1	/*
2	* Compressed RAM block device
3	*
4	* Copyright (C) 2008, 2009, 2010 Nitin Gupta
5	* 2012, 2013 Minchan Kim
6	*
7	* This code is released using a dual license strategy: BSD/GPL
8	* You can choose the licence that better fits your requirements.
9	*
10	* Released under the terms of 3-clause BSD License
11	* Released under the terms of GNU General Public License Version 2.0
12	*
13	*/
14
15	#define KMSG_COMPONENT "zram"
16	#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17
18	#include <linux/module.h>
19	#include <linux/kernel.h>
20	#include <linux/bio.h>
21	#include <linux/bitops.h>
22	#include <linux/blkdev.h>
23	#include <linux/buffer_head.h>
24	#include <linux/device.h>
25	#include <linux/highmem.h>
26	#include <linux/slab.h>
27	#include <linux/backing-dev.h>
28	#include <linux/string.h>
29	#include <linux/vmalloc.h>
30	#include <linux/err.h>
31	#include <linux/idr.h>
32	#include <linux/sysfs.h>
33	#include <linux/debugfs.h>
34	#include <linux/cpuhotplug.h>
35	#include <linux/part_stat.h>
36
37	#include "zram_drv.h"
38
39	static DEFINE_IDR(zram_index_idr);
40	/* idr index must be protected */
41	static DEFINE_MUTEX(zram_index_mutex);
42
43	static int zram_major;
44	static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
45
46	/* Module params (documentation at end) */
47	static unsigned int num_devices = 1;
48	/*
49	* Pages that compress to sizes equals or greater than this are stored
50	* uncompressed in memory.
51	*/
52	static size_t huge_class_size;
53
54	static const struct block_device_operations zram_devops;
55
56	static void zram_free_page(struct zram *zram, size_t index);
57	static int zram_read_page(struct zram *zram, struct page *page, u32 index,
58	struct bio *parent);
59
60	static int zram_slot_trylock(struct zram *zram, u32 index)
61	{
62	return bit_spin_trylock(bitnum: ZRAM_LOCK, addr: &zram->table[index].flags);
63	}
64
65	static void zram_slot_lock(struct zram *zram, u32 index)
66	{
67	bit_spin_lock(bitnum: ZRAM_LOCK, addr: &zram->table[index].flags);
68	}
69
70	static void zram_slot_unlock(struct zram *zram, u32 index)
71	{
72	bit_spin_unlock(bitnum: ZRAM_LOCK, addr: &zram->table[index].flags);
73	}
74
75	static inline bool init_done(struct zram *zram)
76	{
77	return zram->disksize;
78	}
79
80	static inline struct zram *dev_to_zram(struct device *dev)
81	{
82	return (struct zram *)dev_to_disk(dev)->private_data;
83	}
84
85	static unsigned long zram_get_handle(struct zram *zram, u32 index)
86	{
87	return zram->table[index].handle;
88	}
89
90	static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
91	{
92	zram->table[index].handle = handle;
93	}
94
95	/* flag operations require table entry bit_spin_lock() being held */
96	static bool zram_test_flag(struct zram *zram, u32 index,
97	enum zram_pageflags flag)
98	{
99	return zram->table[index].flags & BIT(flag);
100	}
101
102	static void zram_set_flag(struct zram *zram, u32 index,
103	enum zram_pageflags flag)
104	{
105	zram->table[index].flags |= BIT(flag);
106	}
107
108	static void zram_clear_flag(struct zram *zram, u32 index,
109	enum zram_pageflags flag)
110	{
111	zram->table[index].flags &= ~BIT(flag);
112	}
113
114	static inline void zram_set_element(struct zram *zram, u32 index,
115	unsigned long element)
116	{
117	zram->table[index].element = element;
118	}
119
120	static unsigned long zram_get_element(struct zram *zram, u32 index)
121	{
122	return zram->table[index].element;
123	}
124
125	static size_t zram_get_obj_size(struct zram *zram, u32 index)
126	{
127	return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
128	}
129
130	static void zram_set_obj_size(struct zram *zram,
131	u32 index, size_t size)
132	{
133	unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
134
135	zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
136	}
137
138	static inline bool zram_allocated(struct zram *zram, u32 index)
139	{
140	return zram_get_obj_size(zram, index) ||
141	zram_test_flag(zram, index, flag: ZRAM_SAME) ||
142	zram_test_flag(zram, index, flag: ZRAM_WB);
143	}
144
145	#if PAGE_SIZE != 4096
146	static inline bool is_partial_io(struct bio_vec *bvec)
147	{
148	return bvec->bv_len != PAGE_SIZE;
149	}
150	#define ZRAM_PARTIAL_IO 1
151	#else
152	static inline bool is_partial_io(struct bio_vec *bvec)
153	{
154	return false;
155	}
156	#endif
157
158	static inline void zram_set_priority(struct zram *zram, u32 index, u32 prio)
159	{
160	prio &= ZRAM_COMP_PRIORITY_MASK;
161	/*
162	* Clear previous priority value first, in case if we recompress
163	* further an already recompressed page
164	*/
165	zram->table[index].flags &= ~(ZRAM_COMP_PRIORITY_MASK <<
166	ZRAM_COMP_PRIORITY_BIT1);
167	zram->table[index].flags |= (prio << ZRAM_COMP_PRIORITY_BIT1);
168	}
169
170	static inline u32 zram_get_priority(struct zram *zram, u32 index)
171	{
172	u32 prio = zram->table[index].flags >> ZRAM_COMP_PRIORITY_BIT1;
173
174	return prio & ZRAM_COMP_PRIORITY_MASK;
175	}
176
177	static inline void update_used_max(struct zram *zram,
178	const unsigned long pages)
179	{
180	unsigned long cur_max = atomic_long_read(v: &zram->stats.max_used_pages);
181
182	do {
183	if (cur_max >= pages)
184	return;
185	} while (!atomic_long_try_cmpxchg(v: &zram->stats.max_used_pages,
186	old: &cur_max, new: pages));
187	}
188
189	static inline void zram_fill_page(void *ptr, unsigned long len,
190	unsigned long value)
191	{
192	WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
193	memset_l(p: ptr, v: value, n: len / sizeof(unsigned long));
194	}
195
196	static bool page_same_filled(void *ptr, unsigned long *element)
197	{
198	unsigned long *page;
199	unsigned long val;
200	unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
201
202	page = (unsigned long *)ptr;
203	val = page[0];
204
205	if (val != page[last_pos])
206	return false;
207
208	for (pos = 1; pos < last_pos; pos++) {
209	if (val != page[pos])
210	return false;
211	}
212
213	*element = val;
214
215	return true;
216	}
217
218	static ssize_t initstate_show(struct device *dev,
219	struct device_attribute *attr, char *buf)
220	{
221	u32 val;
222	struct zram *zram = dev_to_zram(dev);
223
224	down_read(sem: &zram->init_lock);
225	val = init_done(zram);
226	up_read(sem: &zram->init_lock);
227
228	return scnprintf(buf, PAGE_SIZE, fmt: "%u\n", val);
229	}
230
231	static ssize_t disksize_show(struct device *dev,
232	struct device_attribute *attr, char *buf)
233	{
234	struct zram *zram = dev_to_zram(dev);
235
236	return scnprintf(buf, PAGE_SIZE, fmt: "%llu\n", zram->disksize);
237	}
238
239	static ssize_t mem_limit_store(struct device *dev,
240	struct device_attribute *attr, const char *buf, size_t len)
241	{
242	u64 limit;
243	char *tmp;
244	struct zram *zram = dev_to_zram(dev);
245
246	limit = memparse(ptr: buf, retptr: &tmp);
247	if (buf == tmp) /* no chars parsed, invalid input */
248	return -EINVAL;
249
250	down_write(sem: &zram->init_lock);
251	zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
252	up_write(sem: &zram->init_lock);
253
254	return len;
255	}
256
257	static ssize_t mem_used_max_store(struct device *dev,
258	struct device_attribute *attr, const char *buf, size_t len)
259	{
260	int err;
261	unsigned long val;
262	struct zram *zram = dev_to_zram(dev);
263
264	err = kstrtoul(s: buf, base: 10, res: &val);
265	if (err || val != 0)
266	return -EINVAL;
267
268	down_read(sem: &zram->init_lock);
269	if (init_done(zram)) {
270	atomic_long_set(v: &zram->stats.max_used_pages,
271	i: zs_get_total_pages(pool: zram->mem_pool));
272	}
273	up_read(sem: &zram->init_lock);
274
275	return len;
276	}
277
278	/*
279	* Mark all pages which are older than or equal to cutoff as IDLE.
280	* Callers should hold the zram init lock in read mode
281	*/
282	static void mark_idle(struct zram *zram, ktime_t cutoff)
283	{
284	int is_idle = 1;
285	unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
286	int index;
287
288	for (index = 0; index < nr_pages; index++) {
289	/*
290	* Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
291	* See the comment in writeback_store.
292	*/
293	zram_slot_lock(zram, index);
294	if (zram_allocated(zram, index) &&
295	!zram_test_flag(zram, index, flag: ZRAM_UNDER_WB)) {
296	#ifdef CONFIG_ZRAM_MEMORY_TRACKING
297	is_idle = !cutoff || ktime_after(cmp1: cutoff, cmp2: zram->table[index].ac_time);
298	#endif
299	if (is_idle)
300	zram_set_flag(zram, index, flag: ZRAM_IDLE);
301	}
302	zram_slot_unlock(zram, index);
303	}
304	}
305
306	static ssize_t idle_store(struct device *dev,
307	struct device_attribute *attr, const char *buf, size_t len)
308	{
309	struct zram *zram = dev_to_zram(dev);
310	ktime_t cutoff_time = 0;
311	ssize_t rv = -EINVAL;
312
313	if (!sysfs_streq(s1: buf, s2: "all")) {
314	/*
315	* If it did not parse as 'all' try to treat it as an integer
316	* when we have memory tracking enabled.
317	*/
318	u64 age_sec;
319
320	if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(s: buf, base: 0, res: &age_sec))
321	cutoff_time = ktime_sub(ktime_get_boottime(),
322	ns_to_ktime(age_sec * NSEC_PER_SEC));
323	else
324	goto out;
325	}
326
327	down_read(sem: &zram->init_lock);
328	if (!init_done(zram))
329	goto out_unlock;
330
331	/*
332	* A cutoff_time of 0 marks everything as idle, this is the
333	* "all" behavior.
