Commit | Line | Data |
---|---|---|
306b0c95 | 1 | /* |
f1e3cfff | 2 | * Compressed RAM block device |
306b0c95 | 3 | * |
1130ebba | 4 | * Copyright (C) 2008, 2009, 2010 Nitin Gupta |
306b0c95 NG |
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 | ||
f1e3cfff | 15 | #define KMSG_COMPONENT "zram" |
306b0c95 NG |
16 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt |
17 | ||
b1f5b81e RJ |
18 | #ifdef CONFIG_ZRAM_DEBUG |
19 | #define DEBUG | |
20 | #endif | |
21 | ||
306b0c95 NG |
22 | #include <linux/module.h> |
23 | #include <linux/kernel.h> | |
8946a086 | 24 | #include <linux/bio.h> |
306b0c95 NG |
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> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
306b0c95 | 32 | #include <linux/lzo.h> |
306b0c95 | 33 | #include <linux/string.h> |
306b0c95 | 34 | #include <linux/vmalloc.h> |
306b0c95 | 35 | |
16a4bfb9 | 36 | #include "zram_drv.h" |
306b0c95 NG |
37 | |
38 | /* Globals */ | |
f1e3cfff | 39 | static int zram_major; |
0f0e3ba3 | 40 | static struct zram *zram_devices; |
306b0c95 | 41 | |
306b0c95 | 42 | /* Module params (documentation at end) */ |
ca3d70bd | 43 | static unsigned int num_devices = 1; |
33863c21 | 44 | |
9b3bb7ab SS |
45 | static inline struct zram *dev_to_zram(struct device *dev) |
46 | { | |
47 | return (struct zram *)dev_to_disk(dev)->private_data; | |
48 | } | |
49 | ||
50 | static ssize_t disksize_show(struct device *dev, | |
51 | struct device_attribute *attr, char *buf) | |
52 | { | |
53 | struct zram *zram = dev_to_zram(dev); | |
54 | ||
55 | return sprintf(buf, "%llu\n", zram->disksize); | |
56 | } | |
57 | ||
58 | static ssize_t initstate_show(struct device *dev, | |
59 | struct device_attribute *attr, char *buf) | |
60 | { | |
61 | struct zram *zram = dev_to_zram(dev); | |
62 | ||
63 | return sprintf(buf, "%u\n", zram->init_done); | |
64 | } | |
65 | ||
66 | static ssize_t num_reads_show(struct device *dev, | |
67 | struct device_attribute *attr, char *buf) | |
68 | { | |
69 | struct zram *zram = dev_to_zram(dev); | |
70 | ||
71 | return sprintf(buf, "%llu\n", | |
72 | (u64)atomic64_read(&zram->stats.num_reads)); | |
73 | } | |
74 | ||
75 | static ssize_t num_writes_show(struct device *dev, | |
76 | struct device_attribute *attr, char *buf) | |
77 | { | |
78 | struct zram *zram = dev_to_zram(dev); | |
79 | ||
80 | return sprintf(buf, "%llu\n", | |
81 | (u64)atomic64_read(&zram->stats.num_writes)); | |
82 | } | |
83 | ||
84 | static ssize_t invalid_io_show(struct device *dev, | |
85 | struct device_attribute *attr, char *buf) | |
86 | { | |
87 | struct zram *zram = dev_to_zram(dev); | |
88 | ||
89 | return sprintf(buf, "%llu\n", | |
90 | (u64)atomic64_read(&zram->stats.invalid_io)); | |
91 | } | |
92 | ||
93 | static ssize_t notify_free_show(struct device *dev, | |
94 | struct device_attribute *attr, char *buf) | |
95 | { | |
96 | struct zram *zram = dev_to_zram(dev); | |
97 | ||
98 | return sprintf(buf, "%llu\n", | |
99 | (u64)atomic64_read(&zram->stats.notify_free)); | |
100 | } | |
101 | ||
102 | static ssize_t zero_pages_show(struct device *dev, | |
103 | struct device_attribute *attr, char *buf) | |
104 | { | |
105 | struct zram *zram = dev_to_zram(dev); | |
106 | ||
107 | return sprintf(buf, "%u\n", zram->stats.pages_zero); | |
108 | } | |
109 | ||
110 | static ssize_t orig_data_size_show(struct device *dev, | |
111 | struct device_attribute *attr, char *buf) | |
112 | { | |
113 | struct zram *zram = dev_to_zram(dev); | |
114 | ||
115 | return sprintf(buf, "%llu\n", | |
116 | (u64)(zram->stats.pages_stored) << PAGE_SHIFT); | |
117 | } | |
118 | ||
119 | static ssize_t compr_data_size_show(struct device *dev, | |
120 | struct device_attribute *attr, char *buf) | |
121 | { | |
122 | struct zram *zram = dev_to_zram(dev); | |
123 | ||
124 | return sprintf(buf, "%llu\n", | |
125 | (u64)atomic64_read(&zram->stats.compr_size)); | |
126 | } | |
127 | ||
128 | static ssize_t mem_used_total_show(struct device *dev, | |
129 | struct device_attribute *attr, char *buf) | |
130 | { | |
131 | u64 val = 0; | |
132 | struct zram *zram = dev_to_zram(dev); | |
133 | struct zram_meta *meta = zram->meta; | |
134 | ||
135 | down_read(&zram->init_lock); | |
136 | if (zram->init_done) | |
137 | val = zs_get_total_size_bytes(meta->mem_pool); | |
138 | up_read(&zram->init_lock); | |
139 | ||
140 | return sprintf(buf, "%llu\n", val); | |
141 | } | |
142 | ||
8b3cc3ed | 143 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 144 | enum zram_pageflags flag) |
306b0c95 | 145 | { |
8b3cc3ed | 146 | return meta->table[index].flags & BIT(flag); |
306b0c95 NG |
147 | } |
148 | ||
8b3cc3ed | 149 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 150 | enum zram_pageflags flag) |
306b0c95 | 151 | { |
8b3cc3ed | 152 | meta->table[index].flags |= BIT(flag); |
306b0c95 NG |
153 | } |
154 | ||
8b3cc3ed | 155 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 156 | enum zram_pageflags flag) |
306b0c95 | 157 | { |
8b3cc3ed | 158 | meta->table[index].flags &= ~BIT(flag); |
306b0c95 NG |
159 | } |
160 | ||
9b3bb7ab SS |
161 | static inline int is_partial_io(struct bio_vec *bvec) |
162 | { | |
163 | return bvec->bv_len != PAGE_SIZE; | |
164 | } | |
165 | ||
166 | /* | |
167 | * Check if request is within bounds and aligned on zram logical blocks. | |
168 | */ | |
169 | static inline int valid_io_request(struct zram *zram, struct bio *bio) | |
170 | { | |
171 | u64 start, end, bound; | |
172 | ||
173 | /* unaligned request */ | |
174 | if (unlikely(bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1))) | |
175 | return 0; | |
176 | if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) | |
177 | return 0; | |
178 | ||
179 | start = bio->bi_sector; | |
180 | end = start + (bio->bi_size >> SECTOR_SHIFT); | |
181 | bound = zram->disksize >> SECTOR_SHIFT; | |
182 | /* out of range range */ | |
75c7caf5 | 183 | if (unlikely(start >= bound || end > bound || start > end)) |
9b3bb7ab SS |
184 | return 0; |
185 | ||
186 | /* I/O request is valid */ | |
187 | return 1; | |
188 | } | |
189 | ||
190 | static void zram_meta_free(struct zram_meta *meta) | |
191 | { | |
192 | zs_destroy_pool(meta->mem_pool); | |
193 | kfree(meta->compress_workmem); | |
194 | free_pages((unsigned long)meta->compress_buffer, 1); | |
195 | vfree(meta->table); | |
196 | kfree(meta); | |
197 | } | |
198 | ||
199 | static struct zram_meta *zram_meta_alloc(u64 disksize) | |
200 | { | |
201 | size_t num_pages; | |
202 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); | |
203 | if (!meta) | |
204 | goto out; | |
205 | ||
206 | meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); | |
207 | if (!meta->compress_workmem) | |
208 | goto free_meta; | |
209 | ||
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"); | |
214 | goto free_workmem; | |
215 | } | |
216 | ||
217 | num_pages = disksize >> PAGE_SHIFT; | |
218 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
219 | if (!meta->table) { | |
220 | pr_err("Error allocating zram address table\n"); | |
221 | goto free_buffer; | |
222 | } | |
223 | ||
224 | meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM); | |
225 | if (!meta->mem_pool) { | |
226 | pr_err("Error creating memory pool\n"); | |
227 | goto free_table; | |
228 | } | |
229 | ||
230 | return meta; | |
231 | ||
232 | free_table: | |
233 | vfree(meta->table); | |
234 | free_buffer: | |
235 | free_pages((unsigned long)meta->compress_buffer, 1); | |
236 | free_workmem: | |
237 | kfree(meta->compress_workmem); | |
238 | free_meta: | |
239 | kfree(meta); | |
240 | meta = NULL; | |
241 | out: | |
242 | return meta; | |
243 | } | |
244 | ||
245 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
246 | { | |
247 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
248 | (*index)++; | |
249 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
250 | } | |
251 | ||
306b0c95 NG |
252 | static int page_zero_filled(void *ptr) |
253 | { | |
254 | unsigned int pos; | |
255 | unsigned long *page; | |
256 | ||
257 | page = (unsigned long *)ptr; | |
258 | ||
259 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
260 | if (page[pos]) | |
261 | return 0; | |
262 | } | |
263 | ||
264 | return 1; | |
265 | } | |
266 | ||
9b3bb7ab SS |
267 | static void handle_zero_page(struct bio_vec *bvec) |
268 | { | |
269 | struct page *page = bvec->bv_page; | |
270 | void *user_mem; | |
271 | ||
272 | user_mem = kmap_atomic(page); | |
273 | if (is_partial_io(bvec)) | |
274 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); | |
275 | else | |
276 | clear_page(user_mem); | |
277 | kunmap_atomic(user_mem); | |
278 | ||
279 | flush_dcache_page(page); | |
280 | } | |
281 | ||
f1e3cfff | 282 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 283 | { |
8b3cc3ed MK |
284 | struct zram_meta *meta = zram->meta; |
285 | unsigned long handle = meta->table[index].handle; | |
286 | u16 size = meta->table[index].size; | |
306b0c95 | 287 | |
fd1a30de | 288 | if (unlikely(!handle)) { |
2e882281 NG |
289 | /* |
290 | * No memory is allocated for zero filled pages. | |
291 | * Simply clear zero page flag. | |
292 | */ | |
8b3cc3ed MK |
293 | if (zram_test_flag(meta, index, ZRAM_ZERO)) { |
294 | zram_clear_flag(meta, index, ZRAM_ZERO); | |
d178a07c | 295 | zram->stats.pages_zero--; |
306b0c95 NG |
296 | } |
297 | return; | |
298 | } | |
299 | ||
130f315a | 300 | if (unlikely(size > max_zpage_size)) |
d178a07c | 301 | zram->stats.bad_compress--; |
306b0c95 | 302 | |
8b3cc3ed | 303 | zs_free(meta->mem_pool, handle); |
306b0c95 | 304 | |
130f315a | 305 | if (size <= PAGE_SIZE / 2) |
d178a07c | 306 | zram->stats.good_compress--; |
306b0c95 | 307 | |
da5cc7d3 | 308 | atomic64_sub(meta->table[index].size, &zram->stats.compr_size); |
d178a07c | 309 | zram->stats.pages_stored--; |
306b0c95 | 310 | |
8b3cc3ed MK |
311 | meta->table[index].handle = 0; |
312 | meta->table[index].size = 0; | |
306b0c95 NG |
313 | } |
314 | ||
37b51fdd | 315 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 316 | { |
37b51fdd SS |
317 | int ret = LZO_E_OK; |
318 | size_t clen = PAGE_SIZE; | |
319 | unsigned char *cmem; | |
8b3cc3ed MK |
320 | struct zram_meta *meta = zram->meta; |
321 | unsigned long handle = meta->table[index].handle; | |
306b0c95 | 322 | |
8b3cc3ed | 323 | if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) { |
42e99bd9 | 324 | clear_page(mem); |
8c921b2b JM |
325 | return 0; |
326 | } | |
306b0c95 | 327 | |
8b3cc3ed MK |
328 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
329 | if (meta->table[index].size == PAGE_SIZE) | |
42e99bd9 | 330 | copy_page(mem, cmem); |
37b51fdd | 331 | else |
8b3cc3ed | 332 | ret = lzo1x_decompress_safe(cmem, meta->table[index].size, |
37b51fdd | 333 | mem, &clen); |
8b3cc3ed | 334 | zs_unmap_object(meta->mem_pool, handle); |
a1dd52af | 335 | |
8c921b2b JM |
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); | |
da5cc7d3 | 339 | atomic64_inc(&zram->stats.failed_reads); |
8c921b2b | 340 | return ret; |
a1dd52af | 341 | } |
306b0c95 | 342 | |
8c921b2b | 343 | return 0; |
306b0c95 NG |
344 | } |
345 | ||
37b51fdd SS |
346 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
347 | u32 index, int offset, struct bio *bio) | |
924bd88d JM |
348 | { |
349 | int ret; | |
37b51fdd SS |
350 | struct page *page; |
351 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 352 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
353 | page = bvec->bv_page; |
354 | ||
8b3cc3ed MK |
355 | if (unlikely(!meta->table[index].handle) || |
356 | zram_test_flag(meta, index, ZRAM_ZERO)) { | |
37b51fdd | 357 | handle_zero_page(bvec); |
924bd88d JM |
358 | return 0; |
359 | } | |
360 | ||
37b51fdd SS |
361 | if (is_partial_io(bvec)) |
362 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
363 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
364 | ||
365 | user_mem = kmap_atomic(page); | |
366 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
367 | uncmem = user_mem; |
368 | ||
369 | if (!uncmem) { | |
370 | pr_info("Unable to allocate temp memory\n"); | |
371 | ret = -ENOMEM; | |
372 | goto out_cleanup; | |
373 | } | |
924bd88d | 374 | |
37b51fdd | 375 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 376 | /* Should NEVER happen. Return bio error if it does. */ |
25eeb667 | 377 | if (unlikely(ret != LZO_E_OK)) |
37b51fdd | 378 | goto out_cleanup; |
924bd88d | 379 | |
37b51fdd SS |
380 | if (is_partial_io(bvec)) |
381 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
382 | bvec->bv_len); | |
383 | ||
384 | flush_dcache_page(page); | |
385 | ret = 0; | |
386 | out_cleanup: | |
387 | kunmap_atomic(user_mem); | |
388 | if (is_partial_io(bvec)) | |
389 | kfree(uncmem); | |
390 | return ret; | |
924bd88d JM |
391 | } |
392 | ||
393 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
394 | int offset) | |
306b0c95 | 395 | { |
397c6066 | 396 | int ret = 0; |
8c921b2b | 397 | size_t clen; |
c2344348 | 398 | unsigned long handle; |
130f315a | 399 | struct page *page; |
924bd88d | 400 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 401 | struct zram_meta *meta = zram->meta; |
306b0c95 | 402 | |
8c921b2b | 403 | page = bvec->bv_page; |
8b3cc3ed | 404 | src = meta->compress_buffer; |
306b0c95 | 405 | |
924bd88d JM |
406 | if (is_partial_io(bvec)) { |
407 | /* | |
408 | * This is a partial IO. We need to read the full page | |
409 | * before to write the changes. | |
410 | */ | |
7e5a5104 | 411 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 412 | if (!uncmem) { |
924bd88d JM |
413 | ret = -ENOMEM; |
414 | goto out; | |
415 | } | |
37b51fdd | 416 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 417 | if (ret) |
924bd88d | 418 | goto out; |
924bd88d JM |
419 | } |
420 | ||
8c921b2b JM |
421 | /* |
422 | * System overwrites unused sectors. Free memory associated | |
423 | * with this sector now. | |
424 | */ | |
8b3cc3ed MK |
425 | if (meta->table[index].handle || |
426 | zram_test_flag(meta, index, ZRAM_ZERO)) | |
8c921b2b | 427 | zram_free_page(zram, index); |
306b0c95 | 428 | |
ba82fe2e | 429 | user_mem = kmap_atomic(page); |
924bd88d | 430 | |
397c6066 | 431 | if (is_partial_io(bvec)) { |
924bd88d JM |
432 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
433 | bvec->bv_len); | |
397c6066 NG |
434 | kunmap_atomic(user_mem); |
435 | user_mem = NULL; | |
436 | } else { | |
924bd88d | 437 | uncmem = user_mem; |
397c6066 | 438 | } |
924bd88d JM |
439 | |
440 | if (page_zero_filled(uncmem)) { | |
ba82fe2e | 441 | kunmap_atomic(user_mem); |
d178a07c | 442 | zram->stats.pages_zero++; |
8b3cc3ed | 443 | zram_set_flag(meta, index, ZRAM_ZERO); |
924bd88d JM |
444 | ret = 0; |
445 | goto out; | |
8c921b2b | 446 | } |
306b0c95 | 447 | |
924bd88d | 448 | ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen, |
8b3cc3ed | 449 | meta->compress_workmem); |
306b0c95 | 450 | |
397c6066 NG |
451 | if (!is_partial_io(bvec)) { |
452 | kunmap_atomic(user_mem); | |
453 | user_mem = NULL; | |
454 | uncmem = NULL; | |
455 | } | |
306b0c95 | 456 | |
8c921b2b | 457 | if (unlikely(ret != LZO_E_OK)) { |
8c921b2b | 458 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 459 | goto out; |
8c921b2b | 460 | } |
306b0c95 | 461 | |
c8f2f0db | 462 | if (unlikely(clen > max_zpage_size)) { |
d178a07c | 463 | zram->stats.bad_compress++; |
c8f2f0db | 464 | clen = PAGE_SIZE; |
397c6066 NG |
465 | src = NULL; |
466 | if (is_partial_io(bvec)) | |
467 | src = uncmem; | |
c8f2f0db | 468 | } |
a1dd52af | 469 | |
8b3cc3ed | 470 | handle = zs_malloc(meta->mem_pool, clen); |
fd1a30de | 471 | if (!handle) { |
596b3dd4 MR |
472 | pr_info("Error allocating memory for compressed page: %u, size=%zu\n", |
473 | index, clen); | |
924bd88d JM |
474 | ret = -ENOMEM; |
475 | goto out; | |
8c921b2b | 476 | } |
8b3cc3ed | 477 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 478 | |
42e99bd9 | 479 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 480 | src = kmap_atomic(page); |
42e99bd9 | 481 | copy_page(cmem, src); |
397c6066 | 482 | kunmap_atomic(src); |
42e99bd9 JL |
483 | } else { |
484 | memcpy(cmem, src, clen); | |
485 | } | |
306b0c95 | 486 | |
8b3cc3ed | 487 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 488 | |
8b3cc3ed MK |
489 | meta->table[index].handle = handle; |
490 | meta->table[index].size = clen; | |
306b0c95 | 491 | |
8c921b2b | 492 | /* Update stats */ |
da5cc7d3 | 493 | atomic64_add(clen, &zram->stats.compr_size); |
d178a07c | 494 | zram->stats.pages_stored++; |
8c921b2b | 495 | if (clen <= PAGE_SIZE / 2) |
d178a07c | 496 | zram->stats.good_compress++; |
306b0c95 | 497 | |
924bd88d | 498 | out: |
397c6066 NG |
499 | if (is_partial_io(bvec)) |
500 | kfree(uncmem); | |
501 | ||
924bd88d | 502 | if (ret) |
da5cc7d3 | 503 | atomic64_inc(&zram->stats.failed_writes); |
924bd88d | 504 | return ret; |
8c921b2b JM |
505 | } |
506 | ||
507 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
924bd88d | 508 | int offset, struct bio *bio, int rw) |
8c921b2b | 509 | { |
c5bde238 | 510 | int ret; |
8c921b2b | 511 | |
c5bde238 JM |
512 | if (rw == READ) { |
513 | down_read(&zram->lock); | |
514 | ret = zram_bvec_read(zram, bvec, index, offset, bio); | |
515 | up_read(&zram->lock); | |
516 | } else { | |
517 | down_write(&zram->lock); | |
518 | ret = zram_bvec_write(zram, bvec, index, offset); | |
519 | up_write(&zram->lock); | |
520 | } | |
521 | ||
522 | return ret; | |
924bd88d JM |
523 | } |
524 | ||
9b3bb7ab | 525 | static void zram_reset_device(struct zram *zram) |
924bd88d | 526 | { |
9b3bb7ab SS |
527 | size_t index; |
528 | struct zram_meta *meta; | |
529 | ||
644d4787 SS |
530 | down_write(&zram->init_lock); |
531 | if (!zram->init_done) { | |
532 | up_write(&zram->init_lock); | |
9b3bb7ab | 533 | return; |
644d4787 | 534 | } |
9b3bb7ab SS |
535 | |
536 | meta = zram->meta; | |
537 | zram->init_done = 0; | |
538 | ||
539 | /* Free all pages that are still in this zram device */ | |
540 | for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) { | |
541 | unsigned long handle = meta->table[index].handle; | |
542 | if (!handle) | |
543 | continue; | |
544 | ||
545 | zs_free(meta->mem_pool, handle); | |
546 | } | |
547 | ||
548 | zram_meta_free(zram->meta); | |
549 | zram->meta = NULL; | |
550 | /* Reset stats */ | |
551 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
552 | ||
553 | zram->disksize = 0; | |
554 | set_capacity(zram->disk, 0); | |
644d4787 | 555 | up_write(&zram->init_lock); |
9b3bb7ab SS |
556 | } |
557 | ||
558 | static void zram_init_device(struct zram *zram, struct zram_meta *meta) | |
559 | { | |
560 | if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) { | |
561 | pr_info( | |
562 | "There is little point creating a zram of greater than " | |
563 | "twice the size of memory since we expect a 2:1 compression " | |
564 | "ratio. Note that zram uses about 0.1%% of the size of " | |
565 | "the disk when not in use so a huge zram is " | |
566 | "wasteful.\n" | |
567 | "\tMemory Size: %lu kB\n" | |
568 | "\tSize you selected: %llu kB\n" | |
569 | "Continuing anyway ...\n", | |
570 | (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10 | |
571 | ); | |
572 | } | |
573 | ||
574 | /* zram devices sort of resembles non-rotational disks */ | |
575 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
576 | ||
577 | zram->meta = meta; | |
578 | zram->init_done = 1; | |
579 | ||
580 | pr_debug("Initialization done!\n"); | |
581 | } | |
582 | ||
583 | static ssize_t disksize_store(struct device *dev, | |
584 | struct device_attribute *attr, const char *buf, size_t len) | |
585 | { | |
586 | u64 disksize; | |
587 | struct zram_meta *meta; | |
588 | struct zram *zram = dev_to_zram(dev); | |
589 | ||
590 | disksize = memparse(buf, NULL); | |
591 | if (!disksize) | |
592 | return -EINVAL; | |
593 | ||
594 | disksize = PAGE_ALIGN(disksize); | |
595 | meta = zram_meta_alloc(disksize); | |
596 | down_write(&zram->init_lock); | |
597 | if (zram->init_done) { | |
598 | up_write(&zram->init_lock); | |
599 | zram_meta_free(meta); | |
600 | pr_info("Cannot change disksize for initialized device\n"); | |
601 | return -EBUSY; | |
602 | } | |
603 | ||
604 | zram->disksize = disksize; | |
605 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
606 | zram_init_device(zram, meta); | |
607 | up_write(&zram->init_lock); | |
608 | ||
609 | return len; | |
610 | } | |
611 | ||
612 | static ssize_t reset_store(struct device *dev, | |
613 | struct device_attribute *attr, const char *buf, size_t len) | |
614 | { | |
615 | int ret; | |
616 | unsigned short do_reset; | |
617 | struct zram *zram; | |
618 | struct block_device *bdev; | |
619 | ||
620 | zram = dev_to_zram(dev); | |
621 | bdev = bdget_disk(zram->disk, 0); | |
622 | ||
623 | /* Do not reset an active device! */ | |
624 | if (bdev->bd_holders) | |
625 | return -EBUSY; | |
626 | ||
627 | ret = kstrtou16(buf, 10, &do_reset); | |
628 | if (ret) | |
629 | return ret; | |
630 | ||
631 | if (!do_reset) | |
632 | return -EINVAL; | |
633 | ||
634 | /* Make sure all pending I/O is finished */ | |
635 | if (bdev) | |
636 | fsync_bdev(bdev); | |
637 | ||
638 | zram_reset_device(zram); | |
639 | return len; | |
8c921b2b JM |
640 | } |
641 | ||
642 | static void __zram_make_request(struct zram *zram, struct bio *bio, int rw) | |
643 | { | |
924bd88d | 644 | int i, offset; |
8c921b2b JM |
645 | u32 index; |
646 | struct bio_vec *bvec; | |
647 | ||
648 | switch (rw) { | |
649 | case READ: | |
da5cc7d3 | 650 | atomic64_inc(&zram->stats.num_reads); |
8c921b2b JM |
651 | break; |
652 | case WRITE: | |
da5cc7d3 | 653 | atomic64_inc(&zram->stats.num_writes); |
8c921b2b JM |
654 | break; |
655 | } | |
656 | ||
657 | index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT; | |
924bd88d | 658 | offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; |
8c921b2b JM |
659 | |
660 | bio_for_each_segment(bvec, bio, i) { | |
924bd88d JM |
661 | int max_transfer_size = PAGE_SIZE - offset; |
662 | ||
663 | if (bvec->bv_len > max_transfer_size) { | |
664 | /* | |
665 | * zram_bvec_rw() can only make operation on a single | |
666 | * zram page. Split the bio vector. | |
667 | */ | |
668 | struct bio_vec bv; | |
669 | ||
670 | bv.bv_page = bvec->bv_page; | |
671 | bv.bv_len = max_transfer_size; | |
672 | bv.bv_offset = bvec->bv_offset; | |
673 | ||
674 | if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0) | |
675 | goto out; | |
676 | ||
677 | bv.bv_len = bvec->bv_len - max_transfer_size; | |
678 | bv.bv_offset += max_transfer_size; | |
679 | if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0) | |
680 | goto out; | |
681 | } else | |
682 | if (zram_bvec_rw(zram, bvec, index, offset, bio, rw) | |
683 | < 0) | |
684 | goto out; | |
685 | ||
686 | update_position(&index, &offset, bvec); | |
a1dd52af | 687 | } |
306b0c95 NG |
688 | |
689 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
690 | bio_endio(bio, 0); | |
7d7854b4 | 691 | return; |
306b0c95 NG |
692 | |
693 | out: | |
306b0c95 | 694 | bio_io_error(bio); |
306b0c95 NG |
695 | } |
696 | ||
306b0c95 | 697 | /* |
f1e3cfff | 698 | * Handler function for all zram I/O requests. |
306b0c95 | 699 | */ |
5a7bbad2 | 700 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 701 | { |
f1e3cfff | 702 | struct zram *zram = queue->queuedata; |
306b0c95 | 703 | |
0900beae JM |
704 | down_read(&zram->init_lock); |
705 | if (unlikely(!zram->init_done)) | |
3de738cd | 706 | goto error; |
0900beae | 707 | |
f1e3cfff | 708 | if (!valid_io_request(zram, bio)) { |
da5cc7d3 | 709 | atomic64_inc(&zram->stats.invalid_io); |
3de738cd | 710 | goto error; |
6642a67c JM |
711 | } |
712 | ||
8c921b2b | 713 | __zram_make_request(zram, bio, bio_data_dir(bio)); |
0900beae | 714 | up_read(&zram->init_lock); |
306b0c95 | 715 | |
b4fdcb02 | 716 | return; |
0900beae | 717 | |
0900beae | 718 | error: |
3de738cd | 719 | up_read(&zram->init_lock); |
0900beae | 720 | bio_io_error(bio); |
306b0c95 NG |
721 | } |
722 | ||
2ccbec05 NG |
723 | static void zram_slot_free_notify(struct block_device *bdev, |
724 | unsigned long index) | |
107c161b | 725 | { |
f1e3cfff | 726 | struct zram *zram; |
107c161b | 727 | |
f1e3cfff | 728 | zram = bdev->bd_disk->private_data; |
57ab0485 | 729 | down_write(&zram->lock); |
f1e3cfff | 730 | zram_free_page(zram, index); |
57ab0485 | 731 | up_write(&zram->lock); |
da5cc7d3 | 732 | atomic64_inc(&zram->stats.notify_free); |
107c161b NG |
733 | } |
734 | ||
f1e3cfff | 735 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 736 | .swap_slot_free_notify = zram_slot_free_notify, |
107c161b | 737 | .owner = THIS_MODULE |
306b0c95 NG |
738 | }; |
739 | ||
9b3bb7ab SS |
740 | static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR, |
741 | disksize_show, disksize_store); | |
742 | static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL); | |
743 | static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store); | |
744 | static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL); | |
745 | static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL); | |
746 | static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL); | |
747 | static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL); | |
748 | static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL); | |
749 | static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL); | |
750 | static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL); | |
751 | static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL); | |
752 | ||
753 | static struct attribute *zram_disk_attrs[] = { | |
754 | &dev_attr_disksize.attr, | |
755 | &dev_attr_initstate.attr, | |
756 | &dev_attr_reset.attr, | |
757 | &dev_attr_num_reads.attr, | |
758 | &dev_attr_num_writes.attr, | |
759 | &dev_attr_invalid_io.attr, | |
760 | &dev_attr_notify_free.attr, | |
761 | &dev_attr_zero_pages.attr, | |
762 | &dev_attr_orig_data_size.attr, | |
763 | &dev_attr_compr_data_size.attr, | |
764 | &dev_attr_mem_used_total.attr, | |
765 | NULL, | |
766 | }; | |
767 | ||
768 | static struct attribute_group zram_disk_attr_group = { | |
769 | .attrs = zram_disk_attrs, | |
770 | }; | |
771 | ||
f1e3cfff | 772 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 773 | { |
39a9b8ac | 774 | int ret = -ENOMEM; |
de1a21a0 | 775 | |
c5bde238 | 776 | init_rwsem(&zram->lock); |
0900beae | 777 | init_rwsem(&zram->init_lock); |
306b0c95 | 778 | |
f1e3cfff NG |
779 | zram->queue = blk_alloc_queue(GFP_KERNEL); |
780 | if (!zram->queue) { | |
306b0c95 NG |
781 | pr_err("Error allocating disk queue for device %d\n", |
782 | device_id); | |
de1a21a0 | 783 | goto out; |
306b0c95 NG |
784 | } |
785 | ||
f1e3cfff NG |
786 | blk_queue_make_request(zram->queue, zram_make_request); |
787 | zram->queue->queuedata = zram; | |
306b0c95 NG |
788 | |
789 | /* gendisk structure */ | |
f1e3cfff NG |
790 | zram->disk = alloc_disk(1); |
791 | if (!zram->disk) { | |
94b8435f | 792 | pr_warn("Error allocating disk structure for device %d\n", |
306b0c95 | 793 | device_id); |
39a9b8ac | 794 | goto out_free_queue; |
306b0c95 NG |
795 | } |
796 | ||
f1e3cfff NG |
797 | zram->disk->major = zram_major; |
798 | zram->disk->first_minor = device_id; | |
799 | zram->disk->fops = &zram_devops; | |
800 | zram->disk->queue = zram->queue; | |
801 | zram->disk->private_data = zram; | |
802 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 803 | |
33863c21 | 804 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 805 | set_capacity(zram->disk, 0); |
5d83d5a0 | 806 | |
a1dd52af NG |
807 | /* |
808 | * To ensure that we always get PAGE_SIZE aligned | |
809 | * and n*PAGE_SIZED sized I/O requests. | |
810 | */ | |
f1e3cfff | 811 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
812 | blk_queue_logical_block_size(zram->disk->queue, |
813 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
814 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
815 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
5d83d5a0 | 816 | |
f1e3cfff | 817 | add_disk(zram->disk); |
306b0c95 | 818 | |
33863c21 NG |
819 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
820 | &zram_disk_attr_group); | |
821 | if (ret < 0) { | |
94b8435f | 822 | pr_warn("Error creating sysfs group"); |
39a9b8ac | 823 | goto out_free_disk; |
33863c21 | 824 | } |
33863c21 | 825 | |
f1e3cfff | 826 | zram->init_done = 0; |
39a9b8ac | 827 | return 0; |
de1a21a0 | 828 | |
39a9b8ac JL |
829 | out_free_disk: |
830 | del_gendisk(zram->disk); | |
831 | put_disk(zram->disk); | |
832 | out_free_queue: | |
833 | blk_cleanup_queue(zram->queue); | |
de1a21a0 NG |
834 | out: |
835 | return ret; | |
306b0c95 NG |
836 | } |
837 | ||
f1e3cfff | 838 | static void destroy_device(struct zram *zram) |
306b0c95 | 839 | { |
33863c21 NG |
840 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, |
841 | &zram_disk_attr_group); | |
33863c21 | 842 | |
f1e3cfff NG |
843 | if (zram->disk) { |
844 | del_gendisk(zram->disk); | |
845 | put_disk(zram->disk); | |
306b0c95 NG |
846 | } |
847 | ||
f1e3cfff NG |
848 | if (zram->queue) |
849 | blk_cleanup_queue(zram->queue); | |
306b0c95 NG |
850 | } |
851 | ||
f1e3cfff | 852 | static int __init zram_init(void) |
306b0c95 | 853 | { |
de1a21a0 | 854 | int ret, dev_id; |
306b0c95 | 855 | |
5fa5a901 | 856 | if (num_devices > max_num_devices) { |
94b8435f | 857 | pr_warn("Invalid value for num_devices: %u\n", |
5fa5a901 | 858 | num_devices); |
de1a21a0 NG |
859 | ret = -EINVAL; |
860 | goto out; | |
306b0c95 NG |
861 | } |
862 | ||
f1e3cfff NG |
863 | zram_major = register_blkdev(0, "zram"); |
864 | if (zram_major <= 0) { | |
94b8435f | 865 | pr_warn("Unable to get major number\n"); |
de1a21a0 NG |
866 | ret = -EBUSY; |
867 | goto out; | |
306b0c95 NG |
868 | } |
869 | ||
306b0c95 | 870 | /* Allocate the device array and initialize each one */ |
5fa5a901 | 871 | zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL); |
43801f6e | 872 | if (!zram_devices) { |
de1a21a0 NG |
873 | ret = -ENOMEM; |
874 | goto unregister; | |
875 | } | |
306b0c95 | 876 | |
5fa5a901 | 877 | for (dev_id = 0; dev_id < num_devices; dev_id++) { |
43801f6e | 878 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 879 | if (ret) |
3bf040c7 | 880 | goto free_devices; |
de1a21a0 NG |
881 | } |
882 | ||
ca3d70bd DB |
883 | pr_info("Created %u device(s) ...\n", num_devices); |
884 | ||
306b0c95 | 885 | return 0; |
de1a21a0 | 886 | |
3bf040c7 | 887 | free_devices: |
de1a21a0 | 888 | while (dev_id) |
43801f6e NW |
889 | destroy_device(&zram_devices[--dev_id]); |
890 | kfree(zram_devices); | |
de1a21a0 | 891 | unregister: |
f1e3cfff | 892 | unregister_blkdev(zram_major, "zram"); |
de1a21a0 | 893 | out: |
306b0c95 NG |
894 | return ret; |
895 | } | |
896 | ||
f1e3cfff | 897 | static void __exit zram_exit(void) |
306b0c95 NG |
898 | { |
899 | int i; | |
f1e3cfff | 900 | struct zram *zram; |
306b0c95 | 901 | |
5fa5a901 | 902 | for (i = 0; i < num_devices; i++) { |
43801f6e | 903 | zram = &zram_devices[i]; |
306b0c95 | 904 | |
6030ea9b | 905 | get_disk(zram->disk); |
f1e3cfff | 906 | destroy_device(zram); |
0231c403 | 907 | zram_reset_device(zram); |
6030ea9b | 908 | put_disk(zram->disk); |
306b0c95 NG |
909 | } |
910 | ||
f1e3cfff | 911 | unregister_blkdev(zram_major, "zram"); |
306b0c95 | 912 | |
43801f6e | 913 | kfree(zram_devices); |
306b0c95 NG |
914 | pr_debug("Cleanup done!\n"); |
915 | } | |
916 | ||
f1e3cfff NG |
917 | module_init(zram_init); |
918 | module_exit(zram_exit); | |
306b0c95 | 919 | |
9b3bb7ab SS |
920 | module_param(num_devices, uint, 0); |
921 | MODULE_PARM_DESC(num_devices, "Number of zram devices"); | |
922 | ||
306b0c95 NG |
923 | MODULE_LICENSE("Dual BSD/GPL"); |
924 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 925 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |