2 * Compressed RAM block device
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
6 * This code is released using a dual license strategy: BSD/GPL
7 * You can choose the licence that better fits your requirements.
9 * Released under the terms of 3-clause BSD License
10 * Released under the terms of GNU General Public License Version 2.0
12 * Project home: http://compcache.googlecode.com
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/lzo.h>
33 #include <linux/string.h>
34 #include <linux/vmalloc.h>
39 static int zram_major;
40 static struct zram *zram_devices;
42 /* Module params (documentation at end) */
43 static unsigned int num_devices = 1;
45 static inline struct zram *dev_to_zram(struct device *dev)
47 return (struct zram *)dev_to_disk(dev)->private_data;
50 static ssize_t disksize_show(struct device *dev,
51 struct device_attribute *attr, char *buf)
53 struct zram *zram = dev_to_zram(dev);
55 return sprintf(buf, "%llu\n", zram->disksize);
58 static ssize_t initstate_show(struct device *dev,
59 struct device_attribute *attr, char *buf)
61 struct zram *zram = dev_to_zram(dev);
63 return sprintf(buf, "%u\n", zram->init_done);
66 static ssize_t num_reads_show(struct device *dev,
67 struct device_attribute *attr, char *buf)
69 struct zram *zram = dev_to_zram(dev);
71 return sprintf(buf, "%llu\n",
72 (u64)atomic64_read(&zram->stats.num_reads));
75 static ssize_t num_writes_show(struct device *dev,
76 struct device_attribute *attr, char *buf)
78 struct zram *zram = dev_to_zram(dev);
80 return sprintf(buf, "%llu\n",
81 (u64)atomic64_read(&zram->stats.num_writes));
84 static ssize_t invalid_io_show(struct device *dev,
85 struct device_attribute *attr, char *buf)
87 struct zram *zram = dev_to_zram(dev);
89 return sprintf(buf, "%llu\n",
90 (u64)atomic64_read(&zram->stats.invalid_io));
93 static ssize_t notify_free_show(struct device *dev,
94 struct device_attribute *attr, char *buf)
96 struct zram *zram = dev_to_zram(dev);
98 return sprintf(buf, "%llu\n",
99 (u64)atomic64_read(&zram->stats.notify_free));
102 static ssize_t zero_pages_show(struct device *dev,
103 struct device_attribute *attr, char *buf)
105 struct zram *zram = dev_to_zram(dev);
107 return sprintf(buf, "%u\n", zram->stats.pages_zero);
110 static ssize_t orig_data_size_show(struct device *dev,
111 struct device_attribute *attr, char *buf)
113 struct zram *zram = dev_to_zram(dev);
115 return sprintf(buf, "%llu\n",
116 (u64)(zram->stats.pages_stored) << PAGE_SHIFT);
119 static ssize_t compr_data_size_show(struct device *dev,
120 struct device_attribute *attr, char *buf)
122 struct zram *zram = dev_to_zram(dev);
124 return sprintf(buf, "%llu\n",
125 (u64)atomic64_read(&zram->stats.compr_size));
128 static ssize_t mem_used_total_show(struct device *dev,
129 struct device_attribute *attr, char *buf)
132 struct zram *zram = dev_to_zram(dev);
133 struct zram_meta *meta = zram->meta;
135 down_read(&zram->init_lock);
137 val = zs_get_total_size_bytes(meta->mem_pool);
138 up_read(&zram->init_lock);
140 return sprintf(buf, "%llu\n", val);
143 static int zram_test_flag(struct zram_meta *meta, u32 index,
144 enum zram_pageflags flag)
146 return meta->table[index].flags & BIT(flag);
149 static void zram_set_flag(struct zram_meta *meta, u32 index,
150 enum zram_pageflags flag)
152 meta->table[index].flags |= BIT(flag);
155 static void zram_clear_flag(struct zram_meta *meta, u32 index,
156 enum zram_pageflags flag)
158 meta->table[index].flags &= ~BIT(flag);
161 static inline int is_partial_io(struct bio_vec *bvec)
163 return bvec->bv_len != PAGE_SIZE;
167 * Check if request is within bounds and aligned on zram logical blocks.
169 static inline int valid_io_request(struct zram *zram, struct bio *bio)
171 u64 start, end, bound;
173 /* unaligned request */
174 if (unlikely(bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
176 if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
179 start = bio->bi_sector;
180 end = start + (bio->bi_size >> SECTOR_SHIFT);
181 bound = zram->disksize >> SECTOR_SHIFT;
182 /* out of range range */
183 if (unlikely(start >= bound || end >= bound || start > end))
186 /* I/O request is valid */
190 static void zram_meta_free(struct zram_meta *meta)
192 zs_destroy_pool(meta->mem_pool);
193 kfree(meta->compress_workmem);
194 free_pages((unsigned long)meta->compress_buffer, 1);
199 static struct zram_meta *zram_meta_alloc(u64 disksize)
202 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
206 meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
207 if (!meta->compress_workmem)
210 meta->compress_buffer =
211 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
212 if (!meta->compress_buffer) {
213 pr_err("Error allocating compressor buffer space\n");
217 num_pages = disksize >> PAGE_SHIFT;
218 meta->table = vzalloc(num_pages * sizeof(*meta->table));
220 pr_err("Error allocating zram address table\n");
224 meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
225 if (!meta->mem_pool) {
226 pr_err("Error creating memory pool\n");
235 free_pages((unsigned long)meta->compress_buffer, 1);
237 kfree(meta->compress_workmem);
245 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
247 if (*offset + bvec->bv_len >= PAGE_SIZE)
249 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
252 static int page_zero_filled(void *ptr)
257 page = (unsigned long *)ptr;
259 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
267 static void handle_zero_page(struct bio_vec *bvec)
269 struct page *page = bvec->bv_page;
272 user_mem = kmap_atomic(page);
273 if (is_partial_io(bvec))
274 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
276 clear_page(user_mem);
277 kunmap_atomic(user_mem);
279 flush_dcache_page(page);
282 static void zram_free_page(struct zram *zram, size_t index)
284 struct zram_meta *meta = zram->meta;
285 unsigned long handle = meta->table[index].handle;
286 u16 size = meta->table[index].size;
288 if (unlikely(!handle)) {
290 * No memory is allocated for zero filled pages.
291 * Simply clear zero page flag.
293 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
294 zram_clear_flag(meta, index, ZRAM_ZERO);
295 zram->stats.pages_zero--;
300 if (unlikely(size > max_zpage_size))
301 zram->stats.bad_compress--;
303 zs_free(meta->mem_pool, handle);
305 if (size <= PAGE_SIZE / 2)
306 zram->stats.good_compress--;
308 atomic64_sub(meta->table[index].size, &zram->stats.compr_size);
309 zram->stats.pages_stored--;
311 meta->table[index].handle = 0;
312 meta->table[index].size = 0;
315 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
318 size_t clen = PAGE_SIZE;
320 struct zram_meta *meta = zram->meta;
321 unsigned long handle = meta->table[index].handle;
323 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
328 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
329 if (meta->table[index].size == PAGE_SIZE)
330 copy_page(mem, cmem);
332 ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
334 zs_unmap_object(meta->mem_pool, handle);
336 /* Should NEVER happen. Return bio error if it does. */
337 if (unlikely(ret != LZO_E_OK)) {
338 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
339 atomic64_inc(&zram->stats.failed_reads);
346 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
347 u32 index, int offset, struct bio *bio)
351 unsigned char *user_mem, *uncmem = NULL;
352 struct zram_meta *meta = zram->meta;
353 page = bvec->bv_page;
355 if (unlikely(!meta->table[index].handle) ||
356 zram_test_flag(meta, index, ZRAM_ZERO)) {
357 handle_zero_page(bvec);
361 if (is_partial_io(bvec))
362 /* Use a temporary buffer to decompress the page */
363 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
365 user_mem = kmap_atomic(page);
366 if (!is_partial_io(bvec))
370 pr_info("Unable to allocate temp memory\n");
375 ret = zram_decompress_page(zram, uncmem, index);
376 /* Should NEVER happen. Return bio error if it does. */
377 if (unlikely(ret != LZO_E_OK))
380 if (is_partial_io(bvec))
381 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
384 flush_dcache_page(page);
387 kunmap_atomic(user_mem);
388 if (is_partial_io(bvec))
393 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
398 unsigned long handle;
400 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
401 struct zram_meta *meta = zram->meta;
403 page = bvec->bv_page;
404 src = meta->compress_buffer;
406 if (is_partial_io(bvec)) {
408 * This is a partial IO. We need to read the full page
409 * before to write the changes.
411 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
416 ret = zram_decompress_page(zram, uncmem, index);
421 user_mem = kmap_atomic(page);
423 if (is_partial_io(bvec)) {
424 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
426 kunmap_atomic(user_mem);
432 if (page_zero_filled(uncmem)) {
433 kunmap_atomic(user_mem);
434 /* Free memory associated with this sector now. */
435 zram_free_page(zram, index);
437 zram->stats.pages_zero++;
438 zram_set_flag(meta, index, ZRAM_ZERO);
443 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
444 meta->compress_workmem);
446 if (!is_partial_io(bvec)) {
447 kunmap_atomic(user_mem);
452 if (unlikely(ret != LZO_E_OK)) {
453 pr_err("Compression failed! err=%d\n", ret);
457 if (unlikely(clen > max_zpage_size)) {
458 zram->stats.bad_compress++;
461 if (is_partial_io(bvec))
465 handle = zs_malloc(meta->mem_pool, clen);
467 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
472 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
474 if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
475 src = kmap_atomic(page);
476 copy_page(cmem, src);
479 memcpy(cmem, src, clen);
482 zs_unmap_object(meta->mem_pool, handle);
485 * Free memory associated with this sector
486 * before overwriting unused sectors.
488 zram_free_page(zram, index);
490 meta->table[index].handle = handle;
491 meta->table[index].size = clen;
494 atomic64_add(clen, &zram->stats.compr_size);
495 zram->stats.pages_stored++;
496 if (clen <= PAGE_SIZE / 2)
497 zram->stats.good_compress++;
500 if (is_partial_io(bvec))
504 atomic64_inc(&zram->stats.failed_writes);
508 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
509 int offset, struct bio *bio, int rw)
514 down_read(&zram->lock);
515 ret = zram_bvec_read(zram, bvec, index, offset, bio);
516 up_read(&zram->lock);
518 down_write(&zram->lock);
519 ret = zram_bvec_write(zram, bvec, index, offset);
520 up_write(&zram->lock);
526 static void zram_reset_device(struct zram *zram, bool reset_capacity)
529 struct zram_meta *meta;
531 down_write(&zram->init_lock);
532 if (!zram->init_done) {
533 up_write(&zram->init_lock);
540 /* Free all pages that are still in this zram device */
541 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
542 unsigned long handle = meta->table[index].handle;
546 zs_free(meta->mem_pool, handle);
549 zram_meta_free(zram->meta);
552 memset(&zram->stats, 0, sizeof(zram->stats));
556 set_capacity(zram->disk, 0);
557 up_write(&zram->init_lock);
560 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
562 if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
564 "There is little point creating a zram of greater than "
565 "twice the size of memory since we expect a 2:1 compression "
566 "ratio. Note that zram uses about 0.1%% of the size of "
567 "the disk when not in use so a huge zram is "
569 "\tMemory Size: %lu kB\n"
570 "\tSize you selected: %llu kB\n"
571 "Continuing anyway ...\n",
572 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
576 /* zram devices sort of resembles non-rotational disks */
577 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
582 pr_debug("Initialization done!\n");
585 static ssize_t disksize_store(struct device *dev,
586 struct device_attribute *attr, const char *buf, size_t len)
589 struct zram_meta *meta;
590 struct zram *zram = dev_to_zram(dev);
592 disksize = memparse(buf, NULL);
596 disksize = PAGE_ALIGN(disksize);
597 meta = zram_meta_alloc(disksize);
598 down_write(&zram->init_lock);
599 if (zram->init_done) {
600 up_write(&zram->init_lock);
601 zram_meta_free(meta);
602 pr_info("Cannot change disksize for initialized device\n");
606 zram->disksize = disksize;
607 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
608 zram_init_device(zram, meta);
609 up_write(&zram->init_lock);
614 static ssize_t reset_store(struct device *dev,
615 struct device_attribute *attr, const char *buf, size_t len)
618 unsigned short do_reset;
620 struct block_device *bdev;
622 zram = dev_to_zram(dev);
623 bdev = bdget_disk(zram->disk, 0);
625 /* Do not reset an active device! */
626 if (bdev->bd_holders)
629 ret = kstrtou16(buf, 10, &do_reset);
636 /* Make sure all pending I/O is finished */
640 zram_reset_device(zram, true);
644 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
648 struct bio_vec *bvec;
652 atomic64_inc(&zram->stats.num_reads);
655 atomic64_inc(&zram->stats.num_writes);
659 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
660 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
662 bio_for_each_segment(bvec, bio, i) {
663 int max_transfer_size = PAGE_SIZE - offset;
665 if (bvec->bv_len > max_transfer_size) {
667 * zram_bvec_rw() can only make operation on a single
668 * zram page. Split the bio vector.
672 bv.bv_page = bvec->bv_page;
673 bv.bv_len = max_transfer_size;
674 bv.bv_offset = bvec->bv_offset;
676 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
679 bv.bv_len = bvec->bv_len - max_transfer_size;
680 bv.bv_offset += max_transfer_size;
681 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
684 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
688 update_position(&index, &offset, bvec);
691 set_bit(BIO_UPTODATE, &bio->bi_flags);
700 * Handler function for all zram I/O requests.
702 static void zram_make_request(struct request_queue *queue, struct bio *bio)
704 struct zram *zram = queue->queuedata;
706 down_read(&zram->init_lock);
707 if (unlikely(!zram->init_done))
710 if (!valid_io_request(zram, bio)) {
711 atomic64_inc(&zram->stats.invalid_io);
715 __zram_make_request(zram, bio, bio_data_dir(bio));
716 up_read(&zram->init_lock);
721 up_read(&zram->init_lock);
725 static void zram_slot_free_notify(struct block_device *bdev,
730 zram = bdev->bd_disk->private_data;
731 down_write(&zram->lock);
732 zram_free_page(zram, index);
733 up_write(&zram->lock);
734 atomic64_inc(&zram->stats.notify_free);
737 static const struct block_device_operations zram_devops = {
738 .swap_slot_free_notify = zram_slot_free_notify,
742 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
743 disksize_show, disksize_store);
744 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
745 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
746 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
747 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
748 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
749 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
750 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
751 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
752 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
753 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
755 static struct attribute *zram_disk_attrs[] = {
756 &dev_attr_disksize.attr,
757 &dev_attr_initstate.attr,
758 &dev_attr_reset.attr,
759 &dev_attr_num_reads.attr,
760 &dev_attr_num_writes.attr,
761 &dev_attr_invalid_io.attr,
762 &dev_attr_notify_free.attr,
763 &dev_attr_zero_pages.attr,
764 &dev_attr_orig_data_size.attr,
765 &dev_attr_compr_data_size.attr,
766 &dev_attr_mem_used_total.attr,
770 static struct attribute_group zram_disk_attr_group = {
771 .attrs = zram_disk_attrs,
774 static int create_device(struct zram *zram, int device_id)
778 init_rwsem(&zram->lock);
779 init_rwsem(&zram->init_lock);
781 zram->queue = blk_alloc_queue(GFP_KERNEL);
783 pr_err("Error allocating disk queue for device %d\n",
788 blk_queue_make_request(zram->queue, zram_make_request);
789 zram->queue->queuedata = zram;
791 /* gendisk structure */
792 zram->disk = alloc_disk(1);
794 pr_warn("Error allocating disk structure for device %d\n",
799 zram->disk->major = zram_major;
800 zram->disk->first_minor = device_id;
801 zram->disk->fops = &zram_devops;
802 zram->disk->queue = zram->queue;
803 zram->disk->private_data = zram;
804 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
806 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
807 set_capacity(zram->disk, 0);
810 * To ensure that we always get PAGE_SIZE aligned
811 * and n*PAGE_SIZED sized I/O requests.
813 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
814 blk_queue_logical_block_size(zram->disk->queue,
815 ZRAM_LOGICAL_BLOCK_SIZE);
816 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
817 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
819 add_disk(zram->disk);
821 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
822 &zram_disk_attr_group);
824 pr_warn("Error creating sysfs group");
832 del_gendisk(zram->disk);
833 put_disk(zram->disk);
835 blk_cleanup_queue(zram->queue);
840 static void destroy_device(struct zram *zram)
842 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
843 &zram_disk_attr_group);
846 del_gendisk(zram->disk);
847 put_disk(zram->disk);
851 blk_cleanup_queue(zram->queue);
854 static int __init zram_init(void)
858 if (num_devices > max_num_devices) {
859 pr_warn("Invalid value for num_devices: %u\n",
865 zram_major = register_blkdev(0, "zram");
866 if (zram_major <= 0) {
867 pr_warn("Unable to get major number\n");
872 /* Allocate the device array and initialize each one */
873 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
879 for (dev_id = 0; dev_id < num_devices; dev_id++) {
880 ret = create_device(&zram_devices[dev_id], dev_id);
885 pr_info("Created %u device(s) ...\n", num_devices);
891 destroy_device(&zram_devices[--dev_id]);
894 unregister_blkdev(zram_major, "zram");
899 static void __exit zram_exit(void)
904 for (i = 0; i < num_devices; i++) {
905 zram = &zram_devices[i];
907 destroy_device(zram);
909 * Shouldn't access zram->disk after destroy_device
910 * because destroy_device already released zram->disk.
912 zram_reset_device(zram, false);
915 unregister_blkdev(zram_major, "zram");
918 pr_debug("Cleanup done!\n");
921 module_init(zram_init);
922 module_exit(zram_exit);
924 module_param(num_devices, uint, 0);
925 MODULE_PARM_DESC(num_devices, "Number of zram devices");
927 MODULE_LICENSE("Dual BSD/GPL");
928 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
929 MODULE_DESCRIPTION("Compressed RAM Block Device");
930 MODULE_ALIAS("devname:zram");