offset - the swap offset for the entry.  Index into the red-black tree.
176  * handle - zpool allocation handle that stores the compressed page data
177  * length - the length in bytes of the compressed page data.  Needed during
178  *          decompression
179  */
180 struct zswap_entry {
181 	struct rb_node rbnode;
182 	pgoff_t offset;
183 	int refcount;
184 	unsigned int length;
185 	unsigned long handle;
186 };
187 
188 struct zswap_header {
189 	swp_entry_t swpentry;
190 };
191 
192 /*
193  * The tree lock in the zswap_tree struct protects a few things:
194  * - the rbtree
195  * - the refcount field of each entry in the tree
196  */
197 struct zswap_tree {
198 	struct rb_root rbroot;
199 	spinlock_t lock;
200 };
201 
202 static struct zswap_tree *zswap_trees[MAX_SWAPFILES];
203 
204 /*********************************
205 * zswap entry functions
206 **********************************/
207 static struct kmem_cache *zswap_entry_cache;
208 
zswap_entry_cache_create(void)209 static int zswap_entry_cache_create(void)
210 {
211 	zswap_entry_cache = KMEM_CACHE(zswap_entry, 0);
212 	return zswap_entry_cache == NULL;
213 }
214 
zswap_entry_cache_destroy(void)215 static void __init zswap_entry_cache_destroy(void)
216 {
217 	kmem_cache_destroy(zswap_entry_cache);
218 }
219 
zswap_entry_cache_alloc(gfp_t gfp)220 static struct zswap_entry *zswap_entry_cache_alloc(gfp_t gfp)
221 {
222 	struct zswap_entry *entry;
223 	entry = kmem_cache_alloc(zswap_entry_cache, gfp);
224 	if (!entry)
225 		return NULL;
226 	entry->refcount = 1;
227 	RB_CLEAR_NODE(&entry->rbnode);
228 	return entry;
229 }
230 
zswap_entry_cache_free(struct zswap_entry * entry)231 static void zswap_entry_cache_free(struct zswap_entry *entry)
232 {
233 	kmem_cache_free(zswap_entry_cache, entry);
234 }
235 
236 /*********************************
237 * rbtree functions
238 **********************************/
zswap_rb_search(struct rb_root * root,pgoff_t offset)239 static struct zswap_entry *zswap_rb_search(struct rb_root *root, pgoff_t offset)
240 {
241 	struct rb_node *node = root->rb_node;
242 	struct zswap_entry *entry;
243 
244 	while (node) {
245 		entry = rb_entry(node, struct zswap_entry, rbnode);
246 		if (entry->offset > offset)
247 			node = node->rb_left;
248 		else if (entry->offset < offset)
249 			node = node->rb_right;
250 		else
251 			return entry;
252 	}
253 	return NULL;
254 }
255 
256 /*
257  * In the case that a entry with the same offset is found, a pointer to
258  * the existing entry is stored in dupentry and the function returns -EEXIST
259  */
zswap_rb_insert(struct rb_root * root,struct zswap_entry * entry,struct zswap_entry ** dupentry)260 static int zswap_rb_insert(struct rb_root *root, struct zswap_entry *entry,
261 			struct zswap_entry **dupentry)
262 {
263 	struct rb_node **link = &root->rb_node, *parent = NULL;
264 	struct zswap_entry *myentry;
265 
266 	while (*link) {
267 		parent = *link;
268 		myentry = rb_entry(parent, struct zswap_entry, rbnode);
269 		if (myentry->offset > entry->offset)
270 			link = &(*link)->rb_left;
271 		else if (myentry->offset < entry->offset)
272 			link = &(*link)->rb_right;
273 		else {
274 			*dupentry = myentry;
275 			return -EEXIST;
276 		}
277 	}
278 	rb_link_node(&entry->rbnode, parent, link);
279 	rb_insert_color(&entry->rbnode, root);
280 	return 0;
281 }
282 
zswap_rb_erase(struct rb_root * root,struct zswap_entry * entry)283 static void zswap_rb_erase(struct rb_root *root, struct zswap_entry *entry)
284 {
285 	if (!RB_EMPTY_NODE(&entry->rbnode)) {
286 		rb_erase(&entry->rbnode, root);
287 		RB_CLEAR_NODE(&entry->rbnode);
288 	}
289 }
290 
291 /*
292  * Carries out the common pattern of freeing and entry's zpool allocation,
293  * freeing the entry itself, and decrementing the number of stored pages.
294  */
zswap_free_entry(struct zswap_entry * entry)295 static void zswap_free_entry(struct zswap_entry *entry)
296 {
297 	zpool_free(zswap_pool, entry->handle);
298 	zswap_entry_cache_free(entry);
299 	atomic_dec(&zswap_stored_pages);
300 	zswap_pool_total_size = zpool_get_total_size(zswap_pool);
301 }
302 
303 /* caller must hold the tree lock */
zswap_entry_get(struct zswap_entry * entry)304 static void zswap_entry_get(struct zswap_entry *entry)
305 {
306 	entry->refcount++;
307 }
308 
309 /* caller must hold the tree lock
310 * remove from the tree and free it, if nobody reference the entry
311 */
zswap_entry_put(struct zswap_tree * tree,struct zswap_entry * entry)312 static void zswap_entry_put(struct zswap_tree *tree,
313 			struct zswap_entry *entry)
314 {
315 	int refcount = --entry->refcount;
316 
317 	BUG_ON(refcount < 0);
318 	if (refcount == 0) {
319 		zswap_rb_erase(&tree->rbroot, entry);
320 		zswap_free_entry(entry);
321 	}
322 }
323 
324 /* caller must hold the tree lock */
zswap_entry_find_get(struct rb_root * root,pgoff_t offset)325 static struct zswap_entry *zswap_entry_find_get(struct rb_root *root,
326 				pgoff_t offset)
327 {
328 	struct zswap_entry *entry = NULL;
329 
330 	entry = zswap_rb_search(root, offset);
331 	if (entry)
332 		zswap_entry_get(entry);
333 
334 	return entry;
335 }
336 
337 /*********************************
338 * per-cpu code
339 **********************************/
340 static DEFINE_PER_CPU(u8 *, zswap_dstmem);
341 
__zswap_cpu_notifier(unsigned long action,unsigned long cpu)342 static int __zswap_cpu_notifier(unsigned long action, unsigned long cpu)
343 {
344 	struct crypto_comp *tfm;
345 	u8 *dst;
346 
347 	switch (action) {
348 	case CPU_UP_PREPARE:
349 		tfm = crypto_alloc_comp(zswap_compressor, 0, 0);
350 		if (IS_ERR(tfm)) {
351 			pr_err("can't allocate compressor transform\n");
352 			return NOTIFY_BAD;
353 		}
354 		*per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = tfm;
355 		dst = kmalloc_node(PAGE_SIZE * 2, GFP_KERNEL, cpu_to_node(cpu));
356 		if (!dst) {
357 			pr_err("can't allocate compressor buffer\n");
358 			crypto_free_comp(tfm);
359 			*per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
360 			return NOTIFY_BAD;
361 		}
362 		per_cpu(zswap_dstmem, cpu) = dst;
363 		break;
364 	case CPU_DEAD:
365 	case CPU_UP_CANCELED:
366 		tfm = *per_cpu_ptr(zswap_comp_pcpu_tfms, cpu);
367 		if (tfm) {
368 			crypto_free_comp(tfm);
369 			*per_cpu_ptr(zswap_comp_pcpu_tfms, cpu) = NULL;
370 		}
371 		dst = per_cpu(zswap_dstmem, cpu);
372 		kfree(dst);
373 		per_cpu(zswap_dstmem, cpu) = NULL;
374 		break;
375 	default:
376 		break;
377 	}
378 	return NOTIFY_OK;
379 }
380 
zswap_cpu_notifier(struct notifier_block * nb,unsigned long action,void * pcpu)381 static int zswap_cpu_notifier(struct notifier_block *nb,
382 				unsigned long action, void *pcpu)
383 {
384 	unsigned long cpu = (unsigned long)pcpu;
385 	return __zswap_cpu_notifier(action, cpu);
386 }
387 
388 static struct notifier_block zswap_cpu_notifier_block = {
389 	.notifier_call = zswap_cpu_notifier
390 };
391 
zswap_cpu_init(void)392 static int zswap_cpu_init(void)
393 {
394 	unsigned long cpu;
395 
396 	cpu_notifier_register_begin();
397 	for_each_online_cpu(cpu)
398 		if (__zswap_cpu_notifier(CPU_UP_PREPARE, cpu) != NOTIFY_OK)
399 			goto cleanup;
400 	__register_cpu_notifier(&zswap_cpu_notifier_block);