334	*/
335	mark_idle(zram, cutoff: cutoff_time);
336	rv = len;
337
338	out_unlock:
339	up_read(sem: &zram->init_lock);
340	out:
341	return rv;
342	}
343
344	#ifdef CONFIG_ZRAM_WRITEBACK
345	static ssize_t writeback_limit_enable_store(struct device *dev,
346	struct device_attribute *attr, const char *buf, size_t len)
347	{
348	struct zram *zram = dev_to_zram(dev);
349	u64 val;
350	ssize_t ret = -EINVAL;
351
352	if (kstrtoull(s: buf, base: 10, res: &val))
353	return ret;
354
355	down_read(sem: &zram->init_lock);
356	spin_lock(lock: &zram->wb_limit_lock);
357	zram->wb_limit_enable = val;
358	spin_unlock(lock: &zram->wb_limit_lock);
359	up_read(sem: &zram->init_lock);
360	ret = len;
361
362	return ret;
363	}
364
365	static ssize_t writeback_limit_enable_show(struct device *dev,
366	struct device_attribute *attr, char *buf)
367	{
368	bool val;
369	struct zram *zram = dev_to_zram(dev);
370
371	down_read(sem: &zram->init_lock);
372	spin_lock(lock: &zram->wb_limit_lock);
373	val = zram->wb_limit_enable;
374	spin_unlock(lock: &zram->wb_limit_lock);
375	up_read(sem: &zram->init_lock);
376
377	return scnprintf(buf, PAGE_SIZE, fmt: "%d\n", val);
378	}
379
380	static ssize_t writeback_limit_store(struct device *dev,
381	struct device_attribute *attr, const char *buf, size_t len)
382	{
383	struct zram *zram = dev_to_zram(dev);
384	u64 val;
385	ssize_t ret = -EINVAL;
386
387	if (kstrtoull(s: buf, base: 10, res: &val))
388	return ret;
389
390	down_read(sem: &zram->init_lock);
391	spin_lock(lock: &zram->wb_limit_lock);
392	zram->bd_wb_limit = val;
393	spin_unlock(lock: &zram->wb_limit_lock);
394	up_read(sem: &zram->init_lock);
395	ret = len;
396
397	return ret;
398	}
399
400	static ssize_t writeback_limit_show(struct device *dev,
401	struct device_attribute *attr, char *buf)
402	{
403	u64 val;
404	struct zram *zram = dev_to_zram(dev);
405
406	down_read(sem: &zram->init_lock);
407	spin_lock(lock: &zram->wb_limit_lock);
408	val = zram->bd_wb_limit;
409	spin_unlock(lock: &zram->wb_limit_lock);
410	up_read(sem: &zram->init_lock);
411
412	return scnprintf(buf, PAGE_SIZE, fmt: "%llu\n", val);
413	}
414
415	static void reset_bdev(struct zram *zram)
416	{
417	if (!zram->backing_dev)
418	return;
419
420	bdev_release(handle: zram->bdev_handle);
421	/* hope filp_close flush all of IO */
422	filp_close(zram->backing_dev, NULL);
423	zram->backing_dev = NULL;
424	zram->bdev_handle = NULL;
425	zram->disk->fops = &zram_devops;
426	kvfree(addr: zram->bitmap);
427	zram->bitmap = NULL;
428	}
429
430	static ssize_t backing_dev_show(struct device *dev,
431	struct device_attribute *attr, char *buf)
432	{
433	struct file *file;
434	struct zram *zram = dev_to_zram(dev);
435	char *p;
436	ssize_t ret;
437
438	down_read(sem: &zram->init_lock);
439	file = zram->backing_dev;
440	if (!file) {
441	memcpy(buf, "none\n", 5);
442	up_read(sem: &zram->init_lock);
443	return 5;
444	}
445
446	p = file_path(file, buf, PAGE_SIZE - 1);
447	if (IS_ERR(ptr: p)) {
448	ret = PTR_ERR(ptr: p);
449	goto out;
450	}
451
452	ret = strlen(p);
453	memmove(buf, p, ret);
454	buf[ret++] = '\n';
455	out:
456	up_read(sem: &zram->init_lock);
457	return ret;
458	}
459
460	static ssize_t backing_dev_store(struct device *dev,
461	struct device_attribute *attr, const char *buf, size_t len)
462	{
463	char *file_name;
464	size_t sz;
465	struct file *backing_dev = NULL;
466	struct inode *inode;
467	struct address_space *mapping;
468	unsigned int bitmap_sz;
469	unsigned long nr_pages, *bitmap = NULL;
470	struct bdev_handle *bdev_handle = NULL;
471	int err;
472	struct zram *zram = dev_to_zram(dev);
473
474	file_name = kmalloc(PATH_MAX, GFP_KERNEL);
475	if (!file_name)
476	return -ENOMEM;
477
478	down_write(sem: &zram->init_lock);
479	if (init_done(zram)) {
480	pr_info("Can't setup backing device for initialized device\n");
481	err = -EBUSY;
482	goto out;
483	}
484
485	strscpy(p: file_name, q: buf, PATH_MAX);
486	/* ignore trailing newline */
487	sz = strlen(file_name);
488	if (sz > 0 && file_name[sz - 1] == '\n')
489	file_name[sz - 1] = 0x00;
490
491	backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
492	if (IS_ERR(ptr: backing_dev)) {
493	err = PTR_ERR(ptr: backing_dev);
494	backing_dev = NULL;
495	goto out;
496	}
497
498	mapping = backing_dev->f_mapping;
499	inode = mapping->host;
500
501	/* Support only block device in this moment */
502	if (!S_ISBLK(inode->i_mode)) {
503	err = -ENOTBLK;
504	goto out;
505	}
506
507	bdev_handle = bdev_open_by_dev(dev: inode->i_rdev,
508	BLK_OPEN_READ | BLK_OPEN_WRITE, holder: zram, NULL);
509	if (IS_ERR(ptr: bdev_handle)) {
510	err = PTR_ERR(ptr: bdev_handle);
511	bdev_handle = NULL;
512	goto out;
513	}
514
515	nr_pages = i_size_read(inode) >> PAGE_SHIFT;
516	bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
517	bitmap = kvzalloc(size: bitmap_sz, GFP_KERNEL);
518	if (!bitmap) {
519	err = -ENOMEM;
520	goto out;
521	}
522
523	reset_bdev(zram);
524
525	zram->bdev_handle = bdev_handle;
526	zram->backing_dev = backing_dev;
527	zram->bitmap = bitmap;
528	zram->nr_pages = nr_pages;
529	up_write(sem: &zram->init_lock);
530
531	pr_info("setup backing device %s\n", file_name);
532	kfree(objp: file_name);
533
534	return len;
535	out:
536	kvfree(addr: bitmap);
537
538	if (bdev_handle)
539	bdev_release(handle: bdev_handle);
540
541	if (backing_dev)
542	filp_close(backing_dev, NULL);
543
544	up_write(sem: &zram->init_lock);
545
546	kfree(objp: file_name);
547
548	return err;
549	}
550
551	static unsigned long alloc_block_bdev(struct zram *zram)
552	{
553	unsigned long blk_idx = 1;
554	retry:
555	/* skip 0 bit to confuse zram.handle = 0 */
556	blk_idx = find_next_zero_bit(addr: zram->bitmap, size: zram->nr_pages, offset: blk_idx);
557	if (blk_idx == zram->nr_pages)
558	return 0;
559
560	if (test_and_set_bit(nr: blk_idx, addr: zram->bitmap))
561	goto retry;
562
563	atomic64_inc(v: &zram->stats.bd_count);
564	return blk_idx;
565	}
566
567	static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
568	{
569	int was_set;
570
571	was_set = test_and_clear_bit(nr: blk_idx, addr: zram->bitmap);
572	WARN_ON_ONCE(!was_set);
573	atomic64_dec(v: &zram->stats.bd_count);
574	}
575
576	static void read_from_bdev_async(struct zram *zram, struct page *page,
577	unsigned long entry, struct bio *parent)
578	{
579	struct bio *bio;
580
581	bio = bio_alloc(bdev: zram->bdev_handle->bdev, nr_vecs: 1, opf: parent->bi_opf, GFP_NOIO);
582	bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
583	__bio_add_page(bio, page, PAGE_SIZE, off: 0);
584	bio_chain(bio, parent);
585	submit_bio(bio);
586	}
587
588	#define PAGE_WB_SIG "page_index="
589
590	#define PAGE_WRITEBACK 0
591	#define HUGE_WRITEBACK (1<<0)
592	#define IDLE_WRITEBACK (1<<1)
593	#define INCOMPRESSIBLE_WRITEBACK (1<<2)
594
595	static ssize_t writeback_store(struct device *dev,
596	struct device_attribute *attr, const char *buf, size_t len)
597	{
598	struct zram *zram = dev_to_zram(dev);
599	unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
600	unsigned long index = 0;
601	struct bio bio;
602	struct bio_vec bio_vec;
603	struct page *page;
604	ssize_t ret = len;
605	int mode, err;
606	unsigned long blk_idx = 0;
607
608	if (sysfs_streq(s1: buf, s2: "idle"))
609	mode = IDLE_WRITEBACK;
610	else if (sysfs_streq(s1: buf, s2: "huge"))
611	mode = HUGE_WRITEBACK;
612	else if (sysfs_streq(s1: buf, s2: "huge_idle"))
613	mode = IDLE_WRITEBACK | HUGE_WRITEBACK;
614	else if (sysfs_streq(s1: buf, s2: "incompressible"))
615	mode = INCOMPRESSIBLE_WRITEBACK;
616	else {
617	if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
618	return -EINVAL;
619
620	if (kstrtol(s: buf + sizeof(PAGE_WB_SIG) - 1, base: 10, res: &index) ||
621	index >= nr_pages)
622	return -EINVAL;
623
624	nr_pages = 1;
625	mode = PAGE_WRITEBACK;
626	}
627
628	down_read(sem: &zram->init_lock);
629	if (!init_done(zram)) {
630	ret = -EINVAL;
631	goto release_init_lock;
632	}
633
634	if (!zram->backing_dev) {
635	ret = -ENODEV;
636	goto release_init_lock;
637	}
638
639	page = alloc_page(GFP_KERNEL);
640	if (!page) {
641	ret = -ENOMEM;
642	goto release_init_lock;
643	}
644
645	for (; nr_pages != 0; index++, nr_pages--) {
646	spin_lock(lock: &zram->wb_limit_lock);
647	if (zram->wb_limit_enable && !zram->bd_wb_limit) {
648	spin_unlock(lock: &zram->wb_limit_lock);
649	ret = -EIO;
650	break;
651	}
652	spin_unlock(lock: &zram->wb_limit_lock);
653
654	if (!blk_idx) {
655	blk_idx = alloc_block_bdev(zram);
656	if (!blk_idx) {
657	ret = -ENOSPC;
658	break;
659	}
660	}
661
662	zram_slot_lock(zram, index);
663	if (!zram_allocated(zram, index))
664	goto next;
665
666	if (zram_test_flag(zram, index, flag: ZRAM_WB) ||
667	zram_test_flag(zram, index, flag: ZRAM_SAME) ||
668	zram_test_flag(zram, index, flag: ZRAM_UNDER_WB))
669	goto next;
670
671	if (mode & IDLE_WRITEBACK &&
672	!zram_test_flag(zram, index, flag: ZRAM_IDLE))
673	goto next;
674	if (mode & HUGE_WRITEBACK &&
675	!zram_test_flag(zram, index, flag: ZRAM_HUGE))
676	goto next;
677	if (mode & INCOMPRESSIBLE_WRITEBACK &&
678	!zram_test_flag(zram, index, flag: ZRAM_INCOMPRESSIBLE))
679	goto next;
680
681	/*
682	* Clearing ZRAM_UNDER_WB is duty of caller.
683	* IOW, zram_free_page never clear it.
684	*/
685	zram_set_flag(zram, index, flag: ZRAM_UNDER_WB);
686	/* Need for hugepage writeback racing */
687	zram_set_flag(zram, index, flag: ZRAM_IDLE);
688	zram_slot_unlock(zram, index);
689	if (zram_read_page(zram, page, index, NULL)) {
690	zram_slot_lock(zram, index);
691	zram_clear_flag(zram, index, flag: ZRAM_UNDER_WB);
692	zram_clear_flag(zram, index, flag: ZRAM_IDLE);
693	zram_slot_unlock(zram, index);
694	continue;
695	}
696
697	bio_init(bio: &bio, bdev: zram->bdev_handle->bdev, table: &bio_vec, max_vecs: 1,
698	opf: REQ_OP_WRITE | REQ_SYNC);
699	bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
700	__bio_add_page(bio: &bio, page, PAGE_SIZE, off: 0);
701
702	/*
703	* XXX: A single page IO would be inefficient for write
704	* but it would be not bad as starter.
705	*/
706	err = submit_bio_wait(bio: &bio);
707	if (err) {
708	zram_slot_lock(zram, index);
709	zram_clear_flag(zram, index, flag: ZRAM_UNDER_WB);
710	zram_clear_flag(zram, index, flag: ZRAM_IDLE);
711	zram_slot_unlock(zram, index);
712	/*
713	* BIO errors are not fatal, we continue and simply
714	* attempt to writeback the remaining objects (pages).
715	* At the same time we need to signal user-space that
716	* some writes (at least one, but also could be all of
717	* them) were not successful and we do so by returning
718	* the most recent BIO error.
719	*/
720	ret = err;
721	continue;
722	}
723
724	atomic64_inc(v: &zram->stats.bd_writes);
725	/*
726	* We released zram_slot_lock so need to check if the slot was
727	* changed. If there is freeing for the slot, we can catch it
728	* easily by zram_allocated.
729	* A subtle case is the slot is freed/reallocated/marked as
730	* ZRAM_IDLE again. To close the race, idle_store doesn't
731	* mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
732	* Thus, we could close the race by checking ZRAM_IDLE bit.
733	*/
734	zram_slot_lock(zram, index);
735	if (!zram_allocated(zram, index) ||
736	!zram_test_flag(zram, index, flag: ZRAM_IDLE)) {
737	zram_clear_flag(zram, index, flag: ZRAM_UNDER_WB);
738	zram_clear_flag(zram, index, flag: ZRAM_IDLE);
739	goto next;
740	}
741
742	zram_free_page(zram, index);
743	zram_clear_flag(zram, index, flag: ZRAM_UNDER_WB);
744	zram_set_flag(zram, index, flag: ZRAM_WB);
745	zram_set_element(zram, index, element: blk_idx);
746	blk_idx = 0;
747	atomic64_inc(v: &zram->stats.pages_stored);
748	spin_lock(lock: &zram->wb_limit_lock);
749	if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
750	zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
751	spin_unlock(lock: &zram->wb_limit_lock);
752	next:
753	zram_slot_unlock(zram, index);
754	}
755
756	if (blk_idx)
757	free_block_bdev(zram, blk_idx);
758	__free_page(page);
759	release_init_lock:
760	up_read(sem: &zram->init_lock);
761
762	return ret;
763	}
764
765	struct zram_work {
766	struct work_struct work;
767	struct zram *zram;
768	unsigned long entry;
769	struct page *page;
770	int error;
771	};
772
773	static void zram_sync_read(struct work_struct *work)
774	{
775	struct zram_work *zw = container_of(work, struct zram_work, work);
776	struct bio_vec bv;
777	struct bio bio;
778
779	bio_init(bio: &bio, bdev: zw->zram->bdev_handle->bdev, table: &bv, max_vecs: 1, opf: REQ_OP_READ);
780	bio.bi_iter.bi_sector = zw->entry * (PAGE_SIZE >> 9);
781	__bio_add_page(bio: &bio, page: zw->page, PAGE_SIZE, off: 0);
782	zw->error = submit_bio_wait(bio: &bio);
783	}
784
785	/*
786	* Block layer want one ->submit_bio to be active at a time, so if we use
787	* chained IO with parent IO in same context, it's a deadlock. To avoid that,
788	* use a worker thread context.
789	*/
790	static int read_from_bdev_sync(struct zram *zram, struct page *page,
791	unsigned long entry)
792	{
793	struct zram_work work;
794
795	work.page = page;
796	work.zram = zram;
797	work.entry = entry;
798
799	INIT_WORK_ONSTACK(&work.work, zram_sync_read);
800	queue_work(wq: system_unbound_wq, work: &work.work);
801	flush_work(work: &work.work);
802	destroy_work_on_stack(work: &work.work);
803
804	return work.error;
805	}
806
807	static int read_from_bdev(struct zram *zram, struct page *page,
808	unsigned long entry, struct bio *parent)
809	{
810	atomic64_inc(v: &zram->stats.bd_reads);
811	if (!parent) {
812	if (WARN_ON_ONCE(!IS_ENABLED(ZRAM_PARTIAL_IO)))
813	return -EIO;
814	return read_from_bdev_sync(zram, page, entry);
815	}
816	read_from_bdev_async(zram, page, entry, parent);
817	return 0;
818	}
819	#else
820	static inline void reset_bdev(struct zram *zram) {};
821	static int read_from_bdev(struct zram *zram, struct page *page,
822	unsigned long entry, struct bio *parent)
823	{
824	return -EIO;
825	}
826
827	static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
828	#endif
829
830	#ifdef CONFIG_ZRAM_MEMORY_TRACKING
831
832	static struct dentry *zram_debugfs_root;
833
834	static void zram_debugfs_create(void)
835	{
836	zram_debugfs_root = debugfs_create_dir(name: "zram", NULL);
837	}
838
839	static void zram_debugfs_destroy(void)
840	{
841	debugfs_remove_recursive(dentry: zram_debugfs_root);
842	}
843
844	static void zram_accessed(struct zram *zram, u32 index)
845	{
846	zram_clear_flag(zram, index, flag: ZRAM_IDLE);
847	zram->table[index].ac_time = ktime_get_boottime();
848	}
849
850	static ssize_t read_block_state(struct file *file, char __user *buf,
851	size_t count, loff_t *ppos)
852	{
853	char *kbuf;
854	ssize_t index, written = 0;
855	struct zram *zram = file->private_data;
856	unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
857	struct timespec64 ts;
858
859	kbuf = kvmalloc(size: count, GFP_KERNEL);
860	if (!kbuf)
861	return -ENOMEM;
862
863	down_read(sem: &zram->init_lock);
864	if (!init_done(zram)) {
865	up_read(sem: &zram->init_lock);
866	kvfree(addr: kbuf);
867	return -EINVAL;
868	}
869
870	for (index = *ppos; index < nr_pages; index++) {
871	int copied;
872
873	zram_slot_lock(zram, index);
874	if (!zram_allocated(zram, index))
875	goto next;
876
877	ts = ktime_to_timespec64(zram->table[index].ac_time);
878	copied = snprintf(buf: kbuf + written, size: count,
879	fmt: "%12zd %12lld.%06lu %c%c%c%c%c%c\n",
880	index, (s64)ts.tv_sec,
881	ts.tv_nsec / NSEC_PER_USEC,
882	zram_test_flag(zram, index, flag: ZRAM_SAME) ? 's' : '.',
883	zram_test_flag(zram, index, flag: ZRAM_WB) ? 'w' : '.',
884	zram_test_flag(zram, index, flag: ZRAM_HUGE) ? 'h' : '.',
885	zram_test_flag(zram, index, flag: ZRAM_IDLE) ? 'i' : '.',
886	zram_get_priority(zram, index) ? 'r' : '.',
887	zram_test_flag(zram, index,
888	flag: ZRAM_INCOMPRESSIBLE) ? 'n' : '.');
889
890	if (count <= copied) {
891	zram_slot_unlock(zram, index);
892	break;
893	}
894	written += copied;
895	count -= copied;
896	next:
897	zram_slot_unlock(zram, index);
898	*ppos += 1;
899	}
900
901	up_read(sem: &zram->init_lock);
902	if (copy_to_user(to: buf, from: kbuf, n: written))
903	written = -EFAULT;
904	kvfree(addr: kbuf);
905
906	return written;
907	}
908
909	static const struct file_operations proc_zram_block_state_op = {
910	.open = simple_open,
911	.read = read_block_state,
912	.llseek = default_llseek,
913	};
914
915	static void zram_debugfs_register(struct zram *zram)
916	{
917	if (!zram_debugfs_root)
918	return;
919
920	zram->debugfs_dir = debugfs_create_dir(name: zram->disk->disk_name,
921	parent: zram_debugfs_root);
922	debugfs_create_file(name: "block_state", mode: 0400, parent: zram->debugfs_dir,
923	data: zram, fops: &proc_zram_block_state_op);
924	}
925
926	static void zram_debugfs_unregister(struct zram *zram)
927	{
928	debugfs_remove_recursive(dentry: zram->debugfs_dir);
929	}
930	#else
931	static void zram_debugfs_create(void) {};
932	static void zram_debugfs_destroy(void) {};
933	static void zram_accessed(struct zram *zram, u32 index)
934	{
935	zram_clear_flag(zram, index, ZRAM_IDLE);
936	};
937	static void zram_debugfs_register(struct zram *zram) {};
938	static void zram_debugfs_unregister(struct zram *zram) {};
939	#endif
940
941	/*
942	* We switched to per-cpu streams and this attr is not needed anymore.
943	* However, we will keep it around for some time, because:
944	* a) we may revert per-cpu streams in the future
945	* b) it's visible to user space and we need to follow our 2 years
946	* retirement rule; but we already have a number of 'soon to be
947	* altered' attrs, so max_comp_streams need to wait for the next
948	* layoff cycle.
949	*/
950	static ssize_t max_comp_streams_show(struct device *dev,
951	struct device_attribute *attr, char *buf)
952	{
953	return scnprintf(buf, PAGE_SIZE, fmt: "%d\n", num_online_cpus());
954	}
955
956	static ssize_t max_comp_streams_store(struct device *dev,
957	struct device_attribute *attr, const char *buf, size_t len)
958	{
959	return len;
960	}
961
962	static void comp_algorithm_set(struct zram *zram, u32 prio, const char *alg)
963	{
964	/* Do not free statically defined compression algorithms */
965	if (zram->comp_algs[prio] != default_compressor)
966	kfree(objp: zram->comp_algs[prio]);
967
968	zram->comp_algs[prio] = alg;
969	}
970
971	static ssize_t __comp_algorithm_show(struct zram *zram, u32 prio, char *buf)
972	{
973	ssize_t sz;
974
975	down_read(sem: &zram->init_lock);
976	sz = zcomp_available_show(comp: zram->comp_algs[prio], buf);
977	up_read(sem: &zram->init_lock);
978
979	return sz;
980	}
981
982	static int __comp_algorithm_store(struct zram *zram, u32 prio, const char *buf)
983	{
984	char *compressor;
985	size_t sz;
986
987	sz = strlen(buf);
988	if (sz >= CRYPTO_MAX_ALG_NAME)
989	return -E2BIG;
990
991	compressor = kstrdup(s: buf, GFP_KERNEL);
992	if (!compressor)
993	return -ENOMEM;
994
995	/* ignore trailing newline */
996	if (sz > 0 && compressor[sz - 1] == '\n')
997	compressor[sz - 1] = 0x00;
998
999	if (!zcomp_available_algorithm(comp: compressor)) {
1000	kfree(objp: compressor);
1001	return -EINVAL;
1002	}
1003
1004	down_write(sem: &zram->init_lock);
1005	if (init_done(zram)) {
1006	up_write(sem: &zram->init_lock);
1007	kfree(objp: compressor);
1008	pr_info("Can't change algorithm for initialized device\n");
1009	return -EBUSY;
1010	}
1011
1012	comp_algorithm_set(zram, prio, alg: compressor);
1013	up_write(sem: &zram->init_lock);
1014	return 0;
1015	}
1016
1017	static ssize_t comp_algorithm_show(struct device *dev,
1018	struct device_attribute *attr,
1019	char *buf)
1020	{
1021	struct zram *zram = dev_to_zram(dev);
1022
1023	return __comp_algorithm_show(zram, ZRAM_PRIMARY_COMP, buf);
1024	}
1025
1026	static ssize_t comp_algorithm_store(struct device *dev,
1027	struct device_attribute *attr,
1028	const char *buf,
1029	size_t len)
1030	{
1031	struct zram *zram = dev_to_zram(dev);
1032	int ret;
1033
1034	ret = __comp_algorithm_store(zram, ZRAM_PRIMARY_COMP, buf);
1035	return ret ? ret : len;
1036	}
1037
1038	#ifdef CONFIG_ZRAM_MULTI_COMP
1039	static ssize_t recomp_algorithm_show(struct device *dev,
1040	struct device_attribute *attr,
1041	char *buf)
1042	{
1043	struct zram *zram = dev_to_zram(dev);
1044	ssize_t sz = 0;
1045	u32 prio;
1046
1047	for (prio = ZRAM_SECONDARY_COMP; prio < ZRAM_MAX_COMPS; prio++) {
1048	if (!zram->comp_algs[prio])
1049	continue;
1050
1051	sz += scnprintf(buf: buf + sz, PAGE_SIZE - sz - 2, fmt: "#%d: ", prio);
1052	sz += __comp_algorithm_show(zram, prio, buf: buf + sz);
1053	}
1054
1055	return sz;
1056	}
1057
1058	static ssize_t recomp_algorithm_store(struct device *dev,
1059	struct device_attribute *attr,
1060	const char *buf,
1061	size_t len)
1062	{
1063	struct zram *zram = dev_to_zram(dev);
1064	int prio = ZRAM_SECONDARY_COMP;
1065	char *args, *param, *val;
1066	char *alg = NULL;
1067	int ret;
1068
1069	args = skip_spaces(buf);
1070	while (*args) {
1071	args = next_arg(args, param: &param, val: &val);
1072
1073	if (!val || !*val)
1074	return -EINVAL;
1075
1076	if (!strcmp(param, "algo")) {
1077	alg = val;
1078	continue;
1079	}
1080
1081	if (!strcmp(param, "priority")) {
1082	ret = kstrtoint(s: val, base: 10, res: &prio);
1083	if (ret)
1084	return ret;
1085	continue;
1086	}
1087	}
1088
1089	if (!alg)
1090	return -EINVAL;
1091
1092	if (prio < ZRAM_SECONDARY_COMP || prio >= ZRAM_MAX_COMPS)
1093	return -EINVAL;
1094
1095	ret = __comp_algorithm_store(zram, prio, buf: alg);
1096	return ret ? ret : len;
1097	}
1098	#endif
1099
1100	static ssize_t compact_store(struct device *dev,
1101	struct device_attribute *attr, const char *buf, size_t len)
1102	{
1103	struct zram *zram = dev_to_zram(dev);
1104
1105	down_read(sem: &zram->init_lock);
1106	if (!init_done(zram)) {
1107	up_read(sem: &zram->init_lock);
1108	return -EINVAL;
1109	}
1110
1111	zs_compact(pool: zram->mem_pool);
1112	up_read(sem: &zram->init_lock);
1113
1114	return len;
1115	}
1116
1117	static ssize_t io_stat_show(struct device *dev,
1118	struct device_attribute *attr, char *buf)
1119	{
1120	struct zram *zram = dev_to_zram(dev);
1121	ssize_t ret;
1122
1123	down_read(sem: &zram->init_lock);
1124	ret = scnprintf(buf, PAGE_SIZE,
1125	fmt: "%8llu %8llu 0 %8llu\n",
1126	(u64)atomic64_read(v: &zram->stats.failed_reads),
1127	(u64)atomic64_read(v: &zram->stats.failed_writes),
1128	(u64)atomic64_read(v: &zram->stats.notify_free));
1129	up_read(sem: &zram->init_lock);
1130
1131	return ret;
1132	}
1133
1134	static ssize_t mm_stat_show(struct device *dev,
1135	struct device_attribute *attr, char *buf)
1136	{
1137	struct zram *zram = dev_to_zram(dev);
1138	struct zs_pool_stats pool_stats;
1139	u64 orig_size, mem_used = 0;
1140	long max_used;
1141	ssize_t ret;
1142
1143	memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1144
1145	down_read(sem: &zram->init_lock);
1146	if (init_done(zram)) {
1147	mem_used = zs_get_total_pages(pool: zram->mem_pool);
1148	zs_pool_stats(pool: zram->mem_pool, stats: &pool_stats);
1149	}
1150
1151	orig_size = atomic64_read(v: &zram->stats.pages_stored);
1152	max_used = atomic_long_read(v: &zram->stats.max_used_pages);
1153
1154	ret = scnprintf(buf, PAGE_SIZE,
1155	fmt: "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1156	orig_size << PAGE_SHIFT,
1157	(u64)atomic64_read(v: &zram->stats.compr_data_size),
1158	mem_used << PAGE_SHIFT,
1159	zram->limit_pages << PAGE_SHIFT,
1160	max_used << PAGE_SHIFT,
1161	(u64)atomic64_read(v: &zram->stats.same_pages),
1162	atomic_long_read(v: &pool_stats.pages_compacted),
1163	(u64)atomic64_read(v: &zram->stats.huge_pages),
1164	(u64)atomic64_read(v: &zram->stats.huge_pages_since));
1165	up_read(sem: &zram->init_lock);
1166
1167	return ret;
1168	}
1169
1170	#ifdef CONFIG_ZRAM_WRITEBACK
1171	#define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1172	static ssize_t bd_stat_show(struct device *dev,
1173	struct device_attribute *attr, char *buf)
1174	{
1175	struct zram *zram = dev_to_zram(dev);
1176	ssize_t ret;
1177
1178	down_read(sem: &zram->init_lock);
1179	ret = scnprintf(buf, PAGE_SIZE,
1180	fmt: "%8llu %8llu %8llu\n",
1181	FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1182	FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1183	FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1184	up_read(sem: &zram->init_lock);
1185
1186	return ret;
1187	}
1188	#endif
1189
1190	static ssize_t debug_stat_show(struct device *dev,
1191	struct device_attribute *attr, char *buf)
1192	{
1193	int version = 1;
1194	struct zram *zram = dev_to_zram(dev);
1195	ssize_t ret;
1196
1197	down_read(sem: &zram->init_lock);
1198	ret = scnprintf(buf, PAGE_SIZE,
1199	fmt: "version: %d\n%8llu %8llu\n",
1200	version,
1201	(u64)atomic64_read(v: &zram->stats.writestall),
1202	(u64)atomic64_read(v: &zram->stats.miss_free));
1203	up_read(sem: &zram->init_lock);
1204
1205	return ret;
1206	}
1207
1208	static DEVICE_ATTR_RO(io_stat);
1209	static DEVICE_ATTR_RO(mm_stat);
1210	#ifdef CONFIG_ZRAM_WRITEBACK
1211	static DEVICE_ATTR_RO(bd_stat);
1212	#endif
1213	static DEVICE_ATTR_RO(debug_stat);
1214
1215	static void zram_meta_free(struct zram *zram, u64 disksize)
1216	{
1217	size_t num_pages = disksize >> PAGE_SHIFT;
1218	size_t index;
1219
1220	/* Free all pages that are still in this zram device */
1221	for (index = 0; index < num_pages; index++)
1222	zram_free_page(zram, index);
1223
1224	zs_destroy_pool(pool: zram->mem_pool);
1225	vfree(addr: zram->table);
1226	}
1227
1228	static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1229	{
1230	size_t num_pages;
1231
1232	num_pages = disksize >> PAGE_SHIFT;
1233	zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1234	if (!zram->table)
1235	return false;
1236
1237	zram->mem_pool = zs_create_pool(name: zram->disk->disk_name);
1238	if (!zram->mem_pool) {
1239	vfree(addr: zram->table);
1240	return false;
1241	}
1242
1243	if (!huge_class_size)
1244	huge_class_size = zs_huge_class_size(pool: zram->mem_pool);
1245	return true;
1246	}
1247
1248	/*
1249	* To protect concurrent access to the same index entry,
1250	* caller should hold this table index entry's bit_spinlock to
1251	* indicate this index entry is accessing.
1252	*/
1253	static void zram_free_page(struct zram *zram, size_t index)
1254	{
1255	unsigned long handle;
1256
1257	#ifdef CONFIG_ZRAM_MEMORY_TRACKING
1258	zram->table[index].ac_time = 0;
1259	#endif
1260	if (zram_test_flag(zram, index, flag: ZRAM_IDLE))
1261	zram_clear_flag(zram, index, flag: ZRAM_IDLE);
1262
1263	if (zram_test_flag(zram, index, flag: ZRAM_HUGE)) {
1264	zram_clear_flag(zram, index, flag: ZRAM_HUGE);
1265	atomic64_dec(v: &zram->stats.huge_pages);
1266	}
1267
1268	if (zram_test_flag(zram, index, flag: ZRAM_INCOMPRESSIBLE))
1269	zram_clear_flag(zram, index, flag: ZRAM_INCOMPRESSIBLE);
1270
1271	zram_set_priority(zram, index, prio: 0);
1272
1273	if (zram_test_flag(zram, index, flag: ZRAM_WB)) {
1274	zram_clear_flag(zram, index, flag: ZRAM_WB);
1275	free_block_bdev(zram, blk_idx: zram_get_element(zram, index));
1276	goto out;
1277	}
1278
1279	/*
1280	* No memory is allocated for same element filled pages.
1281	* Simply clear same page flag.
1282	*/
1283	if (zram_test_flag(zram, index, flag: ZRAM_SAME)) {
1284	zram_clear_flag(zram, index, flag: ZRAM_SAME);
1285	atomic64_dec(v: &zram->stats.same_pages);
1286	goto out;
1287	}
1288
1289	handle = zram_get_handle(zram, index);
1290	if (!handle)
1291	return;
1292
1293	zs_free(pool: zram->mem_pool, obj: handle);
1294
1295	atomic64_sub(i: zram_get_obj_size(zram, index),
1296	v: &zram->stats.compr_data_size);
1297	out:
1298	atomic64_dec(v: &zram->stats.pages_stored);
1299	zram_set_handle(zram, index, handle: 0);
1300	zram_set_obj_size(zram, index, size: 0);
1301	WARN_ON_ONCE(zram->table[index].flags &
1302	~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1303	}
1304
1305	/*
1306	* Reads (decompresses if needed) a page from zspool (zsmalloc).
1307	* Corresponding ZRAM slot should be locked.
1308	*/
1309	static int zram_read_from_zspool(struct zram *zram, struct page *page,
1310	u32 index)
1311	{
1312	struct zcomp_strm *zstrm;
1313	unsigned long handle;
1314	unsigned int size;
1315	void *src, *dst;
1316	u32 prio;
1317	int ret;
1318
1319	handle = zram_get_handle(zram, index);
1320	if (!handle || zram_test_flag(zram, index, flag: ZRAM_SAME)) {
1321	unsigned long value;
1322	void *mem;
1323
1324	value = handle ? zram_get_element(zram, index) : 0;
1325	mem = kmap_atomic(page);
1326	zram_fill_page(ptr: mem, PAGE_SIZE, value);
1327	kunmap_atomic(mem);
1328	return 0;
1329	}
1330
1331	size = zram_get_obj_size(zram, index);
1332
1333	if (size != PAGE_SIZE) {
1334	prio = zram_get_priority(zram, index);
1335	zstrm = zcomp_stream_get(comp: zram->comps[prio]);
1336	}
1337
1338	src = zs_map_object(pool: zram->mem_pool, handle, mm: ZS_MM_RO);
1339	if (size == PAGE_SIZE) {
1340	dst = kmap_atomic(page);
1341	memcpy(dst, src, PAGE_SIZE);
1342	kunmap_atomic(dst);
1343	ret = 0;
1344	} else {
1345	dst = kmap_atomic(page);
1346	ret = zcomp_decompress(zstrm, src, src_len: size, dst);
1347	kunmap_atomic(dst);
1348	zcomp_stream_put(comp: zram->comps[prio]);
1349	}
1350	zs_unmap_object(pool: zram->mem_pool, handle);
1351	return ret;
1352	}
1353
1354	static int zram_read_page(struct zram *zram, struct page *page, u32 index,
1355	struct bio *parent)
1356	{
1357	int ret;
1358
1359	zram_slot_lock(zram, index);
1360	if (!zram_test_flag(zram, index, flag: ZRAM_WB)) {
1361	/* Slot should be locked through out the function call */
1362	ret = zram_read_from_zspool(zram, page, index);
1363	zram_slot_unlock(zram, index);
1364	} else {
1365	/*
1366	* The slot should be unlocked before reading from the backing
1367	* device.
1368	*/
1369	zram_slot_unlock(zram, index);
1370
1371	ret = read_from_bdev(zram, page, entry: zram_get_element(zram, index),
1372	parent);
1373	}
1374
1375	/* Should NEVER happen. Return bio error if it does. */
1376	if (WARN_ON(ret < 0))
1377	pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1378
1379	return ret;
1380	}
1381
1382	/*
1383	* Use a temporary buffer to decompress the page, as the decompressor
1384	* always expects a full page for the output.
1385	*/
1386	static int zram_bvec_read_partial(struct zram *zram, struct bio_vec *bvec,
1387	u32 index, int offset)
1388	{
1389	struct page *page = alloc_page(GFP_NOIO);
1390	int ret;
1391
1392	if (!page)
1393	return -ENOMEM;
1394	ret = zram_read_page(zram, page, index, NULL);
1395	if (likely(!ret))
1396	memcpy_to_bvec(bvec, page_address(page) + offset);
1397	__free_page(page);
1398	return ret;
1399	}
1400
1401	static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1402	u32 index, int offset, struct bio *bio)
1403	{
1404	if (is_partial_io(bvec))
1405	return zram_bvec_read_partial(zram, bvec, index, offset);
1406	return zram_read_page(zram, page: bvec->bv_page, index, parent: bio);
1407	}
1408
1409	static int zram_write_page(struct zram *zram, struct page *page, u32 index)
1410	{
1411	int ret = 0;
1412	unsigned long alloced_pages;
1413	unsigned long handle = -ENOMEM;
1414	unsigned int comp_len = 0;
1415	void *src, *dst, *mem;
1416	struct zcomp_strm *zstrm;
1417	unsigned long element = 0;
1418	enum zram_pageflags flags = 0;
1419
1420	mem = kmap_atomic(page);
1421	if (page_same_filled(ptr: mem, element: &element)) {
1422	kunmap_atomic(mem);
1423	/* Free memory associated with this sector now. */
1424	flags = ZRAM_SAME;
1425	atomic64_inc(v: &zram->stats.same_pages);
1426	goto out;
1427	}
1428	kunmap_atomic(mem);
1429
1430	compress_again:
1431	zstrm = zcomp_stream_get(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1432	src = kmap_atomic(page);
1433	ret = zcomp_compress(zstrm, src, dst_len: &comp_len);
1434	kunmap_atomic(src);
1435
1436	if (unlikely(ret)) {
1437	zcomp_stream_put(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1438	pr_err("Compression failed! err=%d\n", ret);
1439	zs_free(pool: zram->mem_pool, obj: handle);
1440	return ret;
1441	}
1442
1443	if (comp_len >= huge_class_size)
1444	comp_len = PAGE_SIZE;
1445	/*
1446	* handle allocation has 2 paths:
1447	* a) fast path is executed with preemption disabled (for
1448	* per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1449	* since we can't sleep;
1450	* b) slow path enables preemption and attempts to allocate
1451	* the page with __GFP_DIRECT_RECLAIM bit set. we have to
1452	* put per-cpu compression stream and, thus, to re-do
1453	* the compression once handle is allocated.
1454	*
1455	* if we have a 'non-null' handle here then we are coming
1456	* from the slow path and handle has already been allocated.
1457	*/
1458	if (IS_ERR_VALUE(handle))
1459	handle = zs_malloc(pool: zram->mem_pool, size: comp_len,
1460	__GFP_KSWAPD_RECLAIM |
1461	__GFP_NOWARN |
1462	__GFP_HIGHMEM |
1463	__GFP_MOVABLE);
1464	if (IS_ERR_VALUE(handle)) {
1465	zcomp_stream_put(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1466	atomic64_inc(v: &zram->stats.writestall);
1467	handle = zs_malloc(pool: zram->mem_pool, size: comp_len,
1468	GFP_NOIO | __GFP_HIGHMEM |
1469	__GFP_MOVABLE);
1470	if (IS_ERR_VALUE(handle))
1471	return PTR_ERR(ptr: (void *)handle);
1472
1473	if (comp_len != PAGE_SIZE)
1474	goto compress_again;
1475	/*
1476	* If the page is not compressible, you need to acquire the
1477	* lock and execute the code below. The zcomp_stream_get()
1478	* call is needed to disable the cpu hotplug and grab the
1479	* zstrm buffer back. It is necessary that the dereferencing
1480	* of the zstrm variable below occurs correctly.
1481	*/
1482	zstrm = zcomp_stream_get(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1483	}
1484
1485	alloced_pages = zs_get_total_pages(pool: zram->mem_pool);
1486	update_used_max(zram, pages: alloced_pages);
1487
1488	if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1489	zcomp_stream_put(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1490	zs_free(pool: zram->mem_pool, obj: handle);
1491	return -ENOMEM;
1492	}
1493
1494	dst = zs_map_object(pool: zram->mem_pool, handle, mm: ZS_MM_WO);
1495
1496	src = zstrm->buffer;
1497	if (comp_len == PAGE_SIZE)
1498	src = kmap_atomic(page);
1499	memcpy(dst, src, comp_len);
1500	if (comp_len == PAGE_SIZE)
1501	kunmap_atomic(src);
1502
1503	zcomp_stream_put(comp: zram->comps[ZRAM_PRIMARY_COMP]);
1504	zs_unmap_object(pool: zram->mem_pool, handle);
1505	atomic64_add(i: comp_len, v: &zram->stats.compr_data_size);
1506	out:
1507	/*
1508	* Free memory associated with this sector
1509	* before overwriting unused sectors.
1510	*/
1511	zram_slot_lock(zram, index);
1512	zram_free_page(zram, index);
1513
1514	if (comp_len == PAGE_SIZE) {
1515	zram_set_flag(zram, index, flag: ZRAM_HUGE);
1516	atomic64_inc(v: &zram->stats.huge_pages);
1517	atomic64_inc(v: &zram->stats.huge_pages_since);
1518	}
1519
1520	if (flags) {
1521	zram_set_flag(zram, index, flag: flags);
1522	zram_set_element(zram, index, element);
1523	} else {
1524	zram_set_handle(zram, index, handle);
1525	zram_set_obj_size(zram, index, size: comp_len);
1526	}
1527	zram_slot_unlock(zram, index);
1528
1529	/* Update stats */
1530	atomic64_inc(v: &zram->stats.pages_stored);
1531	return ret;
1532	}
1533
1534	/*
1535	* This is a partial IO. Read the full page before writing the changes.
1536	*/
1537	static int zram_bvec_write_partial(struct zram *zram, struct bio_vec *bvec,
1538	u32 index, int offset, struct bio *bio)
1539	{
1540	struct page *page = alloc_page(GFP_NOIO);
1541	int ret;
1542
1543	if (!page)
1544	return -ENOMEM;
1545
1546	ret = zram_read_page(zram, page, index, parent: bio);
1547	if (!ret) {
1548	memcpy_from_bvec(page_address(page) + offset, bvec);
1549	ret = zram_write_page(zram, page, index);
1550	}
1551	__free_page(page);
1552	return ret;
1553	}
1554
1555	static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1556	u32 index, int offset, struct bio *bio)
1557	{
1558	if (is_partial_io(bvec))
1559	return zram_bvec_write_partial(zram, bvec, index, offset, bio);
1560	return zram_write_page(zram, page: bvec->bv_page, index);
1561	}
1562
1563	#ifdef CONFIG_ZRAM_MULTI_COMP
1564	/*
1565	* This function will decompress (unless it's ZRAM_HUGE) the page and then
1566	* attempt to compress it using provided compression algorithm priority
1567	* (which is potentially more effective).
1568	*
1569	* Corresponding ZRAM slot should be locked.
1570	*/
1571	static int zram_recompress(struct zram *zram, u32 index, struct page *page,
1572	u32 threshold, u32 prio, u32 prio_max)
1573	{
1574	struct zcomp_strm *zstrm = NULL;
1575	unsigned long handle_old;
1576	unsigned long handle_new;
1577	unsigned int comp_len_old;
1578	unsigned int comp_len_new;
1579	unsigned int class_index_old;
1580	unsigned int class_index_new;
1581	u32 num_recomps = 0;
1582	void *src, *dst;
1583	int ret;
1584
1585	handle_old = zram_get_handle(zram, index);
1586	if (!handle_old)
1587	return -EINVAL;
1588
1589	comp_len_old = zram_get_obj_size(zram, index);
1590	/*
1591	* Do not recompress objects that are already "small enough".
1592	*/
1593	if (comp_len_old < threshold)
1594	return 0;
1595
1596	ret = zram_read_from_zspool(zram, page, index);
1597	if (ret)
1598	return ret;
1599
1600	class_index_old = zs_lookup_class_index(pool: zram->mem_pool, size: comp_len_old);
1601	/*
1602	* Iterate the secondary comp algorithms list (in order of priority)
1603	* and try to recompress the page.
1604	*/
1605	for (; prio < prio_max; prio++) {
1606	if (!zram->comps[prio])
1607	continue;
1608
1609	/*
1610	* Skip if the object is already re-compressed with a higher
1611	* priority algorithm (or same algorithm).
1612	*/
1613	if (prio <= zram_get_priority(zram, index))
1614	continue;
1615
1616	num_recomps++;
1617	zstrm = zcomp_stream_get(comp: zram->comps[prio]);
1618	src = kmap_atomic(page);
1619	ret = zcomp_compress(zstrm, src, dst_len: &comp_len_new);
1620	kunmap_atomic(src);
1621
1622	if (ret) {
1623	zcomp_stream_put(comp: zram->comps[prio]);
1624	return ret;
1625	}
1626
1627	class_index_new = zs_lookup_class_index(pool: zram->mem_pool,
1628	size: comp_len_new);
1629
1630	/* Continue until we make progress */
1631	if (class_index_new >= class_index_old ||
1632	(threshold && comp_len_new >= threshold)) {
1633	zcomp_stream_put(comp: zram->comps[prio]);
1634	continue;
1635	}
1636
1637	/* Recompression was successful so break out */
1638	break;
1639	}
1640
1641	/*
1642	* We did not try to recompress, e.g. when we have only one
1643	* secondary algorithm and the page is already recompressed
1644	* using that algorithm
1645	*/
1646	if (!zstrm)
1647	return 0;
1648
1649	if (class_index_new >= class_index_old) {
1650	/*
1651	* Secondary algorithms failed to re-compress the page
1652	* in a way that would save memory, mark the object as
1653	* incompressible so that we will not try to compress
1654	* it again.
1655	*
1656	* We need to make sure that all secondary algorithms have
1657	* failed, so we test if the number of recompressions matches
1658	* the number of active secondary algorithms.
1659	*/
1660	if (num_recomps == zram->num_active_comps - 1)
1661	zram_set_flag(zram, index, flag: ZRAM_INCOMPRESSIBLE);
1662	return 0;
1663	}
1664
1665	/* Successful recompression but above threshold */
1666	if (threshold && comp_len_new >= threshold)
1667	return 0;
1668
1669	/*
1670	* No direct reclaim (slow path) for handle allocation and no
1671	* re-compression attempt (unlike in zram_write_bvec()) since
1672	* we already have stored that object in zsmalloc. If we cannot
1673	* alloc memory for recompressed object then we bail out and
1674	* simply keep the old (existing) object in zsmalloc.
1675	*/
1676	handle_new = zs_malloc(pool: zram->mem_pool, size: comp_len_new,
1677	__GFP_KSWAPD_RECLAIM |
1678	__GFP_NOWARN |
1679	__GFP_HIGHMEM |
1680	__GFP_MOVABLE);
1681	if (IS_ERR_VALUE(handle_new)) {
1682	zcomp_stream_put(comp: zram->comps[prio]);
1683	return PTR_ERR(ptr: (void *)handle_new);
1684	}
1685
1686	dst = zs_map_object(pool: zram->mem_pool, handle: handle_new, mm: ZS_MM_WO);
1687	memcpy(dst, zstrm->buffer, comp_len_new);
1688	zcomp_stream_put(comp: zram->comps[prio]);
1689
1690	zs_unmap_object(pool: zram->mem_pool, handle: handle_new);
1691
1692	zram_free_page(zram, index);
1693	zram_set_handle(zram, index, handle: handle_new);
1694	zram_set_obj_size(zram, index, size: comp_len_new);
1695	zram_set_priority(zram, index, prio);
1696
1697	atomic64_add(i: comp_len_new, v: &zram->stats.compr_data_size);
1698	atomic64_inc(v: &zram->stats.pages_stored);
1699
1700	return 0;
1701	}
1702
1703	#define RECOMPRESS_IDLE (1 << 0)
1704	#define RECOMPRESS_HUGE (1 << 1)
1705
1706	static ssize_t recompress_store(struct device *dev,
1707	struct device_attribute *attr,
1708	const char *buf, size_t len)
1709	{
1710	u32 prio = ZRAM_SECONDARY_COMP, prio_max = ZRAM_MAX_COMPS;
1711	struct zram *zram = dev_to_zram(dev);
1712	unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
1713	char *args, *param, *val, *algo = NULL;
1714	u32 mode = 0, threshold = 0;
1715	unsigned long index;
1716	struct page *page;
1717	ssize_t ret;
1718
1719	args = skip_spaces(buf);
1720	while (*args) {
1721	args = next_arg(args, param: &param, val: &val);
1722
1723	if (!val || !*val)
1724	return -EINVAL;
1725
1726	if (!strcmp(param, "type")) {
1727	if (!strcmp(val, "idle"))
1728	mode = RECOMPRESS_IDLE;
1729	if (!strcmp(val, "huge"))
1730	mode = RECOMPRESS_HUGE;
1731	if (!strcmp(val, "huge_idle"))
1732	mode = RECOMPRESS_IDLE | RECOMPRESS_HUGE;
1733	continue;
1734	}
1735
1736	if (!strcmp(param, "threshold")) {
1737	/*
1738	* We will re-compress only idle objects equal or
1739	* greater in size than watermark.
1740	*/
1741	ret = kstrtouint(s: val, base: 10, res: &threshold);
1742	if (ret)
1743	return ret;
1744	continue;
1745	}
1746
1747	if (!strcmp(param, "algo")) {
1748	algo = val;
1749	continue;
1750	}
1751	}
1752
1753	if (threshold >= huge_class_size)
1754	return -EINVAL;
1755
1756	down_read(sem: &zram->init_lock);
1757	if (!init_done(zram)) {
1758	ret = -EINVAL;
1759	goto release_init_lock;
1760	}
1761
1762	if (algo) {
1763	bool found = false;
1764
1765	for (; prio < ZRAM_MAX_COMPS; prio++) {
1766	if (!zram->comp_algs[prio])
1767	continue;
1768
1769	if (!strcmp(zram->comp_algs[prio], algo)) {
1770	prio_max = min(prio + 1, ZRAM_MAX_COMPS);
1771	found = true;
1772	break;
1773	}
1774	}
1775
1776	if (!found) {
1777	ret = -EINVAL;
1778	goto release_init_lock;
1779	}
1780	}
1781
1782	page = alloc_page(GFP_KERNEL);
1783	if (!page) {
1784	ret = -ENOMEM;
1785	goto release_init_lock;
1786	}
1787
1788	ret = len;
1789	for (index = 0; index < nr_pages; index++) {
1790	int err = 0;
1791
1792	zram_slot_lock(zram, index);
1793
1794	if (!zram_allocated(zram, index))
1795	goto next;
1796
1797	if (mode & RECOMPRESS_IDLE &&
1798	!zram_test_flag(zram, index, flag: ZRAM_IDLE))
1799	goto next;
1800
1801	if (mode & RECOMPRESS_HUGE &&
1802	!zram_test_flag(zram, index, flag: ZRAM_HUGE))
1803	goto next;
1804
1805	if (zram_test_flag(zram, index, flag: ZRAM_WB) ||
1806	zram_test_flag(zram, index, flag: ZRAM_UNDER_WB) ||
1807	zram_test_flag(zram, index, flag: ZRAM_SAME) ||
1808	zram_test_flag(zram, index, flag: ZRAM_INCOMPRESSIBLE))
1809	goto next;
1810
1811	err = zram_recompress(zram, index, page, threshold,
1812	prio, prio_max);
1813	next:
1814	zram_slot_unlock(zram, index);
1815	if (err) {
1816	ret = err;
1817	break;
1818	}
1819
1820	cond_resched();
1821	}
1822
1823	__free_page(page);
1824
1825	release_init_lock:
1826	up_read(sem: &zram->init_lock);
1827	return ret;
1828	}
1829	#endif
1830
1831	static void zram_bio_discard(struct zram *zram, struct bio *bio)
1832	{
1833	size_t n = bio->bi_iter.bi_size;
1834	u32 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1835	u32 offset = (bio->bi_iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1836	SECTOR_SHIFT;
1837
1838	/*
1839	* zram manages data in physical block size units. Because logical block
1840	* size isn't identical with physical block size on some arch, we
1841	* could get a discard request pointing to a specific offset within a
1842	* certain physical block. Although we can handle this request by
1843	* reading that physiclal block and decompressing and partially zeroing
1844	* and re-compressing and then re-storing it, this isn't reasonable
1845	* because our intent with a discard request is to save memory. So
1846	* skipping this logical block is appropriate here.
1847	*/
1848	if (offset) {
1849	if (n <= (PAGE_SIZE - offset))
1850	return;
1851
1852	n -= (PAGE_SIZE - offset);
1853	index++;
1854	}
1855
1856	while (n >= PAGE_SIZE) {
1857	zram_slot_lock(zram, index);
1858	zram_free_page(zram, index);
1859	zram_slot_unlock(zram, index);
1860	atomic64_inc(v: &zram->stats.notify_free);
1861	index++;
1862	n -= PAGE_SIZE;
1863	}
1864
1865	bio_endio(bio);
1866	}
1867
1868	static void zram_bio_read(struct zram *zram, struct bio *bio)
1869	{
1870	unsigned long start_time = bio_start_io_acct(bio);
1871	struct bvec_iter iter = bio->bi_iter;
1872
1873	do {
1874	u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1875	u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1876	SECTOR_SHIFT;
1877	struct bio_vec bv = bio_iter_iovec(bio, iter);
1878
1879	bv.bv_len = min_t(u32, bv.bv_len, PAGE_SIZE - offset);
1880
1881	if (zram_bvec_read(zram, bvec: &bv, index, offset, bio) < 0) {
1882	atomic64_inc(v: &zram->stats.failed_reads);
1883	bio->bi_status = BLK_STS_IOERR;
1884	break;
1885	}
1886	flush_dcache_page(page: bv.bv_page);
1887
1888	zram_slot_lock(zram, index);
1889	zram_accessed(zram, index);
1890	zram_slot_unlock(zram, index);
1891
1892	bio_advance_iter_single(bio, iter: &iter, bytes: bv.bv_len);
1893	} while (iter.bi_size);
1894
1895	bio_end_io_acct(bio, start_time);
1896	bio_endio(bio);
1897	}
1898
1899	static void zram_bio_write(struct zram *zram, struct bio *bio)
1900	{
1901	unsigned long start_time = bio_start_io_acct(bio);
1902	struct bvec_iter iter = bio->bi_iter;
1903
1904	do {
1905	u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1906	u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1907	SECTOR_SHIFT;
1908	struct bio_vec bv = bio_iter_iovec(bio, iter);
1909
1910	bv.bv_len = min_t(u32, bv.bv_len, PAGE_SIZE - offset);
1911
1912	if (zram_bvec_write(zram, bvec: &bv, index, offset, bio) < 0) {
1913	atomic64_inc(v: &zram->stats.failed_writes);
1914	bio->bi_status = BLK_STS_IOERR;
1915	break;
1916	}
1917
1918	zram_slot_lock(zram, index);
1919	zram_accessed(zram, index);
1920	zram_slot_unlock(zram, index);
1921
1922	bio_advance_iter_single(bio, iter: &iter, bytes: bv.bv_len);
1923	} while (iter.bi_size);
1924
1925	bio_end_io_acct(bio, start_time);
1926	bio_endio(bio);
1927	}
1928
1929	/*
1930	* Handler function for all zram I/O requests.
1931	*/
1932	static void zram_submit_bio(struct bio *bio)
1933	{
1934	struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1935
1936	switch (bio_op(bio)) {
1937	case REQ_OP_READ:
1938	zram_bio_read(zram, bio);
1939	break;
1940	case REQ_OP_WRITE:
1941	zram_bio_write(zram, bio);
1942	break;
1943	case REQ_OP_DISCARD:
1944	case REQ_OP_WRITE_ZEROES:
1945	zram_bio_discard(zram, bio);
1946	break;
1947	default:
1948	WARN_ON_ONCE(1);
1949	bio_endio(bio);
1950	}
1951	}
1952
1953	static void zram_slot_free_notify(struct block_device *bdev,
1954	unsigned long index)
1955	{
1956	struct zram *zram;
1957
1958	zram = bdev->bd_disk->private_data;
1959
1960	atomic64_inc(v: &zram->stats.notify_free);
1961	if (!zram_slot_trylock(zram, index)) {
1962	atomic64_inc(v: &zram->stats.miss_free);
1963	return;
1964	}
1965
1966	zram_free_page(zram, index);
1967	zram_slot_unlock(zram, index);
1968	}
1969
1970	static void zram_destroy_comps(struct zram *zram)
1971	{
1972	u32 prio;
1973
1974	for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
1975	struct zcomp *comp = zram->comps[prio];
1976
1977	zram->comps[prio] = NULL;
1978	if (!comp)
1979	continue;
1980	zcomp_destroy(comp);
1981	zram->num_active_comps--;
1982	}
1983	}
1984
1985	static void zram_reset_device(struct zram *zram)
1986	{
1987	down_write(sem: &zram->init_lock);
1988
1989	zram->limit_pages = 0;
1990
1991	if (!init_done(zram)) {
1992	up_write(sem: &zram->init_lock);
1993	return;
1994	}
1995
1996	set_capacity_and_notify(disk: zram->disk, size: 0);
1997	part_stat_set_all(part: zram->disk->part0, value: 0);
1998
1999	/* I/O operation under all of CPU are done so let's free */
2000	zram_meta_free(zram, disksize: zram->disksize);
2001	zram->disksize = 0;
2002	zram_destroy_comps(zram);
2003	memset(&zram->stats, 0, sizeof(zram->stats));
2004	reset_bdev(zram);
2005
2006	comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, alg: default_compressor);
2007	up_write(sem: &zram->init_lock);
2008	}
2009
2010	static ssize_t disksize_store(struct device *dev,
2011	struct device_attribute *attr, const char *buf, size_t len)
2012	{
2013	u64 disksize;
2014	struct zcomp *comp;
2015	struct zram *zram = dev_to_zram(dev);
2016	int err;
2017	u32 prio;
2018
2019	disksize = memparse(ptr: buf, NULL);
2020	if (!disksize)
2021	return -EINVAL;
2022
2023	down_write(sem: &zram->init_lock);
2024	if (init_done(zram)) {
2025	pr_info("Cannot change disksize for initialized device\n");
2026	err = -EBUSY;
2027	goto out_unlock;
2028	}
2029
2030	disksize = PAGE_ALIGN(disksize);
2031	if (!zram_meta_alloc(zram, disksize)) {
2032	err = -ENOMEM;
2033	goto out_unlock;
2034	}
2035
2036	for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
2037	if (!zram->comp_algs[prio])
2038	continue;
2039
2040	comp = zcomp_create(alg: zram->comp_algs[prio]);
2041	if (IS_ERR(ptr: comp)) {
2042	pr_err("Cannot initialise %s compressing backend\n",
2043	zram->comp_algs[prio]);
2044	err = PTR_ERR(ptr: comp);
2045	goto out_free_comps;
2046	}
2047
2048	zram->comps[prio] = comp;
2049	zram->num_active_comps++;
2050	}
2051	zram->disksize = disksize;
2052	set_capacity_and_notify(disk: zram->disk, size: zram->disksize >> SECTOR_SHIFT);
2053	up_write(sem: &zram->init_lock);
2054
2055	return len;
2056
2057	out_free_comps:
2058	zram_destroy_comps(zram);
2059	zram_meta_free(zram, disksize);
2060	out_unlock:
2061	up_write(sem: &zram->init_lock);
2062	return err;
2063	}
2064
2065	static ssize_t reset_store(struct device *dev,
2066	struct device_attribute *attr, const char *buf, size_t len)
2067	{
2068	int ret;
2069	unsigned short do_reset;
2070	struct zram *zram;
2071	struct gendisk *disk;
2072
2073	ret = kstrtou16(s: buf, base: 10, res: &do_reset);
2074	if (ret)
2075	return ret;
2076
2077	if (!do_reset)
2078	return -EINVAL;
2079
2080	zram = dev_to_zram(dev);
2081	disk = zram->disk;
2082
2083	mutex_lock(&disk->open_mutex);
2084	/* Do not reset an active device or claimed device */
2085	if (disk_openers(disk) || zram->claim) {
2086	mutex_unlock(lock: &disk->open_mutex);
2087	return -EBUSY;
2088	}
2089
2090	/* From now on, anyone can't open /dev/zram[0-9] */
2091	zram->claim = true;
2092	mutex_unlock(lock: &disk->open_mutex);
2093
2094	/* Make sure all the pending I/O are finished */
2095	sync_blockdev(bdev: disk->part0);
2096	zram_reset_device(zram);
2097
2098	mutex_lock(&disk->open_mutex);
2099	zram->claim = false;
2100	mutex_unlock(lock: &disk->open_mutex);
2101
2102	return len;
2103	}
2104
2105	static int zram_open(struct gendisk *disk, blk_mode_t mode)
2106	{
2107	struct zram *zram = disk->private_data;
2108
2109	WARN_ON(!mutex_is_locked(&disk->open_mutex));
2110
2111	/* zram was claimed to reset so open request fails */
2112	if (zram->claim)
2113	return -EBUSY;
2114	return 0;
2115	}
2116
2117	static const struct block_device_operations zram_devops = {
2118	.open = zram_open,
2119	.submit_bio = zram_submit_bio,
2120	.swap_slot_free_notify = zram_slot_free_notify,
2121	.owner = THIS_MODULE
2122	};
2123
2124	static DEVICE_ATTR_WO(compact);
2125	static DEVICE_ATTR_RW(disksize);
2126	static DEVICE_ATTR_RO(initstate);
2127	static DEVICE_ATTR_WO(reset);
2128	static DEVICE_ATTR_WO(mem_limit);
2129	static DEVICE_ATTR_WO(mem_used_max);
2130	static DEVICE_ATTR_WO(idle);
2131	static DEVICE_ATTR_RW(max_comp_streams);
2132	static DEVICE_ATTR_RW(comp_algorithm);
2133	#ifdef CONFIG_ZRAM_WRITEBACK
2134	static DEVICE_ATTR_RW(backing_dev);
2135	static DEVICE_ATTR_WO(writeback);
2136	static DEVICE_ATTR_RW(writeback_limit);
2137	static DEVICE_ATTR_RW(writeback_limit_enable);
2138	#endif
2139	#ifdef CONFIG_ZRAM_MULTI_COMP
2140	static DEVICE_ATTR_RW(recomp_algorithm);
2141	static DEVICE_ATTR_WO(recompress);
2142	#endif
2143
2144	static struct attribute *zram_disk_attrs[] = {
2145	&dev_attr_disksize.attr,
2146	&dev_attr_initstate.attr,
2147	&dev_attr_reset.attr,
2148	&dev_attr_compact.attr,
2149	&dev_attr_mem_limit.attr,
2150	&dev_attr_mem_used_max.attr,
2151	&dev_attr_idle.attr,
2152	&dev_attr_max_comp_streams.attr,
2153	&dev_attr_comp_algorithm.attr,
2154	#ifdef CONFIG_ZRAM_WRITEBACK
2155	&dev_attr_backing_dev.attr,
2156	&dev_attr_writeback.attr,
2157	&dev_attr_writeback_limit.attr,
2158	&dev_attr_writeback_limit_enable.attr,
2159	#endif
2160	&dev_attr_io_stat.attr,
2161	&dev_attr_mm_stat.attr,
2162	#ifdef CONFIG_ZRAM_WRITEBACK
2163	&dev_attr_bd_stat.attr,
2164	#endif
2165	&dev_attr_debug_stat.attr,
2166	#ifdef CONFIG_ZRAM_MULTI_COMP
2167	&dev_attr_recomp_algorithm.attr,
2168	&dev_attr_recompress.attr,
2169	#endif
2170	NULL,
2171	};
2172
2173	ATTRIBUTE_GROUPS(zram_disk);
2174
2175	/*
2176	* Allocate and initialize new zram device. the function returns
2177	* '>= 0' device_id upon success, and negative value otherwise.
2178	*/
2179	static int zram_add(void)
2180	{
2181	struct zram *zram;
2182	int ret, device_id;
2183
2184	zram = kzalloc(size: sizeof(struct zram), GFP_KERNEL);
2185	if (!zram)
2186	return -ENOMEM;
2187
2188	ret = idr_alloc(&zram_index_idr, ptr: zram, start: 0, end: 0, GFP_KERNEL);
2189	if (ret < 0)
2190	goto out_free_dev;
2191	device_id = ret;
2192
2193	init_rwsem(&zram->init_lock);
2194	#ifdef CONFIG_ZRAM_WRITEBACK
2195	spin_lock_init(&zram->wb_limit_lock);
2196	#endif
2197
2198	/* gendisk structure */
2199	zram->disk = blk_alloc_disk(NUMA_NO_NODE);
2200	if (!zram->disk) {
2201	pr_err("Error allocating disk structure for device %d\n",
2202	device_id);
2203	ret = -ENOMEM;
2204	goto out_free_idr;
2205	}
2206
2207	zram->disk->major = zram_major;
2208	zram->disk->first_minor = device_id;
2209	zram->disk->minors = 1;
2210	zram->disk->flags |= GENHD_FL_NO_PART;
2211	zram->disk->fops = &zram_devops;
2212	zram->disk->private_data = zram;
2213	snprintf(buf: zram->disk->disk_name, size: 16, fmt: "zram%d", device_id);
2214
2215	/* Actual capacity set using sysfs (/sys/block/zram<id>/disksize */
2216	set_capacity(disk: zram->disk, size: 0);
2217	/* zram devices sort of resembles non-rotational disks */
2218	blk_queue_flag_set(QUEUE_FLAG_NONROT, q: zram->disk->queue);
2219	blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, q: zram->disk->queue);
2220
2221	/*
2222	* To ensure that we always get PAGE_SIZE aligned
2223	* and n*PAGE_SIZED sized I/O requests.
2224	*/
2225	blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
2226	blk_queue_logical_block_size(zram->disk->queue,
2227	ZRAM_LOGICAL_BLOCK_SIZE);
2228	blk_queue_io_min(q: zram->disk->queue, PAGE_SIZE);
2229	blk_queue_io_opt(q: zram->disk->queue, PAGE_SIZE);
2230	zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
2231	blk_queue_max_discard_sectors(q: zram->disk->queue, UINT_MAX);
2232
2233	/*
2234	* zram_bio_discard() will clear all logical blocks if logical block
2235	* size is identical with physical block size(PAGE_SIZE). But if it is
2236	* different, we will skip discarding some parts of logical blocks in
2237	* the part of the request range which isn't aligned to physical block
2238	* size. So we can't ensure that all discarded logical blocks are
2239	* zeroed.
2240	*/
2241	if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
2242	blk_queue_max_write_zeroes_sectors(q: zram->disk->queue, UINT_MAX);
2243
2244	blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q: zram->disk->queue);
2245	ret = device_add_disk(NULL, disk: zram->disk, groups: zram_disk_groups);
2246	if (ret)
2247	goto out_cleanup_disk;
2248
2249	comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, alg: default_compressor);
2250
2251	zram_debugfs_register(zram);
2252	pr_info("Added device: %s\n", zram->disk->disk_name);
2253	return device_id;
2254
2255	out_cleanup_disk:
2256	put_disk(disk: zram->disk);
2257	out_free_idr:
2258	idr_remove(&zram_index_idr, id: device_id);
2259	out_free_dev:
2260	kfree(objp: zram);
2261	return ret;
2262	}
2263
2264	static int zram_remove(struct zram *zram)
2265	{
2266	bool claimed;
2267
2268	mutex_lock(&zram->disk->open_mutex);
2269	if (disk_openers(disk: zram->disk)) {
2270	mutex_unlock(lock: &zram->disk->open_mutex);
2271	return -EBUSY;
2272	}
2273
2274	claimed = zram->claim;
2275	if (!claimed)
2276	zram->claim = true;
2277	mutex_unlock(lock: &zram->disk->open_mutex);
2278
2279	zram_debugfs_unregister(zram);
2280
2281	if (claimed) {
2282	/*
2283	* If we were claimed by reset_store(), del_gendisk() will
2284	* wait until reset_store() is done, so nothing need to do.
2285	*/
2286	;
2287	} else {
2288	/* Make sure all the pending I/O are finished */
2289	sync_blockdev(bdev: zram->disk->part0);
2290	zram_reset_device(zram);
2291	}
2292
2293	pr_info("Removed device: %s\n", zram->disk->disk_name);
2294
2295	del_gendisk(gp: zram->disk);
2296
2297	/* del_gendisk drains pending reset_store */
2298	WARN_ON_ONCE(claimed && zram->claim);
2299
2300	/*
2301	* disksize_store() may be called in between zram_reset_device()
2302	* and del_gendisk(), so run the last reset to avoid leaking
2303	* anything allocated with disksize_store()
2304	*/
2305	zram_reset_device(zram);
2306
2307	put_disk(disk: zram->disk);
2308	kfree(objp: zram);
2309	return 0;
2310	}
2311
2312	/* zram-control sysfs attributes */
2313
2314	/*
2315	* NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2316	* sense that reading from this file does alter the state of your system -- it
2317	* creates a new un-initialized zram device and returns back this device's
2318	* device_id (or an error code if it fails to create a new device).
2319	*/
2320	static ssize_t hot_add_show(const struct class *class,
2321	const struct class_attribute *attr,
2322	char *buf)
2323	{
2324	int ret;
2325
2326	mutex_lock(&zram_index_mutex);
2327	ret = zram_add();
2328	mutex_unlock(lock: &zram_index_mutex);
2329
2330	if (ret < 0)
2331	return ret;
2332	return scnprintf(buf, PAGE_SIZE, fmt: "%d\n", ret);
2333	}
2334	/* This attribute must be set to 0400, so CLASS_ATTR_RO() can not be used */
2335	static struct class_attribute class_attr_hot_add =
2336	__ATTR(hot_add, 0400, hot_add_show, NULL);
2337
2338	static ssize_t hot_remove_store(const struct class *class,
2339	const struct class_attribute *attr,
2340	const char *buf,
2341	size_t count)
2342	{
2343	struct zram *zram;
2344	int ret, dev_id;
2345
2346	/* dev_id is gendisk->first_minor, which is `int' */
2347	ret = kstrtoint(s: buf, base: 10, res: &dev_id);
2348	if (ret)
2349	return ret;
2350	if (dev_id < 0)
2351	return -EINVAL;
2352
2353	mutex_lock(&zram_index_mutex);
2354
2355	zram = idr_find(&zram_index_idr, id: dev_id);
2356	if (zram) {
2357	ret = zram_remove(zram);
2358	if (!ret)
2359	idr_remove(&zram_index_idr, id: dev_id);
2360	} else {
2361	ret = -ENODEV;
2362	}
2363
2364	mutex_unlock(lock: &zram_index_mutex);
2365	return ret ? ret : count;
2366	}
2367	static CLASS_ATTR_WO(hot_remove);
2368
2369	static struct attribute *zram_control_class_attrs[] = {
2370	&class_attr_hot_add.attr,
2371	&class_attr_hot_remove.attr,
2372	NULL,
2373	};
2374	ATTRIBUTE_GROUPS(zram_control_class);
2375
2376	static struct class zram_control_class = {
2377	.name = "zram-control",
2378	.class_groups = zram_control_class_groups,
2379	};
2380
2381	static int zram_remove_cb(int id, void *ptr, void *data)
2382	{
2383	WARN_ON_ONCE(zram_remove(ptr));
2384	return 0;
2385	}
2386
2387	static void destroy_devices(void)
2388	{
2389	class_unregister(class: &zram_control_class);
2390	idr_for_each(&zram_index_idr, fn: &zram_remove_cb, NULL);
2391	zram_debugfs_destroy();
2392	idr_destroy(&zram_index_idr);
2393	unregister_blkdev(major: zram_major, name: "zram");
2394	cpuhp_remove_multi_state(state: CPUHP_ZCOMP_PREPARE);
2395	}
2396
2397	static int __init zram_init(void)
2398	{
2399	int ret;
2400
2401	BUILD_BUG_ON(__NR_ZRAM_PAGEFLAGS > BITS_PER_LONG);
2402
2403	ret = cpuhp_setup_state_multi(state: CPUHP_ZCOMP_PREPARE, name: "block/zram:prepare",
2404	startup: zcomp_cpu_up_prepare, teardown: zcomp_cpu_dead);
2405	if (ret < 0)
2406	return ret;
2407
2408	ret = class_register(class: &zram_control_class);
2409	if (ret) {
2410	pr_err("Unable to register zram-control class\n");
2411	cpuhp_remove_multi_state(state: CPUHP_ZCOMP_PREPARE);
2412	return ret;
2413	}
2414
2415	zram_debugfs_create();
2416	zram_major = register_blkdev(0, "zram");
2417	if (zram_major <= 0) {
2418	pr_err("Unable to get major number\n");
2419	class_unregister(class: &zram_control_class);
2420	cpuhp_remove_multi_state(state: CPUHP_ZCOMP_PREPARE);
2421	return -EBUSY;
2422	}
2423
2424	while (num_devices != 0) {
2425	mutex_lock(&zram_index_mutex);
2426	ret = zram_add();
2427	mutex_unlock(lock: &zram_index_mutex);
2428	if (ret < 0)
2429	goto out_error;
2430	num_devices--;
2431	}
2432
2433	return 0;
2434
2435	out_error:
2436	destroy_devices();
2437	return ret;
2438	}
2439
2440	static void __exit zram_exit(void)
2441	{
2442	destroy_devices();
2443	}
2444
2445	module_init(zram_init);
2446	module_exit(zram_exit);
2447
2448	module_param(num_devices, uint, 0);
2449	MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2450
2451	MODULE_LICENSE("Dual BSD/GPL");
2452	MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2453	MODULE_DESCRIPTION("Compressed RAM Block Device");
2454