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