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/string.h> |
306b0c95 | 33 | #include <linux/vmalloc.h> |
fcfa8d95 | 34 | #include <linux/err.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; |
b7ca232e | 41 | static const char *default_compressor = "lzo"; |
306b0c95 | 42 | |
306b0c95 | 43 | /* Module params (documentation at end) */ |
ca3d70bd | 44 | static unsigned int num_devices = 1; |
33863c21 | 45 | |
a68eb3b6 | 46 | #define ZRAM_ATTR_RO(name) \ |
083914ea | 47 | static ssize_t name##_show(struct device *d, \ |
a68eb3b6 SS |
48 | struct device_attribute *attr, char *b) \ |
49 | { \ | |
50 | struct zram *zram = dev_to_zram(d); \ | |
56b4e8cb | 51 | return scnprintf(b, PAGE_SIZE, "%llu\n", \ |
a68eb3b6 SS |
52 | (u64)atomic64_read(&zram->stats.name)); \ |
53 | } \ | |
083914ea | 54 | static DEVICE_ATTR_RO(name); |
a68eb3b6 | 55 | |
08eee69f | 56 | static inline bool init_done(struct zram *zram) |
be2d1d56 | 57 | { |
08eee69f | 58 | return zram->disksize; |
be2d1d56 SS |
59 | } |
60 | ||
9b3bb7ab SS |
61 | static inline struct zram *dev_to_zram(struct device *dev) |
62 | { | |
63 | return (struct zram *)dev_to_disk(dev)->private_data; | |
64 | } | |
65 | ||
66 | static ssize_t disksize_show(struct device *dev, | |
67 | struct device_attribute *attr, char *buf) | |
68 | { | |
69 | struct zram *zram = dev_to_zram(dev); | |
70 | ||
56b4e8cb | 71 | return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize); |
9b3bb7ab SS |
72 | } |
73 | ||
74 | static ssize_t initstate_show(struct device *dev, | |
75 | struct device_attribute *attr, char *buf) | |
76 | { | |
a68eb3b6 | 77 | u32 val; |
9b3bb7ab SS |
78 | struct zram *zram = dev_to_zram(dev); |
79 | ||
a68eb3b6 SS |
80 | down_read(&zram->init_lock); |
81 | val = init_done(zram); | |
82 | up_read(&zram->init_lock); | |
9b3bb7ab | 83 | |
56b4e8cb | 84 | return scnprintf(buf, PAGE_SIZE, "%u\n", val); |
9b3bb7ab SS |
85 | } |
86 | ||
87 | static ssize_t orig_data_size_show(struct device *dev, | |
88 | struct device_attribute *attr, char *buf) | |
89 | { | |
90 | struct zram *zram = dev_to_zram(dev); | |
91 | ||
56b4e8cb | 92 | return scnprintf(buf, PAGE_SIZE, "%llu\n", |
90a7806e | 93 | (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT); |
9b3bb7ab SS |
94 | } |
95 | ||
9b3bb7ab SS |
96 | static ssize_t mem_used_total_show(struct device *dev, |
97 | struct device_attribute *attr, char *buf) | |
98 | { | |
99 | u64 val = 0; | |
100 | struct zram *zram = dev_to_zram(dev); | |
9b3bb7ab SS |
101 | |
102 | down_read(&zram->init_lock); | |
5a99e95b WY |
103 | if (init_done(zram)) { |
104 | struct zram_meta *meta = zram->meta; | |
722cdc17 | 105 | val = zs_get_total_pages(meta->mem_pool); |
5a99e95b | 106 | } |
9b3bb7ab SS |
107 | up_read(&zram->init_lock); |
108 | ||
722cdc17 | 109 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); |
9b3bb7ab SS |
110 | } |
111 | ||
beca3ec7 SS |
112 | static ssize_t max_comp_streams_show(struct device *dev, |
113 | struct device_attribute *attr, char *buf) | |
114 | { | |
115 | int val; | |
116 | struct zram *zram = dev_to_zram(dev); | |
117 | ||
118 | down_read(&zram->init_lock); | |
119 | val = zram->max_comp_streams; | |
120 | up_read(&zram->init_lock); | |
121 | ||
56b4e8cb | 122 | return scnprintf(buf, PAGE_SIZE, "%d\n", val); |
beca3ec7 SS |
123 | } |
124 | ||
9ada9da9 MK |
125 | static ssize_t mem_limit_show(struct device *dev, |
126 | struct device_attribute *attr, char *buf) | |
127 | { | |
128 | u64 val; | |
129 | struct zram *zram = dev_to_zram(dev); | |
130 | ||
131 | down_read(&zram->init_lock); | |
132 | val = zram->limit_pages; | |
133 | up_read(&zram->init_lock); | |
134 | ||
135 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
136 | } | |
137 | ||
138 | static ssize_t mem_limit_store(struct device *dev, | |
139 | struct device_attribute *attr, const char *buf, size_t len) | |
140 | { | |
141 | u64 limit; | |
142 | char *tmp; | |
143 | struct zram *zram = dev_to_zram(dev); | |
144 | ||
145 | limit = memparse(buf, &tmp); | |
146 | if (buf == tmp) /* no chars parsed, invalid input */ | |
147 | return -EINVAL; | |
148 | ||
149 | down_write(&zram->init_lock); | |
150 | zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT; | |
151 | up_write(&zram->init_lock); | |
152 | ||
153 | return len; | |
154 | } | |
155 | ||
461a8eee MK |
156 | static ssize_t mem_used_max_show(struct device *dev, |
157 | struct device_attribute *attr, char *buf) | |
158 | { | |
159 | u64 val = 0; | |
160 | struct zram *zram = dev_to_zram(dev); | |
161 | ||
162 | down_read(&zram->init_lock); | |
163 | if (init_done(zram)) | |
164 | val = atomic_long_read(&zram->stats.max_used_pages); | |
165 | up_read(&zram->init_lock); | |
166 | ||
167 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
168 | } | |
169 | ||
170 | static ssize_t mem_used_max_store(struct device *dev, | |
171 | struct device_attribute *attr, const char *buf, size_t len) | |
172 | { | |
173 | int err; | |
174 | unsigned long val; | |
175 | struct zram *zram = dev_to_zram(dev); | |
461a8eee MK |
176 | |
177 | err = kstrtoul(buf, 10, &val); | |
178 | if (err || val != 0) | |
179 | return -EINVAL; | |
180 | ||
181 | down_read(&zram->init_lock); | |
5a99e95b WY |
182 | if (init_done(zram)) { |
183 | struct zram_meta *meta = zram->meta; | |
461a8eee MK |
184 | atomic_long_set(&zram->stats.max_used_pages, |
185 | zs_get_total_pages(meta->mem_pool)); | |
5a99e95b | 186 | } |
461a8eee MK |
187 | up_read(&zram->init_lock); |
188 | ||
189 | return len; | |
