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 struct zram *zram_devices;
42 /* Module params (documentation at end) */
43 unsigned int zram_num_devices;
45 static void zram_stat_inc(u32 *v)
50 static void zram_stat_dec(u32 *v)
55 static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
57 spin_lock(&zram->stat64_lock);
59 spin_unlock(&zram->stat64_lock);
62 static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
64 spin_lock(&zram->stat64_lock);
66 spin_unlock(&zram->stat64_lock);
69 static void zram_stat64_inc(struct zram *zram, u64 *v)
71 zram_stat64_add(zram, v, 1);
74 static int zram_test_flag(struct zram *zram, u32 index,
75 enum zram_pageflags flag)
77 return zram->table[index].flags & BIT(flag);
80 static void zram_set_flag(struct zram *zram, u32 index,
81 enum zram_pageflags flag)
83 zram->table[index].flags |= BIT(flag);
86 static void zram_clear_flag(struct zram *zram, u32 index,
87 enum zram_pageflags flag)
89 zram->table[index].flags &= ~BIT(flag);
92 static int page_zero_filled(void *ptr)
97 page = (unsigned long *)ptr;
99 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
107 static void zram_set_disksize(struct zram *zram, size_t totalram_bytes)
109 if (!zram->disksize) {
111 "disk size not provided. You can use disksize_kb module "
112 "param to specify size.\nUsing default: (%u%% of RAM).\n",
113 default_disksize_perc_ram
115 zram->disksize = default_disksize_perc_ram *
116 (totalram_bytes / 100);
119 if (zram->disksize > 2 * (totalram_bytes)) {
121 "There is little point creating a zram of greater than "
122 "twice the size of memory since we expect a 2:1 compression "
123 "ratio. Note that zram uses about 0.1%% of the size of "
124 "the disk when not in use so a huge zram is "
126 "\tMemory Size: %zu kB\n"
127 "\tSize you selected: %llu kB\n"
128 "Continuing anyway ...\n",
129 totalram_bytes >> 10, zram->disksize
133 zram->disksize &= PAGE_MASK;
136 static void zram_free_page(struct zram *zram, size_t index)
141 struct page *page = zram->table[index].page;
142 u32 offset = zram->table[index].offset;
144 if (unlikely(!page)) {
146 * No memory is allocated for zero filled pages.
147 * Simply clear zero page flag.
149 if (zram_test_flag(zram, index, ZRAM_ZERO)) {
150 zram_clear_flag(zram, index, ZRAM_ZERO);
151 zram_stat_dec(&zram->stats.pages_zero);
156 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
159 zram_clear_flag(zram, index, ZRAM_UNCOMPRESSED);
160 zram_stat_dec(&zram->stats.pages_expand);
164 obj = kmap_atomic(page, KM_USER0) + offset;
165 clen = xv_get_object_size(obj) - sizeof(struct zobj_header);
166 kunmap_atomic(obj, KM_USER0);
168 xv_free(zram->mem_pool, page, offset);
169 if (clen <= PAGE_SIZE / 2)
170 zram_stat_dec(&zram->stats.good_compress);
173 zram_stat64_sub(zram, &zram->stats.compr_size, clen);
174 zram_stat_dec(&zram->stats.pages_stored);
176 zram->table[index].page = NULL;
177 zram->table[index].offset = 0;
180 static void handle_zero_page(struct bio_vec *bvec)
182 struct page *page = bvec->bv_page;
185 user_mem = kmap_atomic(page, KM_USER0);
186 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
187 kunmap_atomic(user_mem, KM_USER0);
189 flush_dcache_page(page);
192 static void handle_uncompressed_page(struct zram *zram, struct bio_vec *bvec,
193 u32 index, int offset)
195 struct page *page = bvec->bv_page;
196 unsigned char *user_mem, *cmem;
198 user_mem = kmap_atomic(page, KM_USER0);
199 cmem = kmap_atomic(zram->table[index].page, KM_USER1);
201 memcpy(user_mem + bvec->bv_offset, cmem + offset, bvec->bv_len);
202 kunmap_atomic(cmem, KM_USER1);
203 kunmap_atomic(user_mem, KM_USER0);
205 flush_dcache_page(page);
208 static inline int is_partial_io(struct bio_vec *bvec)
210 return bvec->bv_len != PAGE_SIZE;
213 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
214 u32 index, int offset, struct bio *bio)
219 struct zobj_header *zheader;
220 unsigned char *user_mem, *cmem, *uncmem = NULL;
222 page = bvec->bv_page;
224 if (zram_test_flag(zram, index, ZRAM_ZERO)) {
225 handle_zero_page(bvec);
229 /* Requested page is not present in compressed area */
230 if (unlikely(!zram->table[index].page)) {
231 pr_debug("Read before write: sector=%lu, size=%u",
232 (ulong)(bio->bi_sector), bio->bi_size);
233 handle_zero_page(bvec);
237 /* Page is stored uncompressed since it's incompressible */
238 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
239 handle_uncompressed_page(zram, bvec, index, offset);
243 if (is_partial_io(bvec)) {
244 /* Use a temporary buffer to decompress the page */
245 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
247 pr_info("Error allocating temp memory!\n");
252 user_mem = kmap_atomic(page, KM_USER0);
253 if (!is_partial_io(bvec))
257 cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
258 zram->table[index].offset;
260 ret = lzo1x_decompress_safe(cmem + sizeof(*zheader),
261 xv_get_object_size(cmem) - sizeof(*zheader),
264 if (is_partial_io(bvec)) {
265 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
270 kunmap_atomic(cmem, KM_USER1);
271 kunmap_atomic(user_mem, KM_USER0);
273 /* Should NEVER happen. Return bio error if it does. */
274 if (unlikely(ret != LZO_E_OK)) {
275 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
276 zram_stat64_inc(zram, &zram->stats.failed_reads);
280 flush_dcache_page(page);
285 static int zram_read_before_write(struct zram *zram, char *mem, u32 index)
288 size_t clen = PAGE_SIZE;
289 struct zobj_header *zheader;
292 if (zram_test_flag(zram, index, ZRAM_ZERO) ||
293 !zram->table[index].page) {
294 memset(mem, 0, PAGE_SIZE);
298 cmem = kmap_atomic(zram->table[index].page, KM_USER0) +
299 zram->table[index].offset;
301 /* Page is stored uncompressed since it's incompressible */
302 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
303 memcpy(mem, cmem, PAGE_SIZE);
304 kunmap_atomic(cmem, KM_USER0);
308 ret = lzo1x_decompress_safe(cmem + sizeof(*zheader),
309 xv_get_object_size(cmem) - sizeof(*zheader),
311 kunmap_atomic(cmem, KM_USER0);
313 /* Should NEVER happen. Return bio error if it does. */
314 if (unlikely(ret != LZO_E_OK)) {
315 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
316 zram_stat64_inc(zram, &zram->stats.failed_reads);
323 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
329 struct zobj_header *zheader;
330 struct page *page, *page_store;
331 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
333 page = bvec->bv_page;
334 src = zram->compress_buffer;
336 if (is_partial_io(bvec)) {
338 * This is a partial IO. We need to read the full page
339 * before to write the changes.
341 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
343 pr_info("Error allocating temp memory!\n");
347 ret = zram_read_before_write(zram, uncmem, index);
355 * System overwrites unused sectors. Free memory associated
356 * with this sector now.
358 if (zram->table[index].page ||
359 zram_test_flag(zram, index, ZRAM_ZERO))
360 zram_free_page(zram, index);
362 user_mem = kmap_atomic(page, KM_USER0);
364 if (is_partial_io(bvec))
365 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
370 if (page_zero_filled(uncmem)) {
371 kunmap_atomic(user_mem, KM_USER0);
372 if (is_partial_io(bvec))
374 zram_stat_inc(&zram->stats.pages_zero);
375 zram_set_flag(zram, index, ZRAM_ZERO);
380 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
381 zram->compress_workmem);
383 kunmap_atomic(user_mem, KM_USER0);
384 if (is_partial_io(bvec))
387 if (unlikely(ret != LZO_E_OK)) {
388 pr_err("Compression failed! err=%d\n", ret);
393 * Page is incompressible. Store it as-is (uncompressed)
394 * since we do not want to return too many disk write
395 * errors which has side effect of hanging the system.
397 if (unlikely(clen > max_zpage_size)) {
399 page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
400 if (unlikely(!page_store)) {
401 pr_info("Error allocating memory for "
402 "incompressible page: %u\n", index);
408 zram_set_flag(zram, index, ZRAM_UNCOMPRESSED);
409 zram_stat_inc(&zram->stats.pages_expand);
410 zram->table[index].page = page_store;
411 src = kmap_atomic(page, KM_USER0);
415 if (xv_malloc(zram->mem_pool, clen + sizeof(*zheader),
416 &zram->table[index].page, &store_offset,
417 GFP_NOIO | __GFP_HIGHMEM)) {
418 pr_info("Error allocating memory for compressed "
419 "page: %u, size=%zu\n", index, clen);
425 zram->table[index].offset = store_offset;
427 cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
428 zram->table[index].offset;
431 /* Back-reference needed for memory defragmentation */
432 if (!zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)) {
433 zheader = (struct zobj_header *)cmem;
434 zheader->table_idx = index;
435 cmem += sizeof(*zheader);
439 memcpy(cmem, src, clen);
441 kunmap_atomic(cmem, KM_USER1);
442 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
443 kunmap_atomic(src, KM_USER0);
446 zram_stat64_add(zram, &zram->stats.compr_size, clen);
447 zram_stat_inc(&zram->stats.pages_stored);
448 if (clen <= PAGE_SIZE / 2)
449 zram_stat_inc(&zram->stats.good_compress);
455 zram_stat64_inc(zram, &zram->stats.failed_writes);
459 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
460 int offset, struct bio *bio, int rw)
465 down_read(&zram->lock);
466 ret = zram_bvec_read(zram, bvec, index, offset, bio);
467 up_read(&zram->lock);
469 down_write(&zram->lock);
470 ret = zram_bvec_write(zram, bvec, index, offset);
471 up_write(&zram->lock);
477 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
479 if (*offset + bvec->bv_len >= PAGE_SIZE)
481 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
484 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
488 struct bio_vec *bvec;
492 zram_stat64_inc(zram, &zram->stats.num_reads);
495 zram_stat64_inc(zram, &zram->stats.num_writes);
499 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
500 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
502 bio_for_each_segment(bvec, bio, i) {
503 int max_transfer_size = PAGE_SIZE - offset;
505 if (bvec->bv_len > max_transfer_size) {
507 * zram_bvec_rw() can only make operation on a single
508 * zram page. Split the bio vector.
512 bv.bv_page = bvec->bv_page;
513 bv.bv_len = max_transfer_size;
514 bv.bv_offset = bvec->bv_offset;
516 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
519 bv.bv_len = bvec->bv_len - max_transfer_size;
520 bv.bv_offset += max_transfer_size;
521 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
524 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
528 update_position(&index, &offset, bvec);
531 set_bit(BIO_UPTODATE, &bio->bi_flags);
540 * Check if request is within bounds and aligned on zram logical blocks.
542 static inline int valid_io_request(struct zram *zram, struct bio *bio)
545 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
546 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
547 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
552 /* I/O request is valid */
557 * Handler function for all zram I/O requests.
559 static int zram_make_request(struct request_queue *queue, struct bio *bio)
561 struct zram *zram = queue->queuedata;
563 if (unlikely(!zram->init_done) && zram_init_device(zram))
566 down_read(&zram->init_lock);
567 if (unlikely(!zram->init_done))
570 if (!valid_io_request(zram, bio)) {
571 zram_stat64_inc(zram, &zram->stats.invalid_io);
575 __zram_make_request(zram, bio, bio_data_dir(bio));
576 up_read(&zram->init_lock);
581 up_read(&zram->init_lock);
587 void __zram_reset_device(struct zram *zram)
593 /* Free various per-device buffers */
594 kfree(zram->compress_workmem);
595 free_pages((unsigned long)zram->compress_buffer, 1);
597 zram->compress_workmem = NULL;
598 zram->compress_buffer = NULL;
600 /* Free all pages that are still in this zram device */
601 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
605 page = zram->table[index].page;
606 offset = zram->table[index].offset;
611 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
614 xv_free(zram->mem_pool, page, offset);
620 xv_destroy_pool(zram->mem_pool);
621 zram->mem_pool = NULL;
624 memset(&zram->stats, 0, sizeof(zram->stats));
629 void zram_reset_device(struct zram *zram)
631 down_write(&zram->init_lock);
632 __zram_reset_device(zram);
633 up_write(&zram->init_lock);
636 int zram_init_device(struct zram *zram)
641 down_write(&zram->init_lock);
643 if (zram->init_done) {
644 up_write(&zram->init_lock);
648 zram_set_disksize(zram, totalram_pages << PAGE_SHIFT);
650 zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
651 if (!zram->compress_workmem) {
652 pr_err("Error allocating compressor working memory!\n");
657 zram->compress_buffer = (void *)__get_free_pages(__GFP_ZERO, 1);
658 if (!zram->compress_buffer) {
659 pr_err("Error allocating compressor buffer space\n");
664 num_pages = zram->disksize >> PAGE_SHIFT;
665 zram->table = vzalloc(num_pages * sizeof(*zram->table));
667 pr_err("Error allocating zram address table\n");
672 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
674 /* zram devices sort of resembles non-rotational disks */
675 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
677 zram->mem_pool = xv_create_pool();
678 if (!zram->mem_pool) {
679 pr_err("Error creating memory pool\n");
685 up_write(&zram->init_lock);
687 pr_debug("Initialization done!\n");
691 /* To prevent accessing table entries during cleanup */
694 __zram_reset_device(zram);
695 up_write(&zram->init_lock);
696 pr_err("Initialization failed: err=%d\n", ret);
700 static void zram_slot_free_notify(struct block_device *bdev,
705 zram = bdev->bd_disk->private_data;
706 zram_free_page(zram, index);
707 zram_stat64_inc(zram, &zram->stats.notify_free);
710 static const struct block_device_operations zram_devops = {
711 .swap_slot_free_notify = zram_slot_free_notify,
715 static int create_device(struct zram *zram, int device_id)
719 init_rwsem(&zram->lock);
720 init_rwsem(&zram->init_lock);
721 spin_lock_init(&zram->stat64_lock);
723 zram->queue = blk_alloc_queue(GFP_KERNEL);
725 pr_err("Error allocating disk queue for device %d\n",
731 blk_queue_make_request(zram->queue, zram_make_request);
732 zram->queue->queuedata = zram;
734 /* gendisk structure */
735 zram->disk = alloc_disk(1);
737 blk_cleanup_queue(zram->queue);
738 pr_warning("Error allocating disk structure for device %d\n",
744 zram->disk->major = zram_major;
745 zram->disk->first_minor = device_id;
746 zram->disk->fops = &zram_devops;
747 zram->disk->queue = zram->queue;
748 zram->disk->private_data = zram;
749 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
751 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
752 set_capacity(zram->disk, 0);
755 * To ensure that we always get PAGE_SIZE aligned
756 * and n*PAGE_SIZED sized I/O requests.
758 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
759 blk_queue_logical_block_size(zram->disk->queue,
760 ZRAM_LOGICAL_BLOCK_SIZE);
761 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
762 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
764 add_disk(zram->disk);
766 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
767 &zram_disk_attr_group);
769 pr_warning("Error creating sysfs group");
779 static void destroy_device(struct zram *zram)
781 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
782 &zram_disk_attr_group);
785 del_gendisk(zram->disk);
786 put_disk(zram->disk);
790 blk_cleanup_queue(zram->queue);
793 static int __init zram_init(void)
797 if (zram_num_devices > max_num_devices) {
798 pr_warning("Invalid value for num_devices: %u\n",
804 zram_major = register_blkdev(0, "zram");
805 if (zram_major <= 0) {
806 pr_warning("Unable to get major number\n");
811 if (!zram_num_devices) {
812 pr_info("num_devices not specified. Using default: 1\n");
813 zram_num_devices = 1;
816 /* Allocate the device array and initialize each one */
817 pr_info("Creating %u devices ...\n", zram_num_devices);
818 zram_devices = kzalloc(zram_num_devices * sizeof(struct zram), GFP_KERNEL);
824 for (dev_id = 0; dev_id < zram_num_devices; dev_id++) {
825 ret = create_device(&zram_devices[dev_id], dev_id);
834 destroy_device(&zram_devices[--dev_id]);
837 unregister_blkdev(zram_major, "zram");
842 static void __exit zram_exit(void)
847 for (i = 0; i < zram_num_devices; i++) {
848 zram = &zram_devices[i];
850 destroy_device(zram);
852 zram_reset_device(zram);
855 unregister_blkdev(zram_major, "zram");
858 pr_debug("Cleanup done!\n");
861 module_param(zram_num_devices, uint, 0);
862 MODULE_PARM_DESC(zram_num_devices, "Number of zram devices");
864 module_init(zram_init);
865 module_exit(zram_exit);
867 MODULE_LICENSE("Dual BSD/GPL");
868 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
869 MODULE_DESCRIPTION("Compressed RAM Block Device");