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 static unsigned int num_devices = 1;
45 static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
47 spin_lock(&zram->stat64_lock);
49 spin_unlock(&zram->stat64_lock);
52 static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
54 spin_lock(&zram->stat64_lock);
56 spin_unlock(&zram->stat64_lock);
59 static void zram_stat64_inc(struct zram *zram, u64 *v)
61 zram_stat64_add(zram, v, 1);
64 static int zram_test_flag(struct zram_meta *meta, u32 index,
65 enum zram_pageflags flag)
67 return meta->table[index].flags & BIT(flag);
70 static void zram_set_flag(struct zram_meta *meta, u32 index,
71 enum zram_pageflags flag)
73 meta->table[index].flags |= BIT(flag);
76 static void zram_clear_flag(struct zram_meta *meta, u32 index,
77 enum zram_pageflags flag)
79 meta->table[index].flags &= ~BIT(flag);
82 static int page_zero_filled(void *ptr)
87 page = (unsigned long *)ptr;
89 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
97 static void zram_free_page(struct zram *zram, size_t index)
99 struct zram_meta *meta = zram->meta;
100 unsigned long handle = meta->table[index].handle;
101 u16 size = meta->table[index].size;
103 if (unlikely(!handle)) {
105 * No memory is allocated for zero filled pages.
106 * Simply clear zero page flag.
108 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
109 zram_clear_flag(meta, index, ZRAM_ZERO);
110 zram->stats.pages_zero--;
115 if (unlikely(size > max_zpage_size))
116 zram->stats.bad_compress--;
118 zs_free(meta->mem_pool, handle);
120 if (size <= PAGE_SIZE / 2)
121 zram->stats.good_compress--;
123 zram_stat64_sub(zram, &zram->stats.compr_size,
124 meta->table[index].size);
125 zram->stats.pages_stored--;
127 meta->table[index].handle = 0;
128 meta->table[index].size = 0;
131 static void handle_zero_page(struct bio_vec *bvec)
133 struct page *page = bvec->bv_page;
136 user_mem = kmap_atomic(page);
137 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
138 kunmap_atomic(user_mem);
140 flush_dcache_page(page);
143 static inline int is_partial_io(struct bio_vec *bvec)
145 return bvec->bv_len != PAGE_SIZE;
148 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
151 size_t clen = PAGE_SIZE;
153 struct zram_meta *meta = zram->meta;
154 unsigned long handle = meta->table[index].handle;
156 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
157 memset(mem, 0, PAGE_SIZE);
161 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
162 if (meta->table[index].size == PAGE_SIZE)
163 memcpy(mem, cmem, PAGE_SIZE);
165 ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
167 zs_unmap_object(meta->mem_pool, handle);
169 /* Should NEVER happen. Return bio error if it does. */
170 if (unlikely(ret != LZO_E_OK)) {
171 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
172 zram_stat64_inc(zram, &zram->stats.failed_reads);
179 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
180 u32 index, int offset, struct bio *bio)
184 unsigned char *user_mem, *uncmem = NULL;
185 struct zram_meta *meta = zram->meta;
186 page = bvec->bv_page;
188 if (unlikely(!meta->table[index].handle) ||
189 zram_test_flag(meta, index, ZRAM_ZERO)) {
190 handle_zero_page(bvec);
194 if (is_partial_io(bvec))
195 /* Use a temporary buffer to decompress the page */
196 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
198 user_mem = kmap_atomic(page);
199 if (!is_partial_io(bvec))
203 pr_info("Unable to allocate temp memory\n");
208 ret = zram_decompress_page(zram, uncmem, index);
209 /* Should NEVER happen. Return bio error if it does. */
210 if (unlikely(ret != LZO_E_OK))
213 if (is_partial_io(bvec))
214 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
217 flush_dcache_page(page);
220 kunmap_atomic(user_mem);
221 if (is_partial_io(bvec))
226 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
231 unsigned long handle;
233 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
234 struct zram_meta *meta = zram->meta;
236 page = bvec->bv_page;
237 src = meta->compress_buffer;
239 if (is_partial_io(bvec)) {
241 * This is a partial IO. We need to read the full page
242 * before to write the changes.
244 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
249 ret = zram_decompress_page(zram, uncmem, index);
255 * System overwrites unused sectors. Free memory associated
256 * with this sector now.
258 if (meta->table[index].handle ||
259 zram_test_flag(meta, index, ZRAM_ZERO))
260 zram_free_page(zram, index);
262 user_mem = kmap_atomic(page);
264 if (is_partial_io(bvec)) {
265 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
267 kunmap_atomic(user_mem);
273 if (page_zero_filled(uncmem)) {
274 kunmap_atomic(user_mem);
275 if (is_partial_io(bvec))
277 zram->stats.pages_zero++;
278 zram_set_flag(meta, index, ZRAM_ZERO);
283 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
284 meta->compress_workmem);
286 if (!is_partial_io(bvec)) {
287 kunmap_atomic(user_mem);
292 if (unlikely(ret != LZO_E_OK)) {
293 pr_err("Compression failed! err=%d\n", ret);
297 if (unlikely(clen > max_zpage_size)) {
298 zram->stats.bad_compress++;
301 if (is_partial_io(bvec))
305 handle = zs_malloc(meta->mem_pool, clen);
307 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
312 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
314 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
315 src = kmap_atomic(page);
316 memcpy(cmem, src, clen);
317 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
320 zs_unmap_object(meta->mem_pool, handle);
322 meta->table[index].handle = handle;
323 meta->table[index].size = clen;
326 zram_stat64_add(zram, &zram->stats.compr_size, clen);
327 zram->stats.pages_stored++;
328 if (clen <= PAGE_SIZE / 2)
329 zram->stats.good_compress++;
332 if (is_partial_io(bvec))
336 zram_stat64_inc(zram, &zram->stats.failed_writes);
340 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
341 int offset, struct bio *bio, int rw)
346 down_read(&zram->lock);
347 ret = zram_bvec_read(zram, bvec, index, offset, bio);
348 up_read(&zram->lock);
350 down_write(&zram->lock);
351 ret = zram_bvec_write(zram, bvec, index, offset);
352 up_write(&zram->lock);
358 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
360 if (*offset + bvec->bv_len >= PAGE_SIZE)
362 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
365 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
369 struct bio_vec *bvec;
373 zram_stat64_inc(zram, &zram->stats.num_reads);
376 zram_stat64_inc(zram, &zram->stats.num_writes);
380 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
381 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
383 bio_for_each_segment(bvec, bio, i) {
384 int max_transfer_size = PAGE_SIZE - offset;
386 if (bvec->bv_len > max_transfer_size) {
388 * zram_bvec_rw() can only make operation on a single
389 * zram page. Split the bio vector.
393 bv.bv_page = bvec->bv_page;
394 bv.bv_len = max_transfer_size;
395 bv.bv_offset = bvec->bv_offset;
397 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
400 bv.bv_len = bvec->bv_len - max_transfer_size;
401 bv.bv_offset += max_transfer_size;
402 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
405 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
409 update_position(&index, &offset, bvec);
412 set_bit(BIO_UPTODATE, &bio->bi_flags);
421 * Check if request is within bounds and aligned on zram logical blocks.
423 static inline int valid_io_request(struct zram *zram, struct bio *bio)
426 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
427 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
428 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
433 /* I/O request is valid */
438 * Handler function for all zram I/O requests.
440 static void zram_make_request(struct request_queue *queue, struct bio *bio)
442 struct zram *zram = queue->queuedata;
444 down_read(&zram->init_lock);
445 if (unlikely(!zram->init_done))
448 if (!valid_io_request(zram, bio)) {
449 zram_stat64_inc(zram, &zram->stats.invalid_io);
453 __zram_make_request(zram, bio, bio_data_dir(bio));
454 up_read(&zram->init_lock);
459 up_read(&zram->init_lock);
463 static void __zram_reset_device(struct zram *zram)
466 struct zram_meta *meta;
468 if (!zram->init_done)
474 /* Free all pages that are still in this zram device */
475 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
476 unsigned long handle = meta->table[index].handle;
480 zs_free(meta->mem_pool, handle);
483 zram_meta_free(zram->meta);
486 memset(&zram->stats, 0, sizeof(zram->stats));
489 set_capacity(zram->disk, 0);
492 void zram_reset_device(struct zram *zram)
494 down_write(&zram->init_lock);
495 __zram_reset_device(zram);
496 up_write(&zram->init_lock);
499 void zram_meta_free(struct zram_meta *meta)
501 zs_destroy_pool(meta->mem_pool);
502 kfree(meta->compress_workmem);
503 free_pages((unsigned long)meta->compress_buffer, 1);
508 struct zram_meta *zram_meta_alloc(u64 disksize)
511 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
515 meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
516 if (!meta->compress_workmem)
519 meta->compress_buffer =
520 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
521 if (!meta->compress_buffer) {
522 pr_err("Error allocating compressor buffer space\n");
526 num_pages = disksize >> PAGE_SHIFT;
527 meta->table = vzalloc(num_pages * sizeof(*meta->table));
529 pr_err("Error allocating zram address table\n");
533 meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
534 if (!meta->mem_pool) {
535 pr_err("Error creating memory pool\n");
544 free_pages((unsigned long)meta->compress_buffer, 1);
546 kfree(meta->compress_workmem);
554 void zram_init_device(struct zram *zram, struct zram_meta *meta)
556 if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
558 "There is little point creating a zram of greater than "
559 "twice the size of memory since we expect a 2:1 compression "
560 "ratio. Note that zram uses about 0.1%% of the size of "
561 "the disk when not in use so a huge zram is "
563 "\tMemory Size: %lu kB\n"
564 "\tSize you selected: %llu kB\n"
565 "Continuing anyway ...\n",
566 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
570 /* zram devices sort of resembles non-rotational disks */
571 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
576 pr_debug("Initialization done!\n");
579 static void zram_slot_free_notify(struct block_device *bdev,
584 zram = bdev->bd_disk->private_data;
585 down_write(&zram->lock);
586 zram_free_page(zram, index);
587 up_write(&zram->lock);
588 zram_stat64_inc(zram, &zram->stats.notify_free);
591 static const struct block_device_operations zram_devops = {
592 .swap_slot_free_notify = zram_slot_free_notify,
596 static int create_device(struct zram *zram, int device_id)
600 init_rwsem(&zram->lock);
601 init_rwsem(&zram->init_lock);
602 spin_lock_init(&zram->stat64_lock);
604 zram->queue = blk_alloc_queue(GFP_KERNEL);
606 pr_err("Error allocating disk queue for device %d\n",
612 blk_queue_make_request(zram->queue, zram_make_request);
613 zram->queue->queuedata = zram;
615 /* gendisk structure */
616 zram->disk = alloc_disk(1);
618 blk_cleanup_queue(zram->queue);
619 pr_warn("Error allocating disk structure for device %d\n",
625 zram->disk->major = zram_major;
626 zram->disk->first_minor = device_id;
627 zram->disk->fops = &zram_devops;
628 zram->disk->queue = zram->queue;
629 zram->disk->private_data = zram;
630 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
632 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
633 set_capacity(zram->disk, 0);
636 * To ensure that we always get PAGE_SIZE aligned
637 * and n*PAGE_SIZED sized I/O requests.
639 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
640 blk_queue_logical_block_size(zram->disk->queue,
641 ZRAM_LOGICAL_BLOCK_SIZE);
642 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
643 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
645 add_disk(zram->disk);
647 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
648 &zram_disk_attr_group);
650 pr_warn("Error creating sysfs group");
660 static void destroy_device(struct zram *zram)
662 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
663 &zram_disk_attr_group);
666 del_gendisk(zram->disk);
667 put_disk(zram->disk);
671 blk_cleanup_queue(zram->queue);
674 unsigned int zram_get_num_devices(void)
679 static int __init zram_init(void)
683 if (num_devices > max_num_devices) {
684 pr_warn("Invalid value for num_devices: %u\n",
690 zram_major = register_blkdev(0, "zram");
691 if (zram_major <= 0) {
692 pr_warn("Unable to get major number\n");
697 /* Allocate the device array and initialize each one */
698 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
704 for (dev_id = 0; dev_id < num_devices; dev_id++) {
705 ret = create_device(&zram_devices[dev_id], dev_id);
710 pr_info("Created %u device(s) ...\n", num_devices);
716 destroy_device(&zram_devices[--dev_id]);
719 unregister_blkdev(zram_major, "zram");
724 static void __exit zram_exit(void)
729 for (i = 0; i < num_devices; i++) {
730 zram = &zram_devices[i];
732 get_disk(zram->disk);
733 destroy_device(zram);
734 zram_reset_device(zram);
735 put_disk(zram->disk);
738 unregister_blkdev(zram_major, "zram");
741 pr_debug("Cleanup done!\n");
744 module_param(num_devices, uint, 0);
745 MODULE_PARM_DESC(num_devices, "Number of zram devices");
747 module_init(zram_init);
748 module_exit(zram_exit);
750 MODULE_LICENSE("Dual BSD/GPL");
751 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
752 MODULE_DESCRIPTION("Compressed RAM Block Device");