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 #ifdef CONFIG_ZRAM_DEBUG
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/bio.h>
25 #include <linux/bitops.h>
26 #include <linux/blkdev.h>
27 #include <linux/buffer_head.h>
28 #include <linux/device.h>
29 #include <linux/genhd.h>
30 #include <linux/highmem.h>
31 #include <linux/slab.h>
32 #include <linux/string.h>
33 #include <linux/vmalloc.h>
34 #include <linux/err.h>
39 static int zram_major;
40 static struct zram *zram_devices;
41 static const char *default_compressor = "lzo";
43 /* Module params (documentation at end) */
44 static unsigned int num_devices = 1;
46 #define ZRAM_ATTR_RO(name) \
47 static ssize_t zram_attr_##name##_show(struct device *d, \
48 struct device_attribute *attr, char *b) \
50 struct zram *zram = dev_to_zram(d); \
51 return sprintf(b, "%llu\n", \
52 (u64)atomic64_read(&zram->stats.name)); \
54 static struct device_attribute dev_attr_##name = \
55 __ATTR(name, S_IRUGO, zram_attr_##name##_show, NULL);
57 static inline int init_done(struct zram *zram)
59 return zram->meta != NULL;
62 static inline struct zram *dev_to_zram(struct device *dev)
64 return (struct zram *)dev_to_disk(dev)->private_data;
67 static ssize_t disksize_show(struct device *dev,
68 struct device_attribute *attr, char *buf)
70 struct zram *zram = dev_to_zram(dev);
72 return sprintf(buf, "%llu\n", zram->disksize);
75 static ssize_t initstate_show(struct device *dev,
76 struct device_attribute *attr, char *buf)
79 struct zram *zram = dev_to_zram(dev);
81 down_read(&zram->init_lock);
82 val = init_done(zram);
83 up_read(&zram->init_lock);
85 return sprintf(buf, "%u\n", val);
88 static ssize_t orig_data_size_show(struct device *dev,
89 struct device_attribute *attr, char *buf)
91 struct zram *zram = dev_to_zram(dev);
93 return sprintf(buf, "%llu\n",
94 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
97 static ssize_t mem_used_total_show(struct device *dev,
98 struct device_attribute *attr, char *buf)
101 struct zram *zram = dev_to_zram(dev);
102 struct zram_meta *meta = zram->meta;
104 down_read(&zram->init_lock);
106 val = zs_get_total_size_bytes(meta->mem_pool);
107 up_read(&zram->init_lock);
109 return sprintf(buf, "%llu\n", val);
112 static ssize_t max_comp_streams_show(struct device *dev,
113 struct device_attribute *attr, char *buf)
116 struct zram *zram = dev_to_zram(dev);
118 down_read(&zram->init_lock);
119 val = zram->max_comp_streams;
120 up_read(&zram->init_lock);
122 return sprintf(buf, "%d\n", val);
125 static ssize_t max_comp_streams_store(struct device *dev,
126 struct device_attribute *attr, const char *buf, size_t len)
129 struct zram *zram = dev_to_zram(dev);
132 ret = kstrtoint(buf, 0, &num);
138 down_write(&zram->init_lock);
139 if (init_done(zram)) {
140 if (!zcomp_set_max_streams(zram->comp, num)) {
141 pr_info("Cannot change max compression streams\n");
147 zram->max_comp_streams = num;
150 up_write(&zram->init_lock);
154 static ssize_t comp_algorithm_show(struct device *dev,
155 struct device_attribute *attr, char *buf)
158 struct zram *zram = dev_to_zram(dev);
160 down_read(&zram->init_lock);
161 sz = zcomp_available_show(zram->compressor, buf);
162 up_read(&zram->init_lock);
167 static ssize_t comp_algorithm_store(struct device *dev,
168 struct device_attribute *attr, const char *buf, size_t len)
170 struct zram *zram = dev_to_zram(dev);
171 down_write(&zram->init_lock);
172 if (init_done(zram)) {
173 up_write(&zram->init_lock);
174 pr_info("Can't change algorithm for initialized device\n");
177 strlcpy(zram->compressor, buf, sizeof(zram->compressor));
178 up_write(&zram->init_lock);
182 /* flag operations needs meta->tb_lock */
183 static int zram_test_flag(struct zram_meta *meta, u32 index,
184 enum zram_pageflags flag)
186 return meta->table[index].flags & BIT(flag);
189 static void zram_set_flag(struct zram_meta *meta, u32 index,
190 enum zram_pageflags flag)
192 meta->table[index].flags |= BIT(flag);
195 static void zram_clear_flag(struct zram_meta *meta, u32 index,
196 enum zram_pageflags flag)
198 meta->table[index].flags &= ~BIT(flag);
201 static inline int is_partial_io(struct bio_vec *bvec)
203 return bvec->bv_len != PAGE_SIZE;
207 * Check if request is within bounds and aligned on zram logical blocks.
209 static inline int valid_io_request(struct zram *zram, struct bio *bio)
211 u64 start, end, bound;
213 /* unaligned request */
214 if (unlikely(bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
216 if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
219 start = bio->bi_sector;
220 end = start + (bio->bi_size >> SECTOR_SHIFT);
221 bound = zram->disksize >> SECTOR_SHIFT;
222 /* out of range range */
223 if (unlikely(start >= bound || end >= bound || start > end))
226 /* I/O request is valid */
230 static void zram_meta_free(struct zram_meta *meta)
232 zs_destroy_pool(meta->mem_pool);
237 static struct zram_meta *zram_meta_alloc(u64 disksize)
240 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
244 num_pages = disksize >> PAGE_SHIFT;
245 meta->table = vzalloc(num_pages * sizeof(*meta->table));
247 pr_err("Error allocating zram address table\n");
251 meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
252 if (!meta->mem_pool) {
253 pr_err("Error creating memory pool\n");
257 rwlock_init(&meta->tb_lock);
269 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
271 if (*offset + bvec->bv_len >= PAGE_SIZE)
273 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
276 static int page_zero_filled(void *ptr)
281 page = (unsigned long *)ptr;
283 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
291 static void handle_zero_page(struct bio_vec *bvec)
293 struct page *page = bvec->bv_page;
296 user_mem = kmap_atomic(page);
297 if (is_partial_io(bvec))
298 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
300 clear_page(user_mem);
301 kunmap_atomic(user_mem);
303 flush_dcache_page(page);
306 /* NOTE: caller should hold meta->tb_lock with write-side */
307 static void zram_free_page(struct zram *zram, size_t index)
309 struct zram_meta *meta = zram->meta;
310 unsigned long handle = meta->table[index].handle;
312 if (unlikely(!handle)) {
314 * No memory is allocated for zero filled pages.
315 * Simply clear zero page flag.
317 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
318 zram_clear_flag(meta, index, ZRAM_ZERO);
319 atomic64_dec(&zram->stats.zero_pages);
324 zs_free(meta->mem_pool, handle);
326 atomic64_sub(meta->table[index].size, &zram->stats.compr_data_size);
327 atomic64_dec(&zram->stats.pages_stored);
329 meta->table[index].handle = 0;
330 meta->table[index].size = 0;
333 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
337 struct zram_meta *meta = zram->meta;
338 unsigned long handle;
341 read_lock(&meta->tb_lock);
342 handle = meta->table[index].handle;
343 size = meta->table[index].size;
345 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
346 read_unlock(&meta->tb_lock);
351 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
352 if (size == PAGE_SIZE)
353 copy_page(mem, cmem);
355 ret = zcomp_decompress(zram->comp, cmem, size, mem);
356 zs_unmap_object(meta->mem_pool, handle);
357 read_unlock(&meta->tb_lock);
359 /* Should NEVER happen. Return bio error if it does. */
361 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
362 atomic64_inc(&zram->stats.failed_reads);
369 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
370 u32 index, int offset, struct bio *bio)
374 unsigned char *user_mem, *uncmem = NULL;
375 struct zram_meta *meta = zram->meta;
376 page = bvec->bv_page;
378 read_lock(&meta->tb_lock);
379 if (unlikely(!meta->table[index].handle) ||
380 zram_test_flag(meta, index, ZRAM_ZERO)) {
381 read_unlock(&meta->tb_lock);
382 handle_zero_page(bvec);
385 read_unlock(&meta->tb_lock);
387 if (is_partial_io(bvec))
388 /* Use a temporary buffer to decompress the page */
389 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
391 user_mem = kmap_atomic(page);
392 if (!is_partial_io(bvec))
396 pr_info("Unable to allocate temp memory\n");
401 ret = zram_decompress_page(zram, uncmem, index);
402 /* Should NEVER happen. Return bio error if it does. */
406 if (is_partial_io(bvec))
407 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
410 flush_dcache_page(page);
413 kunmap_atomic(user_mem);
414 if (is_partial_io(bvec))
419 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
424 unsigned long handle;
426 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
427 struct zram_meta *meta = zram->meta;
428 struct zcomp_strm *zstrm;
431 page = bvec->bv_page;
432 if (is_partial_io(bvec)) {
434 * This is a partial IO. We need to read the full page
435 * before to write the changes.
437 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
442 ret = zram_decompress_page(zram, uncmem, index);
447 zstrm = zcomp_strm_find(zram->comp);
449 user_mem = kmap_atomic(page);
451 if (is_partial_io(bvec)) {
452 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
454 kunmap_atomic(user_mem);
460 if (page_zero_filled(uncmem)) {
461 kunmap_atomic(user_mem);
462 /* Free memory associated with this sector now. */
463 write_lock(&zram->meta->tb_lock);
464 zram_free_page(zram, index);
465 zram_set_flag(meta, index, ZRAM_ZERO);
466 write_unlock(&zram->meta->tb_lock);
468 atomic64_inc(&zram->stats.zero_pages);
473 ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
474 if (!is_partial_io(bvec)) {
475 kunmap_atomic(user_mem);
481 pr_err("Compression failed! err=%d\n", ret);
485 if (unlikely(clen > max_zpage_size)) {
487 if (is_partial_io(bvec))
491 handle = zs_malloc(meta->mem_pool, clen);
493 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
498 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
500 if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
501 src = kmap_atomic(page);
502 copy_page(cmem, src);
505 memcpy(cmem, src, clen);
508 zcomp_strm_release(zram->comp, zstrm);
510 zs_unmap_object(meta->mem_pool, handle);
513 * Free memory associated with this sector
514 * before overwriting unused sectors.
516 write_lock(&zram->meta->tb_lock);
517 zram_free_page(zram, index);
519 meta->table[index].handle = handle;
520 meta->table[index].size = clen;
521 write_unlock(&zram->meta->tb_lock);
524 atomic64_add(clen, &zram->stats.compr_data_size);
525 atomic64_inc(&zram->stats.pages_stored);
528 zcomp_strm_release(zram->comp, zstrm);
529 if (is_partial_io(bvec))
532 atomic64_inc(&zram->stats.failed_writes);
536 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
537 int offset, struct bio *bio)
540 int rw = bio_data_dir(bio);
543 atomic64_inc(&zram->stats.num_reads);
544 ret = zram_bvec_read(zram, bvec, index, offset, bio);
546 atomic64_inc(&zram->stats.num_writes);
547 ret = zram_bvec_write(zram, bvec, index, offset);
553 static void zram_reset_device(struct zram *zram, bool reset_capacity)
556 struct zram_meta *meta;
558 down_write(&zram->init_lock);
559 if (!init_done(zram)) {
560 up_write(&zram->init_lock);
565 /* Free all pages that are still in this zram device */
566 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
567 unsigned long handle = meta->table[index].handle;
571 zs_free(meta->mem_pool, handle);
574 zcomp_destroy(zram->comp);
575 zram->max_comp_streams = 1;
577 zram_meta_free(zram->meta);
580 memset(&zram->stats, 0, sizeof(zram->stats));
584 set_capacity(zram->disk, 0);
585 up_write(&zram->init_lock);
588 static ssize_t disksize_store(struct device *dev,
589 struct device_attribute *attr, const char *buf, size_t len)
593 struct zram_meta *meta;
594 struct zram *zram = dev_to_zram(dev);
597 disksize = memparse(buf, NULL);
601 disksize = PAGE_ALIGN(disksize);
602 meta = zram_meta_alloc(disksize);
606 comp = zcomp_create(zram->compressor, zram->max_comp_streams);
608 pr_info("Cannot initialise %s compressing backend\n",
614 down_write(&zram->init_lock);
615 if (init_done(zram)) {
616 pr_info("Cannot change disksize for initialized device\n");
618 goto out_destroy_comp;
623 zram->disksize = disksize;
624 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
625 up_write(&zram->init_lock);
629 up_write(&zram->init_lock);
632 zram_meta_free(meta);
636 static ssize_t reset_store(struct device *dev,
637 struct device_attribute *attr, const char *buf, size_t len)
640 unsigned short do_reset;
642 struct block_device *bdev;
644 zram = dev_to_zram(dev);
645 bdev = bdget_disk(zram->disk, 0);
650 /* Do not reset an active device! */
651 if (bdev->bd_holders) {
656 ret = kstrtou16(buf, 10, &do_reset);
665 /* Make sure all pending I/O is finished */
669 zram_reset_device(zram, true);
677 static void __zram_make_request(struct zram *zram, struct bio *bio)
681 struct bio_vec *bvec;
684 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
685 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
687 bio_for_each_segment(bvec, bio, i) {
688 int max_transfer_size = PAGE_SIZE - offset;
690 if (bvec->bv_len > max_transfer_size) {
692 * zram_bvec_rw() can only make operation on a single
693 * zram page. Split the bio vector.
697 bv.bv_page = bvec->bv_page;
698 bv.bv_len = max_transfer_size;
699 bv.bv_offset = bvec->bv_offset;
701 if (zram_bvec_rw(zram, &bv, index, offset, bio) < 0)
704 bv.bv_len = bvec->bv_len - max_transfer_size;
705 bv.bv_offset += max_transfer_size;
706 if (zram_bvec_rw(zram, &bv, index + 1, 0, bio) < 0)
709 if (zram_bvec_rw(zram, bvec, index, offset, bio) < 0)
712 update_position(&index, &offset, bvec);
715 set_bit(BIO_UPTODATE, &bio->bi_flags);
724 * Handler function for all zram I/O requests.
726 static void zram_make_request(struct request_queue *queue, struct bio *bio)
728 struct zram *zram = queue->queuedata;
730 down_read(&zram->init_lock);
731 if (unlikely(!init_done(zram)))
734 if (!valid_io_request(zram, bio)) {
735 atomic64_inc(&zram->stats.invalid_io);
739 __zram_make_request(zram, bio);
740 up_read(&zram->init_lock);
745 up_read(&zram->init_lock);
749 static void zram_slot_free_notify(struct block_device *bdev,
753 struct zram_meta *meta;
755 zram = bdev->bd_disk->private_data;
758 write_lock(&meta->tb_lock);
759 zram_free_page(zram, index);
760 write_unlock(&meta->tb_lock);
761 atomic64_inc(&zram->stats.notify_free);
764 static const struct block_device_operations zram_devops = {
765 .swap_slot_free_notify = zram_slot_free_notify,
769 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
770 disksize_show, disksize_store);
771 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
772 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
773 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
774 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
775 static DEVICE_ATTR(max_comp_streams, S_IRUGO | S_IWUSR,
776 max_comp_streams_show, max_comp_streams_store);
777 static DEVICE_ATTR(comp_algorithm, S_IRUGO | S_IWUSR,
778 comp_algorithm_show, comp_algorithm_store);
780 ZRAM_ATTR_RO(num_reads);
781 ZRAM_ATTR_RO(num_writes);
782 ZRAM_ATTR_RO(failed_reads);
783 ZRAM_ATTR_RO(failed_writes);
784 ZRAM_ATTR_RO(invalid_io);
785 ZRAM_ATTR_RO(notify_free);
786 ZRAM_ATTR_RO(zero_pages);
787 ZRAM_ATTR_RO(compr_data_size);
789 static struct attribute *zram_disk_attrs[] = {
790 &dev_attr_disksize.attr,
791 &dev_attr_initstate.attr,
792 &dev_attr_reset.attr,
793 &dev_attr_num_reads.attr,
794 &dev_attr_num_writes.attr,
795 &dev_attr_failed_reads.attr,
796 &dev_attr_failed_writes.attr,
797 &dev_attr_invalid_io.attr,
798 &dev_attr_notify_free.attr,
799 &dev_attr_zero_pages.attr,
800 &dev_attr_orig_data_size.attr,
801 &dev_attr_compr_data_size.attr,
802 &dev_attr_mem_used_total.attr,
803 &dev_attr_max_comp_streams.attr,
804 &dev_attr_comp_algorithm.attr,
808 static struct attribute_group zram_disk_attr_group = {
809 .attrs = zram_disk_attrs,
812 static int create_device(struct zram *zram, int device_id)
816 init_rwsem(&zram->init_lock);
818 zram->queue = blk_alloc_queue(GFP_KERNEL);
820 pr_err("Error allocating disk queue for device %d\n",
825 blk_queue_make_request(zram->queue, zram_make_request);
826 zram->queue->queuedata = zram;
828 /* gendisk structure */
829 zram->disk = alloc_disk(1);
831 pr_warn("Error allocating disk structure for device %d\n",
836 zram->disk->major = zram_major;
837 zram->disk->first_minor = device_id;
838 zram->disk->fops = &zram_devops;
839 zram->disk->queue = zram->queue;
840 zram->disk->private_data = zram;
841 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
843 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
844 set_capacity(zram->disk, 0);
845 /* zram devices sort of resembles non-rotational disks */
846 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
848 * To ensure that we always get PAGE_SIZE aligned
849 * and n*PAGE_SIZED sized I/O requests.
851 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
852 blk_queue_logical_block_size(zram->disk->queue,
853 ZRAM_LOGICAL_BLOCK_SIZE);
854 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
855 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
857 add_disk(zram->disk);
859 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
860 &zram_disk_attr_group);
862 pr_warn("Error creating sysfs group");
865 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
867 zram->max_comp_streams = 1;
871 del_gendisk(zram->disk);
872 put_disk(zram->disk);
874 blk_cleanup_queue(zram->queue);
879 static void destroy_device(struct zram *zram)
881 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
882 &zram_disk_attr_group);
884 del_gendisk(zram->disk);
885 put_disk(zram->disk);
887 blk_cleanup_queue(zram->queue);
890 static int __init zram_init(void)
894 if (num_devices > max_num_devices) {
895 pr_warn("Invalid value for num_devices: %u\n",
901 zram_major = register_blkdev(0, "zram");
902 if (zram_major <= 0) {
903 pr_warn("Unable to get major number\n");
908 /* Allocate the device array and initialize each one */
909 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
915 for (dev_id = 0; dev_id < num_devices; dev_id++) {
916 ret = create_device(&zram_devices[dev_id], dev_id);
921 pr_info("Created %u device(s) ...\n", num_devices);
927 destroy_device(&zram_devices[--dev_id]);
930 unregister_blkdev(zram_major, "zram");
935 static void __exit zram_exit(void)
940 for (i = 0; i < num_devices; i++) {
941 zram = &zram_devices[i];
943 destroy_device(zram);
945 * Shouldn't access zram->disk after destroy_device
946 * because destroy_device already released zram->disk.
948 zram_reset_device(zram, false);
951 unregister_blkdev(zram_major, "zram");
954 pr_debug("Cleanup done!\n");
957 module_init(zram_init);
958 module_exit(zram_exit);
960 module_param(num_devices, uint, 0);
961 MODULE_PARM_DESC(num_devices, "Number of zram devices");
963 MODULE_LICENSE("Dual BSD/GPL");
964 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
965 MODULE_DESCRIPTION("Compressed RAM Block Device");