190 | } | |
191 | ||
beca3ec7 SS |
192 | static ssize_t max_comp_streams_store(struct device *dev, |
193 | struct device_attribute *attr, const char *buf, size_t len) | |
194 | { | |
195 | int num; | |
196 | struct zram *zram = dev_to_zram(dev); | |
60a726e3 | 197 | int ret; |
beca3ec7 | 198 | |
60a726e3 MK |
199 | ret = kstrtoint(buf, 0, &num); |
200 | if (ret < 0) | |
201 | return ret; | |
beca3ec7 SS |
202 | if (num < 1) |
203 | return -EINVAL; | |
60a726e3 | 204 | |
beca3ec7 SS |
205 | down_write(&zram->init_lock); |
206 | if (init_done(zram)) { | |
60a726e3 | 207 | if (!zcomp_set_max_streams(zram->comp, num)) { |
fe8eb122 | 208 | pr_info("Cannot change max compression streams\n"); |
60a726e3 MK |
209 | ret = -EINVAL; |
210 | goto out; | |
211 | } | |
beca3ec7 | 212 | } |
60a726e3 | 213 | |
beca3ec7 | 214 | zram->max_comp_streams = num; |
60a726e3 MK |
215 | ret = len; |
216 | out: | |
beca3ec7 | 217 | up_write(&zram->init_lock); |
60a726e3 | 218 | return ret; |
beca3ec7 SS |
219 | } |
220 | ||
e46b8a03 SS |
221 | static ssize_t comp_algorithm_show(struct device *dev, |
222 | struct device_attribute *attr, char *buf) | |
223 | { | |
224 | size_t sz; | |
225 | struct zram *zram = dev_to_zram(dev); | |
226 | ||
227 | down_read(&zram->init_lock); | |
228 | sz = zcomp_available_show(zram->compressor, buf); | |
229 | up_read(&zram->init_lock); | |
230 | ||
231 | return sz; | |
232 | } | |
233 | ||
234 | static ssize_t comp_algorithm_store(struct device *dev, | |
235 | struct device_attribute *attr, const char *buf, size_t len) | |
236 | { | |
237 | struct zram *zram = dev_to_zram(dev); | |
238 | down_write(&zram->init_lock); | |
239 | if (init_done(zram)) { | |
240 | up_write(&zram->init_lock); | |
241 | pr_info("Can't change algorithm for initialized device\n"); | |
242 | return -EBUSY; | |
243 | } | |
244 | strlcpy(zram->compressor, buf, sizeof(zram->compressor)); | |
245 | up_write(&zram->init_lock); | |
246 | return len; | |
247 | } | |
248 | ||
92967471 | 249 | /* flag operations needs meta->tb_lock */ |
8b3cc3ed | 250 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 251 | enum zram_pageflags flag) |
306b0c95 | 252 | { |
d2d5e762 | 253 | return meta->table[index].value & BIT(flag); |
306b0c95 NG |
254 | } |
255 | ||
8b3cc3ed | 256 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 257 | enum zram_pageflags flag) |
306b0c95 | 258 | { |
d2d5e762 | 259 | meta->table[index].value |= BIT(flag); |
306b0c95 NG |
260 | } |
261 | ||
8b3cc3ed | 262 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 263 | enum zram_pageflags flag) |
306b0c95 | 264 | { |
d2d5e762 WY |
265 | meta->table[index].value &= ~BIT(flag); |
266 | } | |
267 | ||
268 | static size_t zram_get_obj_size(struct zram_meta *meta, u32 index) | |
269 | { | |
270 | return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1); | |
271 | } | |
272 | ||
273 | static void zram_set_obj_size(struct zram_meta *meta, | |
274 | u32 index, size_t size) | |
275 | { | |
276 | unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT; | |
277 | ||
278 | meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size; | |
306b0c95 NG |
279 | } |
280 | ||
9b3bb7ab SS |
281 | static inline int is_partial_io(struct bio_vec *bvec) |
282 | { | |
283 | return bvec->bv_len != PAGE_SIZE; | |
284 | } | |
285 | ||
286 | /* | |
287 | * Check if request is within bounds and aligned on zram logical blocks. | |
288 | */ | |
54850e73 | 289 | static inline int valid_io_request(struct zram *zram, |
290 | sector_t start, unsigned int size) | |
9b3bb7ab | 291 | { |
54850e73 | 292 | u64 end, bound; |
a539c72a | 293 | |
9b3bb7ab | 294 | /* unaligned request */ |
54850e73 | 295 | if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1))) |
9b3bb7ab | 296 | return 0; |
54850e73 | 297 | if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) |
9b3bb7ab SS |
298 | return 0; |
299 | ||
54850e73 | 300 | end = start + (size >> SECTOR_SHIFT); |
9b3bb7ab SS |
301 | bound = zram->disksize >> SECTOR_SHIFT; |
302 | /* out of range range */ | |
75c7caf5 | 303 | if (unlikely(start >= bound || end > bound || start > end)) |
9b3bb7ab SS |
304 | return 0; |
305 | ||
306 | /* I/O request is valid */ | |
307 | return 1; | |
308 | } | |
309 | ||
1fec1172 | 310 | static void zram_meta_free(struct zram_meta *meta, u64 disksize) |
9b3bb7ab | 311 | { |
1fec1172 GM |
312 | size_t num_pages = disksize >> PAGE_SHIFT; |
313 | size_t index; | |
314 | ||
315 | /* Free all pages that are still in this zram device */ | |
316 | for (index = 0; index < num_pages; index++) { | |
317 | unsigned long handle = meta->table[index].handle; | |
318 | ||
319 | if (!handle) | |
320 | continue; | |
321 | ||
322 | zs_free(meta->mem_pool, handle); | |
323 | } | |
324 | ||
9b3bb7ab | 325 | zs_destroy_pool(meta->mem_pool); |
9b3bb7ab SS |
326 | vfree(meta->table); |
327 | kfree(meta); | |
328 | } | |
329 | ||
3eba0c6a | 330 | static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize) |
9b3bb7ab SS |
331 | { |
332 | size_t num_pages; | |
3eba0c6a | 333 | char pool_name[8]; |
9b3bb7ab | 334 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); |
b8179958 | 335 | |
9b3bb7ab | 336 | if (!meta) |
b8179958 | 337 | return NULL; |
9b3bb7ab | 338 | |
9b3bb7ab SS |
339 | num_pages = disksize >> PAGE_SHIFT; |
340 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
341 | if (!meta->table) { | |
342 | pr_err("Error allocating zram address table\n"); | |
b8179958 | 343 | goto out_error; |
9b3bb7ab SS |
344 | } |
345 | ||
3eba0c6a GM |
346 | snprintf(pool_name, sizeof(pool_name), "zram%d", device_id); |
347 | meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM); | |
9b3bb7ab SS |
348 | if (!meta->mem_pool) { |
349 | pr_err("Error creating memory pool\n"); | |
b8179958 | 350 | goto out_error; |
9b3bb7ab SS |
351 | } |
352 | ||
353 | return meta; | |
354 | ||
b8179958 | 355 | out_error: |
9b3bb7ab | 356 | vfree(meta->table); |
9b3bb7ab | 357 | kfree(meta); |
b8179958 | 358 | return NULL; |
9b3bb7ab SS |
359 | } |
360 | ||
08eee69f MK |
361 | static inline bool zram_meta_get(struct zram *zram) |
362 | { | |
363 | if (atomic_inc_not_zero(&zram->refcount)) | |
364 | return true; | |
365 | return false; | |
366 | } | |
367 | ||
368 | static inline void zram_meta_put(struct zram *zram) | |
369 | { | |
370 | atomic_dec(&zram->refcount); | |
371 | } | |
372 | ||
9b3bb7ab SS |
373 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) |
374 | { | |
375 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
376 | (*index)++; | |
377 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
378 | } | |
379 | ||
306b0c95 NG |
380 | static int page_zero_filled(void *ptr) |
381 | { | |
382 | unsigned int pos; | |
383 | unsigned long *page; | |
384 | ||
385 | page = (unsigned long *)ptr; | |
386 | ||
387 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
388 | if (page[pos]) | |
389 | return 0; | |
390 | } | |
391 | ||
392 | return 1; | |
393 | } | |
394 | ||
9b3bb7ab SS |
395 | static void handle_zero_page(struct bio_vec *bvec) |
396 | { | |
397 | struct page *page = bvec->bv_page; | |
398 | void *user_mem; | |
399 | ||
400 | user_mem = kmap_atomic(page); | |
401 | if (is_partial_io(bvec)) | |
402 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); | |
403 | else | |
404 | clear_page(user_mem); | |
405 | kunmap_atomic(user_mem); | |
406 | ||
407 | flush_dcache_page(page); | |
408 | } | |
409 | ||
d2d5e762 WY |
410 | |
411 | /* | |
412 | * To protect concurrent access to the same index entry, | |
413 | * caller should hold this table index entry's bit_spinlock to | |
414 | * indicate this index entry is accessing. | |
415 | */ | |
f1e3cfff | 416 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 417 | { |
8b3cc3ed MK |
418 | struct zram_meta *meta = zram->meta; |
419 | unsigned long handle = meta->table[index].handle; | |
306b0c95 | 420 | |
fd1a30de | 421 | if (unlikely(!handle)) { |
2e882281 NG |
422 | /* |
423 | * No memory is allocated for zero filled pages. | |
424 | * Simply clear zero page flag. | |
425 | */ | |
8b3cc3ed MK |
426 | if (zram_test_flag(meta, index, ZRAM_ZERO)) { |
427 | zram_clear_flag(meta, index, ZRAM_ZERO); | |
90a7806e | 428 | atomic64_dec(&zram->stats.zero_pages); |
306b0c95 NG |
429 | } |
430 | return; | |
431 | } | |
432 | ||
8b3cc3ed | 433 | zs_free(meta->mem_pool, handle); |
306b0c95 | 434 | |
d2d5e762 WY |
435 | atomic64_sub(zram_get_obj_size(meta, index), |
436 | &zram->stats.compr_data_size); | |
90a7806e | 437 | atomic64_dec(&zram->stats.pages_stored); |
306b0c95 | 438 | |
8b3cc3ed | 439 | meta->table[index].handle = 0; |
d2d5e762 | 440 | zram_set_obj_size(meta, index, 0); |
306b0c95 NG |
441 | } |
442 | ||
37b51fdd | 443 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 444 | { |
b7ca232e | 445 | int ret = 0; |
37b51fdd | 446 | unsigned char *cmem; |
8b3cc3ed | 447 | struct zram_meta *meta = zram->meta; |
92967471 | 448 | unsigned long handle; |
023b409f | 449 | size_t size; |
92967471 | 450 | |
d2d5e762 | 451 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
92967471 | 452 | handle = meta->table[index].handle; |
d2d5e762 | 453 | size = zram_get_obj_size(meta, index); |
306b0c95 | 454 | |
8b3cc3ed | 455 | if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) { |
d2d5e762 | 456 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
42e99bd9 | 457 | clear_page(mem); |
8c921b2b JM |
458 | return 0; |
459 | } | |
306b0c95 | 460 | |
8b3cc3ed | 461 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
92967471 | 462 | if (size == PAGE_SIZE) |
42e99bd9 | 463 | copy_page(mem, cmem); |
37b51fdd | 464 | else |
b7ca232e | 465 | ret = zcomp_decompress(zram->comp, cmem, size, mem); |
8b3cc3ed | 466 | zs_unmap_object(meta->mem_pool, handle); |
d2d5e762 | 467 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
a1dd52af | 468 | |
8c921b2b | 469 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 470 | if (unlikely(ret)) { |
8c921b2b | 471 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); |
8c921b2b | 472 | return ret; |
a1dd52af | 473 | } |
306b0c95 | 474 | |
8c921b2b | 475 | return 0; |
306b0c95 NG |
476 | } |
477 | ||
37b51fdd | 478 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
b627cff3 | 479 | u32 index, int offset) |
924bd88d JM |
480 | { |
481 | int ret; | |
37b51fdd SS |
482 | struct page *page; |
483 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 484 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
485 | page = bvec->bv_page; |
486 | ||
d2d5e762 | 487 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
8b3cc3ed MK |
488 | if (unlikely(!meta->table[index].handle) || |
489 | zram_test_flag(meta, index, ZRAM_ZERO)) { | |
d2d5e762 | 490 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
37b51fdd | 491 | handle_zero_page(bvec); |
924bd88d JM |
492 | return 0; |
493 | } | |
d2d5e762 | 494 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
924bd88d | 495 | |
37b51fdd SS |
496 | if (is_partial_io(bvec)) |
497 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
498 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
499 | ||
500 | user_mem = kmap_atomic(page); | |
501 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
502 | uncmem = user_mem; |
503 | ||
504 | if (!uncmem) { | |
505 | pr_info("Unable to allocate temp memory\n"); | |
506 | ret = -ENOMEM; | |
507 | goto out_cleanup; | |
508 | } | |
924bd88d | 509 | |
37b51fdd | 510 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 511 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 512 | if (unlikely(ret)) |
37b51fdd | 513 | goto out_cleanup; |
924bd88d | 514 | |
37b51fdd SS |
515 | if (is_partial_io(bvec)) |
516 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
517 | bvec->bv_len); | |
518 | ||
519 | flush_dcache_page(page); | |
520 | ret = 0; | |
521 | out_cleanup: | |
522 | kunmap_atomic(user_mem); | |
523 | if (is_partial_io(bvec)) | |
524 | kfree(uncmem); | |
525 | return ret; | |
924bd88d JM |
526 | } |
527 | ||
461a8eee MK |
528 | static inline void update_used_max(struct zram *zram, |
529 | const unsigned long pages) | |
530 | { | |
2ea55a2c | 531 | unsigned long old_max, cur_max; |
461a8eee MK |
532 | |
533 | old_max = atomic_long_read(&zram->stats.max_used_pages); | |
534 | ||
535 | do { | |
536 | cur_max = old_max; | |
537 | if (pages > cur_max) | |
538 | old_max = atomic_long_cmpxchg( | |
539 | &zram->stats.max_used_pages, cur_max, pages); | |
540 | } while (old_max != cur_max); | |
541 | } | |
542 | ||
924bd88d JM |
543 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, |
544 | int offset) | |
306b0c95 | 545 | { |
397c6066 | 546 | int ret = 0; |
8c921b2b | 547 | size_t clen; |
c2344348 | 548 | unsigned long handle; |
130f315a | 549 | struct page *page; |
924bd88d | 550 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 551 | struct zram_meta *meta = zram->meta; |
b7ca232e | 552 | struct zcomp_strm *zstrm; |
e46e3315 | 553 | bool locked = false; |
461a8eee | 554 | unsigned long alloced_pages; |
306b0c95 | 555 | |
8c921b2b | 556 | page = bvec->bv_page; |
924bd88d JM |
557 | if (is_partial_io(bvec)) { |
558 | /* | |
559 | * This is a partial IO. We need to read the full page | |
560 | * before to write the changes. | |
561 | */ | |
7e5a5104 | 562 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 563 | if (!uncmem) { |
924bd88d JM |
564 | ret = -ENOMEM; |
565 | goto out; | |
566 | } | |
37b51fdd | 567 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 568 | if (ret) |
924bd88d | 569 | goto out; |
924bd88d JM |
570 | } |
571 | ||
b7ca232e | 572 | zstrm = zcomp_strm_find(zram->comp); |
e46e3315 | 573 | locked = true; |
ba82fe2e | 574 | user_mem = kmap_atomic(page); |
924bd88d | 575 | |
397c6066 | 576 | if (is_partial_io(bvec)) { |
924bd88d JM |
577 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
578 | bvec->bv_len); | |
397c6066 NG |
579 | kunmap_atomic(user_mem); |
580 | user_mem = NULL; | |
581 | } else { | |
924bd88d | 582 | uncmem = user_mem; |
397c6066 | 583 | } |
924bd88d JM |
584 | |
585 | if (page_zero_filled(uncmem)) { | |
c4065152 WY |
586 | if (user_mem) |
587 | kunmap_atomic(user_mem); | |
f40ac2ae | 588 | /* Free memory associated with this sector now. */ |
d2d5e762 | 589 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 590 | zram_free_page(zram, index); |
92967471 | 591 | zram_set_flag(meta, index, ZRAM_ZERO); |
d2d5e762 | 592 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 593 | |
90a7806e | 594 | atomic64_inc(&zram->stats.zero_pages); |
924bd88d JM |
595 | ret = 0; |
596 | goto out; | |
8c921b2b | 597 | } |
306b0c95 | 598 | |
b7ca232e | 599 | ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen); |
397c6066 NG |
600 | if (!is_partial_io(bvec)) { |
601 | kunmap_atomic(user_mem); | |
602 | user_mem = NULL; | |
603 | uncmem = NULL; | |
604 | } | |
306b0c95 | 605 | |
b7ca232e | 606 | if (unlikely(ret)) { |
8c921b2b | 607 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 608 | goto out; |
8c921b2b | 609 | } |
b7ca232e | 610 | src = zstrm->buffer; |
c8f2f0db | 611 | if (unlikely(clen > max_zpage_size)) { |
c8f2f0db | 612 | clen = PAGE_SIZE; |
397c6066 NG |
613 | if (is_partial_io(bvec)) |
614 | src = uncmem; | |
c8f2f0db | 615 | } |
a1dd52af | 616 | |
8b3cc3ed | 617 | handle = zs_malloc(meta->mem_pool, clen); |
fd1a30de | 618 | if (!handle) { |
596b3dd4 MR |
619 | pr_info("Error allocating memory for compressed page: %u, size=%zu\n", |
620 | index, clen); | |
924bd88d JM |
621 | ret = -ENOMEM; |
622 | goto out; | |
8c921b2b | 623 | } |
9ada9da9 | 624 | |
461a8eee MK |
625 | alloced_pages = zs_get_total_pages(meta->mem_pool); |
626 | if (zram->limit_pages && alloced_pages > zram->limit_pages) { | |
9ada9da9 MK |
627 | zs_free(meta->mem_pool, handle); |
628 | ret = -ENOMEM; | |
629 | goto out; | |
630 | } | |
631 | ||
461a8eee MK |
632 | update_used_max(zram, alloced_pages); |
633 | ||
8b3cc3ed | 634 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 635 | |
42e99bd9 | 636 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 637 | src = kmap_atomic(page); |
42e99bd9 | 638 | copy_page(cmem, src); |
397c6066 | 639 | kunmap_atomic(src); |
42e99bd9 JL |
640 | } else { |
641 | memcpy(cmem, src, clen); | |
642 | } | |
306b0c95 | 643 | |
b7ca232e SS |
644 | zcomp_strm_release(zram->comp, zstrm); |
645 | locked = false; | |
8b3cc3ed | 646 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 647 | |
f40ac2ae SS |
648 | /* |
649 | * Free memory associated with this sector | |
650 | * before overwriting unused sectors. | |
651 | */ | |
d2d5e762 | 652 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae SS |
653 | zram_free_page(zram, index); |
654 | ||
8b3cc3ed | 655 | meta->table[index].handle = handle; |
d2d5e762 WY |
656 | zram_set_obj_size(meta, index, clen); |
657 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); | |
306b0c95 | 658 | |
8c921b2b | 659 | /* Update stats */ |
90a7806e SS |
660 | atomic64_add(clen, &zram->stats.compr_data_size); |
661 | atomic64_inc(&zram->stats.pages_stored); | |
924bd88d | 662 | out: |
e46e3315 | 663 | if (locked) |
b7ca232e | 664 | zcomp_strm_release(zram->comp, zstrm); |
397c6066 NG |
665 | if (is_partial_io(bvec)) |
666 | kfree(uncmem); | |
924bd88d | 667 | return ret; |
8c921b2b JM |
668 | } |
669 | ||
670 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
b627cff3 | 671 | int offset, int rw) |
8c921b2b | 672 | { |
c5bde238 | 673 | int ret; |
8c921b2b | 674 | |
be257c61 SS |
675 | if (rw == READ) { |
676 | atomic64_inc(&zram->stats.num_reads); | |
b627cff3 | 677 | ret = zram_bvec_read(zram, bvec, index, offset); |
be257c61 SS |
678 | } else { |
679 | atomic64_inc(&zram->stats.num_writes); | |
c5bde238 | 680 | ret = zram_bvec_write(zram, bvec, index, offset); |
be257c61 | 681 | } |
c5bde238 | 682 | |
0cf1e9d6 CY |
683 | if (unlikely(ret)) { |
684 | if (rw == READ) | |
685 | atomic64_inc(&zram->stats.failed_reads); | |
686 | else | |
687 | atomic64_inc(&zram->stats.failed_writes); | |
688 | } | |
689 | ||
c5bde238 | 690 | return ret; |
924bd88d JM |
691 | } |
692 | ||
f4659d8e JK |
693 | /* |
694 | * zram_bio_discard - handler on discard request | |
695 | * @index: physical block index in PAGE_SIZE units | |
696 | * @offset: byte offset within physical block | |
697 | */ | |
698 | static void zram_bio_discard(struct zram *zram, u32 index, | |
699 | int offset, struct bio *bio) | |
700 | { | |
701 | size_t n = bio->bi_iter.bi_size; | |
d2d5e762 | 702 | struct zram_meta *meta = zram->meta; |
f4659d8e JK |
703 | |
704 | /* | |
705 | * zram manages data in physical block size units. Because logical block | |
706 | * size isn't identical with physical block size on some arch, we | |
707 | * could get a discard request pointing to a specific offset within a | |
708 | * certain physical block. Although we can handle this request by | |
709 | * reading that physiclal block and decompressing and partially zeroing | |
710 | * and re-compressing and then re-storing it, this isn't reasonable | |
711 | * because our intent with a discard request is to save memory. So | |
712 | * skipping this logical block is appropriate here. | |
713 | */ | |
714 | if (offset) { | |
38515c73 | 715 | if (n <= (PAGE_SIZE - offset)) |
f4659d8e JK |
716 | return; |
717 | ||
38515c73 | 718 | n -= (PAGE_SIZE - offset); |
f4659d8e JK |
719 | index++; |
720 | } | |
721 | ||
722 | while (n >= PAGE_SIZE) { | |
d2d5e762 | 723 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f4659d8e | 724 | zram_free_page(zram, index); |
d2d5e762 | 725 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
015254da | 726 | atomic64_inc(&zram->stats.notify_free); |
f4659d8e JK |
727 | index++; |
728 | n -= PAGE_SIZE; | |
729 | } | |
730 | } | |
731 | ||
ba6b17d6 | 732 | static void zram_reset_device(struct zram *zram) |
924bd88d | 733 | { |
08eee69f MK |
734 | struct zram_meta *meta; |
735 | struct zcomp *comp; | |
736 | u64 disksize; | |
737 | ||
644d4787 | 738 | down_write(&zram->init_lock); |
9ada9da9 MK |
739 | |
740 | zram->limit_pages = 0; | |
741 | ||
be2d1d56 | 742 | if (!init_done(zram)) { |
644d4787 | 743 | up_write(&zram->init_lock); |
9b3bb7ab | 744 | return; |
644d4787 | 745 | } |
9b3bb7ab | 746 | |
08eee69f MK |
747 | meta = zram->meta; |
748 | comp = zram->comp; | |
749 | disksize = zram->disksize; | |
750 | /* | |
751 | * Refcount will go down to 0 eventually and r/w handler | |
752 | * cannot handle further I/O so it will bail out by | |
753 | * check zram_meta_get. | |
754 | */ | |
755 | zram_meta_put(zram); | |
756 | /* | |
757 | * We want to free zram_meta in process context to avoid | |
758 | * deadlock between reclaim path and any other locks. | |
759 | */ | |
760 | wait_event(zram->io_done, atomic_read(&zram->refcount) == 0); | |
761 | ||
9b3bb7ab SS |
762 | /* Reset stats */ |
763 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
9b3bb7ab | 764 | zram->disksize = 0; |
08eee69f | 765 | zram->max_comp_streams = 1; |
a096cafc SS |
766 | set_capacity(zram->disk, 0); |
767 | ||
644d4787 | 768 | up_write(&zram->init_lock); |
08eee69f MK |
769 | /* I/O operation under all of CPU are done so let's free */ |
770 | zram_meta_free(meta, disksize); | |
771 | zcomp_destroy(comp); | |
9b3bb7ab SS |
772 | } |
773 | ||
9b3bb7ab SS |
774 | static ssize_t disksize_store(struct device *dev, |
775 | struct device_attribute *attr, const char *buf, size_t len) | |
776 | { | |
777 | u64 disksize; | |
d61f98c7 | 778 | struct zcomp *comp; |
9b3bb7ab SS |
779 | struct zram_meta *meta; |
780 | struct zram *zram = dev_to_zram(dev); | |
fcfa8d95 | 781 | int err; |
9b3bb7ab SS |
782 | |
783 | disksize = memparse(buf, NULL); | |
784 | if (!disksize) | |
785 | return -EINVAL; | |
786 | ||
787 | disksize = PAGE_ALIGN(disksize); | |
3eba0c6a | 788 | meta = zram_meta_alloc(zram->disk->first_minor, disksize); |
db5d711e MK |
789 | if (!meta) |
790 | return -ENOMEM; | |
b67d1ec1 | 791 | |
d61f98c7 | 792 | comp = zcomp_create(zram->compressor, zram->max_comp_streams); |
fcfa8d95 | 793 | if (IS_ERR(comp)) { |
d61f98c7 SS |
794 | pr_info("Cannot initialise %s compressing backend\n", |
795 | zram->compressor); | |
fcfa8d95 SS |
796 | err = PTR_ERR(comp); |
797 | goto out_free_meta; | |
d61f98c7 SS |
798 | } |
799 | ||
9b3bb7ab | 800 | down_write(&zram->init_lock); |
be2d1d56 | 801 | if (init_done(zram)) { |
9b3bb7ab | 802 | pr_info("Cannot change disksize for initialized device\n"); |
b7ca232e | 803 | err = -EBUSY; |
fcfa8d95 | 804 | goto out_destroy_comp; |
9b3bb7ab SS |
805 | } |
806 | ||
08eee69f MK |
807 | init_waitqueue_head(&zram->io_done); |
808 | atomic_set(&zram->refcount, 1); | |
b67d1ec1 | 809 | zram->meta = meta; |
d61f98c7 | 810 | zram->comp = comp; |
9b3bb7ab SS |
811 | zram->disksize = disksize; |
812 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
9b3bb7ab | 813 | up_write(&zram->init_lock); |
b4c5c609 MK |
814 | |
815 | /* | |
816 | * Revalidate disk out of the init_lock to avoid lockdep splat. | |
817 | * It's okay because disk's capacity is protected by init_lock | |
818 | * so that revalidate_disk always sees up-to-date capacity. | |
819 | */ | |
820 | revalidate_disk(zram->disk); | |
821 | ||
9b3bb7ab | 822 | return len; |
b7ca232e | 823 | |
fcfa8d95 SS |
824 | out_destroy_comp: |
825 | up_write(&zram->init_lock); | |
826 | zcomp_destroy(comp); | |
827 | out_free_meta: | |
1fec1172 | 828 | zram_meta_free(meta, disksize); |
b7ca232e | 829 | return err; |
9b3bb7ab SS |
830 | } |
831 | ||
832 | static ssize_t reset_store(struct device *dev, | |
833 | struct device_attribute *attr, const char *buf, size_t len) | |
834 | { | |
835 | int ret; | |
836 | unsigned short do_reset; | |
837 | struct zram *zram; | |
838 | struct block_device *bdev; | |
839 | ||
840 | zram = dev_to_zram(dev); | |
841 | bdev = bdget_disk(zram->disk, 0); | |
842 | ||
46a51c80 RK |
843 | if (!bdev) |
844 | return -ENOMEM; | |
845 | ||
ba6b17d6 | 846 | mutex_lock(&bdev->bd_mutex); |
9b3bb7ab | 847 | /* Do not reset an active device! */ |
2b269ce6 | 848 | if (bdev->bd_openers) { |
1b672224 RK |
849 | ret = -EBUSY; |
850 | goto out; | |
851 | } | |
9b3bb7ab SS |
852 | |
853 | ret = kstrtou16(buf, 10, &do_reset); | |
854 | if (ret) | |
1b672224 | 855 | goto out; |
9b3bb7ab | 856 | |
1b672224 RK |
857 | if (!do_reset) { |
858 | ret = -EINVAL; | |
859 | goto out; | |
860 | } | |
9b3bb7ab SS |
861 | |
862 | /* Make sure all pending I/O is finished */ | |
46a51c80 | 863 | fsync_bdev(bdev); |
ba6b17d6 | 864 | zram_reset_device(zram); |
ba6b17d6 SS |
865 | |
866 | mutex_unlock(&bdev->bd_mutex); | |
867 | revalidate_disk(zram->disk); | |
1b672224 | 868 | bdput(bdev); |
9b3bb7ab | 869 | |
9b3bb7ab | 870 | return len; |
1b672224 RK |
871 | |
872 | out: | |
ba6b17d6 | 873 | mutex_unlock(&bdev->bd_mutex); |
1b672224 RK |
874 | bdput(bdev); |
875 | return ret; | |
8c921b2b JM |
876 | } |
877 | ||
be257c61 | 878 | static void __zram_make_request(struct zram *zram, struct bio *bio) |
8c921b2b | 879 | { |
b627cff3 | 880 | int offset, rw; |
8c921b2b | 881 | u32 index; |
7988613b KO |
882 | struct bio_vec bvec; |
883 | struct bvec_iter iter; | |
8c921b2b | 884 | |
4f024f37 KO |
885 | index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT; |
886 | offset = (bio->bi_iter.bi_sector & | |
887 | (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; | |
8c921b2b | 888 | |
f4659d8e JK |
889 | if (unlikely(bio->bi_rw & REQ_DISCARD)) { |
890 | zram_bio_discard(zram, index, offset, bio); | |
891 | bio_endio(bio, 0); | |
892 | return; | |
893 | } | |
894 | ||
b627cff3 | 895 | rw = bio_data_dir(bio); |
7988613b | 896 | bio_for_each_segment(bvec, bio, iter) { |
924bd88d JM |
897 | int max_transfer_size = PAGE_SIZE - offset; |
898 | ||
7988613b | 899 | if (bvec.bv_len > max_transfer_size) { |
924bd88d JM |
900 | /* |
901 | * zram_bvec_rw() can only make operation on a single | |
902 | * zram page. Split the bio vector. | |
903 | */ | |
904 | struct bio_vec bv; | |
905 | ||
7988613b | 906 | bv.bv_page = bvec.bv_page; |
924bd88d | 907 | bv.bv_len = max_transfer_size; |
7988613b | 908 | bv.bv_offset = bvec.bv_offset; |
924bd88d | 909 | |
b627cff3 | 910 | if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0) |
924bd88d JM |
911 | goto out; |
912 | ||
7988613b | 913 | bv.bv_len = bvec.bv_len - max_transfer_size; |
924bd88d | 914 | bv.bv_offset += max_transfer_size; |
b627cff3 | 915 | if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0) |
924bd88d JM |
916 | goto out; |
917 | } else | |
b627cff3 | 918 | if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0) |
924bd88d JM |
919 | goto out; |
920 | ||
7988613b | 921 | update_position(&index, &offset, &bvec); |
a1dd52af | 922 | } |
306b0c95 NG |
923 | |
924 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
925 | bio_endio(bio, 0); | |
7d7854b4 | 926 | return; |
306b0c95 NG |
927 | |
928 | out: | |
306b0c95 | 929 | bio_io_error(bio); |
306b0c95 NG |
930 | } |
931 | ||
306b0c95 | 932 | /* |
f1e3cfff | 933 | * Handler function for all zram I/O requests. |
306b0c95 | 934 | */ |
5a7bbad2 | 935 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 936 | { |
f1e3cfff | 937 | struct zram *zram = queue->queuedata; |
306b0c95 | 938 | |
08eee69f | 939 | if (unlikely(!zram_meta_get(zram))) |
3de738cd | 940 | goto error; |
0900beae | 941 | |
54850e73 | 942 | if (!valid_io_request(zram, bio->bi_iter.bi_sector, |
943 | bio->bi_iter.bi_size)) { | |
da5cc7d3 | 944 | atomic64_inc(&zram->stats.invalid_io); |
08eee69f | 945 | goto put_zram; |
6642a67c JM |
946 | } |
947 | ||
be257c61 | 948 | __zram_make_request(zram, bio); |
08eee69f | 949 | zram_meta_put(zram); |
b4fdcb02 | 950 | return; |
08eee69f MK |
951 | put_zram: |
952 | zram_meta_put(zram); | |
0900beae JM |
953 | error: |
954 | bio_io_error(bio); | |
306b0c95 NG |
955 | } |
956 | ||
2ccbec05 NG |
957 | static void zram_slot_free_notify(struct block_device *bdev, |
958 | unsigned long index) | |
107c161b | 959 | { |
f1e3cfff | 960 | struct zram *zram; |
f614a9f4 | 961 | struct zram_meta *meta; |
107c161b | 962 | |
f1e3cfff | 963 | zram = bdev->bd_disk->private_data; |
f614a9f4 | 964 | meta = zram->meta; |
a0c516cb | 965 | |
d2d5e762 | 966 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 967 | zram_free_page(zram, index); |
d2d5e762 | 968 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 969 | atomic64_inc(&zram->stats.notify_free); |
107c161b NG |
970 | } |
971 | ||
8c7f0102 | 972 | static int zram_rw_page(struct block_device *bdev, sector_t sector, |
973 | struct page *page, int rw) | |
974 | { | |
08eee69f | 975 | int offset, err = -EIO; |
8c7f0102 | 976 | u32 index; |
977 | struct zram *zram; | |
978 | struct bio_vec bv; | |
979 | ||
980 | zram = bdev->bd_disk->private_data; | |
08eee69f MK |
981 | if (unlikely(!zram_meta_get(zram))) |
982 | goto out; | |
983 | ||
8c7f0102 | 984 | if (!valid_io_request(zram, sector, PAGE_SIZE)) { |
985 | atomic64_inc(&zram->stats.invalid_io); | |
08eee69f MK |
986 | err = -EINVAL; |
987 | goto put_zram; | |
8c7f0102 | 988 | } |
989 | ||
990 | index = sector >> SECTORS_PER_PAGE_SHIFT; | |
991 | offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT; | |
992 | ||
993 | bv.bv_page = page; | |
994 | bv.bv_len = PAGE_SIZE; | |
995 | bv.bv_offset = 0; | |
996 | ||
997 | err = zram_bvec_rw(zram, &bv, index, offset, rw); | |
08eee69f MK |
998 | put_zram: |
999 | zram_meta_put(zram); | |
1000 | out: | |
8c7f0102 | 1001 | /* |
1002 | * If I/O fails, just return error(ie, non-zero) without | |
1003 | * calling page_endio. | |
1004 | * It causes resubmit the I/O with bio request by upper functions | |
1005 | * of rw_page(e.g., swap_readpage, __swap_writepage) and | |
1006 | * bio->bi_end_io does things to handle the error | |
1007 | * (e.g., SetPageError, set_page_dirty and extra works). | |
1008 | */ | |
1009 | if (err == 0) | |
1010 | page_endio(page, rw, 0); | |
1011 | return err; | |
1012 | } | |
1013 | ||
f1e3cfff | 1014 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 1015 | .swap_slot_free_notify = zram_slot_free_notify, |
8c7f0102 | 1016 | .rw_page = zram_rw_page, |
107c161b | 1017 | .owner = THIS_MODULE |
306b0c95 NG |
1018 | }; |
1019 | ||
083914ea GM |
1020 | static DEVICE_ATTR_RW(disksize); |
1021 | static DEVICE_ATTR_RO(initstate); | |
1022 | static DEVICE_ATTR_WO(reset); | |
1023 | static DEVICE_ATTR_RO(orig_data_size); | |
1024 | static DEVICE_ATTR_RO(mem_used_total); | |
1025 | static DEVICE_ATTR_RW(mem_limit); | |
1026 | static DEVICE_ATTR_RW(mem_used_max); | |
1027 | static DEVICE_ATTR_RW(max_comp_streams); | |
1028 | static DEVICE_ATTR_RW(comp_algorithm); | |
9b3bb7ab | 1029 | |
a68eb3b6 SS |
1030 | ZRAM_ATTR_RO(num_reads); |
1031 | ZRAM_ATTR_RO(num_writes); | |
64447249 SS |
1032 | ZRAM_ATTR_RO(failed_reads); |
1033 | ZRAM_ATTR_RO(failed_writes); | |
a68eb3b6 SS |
1034 | ZRAM_ATTR_RO(invalid_io); |
1035 | ZRAM_ATTR_RO(notify_free); | |
1036 | ZRAM_ATTR_RO(zero_pages); | |
1037 | ZRAM_ATTR_RO(compr_data_size); | |
1038 | ||
9b3bb7ab SS |
1039 | static struct attribute *zram_disk_attrs[] = { |
1040 | &dev_attr_disksize.attr, | |
1041 | &dev_attr_initstate.attr, | |
1042 | &dev_attr_reset.attr, | |
1043 | &dev_attr_num_reads.attr, | |
1044 | &dev_attr_num_writes.attr, | |
64447249 SS |
1045 | &dev_attr_failed_reads.attr, |
1046 | &dev_attr_failed_writes.attr, | |
9b3bb7ab SS |
1047 | &dev_attr_invalid_io.attr, |
1048 | &dev_attr_notify_free.attr, | |
1049 | &dev_attr_zero_pages.attr, | |
1050 | &dev_attr_orig_data_size.attr, | |
1051 | &dev_attr_compr_data_size.attr, | |
1052 | &dev_attr_mem_used_total.attr, | |
9ada9da9 | 1053 | &dev_attr_mem_limit.attr, |
461a8eee | 1054 | &dev_attr_mem_used_max.attr, |
beca3ec7 | 1055 | &dev_attr_max_comp_streams.attr, |
e46b8a03 | 1056 | &dev_attr_comp_algorithm.attr, |
9b3bb7ab SS |
1057 | NULL, |
1058 | }; | |
1059 | ||
1060 | static struct attribute_group zram_disk_attr_group = { | |
1061 | .attrs = zram_disk_attrs, | |
1062 | }; | |
1063 | ||
f1e3cfff | 1064 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 1065 | { |
ee980160 | 1066 | struct request_queue *queue; |
39a9b8ac | 1067 | int ret = -ENOMEM; |
de1a21a0 | 1068 | |
0900beae | 1069 | init_rwsem(&zram->init_lock); |
306b0c95 | 1070 | |
ee980160 SS |
1071 | queue = blk_alloc_queue(GFP_KERNEL); |
1072 | if (!queue) { | |
306b0c95 NG |
1073 | pr_err("Error allocating disk queue for device %d\n", |
1074 | device_id); | |
de1a21a0 | 1075 | goto out; |
306b0c95 NG |
1076 | } |
1077 | ||
ee980160 | 1078 | blk_queue_make_request(queue, zram_make_request); |
306b0c95 NG |
1079 | |
1080 | /* gendisk structure */ | |
f1e3cfff NG |
1081 | zram->disk = alloc_disk(1); |
1082 | if (!zram->disk) { | |
94b8435f | 1083 | pr_warn("Error allocating disk structure for device %d\n", |
306b0c95 | 1084 | device_id); |
39a9b8ac | 1085 | goto out_free_queue; |
306b0c95 NG |
1086 | } |
1087 | ||
f1e3cfff NG |
1088 | zram->disk->major = zram_major; |
1089 | zram->disk->first_minor = device_id; | |
1090 | zram->disk->fops = &zram_devops; | |
ee980160 SS |
1091 | zram->disk->queue = queue; |
1092 | zram->disk->queue->queuedata = zram; | |
f1e3cfff NG |
1093 | zram->disk->private_data = zram; |
1094 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 1095 | |
33863c21 | 1096 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 1097 | set_capacity(zram->disk, 0); |
b67d1ec1 SS |
1098 | /* zram devices sort of resembles non-rotational disks */ |
1099 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
b277da0a | 1100 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue); |
a1dd52af NG |
1101 | /* |
1102 | * To ensure that we always get PAGE_SIZE aligned | |
1103 | * and n*PAGE_SIZED sized I/O requests. | |
1104 | */ | |
f1e3cfff | 1105 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
1106 | blk_queue_logical_block_size(zram->disk->queue, |
1107 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
1108 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
1109 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
f4659d8e JK |
1110 | zram->disk->queue->limits.discard_granularity = PAGE_SIZE; |
1111 | zram->disk->queue->limits.max_discard_sectors = UINT_MAX; | |
1112 | /* | |
1113 | * zram_bio_discard() will clear all logical blocks if logical block | |
1114 | * size is identical with physical block size(PAGE_SIZE). But if it is | |
1115 | * different, we will skip discarding some parts of logical blocks in | |
1116 | * the part of the request range which isn't aligned to physical block | |
1117 | * size. So we can't ensure that all discarded logical blocks are | |
1118 | * zeroed. | |
1119 | */ | |
1120 | if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE) | |
1121 | zram->disk->queue->limits.discard_zeroes_data = 1; | |
1122 | else | |
1123 | zram->disk->queue->limits.discard_zeroes_data = 0; | |
1124 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue); | |
5d83d5a0 | 1125 | |
f1e3cfff | 1126 | add_disk(zram->disk); |
306b0c95 | 1127 | |
33863c21 NG |
1128 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
1129 | &zram_disk_attr_group); | |
1130 | if (ret < 0) { | |
94b8435f | 1131 | pr_warn("Error creating sysfs group"); |
39a9b8ac | 1132 | goto out_free_disk; |
33863c21 | 1133 | } |
e46b8a03 | 1134 | strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor)); |
be2d1d56 | 1135 | zram->meta = NULL; |
beca3ec7 | 1136 | zram->max_comp_streams = 1; |
39a9b8ac | 1137 | return 0; |
de1a21a0 | 1138 | |
39a9b8ac JL |
1139 | out_free_disk: |
1140 | del_gendisk(zram->disk); | |
1141 | put_disk(zram->disk); | |
1142 | out_free_queue: | |
ee980160 | 1143 | blk_cleanup_queue(queue); |
de1a21a0 NG |
1144 | out: |
1145 | return ret; | |
306b0c95 NG |
1146 | } |
1147 | ||
a096cafc | 1148 | static void destroy_devices(unsigned int nr) |
306b0c95 | 1149 | { |
a096cafc SS |
1150 | struct zram *zram; |
1151 | unsigned int i; | |
33863c21 | 1152 | |
a096cafc SS |
1153 | for (i = 0; i < nr; i++) { |
1154 | zram = &zram_devices[i]; | |
1155 | /* | |
1156 | * Remove sysfs first, so no one will perform a disksize | |
1157 | * store while we destroy the devices | |
1158 | */ | |
1159 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, | |
1160 | &zram_disk_attr_group); | |
306b0c95 | 1161 | |
a096cafc SS |
1162 | zram_reset_device(zram); |
1163 | ||
ee980160 | 1164 | blk_cleanup_queue(zram->disk->queue); |
a096cafc SS |
1165 | del_gendisk(zram->disk); |
1166 | put_disk(zram->disk); | |
a096cafc SS |
1167 | } |
1168 | ||
1169 | kfree(zram_devices); | |
1170 | unregister_blkdev(zram_major, "zram"); | |
1171 | pr_info("Destroyed %u device(s)\n", nr); | |
306b0c95 NG |
1172 | } |
1173 | ||
f1e3cfff | 1174 | static int __init zram_init(void) |
306b0c95 | 1175 | { |
de1a21a0 | 1176 | int ret, dev_id; |
306b0c95 | 1177 | |
5fa5a901 | 1178 | if (num_devices > max_num_devices) { |
94b8435f | 1179 | pr_warn("Invalid value for num_devices: %u\n", |
5fa5a901 | 1180 | num_devices); |
a096cafc | 1181 | return -EINVAL; |
306b0c95 NG |
1182 | } |
1183 | ||
f1e3cfff NG |
1184 | zram_major = register_blkdev(0, "zram"); |
1185 | if (zram_major <= 0) { | |
94b8435f | 1186 | pr_warn("Unable to get major number\n"); |
a096cafc | 1187 | return -EBUSY; |
306b0c95 NG |
1188 | } |
1189 | ||
306b0c95 | 1190 | /* Allocate the device array and initialize each one */ |
5fa5a901 | 1191 | zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL); |
43801f6e | 1192 | if (!zram_devices) { |
a096cafc SS |
1193 | unregister_blkdev(zram_major, "zram"); |
1194 | return -ENOMEM; | |
de1a21a0 | 1195 | } |
306b0c95 | 1196 | |
5fa5a901 | 1197 | for (dev_id = 0; dev_id < num_devices; dev_id++) { |
43801f6e | 1198 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 1199 | if (ret) |
a096cafc | 1200 | goto out_error; |
de1a21a0 NG |
1201 | } |
1202 | ||
a096cafc | 1203 | pr_info("Created %u device(s)\n", num_devices); |
306b0c95 | 1204 | return 0; |
de1a21a0 | 1205 | |
a096cafc SS |
1206 | out_error: |
1207 | destroy_devices(dev_id); | |
306b0c95 NG |
1208 | return ret; |
1209 | } | |
1210 | ||
f1e3cfff | 1211 | static void __exit zram_exit(void) |
306b0c95 | 1212 | { |
a096cafc | 1213 | destroy_devices(num_devices); |
306b0c95 NG |
1214 | } |
1215 | ||
f1e3cfff NG |
1216 | module_init(zram_init); |
1217 | module_exit(zram_exit); | |
306b0c95 | 1218 | |
9b3bb7ab SS |
1219 | module_param(num_devices, uint, 0); |
1220 | MODULE_PARM_DESC(num_devices, "Number of zram devices"); | |
1221 | ||
306b0c95 NG |
1222 | MODULE_LICENSE("Dual BSD/GPL"); |
1223 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 1224 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |