2 * Compressed RAM block device
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
5 * 2012, 2013 Minchan Kim
7 * This code is released using a dual license strategy: BSD/GPL
8 * You can choose the licence that better fits your requirements.
10 * Released under the terms of 3-clause BSD License
11 * Released under the terms of GNU General Public License Version 2.0
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
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/genhd.h>
26 #include <linux/highmem.h>
27 #include <linux/slab.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>
36 static DEFINE_IDR(zram_index_idr
);
37 /* idr index must be protected */
38 static DEFINE_MUTEX(zram_index_mutex
);
40 static int zram_major
;
41 static const char *default_compressor
= "lzo";
43 /* Module params (documentation at end) */
44 static unsigned int num_devices
= 1;
46 static inline void deprecated_attr_warn(const char *name
)
48 pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
52 "See zram documentation.");
55 #define ZRAM_ATTR_RO(name) \
56 static ssize_t name##_show(struct device *d, \
57 struct device_attribute *attr, char *b) \
59 struct zram *zram = dev_to_zram(d); \
61 deprecated_attr_warn(__stringify(name)); \
62 return scnprintf(b, PAGE_SIZE, "%llu\n", \
63 (u64)atomic64_read(&zram->stats.name)); \
65 static DEVICE_ATTR_RO(name);
67 static inline bool init_done(struct zram
*zram
)
69 return zram
->disksize
;
72 static inline struct zram
*dev_to_zram(struct device
*dev
)
74 return (struct zram
*)dev_to_disk(dev
)->private_data
;
77 /* flag operations require table entry bit_spin_lock() being held */
78 static int zram_test_flag(struct zram_meta
*meta
, u32 index
,
79 enum zram_pageflags flag
)
81 return meta
->table
[index
].value
& BIT(flag
);
84 static void zram_set_flag(struct zram_meta
*meta
, u32 index
,
85 enum zram_pageflags flag
)
87 meta
->table
[index
].value
|= BIT(flag
);
90 static void zram_clear_flag(struct zram_meta
*meta
, u32 index
,
91 enum zram_pageflags flag
)
93 meta
->table
[index
].value
&= ~BIT(flag
);
96 static size_t zram_get_obj_size(struct zram_meta
*meta
, u32 index
)
98 return meta
->table
[index
].value
& (BIT(ZRAM_FLAG_SHIFT
) - 1);
101 static void zram_set_obj_size(struct zram_meta
*meta
,
102 u32 index
, size_t size
)
104 unsigned long flags
= meta
->table
[index
].value
>> ZRAM_FLAG_SHIFT
;
106 meta
->table
[index
].value
= (flags
<< ZRAM_FLAG_SHIFT
) | size
;
109 static inline bool is_partial_io(struct bio_vec
*bvec
)
111 return bvec
->bv_len
!= PAGE_SIZE
;
115 * Check if request is within bounds and aligned on zram logical blocks.
117 static inline bool valid_io_request(struct zram
*zram
,
118 sector_t start
, unsigned int size
)
122 /* unaligned request */
123 if (unlikely(start
& (ZRAM_SECTOR_PER_LOGICAL_BLOCK
- 1)))
125 if (unlikely(size
& (ZRAM_LOGICAL_BLOCK_SIZE
- 1)))
128 end
= start
+ (size
>> SECTOR_SHIFT
);
129 bound
= zram
->disksize
>> SECTOR_SHIFT
;
130 /* out of range range */
131 if (unlikely(start
>= bound
|| end
> bound
|| start
> end
))
134 /* I/O request is valid */
138 static void update_position(u32
*index
, int *offset
, struct bio_vec
*bvec
)
140 if (*offset
+ bvec
->bv_len
>= PAGE_SIZE
)
142 *offset
= (*offset
+ bvec
->bv_len
) % PAGE_SIZE
;
145 static inline void update_used_max(struct zram
*zram
,
146 const unsigned long pages
)
148 unsigned long old_max
, cur_max
;
150 old_max
= atomic_long_read(&zram
->stats
.max_used_pages
);
155 old_max
= atomic_long_cmpxchg(
156 &zram
->stats
.max_used_pages
, cur_max
, pages
);
157 } while (old_max
!= cur_max
);
160 static bool page_zero_filled(void *ptr
)
165 page
= (unsigned long *)ptr
;
167 for (pos
= 0; pos
!= PAGE_SIZE
/ sizeof(*page
); pos
++) {
175 static void handle_zero_page(struct bio_vec
*bvec
)
177 struct page
*page
= bvec
->bv_page
;
180 user_mem
= kmap_atomic(page
);
181 if (is_partial_io(bvec
))
182 memset(user_mem
+ bvec
->bv_offset
, 0, bvec
->bv_len
);
184 clear_page(user_mem
);
185 kunmap_atomic(user_mem
);
187 flush_dcache_page(page
);
190 static ssize_t
initstate_show(struct device
*dev
,
191 struct device_attribute
*attr
, char *buf
)
194 struct zram
*zram
= dev_to_zram(dev
);
196 down_read(&zram
->init_lock
);
197 val
= init_done(zram
);
198 up_read(&zram
->init_lock
);
200 return scnprintf(buf
, PAGE_SIZE
, "%u\n", val
);
203 static ssize_t
disksize_show(struct device
*dev
,
204 struct device_attribute
*attr
, char *buf
)
206 struct zram
*zram
= dev_to_zram(dev
);
208 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", zram
->disksize
);
211 static ssize_t
orig_data_size_show(struct device
*dev
,
212 struct device_attribute
*attr
, char *buf
)
214 struct zram
*zram
= dev_to_zram(dev
);
216 deprecated_attr_warn("orig_data_size");
217 return scnprintf(buf
, PAGE_SIZE
, "%llu\n",
218 (u64
)(atomic64_read(&zram
->stats
.pages_stored
)) << PAGE_SHIFT
);
221 static ssize_t
mem_used_total_show(struct device
*dev
,
222 struct device_attribute
*attr
, char *buf
)
225 struct zram
*zram
= dev_to_zram(dev
);
227 deprecated_attr_warn("mem_used_total");
228 down_read(&zram
->init_lock
);
229 if (init_done(zram
)) {
230 struct zram_meta
*meta
= zram
->meta
;
231 val
= zs_get_total_pages(meta
->mem_pool
);
233 up_read(&zram
->init_lock
);
235 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
238 static ssize_t
mem_limit_show(struct device
*dev
,
239 struct device_attribute
*attr
, char *buf
)
242 struct zram
*zram
= dev_to_zram(dev
);
244 deprecated_attr_warn("mem_limit");
245 down_read(&zram
->init_lock
);
246 val
= zram
->limit_pages
;
247 up_read(&zram
->init_lock
);
249 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
252 static ssize_t
mem_limit_store(struct device
*dev
,
253 struct device_attribute
*attr
, const char *buf
, size_t len
)
257 struct zram
*zram
= dev_to_zram(dev
);
259 limit
= memparse(buf
, &tmp
);
260 if (buf
== tmp
) /* no chars parsed, invalid input */
263 down_write(&zram
->init_lock
);
264 zram
->limit_pages
= PAGE_ALIGN(limit
) >> PAGE_SHIFT
;
265 up_write(&zram
->init_lock
);
270 static ssize_t
mem_used_max_show(struct device
*dev
,
271 struct device_attribute
*attr
, char *buf
)
274 struct zram
*zram
= dev_to_zram(dev
);
276 deprecated_attr_warn("mem_used_max");
277 down_read(&zram
->init_lock
);
279 val
= atomic_long_read(&zram
->stats
.max_used_pages
);
280 up_read(&zram
->init_lock
);
282 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
285 static ssize_t
mem_used_max_store(struct device
*dev
,
286 struct device_attribute
*attr
, const char *buf
, size_t len
)
290 struct zram
*zram
= dev_to_zram(dev
);
292 err
= kstrtoul(buf
, 10, &val
);
296 down_read(&zram
->init_lock
);
297 if (init_done(zram
)) {
298 struct zram_meta
*meta
= zram
->meta
;
299 atomic_long_set(&zram
->stats
.max_used_pages
,
300 zs_get_total_pages(meta
->mem_pool
));
302 up_read(&zram
->init_lock
);
308 * We switched to per-cpu streams and this attr is not needed anymore.
309 * However, we will keep it around for some time, because:
310 * a) we may revert per-cpu streams in the future
311 * b) it's visible to user space and we need to follow our 2 years
312 * retirement rule; but we already have a number of 'soon to be
313 * altered' attrs, so max_comp_streams need to wait for the next
316 static ssize_t
max_comp_streams_show(struct device
*dev
,
317 struct device_attribute
*attr
, char *buf
)
319 return scnprintf(buf
, PAGE_SIZE
, "%d\n", num_online_cpus());
322 static ssize_t
max_comp_streams_store(struct device
*dev
,
323 struct device_attribute
*attr
, const char *buf
, size_t len
)
328 static ssize_t
comp_algorithm_show(struct device
*dev
,
329 struct device_attribute
*attr
, char *buf
)
332 struct zram
*zram
= dev_to_zram(dev
);
334 down_read(&zram
->init_lock
);
335 sz
= zcomp_available_show(zram
->compressor
, buf
);
336 up_read(&zram
->init_lock
);
341 static ssize_t
comp_algorithm_store(struct device
*dev
,
342 struct device_attribute
*attr
, const char *buf
, size_t len
)
344 struct zram
*zram
= dev_to_zram(dev
);
345 char compressor
[CRYPTO_MAX_ALG_NAME
];
348 strlcpy(compressor
, buf
, sizeof(compressor
));
349 /* ignore trailing newline */
350 sz
= strlen(compressor
);
351 if (sz
> 0 && compressor
[sz
- 1] == '\n')
352 compressor
[sz
- 1] = 0x00;
354 if (!zcomp_available_algorithm(compressor
))
357 down_write(&zram
->init_lock
);
358 if (init_done(zram
)) {
359 up_write(&zram
->init_lock
);
360 pr_info("Can't change algorithm for initialized device\n");
364 strlcpy(zram
->compressor
, compressor
, sizeof(compressor
));
365 up_write(&zram
->init_lock
);
369 static ssize_t
compact_store(struct device
*dev
,
370 struct device_attribute
*attr
, const char *buf
, size_t len
)
372 struct zram
*zram
= dev_to_zram(dev
);
373 struct zram_meta
*meta
;
375 down_read(&zram
->init_lock
);
376 if (!init_done(zram
)) {
377 up_read(&zram
->init_lock
);
382 zs_compact(meta
->mem_pool
);
383 up_read(&zram
->init_lock
);
388 static ssize_t
io_stat_show(struct device
*dev
,
389 struct device_attribute
*attr
, char *buf
)
391 struct zram
*zram
= dev_to_zram(dev
);
394 down_read(&zram
->init_lock
);
395 ret
= scnprintf(buf
, PAGE_SIZE
,
396 "%8llu %8llu %8llu %8llu\n",
397 (u64
)atomic64_read(&zram
->stats
.failed_reads
),
398 (u64
)atomic64_read(&zram
->stats
.failed_writes
),
399 (u64
)atomic64_read(&zram
->stats
.invalid_io
),
400 (u64
)atomic64_read(&zram
->stats
.notify_free
));
401 up_read(&zram
->init_lock
);
406 static ssize_t
mm_stat_show(struct device
*dev
,
407 struct device_attribute
*attr
, char *buf
)
409 struct zram
*zram
= dev_to_zram(dev
);
410 struct zs_pool_stats pool_stats
;
411 u64 orig_size
, mem_used
= 0;
415 memset(&pool_stats
, 0x00, sizeof(struct zs_pool_stats
));
417 down_read(&zram
->init_lock
);
418 if (init_done(zram
)) {
419 mem_used
= zs_get_total_pages(zram
->meta
->mem_pool
);
420 zs_pool_stats(zram
->meta
->mem_pool
, &pool_stats
);
423 orig_size
= atomic64_read(&zram
->stats
.pages_stored
);
424 max_used
= atomic_long_read(&zram
->stats
.max_used_pages
);
426 ret
= scnprintf(buf
, PAGE_SIZE
,
427 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
428 orig_size
<< PAGE_SHIFT
,
429 (u64
)atomic64_read(&zram
->stats
.compr_data_size
),
430 mem_used
<< PAGE_SHIFT
,
431 zram
->limit_pages
<< PAGE_SHIFT
,
432 max_used
<< PAGE_SHIFT
,
433 (u64
)atomic64_read(&zram
->stats
.zero_pages
),
434 pool_stats
.pages_compacted
);
435 up_read(&zram
->init_lock
);
440 static ssize_t
debug_stat_show(struct device
*dev
,
441 struct device_attribute
*attr
, char *buf
)
444 struct zram
*zram
= dev_to_zram(dev
);
447 down_read(&zram
->init_lock
);
448 ret
= scnprintf(buf
, PAGE_SIZE
,
449 "version: %d\n%8llu\n",
451 (u64
)atomic64_read(&zram
->stats
.writestall
));
452 up_read(&zram
->init_lock
);
457 static DEVICE_ATTR_RO(io_stat
);
458 static DEVICE_ATTR_RO(mm_stat
);
459 static DEVICE_ATTR_RO(debug_stat
);
460 ZRAM_ATTR_RO(num_reads
);
461 ZRAM_ATTR_RO(num_writes
);
462 ZRAM_ATTR_RO(failed_reads
);
463 ZRAM_ATTR_RO(failed_writes
);
464 ZRAM_ATTR_RO(invalid_io
);
465 ZRAM_ATTR_RO(notify_free
);
466 ZRAM_ATTR_RO(zero_pages
);
467 ZRAM_ATTR_RO(compr_data_size
);
469 static inline bool zram_meta_get(struct zram
*zram
)
471 if (atomic_inc_not_zero(&zram
->refcount
))
476 static inline void zram_meta_put(struct zram
*zram
)
478 atomic_dec(&zram
->refcount
);
481 static void zram_meta_free(struct zram_meta
*meta
, u64 disksize
)
483 size_t num_pages
= disksize
>> PAGE_SHIFT
;
486 /* Free all pages that are still in this zram device */
487 for (index
= 0; index
< num_pages
; index
++) {
488 unsigned long handle
= meta
->table
[index
].handle
;
493 zs_free(meta
->mem_pool
, handle
);
496 zs_destroy_pool(meta
->mem_pool
);
501 static struct zram_meta
*zram_meta_alloc(char *pool_name
, u64 disksize
)
504 struct zram_meta
*meta
= kmalloc(sizeof(*meta
), GFP_KERNEL
);
509 num_pages
= disksize
>> PAGE_SHIFT
;
510 meta
->table
= vzalloc(num_pages
* sizeof(*meta
->table
));
512 pr_err("Error allocating zram address table\n");
516 meta
->mem_pool
= zs_create_pool(pool_name
);
517 if (!meta
->mem_pool
) {
518 pr_err("Error creating memory pool\n");
531 * To protect concurrent access to the same index entry,
532 * caller should hold this table index entry's bit_spinlock to
533 * indicate this index entry is accessing.
535 static void zram_free_page(struct zram
*zram
, size_t index
)
537 struct zram_meta
*meta
= zram
->meta
;
538 unsigned long handle
= meta
->table
[index
].handle
;
540 if (unlikely(!handle
)) {
542 * No memory is allocated for zero filled pages.
543 * Simply clear zero page flag.
545 if (zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
546 zram_clear_flag(meta
, index
, ZRAM_ZERO
);
547 atomic64_dec(&zram
->stats
.zero_pages
);
552 zs_free(meta
->mem_pool
, handle
);
554 atomic64_sub(zram_get_obj_size(meta
, index
),
555 &zram
->stats
.compr_data_size
);
556 atomic64_dec(&zram
->stats
.pages_stored
);
558 meta
->table
[index
].handle
= 0;
559 zram_set_obj_size(meta
, index
, 0);
562 static int zram_decompress_page(struct zram
*zram
, char *mem
, u32 index
)
566 struct zram_meta
*meta
= zram
->meta
;
567 unsigned long handle
;
570 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
571 handle
= meta
->table
[index
].handle
;
572 size
= zram_get_obj_size(meta
, index
);
574 if (!handle
|| zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
575 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
580 cmem
= zs_map_object(meta
->mem_pool
, handle
, ZS_MM_RO
);
581 if (size
== PAGE_SIZE
) {
582 copy_page(mem
, cmem
);
584 struct zcomp_strm
*zstrm
= zcomp_stream_get(zram
->comp
);
586 ret
= zcomp_decompress(zstrm
, cmem
, size
, mem
);
587 zcomp_stream_put(zram
->comp
);
589 zs_unmap_object(meta
->mem_pool
, handle
);
590 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
592 /* Should NEVER happen. Return bio error if it does. */
594 pr_err("Decompression failed! err=%d, page=%u\n", ret
, index
);
601 static int zram_bvec_read(struct zram
*zram
, struct bio_vec
*bvec
,
602 u32 index
, int offset
)
606 unsigned char *user_mem
, *uncmem
= NULL
;
607 struct zram_meta
*meta
= zram
->meta
;
608 page
= bvec
->bv_page
;
610 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
611 if (unlikely(!meta
->table
[index
].handle
) ||
612 zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
613 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
614 handle_zero_page(bvec
);
617 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
619 if (is_partial_io(bvec
))
620 /* Use a temporary buffer to decompress the page */
621 uncmem
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
623 user_mem
= kmap_atomic(page
);
624 if (!is_partial_io(bvec
))
628 pr_err("Unable to allocate temp memory\n");
633 ret
= zram_decompress_page(zram
, uncmem
, index
);
634 /* Should NEVER happen. Return bio error if it does. */
638 if (is_partial_io(bvec
))
639 memcpy(user_mem
+ bvec
->bv_offset
, uncmem
+ offset
,
642 flush_dcache_page(page
);
645 kunmap_atomic(user_mem
);
646 if (is_partial_io(bvec
))
651 static int zram_bvec_write(struct zram
*zram
, struct bio_vec
*bvec
, u32 index
,
656 unsigned long handle
= 0;
658 unsigned char *user_mem
, *cmem
, *src
, *uncmem
= NULL
;
659 struct zram_meta
*meta
= zram
->meta
;
660 struct zcomp_strm
*zstrm
= NULL
;
661 unsigned long alloced_pages
;
663 page
= bvec
->bv_page
;
664 if (is_partial_io(bvec
)) {
666 * This is a partial IO. We need to read the full page
667 * before to write the changes.
669 uncmem
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
674 ret
= zram_decompress_page(zram
, uncmem
, index
);
680 user_mem
= kmap_atomic(page
);
681 if (is_partial_io(bvec
)) {
682 memcpy(uncmem
+ offset
, user_mem
+ bvec
->bv_offset
,
684 kunmap_atomic(user_mem
);
690 if (page_zero_filled(uncmem
)) {
692 kunmap_atomic(user_mem
);
693 /* Free memory associated with this sector now. */
694 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
695 zram_free_page(zram
, index
);
696 zram_set_flag(meta
, index
, ZRAM_ZERO
);
697 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
699 atomic64_inc(&zram
->stats
.zero_pages
);
704 zstrm
= zcomp_stream_get(zram
->comp
);
705 ret
= zcomp_compress(zstrm
, uncmem
, &clen
);
706 if (!is_partial_io(bvec
)) {
707 kunmap_atomic(user_mem
);
713 pr_err("Compression failed! err=%d\n", ret
);
718 if (unlikely(clen
> max_zpage_size
)) {
720 if (is_partial_io(bvec
))
725 * handle allocation has 2 paths:
726 * a) fast path is executed with preemption disabled (for
727 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
728 * since we can't sleep;
729 * b) slow path enables preemption and attempts to allocate
730 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
731 * put per-cpu compression stream and, thus, to re-do
732 * the compression once handle is allocated.
734 * if we have a 'non-null' handle here then we are coming
735 * from the slow path and handle has already been allocated.
738 handle
= zs_malloc(meta
->mem_pool
, clen
,
739 __GFP_KSWAPD_RECLAIM
|
744 zcomp_stream_put(zram
->comp
);
747 atomic64_inc(&zram
->stats
.writestall
);
749 handle
= zs_malloc(meta
->mem_pool
, clen
,
750 GFP_NOIO
| __GFP_HIGHMEM
|
755 pr_err("Error allocating memory for compressed page: %u, size=%u\n",
761 alloced_pages
= zs_get_total_pages(meta
->mem_pool
);
762 update_used_max(zram
, alloced_pages
);
764 if (zram
->limit_pages
&& alloced_pages
> zram
->limit_pages
) {
765 zs_free(meta
->mem_pool
, handle
);
770 cmem
= zs_map_object(meta
->mem_pool
, handle
, ZS_MM_WO
);
772 if ((clen
== PAGE_SIZE
) && !is_partial_io(bvec
)) {
773 src
= kmap_atomic(page
);
774 copy_page(cmem
, src
);
777 memcpy(cmem
, src
, clen
);
780 zcomp_stream_put(zram
->comp
);
782 zs_unmap_object(meta
->mem_pool
, handle
);
785 * Free memory associated with this sector
786 * before overwriting unused sectors.
788 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
789 zram_free_page(zram
, index
);
791 meta
->table
[index
].handle
= handle
;
792 zram_set_obj_size(meta
, index
, clen
);
793 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
796 atomic64_add(clen
, &zram
->stats
.compr_data_size
);
797 atomic64_inc(&zram
->stats
.pages_stored
);
800 zcomp_stream_put(zram
->comp
);
801 if (is_partial_io(bvec
))
807 * zram_bio_discard - handler on discard request
808 * @index: physical block index in PAGE_SIZE units
809 * @offset: byte offset within physical block
811 static void zram_bio_discard(struct zram
*zram
, u32 index
,
812 int offset
, struct bio
*bio
)
814 size_t n
= bio
->bi_iter
.bi_size
;
815 struct zram_meta
*meta
= zram
->meta
;
818 * zram manages data in physical block size units. Because logical block
819 * size isn't identical with physical block size on some arch, we
820 * could get a discard request pointing to a specific offset within a
821 * certain physical block. Although we can handle this request by
822 * reading that physiclal block and decompressing and partially zeroing
823 * and re-compressing and then re-storing it, this isn't reasonable
824 * because our intent with a discard request is to save memory. So
825 * skipping this logical block is appropriate here.
828 if (n
<= (PAGE_SIZE
- offset
))
831 n
-= (PAGE_SIZE
- offset
);
835 while (n
>= PAGE_SIZE
) {
836 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
837 zram_free_page(zram
, index
);
838 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
839 atomic64_inc(&zram
->stats
.notify_free
);
845 static int zram_bvec_rw(struct zram
*zram
, struct bio_vec
*bvec
, u32 index
,
846 int offset
, bool is_write
)
848 unsigned long start_time
= jiffies
;
849 int rw_acct
= is_write
? REQ_OP_WRITE
: REQ_OP_READ
;
852 generic_start_io_acct(rw_acct
, bvec
->bv_len
>> SECTOR_SHIFT
,
856 atomic64_inc(&zram
->stats
.num_reads
);
857 ret
= zram_bvec_read(zram
, bvec
, index
, offset
);
859 atomic64_inc(&zram
->stats
.num_writes
);
860 ret
= zram_bvec_write(zram
, bvec
, index
, offset
);
863 generic_end_io_acct(rw_acct
, &zram
->disk
->part0
, start_time
);
867 atomic64_inc(&zram
->stats
.failed_reads
);
869 atomic64_inc(&zram
->stats
.failed_writes
);
875 static void __zram_make_request(struct zram
*zram
, struct bio
*bio
)
880 struct bvec_iter iter
;
882 index
= bio
->bi_iter
.bi_sector
>> SECTORS_PER_PAGE_SHIFT
;
883 offset
= (bio
->bi_iter
.bi_sector
&
884 (SECTORS_PER_PAGE
- 1)) << SECTOR_SHIFT
;
886 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
)) {
887 zram_bio_discard(zram
, index
, offset
, bio
);
892 bio_for_each_segment(bvec
, bio
, iter
) {
893 int max_transfer_size
= PAGE_SIZE
- offset
;
895 if (bvec
.bv_len
> max_transfer_size
) {
897 * zram_bvec_rw() can only make operation on a single
898 * zram page. Split the bio vector.
902 bv
.bv_page
= bvec
.bv_page
;
903 bv
.bv_len
= max_transfer_size
;
904 bv
.bv_offset
= bvec
.bv_offset
;
906 if (zram_bvec_rw(zram
, &bv
, index
, offset
,
907 op_is_write(bio_op(bio
))) < 0)
910 bv
.bv_len
= bvec
.bv_len
- max_transfer_size
;
911 bv
.bv_offset
+= max_transfer_size
;
912 if (zram_bvec_rw(zram
, &bv
, index
+ 1, 0,
913 op_is_write(bio_op(bio
))) < 0)
916 if (zram_bvec_rw(zram
, &bvec
, index
, offset
,
917 op_is_write(bio_op(bio
))) < 0)
920 update_position(&index
, &offset
, &bvec
);
931 * Handler function for all zram I/O requests.
933 static blk_qc_t
zram_make_request(struct request_queue
*queue
, struct bio
*bio
)
935 struct zram
*zram
= queue
->queuedata
;
937 if (unlikely(!zram_meta_get(zram
)))
940 blk_queue_split(queue
, &bio
, queue
->bio_split
);
942 if (!valid_io_request(zram
, bio
->bi_iter
.bi_sector
,
943 bio
->bi_iter
.bi_size
)) {
944 atomic64_inc(&zram
->stats
.invalid_io
);
948 __zram_make_request(zram
, bio
);
950 return BLK_QC_T_NONE
;
955 return BLK_QC_T_NONE
;
958 static void zram_slot_free_notify(struct block_device
*bdev
,
962 struct zram_meta
*meta
;
964 zram
= bdev
->bd_disk
->private_data
;
967 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
968 zram_free_page(zram
, index
);
969 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
970 atomic64_inc(&zram
->stats
.notify_free
);
973 static int zram_rw_page(struct block_device
*bdev
, sector_t sector
,
974 struct page
*page
, bool is_write
)
976 int offset
, err
= -EIO
;
981 zram
= bdev
->bd_disk
->private_data
;
982 if (unlikely(!zram_meta_get(zram
)))
985 if (!valid_io_request(zram
, sector
, PAGE_SIZE
)) {
986 atomic64_inc(&zram
->stats
.invalid_io
);
991 index
= sector
>> SECTORS_PER_PAGE_SHIFT
;
992 offset
= sector
& (SECTORS_PER_PAGE
- 1) << SECTOR_SHIFT
;
995 bv
.bv_len
= PAGE_SIZE
;
998 err
= zram_bvec_rw(zram
, &bv
, index
, offset
, is_write
);
1000 zram_meta_put(zram
);
1003 * If I/O fails, just return error(ie, non-zero) without
1004 * calling page_endio.
1005 * It causes resubmit the I/O with bio request by upper functions
1006 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1007 * bio->bi_end_io does things to handle the error
1008 * (e.g., SetPageError, set_page_dirty and extra works).
1011 page_endio(page
, is_write
, 0);
1015 static void zram_reset_device(struct zram
*zram
)
1017 struct zram_meta
*meta
;
1021 down_write(&zram
->init_lock
);
1023 zram
->limit_pages
= 0;
1025 if (!init_done(zram
)) {
1026 up_write(&zram
->init_lock
);
1032 disksize
= zram
->disksize
;
1034 * Refcount will go down to 0 eventually and r/w handler
1035 * cannot handle further I/O so it will bail out by
1036 * check zram_meta_get.
1038 zram_meta_put(zram
);
1040 * We want to free zram_meta in process context to avoid
1041 * deadlock between reclaim path and any other locks.
1043 wait_event(zram
->io_done
, atomic_read(&zram
->refcount
) == 0);
1046 memset(&zram
->stats
, 0, sizeof(zram
->stats
));
1049 set_capacity(zram
->disk
, 0);
1050 part_stat_set_all(&zram
->disk
->part0
, 0);
1052 up_write(&zram
->init_lock
);
1053 /* I/O operation under all of CPU are done so let's free */
1054 zram_meta_free(meta
, disksize
);
1055 zcomp_destroy(comp
);
1058 static ssize_t
disksize_store(struct device
*dev
,
1059 struct device_attribute
*attr
, const char *buf
, size_t len
)
1063 struct zram_meta
*meta
;
1064 struct zram
*zram
= dev_to_zram(dev
);
1067 disksize
= memparse(buf
, NULL
);
1071 disksize
= PAGE_ALIGN(disksize
);
1072 meta
= zram_meta_alloc(zram
->disk
->disk_name
, disksize
);
1076 comp
= zcomp_create(zram
->compressor
);
1078 pr_err("Cannot initialise %s compressing backend\n",
1080 err
= PTR_ERR(comp
);
1084 down_write(&zram
->init_lock
);
1085 if (init_done(zram
)) {
1086 pr_info("Cannot change disksize for initialized device\n");
1088 goto out_destroy_comp
;
1091 init_waitqueue_head(&zram
->io_done
);
1092 atomic_set(&zram
->refcount
, 1);
1095 zram
->disksize
= disksize
;
1096 set_capacity(zram
->disk
, zram
->disksize
>> SECTOR_SHIFT
);
1097 up_write(&zram
->init_lock
);
1100 * Revalidate disk out of the init_lock to avoid lockdep splat.
1101 * It's okay because disk's capacity is protected by init_lock
1102 * so that revalidate_disk always sees up-to-date capacity.
1104 revalidate_disk(zram
->disk
);
1109 up_write(&zram
->init_lock
);
1110 zcomp_destroy(comp
);
1112 zram_meta_free(meta
, disksize
);
1116 static ssize_t
reset_store(struct device
*dev
,
1117 struct device_attribute
*attr
, const char *buf
, size_t len
)
1120 unsigned short do_reset
;
1122 struct block_device
*bdev
;
1124 ret
= kstrtou16(buf
, 10, &do_reset
);
1131 zram
= dev_to_zram(dev
);
1132 bdev
= bdget_disk(zram
->disk
, 0);
1136 mutex_lock(&bdev
->bd_mutex
);
1137 /* Do not reset an active device or claimed device */
1138 if (bdev
->bd_openers
|| zram
->claim
) {
1139 mutex_unlock(&bdev
->bd_mutex
);
1144 /* From now on, anyone can't open /dev/zram[0-9] */
1146 mutex_unlock(&bdev
->bd_mutex
);
1148 /* Make sure all the pending I/O are finished */
1150 zram_reset_device(zram
);
1151 revalidate_disk(zram
->disk
);
1154 mutex_lock(&bdev
->bd_mutex
);
1155 zram
->claim
= false;
1156 mutex_unlock(&bdev
->bd_mutex
);
1161 static int zram_open(struct block_device
*bdev
, fmode_t mode
)
1166 WARN_ON(!mutex_is_locked(&bdev
->bd_mutex
));
1168 zram
= bdev
->bd_disk
->private_data
;
1169 /* zram was claimed to reset so open request fails */
1176 static const struct block_device_operations zram_devops
= {
1178 .swap_slot_free_notify
= zram_slot_free_notify
,
1179 .rw_page
= zram_rw_page
,
1180 .owner
= THIS_MODULE
1183 static DEVICE_ATTR_WO(compact
);
1184 static DEVICE_ATTR_RW(disksize
);
1185 static DEVICE_ATTR_RO(initstate
);
1186 static DEVICE_ATTR_WO(reset
);
1187 static DEVICE_ATTR_RO(orig_data_size
);
1188 static DEVICE_ATTR_RO(mem_used_total
);
1189 static DEVICE_ATTR_RW(mem_limit
);
1190 static DEVICE_ATTR_RW(mem_used_max
);
1191 static DEVICE_ATTR_RW(max_comp_streams
);
1192 static DEVICE_ATTR_RW(comp_algorithm
);
1194 static struct attribute
*zram_disk_attrs
[] = {
1195 &dev_attr_disksize
.attr
,
1196 &dev_attr_initstate
.attr
,
1197 &dev_attr_reset
.attr
,
1198 &dev_attr_num_reads
.attr
,
1199 &dev_attr_num_writes
.attr
,
1200 &dev_attr_failed_reads
.attr
,
1201 &dev_attr_failed_writes
.attr
,
1202 &dev_attr_compact
.attr
,
1203 &dev_attr_invalid_io
.attr
,
1204 &dev_attr_notify_free
.attr
,
1205 &dev_attr_zero_pages
.attr
,
1206 &dev_attr_orig_data_size
.attr
,
1207 &dev_attr_compr_data_size
.attr
,
1208 &dev_attr_mem_used_total
.attr
,
1209 &dev_attr_mem_limit
.attr
,
1210 &dev_attr_mem_used_max
.attr
,
1211 &dev_attr_max_comp_streams
.attr
,
1212 &dev_attr_comp_algorithm
.attr
,
1213 &dev_attr_io_stat
.attr
,
1214 &dev_attr_mm_stat
.attr
,
1215 &dev_attr_debug_stat
.attr
,
1219 static struct attribute_group zram_disk_attr_group
= {
1220 .attrs
= zram_disk_attrs
,
1224 * Allocate and initialize new zram device. the function returns
1225 * '>= 0' device_id upon success, and negative value otherwise.
1227 static int zram_add(void)
1230 struct request_queue
*queue
;
1233 zram
= kzalloc(sizeof(struct zram
), GFP_KERNEL
);
1237 ret
= idr_alloc(&zram_index_idr
, zram
, 0, 0, GFP_KERNEL
);
1242 init_rwsem(&zram
->init_lock
);
1244 queue
= blk_alloc_queue(GFP_KERNEL
);
1246 pr_err("Error allocating disk queue for device %d\n",
1252 blk_queue_make_request(queue
, zram_make_request
);
1254 /* gendisk structure */
1255 zram
->disk
= alloc_disk(1);
1257 pr_err("Error allocating disk structure for device %d\n",
1260 goto out_free_queue
;
1263 zram
->disk
->major
= zram_major
;
1264 zram
->disk
->first_minor
= device_id
;
1265 zram
->disk
->fops
= &zram_devops
;
1266 zram
->disk
->queue
= queue
;
1267 zram
->disk
->queue
->queuedata
= zram
;
1268 zram
->disk
->private_data
= zram
;
1269 snprintf(zram
->disk
->disk_name
, 16, "zram%d", device_id
);
1271 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1272 set_capacity(zram
->disk
, 0);
1273 /* zram devices sort of resembles non-rotational disks */
1274 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, zram
->disk
->queue
);
1275 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, zram
->disk
->queue
);
1277 * To ensure that we always get PAGE_SIZE aligned
1278 * and n*PAGE_SIZED sized I/O requests.
1280 blk_queue_physical_block_size(zram
->disk
->queue
, PAGE_SIZE
);
1281 blk_queue_logical_block_size(zram
->disk
->queue
,
1282 ZRAM_LOGICAL_BLOCK_SIZE
);
1283 blk_queue_io_min(zram
->disk
->queue
, PAGE_SIZE
);
1284 blk_queue_io_opt(zram
->disk
->queue
, PAGE_SIZE
);
1285 zram
->disk
->queue
->limits
.discard_granularity
= PAGE_SIZE
;
1286 blk_queue_max_discard_sectors(zram
->disk
->queue
, UINT_MAX
);
1288 * zram_bio_discard() will clear all logical blocks if logical block
1289 * size is identical with physical block size(PAGE_SIZE). But if it is
1290 * different, we will skip discarding some parts of logical blocks in
1291 * the part of the request range which isn't aligned to physical block
1292 * size. So we can't ensure that all discarded logical blocks are
1295 if (ZRAM_LOGICAL_BLOCK_SIZE
== PAGE_SIZE
)
1296 zram
->disk
->queue
->limits
.discard_zeroes_data
= 1;
1298 zram
->disk
->queue
->limits
.discard_zeroes_data
= 0;
1299 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, zram
->disk
->queue
);
1301 add_disk(zram
->disk
);
1303 ret
= sysfs_create_group(&disk_to_dev(zram
->disk
)->kobj
,
1304 &zram_disk_attr_group
);
1306 pr_err("Error creating sysfs group for device %d\n",
1310 strlcpy(zram
->compressor
, default_compressor
, sizeof(zram
->compressor
));
1313 pr_info("Added device: %s\n", zram
->disk
->disk_name
);
1317 del_gendisk(zram
->disk
);
1318 put_disk(zram
->disk
);
1320 blk_cleanup_queue(queue
);
1322 idr_remove(&zram_index_idr
, device_id
);
1328 static int zram_remove(struct zram
*zram
)
1330 struct block_device
*bdev
;
1332 bdev
= bdget_disk(zram
->disk
, 0);
1336 mutex_lock(&bdev
->bd_mutex
);
1337 if (bdev
->bd_openers
|| zram
->claim
) {
1338 mutex_unlock(&bdev
->bd_mutex
);
1344 mutex_unlock(&bdev
->bd_mutex
);
1347 * Remove sysfs first, so no one will perform a disksize
1348 * store while we destroy the devices. This also helps during
1349 * hot_remove -- zram_reset_device() is the last holder of
1350 * ->init_lock, no later/concurrent disksize_store() or any
1351 * other sysfs handlers are possible.
1353 sysfs_remove_group(&disk_to_dev(zram
->disk
)->kobj
,
1354 &zram_disk_attr_group
);
1356 /* Make sure all the pending I/O are finished */
1358 zram_reset_device(zram
);
1361 pr_info("Removed device: %s\n", zram
->disk
->disk_name
);
1363 blk_cleanup_queue(zram
->disk
->queue
);
1364 del_gendisk(zram
->disk
);
1365 put_disk(zram
->disk
);
1370 /* zram-control sysfs attributes */
1371 static ssize_t
hot_add_show(struct class *class,
1372 struct class_attribute
*attr
,
1377 mutex_lock(&zram_index_mutex
);
1379 mutex_unlock(&zram_index_mutex
);
1383 return scnprintf(buf
, PAGE_SIZE
, "%d\n", ret
);
1386 static ssize_t
hot_remove_store(struct class *class,
1387 struct class_attribute
*attr
,
1394 /* dev_id is gendisk->first_minor, which is `int' */
1395 ret
= kstrtoint(buf
, 10, &dev_id
);
1401 mutex_lock(&zram_index_mutex
);
1403 zram
= idr_find(&zram_index_idr
, dev_id
);
1405 ret
= zram_remove(zram
);
1406 idr_remove(&zram_index_idr
, dev_id
);
1411 mutex_unlock(&zram_index_mutex
);
1412 return ret
? ret
: count
;
1415 static struct class_attribute zram_control_class_attrs
[] = {
1417 __ATTR_WO(hot_remove
),
1421 static struct class zram_control_class
= {
1422 .name
= "zram-control",
1423 .owner
= THIS_MODULE
,
1424 .class_attrs
= zram_control_class_attrs
,
1427 static int zram_remove_cb(int id
, void *ptr
, void *data
)
1433 static void destroy_devices(void)
1435 class_unregister(&zram_control_class
);
1436 idr_for_each(&zram_index_idr
, &zram_remove_cb
, NULL
);
1437 idr_destroy(&zram_index_idr
);
1438 unregister_blkdev(zram_major
, "zram");
1441 static int __init
zram_init(void)
1445 ret
= class_register(&zram_control_class
);
1447 pr_err("Unable to register zram-control class\n");
1451 zram_major
= register_blkdev(0, "zram");
1452 if (zram_major
<= 0) {
1453 pr_err("Unable to get major number\n");
1454 class_unregister(&zram_control_class
);
1458 while (num_devices
!= 0) {
1459 mutex_lock(&zram_index_mutex
);
1461 mutex_unlock(&zram_index_mutex
);
1474 static void __exit
zram_exit(void)
1479 module_init(zram_init
);
1480 module_exit(zram_exit
);
1482 module_param(num_devices
, uint
, 0);
1483 MODULE_PARM_DESC(num_devices
, "Number of pre-created zram devices");
1485 MODULE_LICENSE("Dual BSD/GPL");
1486 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1487 MODULE_DESCRIPTION("Compressed RAM Block Device");