/drivers/staging/zram/zram_drv.c

https://bitbucket.org/wisechild/galaxy-nexus · C · 732 lines · 537 code · 143 blank · 52 comment · 51 complexity · b061b07ce824f04f240ed520d2f72225 MD5 · raw file

  1. /*
  2. * Compressed RAM block device
  3. *
  4. * Copyright (C) 2008, 2009, 2010 Nitin Gupta
  5. *
  6. * This code is released using a dual license strategy: BSD/GPL
  7. * You can choose the licence that better fits your requirements.
  8. *
  9. * Released under the terms of 3-clause BSD License
  10. * Released under the terms of GNU General Public License Version 2.0
  11. *
  12. * Project home: http://compcache.googlecode.com
  13. */
  14. #define KMSG_COMPONENT "zram"
  15. #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  16. #ifdef CONFIG_ZRAM_DEBUG
  17. #define DEBUG
  18. #endif
  19. #include <linux/module.h>
  20. #include <linux/kernel.h>
  21. #include <linux/bio.h>
  22. #include <linux/bitops.h>
  23. #include <linux/blkdev.h>
  24. #include <linux/buffer_head.h>
  25. #include <linux/device.h>
  26. #include <linux/genhd.h>
  27. #include <linux/highmem.h>
  28. #include <linux/slab.h>
  29. #include <linux/lzo.h>
  30. #include <linux/string.h>
  31. #include <linux/vmalloc.h>
  32. #include "zram_drv.h"
  33. /* Globals */
  34. static int zram_major;
  35. struct zram *devices;
  36. /* Module params (documentation at end) */
  37. unsigned int num_devices;
  38. static void zram_stat_inc(u32 *v)
  39. {
  40. *v = *v + 1;
  41. }
  42. static void zram_stat_dec(u32 *v)
  43. {
  44. *v = *v - 1;
  45. }
  46. static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
  47. {
  48. spin_lock(&zram->stat64_lock);
  49. *v = *v + inc;
  50. spin_unlock(&zram->stat64_lock);
  51. }
  52. static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
  53. {
  54. spin_lock(&zram->stat64_lock);
  55. *v = *v - dec;
  56. spin_unlock(&zram->stat64_lock);
  57. }
  58. static void zram_stat64_inc(struct zram *zram, u64 *v)
  59. {
  60. zram_stat64_add(zram, v, 1);
  61. }
  62. static int zram_test_flag(struct zram *zram, u32 index,
  63. enum zram_pageflags flag)
  64. {
  65. return zram->table[index].flags & BIT(flag);
  66. }
  67. static void zram_set_flag(struct zram *zram, u32 index,
  68. enum zram_pageflags flag)
  69. {
  70. zram->table[index].flags |= BIT(flag);
  71. }
  72. static void zram_clear_flag(struct zram *zram, u32 index,
  73. enum zram_pageflags flag)
  74. {
  75. zram->table[index].flags &= ~BIT(flag);
  76. }
  77. static int page_zero_filled(void *ptr)
  78. {
  79. unsigned int pos;
  80. unsigned long *page;
  81. page = (unsigned long *)ptr;
  82. for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
  83. if (page[pos])
  84. return 0;
  85. }
  86. return 1;
  87. }
  88. static void zram_set_disksize(struct zram *zram, size_t totalram_bytes)
  89. {
  90. if (!zram->disksize) {
  91. pr_info(
  92. "disk size not provided. You can use disksize_kb module "
  93. "param to specify size.\nUsing default: (%u%% of RAM).\n",
  94. default_disksize_perc_ram
  95. );
  96. zram->disksize = default_disksize_perc_ram *
  97. (totalram_bytes / 100);
  98. }
  99. if (zram->disksize > 2 * (totalram_bytes)) {
  100. pr_info(
  101. "There is little point creating a zram of greater than "
  102. "twice the size of memory since we expect a 2:1 compression "
  103. "ratio. Note that zram uses about 0.1%% of the size of "
  104. "the disk when not in use so a huge zram is "
  105. "wasteful.\n"
  106. "\tMemory Size: %zu kB\n"
  107. "\tSize you selected: %llu kB\n"
  108. "Continuing anyway ...\n",
  109. totalram_bytes >> 10, zram->disksize
  110. );
  111. }
  112. zram->disksize &= PAGE_MASK;
  113. }
  114. static void zram_free_page(struct zram *zram, size_t index)
  115. {
  116. u32 clen;
  117. void *obj;
  118. struct page *page = zram->table[index].page;
  119. u32 offset = zram->table[index].offset;
  120. if (unlikely(!page)) {
  121. /*
  122. * No memory is allocated for zero filled pages.
  123. * Simply clear zero page flag.
  124. */
  125. if (zram_test_flag(zram, index, ZRAM_ZERO)) {
  126. zram_clear_flag(zram, index, ZRAM_ZERO);
  127. zram_stat_dec(&zram->stats.pages_zero);
  128. }
  129. return;
  130. }
  131. if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
  132. clen = PAGE_SIZE;
  133. __free_page(page);
  134. zram_clear_flag(zram, index, ZRAM_UNCOMPRESSED);
  135. zram_stat_dec(&zram->stats.pages_expand);
  136. goto out;
  137. }
  138. obj = kmap_atomic(page, KM_USER0) + offset;
  139. clen = xv_get_object_size(obj) - sizeof(struct zobj_header);
  140. kunmap_atomic(obj, KM_USER0);
  141. xv_free(zram->mem_pool, page, offset);
  142. if (clen <= PAGE_SIZE / 2)
  143. zram_stat_dec(&zram->stats.good_compress);
  144. out:
  145. zram_stat64_sub(zram, &zram->stats.compr_size, clen);
  146. zram_stat_dec(&zram->stats.pages_stored);
  147. zram->table[index].page = NULL;
  148. zram->table[index].offset = 0;
  149. }
  150. static void handle_zero_page(struct page *page)
  151. {
  152. void *user_mem;
  153. user_mem = kmap_atomic(page, KM_USER0);
  154. memset(user_mem, 0, PAGE_SIZE);
  155. kunmap_atomic(user_mem, KM_USER0);
  156. flush_dcache_page(page);
  157. }
  158. static void handle_uncompressed_page(struct zram *zram,
  159. struct page *page, u32 index)
  160. {
  161. unsigned char *user_mem, *cmem;
  162. user_mem = kmap_atomic(page, KM_USER0);
  163. cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
  164. zram->table[index].offset;
  165. memcpy(user_mem, cmem, PAGE_SIZE);
  166. kunmap_atomic(user_mem, KM_USER0);
  167. kunmap_atomic(cmem, KM_USER1);
  168. flush_dcache_page(page);
  169. }
  170. static void zram_read(struct zram *zram, struct bio *bio)
  171. {
  172. int i;
  173. u32 index;
  174. struct bio_vec *bvec;
  175. zram_stat64_inc(zram, &zram->stats.num_reads);
  176. index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
  177. bio_for_each_segment(bvec, bio, i) {
  178. int ret;
  179. size_t clen;
  180. struct page *page;
  181. struct zobj_header *zheader;
  182. unsigned char *user_mem, *cmem;
  183. page = bvec->bv_page;
  184. if (zram_test_flag(zram, index, ZRAM_ZERO)) {
  185. handle_zero_page(page);
  186. index++;
  187. continue;
  188. }
  189. /* Requested page is not present in compressed area */
  190. if (unlikely(!zram->table[index].page)) {
  191. pr_debug("Read before write: sector=%lu, size=%u",
  192. (ulong)(bio->bi_sector), bio->bi_size);
  193. handle_zero_page(page);
  194. index++;
  195. continue;
  196. }
  197. /* Page is stored uncompressed since it's incompressible */
  198. if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
  199. handle_uncompressed_page(zram, page, index);
  200. index++;
  201. continue;
  202. }
  203. user_mem = kmap_atomic(page, KM_USER0);
  204. clen = PAGE_SIZE;
  205. cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
  206. zram->table[index].offset;
  207. ret = lzo1x_decompress_safe(
  208. cmem + sizeof(*zheader),
  209. xv_get_object_size(cmem) - sizeof(*zheader),
  210. user_mem, &clen);
  211. kunmap_atomic(user_mem, KM_USER0);
  212. kunmap_atomic(cmem, KM_USER1);
  213. /* Should NEVER happen. Return bio error if it does. */
  214. if (unlikely(ret != LZO_E_OK)) {
  215. pr_err("Decompression failed! err=%d, page=%u\n",
  216. ret, index);
  217. zram_stat64_inc(zram, &zram->stats.failed_reads);
  218. goto out;
  219. }
  220. flush_dcache_page(page);
  221. index++;
  222. }
  223. set_bit(BIO_UPTODATE, &bio->bi_flags);
  224. bio_endio(bio, 0);
  225. return;
  226. out:
  227. bio_io_error(bio);
  228. }
  229. static void zram_write(struct zram *zram, struct bio *bio)
  230. {
  231. int i;
  232. u32 index;
  233. struct bio_vec *bvec;
  234. zram_stat64_inc(zram, &zram->stats.num_writes);
  235. index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
  236. bio_for_each_segment(bvec, bio, i) {
  237. int ret;
  238. u32 offset;
  239. size_t clen;
  240. struct zobj_header *zheader;
  241. struct page *page, *page_store;
  242. unsigned char *user_mem, *cmem, *src;
  243. page = bvec->bv_page;
  244. src = zram->compress_buffer;
  245. /*
  246. * System overwrites unused sectors. Free memory associated
  247. * with this sector now.
  248. */
  249. if (zram->table[index].page ||
  250. zram_test_flag(zram, index, ZRAM_ZERO))
  251. zram_free_page(zram, index);
  252. mutex_lock(&zram->lock);
  253. user_mem = kmap_atomic(page, KM_USER0);
  254. if (page_zero_filled(user_mem)) {
  255. kunmap_atomic(user_mem, KM_USER0);
  256. mutex_unlock(&zram->lock);
  257. zram_stat_inc(&zram->stats.pages_zero);
  258. zram_set_flag(zram, index, ZRAM_ZERO);
  259. index++;
  260. continue;
  261. }
  262. ret = lzo1x_1_compress(user_mem, PAGE_SIZE, src, &clen,
  263. zram->compress_workmem);
  264. kunmap_atomic(user_mem, KM_USER0);
  265. if (unlikely(ret != LZO_E_OK)) {
  266. mutex_unlock(&zram->lock);
  267. pr_err("Compression failed! err=%d\n", ret);
  268. zram_stat64_inc(zram, &zram->stats.failed_writes);
  269. goto out;
  270. }
  271. /*
  272. * Page is incompressible. Store it as-is (uncompressed)
  273. * since we do not want to return too many disk write
  274. * errors which has side effect of hanging the system.
  275. */
  276. if (unlikely(clen > max_zpage_size)) {
  277. clen = PAGE_SIZE;
  278. page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
  279. if (unlikely(!page_store)) {
  280. mutex_unlock(&zram->lock);
  281. pr_info("Error allocating memory for "
  282. "incompressible page: %u\n", index);
  283. zram_stat64_inc(zram,
  284. &zram->stats.failed_writes);
  285. goto out;
  286. }
  287. offset = 0;
  288. zram_set_flag(zram, index, ZRAM_UNCOMPRESSED);
  289. zram_stat_inc(&zram->stats.pages_expand);
  290. zram->table[index].page = page_store;
  291. src = kmap_atomic(page, KM_USER0);
  292. goto memstore;
  293. }
  294. if (xv_malloc(zram->mem_pool, clen + sizeof(*zheader),
  295. &zram->table[index].page, &offset,
  296. GFP_NOIO | __GFP_HIGHMEM)) {
  297. mutex_unlock(&zram->lock);
  298. pr_info("Error allocating memory for compressed "
  299. "page: %u, size=%zu\n", index, clen);
  300. zram_stat64_inc(zram, &zram->stats.failed_writes);
  301. goto out;
  302. }
  303. memstore:
  304. zram->table[index].offset = offset;
  305. cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
  306. zram->table[index].offset;
  307. #if 0
  308. /* Back-reference needed for memory defragmentation */
  309. if (!zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)) {
  310. zheader = (struct zobj_header *)cmem;
  311. zheader->table_idx = index;
  312. cmem += sizeof(*zheader);
  313. }
  314. #endif
  315. memcpy(cmem, src, clen);
  316. kunmap_atomic(cmem, KM_USER1);
  317. if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
  318. kunmap_atomic(src, KM_USER0);
  319. /* Update stats */
  320. zram_stat64_add(zram, &zram->stats.compr_size, clen);
  321. zram_stat_inc(&zram->stats.pages_stored);
  322. if (clen <= PAGE_SIZE / 2)
  323. zram_stat_inc(&zram->stats.good_compress);
  324. mutex_unlock(&zram->lock);
  325. index++;
  326. }
  327. set_bit(BIO_UPTODATE, &bio->bi_flags);
  328. bio_endio(bio, 0);
  329. return;
  330. out:
  331. bio_io_error(bio);
  332. }
  333. /*
  334. * Check if request is within bounds and page aligned.
  335. */
  336. static inline int valid_io_request(struct zram *zram, struct bio *bio)
  337. {
  338. if (unlikely(
  339. (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
  340. (bio->bi_sector & (SECTORS_PER_PAGE - 1)) ||
  341. (bio->bi_size & (PAGE_SIZE - 1)))) {
  342. return 0;
  343. }
  344. /* I/O request is valid */
  345. return 1;
  346. }
  347. /*
  348. * Handler function for all zram I/O requests.
  349. */
  350. static int zram_make_request(struct request_queue *queue, struct bio *bio)
  351. {
  352. struct zram *zram = queue->queuedata;
  353. if (!valid_io_request(zram, bio)) {
  354. zram_stat64_inc(zram, &zram->stats.invalid_io);
  355. bio_io_error(bio);
  356. return 0;
  357. }
  358. if (unlikely(!zram->init_done) && zram_init_device(zram)) {
  359. bio_io_error(bio);
  360. return 0;
  361. }
  362. switch (bio_data_dir(bio)) {
  363. case READ:
  364. zram_read(zram, bio);
  365. break;
  366. case WRITE:
  367. zram_write(zram, bio);
  368. break;
  369. }
  370. return 0;
  371. }
  372. void zram_reset_device(struct zram *zram)
  373. {
  374. size_t index;
  375. mutex_lock(&zram->init_lock);
  376. zram->init_done = 0;
  377. /* Free various per-device buffers */
  378. kfree(zram->compress_workmem);
  379. free_pages((unsigned long)zram->compress_buffer, 1);
  380. zram->compress_workmem = NULL;
  381. zram->compress_buffer = NULL;
  382. /* Free all pages that are still in this zram device */
  383. for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
  384. struct page *page;
  385. u16 offset;
  386. page = zram->table[index].page;
  387. offset = zram->table[index].offset;
  388. if (!page)
  389. continue;
  390. if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
  391. __free_page(page);
  392. else
  393. xv_free(zram->mem_pool, page, offset);
  394. }
  395. vfree(zram->table);
  396. zram->table = NULL;
  397. xv_destroy_pool(zram->mem_pool);
  398. zram->mem_pool = NULL;
  399. /* Reset stats */
  400. memset(&zram->stats, 0, sizeof(zram->stats));
  401. zram->disksize = 0;
  402. mutex_unlock(&zram->init_lock);
  403. }
  404. int zram_init_device(struct zram *zram)
  405. {
  406. int ret;
  407. size_t num_pages;
  408. mutex_lock(&zram->init_lock);
  409. if (zram->init_done) {
  410. mutex_unlock(&zram->init_lock);
  411. return 0;
  412. }
  413. zram_set_disksize(zram, totalram_pages << PAGE_SHIFT);
  414. zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
  415. if (!zram->compress_workmem) {
  416. pr_err("Error allocating compressor working memory!\n");
  417. ret = -ENOMEM;
  418. goto fail;
  419. }
  420. zram->compress_buffer = (void *)__get_free_pages(__GFP_ZERO, 1);
  421. if (!zram->compress_buffer) {
  422. pr_err("Error allocating compressor buffer space\n");
  423. ret = -ENOMEM;
  424. goto fail;
  425. }
  426. num_pages = zram->disksize >> PAGE_SHIFT;
  427. zram->table = vzalloc(num_pages * sizeof(*zram->table));
  428. if (!zram->table) {
  429. pr_err("Error allocating zram address table\n");
  430. /* To prevent accessing table entries during cleanup */
  431. zram->disksize = 0;
  432. ret = -ENOMEM;
  433. goto fail;
  434. }
  435. set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
  436. /* zram devices sort of resembles non-rotational disks */
  437. queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
  438. zram->mem_pool = xv_create_pool();
  439. if (!zram->mem_pool) {
  440. pr_err("Error creating memory pool\n");
  441. ret = -ENOMEM;
  442. goto fail;
  443. }
  444. zram->init_done = 1;
  445. mutex_unlock(&zram->init_lock);
  446. pr_debug("Initialization done!\n");
  447. return 0;
  448. fail:
  449. mutex_unlock(&zram->init_lock);
  450. zram_reset_device(zram);
  451. pr_err("Initialization failed: err=%d\n", ret);
  452. return ret;
  453. }
  454. void zram_slot_free_notify(struct block_device *bdev, unsigned long index)
  455. {
  456. struct zram *zram;
  457. zram = bdev->bd_disk->private_data;
  458. zram_free_page(zram, index);
  459. zram_stat64_inc(zram, &zram->stats.notify_free);
  460. }
  461. static const struct block_device_operations zram_devops = {
  462. .swap_slot_free_notify = zram_slot_free_notify,
  463. .owner = THIS_MODULE
  464. };
  465. static int create_device(struct zram *zram, int device_id)
  466. {
  467. int ret = 0;
  468. mutex_init(&zram->lock);
  469. mutex_init(&zram->init_lock);
  470. spin_lock_init(&zram->stat64_lock);
  471. zram->queue = blk_alloc_queue(GFP_KERNEL);
  472. if (!zram->queue) {
  473. pr_err("Error allocating disk queue for device %d\n",
  474. device_id);
  475. ret = -ENOMEM;
  476. goto out;
  477. }
  478. blk_queue_make_request(zram->queue, zram_make_request);
  479. zram->queue->queuedata = zram;
  480. /* gendisk structure */
  481. zram->disk = alloc_disk(1);
  482. if (!zram->disk) {
  483. blk_cleanup_queue(zram->queue);
  484. pr_warning("Error allocating disk structure for device %d\n",
  485. device_id);
  486. ret = -ENOMEM;
  487. goto out;
  488. }
  489. zram->disk->major = zram_major;
  490. zram->disk->first_minor = device_id;
  491. zram->disk->fops = &zram_devops;
  492. zram->disk->queue = zram->queue;
  493. zram->disk->private_data = zram;
  494. snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
  495. /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
  496. set_capacity(zram->disk, 0);
  497. /*
  498. * To ensure that we always get PAGE_SIZE aligned
  499. * and n*PAGE_SIZED sized I/O requests.
  500. */
  501. blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
  502. blk_queue_logical_block_size(zram->disk->queue,
  503. ZRAM_LOGICAL_BLOCK_SIZE);
  504. blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
  505. blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
  506. add_disk(zram->disk);
  507. ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
  508. &zram_disk_attr_group);
  509. if (ret < 0) {
  510. pr_warning("Error creating sysfs group");
  511. goto out;
  512. }
  513. zram->init_done = 0;
  514. out:
  515. return ret;
  516. }
  517. static void destroy_device(struct zram *zram)
  518. {
  519. sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
  520. &zram_disk_attr_group);
  521. if (zram->disk) {
  522. del_gendisk(zram->disk);
  523. put_disk(zram->disk);
  524. }
  525. if (zram->queue)
  526. blk_cleanup_queue(zram->queue);
  527. }
  528. static int __init zram_init(void)
  529. {
  530. int ret, dev_id;
  531. if (num_devices > max_num_devices) {
  532. pr_warning("Invalid value for num_devices: %u\n",
  533. num_devices);
  534. ret = -EINVAL;
  535. goto out;
  536. }
  537. zram_major = register_blkdev(0, "zram");
  538. if (zram_major <= 0) {
  539. pr_warning("Unable to get major number\n");
  540. ret = -EBUSY;
  541. goto out;
  542. }
  543. if (!num_devices) {
  544. pr_info("num_devices not specified. Using default: 1\n");
  545. num_devices = 1;
  546. }
  547. /* Allocate the device array and initialize each one */
  548. pr_info("Creating %u devices ...\n", num_devices);
  549. devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
  550. if (!devices) {
  551. ret = -ENOMEM;
  552. goto unregister;
  553. }
  554. for (dev_id = 0; dev_id < num_devices; dev_id++) {
  555. ret = create_device(&devices[dev_id], dev_id);
  556. if (ret)
  557. goto free_devices;
  558. }
  559. return 0;
  560. free_devices:
  561. while (dev_id)
  562. destroy_device(&devices[--dev_id]);
  563. kfree(devices);
  564. unregister:
  565. unregister_blkdev(zram_major, "zram");
  566. out:
  567. return ret;
  568. }
  569. static void __exit zram_exit(void)
  570. {
  571. int i;
  572. struct zram *zram;
  573. for (i = 0; i < num_devices; i++) {
  574. zram = &devices[i];
  575. destroy_device(zram);
  576. if (zram->init_done)
  577. zram_reset_device(zram);
  578. }
  579. unregister_blkdev(zram_major, "zram");
  580. kfree(devices);
  581. pr_debug("Cleanup done!\n");
  582. }
  583. module_param(num_devices, uint, 0);
  584. MODULE_PARM_DESC(num_devices, "Number of zram devices");
  585. module_init(zram_init);
  586. module_exit(zram_exit);
  587. MODULE_LICENSE("Dual BSD/GPL");
  588. MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
  589. MODULE_DESCRIPTION("Compressed RAM Block Device");