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/backing-dev.h>
29 #include <linux/string.h>
30 #include <linux/vmalloc.h>
31 #include <linux/err.h>
32 #include <linux/idr.h>
33 #include <linux/sysfs.h>
34 #include <linux/cpuhotplug.h>
38 static DEFINE_IDR(zram_index_idr
);
39 /* idr index must be protected */
40 static DEFINE_MUTEX(zram_index_mutex
);
42 static int zram_major
;
43 static const char *default_compressor
= "lzo";
45 /* Module params (documentation at end) */
46 static unsigned int num_devices
= 1;
48 static inline void deprecated_attr_warn(const char *name
)
50 pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
54 "See zram documentation.");
57 #define ZRAM_ATTR_RO(name) \
58 static ssize_t name##_show(struct device *d, \
59 struct device_attribute *attr, char *b) \
61 struct zram *zram = dev_to_zram(d); \
63 deprecated_attr_warn(__stringify(name)); \
64 return scnprintf(b, PAGE_SIZE, "%llu\n", \
65 (u64)atomic64_read(&zram->stats.name)); \
67 static DEVICE_ATTR_RO(name);
69 static inline bool init_done(struct zram
*zram
)
71 return zram
->disksize
;
74 static inline struct zram
*dev_to_zram(struct device
*dev
)
76 return (struct zram
*)dev_to_disk(dev
)->private_data
;
79 /* flag operations require table entry bit_spin_lock() being held */
80 static int zram_test_flag(struct zram_meta
*meta
, u32 index
,
81 enum zram_pageflags flag
)
83 return meta
->table
[index
].value
& BIT(flag
);
86 static void zram_set_flag(struct zram_meta
*meta
, u32 index
,
87 enum zram_pageflags flag
)
89 meta
->table
[index
].value
|= BIT(flag
);
92 static void zram_clear_flag(struct zram_meta
*meta
, u32 index
,
93 enum zram_pageflags flag
)
95 meta
->table
[index
].value
&= ~BIT(flag
);
98 static size_t zram_get_obj_size(struct zram_meta
*meta
, u32 index
)
100 return meta
->table
[index
].value
& (BIT(ZRAM_FLAG_SHIFT
) - 1);
103 static void zram_set_obj_size(struct zram_meta
*meta
,
104 u32 index
, size_t size
)
106 unsigned long flags
= meta
->table
[index
].value
>> ZRAM_FLAG_SHIFT
;
108 meta
->table
[index
].value
= (flags
<< ZRAM_FLAG_SHIFT
) | size
;
111 static inline bool is_partial_io(struct bio_vec
*bvec
)
113 return bvec
->bv_len
!= PAGE_SIZE
;
116 static void zram_revalidate_disk(struct zram
*zram
)
118 revalidate_disk(zram
->disk
);
119 /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */
120 zram
->disk
->queue
->backing_dev_info
->capabilities
|=
121 BDI_CAP_STABLE_WRITES
;
125 * Check if request is within bounds and aligned on zram logical blocks.
127 static inline bool valid_io_request(struct zram
*zram
,
128 sector_t start
, unsigned int size
)
132 /* unaligned request */
133 if (unlikely(start
& (ZRAM_SECTOR_PER_LOGICAL_BLOCK
- 1)))
135 if (unlikely(size
& (ZRAM_LOGICAL_BLOCK_SIZE
- 1)))
138 end
= start
+ (size
>> SECTOR_SHIFT
);
139 bound
= zram
->disksize
>> SECTOR_SHIFT
;
140 /* out of range range */
141 if (unlikely(start
>= bound
|| end
> bound
|| start
> end
))
144 /* I/O request is valid */
148 static void update_position(u32
*index
, int *offset
, struct bio_vec
*bvec
)
150 if (*offset
+ bvec
->bv_len
>= PAGE_SIZE
)
152 *offset
= (*offset
+ bvec
->bv_len
) % PAGE_SIZE
;
155 static inline void update_used_max(struct zram
*zram
,
156 const unsigned long pages
)
158 unsigned long old_max
, cur_max
;
160 old_max
= atomic_long_read(&zram
->stats
.max_used_pages
);
165 old_max
= atomic_long_cmpxchg(
166 &zram
->stats
.max_used_pages
, cur_max
, pages
);
167 } while (old_max
!= cur_max
);
170 static bool page_zero_filled(void *ptr
)
175 page
= (unsigned long *)ptr
;
177 for (pos
= 0; pos
!= PAGE_SIZE
/ sizeof(*page
); pos
++) {
185 static void handle_zero_page(struct bio_vec
*bvec
)
187 struct page
*page
= bvec
->bv_page
;
190 user_mem
= kmap_atomic(page
);
191 if (is_partial_io(bvec
))
192 memset(user_mem
+ bvec
->bv_offset
, 0, bvec
->bv_len
);
194 clear_page(user_mem
);
195 kunmap_atomic(user_mem
);
197 flush_dcache_page(page
);
200 static ssize_t
initstate_show(struct device
*dev
,
201 struct device_attribute
*attr
, char *buf
)
204 struct zram
*zram
= dev_to_zram(dev
);
206 down_read(&zram
->init_lock
);
207 val
= init_done(zram
);
208 up_read(&zram
->init_lock
);
210 return scnprintf(buf
, PAGE_SIZE
, "%u\n", val
);
213 static ssize_t
disksize_show(struct device
*dev
,
214 struct device_attribute
*attr
, char *buf
)
216 struct zram
*zram
= dev_to_zram(dev
);
218 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", zram
->disksize
);
221 static ssize_t
orig_data_size_show(struct device
*dev
,
222 struct device_attribute
*attr
, char *buf
)
224 struct zram
*zram
= dev_to_zram(dev
);
226 deprecated_attr_warn("orig_data_size");
227 return scnprintf(buf
, PAGE_SIZE
, "%llu\n",
228 (u64
)(atomic64_read(&zram
->stats
.pages_stored
)) << PAGE_SHIFT
);
231 static ssize_t
mem_used_total_show(struct device
*dev
,
232 struct device_attribute
*attr
, char *buf
)
235 struct zram
*zram
= dev_to_zram(dev
);
237 deprecated_attr_warn("mem_used_total");
238 down_read(&zram
->init_lock
);
239 if (init_done(zram
)) {
240 struct zram_meta
*meta
= zram
->meta
;
241 val
= zs_get_total_pages(meta
->mem_pool
);
243 up_read(&zram
->init_lock
);
245 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
248 static ssize_t
mem_limit_show(struct device
*dev
,
249 struct device_attribute
*attr
, char *buf
)
252 struct zram
*zram
= dev_to_zram(dev
);
254 deprecated_attr_warn("mem_limit");
255 down_read(&zram
->init_lock
);
256 val
= zram
->limit_pages
;
257 up_read(&zram
->init_lock
);
259 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
262 static ssize_t
mem_limit_store(struct device
*dev
,
263 struct device_attribute
*attr
, const char *buf
, size_t len
)
267 struct zram
*zram
= dev_to_zram(dev
);
269 limit
= memparse(buf
, &tmp
);
270 if (buf
== tmp
) /* no chars parsed, invalid input */
273 down_write(&zram
->init_lock
);
274 zram
->limit_pages
= PAGE_ALIGN(limit
) >> PAGE_SHIFT
;
275 up_write(&zram
->init_lock
);
280 static ssize_t
mem_used_max_show(struct device
*dev
,
281 struct device_attribute
*attr
, char *buf
)
284 struct zram
*zram
= dev_to_zram(dev
);
286 deprecated_attr_warn("mem_used_max");
287 down_read(&zram
->init_lock
);
289 val
= atomic_long_read(&zram
->stats
.max_used_pages
);
290 up_read(&zram
->init_lock
);
292 return scnprintf(buf
, PAGE_SIZE
, "%llu\n", val
<< PAGE_SHIFT
);
295 static ssize_t
mem_used_max_store(struct device
*dev
,
296 struct device_attribute
*attr
, const char *buf
, size_t len
)
300 struct zram
*zram
= dev_to_zram(dev
);
302 err
= kstrtoul(buf
, 10, &val
);
306 down_read(&zram
->init_lock
);
307 if (init_done(zram
)) {
308 struct zram_meta
*meta
= zram
->meta
;
309 atomic_long_set(&zram
->stats
.max_used_pages
,
310 zs_get_total_pages(meta
->mem_pool
));
312 up_read(&zram
->init_lock
);
318 * We switched to per-cpu streams and this attr is not needed anymore.
319 * However, we will keep it around for some time, because:
320 * a) we may revert per-cpu streams in the future
321 * b) it's visible to user space and we need to follow our 2 years
322 * retirement rule; but we already have a number of 'soon to be
323 * altered' attrs, so max_comp_streams need to wait for the next
326 static ssize_t
max_comp_streams_show(struct device
*dev
,
327 struct device_attribute
*attr
, char *buf
)
329 return scnprintf(buf
, PAGE_SIZE
, "%d\n", num_online_cpus());
332 static ssize_t
max_comp_streams_store(struct device
*dev
,
333 struct device_attribute
*attr
, const char *buf
, size_t len
)
338 static ssize_t
comp_algorithm_show(struct device
*dev
,
339 struct device_attribute
*attr
, char *buf
)
342 struct zram
*zram
= dev_to_zram(dev
);
344 down_read(&zram
->init_lock
);
345 sz
= zcomp_available_show(zram
->compressor
, buf
);
346 up_read(&zram
->init_lock
);
351 static ssize_t
comp_algorithm_store(struct device
*dev
,
352 struct device_attribute
*attr
, const char *buf
, size_t len
)
354 struct zram
*zram
= dev_to_zram(dev
);
355 char compressor
[CRYPTO_MAX_ALG_NAME
];
358 strlcpy(compressor
, buf
, sizeof(compressor
));
359 /* ignore trailing newline */
360 sz
= strlen(compressor
);
361 if (sz
> 0 && compressor
[sz
- 1] == '\n')
362 compressor
[sz
- 1] = 0x00;
364 if (!zcomp_available_algorithm(compressor
))
367 down_write(&zram
->init_lock
);
368 if (init_done(zram
)) {
369 up_write(&zram
->init_lock
);
370 pr_info("Can't change algorithm for initialized device\n");
374 strlcpy(zram
->compressor
, compressor
, sizeof(compressor
));
375 up_write(&zram
->init_lock
);
379 static ssize_t
compact_store(struct device
*dev
,
380 struct device_attribute
*attr
, const char *buf
, size_t len
)
382 struct zram
*zram
= dev_to_zram(dev
);
383 struct zram_meta
*meta
;
385 down_read(&zram
->init_lock
);
386 if (!init_done(zram
)) {
387 up_read(&zram
->init_lock
);
392 zs_compact(meta
->mem_pool
);
393 up_read(&zram
->init_lock
);
398 static ssize_t
io_stat_show(struct device
*dev
,
399 struct device_attribute
*attr
, char *buf
)
401 struct zram
*zram
= dev_to_zram(dev
);
404 down_read(&zram
->init_lock
);
405 ret
= scnprintf(buf
, PAGE_SIZE
,
406 "%8llu %8llu %8llu %8llu\n",
407 (u64
)atomic64_read(&zram
->stats
.failed_reads
),
408 (u64
)atomic64_read(&zram
->stats
.failed_writes
),
409 (u64
)atomic64_read(&zram
->stats
.invalid_io
),
410 (u64
)atomic64_read(&zram
->stats
.notify_free
));
411 up_read(&zram
->init_lock
);
416 static ssize_t
mm_stat_show(struct device
*dev
,
417 struct device_attribute
*attr
, char *buf
)
419 struct zram
*zram
= dev_to_zram(dev
);
420 struct zs_pool_stats pool_stats
;
421 u64 orig_size
, mem_used
= 0;
425 memset(&pool_stats
, 0x00, sizeof(struct zs_pool_stats
));
427 down_read(&zram
->init_lock
);
428 if (init_done(zram
)) {
429 mem_used
= zs_get_total_pages(zram
->meta
->mem_pool
);
430 zs_pool_stats(zram
->meta
->mem_pool
, &pool_stats
);
433 orig_size
= atomic64_read(&zram
->stats
.pages_stored
);
434 max_used
= atomic_long_read(&zram
->stats
.max_used_pages
);
436 ret
= scnprintf(buf
, PAGE_SIZE
,
437 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
438 orig_size
<< PAGE_SHIFT
,
439 (u64
)atomic64_read(&zram
->stats
.compr_data_size
),
440 mem_used
<< PAGE_SHIFT
,
441 zram
->limit_pages
<< PAGE_SHIFT
,
442 max_used
<< PAGE_SHIFT
,
443 (u64
)atomic64_read(&zram
->stats
.zero_pages
),
444 pool_stats
.pages_compacted
);
445 up_read(&zram
->init_lock
);
450 static ssize_t
debug_stat_show(struct device
*dev
,
451 struct device_attribute
*attr
, char *buf
)
454 struct zram
*zram
= dev_to_zram(dev
);
457 down_read(&zram
->init_lock
);
458 ret
= scnprintf(buf
, PAGE_SIZE
,
459 "version: %d\n%8llu\n",
461 (u64
)atomic64_read(&zram
->stats
.writestall
));
462 up_read(&zram
->init_lock
);
467 static DEVICE_ATTR_RO(io_stat
);
468 static DEVICE_ATTR_RO(mm_stat
);
469 static DEVICE_ATTR_RO(debug_stat
);
470 ZRAM_ATTR_RO(num_reads
);
471 ZRAM_ATTR_RO(num_writes
);
472 ZRAM_ATTR_RO(failed_reads
);
473 ZRAM_ATTR_RO(failed_writes
);
474 ZRAM_ATTR_RO(invalid_io
);
475 ZRAM_ATTR_RO(notify_free
);
476 ZRAM_ATTR_RO(zero_pages
);
477 ZRAM_ATTR_RO(compr_data_size
);
479 static inline bool zram_meta_get(struct zram
*zram
)
481 if (atomic_inc_not_zero(&zram
->refcount
))
486 static inline void zram_meta_put(struct zram
*zram
)
488 atomic_dec(&zram
->refcount
);
491 static void zram_meta_free(struct zram_meta
*meta
, u64 disksize
)
493 size_t num_pages
= disksize
>> PAGE_SHIFT
;
496 /* Free all pages that are still in this zram device */
497 for (index
= 0; index
< num_pages
; index
++) {
498 unsigned long handle
= meta
->table
[index
].handle
;
503 zs_free(meta
->mem_pool
, handle
);
506 zs_destroy_pool(meta
->mem_pool
);
511 static struct zram_meta
*zram_meta_alloc(char *pool_name
, u64 disksize
)
514 struct zram_meta
*meta
= kmalloc(sizeof(*meta
), GFP_KERNEL
);
519 num_pages
= disksize
>> PAGE_SHIFT
;
520 meta
->table
= vzalloc(num_pages
* sizeof(*meta
->table
));
522 pr_err("Error allocating zram address table\n");
526 meta
->mem_pool
= zs_create_pool(pool_name
);
527 if (!meta
->mem_pool
) {
528 pr_err("Error creating memory pool\n");
541 * To protect concurrent access to the same index entry,
542 * caller should hold this table index entry's bit_spinlock to
543 * indicate this index entry is accessing.
545 static void zram_free_page(struct zram
*zram
, size_t index
)
547 struct zram_meta
*meta
= zram
->meta
;
548 unsigned long handle
= meta
->table
[index
].handle
;
550 if (unlikely(!handle
)) {
552 * No memory is allocated for zero filled pages.
553 * Simply clear zero page flag.
555 if (zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
556 zram_clear_flag(meta
, index
, ZRAM_ZERO
);
557 atomic64_dec(&zram
->stats
.zero_pages
);
562 zs_free(meta
->mem_pool
, handle
);
564 atomic64_sub(zram_get_obj_size(meta
, index
),
565 &zram
->stats
.compr_data_size
);
566 atomic64_dec(&zram
->stats
.pages_stored
);
568 meta
->table
[index
].handle
= 0;
569 zram_set_obj_size(meta
, index
, 0);
572 static int zram_decompress_page(struct zram
*zram
, char *mem
, u32 index
)
576 struct zram_meta
*meta
= zram
->meta
;
577 unsigned long handle
;
580 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
581 handle
= meta
->table
[index
].handle
;
582 size
= zram_get_obj_size(meta
, index
);
584 if (!handle
|| zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
585 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
590 cmem
= zs_map_object(meta
->mem_pool
, handle
, ZS_MM_RO
);
591 if (size
== PAGE_SIZE
) {
592 copy_page(mem
, cmem
);
594 struct zcomp_strm
*zstrm
= zcomp_stream_get(zram
->comp
);
596 ret
= zcomp_decompress(zstrm
, cmem
, size
, mem
);
597 zcomp_stream_put(zram
->comp
);
599 zs_unmap_object(meta
->mem_pool
, handle
);
600 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
602 /* Should NEVER happen. Return bio error if it does. */
604 pr_err("Decompression failed! err=%d, page=%u\n", ret
, index
);
611 static int zram_bvec_read(struct zram
*zram
, struct bio_vec
*bvec
,
612 u32 index
, int offset
)
616 unsigned char *user_mem
, *uncmem
= NULL
;
617 struct zram_meta
*meta
= zram
->meta
;
618 page
= bvec
->bv_page
;
620 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
621 if (unlikely(!meta
->table
[index
].handle
) ||
622 zram_test_flag(meta
, index
, ZRAM_ZERO
)) {
623 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
624 handle_zero_page(bvec
);
627 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
629 if (is_partial_io(bvec
))
630 /* Use a temporary buffer to decompress the page */
631 uncmem
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
633 user_mem
= kmap_atomic(page
);
634 if (!is_partial_io(bvec
))
638 pr_err("Unable to allocate temp memory\n");
643 ret
= zram_decompress_page(zram
, uncmem
, index
);
644 /* Should NEVER happen. Return bio error if it does. */
648 if (is_partial_io(bvec
))
649 memcpy(user_mem
+ bvec
->bv_offset
, uncmem
+ offset
,
652 flush_dcache_page(page
);
655 kunmap_atomic(user_mem
);
656 if (is_partial_io(bvec
))
661 static int zram_bvec_write(struct zram
*zram
, struct bio_vec
*bvec
, u32 index
,
666 unsigned long handle
= 0;
668 unsigned char *user_mem
, *cmem
, *src
, *uncmem
= NULL
;
669 struct zram_meta
*meta
= zram
->meta
;
670 struct zcomp_strm
*zstrm
= NULL
;
671 unsigned long alloced_pages
;
673 page
= bvec
->bv_page
;
674 if (is_partial_io(bvec
)) {
676 * This is a partial IO. We need to read the full page
677 * before to write the changes.
679 uncmem
= kmalloc(PAGE_SIZE
, GFP_NOIO
);
684 ret
= zram_decompress_page(zram
, uncmem
, index
);
690 user_mem
= kmap_atomic(page
);
691 if (is_partial_io(bvec
)) {
692 memcpy(uncmem
+ offset
, user_mem
+ bvec
->bv_offset
,
694 kunmap_atomic(user_mem
);
700 if (page_zero_filled(uncmem
)) {
702 kunmap_atomic(user_mem
);
703 /* Free memory associated with this sector now. */
704 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
705 zram_free_page(zram
, index
);
706 zram_set_flag(meta
, index
, ZRAM_ZERO
);
707 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
709 atomic64_inc(&zram
->stats
.zero_pages
);
714 zstrm
= zcomp_stream_get(zram
->comp
);
715 ret
= zcomp_compress(zstrm
, uncmem
, &clen
);
716 if (!is_partial_io(bvec
)) {
717 kunmap_atomic(user_mem
);
723 pr_err("Compression failed! err=%d\n", ret
);
728 if (unlikely(clen
> max_zpage_size
)) {
730 if (is_partial_io(bvec
))
735 * handle allocation has 2 paths:
736 * a) fast path is executed with preemption disabled (for
737 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
738 * since we can't sleep;
739 * b) slow path enables preemption and attempts to allocate
740 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
741 * put per-cpu compression stream and, thus, to re-do
742 * the compression once handle is allocated.
744 * if we have a 'non-null' handle here then we are coming
745 * from the slow path and handle has already been allocated.
748 handle
= zs_malloc(meta
->mem_pool
, clen
,
749 __GFP_KSWAPD_RECLAIM
|
754 zcomp_stream_put(zram
->comp
);
757 atomic64_inc(&zram
->stats
.writestall
);
759 handle
= zs_malloc(meta
->mem_pool
, clen
,
760 GFP_NOIO
| __GFP_HIGHMEM
|
765 pr_err("Error allocating memory for compressed page: %u, size=%u\n",
771 alloced_pages
= zs_get_total_pages(meta
->mem_pool
);
772 update_used_max(zram
, alloced_pages
);
774 if (zram
->limit_pages
&& alloced_pages
> zram
->limit_pages
) {
775 zs_free(meta
->mem_pool
, handle
);
780 cmem
= zs_map_object(meta
->mem_pool
, handle
, ZS_MM_WO
);
782 if ((clen
== PAGE_SIZE
) && !is_partial_io(bvec
)) {
783 src
= kmap_atomic(page
);
784 copy_page(cmem
, src
);
787 memcpy(cmem
, src
, clen
);
790 zcomp_stream_put(zram
->comp
);
792 zs_unmap_object(meta
->mem_pool
, handle
);
795 * Free memory associated with this sector
796 * before overwriting unused sectors.
798 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
799 zram_free_page(zram
, index
);
801 meta
->table
[index
].handle
= handle
;
802 zram_set_obj_size(meta
, index
, clen
);
803 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
806 atomic64_add(clen
, &zram
->stats
.compr_data_size
);
807 atomic64_inc(&zram
->stats
.pages_stored
);
810 zcomp_stream_put(zram
->comp
);
811 if (is_partial_io(bvec
))
817 * zram_bio_discard - handler on discard request
818 * @index: physical block index in PAGE_SIZE units
819 * @offset: byte offset within physical block
821 static void zram_bio_discard(struct zram
*zram
, u32 index
,
822 int offset
, struct bio
*bio
)
824 size_t n
= bio
->bi_iter
.bi_size
;
825 struct zram_meta
*meta
= zram
->meta
;
828 * zram manages data in physical block size units. Because logical block
829 * size isn't identical with physical block size on some arch, we
830 * could get a discard request pointing to a specific offset within a
831 * certain physical block. Although we can handle this request by
832 * reading that physiclal block and decompressing and partially zeroing
833 * and re-compressing and then re-storing it, this isn't reasonable
834 * because our intent with a discard request is to save memory. So
835 * skipping this logical block is appropriate here.
838 if (n
<= (PAGE_SIZE
- offset
))
841 n
-= (PAGE_SIZE
- offset
);
845 while (n
>= PAGE_SIZE
) {
846 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
847 zram_free_page(zram
, index
);
848 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
849 atomic64_inc(&zram
->stats
.notify_free
);
855 static int zram_bvec_rw(struct zram
*zram
, struct bio_vec
*bvec
, u32 index
,
856 int offset
, bool is_write
)
858 unsigned long start_time
= jiffies
;
859 int rw_acct
= is_write
? REQ_OP_WRITE
: REQ_OP_READ
;
862 generic_start_io_acct(rw_acct
, bvec
->bv_len
>> SECTOR_SHIFT
,
866 atomic64_inc(&zram
->stats
.num_reads
);
867 ret
= zram_bvec_read(zram
, bvec
, index
, offset
);
869 atomic64_inc(&zram
->stats
.num_writes
);
870 ret
= zram_bvec_write(zram
, bvec
, index
, offset
);
873 generic_end_io_acct(rw_acct
, &zram
->disk
->part0
, start_time
);
877 atomic64_inc(&zram
->stats
.failed_reads
);
879 atomic64_inc(&zram
->stats
.failed_writes
);
885 static void __zram_make_request(struct zram
*zram
, struct bio
*bio
)
890 struct bvec_iter iter
;
892 index
= bio
->bi_iter
.bi_sector
>> SECTORS_PER_PAGE_SHIFT
;
893 offset
= (bio
->bi_iter
.bi_sector
&
894 (SECTORS_PER_PAGE
- 1)) << SECTOR_SHIFT
;
896 if (unlikely(bio_op(bio
) == REQ_OP_DISCARD
)) {
897 zram_bio_discard(zram
, index
, offset
, bio
);
902 bio_for_each_segment(bvec
, bio
, iter
) {
903 int max_transfer_size
= PAGE_SIZE
- offset
;
905 if (bvec
.bv_len
> max_transfer_size
) {
907 * zram_bvec_rw() can only make operation on a single
908 * zram page. Split the bio vector.
912 bv
.bv_page
= bvec
.bv_page
;
913 bv
.bv_len
= max_transfer_size
;
914 bv
.bv_offset
= bvec
.bv_offset
;
916 if (zram_bvec_rw(zram
, &bv
, index
, offset
,
917 op_is_write(bio_op(bio
))) < 0)
920 bv
.bv_len
= bvec
.bv_len
- max_transfer_size
;
921 bv
.bv_offset
+= max_transfer_size
;
922 if (zram_bvec_rw(zram
, &bv
, index
+ 1, 0,
923 op_is_write(bio_op(bio
))) < 0)
926 if (zram_bvec_rw(zram
, &bvec
, index
, offset
,
927 op_is_write(bio_op(bio
))) < 0)
930 update_position(&index
, &offset
, &bvec
);
941 * Handler function for all zram I/O requests.
943 static blk_qc_t
zram_make_request(struct request_queue
*queue
, struct bio
*bio
)
945 struct zram
*zram
= queue
->queuedata
;
947 if (unlikely(!zram_meta_get(zram
)))
950 blk_queue_split(queue
, &bio
, queue
->bio_split
);
952 if (!valid_io_request(zram
, bio
->bi_iter
.bi_sector
,
953 bio
->bi_iter
.bi_size
)) {
954 atomic64_inc(&zram
->stats
.invalid_io
);
958 __zram_make_request(zram
, bio
);
960 return BLK_QC_T_NONE
;
965 return BLK_QC_T_NONE
;
968 static void zram_slot_free_notify(struct block_device
*bdev
,
972 struct zram_meta
*meta
;
974 zram
= bdev
->bd_disk
->private_data
;
977 bit_spin_lock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
978 zram_free_page(zram
, index
);
979 bit_spin_unlock(ZRAM_ACCESS
, &meta
->table
[index
].value
);
980 atomic64_inc(&zram
->stats
.notify_free
);
983 static int zram_rw_page(struct block_device
*bdev
, sector_t sector
,
984 struct page
*page
, bool is_write
)
986 int offset
, err
= -EIO
;
991 zram
= bdev
->bd_disk
->private_data
;
992 if (unlikely(!zram_meta_get(zram
)))
995 if (!valid_io_request(zram
, sector
, PAGE_SIZE
)) {
996 atomic64_inc(&zram
->stats
.invalid_io
);
1001 index
= sector
>> SECTORS_PER_PAGE_SHIFT
;
1002 offset
= sector
& (SECTORS_PER_PAGE
- 1) << SECTOR_SHIFT
;
1005 bv
.bv_len
= PAGE_SIZE
;
1008 err
= zram_bvec_rw(zram
, &bv
, index
, offset
, is_write
);
1010 zram_meta_put(zram
);
1013 * If I/O fails, just return error(ie, non-zero) without
1014 * calling page_endio.
1015 * It causes resubmit the I/O with bio request by upper functions
1016 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1017 * bio->bi_end_io does things to handle the error
1018 * (e.g., SetPageError, set_page_dirty and extra works).
1021 page_endio(page
, is_write
, 0);
1025 static void zram_reset_device(struct zram
*zram
)
1027 struct zram_meta
*meta
;
1031 down_write(&zram
->init_lock
);
1033 zram
->limit_pages
= 0;
1035 if (!init_done(zram
)) {
1036 up_write(&zram
->init_lock
);
1042 disksize
= zram
->disksize
;
1044 * Refcount will go down to 0 eventually and r/w handler
1045 * cannot handle further I/O so it will bail out by
1046 * check zram_meta_get.
1048 zram_meta_put(zram
);
1050 * We want to free zram_meta in process context to avoid
1051 * deadlock between reclaim path and any other locks.
1053 wait_event(zram
->io_done
, atomic_read(&zram
->refcount
) == 0);
1056 memset(&zram
->stats
, 0, sizeof(zram
->stats
));
1059 set_capacity(zram
->disk
, 0);
1060 part_stat_set_all(&zram
->disk
->part0
, 0);
1062 up_write(&zram
->init_lock
);
1063 /* I/O operation under all of CPU are done so let's free */
1064 zram_meta_free(meta
, disksize
);
1065 zcomp_destroy(comp
);
1068 static ssize_t
disksize_store(struct device
*dev
,
1069 struct device_attribute
*attr
, const char *buf
, size_t len
)
1073 struct zram_meta
*meta
;
1074 struct zram
*zram
= dev_to_zram(dev
);
1077 disksize
= memparse(buf
, NULL
);
1081 disksize
= PAGE_ALIGN(disksize
);
1082 meta
= zram_meta_alloc(zram
->disk
->disk_name
, disksize
);
1086 comp
= zcomp_create(zram
->compressor
);
1088 pr_err("Cannot initialise %s compressing backend\n",
1090 err
= PTR_ERR(comp
);
1094 down_write(&zram
->init_lock
);
1095 if (init_done(zram
)) {
1096 pr_info("Cannot change disksize for initialized device\n");
1098 goto out_destroy_comp
;
1101 init_waitqueue_head(&zram
->io_done
);
1102 atomic_set(&zram
->refcount
, 1);
1105 zram
->disksize
= disksize
;
1106 set_capacity(zram
->disk
, zram
->disksize
>> SECTOR_SHIFT
);
1107 zram_revalidate_disk(zram
);
1108 up_write(&zram
->init_lock
);
1113 up_write(&zram
->init_lock
);
1114 zcomp_destroy(comp
);
1116 zram_meta_free(meta
, disksize
);
1120 static ssize_t
reset_store(struct device
*dev
,
1121 struct device_attribute
*attr
, const char *buf
, size_t len
)
1124 unsigned short do_reset
;
1126 struct block_device
*bdev
;
1128 ret
= kstrtou16(buf
, 10, &do_reset
);
1135 zram
= dev_to_zram(dev
);
1136 bdev
= bdget_disk(zram
->disk
, 0);
1140 mutex_lock(&bdev
->bd_mutex
);
1141 /* Do not reset an active device or claimed device */
1142 if (bdev
->bd_openers
|| zram
->claim
) {
1143 mutex_unlock(&bdev
->bd_mutex
);
1148 /* From now on, anyone can't open /dev/zram[0-9] */
1150 mutex_unlock(&bdev
->bd_mutex
);
1152 /* Make sure all the pending I/O are finished */
1154 zram_reset_device(zram
);
1155 zram_revalidate_disk(zram
);
1158 mutex_lock(&bdev
->bd_mutex
);
1159 zram
->claim
= false;
1160 mutex_unlock(&bdev
->bd_mutex
);
1165 static int zram_open(struct block_device
*bdev
, fmode_t mode
)
1170 WARN_ON(!mutex_is_locked(&bdev
->bd_mutex
));
1172 zram
= bdev
->bd_disk
->private_data
;
1173 /* zram was claimed to reset so open request fails */
1180 static const struct block_device_operations zram_devops
= {
1182 .swap_slot_free_notify
= zram_slot_free_notify
,
1183 .rw_page
= zram_rw_page
,
1184 .owner
= THIS_MODULE
1187 static DEVICE_ATTR_WO(compact
);
1188 static DEVICE_ATTR_RW(disksize
);
1189 static DEVICE_ATTR_RO(initstate
);
1190 static DEVICE_ATTR_WO(reset
);
1191 static DEVICE_ATTR_RO(orig_data_size
);
1192 static DEVICE_ATTR_RO(mem_used_total
);
1193 static DEVICE_ATTR_RW(mem_limit
);
1194 static DEVICE_ATTR_RW(mem_used_max
);
1195 static DEVICE_ATTR_RW(max_comp_streams
);
1196 static DEVICE_ATTR_RW(comp_algorithm
);
1198 static struct attribute
*zram_disk_attrs
[] = {
1199 &dev_attr_disksize
.attr
,
1200 &dev_attr_initstate
.attr
,
1201 &dev_attr_reset
.attr
,
1202 &dev_attr_num_reads
.attr
,
1203 &dev_attr_num_writes
.attr
,
1204 &dev_attr_failed_reads
.attr
,
1205 &dev_attr_failed_writes
.attr
,
1206 &dev_attr_compact
.attr
,
1207 &dev_attr_invalid_io
.attr
,
1208 &dev_attr_notify_free
.attr
,
1209 &dev_attr_zero_pages
.attr
,
1210 &dev_attr_orig_data_size
.attr
,
1211 &dev_attr_compr_data_size
.attr
,
1212 &dev_attr_mem_used_total
.attr
,
1213 &dev_attr_mem_limit
.attr
,
1214 &dev_attr_mem_used_max
.attr
,
1215 &dev_attr_max_comp_streams
.attr
,
1216 &dev_attr_comp_algorithm
.attr
,
1217 &dev_attr_io_stat
.attr
,
1218 &dev_attr_mm_stat
.attr
,
1219 &dev_attr_debug_stat
.attr
,
1223 static struct attribute_group zram_disk_attr_group
= {
1224 .attrs
= zram_disk_attrs
,
1228 * Allocate and initialize new zram device. the function returns
1229 * '>= 0' device_id upon success, and negative value otherwise.
1231 static int zram_add(void)
1234 struct request_queue
*queue
;
1237 zram
= kzalloc(sizeof(struct zram
), GFP_KERNEL
);
1241 ret
= idr_alloc(&zram_index_idr
, zram
, 0, 0, GFP_KERNEL
);
1246 init_rwsem(&zram
->init_lock
);
1248 queue
= blk_alloc_queue(GFP_KERNEL
);
1250 pr_err("Error allocating disk queue for device %d\n",
1256 blk_queue_make_request(queue
, zram_make_request
);
1258 /* gendisk structure */
1259 zram
->disk
= alloc_disk(1);
1261 pr_err("Error allocating disk structure for device %d\n",
1264 goto out_free_queue
;
1267 zram
->disk
->major
= zram_major
;
1268 zram
->disk
->first_minor
= device_id
;
1269 zram
->disk
->fops
= &zram_devops
;
1270 zram
->disk
->queue
= queue
;
1271 zram
->disk
->queue
->queuedata
= zram
;
1272 zram
->disk
->private_data
= zram
;
1273 snprintf(zram
->disk
->disk_name
, 16, "zram%d", device_id
);
1275 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1276 set_capacity(zram
->disk
, 0);
1277 /* zram devices sort of resembles non-rotational disks */
1278 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, zram
->disk
->queue
);
1279 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM
, zram
->disk
->queue
);
1281 * To ensure that we always get PAGE_SIZE aligned
1282 * and n*PAGE_SIZED sized I/O requests.
1284 blk_queue_physical_block_size(zram
->disk
->queue
, PAGE_SIZE
);
1285 blk_queue_logical_block_size(zram
->disk
->queue
,
1286 ZRAM_LOGICAL_BLOCK_SIZE
);
1287 blk_queue_io_min(zram
->disk
->queue
, PAGE_SIZE
);
1288 blk_queue_io_opt(zram
->disk
->queue
, PAGE_SIZE
);
1289 zram
->disk
->queue
->limits
.discard_granularity
= PAGE_SIZE
;
1290 blk_queue_max_discard_sectors(zram
->disk
->queue
, UINT_MAX
);
1292 * zram_bio_discard() will clear all logical blocks if logical block
1293 * size is identical with physical block size(PAGE_SIZE). But if it is
1294 * different, we will skip discarding some parts of logical blocks in
1295 * the part of the request range which isn't aligned to physical block
1296 * size. So we can't ensure that all discarded logical blocks are
1299 if (ZRAM_LOGICAL_BLOCK_SIZE
== PAGE_SIZE
)
1300 zram
->disk
->queue
->limits
.discard_zeroes_data
= 1;
1302 zram
->disk
->queue
->limits
.discard_zeroes_data
= 0;
1303 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, zram
->disk
->queue
);
1305 add_disk(zram
->disk
);
1307 ret
= sysfs_create_group(&disk_to_dev(zram
->disk
)->kobj
,
1308 &zram_disk_attr_group
);
1310 pr_err("Error creating sysfs group for device %d\n",
1314 strlcpy(zram
->compressor
, default_compressor
, sizeof(zram
->compressor
));
1317 pr_info("Added device: %s\n", zram
->disk
->disk_name
);
1321 del_gendisk(zram
->disk
);
1322 put_disk(zram
->disk
);
1324 blk_cleanup_queue(queue
);
1326 idr_remove(&zram_index_idr
, device_id
);
1332 static int zram_remove(struct zram
*zram
)
1334 struct block_device
*bdev
;
1336 bdev
= bdget_disk(zram
->disk
, 0);
1340 mutex_lock(&bdev
->bd_mutex
);
1341 if (bdev
->bd_openers
|| zram
->claim
) {
1342 mutex_unlock(&bdev
->bd_mutex
);
1348 mutex_unlock(&bdev
->bd_mutex
);
1351 * Remove sysfs first, so no one will perform a disksize
1352 * store while we destroy the devices. This also helps during
1353 * hot_remove -- zram_reset_device() is the last holder of
1354 * ->init_lock, no later/concurrent disksize_store() or any
1355 * other sysfs handlers are possible.
1357 sysfs_remove_group(&disk_to_dev(zram
->disk
)->kobj
,
1358 &zram_disk_attr_group
);
1360 /* Make sure all the pending I/O are finished */
1362 zram_reset_device(zram
);
1365 pr_info("Removed device: %s\n", zram
->disk
->disk_name
);
1367 blk_cleanup_queue(zram
->disk
->queue
);
1368 del_gendisk(zram
->disk
);
1369 put_disk(zram
->disk
);
1374 /* zram-control sysfs attributes */
1375 static ssize_t
hot_add_show(struct class *class,
1376 struct class_attribute
*attr
,
1381 mutex_lock(&zram_index_mutex
);
1383 mutex_unlock(&zram_index_mutex
);
1387 return scnprintf(buf
, PAGE_SIZE
, "%d\n", ret
);
1390 static ssize_t
hot_remove_store(struct class *class,
1391 struct class_attribute
*attr
,
1398 /* dev_id is gendisk->first_minor, which is `int' */
1399 ret
= kstrtoint(buf
, 10, &dev_id
);
1405 mutex_lock(&zram_index_mutex
);
1407 zram
= idr_find(&zram_index_idr
, dev_id
);
1409 ret
= zram_remove(zram
);
1411 idr_remove(&zram_index_idr
, dev_id
);
1416 mutex_unlock(&zram_index_mutex
);
1417 return ret
? ret
: count
;
1421 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1422 * sense that reading from this file does alter the state of your system -- it
1423 * creates a new un-initialized zram device and returns back this device's
1424 * device_id (or an error code if it fails to create a new device).
1426 static struct class_attribute zram_control_class_attrs
[] = {
1427 __ATTR(hot_add
, 0400, hot_add_show
, NULL
),
1428 __ATTR_WO(hot_remove
),
1432 static struct class zram_control_class
= {
1433 .name
= "zram-control",
1434 .owner
= THIS_MODULE
,
1435 .class_attrs
= zram_control_class_attrs
,
1438 static int zram_remove_cb(int id
, void *ptr
, void *data
)
1444 static void destroy_devices(void)
1446 class_unregister(&zram_control_class
);
1447 idr_for_each(&zram_index_idr
, &zram_remove_cb
, NULL
);
1448 idr_destroy(&zram_index_idr
);
1449 unregister_blkdev(zram_major
, "zram");
1450 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE
);
1453 static int __init
zram_init(void)
1457 ret
= cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE
, "block/zram:prepare",
1458 zcomp_cpu_up_prepare
, zcomp_cpu_dead
);
1462 ret
= class_register(&zram_control_class
);
1464 pr_err("Unable to register zram-control class\n");
1465 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE
);
1469 zram_major
= register_blkdev(0, "zram");
1470 if (zram_major
<= 0) {
1471 pr_err("Unable to get major number\n");
1472 class_unregister(&zram_control_class
);
1473 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE
);
1477 while (num_devices
!= 0) {
1478 mutex_lock(&zram_index_mutex
);
1480 mutex_unlock(&zram_index_mutex
);
1493 static void __exit
zram_exit(void)
1498 module_init(zram_init
);
1499 module_exit(zram_exit
);
1501 module_param(num_devices
, uint
, 0);
1502 MODULE_PARM_DESC(num_devices
, "Number of pre-created zram devices");
1504 MODULE_LICENSE("Dual BSD/GPL");
1505 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1506 MODULE_DESCRIPTION("Compressed RAM Block Device");