401 	cpu_notifier_register_done();
402 	return 0;
403 
404 cleanup:
405 	for_each_online_cpu(cpu)
406 		__zswap_cpu_notifier(CPU_UP_CANCELED, cpu);
407 	cpu_notifier_register_done();
408 	return -ENOMEM;
409 }
410 
411 /*********************************
412 * helpers
413 **********************************/
zswap_is_full(void)414 static bool zswap_is_full(void)
415 {
416 	return totalram_pages * zswap_max_pool_percent / 100 <
417 		DIV_ROUND_UP(zswap_pool_total_size, PAGE_SIZE);
418 }
419 
420 /*********************************
421 * writeback code
422 **********************************/
423 /* return enum for zswap_get_swap_cache_page */
424 enum zswap_get_swap_ret {
425 	ZSWAP_SWAPCACHE_NEW,
426 	ZSWAP_SWAPCACHE_EXIST,
427 	ZSWAP_SWAPCACHE_FAIL,
428 };
429 
430 /*
431  * zswap_get_swap_cache_page
432  *
433  * This is an adaption of read_swap_cache_async()
434  *
435  * This function tries to find a page with the given swap entry
436  * in the swapper_space address space (the swap cache).  If the page
437  * is found, it is returned in retpage.  Otherwise, a page is allocated,
438  * added to the swap cache, and returned in retpage.
439  *
440  * If success, the swap cache page is returned in retpage
441  * Returns ZSWAP_SWAPCACHE_EXIST if page was already in the swap cache
442  * Returns ZSWAP_SWAPCACHE_NEW if the new page needs to be populated,
443  *     the new page is added to swapcache and locked
444  * Returns ZSWAP_SWAPCACHE_FAIL on error
445  */
zswap_get_swap_cache_page(swp_entry_t entry,struct page ** retpage)446 static int zswap_get_swap_cache_page(swp_entry_t entry,
447 				struct page **retpage)
448 {
449 	struct page *found_page, *new_page = NULL;
450 	struct address_space *swapper_space = swap_address_space(entry);
451 	int err;
452 
453 	*retpage = NULL;
454 	do {
455 		/*
456 		 * First check the swap cache.  Since this is normally
457 		 * called after lookup_swap_cache() failed, re-calling
458 		 * that would confuse statistics.
459 		 */
460 		found_page = find_get_page(swapper_space, entry.val);
461 		if (found_page)
462 			break;
463 
464 		/*
465 		 * Get a new page to read into from swap.
466 		 */
467 		if (!new_page) {
468 			new_page = alloc_page(GFP_KERNEL);
469 			if (!new_page)
470 				break; /* Out of memory */
471 		}
472 
473 		/*
474 		 * call radix_tree_preload() while we can wait.
475 		 */
476 		err = radix_tree_preload(GFP_KERNEL);
477 		if (err)
478 			break;
479 
480 		/*
481 		 * Swap entry may have been freed since our caller observed it.
482 		 */
483 		err = swapcache_prepare(entry);
484 		if (err == -EEXIST) { /* seems racy */
485 			radix_tree_preload_end();
486 			continue;
487 		}
488 		if (err) { /* swp entry is obsolete ? */
489 			radix_tree_preload_end();
490 			break;
491 		}
492 
493 		/* May fail (-ENOMEM) if radix-tree node allocation failed. */
494 		__set_page_locked(new_page);
495 		SetPageSwapBacked(new_page);
496 		err = __add_to_swap_cache(new_page, entry);
497 		if (likely(!err)) {
498 			radix_tree_preload_end();
499 			lru_cache_add_anon(new_page);
500 			*retpage = new_page;
501 			return ZSWAP_SWAPCACHE_NEW;
502 		}
503 		radix_tree_preload_end();
504 		ClearPageSwapBacked(new_page);
505 		__clear_page_locked(new_page);
506 		/*
507 		 * add_to_swap_cache() doesn't return -EEXIST, so we can safely
508 		 * clear SWAP_HAS_CACHE flag.
509 		 */
510 		swapcache_free(entry);
511 	} while (err != -ENOMEM);
512 
513 	if (new_page)
514 		page_cache_release(new_page);
515 	if (!found_page)
516 		return ZSWAP_SWAPCACHE_FAIL;
517 	*retpage = found_page;
518 	return ZSWAP_SWAPCACHE_EXIST;
519 }
520 
521 /*
522  * Attempts to free an entry by adding a page to the swap cache,
523  * decompressing the entry data into the page, and issuing a
524  * bio write to write the page back to the swap device.
525  *
526  * This can be thought of as a "resumed writeback" of the page
527  * to the swap device.  We are basically resuming the same swap
528  * writeback path that was intercepted with the frontswap_store()
529  * in the first place.  After the page has been decompressed into
530  * the swap cache, the compressed version stored by zswap can be
531  * freed.
532  */
zswap_writeback_entry(struct zpool * pool,unsigned long handle)533 static int zswap_writeback_entry(struct zpool *pool, unsigned long handle)
534 {
535 	struct zswap_header *zhdr;
536 	swp_entry_t swpentry;
537 	struct zswap_tree *tree;
538 	pgoff_t offset;
539 	struct zswap_entry *entry;
540 	struct page *page;
541 	u8 *src, *dst;
542 	unsigned int dlen;
543 	int ret;
544 	struct writeback_control wbc = {
545 		.sync_mode = WB_SYNC_NONE,
546 	};
547 
548 	/* extract swpentry from data */
549 	zhdr = zpool_map_handle(pool, handle, ZPOOL_MM_RO);
550 	swpentry = zhdr->swpentry; /* here */
551 	zpool_unmap_handle(pool, handle);
552 	tree = zswap_trees[swp_type(swpentry)];
553 	offset = swp_offset(swpentry);
554 
555 	/* find and ref zswap entry */
556 	spin_lock(&tree->lock);
557 	entry = zswap_entry_find_get(&tree->rbroot, offset);
558 	if (!entry) {
559 		/* entry was invalidated */
560 		spin_unlock(&tree->lock);
561 		return 0;
562 	}
563 	spin_unlock(&tree->lock);
564 	BUG_ON(offset != entry->offset);
565 
566 	/* try to allocate swap cache page */
567 	switch (zswap_get_swap_cache_page(swpentry, &page)) {
568 	case ZSWAP_SWAPCACHE_FAIL: /* no memory or invalidate happened */
569 		ret = -ENOMEM;
570 		goto fail;
571 
572 	case ZSWAP_SWAPCACHE_EXIST:
573 		/* page is already in the swap cache, ignore for now */
574 		page_cache_release(page);
575 		ret = -EEXIST;
576 		goto fail;
577 
578 	case ZSWAP_SWAPCACHE_NEW: /* page is locked */
579 		/* decompress */
580 		dlen = PAGE_SIZE;
581 		src = (u8 *)zpool_map_handle(zswap_pool, entry->handle,
582 				ZPOOL_MM_RO) + sizeof(struct zswap_header);
583 		dst = kmap_atomic(page);
584 		ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src,
585 				entry->length, dst, &dlen);
586 		kunmap_atomic(dst);
587 		zpool_unmap_handle(zswap_pool, entry->handle);
588 		BUG_ON(ret);
589 		BUG_ON(dlen != PAGE_SIZE);
590 
591 		/* page is up to date */
592 		SetPageUptodate(page);
593 	}
594 
595 	/* move it to the tail of the inactive list after end_writeback */
596 	SetPageReclaim(page);
597 
598 	/* start writeback */
599 	__swap_writepage(page, &wbc, end_swap_bio_write);
600 	page_cache_release(page);
601 	zswap_written_back_pages++;
602 
603 	spin_lock(&tree->lock);
604 	/* drop local reference */
605 	zswap_entry_put(tree, entry);
606 
607 	/*
608 	* There are two possible situations for entry here:
609 	* (1) refcount is 1(normal case),  entry is valid and on the tree
610 	* (2) refcount is 0, entry is freed and not on the tree
611 	*     because invalidate happened during writeback
612 	*  search the tree and free the entry if find entry
613 	*/
614 	if (entry == zswap_rb_search(&tree->rbroot, offset))
615 		zswap_entry_put(tree, entry);
616 	spin_unlock(&tree->lock);
617 
618 	goto end;
619 
620 	/*
621 	* if we get here due to ZSWAP_SWAPCACHE_EXIST
622 	* a load may happening concurrently
623 	* it is safe and okay to not free the entry
624 	* if we free the entry in the following put
625 	* it it either okay to return !0
626 	*/
627 fail:
628 	spin_lock(&tree->lock);
629 	zswap_entry_put(tree, entry);
630 	spin_unlock(&tree->lock);
631 
632 end:
633 	return ret;
634 }
635 
636 /*********************************
637 * frontswap hooks
638 **********************************/
639 /* attempts to compress and store an single page */
zswap_frontswap_store(unsigned type,pgoff_t offset,struct page * page)640 static int zswap_frontswap_store(unsigned type, pgoff_t offset,
641 				struct page *page)
642 {
643 	struct zswap_tree *tree = zswap_trees[type];
644 	struct zswap_entry *entry, *dupentry;
645 	int ret;
646 	unsigned int dlen = PAGE_SIZE, len;
647 	unsigned long handle;
648 	char *buf;
649 	u8 *src, *dst;
650 	struct zswap_header *zhdr;
651 
652 	if (!tree) {
653 		ret = -ENODEV;
654 		goto reject;
655 	}
656 
657 	/* reclaim space if needed */
658 	if (zswap_is_full()) {
659 		zswap_pool_limit_hit++;
660 		if (zpool_shrink(zswap_pool, 1, NULL)) {
661 			zswap_reject_reclaim_fail++;
662 			ret = -ENOMEM;
663 			goto reject;
664 		}
665 	}
666 
667 	/* allocate entry */
668 	entry = zswap_entry_cache_alloc(GFP_KERNEL);
669 	if (!entry) {
670 		zswap_reject_kmemcache_fail++;
671 		ret = -ENOMEM;
672 		goto reject;
673 	}
674 
675 	/* compress */
676 	dst = get_cpu_var(zswap_dstmem);
677 	src = kmap_atomic(page);
678 	ret = zswap_comp_op(ZSWAP_COMPOP_COMPRESS, src, PAGE_SIZE, dst, &dlen);
679 	kunmap_atomic(src);
680 	if (ret) {
681 		ret = -EINVAL;
682 		goto freepage;
683 	}
684 
685 	/* store */
686 	len = dlen + sizeof(struct zswap_header);
687 	ret = zpool_malloc(zswap_pool, len, __GFP_NORETRY | __GFP_NOWARN,
688 		&handle);
689 	if (ret == -ENOSPC) {
690 		zswap_reject_compress_poor++;
691 		goto freepage;
692 	}
693 	if (ret) {
694 		zswap_reject_alloc_fail++;
695 		goto freepage;
696 	}
697 	zhdr = zpool_map_handle(zswap_pool, handle, ZPOOL_MM_RW);
698 	zhdr->swpentry = swp_entry(type, offset);
699 	buf = (u8 *)(zhdr + 1);
700 	memcpy(buf, dst, dlen);
701 	zpool_unmap_handle(zswap_pool, handle);
702 	put_cpu_var(zswap_dstmem);
703 
704 	/* populate entry */
705 	entry->offset = offset;
706 	entry->handle = handle;
707 	entry->length = dlen;
708 
709 	/* map */
710 	spin_lock(&tree->lock);
711 	do {
712 		ret = zswap_rb_insert(&tree->rbroot, entry, &dupentry);
713 		if (ret == -EEXIST) {
714 			zswap_duplicate_entry++;
715 			/* remove from rbtree */
716 			zswap_rb_erase(&tree->rbroot, dupentry);
717 			zswap_entry_put(tree, dupentry);
718 		}
719 	} while (ret == -EEXIST);
720 	spin_unlock(&tree->lock);
721 
722 	/* update stats */
723 	atomic_inc(&zswap_stored_pages);
724 	zswap_pool_total_size = zpool_get_total_size(zswap_pool);
725 
726 	return 0;
727 
728 freepage:
729 	put_cpu_var(zswap_dstmem);
730 	zswap_entry_cache_free(entry);
731 reject:
732 	return ret;
733 }
734 
735 /*
736  * returns 0 if the page was successfully decompressed
737  * return -1 on entry not found or error
738 */
zswap_frontswap_load(unsigned type,pgoff_t offset,struct page * page)739 static int zswap_frontswap_load(unsigned type, pgoff_t offset,
740 				struct page *page)
741 {
742 	struct zswap_tree *tree = zswap_trees[type];
743 	struct zswap_entry *entry;
744 	u8 *src, *dst;
745 	unsigned int dlen;
746 	int ret;
747 
748 	/* find */
749 	spin_lock(&tree->lock);
750 	entry = zswap_entry_find_get(&tree->rbroot, offset);
751 	if (!entry) {
752 		/* entry was written back */
753 		spin_unlock(&tree->lock);
754 		return -1;
755 	}
756 	spin_unlock(&tree->lock);
757 
758 	/* decompress */
759 	dlen = PAGE_SIZE;
760 	src = (u8 *)zpool_map_handle(zswap_pool, entry->handle,
761 			ZPOOL_MM_RO) + sizeof(struct zswap_header);
762 	dst = kmap_atomic(page);
763 	ret = zswap_comp_op(ZSWAP_COMPOP_DECOMPRESS, src, entry->length,
764 		dst, &dlen);
765 	kunmap_atomic(dst);
766 	zpool_unmap_handle(zswap_pool, entry->handle);
767 	BUG_ON(ret);
768 
769 	spin_lock(&tree->lock);
770 	zswap_entry_put(tree, entry);
771 	spin_unlock(&tree->lock);
772 
773 	return 0;
774 }
775 
776 /* frees an entry in zswap */
zswap_frontswap_invalidate_page(unsigned type,pgoff_t offset)777 static void zswap_frontswap_invalidate_page(unsigned type, pgoff_t offset)
778 {
779 	struct zswap_tree *tree = zswap_trees[type];
780 	struct zswap_entry *entry;
781 
782 	/* find */
783 	spin_lock(&tree->lock);
784 	entry = zswap_rb_search(&tree->rbroot, offset);
785 	if (!entry) {
786 		/* entry was written back */
787 		spin_unlock(&tree->lock);
788 		return;
789 	}
790 
791 	/* remove from rbtree */
792 	zswap_rb_erase(&tree->rbroot, entry);
793 
794 	/* drop the initial reference from entry creation */
795 	zswap_entry_put(tree, entry);
796 
797 	spin_unlock(&tree->lock);
798 }
799 
800 /* frees all zswap entries for the given swap type */
zswap_frontswap_invalidate_area(unsigned type)801 static void zswap_frontswap_invalidate_area(unsigned type)
802 {
803 	struct zswap_tree *tree = zswap_trees[type];
804 	struct zswap_entry *entry, *n;
805 
806 	if (!tree)
807 		return;
808 
809 	/* walk the tree and free everything */
810 	spin_lock(&tree->lock);
811 	rbtree_postorder_for_each_entry_safe(entry, n, &tree->rbroot, rbnode)
812 		zswap_free_entry(entry);
813 	tree->rbroot = RB_ROOT;
814 	spin_unlock(&tree->lock);
815 	kfree(tree);
816 	zswap_trees[type] = NULL;
817 }
818 
819 static struct zpool_ops zswap_zpool_ops = {
820 	.evict = zswap_writeback_entry
821 };
822 
zswap_frontswap_init(unsigned type)823 static void zswap_frontswap_init(unsigned type)
824 {
825 	struct zswap_tree *tree;
826 
827 	tree = kzalloc(sizeof(struct zswap_tree), GFP_KERNEL);
828 	if (!tree) {
829 		pr_err("alloc failed, zswap disabled for swap type %d\n", type);
830 		return;
831 	}
832 
833 	tree->rbroot = RB_ROOT;
834 	spin_lock_init(&tree->lock);
835 	zswap_trees[type] = tree;
836 }
837 
838 static struct frontswap_ops zswap_frontswap_ops = {
839 	.store = zswap_frontswap_store,
840 	.load = zswap_frontswap_load,
841 	.invalidate_page = zswap_frontswap_invalidate_page,
842 	.invalidate_area = zswap_frontswap_invalidate_area,
843 	.init = zswap_frontswap_init
844 };
845 
846 /*********************************
847 * debugfs functions
848 **********************************/
849 #ifdef CONFIG_DEBUG_FS
850 #include <linux/debugfs.h>
851 
852 static struct dentry *zswap_debugfs_root;
853 
zswap_debugfs_init(void)854 static int __init zswap_debugfs_init(void)
855 {
856 	if (!debugfs_initialized())
857 		return -ENODEV;
858 
859 	zswap_debugfs_root = debugfs_create_dir("zswap", NULL);
860 	if (!zswap_debugfs_root)
861 		return -ENOMEM;
862 
863 	debugfs_create_u64("pool_limit_hit", S_IRUGO,
864 			zswap_debugfs_root, &zswap_pool_limit_hit);
865 	debugfs_create_u64("reject_reclaim_fail", S_IRUGO,
866 			zswap_debugfs_root, &zswap_reject_reclaim_fail);
867 	debugfs_create_u64("reject_alloc_fail", S_IRUGO,
868 			zswap_debugfs_root, &zswap_reject_alloc_fail);
869 	debugfs_create_u64("reject_kmemcache_fail", S_IRUGO,
870 			zswap_debugfs_root, &zswap_reject_kmemcache_fail);
871 	debugfs_create_u64("reject_compress_poor", S_IRUGO,
872 			zswap_debugfs_root, &zswap_reject_compress_poor);
873 	debugfs_create_u64("written_back_pages", S_IRUGO,
874 			zswap_debugfs_root, &zswap_written_back_pages);
875 	debugfs_create_u64("duplicate_entry", S_IRUGO,
876 			zswap_debugfs_root, &zswap_duplicate_entry);
877 	debugfs_create_u64("pool_total_size", S_IRUGO,
878 			zswap_debugfs_root, &zswap_pool_total_size);
879 	debugfs_create_atomic_t("stored_pages", S_IRUGO,
880 			zswap_debugfs_root, &zswap_stored_pages);
881 
882 	return 0;
883 }
884 
zswap_debugfs_exit(void)885 static void __exit zswap_debugfs_exit(void)
886 {
887 	debugfs_remove_recursive(zswap_debugfs_root);
888 }
889 #else
zswap_debugfs_init(void)890 static int __init zswap_debugfs_init(void)
891 {
892 	return 0;
893 }
894 
zswap_debugfs_exit(void)895 static void __exit zswap_debugfs_exit(void) { }
896 #endif
897 
898 /*********************************
899 * module init and exit
900 **********************************/
init_zswap(void)901 static int __init init_zswap(void)
902 {
903 	gfp_t gfp = __GFP_NORETRY | __GFP_NOWARN;
904 
905 	if (!zswap_enabled)
906 		return 0;
907 
908 	pr_info("loading zswap\n");
909 
910 	zswap_pool = zpool_create_pool(zswap_zpool_type, "zswap", gfp,
911 					&zswap_zpool_ops);
912 	if (!zswap_pool && strcmp(zswap_zpool_type, ZSWAP_ZPOOL_DEFAULT)) {
913 		pr_info("%s zpool not available\n", zswap_zpool_type);
914 		zswap_zpool_type = ZSWAP_ZPOOL_DEFAULT;
915 		zswap_pool = zpool_create_pool(zswap_zpool_type, "zswap", gfp,
916 					&zswap_zpool_ops);
917 	}
918 	if (!zswap_pool) {
919 		pr_err("%s zpool not available\n", zswap_zpool_type);
920 		pr_err("zpool creation failed\n");
921 		goto error;
922 	}
923 	pr_info("using %s pool\n", zswap_zpool_type);
924 
925 	if (zswap_entry_cache_create()) {
926 		pr_err("entry cache creation failed\n");
927 		goto cachefail;
928 	}
929 	if (zswap_comp_init()) {
930 		pr_err("compressor initialization failed\n");
931 		goto compfail;
932 	}
933 	if (zswap_cpu_init()) {
934 		pr_err("per-cpu initialization failed\n");
935 		goto pcpufail;
936 	}
937 
938 	frontswap_register_ops(&zswap_frontswap_ops);
939 	if (zswap_debugfs_init())
940 		pr_warn("debugfs initialization failed\n");
941 	return 0;
942 pcpufail:
943 	zswap_comp_exit();
944 compfail:
945 	zswap_entry_cache_destroy();
946 cachefail:
947 	zpool_destroy_pool(zswap_pool);
948 error:
949 	return -ENOMEM;
950 }
951 /* must be late so crypto has time to come up */
952 late_initcall(init_zswap);
953 
954 MODULE_LICENSE("GPL");
955 MODULE_AUTHOR("Seth Jennings <sjenning@linux.vnet.ibm.com>");
956 MODULE_DESCRIPTION("Compressed cache for swap pages");
957