]>
Commit | Line | Data |
---|---|---|
306b0c95 | 1 | /* |
f1e3cfff | 2 | * Compressed RAM block device |
306b0c95 | 3 | * |
1130ebba | 4 | * Copyright (C) 2008, 2009, 2010 Nitin Gupta |
7bfb3de8 | 5 | * 2012, 2013 Minchan Kim |
306b0c95 NG |
6 | * |
7 | * This code is released using a dual license strategy: BSD/GPL | |
8 | * You can choose the licence that better fits your requirements. | |
9 | * | |
10 | * Released under the terms of 3-clause BSD License | |
11 | * Released under the terms of GNU General Public License Version 2.0 | |
12 | * | |
306b0c95 NG |
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 | ||
deb0bdeb | 107 | return sprintf(buf, "%u\n", atomic_read(&zram->stats.pages_zero)); |
9b3bb7ab SS |
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", | |
deb0bdeb | 116 | (u64)(atomic_read(&zram->stats.pages_stored)) << PAGE_SHIFT); |
9b3bb7ab SS |
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 | ||
92967471 | 143 | /* flag operations needs meta->tb_lock */ |
8b3cc3ed | 144 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 145 | enum zram_pageflags flag) |
306b0c95 | 146 | { |
8b3cc3ed | 147 | return meta->table[index].flags & BIT(flag); |
306b0c95 NG |
148 | } |
149 | ||
8b3cc3ed | 150 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 151 | enum zram_pageflags flag) |
306b0c95 | 152 | { |
8b3cc3ed | 153 | meta->table[index].flags |= BIT(flag); |
306b0c95 NG |
154 | } |
155 | ||
8b3cc3ed | 156 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 157 | enum zram_pageflags flag) |
306b0c95 | 158 | { |
8b3cc3ed | 159 | meta->table[index].flags &= ~BIT(flag); |
306b0c95 NG |
160 | } |
161 | ||
9b3bb7ab SS |
162 | static inline int is_partial_io(struct bio_vec *bvec) |
163 | { | |
164 | return bvec->bv_len != PAGE_SIZE; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Check if request is within bounds and aligned on zram logical blocks. | |
169 | */ | |
170 | static inline int valid_io_request(struct zram *zram, struct bio *bio) | |
171 | { | |
172 | u64 start, end, bound; | |
a539c72a | 173 | |
9b3bb7ab | 174 | /* unaligned request */ |
4f024f37 KO |
175 | if (unlikely(bio->bi_iter.bi_sector & |
176 | (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1))) | |
9b3bb7ab | 177 | return 0; |
4f024f37 | 178 | if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) |
9b3bb7ab SS |
179 | return 0; |
180 | ||
4f024f37 KO |
181 | start = bio->bi_iter.bi_sector; |
182 | end = start + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
9b3bb7ab SS |
183 | bound = zram->disksize >> SECTOR_SHIFT; |
184 | /* out of range range */ | |
75c7caf5 | 185 | if (unlikely(start >= bound || end > bound || start > end)) |
9b3bb7ab SS |
186 | return 0; |
187 | ||
188 | /* I/O request is valid */ | |
189 | return 1; | |
190 | } | |
191 | ||
192 | static void zram_meta_free(struct zram_meta *meta) | |
193 | { | |
194 | zs_destroy_pool(meta->mem_pool); | |
195 | kfree(meta->compress_workmem); | |
196 | free_pages((unsigned long)meta->compress_buffer, 1); | |
197 | vfree(meta->table); | |
198 | kfree(meta); | |
199 | } | |
200 | ||
201 | static struct zram_meta *zram_meta_alloc(u64 disksize) | |
202 | { | |
203 | size_t num_pages; | |
204 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); | |
205 | if (!meta) | |
206 | goto out; | |
207 | ||
208 | meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); | |
209 | if (!meta->compress_workmem) | |
210 | goto free_meta; | |
211 | ||
212 | meta->compress_buffer = | |
213 | (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1); | |
214 | if (!meta->compress_buffer) { | |
215 | pr_err("Error allocating compressor buffer space\n"); | |
216 | goto free_workmem; | |
217 | } | |
218 | ||
219 | num_pages = disksize >> PAGE_SHIFT; | |
220 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
221 | if (!meta->table) { | |
222 | pr_err("Error allocating zram address table\n"); | |
223 | goto free_buffer; | |
224 | } | |
225 | ||
226 | meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM); | |
227 | if (!meta->mem_pool) { | |
228 | pr_err("Error creating memory pool\n"); | |
229 | goto free_table; | |
230 | } | |
231 | ||
92967471 | 232 | rwlock_init(&meta->tb_lock); |
e46e3315 | 233 | mutex_init(&meta->buffer_lock); |
9b3bb7ab SS |
234 | return meta; |
235 | ||
236 | free_table: | |
237 | vfree(meta->table); | |
238 | free_buffer: | |
239 | free_pages((unsigned long)meta->compress_buffer, 1); | |
240 | free_workmem: | |
241 | kfree(meta->compress_workmem); | |
242 | free_meta: | |
243 | kfree(meta); | |
244 | meta = NULL; | |
245 | out: | |
246 | return meta; | |
247 | } | |
248 | ||
249 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
250 | { | |
251 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
252 | (*index)++; | |
253 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
254 | } | |
255 | ||
306b0c95 NG |
256 | static int page_zero_filled(void *ptr) |
257 | { | |
258 | unsigned int pos; | |
259 | unsigned long *page; | |
260 | ||
261 | page = (unsigned long *)ptr; | |
262 | ||
263 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
264 | if (page[pos]) | |
265 | return 0; | |
266 | } | |
267 | ||
268 | return 1; | |
269 | } | |
270 | ||
9b3bb7ab SS |
271 | static void handle_zero_page(struct bio_vec *bvec) |
272 | { | |
273 | struct page *page = bvec->bv_page; | |
274 | void *user_mem; | |
275 | ||
276 | user_mem = kmap_atomic(page); | |
277 | if (is_partial_io(bvec)) | |
278 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); | |
279 | else | |
280 | clear_page(user_mem); | |
281 | kunmap_atomic(user_mem); | |
282 | ||
283 | flush_dcache_page(page); | |
284 | } | |
285 | ||
92967471 | 286 | /* NOTE: caller should hold meta->tb_lock with write-side */ |
f1e3cfff | 287 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 288 | { |
8b3cc3ed MK |
289 | struct zram_meta *meta = zram->meta; |
290 | unsigned long handle = meta->table[index].handle; | |
291 | u16 size = meta->table[index].size; | |
306b0c95 | 292 | |
fd1a30de | 293 | if (unlikely(!handle)) { |
2e882281 NG |
294 | /* |
295 | * No memory is allocated for zero filled pages. | |
296 | * Simply clear zero page flag. | |
297 | */ | |
8b3cc3ed MK |
298 | if (zram_test_flag(meta, index, ZRAM_ZERO)) { |
299 | zram_clear_flag(meta, index, ZRAM_ZERO); | |
deb0bdeb | 300 | atomic_dec(&zram->stats.pages_zero); |
306b0c95 NG |
301 | } |
302 | return; | |
303 | } | |
304 | ||
130f315a | 305 | if (unlikely(size > max_zpage_size)) |
deb0bdeb | 306 | atomic_dec(&zram->stats.bad_compress); |
306b0c95 | 307 | |
8b3cc3ed | 308 | zs_free(meta->mem_pool, handle); |
306b0c95 | 309 | |
130f315a | 310 | if (size <= PAGE_SIZE / 2) |
deb0bdeb | 311 | atomic_dec(&zram->stats.good_compress); |
306b0c95 | 312 | |
da5cc7d3 | 313 | atomic64_sub(meta->table[index].size, &zram->stats.compr_size); |
deb0bdeb | 314 | atomic_dec(&zram->stats.pages_stored); |
306b0c95 | 315 | |
8b3cc3ed MK |
316 | meta->table[index].handle = 0; |
317 | meta->table[index].size = 0; | |
306b0c95 NG |
318 | } |
319 | ||
37b51fdd | 320 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 321 | { |
37b51fdd SS |
322 | int ret = LZO_E_OK; |
323 | size_t clen = PAGE_SIZE; | |
324 | unsigned char *cmem; | |
8b3cc3ed | 325 | struct zram_meta *meta = zram->meta; |
92967471 MK |
326 | unsigned long handle; |
327 | u16 size; | |
328 | ||
329 | read_lock(&meta->tb_lock); | |
330 | handle = meta->table[index].handle; | |
331 | size = meta->table[index].size; | |
306b0c95 | 332 | |
8b3cc3ed | 333 | if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) { |
92967471 | 334 | read_unlock(&meta->tb_lock); |
42e99bd9 | 335 | clear_page(mem); |
8c921b2b JM |
336 | return 0; |
337 | } | |
306b0c95 | 338 | |
8b3cc3ed | 339 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
92967471 | 340 | if (size == PAGE_SIZE) |
42e99bd9 | 341 | copy_page(mem, cmem); |
37b51fdd | 342 | else |
92967471 | 343 | ret = lzo1x_decompress_safe(cmem, size, mem, &clen); |
8b3cc3ed | 344 | zs_unmap_object(meta->mem_pool, handle); |
92967471 | 345 | read_unlock(&meta->tb_lock); |
a1dd52af | 346 | |
8c921b2b JM |
347 | /* Should NEVER happen. Return bio error if it does. */ |
348 | if (unlikely(ret != LZO_E_OK)) { | |
349 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); | |
da5cc7d3 | 350 | atomic64_inc(&zram->stats.failed_reads); |
8c921b2b | 351 | return ret; |
a1dd52af | 352 | } |
306b0c95 | 353 | |
8c921b2b | 354 | return 0; |
306b0c95 NG |
355 | } |
356 | ||
37b51fdd SS |
357 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
358 | u32 index, int offset, struct bio *bio) | |
924bd88d JM |
359 | { |
360 | int ret; | |
37b51fdd SS |
361 | struct page *page; |
362 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 363 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
364 | page = bvec->bv_page; |
365 | ||
92967471 | 366 | read_lock(&meta->tb_lock); |
8b3cc3ed MK |
367 | if (unlikely(!meta->table[index].handle) || |
368 | zram_test_flag(meta, index, ZRAM_ZERO)) { | |
92967471 | 369 | read_unlock(&meta->tb_lock); |
37b51fdd | 370 | handle_zero_page(bvec); |
924bd88d JM |
371 | return 0; |
372 | } | |
92967471 | 373 | read_unlock(&meta->tb_lock); |
924bd88d | 374 | |
37b51fdd SS |
375 | if (is_partial_io(bvec)) |
376 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
377 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
378 | ||
379 | user_mem = kmap_atomic(page); | |
380 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
381 | uncmem = user_mem; |
382 | ||
383 | if (!uncmem) { | |
384 | pr_info("Unable to allocate temp memory\n"); | |
385 | ret = -ENOMEM; | |
386 | goto out_cleanup; | |
387 | } | |
924bd88d | 388 | |
37b51fdd | 389 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 390 | /* Should NEVER happen. Return bio error if it does. */ |
25eeb667 | 391 | if (unlikely(ret != LZO_E_OK)) |
37b51fdd | 392 | goto out_cleanup; |
924bd88d | 393 | |
37b51fdd SS |
394 | if (is_partial_io(bvec)) |
395 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
396 | bvec->bv_len); | |
397 | ||
398 | flush_dcache_page(page); | |
399 | ret = 0; | |
400 | out_cleanup: | |
401 | kunmap_atomic(user_mem); | |
402 | if (is_partial_io(bvec)) | |
403 | kfree(uncmem); | |
404 | return ret; | |
924bd88d JM |
405 | } |
406 | ||
407 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
408 | int offset) | |
306b0c95 | 409 | { |
397c6066 | 410 | int ret = 0; |
8c921b2b | 411 | size_t clen; |
c2344348 | 412 | unsigned long handle; |
130f315a | 413 | struct page *page; |
924bd88d | 414 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 415 | struct zram_meta *meta = zram->meta; |
e46e3315 | 416 | bool locked = false; |
306b0c95 | 417 | |
8c921b2b | 418 | page = bvec->bv_page; |
8b3cc3ed | 419 | src = meta->compress_buffer; |
306b0c95 | 420 | |
924bd88d JM |
421 | if (is_partial_io(bvec)) { |
422 | /* | |
423 | * This is a partial IO. We need to read the full page | |
424 | * before to write the changes. | |
425 | */ | |
7e5a5104 | 426 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 427 | if (!uncmem) { |
924bd88d JM |
428 | ret = -ENOMEM; |
429 | goto out; | |
430 | } | |
37b51fdd | 431 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 432 | if (ret) |
924bd88d | 433 | goto out; |
924bd88d JM |
434 | } |
435 | ||
e46e3315 MK |
436 | mutex_lock(&meta->buffer_lock); |
437 | locked = true; | |
ba82fe2e | 438 | user_mem = kmap_atomic(page); |
924bd88d | 439 | |
397c6066 | 440 | if (is_partial_io(bvec)) { |
924bd88d JM |
441 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
442 | bvec->bv_len); | |
397c6066 NG |
443 | kunmap_atomic(user_mem); |
444 | user_mem = NULL; | |
445 | } else { | |
924bd88d | 446 | uncmem = user_mem; |
397c6066 | 447 | } |
924bd88d JM |
448 | |
449 | if (page_zero_filled(uncmem)) { | |
ba82fe2e | 450 | kunmap_atomic(user_mem); |
f40ac2ae | 451 | /* Free memory associated with this sector now. */ |
92967471 | 452 | write_lock(&zram->meta->tb_lock); |
f40ac2ae | 453 | zram_free_page(zram, index); |
92967471 MK |
454 | zram_set_flag(meta, index, ZRAM_ZERO); |
455 | write_unlock(&zram->meta->tb_lock); | |
f40ac2ae | 456 | |
deb0bdeb | 457 | atomic_inc(&zram->stats.pages_zero); |
924bd88d JM |
458 | ret = 0; |
459 | goto out; | |
8c921b2b | 460 | } |
306b0c95 | 461 | |
924bd88d | 462 | ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen, |
8b3cc3ed | 463 | meta->compress_workmem); |
397c6066 NG |
464 | if (!is_partial_io(bvec)) { |
465 | kunmap_atomic(user_mem); | |
466 | user_mem = NULL; | |
467 | uncmem = NULL; | |
468 | } | |
306b0c95 | 469 | |
8c921b2b | 470 | if (unlikely(ret != LZO_E_OK)) { |
8c921b2b | 471 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 472 | goto out; |
8c921b2b | 473 | } |
306b0c95 | 474 | |
c8f2f0db | 475 | if (unlikely(clen > max_zpage_size)) { |
deb0bdeb | 476 | atomic_inc(&zram->stats.bad_compress); |
c8f2f0db | 477 | clen = PAGE_SIZE; |
397c6066 NG |
478 | src = NULL; |
479 | if (is_partial_io(bvec)) | |
480 | src = uncmem; | |
c8f2f0db | 481 | } |
a1dd52af | 482 | |
8b3cc3ed | 483 | handle = zs_malloc(meta->mem_pool, clen); |
fd1a30de | 484 | if (!handle) { |
596b3dd4 MR |
485 | pr_info("Error allocating memory for compressed page: %u, size=%zu\n", |
486 | index, clen); | |
924bd88d JM |
487 | ret = -ENOMEM; |
488 | goto out; | |
8c921b2b | 489 | } |
8b3cc3ed | 490 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 491 | |
42e99bd9 | 492 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 493 | src = kmap_atomic(page); |
42e99bd9 | 494 | copy_page(cmem, src); |
397c6066 | 495 | kunmap_atomic(src); |
42e99bd9 JL |
496 | } else { |
497 | memcpy(cmem, src, clen); | |
498 | } | |
306b0c95 | 499 | |
8b3cc3ed | 500 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 501 | |
f40ac2ae SS |
502 | /* |
503 | * Free memory associated with this sector | |
504 | * before overwriting unused sectors. | |
505 | */ | |
92967471 | 506 | write_lock(&zram->meta->tb_lock); |
f40ac2ae SS |
507 | zram_free_page(zram, index); |
508 | ||
8b3cc3ed MK |
509 | meta->table[index].handle = handle; |
510 | meta->table[index].size = clen; | |
92967471 | 511 | write_unlock(&zram->meta->tb_lock); |
306b0c95 | 512 | |
8c921b2b | 513 | /* Update stats */ |
da5cc7d3 | 514 | atomic64_add(clen, &zram->stats.compr_size); |
deb0bdeb | 515 | atomic_inc(&zram->stats.pages_stored); |
8c921b2b | 516 | if (clen <= PAGE_SIZE / 2) |
deb0bdeb | 517 | atomic_inc(&zram->stats.good_compress); |
306b0c95 | 518 | |
924bd88d | 519 | out: |
e46e3315 MK |
520 | if (locked) |
521 | mutex_unlock(&meta->buffer_lock); | |
397c6066 NG |
522 | if (is_partial_io(bvec)) |
523 | kfree(uncmem); | |
524 | ||
924bd88d | 525 | if (ret) |
da5cc7d3 | 526 | atomic64_inc(&zram->stats.failed_writes); |
924bd88d | 527 | return ret; |
8c921b2b JM |
528 | } |
529 | ||
530 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
924bd88d | 531 | int offset, struct bio *bio, int rw) |
8c921b2b | 532 | { |
c5bde238 | 533 | int ret; |
8c921b2b | 534 | |
e46e3315 | 535 | if (rw == READ) |
c5bde238 | 536 | ret = zram_bvec_read(zram, bvec, index, offset, bio); |
e46e3315 | 537 | else |
c5bde238 | 538 | ret = zram_bvec_write(zram, bvec, index, offset); |
c5bde238 JM |
539 | |
540 | return ret; | |
924bd88d JM |
541 | } |
542 | ||
2b86ab9c | 543 | static void zram_reset_device(struct zram *zram, bool reset_capacity) |
924bd88d | 544 | { |
9b3bb7ab SS |
545 | size_t index; |
546 | struct zram_meta *meta; | |
547 | ||
644d4787 SS |
548 | down_write(&zram->init_lock); |
549 | if (!zram->init_done) { | |
550 | up_write(&zram->init_lock); | |
9b3bb7ab | 551 | return; |
644d4787 | 552 | } |
9b3bb7ab SS |
553 | |
554 | meta = zram->meta; | |
555 | zram->init_done = 0; | |
556 | ||
557 | /* Free all pages that are still in this zram device */ | |
558 | for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) { | |
559 | unsigned long handle = meta->table[index].handle; | |
560 | if (!handle) | |
561 | continue; | |
562 | ||
563 | zs_free(meta->mem_pool, handle); | |
564 | } | |
565 | ||
566 | zram_meta_free(zram->meta); | |
567 | zram->meta = NULL; | |
568 | /* Reset stats */ | |
569 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
570 | ||
571 | zram->disksize = 0; | |
2b86ab9c MK |
572 | if (reset_capacity) |
573 | set_capacity(zram->disk, 0); | |
644d4787 | 574 | up_write(&zram->init_lock); |
9b3bb7ab SS |
575 | } |
576 | ||
577 | static void zram_init_device(struct zram *zram, struct zram_meta *meta) | |
578 | { | |
579 | if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) { | |
580 | pr_info( | |
581 | "There is little point creating a zram of greater than " | |
582 | "twice the size of memory since we expect a 2:1 compression " | |
583 | "ratio. Note that zram uses about 0.1%% of the size of " | |
584 | "the disk when not in use so a huge zram is " | |
585 | "wasteful.\n" | |
586 | "\tMemory Size: %lu kB\n" | |
587 | "\tSize you selected: %llu kB\n" | |
588 | "Continuing anyway ...\n", | |
589 | (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10 | |
590 | ); | |
591 | } | |
592 | ||
593 | /* zram devices sort of resembles non-rotational disks */ | |
594 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
595 | ||
596 | zram->meta = meta; | |
597 | zram->init_done = 1; | |
598 | ||
599 | pr_debug("Initialization done!\n"); | |
600 | } | |
601 | ||
602 | static ssize_t disksize_store(struct device *dev, | |
603 | struct device_attribute *attr, const char *buf, size_t len) | |
604 | { | |
605 | u64 disksize; | |
606 | struct zram_meta *meta; | |
607 | struct zram *zram = dev_to_zram(dev); | |
608 | ||
609 | disksize = memparse(buf, NULL); | |
610 | if (!disksize) | |
611 | return -EINVAL; | |
612 | ||
613 | disksize = PAGE_ALIGN(disksize); | |
614 | meta = zram_meta_alloc(disksize); | |
db5d711e MK |
615 | if (!meta) |
616 | return -ENOMEM; | |
9b3bb7ab SS |
617 | down_write(&zram->init_lock); |
618 | if (zram->init_done) { | |
619 | up_write(&zram->init_lock); | |
620 | zram_meta_free(meta); | |
621 | pr_info("Cannot change disksize for initialized device\n"); | |
622 | return -EBUSY; | |
623 | } | |
624 | ||
625 | zram->disksize = disksize; | |
626 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
627 | zram_init_device(zram, meta); | |
628 | up_write(&zram->init_lock); | |
629 | ||
630 | return len; | |
631 | } | |
632 | ||
633 | static ssize_t reset_store(struct device *dev, | |
634 | struct device_attribute *attr, const char *buf, size_t len) | |
635 | { | |
636 | int ret; | |
637 | unsigned short do_reset; | |
638 | struct zram *zram; | |
639 | struct block_device *bdev; | |
640 | ||
641 | zram = dev_to_zram(dev); | |
642 | bdev = bdget_disk(zram->disk, 0); | |
643 | ||
46a51c80 RK |
644 | if (!bdev) |
645 | return -ENOMEM; | |
646 | ||
9b3bb7ab | 647 | /* Do not reset an active device! */ |
1b672224 RK |
648 | if (bdev->bd_holders) { |
649 | ret = -EBUSY; | |
650 | goto out; | |
651 | } | |
9b3bb7ab SS |
652 | |
653 | ret = kstrtou16(buf, 10, &do_reset); | |
654 | if (ret) | |
1b672224 | 655 | goto out; |
9b3bb7ab | 656 | |
1b672224 RK |
657 | if (!do_reset) { |
658 | ret = -EINVAL; | |
659 | goto out; | |
660 | } | |
9b3bb7ab SS |
661 | |
662 | /* Make sure all pending I/O is finished */ | |
46a51c80 | 663 | fsync_bdev(bdev); |
1b672224 | 664 | bdput(bdev); |
9b3bb7ab | 665 | |
2b86ab9c | 666 | zram_reset_device(zram, true); |
9b3bb7ab | 667 | return len; |
1b672224 RK |
668 | |
669 | out: | |
670 | bdput(bdev); | |
671 | return ret; | |
8c921b2b JM |
672 | } |
673 | ||
674 | static void __zram_make_request(struct zram *zram, struct bio *bio, int rw) | |
675 | { | |
7988613b | 676 | int offset; |
8c921b2b | 677 | u32 index; |
7988613b KO |
678 | struct bio_vec bvec; |
679 | struct bvec_iter iter; | |
8c921b2b JM |
680 | |
681 | switch (rw) { | |
682 | case READ: | |
da5cc7d3 | 683 | atomic64_inc(&zram->stats.num_reads); |
8c921b2b JM |
684 | break; |
685 | case WRITE: | |
da5cc7d3 | 686 | atomic64_inc(&zram->stats.num_writes); |
8c921b2b JM |
687 | break; |
688 | } | |
689 | ||
4f024f37 KO |
690 | index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT; |
691 | offset = (bio->bi_iter.bi_sector & | |
692 | (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; | |
8c921b2b | 693 | |
7988613b | 694 | bio_for_each_segment(bvec, bio, iter) { |
924bd88d JM |
695 | int max_transfer_size = PAGE_SIZE - offset; |
696 | ||
7988613b | 697 | if (bvec.bv_len > max_transfer_size) { |
924bd88d JM |
698 | /* |
699 | * zram_bvec_rw() can only make operation on a single | |
700 | * zram page. Split the bio vector. | |
701 | */ | |
702 | struct bio_vec bv; | |
703 | ||
7988613b | 704 | bv.bv_page = bvec.bv_page; |
924bd88d | 705 | bv.bv_len = max_transfer_size; |
7988613b | 706 | bv.bv_offset = bvec.bv_offset; |
924bd88d JM |
707 | |
708 | if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0) | |
709 | goto out; | |
710 | ||
7988613b | 711 | bv.bv_len = bvec.bv_len - max_transfer_size; |
924bd88d JM |
712 | bv.bv_offset += max_transfer_size; |
713 | if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0) | |
714 | goto out; | |
715 | } else | |
7988613b | 716 | if (zram_bvec_rw(zram, &bvec, index, offset, bio, rw) |
924bd88d JM |
717 | < 0) |
718 | goto out; | |
719 | ||
7988613b | 720 | update_position(&index, &offset, &bvec); |
a1dd52af | 721 | } |
306b0c95 NG |
722 | |
723 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
724 | bio_endio(bio, 0); | |
7d7854b4 | 725 | return; |
306b0c95 NG |
726 | |
727 | out: | |
306b0c95 | 728 | bio_io_error(bio); |
306b0c95 NG |
729 | } |
730 | ||
306b0c95 | 731 | /* |
f1e3cfff | 732 | * Handler function for all zram I/O requests. |
306b0c95 | 733 | */ |
5a7bbad2 | 734 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 735 | { |
f1e3cfff | 736 | struct zram *zram = queue->queuedata; |
306b0c95 | 737 | |
0900beae JM |
738 | down_read(&zram->init_lock); |
739 | if (unlikely(!zram->init_done)) | |
3de738cd | 740 | goto error; |
0900beae | 741 | |
f1e3cfff | 742 | if (!valid_io_request(zram, bio)) { |
da5cc7d3 | 743 | atomic64_inc(&zram->stats.invalid_io); |
3de738cd | 744 | goto error; |
6642a67c JM |
745 | } |
746 | ||
8c921b2b | 747 | __zram_make_request(zram, bio, bio_data_dir(bio)); |
0900beae | 748 | up_read(&zram->init_lock); |
306b0c95 | 749 | |
b4fdcb02 | 750 | return; |
0900beae | 751 | |
0900beae | 752 | error: |
3de738cd | 753 | up_read(&zram->init_lock); |
0900beae | 754 | bio_io_error(bio); |
306b0c95 NG |
755 | } |
756 | ||
2ccbec05 NG |
757 | static void zram_slot_free_notify(struct block_device *bdev, |
758 | unsigned long index) | |
107c161b | 759 | { |
f1e3cfff | 760 | struct zram *zram; |
f614a9f4 | 761 | struct zram_meta *meta; |
107c161b | 762 | |
f1e3cfff | 763 | zram = bdev->bd_disk->private_data; |
f614a9f4 | 764 | meta = zram->meta; |
a0c516cb | 765 | |
f614a9f4 MK |
766 | write_lock(&meta->tb_lock); |
767 | zram_free_page(zram, index); | |
768 | write_unlock(&meta->tb_lock); | |
769 | atomic64_inc(&zram->stats.notify_free); | |
107c161b NG |
770 | } |
771 | ||
f1e3cfff | 772 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 773 | .swap_slot_free_notify = zram_slot_free_notify, |
107c161b | 774 | .owner = THIS_MODULE |
306b0c95 NG |
775 | }; |
776 | ||
9b3bb7ab SS |
777 | static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR, |
778 | disksize_show, disksize_store); | |
779 | static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL); | |
780 | static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store); | |
781 | static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL); | |
782 | static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL); | |
783 | static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL); | |
784 | static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL); | |
785 | static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL); | |
786 | static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL); | |
787 | static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL); | |
788 | static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL); | |
789 | ||
790 | static struct attribute *zram_disk_attrs[] = { | |
791 | &dev_attr_disksize.attr, | |
792 | &dev_attr_initstate.attr, | |
793 | &dev_attr_reset.attr, | |
794 | &dev_attr_num_reads.attr, | |
795 | &dev_attr_num_writes.attr, | |
796 | &dev_attr_invalid_io.attr, | |
797 | &dev_attr_notify_free.attr, | |
798 | &dev_attr_zero_pages.attr, | |
799 | &dev_attr_orig_data_size.attr, | |
800 | &dev_attr_compr_data_size.attr, | |
801 | &dev_attr_mem_used_total.attr, | |
802 | NULL, | |
803 | }; | |
804 | ||
805 | static struct attribute_group zram_disk_attr_group = { | |
806 | .attrs = zram_disk_attrs, | |
807 | }; | |
808 | ||
f1e3cfff | 809 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 810 | { |
39a9b8ac | 811 | int ret = -ENOMEM; |
de1a21a0 | 812 | |
0900beae | 813 | init_rwsem(&zram->init_lock); |
306b0c95 | 814 | |
f1e3cfff NG |
815 | zram->queue = blk_alloc_queue(GFP_KERNEL); |
816 | if (!zram->queue) { | |
306b0c95 NG |
817 | pr_err("Error allocating disk queue for device %d\n", |
818 | device_id); | |
de1a21a0 | 819 | goto out; |
306b0c95 NG |
820 | } |
821 | ||
f1e3cfff NG |
822 | blk_queue_make_request(zram->queue, zram_make_request); |
823 | zram->queue->queuedata = zram; | |
306b0c95 NG |
824 | |
825 | /* gendisk structure */ | |
f1e3cfff NG |
826 | zram->disk = alloc_disk(1); |
827 | if (!zram->disk) { | |
94b8435f | 828 | pr_warn("Error allocating disk structure for device %d\n", |
306b0c95 | 829 | device_id); |
39a9b8ac | 830 | goto out_free_queue; |
306b0c95 NG |
831 | } |
832 | ||
f1e3cfff NG |
833 | zram->disk->major = zram_major; |
834 | zram->disk->first_minor = device_id; | |
835 | zram->disk->fops = &zram_devops; | |
836 | zram->disk->queue = zram->queue; | |
837 | zram->disk->private_data = zram; | |
838 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 839 | |
33863c21 | 840 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 841 | set_capacity(zram->disk, 0); |
5d83d5a0 | 842 | |
a1dd52af NG |
843 | /* |
844 | * To ensure that we always get PAGE_SIZE aligned | |
845 | * and n*PAGE_SIZED sized I/O requests. | |
846 | */ | |
f1e3cfff | 847 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
848 | blk_queue_logical_block_size(zram->disk->queue, |
849 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
850 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
851 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
5d83d5a0 | 852 | |
f1e3cfff | 853 | add_disk(zram->disk); |
306b0c95 | 854 | |
33863c21 NG |
855 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
856 | &zram_disk_attr_group); | |
857 | if (ret < 0) { | |
94b8435f | 858 | pr_warn("Error creating sysfs group"); |
39a9b8ac | 859 | goto out_free_disk; |
33863c21 | 860 | } |
33863c21 | 861 | |
f1e3cfff | 862 | zram->init_done = 0; |
39a9b8ac | 863 | return 0; |
de1a21a0 | 864 | |
39a9b8ac JL |
865 | out_free_disk: |
866 | del_gendisk(zram->disk); | |
867 | put_disk(zram->disk); | |
868 | out_free_queue: | |
869 | blk_cleanup_queue(zram->queue); | |
de1a21a0 NG |
870 | out: |
871 | return ret; | |
306b0c95 NG |
872 | } |
873 | ||
f1e3cfff | 874 | static void destroy_device(struct zram *zram) |
306b0c95 | 875 | { |
33863c21 NG |
876 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, |
877 | &zram_disk_attr_group); | |
33863c21 | 878 | |
59d3fe54 RK |
879 | del_gendisk(zram->disk); |
880 | put_disk(zram->disk); | |
306b0c95 | 881 | |
59d3fe54 | 882 | blk_cleanup_queue(zram->queue); |
306b0c95 NG |
883 | } |
884 | ||
f1e3cfff | 885 | static int __init zram_init(void) |
306b0c95 | 886 | { |
de1a21a0 | 887 | int ret, dev_id; |
306b0c95 | 888 | |
5fa5a901 | 889 | if (num_devices > max_num_devices) { |
94b8435f | 890 | pr_warn("Invalid value for num_devices: %u\n", |
5fa5a901 | 891 | num_devices); |
de1a21a0 NG |
892 | ret = -EINVAL; |
893 | goto out; | |
306b0c95 NG |
894 | } |
895 | ||
f1e3cfff NG |
896 | zram_major = register_blkdev(0, "zram"); |
897 | if (zram_major <= 0) { | |
94b8435f | 898 | pr_warn("Unable to get major number\n"); |
de1a21a0 NG |
899 | ret = -EBUSY; |
900 | goto out; | |
306b0c95 NG |
901 | } |
902 | ||
306b0c95 | 903 | /* Allocate the device array and initialize each one */ |
5fa5a901 | 904 | zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL); |
43801f6e | 905 | if (!zram_devices) { |
de1a21a0 NG |
906 | ret = -ENOMEM; |
907 | goto unregister; | |
908 | } | |
306b0c95 | 909 | |
5fa5a901 | 910 | for (dev_id = 0; dev_id < num_devices; dev_id++) { |
43801f6e | 911 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 912 | if (ret) |
3bf040c7 | 913 | goto free_devices; |
de1a21a0 NG |
914 | } |
915 | ||
ca3d70bd DB |
916 | pr_info("Created %u device(s) ...\n", num_devices); |
917 | ||
306b0c95 | 918 | return 0; |
de1a21a0 | 919 | |
3bf040c7 | 920 | free_devices: |
de1a21a0 | 921 | while (dev_id) |
43801f6e NW |
922 | destroy_device(&zram_devices[--dev_id]); |
923 | kfree(zram_devices); | |
de1a21a0 | 924 | unregister: |
f1e3cfff | 925 | unregister_blkdev(zram_major, "zram"); |
de1a21a0 | 926 | out: |
306b0c95 NG |
927 | return ret; |
928 | } | |
929 | ||
f1e3cfff | 930 | static void __exit zram_exit(void) |
306b0c95 NG |
931 | { |
932 | int i; | |
f1e3cfff | 933 | struct zram *zram; |
306b0c95 | 934 | |
5fa5a901 | 935 | for (i = 0; i < num_devices; i++) { |
43801f6e | 936 | zram = &zram_devices[i]; |
306b0c95 | 937 | |
f1e3cfff | 938 | destroy_device(zram); |
2b86ab9c MK |
939 | /* |
940 | * Shouldn't access zram->disk after destroy_device | |
941 | * because destroy_device already released zram->disk. | |
942 | */ | |
943 | zram_reset_device(zram, false); | |
306b0c95 NG |
944 | } |
945 | ||
f1e3cfff | 946 | unregister_blkdev(zram_major, "zram"); |
306b0c95 | 947 | |
43801f6e | 948 | kfree(zram_devices); |
306b0c95 NG |
949 | pr_debug("Cleanup done!\n"); |
950 | } | |
951 | ||
f1e3cfff NG |
952 | module_init(zram_init); |
953 | module_exit(zram_exit); | |
306b0c95 | 954 | |
9b3bb7ab SS |
955 | module_param(num_devices, uint, 0); |
956 | MODULE_PARM_DESC(num_devices, "Number of zram devices"); | |
957 | ||
306b0c95 NG |
958 | MODULE_LICENSE("Dual BSD/GPL"); |
959 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 960 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |