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 SS |
46 | #define ZRAM_ATTR_RO(name) \ |
47 | static ssize_t zram_attr_##name##_show(struct device *d, \ | |
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 | } \ | |
54 | static struct device_attribute dev_attr_##name = \ | |
55 | __ATTR(name, S_IRUGO, zram_attr_##name##_show, NULL); | |
56 | ||
be2d1d56 SS |
57 | static inline int init_done(struct zram *zram) |
58 | { | |
59 | return zram->meta != NULL; | |
60 | } | |
61 | ||
9b3bb7ab SS |
62 | static inline struct zram *dev_to_zram(struct device *dev) |
63 | { | |
64 | return (struct zram *)dev_to_disk(dev)->private_data; | |
65 | } | |
66 | ||
67 | static ssize_t disksize_show(struct device *dev, | |
68 | struct device_attribute *attr, char *buf) | |
69 | { | |
70 | struct zram *zram = dev_to_zram(dev); | |
71 | ||
56b4e8cb | 72 | return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize); |
9b3bb7ab SS |
73 | } |
74 | ||
75 | static ssize_t initstate_show(struct device *dev, | |
76 | struct device_attribute *attr, char *buf) | |
77 | { | |
a68eb3b6 | 78 | u32 val; |
9b3bb7ab SS |
79 | struct zram *zram = dev_to_zram(dev); |
80 | ||
a68eb3b6 SS |
81 | down_read(&zram->init_lock); |
82 | val = init_done(zram); | |
83 | up_read(&zram->init_lock); | |
9b3bb7ab | 84 | |
56b4e8cb | 85 | return scnprintf(buf, PAGE_SIZE, "%u\n", val); |
9b3bb7ab SS |
86 | } |
87 | ||
88 | static ssize_t orig_data_size_show(struct device *dev, | |
89 | struct device_attribute *attr, char *buf) | |
90 | { | |
91 | struct zram *zram = dev_to_zram(dev); | |
92 | ||
56b4e8cb | 93 | return scnprintf(buf, PAGE_SIZE, "%llu\n", |
90a7806e | 94 | (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT); |
9b3bb7ab SS |
95 | } |
96 | ||
9b3bb7ab SS |
97 | static ssize_t mem_used_total_show(struct device *dev, |
98 | struct device_attribute *attr, char *buf) | |
99 | { | |
100 | u64 val = 0; | |
101 | struct zram *zram = dev_to_zram(dev); | |
9b3bb7ab SS |
102 | |
103 | down_read(&zram->init_lock); | |
5a99e95b WY |
104 | if (init_done(zram)) { |
105 | struct zram_meta *meta = zram->meta; | |
722cdc17 | 106 | val = zs_get_total_pages(meta->mem_pool); |
5a99e95b | 107 | } |
9b3bb7ab SS |
108 | up_read(&zram->init_lock); |
109 | ||
722cdc17 | 110 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); |
9b3bb7ab SS |
111 | } |
112 | ||
beca3ec7 SS |
113 | static ssize_t max_comp_streams_show(struct device *dev, |
114 | struct device_attribute *attr, char *buf) | |
115 | { | |
116 | int val; | |
117 | struct zram *zram = dev_to_zram(dev); | |
118 | ||
119 | down_read(&zram->init_lock); | |
120 | val = zram->max_comp_streams; | |
121 | up_read(&zram->init_lock); | |
122 | ||
56b4e8cb | 123 | return scnprintf(buf, PAGE_SIZE, "%d\n", val); |
beca3ec7 SS |
124 | } |
125 | ||
9ada9da9 MK |
126 | static ssize_t mem_limit_show(struct device *dev, |
127 | struct device_attribute *attr, char *buf) | |
128 | { | |
129 | u64 val; | |
130 | struct zram *zram = dev_to_zram(dev); | |
131 | ||
132 | down_read(&zram->init_lock); | |
133 | val = zram->limit_pages; | |
134 | up_read(&zram->init_lock); | |
135 | ||
136 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
137 | } | |
138 | ||
139 | static ssize_t mem_limit_store(struct device *dev, | |
140 | struct device_attribute *attr, const char *buf, size_t len) | |
141 | { | |
142 | u64 limit; | |
143 | char *tmp; | |
144 | struct zram *zram = dev_to_zram(dev); | |
145 | ||
146 | limit = memparse(buf, &tmp); | |
147 | if (buf == tmp) /* no chars parsed, invalid input */ | |
148 | return -EINVAL; | |
149 | ||
150 | down_write(&zram->init_lock); | |
151 | zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT; | |
152 | up_write(&zram->init_lock); | |
153 | ||
154 | return len; | |
155 | } | |
156 | ||
461a8eee MK |
157 | static ssize_t mem_used_max_show(struct device *dev, |
158 | struct device_attribute *attr, char *buf) | |
159 | { | |
160 | u64 val = 0; | |
161 | struct zram *zram = dev_to_zram(dev); | |
162 | ||
163 | down_read(&zram->init_lock); | |
164 | if (init_done(zram)) | |
165 | val = atomic_long_read(&zram->stats.max_used_pages); | |
166 | up_read(&zram->init_lock); | |
167 | ||
168 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
169 | } | |
170 | ||
171 | static ssize_t mem_used_max_store(struct device *dev, | |
172 | struct device_attribute *attr, const char *buf, size_t len) | |
173 | { | |
174 | int err; | |
175 | unsigned long val; | |
176 | struct zram *zram = dev_to_zram(dev); | |
461a8eee MK |
177 | |
178 | err = kstrtoul(buf, 10, &val); | |
179 | if (err || val != 0) | |
180 | return -EINVAL; | |
181 | ||
182 | down_read(&zram->init_lock); | |
5a99e95b WY |
183 | if (init_done(zram)) { |
184 | struct zram_meta *meta = zram->meta; | |
461a8eee MK |
185 | atomic_long_set(&zram->stats.max_used_pages, |
186 | zs_get_total_pages(meta->mem_pool)); | |
5a99e95b | 187 | } |
461a8eee MK |
188 | up_read(&zram->init_lock); |
189 | ||
190 | return len; | |
191 | } | |
192 | ||
beca3ec7 SS |
193 | static ssize_t max_comp_streams_store(struct device *dev, |
194 | struct device_attribute *attr, const char *buf, size_t len) | |
195 | { | |
196 | int num; | |
197 | struct zram *zram = dev_to_zram(dev); | |
60a726e3 | 198 | int ret; |
beca3ec7 | 199 | |
60a726e3 MK |
200 | ret = kstrtoint(buf, 0, &num); |
201 | if (ret < 0) | |
202 | return ret; | |
beca3ec7 SS |
203 | if (num < 1) |
204 | return -EINVAL; | |
60a726e3 | 205 | |
beca3ec7 SS |
206 | down_write(&zram->init_lock); |
207 | if (init_done(zram)) { | |
60a726e3 | 208 | if (!zcomp_set_max_streams(zram->comp, num)) { |
fe8eb122 | 209 | pr_info("Cannot change max compression streams\n"); |
60a726e3 MK |
210 | ret = -EINVAL; |
211 | goto out; | |
212 | } | |
beca3ec7 | 213 | } |
60a726e3 | 214 | |
beca3ec7 | 215 | zram->max_comp_streams = num; |
60a726e3 MK |
216 | ret = len; |
217 | out: | |
beca3ec7 | 218 | up_write(&zram->init_lock); |
60a726e3 | 219 | return ret; |
beca3ec7 SS |
220 | } |
221 | ||
e46b8a03 SS |
222 | static ssize_t comp_algorithm_show(struct device *dev, |
223 | struct device_attribute *attr, char *buf) | |
224 | { | |
225 | size_t sz; | |
226 | struct zram *zram = dev_to_zram(dev); | |
227 | ||
228 | down_read(&zram->init_lock); | |
229 | sz = zcomp_available_show(zram->compressor, buf); | |
230 | up_read(&zram->init_lock); | |
231 | ||
232 | return sz; | |
233 | } | |
234 | ||
235 | static ssize_t comp_algorithm_store(struct device *dev, | |
236 | struct device_attribute *attr, const char *buf, size_t len) | |
237 | { | |
238 | struct zram *zram = dev_to_zram(dev); | |
239 | down_write(&zram->init_lock); | |
240 | if (init_done(zram)) { | |
241 | up_write(&zram->init_lock); | |
242 | pr_info("Can't change algorithm for initialized device\n"); | |
243 | return -EBUSY; | |
244 | } | |
245 | strlcpy(zram->compressor, buf, sizeof(zram->compressor)); | |
246 | up_write(&zram->init_lock); | |
247 | return len; | |
248 | } | |
249 | ||
92967471 | 250 | /* flag operations needs meta->tb_lock */ |
8b3cc3ed | 251 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 252 | enum zram_pageflags flag) |
306b0c95 | 253 | { |
d2d5e762 | 254 | return meta->table[index].value & BIT(flag); |
306b0c95 NG |
255 | } |
256 | ||
8b3cc3ed | 257 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 258 | enum zram_pageflags flag) |
306b0c95 | 259 | { |
d2d5e762 | 260 | meta->table[index].value |= BIT(flag); |
306b0c95 NG |
261 | } |
262 | ||
8b3cc3ed | 263 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
f1e3cfff | 264 | enum zram_pageflags flag) |
306b0c95 | 265 | { |
d2d5e762 WY |
266 | meta->table[index].value &= ~BIT(flag); |
267 | } | |
268 | ||
269 | static size_t zram_get_obj_size(struct zram_meta *meta, u32 index) | |
270 | { | |
271 | return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1); | |
272 | } | |
273 | ||
274 | static void zram_set_obj_size(struct zram_meta *meta, | |
275 | u32 index, size_t size) | |
276 | { | |
277 | unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT; | |
278 | ||
279 | meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size; | |
306b0c95 NG |
280 | } |
281 | ||
9b3bb7ab SS |
282 | static inline int is_partial_io(struct bio_vec *bvec) |
283 | { | |
284 | return bvec->bv_len != PAGE_SIZE; | |
285 | } | |
286 | ||
287 | /* | |
288 | * Check if request is within bounds and aligned on zram logical blocks. | |
289 | */ | |
290 | static inline int valid_io_request(struct zram *zram, struct bio *bio) | |
291 | { | |
292 | u64 start, end, bound; | |
a539c72a | 293 | |
9b3bb7ab | 294 | /* unaligned request */ |
4f024f37 KO |
295 | if (unlikely(bio->bi_iter.bi_sector & |
296 | (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1))) | |
9b3bb7ab | 297 | return 0; |
4f024f37 | 298 | if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) |
9b3bb7ab SS |
299 | return 0; |
300 | ||
4f024f37 KO |
301 | start = bio->bi_iter.bi_sector; |
302 | end = start + (bio->bi_iter.bi_size >> SECTOR_SHIFT); | |
9b3bb7ab SS |
303 | bound = zram->disksize >> SECTOR_SHIFT; |
304 | /* out of range range */ | |
75c7caf5 | 305 | if (unlikely(start >= bound || end > bound || start > end)) |
9b3bb7ab SS |
306 | return 0; |
307 | ||
308 | /* I/O request is valid */ | |
309 | return 1; | |
310 | } | |
311 | ||
312 | static void zram_meta_free(struct zram_meta *meta) | |
313 | { | |
314 | zs_destroy_pool(meta->mem_pool); | |
9b3bb7ab SS |
315 | vfree(meta->table); |
316 | kfree(meta); | |
317 | } | |
318 | ||
319 | static struct zram_meta *zram_meta_alloc(u64 disksize) | |
320 | { | |
321 | size_t num_pages; | |
322 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); | |
323 | if (!meta) | |
324 | goto out; | |
325 | ||
9b3bb7ab SS |
326 | num_pages = disksize >> PAGE_SHIFT; |
327 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
328 | if (!meta->table) { | |
329 | pr_err("Error allocating zram address table\n"); | |
b7ca232e | 330 | goto free_meta; |
9b3bb7ab SS |
331 | } |
332 | ||
333 | meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM); | |
334 | if (!meta->mem_pool) { | |
335 | pr_err("Error creating memory pool\n"); | |
336 | goto free_table; | |
337 | } | |
338 | ||
339 | return meta; | |
340 | ||
341 | free_table: | |
342 | vfree(meta->table); | |
9b3bb7ab SS |
343 | free_meta: |
344 | kfree(meta); | |
345 | meta = NULL; | |
346 | out: | |
347 | return meta; | |
348 | } | |
349 | ||
350 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
351 | { | |
352 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
353 | (*index)++; | |
354 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
355 | } | |
356 | ||
306b0c95 NG |
357 | static int page_zero_filled(void *ptr) |
358 | { | |
359 | unsigned int pos; | |
360 | unsigned long *page; | |
361 | ||
362 | page = (unsigned long *)ptr; | |
363 | ||
364 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
365 | if (page[pos]) | |
366 | return 0; | |
367 | } | |
368 | ||
369 | return 1; | |
370 | } | |
371 | ||
9b3bb7ab SS |
372 | static void handle_zero_page(struct bio_vec *bvec) |
373 | { | |
374 | struct page *page = bvec->bv_page; | |
375 | void *user_mem; | |
376 | ||
377 | user_mem = kmap_atomic(page); | |
378 | if (is_partial_io(bvec)) | |
379 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); | |
380 | else | |
381 | clear_page(user_mem); | |
382 | kunmap_atomic(user_mem); | |
383 | ||
384 | flush_dcache_page(page); | |
385 | } | |
386 | ||
d2d5e762 WY |
387 | |
388 | /* | |
389 | * To protect concurrent access to the same index entry, | |
390 | * caller should hold this table index entry's bit_spinlock to | |
391 | * indicate this index entry is accessing. | |
392 | */ | |
f1e3cfff | 393 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 394 | { |
8b3cc3ed MK |
395 | struct zram_meta *meta = zram->meta; |
396 | unsigned long handle = meta->table[index].handle; | |
306b0c95 | 397 | |
fd1a30de | 398 | if (unlikely(!handle)) { |
2e882281 NG |
399 | /* |
400 | * No memory is allocated for zero filled pages. | |
401 | * Simply clear zero page flag. | |
402 | */ | |
8b3cc3ed MK |
403 | if (zram_test_flag(meta, index, ZRAM_ZERO)) { |
404 | zram_clear_flag(meta, index, ZRAM_ZERO); | |
90a7806e | 405 | atomic64_dec(&zram->stats.zero_pages); |
306b0c95 NG |
406 | } |
407 | return; | |
408 | } | |
409 | ||
8b3cc3ed | 410 | zs_free(meta->mem_pool, handle); |
306b0c95 | 411 | |
d2d5e762 WY |
412 | atomic64_sub(zram_get_obj_size(meta, index), |
413 | &zram->stats.compr_data_size); | |
90a7806e | 414 | atomic64_dec(&zram->stats.pages_stored); |
306b0c95 | 415 | |
8b3cc3ed | 416 | meta->table[index].handle = 0; |
d2d5e762 | 417 | zram_set_obj_size(meta, index, 0); |
306b0c95 NG |
418 | } |
419 | ||
37b51fdd | 420 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 421 | { |
b7ca232e | 422 | int ret = 0; |
37b51fdd | 423 | unsigned char *cmem; |
8b3cc3ed | 424 | struct zram_meta *meta = zram->meta; |
92967471 | 425 | unsigned long handle; |
023b409f | 426 | size_t size; |
92967471 | 427 | |
d2d5e762 | 428 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
92967471 | 429 | handle = meta->table[index].handle; |
d2d5e762 | 430 | size = zram_get_obj_size(meta, index); |
306b0c95 | 431 | |
8b3cc3ed | 432 | if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) { |
d2d5e762 | 433 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
42e99bd9 | 434 | clear_page(mem); |
8c921b2b JM |
435 | return 0; |
436 | } | |
306b0c95 | 437 | |
8b3cc3ed | 438 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
92967471 | 439 | if (size == PAGE_SIZE) |
42e99bd9 | 440 | copy_page(mem, cmem); |
37b51fdd | 441 | else |
b7ca232e | 442 | ret = zcomp_decompress(zram->comp, cmem, size, mem); |
8b3cc3ed | 443 | zs_unmap_object(meta->mem_pool, handle); |
d2d5e762 | 444 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
a1dd52af | 445 | |
8c921b2b | 446 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 447 | if (unlikely(ret)) { |
8c921b2b | 448 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); |
8c921b2b | 449 | return ret; |
a1dd52af | 450 | } |
306b0c95 | 451 | |
8c921b2b | 452 | return 0; |
306b0c95 NG |
453 | } |
454 | ||
37b51fdd | 455 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
b627cff3 | 456 | u32 index, int offset) |
924bd88d JM |
457 | { |
458 | int ret; | |
37b51fdd SS |
459 | struct page *page; |
460 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 461 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
462 | page = bvec->bv_page; |
463 | ||
d2d5e762 | 464 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
8b3cc3ed MK |
465 | if (unlikely(!meta->table[index].handle) || |
466 | zram_test_flag(meta, index, ZRAM_ZERO)) { | |
d2d5e762 | 467 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
37b51fdd | 468 | handle_zero_page(bvec); |
924bd88d JM |
469 | return 0; |
470 | } | |
d2d5e762 | 471 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
924bd88d | 472 | |
37b51fdd SS |
473 | if (is_partial_io(bvec)) |
474 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
475 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
476 | ||
477 | user_mem = kmap_atomic(page); | |
478 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
479 | uncmem = user_mem; |
480 | ||
481 | if (!uncmem) { | |
482 | pr_info("Unable to allocate temp memory\n"); | |
483 | ret = -ENOMEM; | |
484 | goto out_cleanup; | |
485 | } | |
924bd88d | 486 | |
37b51fdd | 487 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 488 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 489 | if (unlikely(ret)) |
37b51fdd | 490 | goto out_cleanup; |
924bd88d | 491 | |
37b51fdd SS |
492 | if (is_partial_io(bvec)) |
493 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
494 | bvec->bv_len); | |
495 | ||
496 | flush_dcache_page(page); | |
497 | ret = 0; | |
498 | out_cleanup: | |
499 | kunmap_atomic(user_mem); | |
500 | if (is_partial_io(bvec)) | |
501 | kfree(uncmem); | |
502 | return ret; | |
924bd88d JM |
503 | } |
504 | ||
461a8eee MK |
505 | static inline void update_used_max(struct zram *zram, |
506 | const unsigned long pages) | |
507 | { | |
508 | int old_max, cur_max; | |
509 | ||
510 | old_max = atomic_long_read(&zram->stats.max_used_pages); | |
511 | ||
512 | do { | |
513 | cur_max = old_max; | |
514 | if (pages > cur_max) | |
515 | old_max = atomic_long_cmpxchg( | |
516 | &zram->stats.max_used_pages, cur_max, pages); | |
517 | } while (old_max != cur_max); | |
518 | } | |
519 | ||
924bd88d JM |
520 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, |
521 | int offset) | |
306b0c95 | 522 | { |
397c6066 | 523 | int ret = 0; |
8c921b2b | 524 | size_t clen; |
c2344348 | 525 | unsigned long handle; |
130f315a | 526 | struct page *page; |
924bd88d | 527 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 528 | struct zram_meta *meta = zram->meta; |
b7ca232e | 529 | struct zcomp_strm *zstrm; |
e46e3315 | 530 | bool locked = false; |
461a8eee | 531 | unsigned long alloced_pages; |
306b0c95 | 532 | |
8c921b2b | 533 | page = bvec->bv_page; |
924bd88d JM |
534 | if (is_partial_io(bvec)) { |
535 | /* | |
536 | * This is a partial IO. We need to read the full page | |
537 | * before to write the changes. | |
538 | */ | |
7e5a5104 | 539 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 540 | if (!uncmem) { |
924bd88d JM |
541 | ret = -ENOMEM; |
542 | goto out; | |
543 | } | |
37b51fdd | 544 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 545 | if (ret) |
924bd88d | 546 | goto out; |
924bd88d JM |
547 | } |
548 | ||
b7ca232e | 549 | zstrm = zcomp_strm_find(zram->comp); |
e46e3315 | 550 | locked = true; |
ba82fe2e | 551 | user_mem = kmap_atomic(page); |
924bd88d | 552 | |
397c6066 | 553 | if (is_partial_io(bvec)) { |
924bd88d JM |
554 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
555 | bvec->bv_len); | |
397c6066 NG |
556 | kunmap_atomic(user_mem); |
557 | user_mem = NULL; | |
558 | } else { | |
924bd88d | 559 | uncmem = user_mem; |
397c6066 | 560 | } |
924bd88d JM |
561 | |
562 | if (page_zero_filled(uncmem)) { | |
c4065152 WY |
563 | if (user_mem) |
564 | kunmap_atomic(user_mem); | |
f40ac2ae | 565 | /* Free memory associated with this sector now. */ |
d2d5e762 | 566 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 567 | zram_free_page(zram, index); |
92967471 | 568 | zram_set_flag(meta, index, ZRAM_ZERO); |
d2d5e762 | 569 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 570 | |
90a7806e | 571 | atomic64_inc(&zram->stats.zero_pages); |
924bd88d JM |
572 | ret = 0; |
573 | goto out; | |
8c921b2b | 574 | } |
306b0c95 | 575 | |
b7ca232e | 576 | ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen); |
397c6066 NG |
577 | if (!is_partial_io(bvec)) { |
578 | kunmap_atomic(user_mem); | |
579 | user_mem = NULL; | |
580 | uncmem = NULL; | |
581 | } | |
306b0c95 | 582 | |
b7ca232e | 583 | if (unlikely(ret)) { |
8c921b2b | 584 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 585 | goto out; |
8c921b2b | 586 | } |
b7ca232e | 587 | src = zstrm->buffer; |
c8f2f0db | 588 | if (unlikely(clen > max_zpage_size)) { |
c8f2f0db | 589 | clen = PAGE_SIZE; |
397c6066 NG |
590 | if (is_partial_io(bvec)) |
591 | src = uncmem; | |
c8f2f0db | 592 | } |
a1dd52af | 593 | |
8b3cc3ed | 594 | handle = zs_malloc(meta->mem_pool, clen); |
fd1a30de | 595 | if (!handle) { |
596b3dd4 MR |
596 | pr_info("Error allocating memory for compressed page: %u, size=%zu\n", |
597 | index, clen); | |
924bd88d JM |
598 | ret = -ENOMEM; |
599 | goto out; | |
8c921b2b | 600 | } |
9ada9da9 | 601 | |
461a8eee MK |
602 | alloced_pages = zs_get_total_pages(meta->mem_pool); |
603 | if (zram->limit_pages && alloced_pages > zram->limit_pages) { | |
9ada9da9 MK |
604 | zs_free(meta->mem_pool, handle); |
605 | ret = -ENOMEM; | |
606 | goto out; | |
607 | } | |
608 | ||
461a8eee MK |
609 | update_used_max(zram, alloced_pages); |
610 | ||
8b3cc3ed | 611 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 612 | |
42e99bd9 | 613 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 614 | src = kmap_atomic(page); |
42e99bd9 | 615 | copy_page(cmem, src); |
397c6066 | 616 | kunmap_atomic(src); |
42e99bd9 JL |
617 | } else { |
618 | memcpy(cmem, src, clen); | |
619 | } | |
306b0c95 | 620 | |
b7ca232e SS |
621 | zcomp_strm_release(zram->comp, zstrm); |
622 | locked = false; | |
8b3cc3ed | 623 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 624 | |
f40ac2ae SS |
625 | /* |
626 | * Free memory associated with this sector | |
627 | * before overwriting unused sectors. | |
628 | */ | |
d2d5e762 | 629 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae SS |
630 | zram_free_page(zram, index); |
631 | ||
8b3cc3ed | 632 | meta->table[index].handle = handle; |
d2d5e762 WY |
633 | zram_set_obj_size(meta, index, clen); |
634 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); | |
306b0c95 | 635 | |
8c921b2b | 636 | /* Update stats */ |
90a7806e SS |
637 | atomic64_add(clen, &zram->stats.compr_data_size); |
638 | atomic64_inc(&zram->stats.pages_stored); | |
924bd88d | 639 | out: |
e46e3315 | 640 | if (locked) |
b7ca232e | 641 | zcomp_strm_release(zram->comp, zstrm); |
397c6066 NG |
642 | if (is_partial_io(bvec)) |
643 | kfree(uncmem); | |
924bd88d | 644 | return ret; |
8c921b2b JM |
645 | } |
646 | ||
647 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
b627cff3 | 648 | int offset, int rw) |
8c921b2b | 649 | { |
c5bde238 | 650 | int ret; |
8c921b2b | 651 | |
be257c61 SS |
652 | if (rw == READ) { |
653 | atomic64_inc(&zram->stats.num_reads); | |
b627cff3 | 654 | ret = zram_bvec_read(zram, bvec, index, offset); |
be257c61 SS |
655 | } else { |
656 | atomic64_inc(&zram->stats.num_writes); | |
c5bde238 | 657 | ret = zram_bvec_write(zram, bvec, index, offset); |
be257c61 | 658 | } |
c5bde238 | 659 | |
0cf1e9d6 CY |
660 | if (unlikely(ret)) { |
661 | if (rw == READ) | |
662 | atomic64_inc(&zram->stats.failed_reads); | |
663 | else | |
664 | atomic64_inc(&zram->stats.failed_writes); | |
665 | } | |
666 | ||
c5bde238 | 667 | return ret; |
924bd88d JM |
668 | } |
669 | ||
f4659d8e JK |
670 | /* |
671 | * zram_bio_discard - handler on discard request | |
672 | * @index: physical block index in PAGE_SIZE units | |
673 | * @offset: byte offset within physical block | |
674 | */ | |
675 | static void zram_bio_discard(struct zram *zram, u32 index, | |
676 | int offset, struct bio *bio) | |
677 | { | |
678 | size_t n = bio->bi_iter.bi_size; | |
d2d5e762 | 679 | struct zram_meta *meta = zram->meta; |
f4659d8e JK |
680 | |
681 | /* | |
682 | * zram manages data in physical block size units. Because logical block | |
683 | * size isn't identical with physical block size on some arch, we | |
684 | * could get a discard request pointing to a specific offset within a | |
685 | * certain physical block. Although we can handle this request by | |
686 | * reading that physiclal block and decompressing and partially zeroing | |
687 | * and re-compressing and then re-storing it, this isn't reasonable | |
688 | * because our intent with a discard request is to save memory. So | |
689 | * skipping this logical block is appropriate here. | |
690 | */ | |
691 | if (offset) { | |
38515c73 | 692 | if (n <= (PAGE_SIZE - offset)) |
f4659d8e JK |
693 | return; |
694 | ||
38515c73 | 695 | n -= (PAGE_SIZE - offset); |
f4659d8e JK |
696 | index++; |
697 | } | |
698 | ||
699 | while (n >= PAGE_SIZE) { | |
d2d5e762 | 700 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f4659d8e | 701 | zram_free_page(zram, index); |
d2d5e762 | 702 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
015254da | 703 | atomic64_inc(&zram->stats.notify_free); |
f4659d8e JK |
704 | index++; |
705 | n -= PAGE_SIZE; | |
706 | } | |
707 | } | |
708 | ||
2b86ab9c | 709 | static void zram_reset_device(struct zram *zram, bool reset_capacity) |
924bd88d | 710 | { |
9b3bb7ab SS |
711 | size_t index; |
712 | struct zram_meta *meta; | |
713 | ||
644d4787 | 714 | down_write(&zram->init_lock); |
9ada9da9 MK |
715 | |
716 | zram->limit_pages = 0; | |
717 | ||
be2d1d56 | 718 | if (!init_done(zram)) { |
644d4787 | 719 | up_write(&zram->init_lock); |
9b3bb7ab | 720 | return; |
644d4787 | 721 | } |
9b3bb7ab SS |
722 | |
723 | meta = zram->meta; | |
9b3bb7ab SS |
724 | /* Free all pages that are still in this zram device */ |
725 | for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) { | |
726 | unsigned long handle = meta->table[index].handle; | |
727 | if (!handle) | |
728 | continue; | |
729 | ||
730 | zs_free(meta->mem_pool, handle); | |
731 | } | |
732 | ||
b7ca232e | 733 | zcomp_destroy(zram->comp); |
beca3ec7 SS |
734 | zram->max_comp_streams = 1; |
735 | ||
9b3bb7ab SS |
736 | zram_meta_free(zram->meta); |
737 | zram->meta = NULL; | |
738 | /* Reset stats */ | |
739 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
740 | ||
741 | zram->disksize = 0; | |
b4c5c609 | 742 | if (reset_capacity) |
2b86ab9c | 743 | set_capacity(zram->disk, 0); |
b4c5c609 | 744 | |
644d4787 | 745 | up_write(&zram->init_lock); |
b4c5c609 MK |
746 | |
747 | /* | |
748 | * Revalidate disk out of the init_lock to avoid lockdep splat. | |
749 | * It's okay because disk's capacity is protected by init_lock | |
750 | * so that revalidate_disk always sees up-to-date capacity. | |
751 | */ | |
752 | if (reset_capacity) | |
753 | revalidate_disk(zram->disk); | |
9b3bb7ab SS |
754 | } |
755 | ||
9b3bb7ab SS |
756 | static ssize_t disksize_store(struct device *dev, |
757 | struct device_attribute *attr, const char *buf, size_t len) | |
758 | { | |
759 | u64 disksize; | |
d61f98c7 | 760 | struct zcomp *comp; |
9b3bb7ab SS |
761 | struct zram_meta *meta; |
762 | struct zram *zram = dev_to_zram(dev); | |
fcfa8d95 | 763 | int err; |
9b3bb7ab SS |
764 | |
765 | disksize = memparse(buf, NULL); | |
766 | if (!disksize) | |
767 | return -EINVAL; | |
768 | ||
769 | disksize = PAGE_ALIGN(disksize); | |
770 | meta = zram_meta_alloc(disksize); | |
db5d711e MK |
771 | if (!meta) |
772 | return -ENOMEM; | |
b67d1ec1 | 773 | |
d61f98c7 | 774 | comp = zcomp_create(zram->compressor, zram->max_comp_streams); |
fcfa8d95 | 775 | if (IS_ERR(comp)) { |
d61f98c7 SS |
776 | pr_info("Cannot initialise %s compressing backend\n", |
777 | zram->compressor); | |
fcfa8d95 SS |
778 | err = PTR_ERR(comp); |
779 | goto out_free_meta; | |
d61f98c7 SS |
780 | } |
781 | ||
9b3bb7ab | 782 | down_write(&zram->init_lock); |
be2d1d56 | 783 | if (init_done(zram)) { |
9b3bb7ab | 784 | pr_info("Cannot change disksize for initialized device\n"); |
b7ca232e | 785 | err = -EBUSY; |
fcfa8d95 | 786 | goto out_destroy_comp; |
9b3bb7ab SS |
787 | } |
788 | ||
b67d1ec1 | 789 | zram->meta = meta; |
d61f98c7 | 790 | zram->comp = comp; |
9b3bb7ab SS |
791 | zram->disksize = disksize; |
792 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
9b3bb7ab | 793 | up_write(&zram->init_lock); |
b4c5c609 MK |
794 | |
795 | /* | |
796 | * Revalidate disk out of the init_lock to avoid lockdep splat. | |
797 | * It's okay because disk's capacity is protected by init_lock | |
798 | * so that revalidate_disk always sees up-to-date capacity. | |
799 | */ | |
800 | revalidate_disk(zram->disk); | |
801 | ||
9b3bb7ab | 802 | return len; |
b7ca232e | 803 | |
fcfa8d95 SS |
804 | out_destroy_comp: |
805 | up_write(&zram->init_lock); | |
806 | zcomp_destroy(comp); | |
807 | out_free_meta: | |
b7ca232e SS |
808 | zram_meta_free(meta); |
809 | return err; | |
9b3bb7ab SS |
810 | } |
811 | ||
812 | static ssize_t reset_store(struct device *dev, | |
813 | struct device_attribute *attr, const char *buf, size_t len) | |
814 | { | |
815 | int ret; | |
816 | unsigned short do_reset; | |
817 | struct zram *zram; | |
818 | struct block_device *bdev; | |
819 | ||
820 | zram = dev_to_zram(dev); | |
821 | bdev = bdget_disk(zram->disk, 0); | |
822 | ||
46a51c80 RK |
823 | if (!bdev) |
824 | return -ENOMEM; | |
825 | ||
9b3bb7ab | 826 | /* Do not reset an active device! */ |
1b672224 RK |
827 | if (bdev->bd_holders) { |
828 | ret = -EBUSY; | |
829 | goto out; | |
830 | } | |
9b3bb7ab SS |
831 | |
832 | ret = kstrtou16(buf, 10, &do_reset); | |
833 | if (ret) | |
1b672224 | 834 | goto out; |
9b3bb7ab | 835 | |
1b672224 RK |
836 | if (!do_reset) { |
837 | ret = -EINVAL; | |
838 | goto out; | |
839 | } | |
9b3bb7ab SS |
840 | |
841 | /* Make sure all pending I/O is finished */ | |
46a51c80 | 842 | fsync_bdev(bdev); |
1b672224 | 843 | bdput(bdev); |
9b3bb7ab | 844 | |
2b86ab9c | 845 | zram_reset_device(zram, true); |
9b3bb7ab | 846 | return len; |
1b672224 RK |
847 | |
848 | out: | |
849 | bdput(bdev); | |
850 | return ret; | |
8c921b2b JM |
851 | } |
852 | ||
be257c61 | 853 | static void __zram_make_request(struct zram *zram, struct bio *bio) |
8c921b2b | 854 | { |
b627cff3 | 855 | int offset, rw; |
8c921b2b | 856 | u32 index; |
7988613b KO |
857 | struct bio_vec bvec; |
858 | struct bvec_iter iter; | |
8c921b2b | 859 | |
4f024f37 KO |
860 | index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT; |
861 | offset = (bio->bi_iter.bi_sector & | |
862 | (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; | |
8c921b2b | 863 | |
f4659d8e JK |
864 | if (unlikely(bio->bi_rw & REQ_DISCARD)) { |
865 | zram_bio_discard(zram, index, offset, bio); | |
866 | bio_endio(bio, 0); | |
867 | return; | |
868 | } | |
869 | ||
b627cff3 | 870 | rw = bio_data_dir(bio); |
7988613b | 871 | bio_for_each_segment(bvec, bio, iter) { |
924bd88d JM |
872 | int max_transfer_size = PAGE_SIZE - offset; |
873 | ||
7988613b | 874 | if (bvec.bv_len > max_transfer_size) { |
924bd88d JM |
875 | /* |
876 | * zram_bvec_rw() can only make operation on a single | |
877 | * zram page. Split the bio vector. | |
878 | */ | |
879 | struct bio_vec bv; | |
880 | ||
7988613b | 881 | bv.bv_page = bvec.bv_page; |
924bd88d | 882 | bv.bv_len = max_transfer_size; |
7988613b | 883 | bv.bv_offset = bvec.bv_offset; |
924bd88d | 884 | |
b627cff3 | 885 | if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0) |
924bd88d JM |
886 | goto out; |
887 | ||
7988613b | 888 | bv.bv_len = bvec.bv_len - max_transfer_size; |
924bd88d | 889 | bv.bv_offset += max_transfer_size; |
b627cff3 | 890 | if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0) |
924bd88d JM |
891 | goto out; |
892 | } else | |
b627cff3 | 893 | if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0) |
924bd88d JM |
894 | goto out; |
895 | ||
7988613b | 896 | update_position(&index, &offset, &bvec); |
a1dd52af | 897 | } |
306b0c95 NG |
898 | |
899 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
900 | bio_endio(bio, 0); | |
7d7854b4 | 901 | return; |
306b0c95 NG |
902 | |
903 | out: | |
306b0c95 | 904 | bio_io_error(bio); |
306b0c95 NG |
905 | } |
906 | ||
306b0c95 | 907 | /* |
f1e3cfff | 908 | * Handler function for all zram I/O requests. |
306b0c95 | 909 | */ |
5a7bbad2 | 910 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 911 | { |
f1e3cfff | 912 | struct zram *zram = queue->queuedata; |
306b0c95 | 913 | |
0900beae | 914 | down_read(&zram->init_lock); |
be2d1d56 | 915 | if (unlikely(!init_done(zram))) |
3de738cd | 916 | goto error; |
0900beae | 917 | |
f1e3cfff | 918 | if (!valid_io_request(zram, bio)) { |
da5cc7d3 | 919 | atomic64_inc(&zram->stats.invalid_io); |
3de738cd | 920 | goto error; |
6642a67c JM |
921 | } |
922 | ||
be257c61 | 923 | __zram_make_request(zram, bio); |
0900beae | 924 | up_read(&zram->init_lock); |
306b0c95 | 925 | |
b4fdcb02 | 926 | return; |
0900beae | 927 | |
0900beae | 928 | error: |
3de738cd | 929 | up_read(&zram->init_lock); |
0900beae | 930 | bio_io_error(bio); |
306b0c95 NG |
931 | } |
932 | ||
2ccbec05 NG |
933 | static void zram_slot_free_notify(struct block_device *bdev, |
934 | unsigned long index) | |
107c161b | 935 | { |
f1e3cfff | 936 | struct zram *zram; |
f614a9f4 | 937 | struct zram_meta *meta; |
107c161b | 938 | |
f1e3cfff | 939 | zram = bdev->bd_disk->private_data; |
f614a9f4 | 940 | meta = zram->meta; |
a0c516cb | 941 | |
d2d5e762 | 942 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 943 | zram_free_page(zram, index); |
d2d5e762 | 944 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 945 | atomic64_inc(&zram->stats.notify_free); |
107c161b NG |
946 | } |
947 | ||
f1e3cfff | 948 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 949 | .swap_slot_free_notify = zram_slot_free_notify, |
107c161b | 950 | .owner = THIS_MODULE |
306b0c95 NG |
951 | }; |
952 | ||
9b3bb7ab SS |
953 | static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR, |
954 | disksize_show, disksize_store); | |
955 | static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL); | |
956 | static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store); | |
9b3bb7ab | 957 | static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL); |
9b3bb7ab | 958 | static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL); |
9ada9da9 MK |
959 | static DEVICE_ATTR(mem_limit, S_IRUGO | S_IWUSR, mem_limit_show, |
960 | mem_limit_store); | |
461a8eee MK |
961 | static DEVICE_ATTR(mem_used_max, S_IRUGO | S_IWUSR, mem_used_max_show, |
962 | mem_used_max_store); | |
beca3ec7 SS |
963 | static DEVICE_ATTR(max_comp_streams, S_IRUGO | S_IWUSR, |
964 | max_comp_streams_show, max_comp_streams_store); | |
e46b8a03 SS |
965 | static DEVICE_ATTR(comp_algorithm, S_IRUGO | S_IWUSR, |
966 | comp_algorithm_show, comp_algorithm_store); | |
9b3bb7ab | 967 | |
a68eb3b6 SS |
968 | ZRAM_ATTR_RO(num_reads); |
969 | ZRAM_ATTR_RO(num_writes); | |
64447249 SS |
970 | ZRAM_ATTR_RO(failed_reads); |
971 | ZRAM_ATTR_RO(failed_writes); | |
a68eb3b6 SS |
972 | ZRAM_ATTR_RO(invalid_io); |
973 | ZRAM_ATTR_RO(notify_free); | |
974 | ZRAM_ATTR_RO(zero_pages); | |
975 | ZRAM_ATTR_RO(compr_data_size); | |
976 | ||
9b3bb7ab SS |
977 | static struct attribute *zram_disk_attrs[] = { |
978 | &dev_attr_disksize.attr, | |
979 | &dev_attr_initstate.attr, | |
980 | &dev_attr_reset.attr, | |
981 | &dev_attr_num_reads.attr, | |
982 | &dev_attr_num_writes.attr, | |
64447249 SS |
983 | &dev_attr_failed_reads.attr, |
984 | &dev_attr_failed_writes.attr, | |
9b3bb7ab SS |
985 | &dev_attr_invalid_io.attr, |
986 | &dev_attr_notify_free.attr, | |
987 | &dev_attr_zero_pages.attr, | |
988 | &dev_attr_orig_data_size.attr, | |
989 | &dev_attr_compr_data_size.attr, | |
990 | &dev_attr_mem_used_total.attr, | |
9ada9da9 | 991 | &dev_attr_mem_limit.attr, |
461a8eee | 992 | &dev_attr_mem_used_max.attr, |
beca3ec7 | 993 | &dev_attr_max_comp_streams.attr, |
e46b8a03 | 994 | &dev_attr_comp_algorithm.attr, |
9b3bb7ab SS |
995 | NULL, |
996 | }; | |
997 | ||
998 | static struct attribute_group zram_disk_attr_group = { | |
999 | .attrs = zram_disk_attrs, | |
1000 | }; | |
1001 | ||
f1e3cfff | 1002 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 1003 | { |
39a9b8ac | 1004 | int ret = -ENOMEM; |
de1a21a0 | 1005 | |
0900beae | 1006 | init_rwsem(&zram->init_lock); |
306b0c95 | 1007 | |
f1e3cfff NG |
1008 | zram->queue = blk_alloc_queue(GFP_KERNEL); |
1009 | if (!zram->queue) { | |
306b0c95 NG |
1010 | pr_err("Error allocating disk queue for device %d\n", |
1011 | device_id); | |
de1a21a0 | 1012 | goto out; |
306b0c95 NG |
1013 | } |
1014 | ||
f1e3cfff NG |
1015 | blk_queue_make_request(zram->queue, zram_make_request); |
1016 | zram->queue->queuedata = zram; | |
306b0c95 NG |
1017 | |
1018 | /* gendisk structure */ | |
f1e3cfff NG |
1019 | zram->disk = alloc_disk(1); |
1020 | if (!zram->disk) { | |
94b8435f | 1021 | pr_warn("Error allocating disk structure for device %d\n", |
306b0c95 | 1022 | device_id); |
39a9b8ac | 1023 | goto out_free_queue; |
306b0c95 NG |
1024 | } |
1025 | ||
f1e3cfff NG |
1026 | zram->disk->major = zram_major; |
1027 | zram->disk->first_minor = device_id; | |
1028 | zram->disk->fops = &zram_devops; | |
1029 | zram->disk->queue = zram->queue; | |
1030 | zram->disk->private_data = zram; | |
1031 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 1032 | |
33863c21 | 1033 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 1034 | set_capacity(zram->disk, 0); |
b67d1ec1 SS |
1035 | /* zram devices sort of resembles non-rotational disks */ |
1036 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
b277da0a | 1037 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue); |
a1dd52af NG |
1038 | /* |
1039 | * To ensure that we always get PAGE_SIZE aligned | |
1040 | * and n*PAGE_SIZED sized I/O requests. | |
1041 | */ | |
f1e3cfff | 1042 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
1043 | blk_queue_logical_block_size(zram->disk->queue, |
1044 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
1045 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
1046 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
f4659d8e JK |
1047 | zram->disk->queue->limits.discard_granularity = PAGE_SIZE; |
1048 | zram->disk->queue->limits.max_discard_sectors = UINT_MAX; | |
1049 | /* | |
1050 | * zram_bio_discard() will clear all logical blocks if logical block | |
1051 | * size is identical with physical block size(PAGE_SIZE). But if it is | |
1052 | * different, we will skip discarding some parts of logical blocks in | |
1053 | * the part of the request range which isn't aligned to physical block | |
1054 | * size. So we can't ensure that all discarded logical blocks are | |
1055 | * zeroed. | |
1056 | */ | |
1057 | if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE) | |
1058 | zram->disk->queue->limits.discard_zeroes_data = 1; | |
1059 | else | |
1060 | zram->disk->queue->limits.discard_zeroes_data = 0; | |
1061 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue); | |
5d83d5a0 | 1062 | |
f1e3cfff | 1063 | add_disk(zram->disk); |
306b0c95 | 1064 | |
33863c21 NG |
1065 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
1066 | &zram_disk_attr_group); | |
1067 | if (ret < 0) { | |
94b8435f | 1068 | pr_warn("Error creating sysfs group"); |
39a9b8ac | 1069 | goto out_free_disk; |
33863c21 | 1070 | } |
e46b8a03 | 1071 | strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor)); |
be2d1d56 | 1072 | zram->meta = NULL; |
beca3ec7 | 1073 | zram->max_comp_streams = 1; |
39a9b8ac | 1074 | return 0; |
de1a21a0 | 1075 | |
39a9b8ac JL |
1076 | out_free_disk: |
1077 | del_gendisk(zram->disk); | |
1078 | put_disk(zram->disk); | |
1079 | out_free_queue: | |
1080 | blk_cleanup_queue(zram->queue); | |
de1a21a0 NG |
1081 | out: |
1082 | return ret; | |
306b0c95 NG |
1083 | } |
1084 | ||
f1e3cfff | 1085 | static void destroy_device(struct zram *zram) |
306b0c95 | 1086 | { |
33863c21 NG |
1087 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, |
1088 | &zram_disk_attr_group); | |
33863c21 | 1089 | |
59d3fe54 RK |
1090 | del_gendisk(zram->disk); |
1091 | put_disk(zram->disk); | |
306b0c95 | 1092 | |
59d3fe54 | 1093 | blk_cleanup_queue(zram->queue); |
306b0c95 NG |
1094 | } |
1095 | ||
f1e3cfff | 1096 | static int __init zram_init(void) |
306b0c95 | 1097 | { |
de1a21a0 | 1098 | int ret, dev_id; |
306b0c95 | 1099 | |
5fa5a901 | 1100 | if (num_devices > max_num_devices) { |
94b8435f | 1101 | pr_warn("Invalid value for num_devices: %u\n", |
5fa5a901 | 1102 | num_devices); |
de1a21a0 NG |
1103 | ret = -EINVAL; |
1104 | goto out; | |
306b0c95 NG |
1105 | } |
1106 | ||
f1e3cfff NG |
1107 | zram_major = register_blkdev(0, "zram"); |
1108 | if (zram_major <= 0) { | |
94b8435f | 1109 | pr_warn("Unable to get major number\n"); |
de1a21a0 NG |
1110 | ret = -EBUSY; |
1111 | goto out; | |
306b0c95 NG |
1112 | } |
1113 | ||
306b0c95 | 1114 | /* Allocate the device array and initialize each one */ |
5fa5a901 | 1115 | zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL); |
43801f6e | 1116 | if (!zram_devices) { |
de1a21a0 NG |
1117 | ret = -ENOMEM; |
1118 | goto unregister; | |
1119 | } | |
306b0c95 | 1120 | |
5fa5a901 | 1121 | for (dev_id = 0; dev_id < num_devices; dev_id++) { |
43801f6e | 1122 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 1123 | if (ret) |
3bf040c7 | 1124 | goto free_devices; |
de1a21a0 NG |
1125 | } |
1126 | ||
ca3d70bd DB |
1127 | pr_info("Created %u device(s) ...\n", num_devices); |
1128 | ||
306b0c95 | 1129 | return 0; |
de1a21a0 | 1130 | |
3bf040c7 | 1131 | free_devices: |
de1a21a0 | 1132 | while (dev_id) |
43801f6e NW |
1133 | destroy_device(&zram_devices[--dev_id]); |
1134 | kfree(zram_devices); | |
de1a21a0 | 1135 | unregister: |
f1e3cfff | 1136 | unregister_blkdev(zram_major, "zram"); |
de1a21a0 | 1137 | out: |
306b0c95 NG |
1138 | return ret; |
1139 | } | |
1140 | ||
f1e3cfff | 1141 | static void __exit zram_exit(void) |
306b0c95 NG |
1142 | { |
1143 | int i; | |
f1e3cfff | 1144 | struct zram *zram; |
306b0c95 | 1145 | |
5fa5a901 | 1146 | for (i = 0; i < num_devices; i++) { |
43801f6e | 1147 | zram = &zram_devices[i]; |
306b0c95 | 1148 | |
f1e3cfff | 1149 | destroy_device(zram); |
2b86ab9c MK |
1150 | /* |
1151 | * Shouldn't access zram->disk after destroy_device | |
1152 | * because destroy_device already released zram->disk. | |
1153 | */ | |
1154 | zram_reset_device(zram, false); | |
306b0c95 NG |
1155 | } |
1156 | ||
f1e3cfff | 1157 | unregister_blkdev(zram_major, "zram"); |
306b0c95 | 1158 | |
43801f6e | 1159 | kfree(zram_devices); |
306b0c95 NG |
1160 | pr_debug("Cleanup done!\n"); |
1161 | } | |
1162 | ||
f1e3cfff NG |
1163 | module_init(zram_init); |
1164 | module_exit(zram_exit); | |
306b0c95 | 1165 | |
9b3bb7ab SS |
1166 | module_param(num_devices, uint, 0); |
1167 | MODULE_PARM_DESC(num_devices, "Number of zram devices"); | |
1168 | ||
306b0c95 NG |
1169 | MODULE_LICENSE("Dual BSD/GPL"); |
1170 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 1171 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |