]>
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 | ||
18 | #include <linux/module.h> | |
19 | #include <linux/kernel.h> | |
8946a086 | 20 | #include <linux/bio.h> |
306b0c95 NG |
21 | #include <linux/bitops.h> |
22 | #include <linux/blkdev.h> | |
23 | #include <linux/buffer_head.h> | |
24 | #include <linux/device.h> | |
25 | #include <linux/genhd.h> | |
26 | #include <linux/highmem.h> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
b09ab054 | 28 | #include <linux/backing-dev.h> |
306b0c95 | 29 | #include <linux/string.h> |
306b0c95 | 30 | #include <linux/vmalloc.h> |
fcfa8d95 | 31 | #include <linux/err.h> |
85508ec6 | 32 | #include <linux/idr.h> |
6566d1a3 | 33 | #include <linux/sysfs.h> |
1dd6c834 | 34 | #include <linux/cpuhotplug.h> |
306b0c95 | 35 | |
16a4bfb9 | 36 | #include "zram_drv.h" |
306b0c95 | 37 | |
85508ec6 | 38 | static DEFINE_IDR(zram_index_idr); |
6566d1a3 SS |
39 | /* idr index must be protected */ |
40 | static DEFINE_MUTEX(zram_index_mutex); | |
41 | ||
f1e3cfff | 42 | static int zram_major; |
b7ca232e | 43 | static const char *default_compressor = "lzo"; |
306b0c95 | 44 | |
306b0c95 | 45 | /* Module params (documentation at end) */ |
ca3d70bd | 46 | static unsigned int num_devices = 1; |
33863c21 | 47 | |
08eee69f | 48 | static inline bool init_done(struct zram *zram) |
be2d1d56 | 49 | { |
08eee69f | 50 | return zram->disksize; |
be2d1d56 SS |
51 | } |
52 | ||
9b3bb7ab SS |
53 | static inline struct zram *dev_to_zram(struct device *dev) |
54 | { | |
55 | return (struct zram *)dev_to_disk(dev)->private_data; | |
56 | } | |
57 | ||
b31177f2 | 58 | /* flag operations require table entry bit_spin_lock() being held */ |
522698d7 SS |
59 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
60 | enum zram_pageflags flag) | |
99ebbd30 | 61 | { |
522698d7 SS |
62 | return meta->table[index].value & BIT(flag); |
63 | } | |
99ebbd30 | 64 | |
522698d7 SS |
65 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
66 | enum zram_pageflags flag) | |
67 | { | |
68 | meta->table[index].value |= BIT(flag); | |
69 | } | |
99ebbd30 | 70 | |
522698d7 SS |
71 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
72 | enum zram_pageflags flag) | |
73 | { | |
74 | meta->table[index].value &= ~BIT(flag); | |
75 | } | |
99ebbd30 | 76 | |
8e19d540 | 77 | static inline void zram_set_element(struct zram_meta *meta, u32 index, |
78 | unsigned long element) | |
79 | { | |
80 | meta->table[index].element = element; | |
81 | } | |
82 | ||
83 | static inline void zram_clear_element(struct zram_meta *meta, u32 index) | |
84 | { | |
85 | meta->table[index].element = 0; | |
86 | } | |
87 | ||
522698d7 SS |
88 | static size_t zram_get_obj_size(struct zram_meta *meta, u32 index) |
89 | { | |
90 | return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1); | |
99ebbd30 AM |
91 | } |
92 | ||
522698d7 SS |
93 | static void zram_set_obj_size(struct zram_meta *meta, |
94 | u32 index, size_t size) | |
9b3bb7ab | 95 | { |
522698d7 | 96 | unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT; |
9b3bb7ab | 97 | |
522698d7 SS |
98 | meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size; |
99 | } | |
100 | ||
1c53e0d2 | 101 | static inline bool is_partial_io(struct bio_vec *bvec) |
522698d7 SS |
102 | { |
103 | return bvec->bv_len != PAGE_SIZE; | |
104 | } | |
105 | ||
b09ab054 MK |
106 | static void zram_revalidate_disk(struct zram *zram) |
107 | { | |
108 | revalidate_disk(zram->disk); | |
109 | /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */ | |
e1735496 | 110 | zram->disk->queue->backing_dev_info->capabilities |= |
b09ab054 MK |
111 | BDI_CAP_STABLE_WRITES; |
112 | } | |
113 | ||
522698d7 SS |
114 | /* |
115 | * Check if request is within bounds and aligned on zram logical blocks. | |
116 | */ | |
1c53e0d2 | 117 | static inline bool valid_io_request(struct zram *zram, |
522698d7 SS |
118 | sector_t start, unsigned int size) |
119 | { | |
120 | u64 end, bound; | |
121 | ||
122 | /* unaligned request */ | |
123 | if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1))) | |
1c53e0d2 | 124 | return false; |
522698d7 | 125 | if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) |
1c53e0d2 | 126 | return false; |
522698d7 SS |
127 | |
128 | end = start + (size >> SECTOR_SHIFT); | |
129 | bound = zram->disksize >> SECTOR_SHIFT; | |
130 | /* out of range range */ | |
131 | if (unlikely(start >= bound || end > bound || start > end)) | |
1c53e0d2 | 132 | return false; |
522698d7 SS |
133 | |
134 | /* I/O request is valid */ | |
1c53e0d2 | 135 | return true; |
522698d7 SS |
136 | } |
137 | ||
138 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
139 | { | |
140 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
141 | (*index)++; | |
142 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
143 | } | |
144 | ||
145 | static inline void update_used_max(struct zram *zram, | |
146 | const unsigned long pages) | |
147 | { | |
148 | unsigned long old_max, cur_max; | |
149 | ||
150 | old_max = atomic_long_read(&zram->stats.max_used_pages); | |
151 | ||
152 | do { | |
153 | cur_max = old_max; | |
154 | if (pages > cur_max) | |
155 | old_max = atomic_long_cmpxchg( | |
156 | &zram->stats.max_used_pages, cur_max, pages); | |
157 | } while (old_max != cur_max); | |
158 | } | |
159 | ||
8e19d540 | 160 | static inline void zram_fill_page(char *ptr, unsigned long len, |
161 | unsigned long value) | |
162 | { | |
163 | int i; | |
164 | unsigned long *page = (unsigned long *)ptr; | |
165 | ||
166 | WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long))); | |
167 | ||
168 | if (likely(value == 0)) { | |
169 | memset(ptr, 0, len); | |
170 | } else { | |
171 | for (i = 0; i < len / sizeof(*page); i++) | |
172 | page[i] = value; | |
173 | } | |
174 | } | |
175 | ||
176 | static bool page_same_filled(void *ptr, unsigned long *element) | |
522698d7 SS |
177 | { |
178 | unsigned int pos; | |
179 | unsigned long *page; | |
180 | ||
181 | page = (unsigned long *)ptr; | |
182 | ||
8e19d540 | 183 | for (pos = 0; pos < PAGE_SIZE / sizeof(*page) - 1; pos++) { |
184 | if (page[pos] != page[pos + 1]) | |
1c53e0d2 | 185 | return false; |
522698d7 SS |
186 | } |
187 | ||
8e19d540 | 188 | *element = page[pos]; |
189 | ||
1c53e0d2 | 190 | return true; |
522698d7 SS |
191 | } |
192 | ||
8e19d540 | 193 | static void handle_same_page(struct bio_vec *bvec, unsigned long element) |
522698d7 SS |
194 | { |
195 | struct page *page = bvec->bv_page; | |
196 | void *user_mem; | |
197 | ||
198 | user_mem = kmap_atomic(page); | |
8e19d540 | 199 | zram_fill_page(user_mem + bvec->bv_offset, bvec->bv_len, element); |
522698d7 SS |
200 | kunmap_atomic(user_mem); |
201 | ||
202 | flush_dcache_page(page); | |
9b3bb7ab SS |
203 | } |
204 | ||
205 | static ssize_t initstate_show(struct device *dev, | |
206 | struct device_attribute *attr, char *buf) | |
207 | { | |
a68eb3b6 | 208 | u32 val; |
9b3bb7ab SS |
209 | struct zram *zram = dev_to_zram(dev); |
210 | ||
a68eb3b6 SS |
211 | down_read(&zram->init_lock); |
212 | val = init_done(zram); | |
213 | up_read(&zram->init_lock); | |
9b3bb7ab | 214 | |
56b4e8cb | 215 | return scnprintf(buf, PAGE_SIZE, "%u\n", val); |
9b3bb7ab SS |
216 | } |
217 | ||
522698d7 SS |
218 | static ssize_t disksize_show(struct device *dev, |
219 | struct device_attribute *attr, char *buf) | |
220 | { | |
221 | struct zram *zram = dev_to_zram(dev); | |
222 | ||
223 | return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize); | |
224 | } | |
225 | ||
9ada9da9 MK |
226 | static ssize_t mem_limit_store(struct device *dev, |
227 | struct device_attribute *attr, const char *buf, size_t len) | |
228 | { | |
229 | u64 limit; | |
230 | char *tmp; | |
231 | struct zram *zram = dev_to_zram(dev); | |
232 | ||
233 | limit = memparse(buf, &tmp); | |
234 | if (buf == tmp) /* no chars parsed, invalid input */ | |
235 | return -EINVAL; | |
236 | ||
237 | down_write(&zram->init_lock); | |
238 | zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT; | |
239 | up_write(&zram->init_lock); | |
240 | ||
241 | return len; | |
242 | } | |
243 | ||
461a8eee MK |
244 | static ssize_t mem_used_max_store(struct device *dev, |
245 | struct device_attribute *attr, const char *buf, size_t len) | |
246 | { | |
247 | int err; | |
248 | unsigned long val; | |
249 | struct zram *zram = dev_to_zram(dev); | |
461a8eee MK |
250 | |
251 | err = kstrtoul(buf, 10, &val); | |
252 | if (err || val != 0) | |
253 | return -EINVAL; | |
254 | ||
255 | down_read(&zram->init_lock); | |
5a99e95b WY |
256 | if (init_done(zram)) { |
257 | struct zram_meta *meta = zram->meta; | |
461a8eee MK |
258 | atomic_long_set(&zram->stats.max_used_pages, |
259 | zs_get_total_pages(meta->mem_pool)); | |
5a99e95b | 260 | } |
461a8eee MK |
261 | up_read(&zram->init_lock); |
262 | ||
263 | return len; | |
264 | } | |
265 | ||
43209ea2 SS |
266 | /* |
267 | * We switched to per-cpu streams and this attr is not needed anymore. | |
268 | * However, we will keep it around for some time, because: | |
269 | * a) we may revert per-cpu streams in the future | |
270 | * b) it's visible to user space and we need to follow our 2 years | |
271 | * retirement rule; but we already have a number of 'soon to be | |
272 | * altered' attrs, so max_comp_streams need to wait for the next | |
273 | * layoff cycle. | |
274 | */ | |
522698d7 SS |
275 | static ssize_t max_comp_streams_show(struct device *dev, |
276 | struct device_attribute *attr, char *buf) | |
277 | { | |
43209ea2 | 278 | return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus()); |
522698d7 SS |
279 | } |
280 | ||
beca3ec7 SS |
281 | static ssize_t max_comp_streams_store(struct device *dev, |
282 | struct device_attribute *attr, const char *buf, size_t len) | |
283 | { | |
43209ea2 | 284 | return len; |
beca3ec7 SS |
285 | } |
286 | ||
e46b8a03 SS |
287 | static ssize_t comp_algorithm_show(struct device *dev, |
288 | struct device_attribute *attr, char *buf) | |
289 | { | |
290 | size_t sz; | |
291 | struct zram *zram = dev_to_zram(dev); | |
292 | ||
293 | down_read(&zram->init_lock); | |
294 | sz = zcomp_available_show(zram->compressor, buf); | |
295 | up_read(&zram->init_lock); | |
296 | ||
297 | return sz; | |
298 | } | |
299 | ||
300 | static ssize_t comp_algorithm_store(struct device *dev, | |
301 | struct device_attribute *attr, const char *buf, size_t len) | |
302 | { | |
303 | struct zram *zram = dev_to_zram(dev); | |
415403be | 304 | char compressor[CRYPTO_MAX_ALG_NAME]; |
4bbacd51 SS |
305 | size_t sz; |
306 | ||
415403be SS |
307 | strlcpy(compressor, buf, sizeof(compressor)); |
308 | /* ignore trailing newline */ | |
309 | sz = strlen(compressor); | |
310 | if (sz > 0 && compressor[sz - 1] == '\n') | |
311 | compressor[sz - 1] = 0x00; | |
312 | ||
313 | if (!zcomp_available_algorithm(compressor)) | |
1d5b43bf LH |
314 | return -EINVAL; |
315 | ||
e46b8a03 SS |
316 | down_write(&zram->init_lock); |
317 | if (init_done(zram)) { | |
318 | up_write(&zram->init_lock); | |
319 | pr_info("Can't change algorithm for initialized device\n"); | |
320 | return -EBUSY; | |
321 | } | |
4bbacd51 | 322 | |
415403be | 323 | strlcpy(zram->compressor, compressor, sizeof(compressor)); |
e46b8a03 SS |
324 | up_write(&zram->init_lock); |
325 | return len; | |
326 | } | |
327 | ||
522698d7 SS |
328 | static ssize_t compact_store(struct device *dev, |
329 | struct device_attribute *attr, const char *buf, size_t len) | |
306b0c95 | 330 | { |
522698d7 SS |
331 | struct zram *zram = dev_to_zram(dev); |
332 | struct zram_meta *meta; | |
306b0c95 | 333 | |
522698d7 SS |
334 | down_read(&zram->init_lock); |
335 | if (!init_done(zram)) { | |
336 | up_read(&zram->init_lock); | |
337 | return -EINVAL; | |
338 | } | |
306b0c95 | 339 | |
522698d7 | 340 | meta = zram->meta; |
7d3f3938 | 341 | zs_compact(meta->mem_pool); |
522698d7 | 342 | up_read(&zram->init_lock); |
d2d5e762 | 343 | |
522698d7 | 344 | return len; |
d2d5e762 WY |
345 | } |
346 | ||
522698d7 SS |
347 | static ssize_t io_stat_show(struct device *dev, |
348 | struct device_attribute *attr, char *buf) | |
d2d5e762 | 349 | { |
522698d7 SS |
350 | struct zram *zram = dev_to_zram(dev); |
351 | ssize_t ret; | |
d2d5e762 | 352 | |
522698d7 SS |
353 | down_read(&zram->init_lock); |
354 | ret = scnprintf(buf, PAGE_SIZE, | |
355 | "%8llu %8llu %8llu %8llu\n", | |
356 | (u64)atomic64_read(&zram->stats.failed_reads), | |
357 | (u64)atomic64_read(&zram->stats.failed_writes), | |
358 | (u64)atomic64_read(&zram->stats.invalid_io), | |
359 | (u64)atomic64_read(&zram->stats.notify_free)); | |
360 | up_read(&zram->init_lock); | |
306b0c95 | 361 | |
522698d7 | 362 | return ret; |
9b3bb7ab SS |
363 | } |
364 | ||
522698d7 SS |
365 | static ssize_t mm_stat_show(struct device *dev, |
366 | struct device_attribute *attr, char *buf) | |
9b3bb7ab | 367 | { |
522698d7 | 368 | struct zram *zram = dev_to_zram(dev); |
7d3f3938 | 369 | struct zs_pool_stats pool_stats; |
522698d7 SS |
370 | u64 orig_size, mem_used = 0; |
371 | long max_used; | |
372 | ssize_t ret; | |
a539c72a | 373 | |
7d3f3938 SS |
374 | memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats)); |
375 | ||
522698d7 | 376 | down_read(&zram->init_lock); |
7d3f3938 | 377 | if (init_done(zram)) { |
522698d7 | 378 | mem_used = zs_get_total_pages(zram->meta->mem_pool); |
7d3f3938 SS |
379 | zs_pool_stats(zram->meta->mem_pool, &pool_stats); |
380 | } | |
9b3bb7ab | 381 | |
522698d7 SS |
382 | orig_size = atomic64_read(&zram->stats.pages_stored); |
383 | max_used = atomic_long_read(&zram->stats.max_used_pages); | |
9b3bb7ab | 384 | |
522698d7 | 385 | ret = scnprintf(buf, PAGE_SIZE, |
7d3f3938 | 386 | "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n", |
522698d7 SS |
387 | orig_size << PAGE_SHIFT, |
388 | (u64)atomic64_read(&zram->stats.compr_data_size), | |
389 | mem_used << PAGE_SHIFT, | |
390 | zram->limit_pages << PAGE_SHIFT, | |
391 | max_used << PAGE_SHIFT, | |
8e19d540 | 392 | (u64)atomic64_read(&zram->stats.same_pages), |
860c707d | 393 | pool_stats.pages_compacted); |
522698d7 | 394 | up_read(&zram->init_lock); |
9b3bb7ab | 395 | |
522698d7 SS |
396 | return ret; |
397 | } | |
398 | ||
623e47fc SS |
399 | static ssize_t debug_stat_show(struct device *dev, |
400 | struct device_attribute *attr, char *buf) | |
401 | { | |
402 | int version = 1; | |
403 | struct zram *zram = dev_to_zram(dev); | |
404 | ssize_t ret; | |
405 | ||
406 | down_read(&zram->init_lock); | |
407 | ret = scnprintf(buf, PAGE_SIZE, | |
408 | "version: %d\n%8llu\n", | |
409 | version, | |
410 | (u64)atomic64_read(&zram->stats.writestall)); | |
411 | up_read(&zram->init_lock); | |
412 | ||
413 | return ret; | |
414 | } | |
415 | ||
522698d7 SS |
416 | static DEVICE_ATTR_RO(io_stat); |
417 | static DEVICE_ATTR_RO(mm_stat); | |
623e47fc | 418 | static DEVICE_ATTR_RO(debug_stat); |
522698d7 | 419 | |
522698d7 SS |
420 | static void zram_meta_free(struct zram_meta *meta, u64 disksize) |
421 | { | |
422 | size_t num_pages = disksize >> PAGE_SHIFT; | |
423 | size_t index; | |
1fec1172 GM |
424 | |
425 | /* Free all pages that are still in this zram device */ | |
426 | for (index = 0; index < num_pages; index++) { | |
427 | unsigned long handle = meta->table[index].handle; | |
8e19d540 | 428 | /* |
429 | * No memory is allocated for same element filled pages. | |
430 | * Simply clear same page flag. | |
431 | */ | |
432 | if (!handle || zram_test_flag(meta, index, ZRAM_SAME)) | |
1fec1172 GM |
433 | continue; |
434 | ||
435 | zs_free(meta->mem_pool, handle); | |
436 | } | |
437 | ||
9b3bb7ab | 438 | zs_destroy_pool(meta->mem_pool); |
9b3bb7ab SS |
439 | vfree(meta->table); |
440 | kfree(meta); | |
441 | } | |
442 | ||
4ce321f5 | 443 | static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize) |
9b3bb7ab SS |
444 | { |
445 | size_t num_pages; | |
446 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); | |
b8179958 | 447 | |
9b3bb7ab | 448 | if (!meta) |
b8179958 | 449 | return NULL; |
9b3bb7ab | 450 | |
9b3bb7ab SS |
451 | num_pages = disksize >> PAGE_SHIFT; |
452 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
453 | if (!meta->table) { | |
454 | pr_err("Error allocating zram address table\n"); | |
b8179958 | 455 | goto out_error; |
9b3bb7ab SS |
456 | } |
457 | ||
d0d8da2d | 458 | meta->mem_pool = zs_create_pool(pool_name); |
9b3bb7ab SS |
459 | if (!meta->mem_pool) { |
460 | pr_err("Error creating memory pool\n"); | |
b8179958 | 461 | goto out_error; |
9b3bb7ab SS |
462 | } |
463 | ||
464 | return meta; | |
465 | ||
b8179958 | 466 | out_error: |
9b3bb7ab | 467 | vfree(meta->table); |
9b3bb7ab | 468 | kfree(meta); |
b8179958 | 469 | return NULL; |
9b3bb7ab SS |
470 | } |
471 | ||
d2d5e762 WY |
472 | /* |
473 | * To protect concurrent access to the same index entry, | |
474 | * caller should hold this table index entry's bit_spinlock to | |
475 | * indicate this index entry is accessing. | |
476 | */ | |
f1e3cfff | 477 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 478 | { |
8b3cc3ed MK |
479 | struct zram_meta *meta = zram->meta; |
480 | unsigned long handle = meta->table[index].handle; | |
306b0c95 | 481 | |
8e19d540 | 482 | /* |
483 | * No memory is allocated for same element filled pages. | |
484 | * Simply clear same page flag. | |
485 | */ | |
486 | if (zram_test_flag(meta, index, ZRAM_SAME)) { | |
487 | zram_clear_flag(meta, index, ZRAM_SAME); | |
488 | zram_clear_element(meta, index); | |
489 | atomic64_dec(&zram->stats.same_pages); | |
306b0c95 NG |
490 | return; |
491 | } | |
492 | ||
8e19d540 | 493 | if (!handle) |
494 | return; | |
495 | ||
8b3cc3ed | 496 | zs_free(meta->mem_pool, handle); |
306b0c95 | 497 | |
d2d5e762 WY |
498 | atomic64_sub(zram_get_obj_size(meta, index), |
499 | &zram->stats.compr_data_size); | |
90a7806e | 500 | atomic64_dec(&zram->stats.pages_stored); |
306b0c95 | 501 | |
8b3cc3ed | 502 | meta->table[index].handle = 0; |
d2d5e762 | 503 | zram_set_obj_size(meta, index, 0); |
306b0c95 NG |
504 | } |
505 | ||
37b51fdd | 506 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 507 | { |
b7ca232e | 508 | int ret = 0; |
37b51fdd | 509 | unsigned char *cmem; |
8b3cc3ed | 510 | struct zram_meta *meta = zram->meta; |
92967471 | 511 | unsigned long handle; |
ebaf9ab5 | 512 | unsigned int size; |
92967471 | 513 | |
d2d5e762 | 514 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
92967471 | 515 | handle = meta->table[index].handle; |
d2d5e762 | 516 | size = zram_get_obj_size(meta, index); |
306b0c95 | 517 | |
8e19d540 | 518 | if (!handle || zram_test_flag(meta, index, ZRAM_SAME)) { |
d2d5e762 | 519 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
8e19d540 | 520 | zram_fill_page(mem, PAGE_SIZE, meta->table[index].element); |
8c921b2b JM |
521 | return 0; |
522 | } | |
306b0c95 | 523 | |
8b3cc3ed | 524 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
ebaf9ab5 | 525 | if (size == PAGE_SIZE) { |
42e99bd9 | 526 | copy_page(mem, cmem); |
ebaf9ab5 SS |
527 | } else { |
528 | struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp); | |
529 | ||
530 | ret = zcomp_decompress(zstrm, cmem, size, mem); | |
531 | zcomp_stream_put(zram->comp); | |
532 | } | |
8b3cc3ed | 533 | zs_unmap_object(meta->mem_pool, handle); |
d2d5e762 | 534 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
a1dd52af | 535 | |
8c921b2b | 536 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 537 | if (unlikely(ret)) { |
8c921b2b | 538 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); |
8c921b2b | 539 | return ret; |
a1dd52af | 540 | } |
306b0c95 | 541 | |
8c921b2b | 542 | return 0; |
306b0c95 NG |
543 | } |
544 | ||
37b51fdd | 545 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
b627cff3 | 546 | u32 index, int offset) |
924bd88d JM |
547 | { |
548 | int ret; | |
37b51fdd SS |
549 | struct page *page; |
550 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 551 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
552 | page = bvec->bv_page; |
553 | ||
d2d5e762 | 554 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
8b3cc3ed | 555 | if (unlikely(!meta->table[index].handle) || |
8e19d540 | 556 | zram_test_flag(meta, index, ZRAM_SAME)) { |
d2d5e762 | 557 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
8e19d540 | 558 | handle_same_page(bvec, meta->table[index].element); |
924bd88d JM |
559 | return 0; |
560 | } | |
d2d5e762 | 561 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
924bd88d | 562 | |
37b51fdd SS |
563 | if (is_partial_io(bvec)) |
564 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
565 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
566 | ||
567 | user_mem = kmap_atomic(page); | |
568 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
569 | uncmem = user_mem; |
570 | ||
571 | if (!uncmem) { | |
70864969 | 572 | pr_err("Unable to allocate temp memory\n"); |
37b51fdd SS |
573 | ret = -ENOMEM; |
574 | goto out_cleanup; | |
575 | } | |
924bd88d | 576 | |
37b51fdd | 577 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 578 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 579 | if (unlikely(ret)) |
37b51fdd | 580 | goto out_cleanup; |
924bd88d | 581 | |
37b51fdd SS |
582 | if (is_partial_io(bvec)) |
583 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
584 | bvec->bv_len); | |
585 | ||
586 | flush_dcache_page(page); | |
587 | ret = 0; | |
588 | out_cleanup: | |
589 | kunmap_atomic(user_mem); | |
590 | if (is_partial_io(bvec)) | |
591 | kfree(uncmem); | |
592 | return ret; | |
924bd88d JM |
593 | } |
594 | ||
595 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
596 | int offset) | |
306b0c95 | 597 | { |
397c6066 | 598 | int ret = 0; |
ebaf9ab5 | 599 | unsigned int clen; |
da9556a2 | 600 | unsigned long handle = 0; |
130f315a | 601 | struct page *page; |
924bd88d | 602 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 603 | struct zram_meta *meta = zram->meta; |
17162f41 | 604 | struct zcomp_strm *zstrm = NULL; |
461a8eee | 605 | unsigned long alloced_pages; |
8e19d540 | 606 | unsigned long element; |
306b0c95 | 607 | |
8c921b2b | 608 | page = bvec->bv_page; |
924bd88d JM |
609 | if (is_partial_io(bvec)) { |
610 | /* | |
611 | * This is a partial IO. We need to read the full page | |
612 | * before to write the changes. | |
613 | */ | |
7e5a5104 | 614 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 615 | if (!uncmem) { |
924bd88d JM |
616 | ret = -ENOMEM; |
617 | goto out; | |
618 | } | |
37b51fdd | 619 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 620 | if (ret) |
924bd88d | 621 | goto out; |
924bd88d JM |
622 | } |
623 | ||
da9556a2 | 624 | compress_again: |
ba82fe2e | 625 | user_mem = kmap_atomic(page); |
397c6066 | 626 | if (is_partial_io(bvec)) { |
924bd88d JM |
627 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
628 | bvec->bv_len); | |
397c6066 NG |
629 | kunmap_atomic(user_mem); |
630 | user_mem = NULL; | |
631 | } else { | |
924bd88d | 632 | uncmem = user_mem; |
397c6066 | 633 | } |
924bd88d | 634 | |
8e19d540 | 635 | if (page_same_filled(uncmem, &element)) { |
c4065152 WY |
636 | if (user_mem) |
637 | kunmap_atomic(user_mem); | |
f40ac2ae | 638 | /* Free memory associated with this sector now. */ |
d2d5e762 | 639 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 640 | zram_free_page(zram, index); |
8e19d540 | 641 | zram_set_flag(meta, index, ZRAM_SAME); |
642 | zram_set_element(meta, index, element); | |
d2d5e762 | 643 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 644 | |
8e19d540 | 645 | atomic64_inc(&zram->stats.same_pages); |
924bd88d JM |
646 | ret = 0; |
647 | goto out; | |
8c921b2b | 648 | } |
306b0c95 | 649 | |
2aea8493 | 650 | zstrm = zcomp_stream_get(zram->comp); |
ebaf9ab5 | 651 | ret = zcomp_compress(zstrm, uncmem, &clen); |
397c6066 NG |
652 | if (!is_partial_io(bvec)) { |
653 | kunmap_atomic(user_mem); | |
654 | user_mem = NULL; | |
655 | uncmem = NULL; | |
656 | } | |
306b0c95 | 657 | |
b7ca232e | 658 | if (unlikely(ret)) { |
8c921b2b | 659 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 660 | goto out; |
8c921b2b | 661 | } |
da9556a2 | 662 | |
b7ca232e | 663 | src = zstrm->buffer; |
c8f2f0db | 664 | if (unlikely(clen > max_zpage_size)) { |
c8f2f0db | 665 | clen = PAGE_SIZE; |
397c6066 NG |
666 | if (is_partial_io(bvec)) |
667 | src = uncmem; | |
c8f2f0db | 668 | } |
a1dd52af | 669 | |
da9556a2 SS |
670 | /* |
671 | * handle allocation has 2 paths: | |
672 | * a) fast path is executed with preemption disabled (for | |
673 | * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear, | |
674 | * since we can't sleep; | |
675 | * b) slow path enables preemption and attempts to allocate | |
676 | * the page with __GFP_DIRECT_RECLAIM bit set. we have to | |
677 | * put per-cpu compression stream and, thus, to re-do | |
678 | * the compression once handle is allocated. | |
679 | * | |
680 | * if we have a 'non-null' handle here then we are coming | |
681 | * from the slow path and handle has already been allocated. | |
682 | */ | |
683 | if (!handle) | |
684 | handle = zs_malloc(meta->mem_pool, clen, | |
685 | __GFP_KSWAPD_RECLAIM | | |
686 | __GFP_NOWARN | | |
9bc482d3 MK |
687 | __GFP_HIGHMEM | |
688 | __GFP_MOVABLE); | |
fd1a30de | 689 | if (!handle) { |
2aea8493 | 690 | zcomp_stream_put(zram->comp); |
da9556a2 SS |
691 | zstrm = NULL; |
692 | ||
623e47fc SS |
693 | atomic64_inc(&zram->stats.writestall); |
694 | ||
da9556a2 | 695 | handle = zs_malloc(meta->mem_pool, clen, |
9bc482d3 MK |
696 | GFP_NOIO | __GFP_HIGHMEM | |
697 | __GFP_MOVABLE); | |
da9556a2 SS |
698 | if (handle) |
699 | goto compress_again; | |
700 | ||
ebaf9ab5 | 701 | pr_err("Error allocating memory for compressed page: %u, size=%u\n", |
596b3dd4 | 702 | index, clen); |
924bd88d JM |
703 | ret = -ENOMEM; |
704 | goto out; | |
8c921b2b | 705 | } |
9ada9da9 | 706 | |
461a8eee | 707 | alloced_pages = zs_get_total_pages(meta->mem_pool); |
12372755 SS |
708 | update_used_max(zram, alloced_pages); |
709 | ||
461a8eee | 710 | if (zram->limit_pages && alloced_pages > zram->limit_pages) { |
9ada9da9 MK |
711 | zs_free(meta->mem_pool, handle); |
712 | ret = -ENOMEM; | |
713 | goto out; | |
714 | } | |
715 | ||
8b3cc3ed | 716 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 717 | |
42e99bd9 | 718 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 719 | src = kmap_atomic(page); |
42e99bd9 | 720 | copy_page(cmem, src); |
397c6066 | 721 | kunmap_atomic(src); |
42e99bd9 JL |
722 | } else { |
723 | memcpy(cmem, src, clen); | |
724 | } | |
306b0c95 | 725 | |
2aea8493 | 726 | zcomp_stream_put(zram->comp); |
17162f41 | 727 | zstrm = NULL; |
8b3cc3ed | 728 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 729 | |
f40ac2ae SS |
730 | /* |
731 | * Free memory associated with this sector | |
732 | * before overwriting unused sectors. | |
733 | */ | |
d2d5e762 | 734 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae SS |
735 | zram_free_page(zram, index); |
736 | ||
8b3cc3ed | 737 | meta->table[index].handle = handle; |
d2d5e762 WY |
738 | zram_set_obj_size(meta, index, clen); |
739 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); | |
306b0c95 | 740 | |
8c921b2b | 741 | /* Update stats */ |
90a7806e SS |
742 | atomic64_add(clen, &zram->stats.compr_data_size); |
743 | atomic64_inc(&zram->stats.pages_stored); | |
924bd88d | 744 | out: |
17162f41 | 745 | if (zstrm) |
2aea8493 | 746 | zcomp_stream_put(zram->comp); |
397c6066 NG |
747 | if (is_partial_io(bvec)) |
748 | kfree(uncmem); | |
924bd88d | 749 | return ret; |
8c921b2b JM |
750 | } |
751 | ||
f4659d8e JK |
752 | /* |
753 | * zram_bio_discard - handler on discard request | |
754 | * @index: physical block index in PAGE_SIZE units | |
755 | * @offset: byte offset within physical block | |
756 | */ | |
757 | static void zram_bio_discard(struct zram *zram, u32 index, | |
758 | int offset, struct bio *bio) | |
759 | { | |
760 | size_t n = bio->bi_iter.bi_size; | |
d2d5e762 | 761 | struct zram_meta *meta = zram->meta; |
f4659d8e JK |
762 | |
763 | /* | |
764 | * zram manages data in physical block size units. Because logical block | |
765 | * size isn't identical with physical block size on some arch, we | |
766 | * could get a discard request pointing to a specific offset within a | |
767 | * certain physical block. Although we can handle this request by | |
768 | * reading that physiclal block and decompressing and partially zeroing | |
769 | * and re-compressing and then re-storing it, this isn't reasonable | |
770 | * because our intent with a discard request is to save memory. So | |
771 | * skipping this logical block is appropriate here. | |
772 | */ | |
773 | if (offset) { | |
38515c73 | 774 | if (n <= (PAGE_SIZE - offset)) |
f4659d8e JK |
775 | return; |
776 | ||
38515c73 | 777 | n -= (PAGE_SIZE - offset); |
f4659d8e JK |
778 | index++; |
779 | } | |
780 | ||
781 | while (n >= PAGE_SIZE) { | |
d2d5e762 | 782 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f4659d8e | 783 | zram_free_page(zram, index); |
d2d5e762 | 784 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
015254da | 785 | atomic64_inc(&zram->stats.notify_free); |
f4659d8e JK |
786 | index++; |
787 | n -= PAGE_SIZE; | |
788 | } | |
789 | } | |
790 | ||
522698d7 | 791 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, |
c11f0c0b | 792 | int offset, bool is_write) |
9b3bb7ab | 793 | { |
522698d7 | 794 | unsigned long start_time = jiffies; |
c11f0c0b | 795 | int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ; |
9b3bb7ab | 796 | int ret; |
9b3bb7ab | 797 | |
c11f0c0b | 798 | generic_start_io_acct(rw_acct, bvec->bv_len >> SECTOR_SHIFT, |
522698d7 | 799 | &zram->disk->part0); |
46a51c80 | 800 | |
c11f0c0b | 801 | if (!is_write) { |
522698d7 SS |
802 | atomic64_inc(&zram->stats.num_reads); |
803 | ret = zram_bvec_read(zram, bvec, index, offset); | |
804 | } else { | |
805 | atomic64_inc(&zram->stats.num_writes); | |
806 | ret = zram_bvec_write(zram, bvec, index, offset); | |
1b672224 | 807 | } |
9b3bb7ab | 808 | |
c11f0c0b | 809 | generic_end_io_acct(rw_acct, &zram->disk->part0, start_time); |
9b3bb7ab | 810 | |
522698d7 | 811 | if (unlikely(ret)) { |
c11f0c0b | 812 | if (!is_write) |
522698d7 SS |
813 | atomic64_inc(&zram->stats.failed_reads); |
814 | else | |
815 | atomic64_inc(&zram->stats.failed_writes); | |
1b672224 | 816 | } |
9b3bb7ab | 817 | |
1b672224 | 818 | return ret; |
8c921b2b JM |
819 | } |
820 | ||
be257c61 | 821 | static void __zram_make_request(struct zram *zram, struct bio *bio) |
8c921b2b | 822 | { |
abf54548 | 823 | int offset; |
8c921b2b | 824 | u32 index; |
7988613b KO |
825 | struct bio_vec bvec; |
826 | struct bvec_iter iter; | |
8c921b2b | 827 | |
4f024f37 KO |
828 | index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT; |
829 | offset = (bio->bi_iter.bi_sector & | |
830 | (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; | |
8c921b2b | 831 | |
95fe6c1a | 832 | if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) { |
f4659d8e | 833 | zram_bio_discard(zram, index, offset, bio); |
4246a0b6 | 834 | bio_endio(bio); |
f4659d8e JK |
835 | return; |
836 | } | |
837 | ||
7988613b | 838 | bio_for_each_segment(bvec, bio, iter) { |
924bd88d JM |
839 | int max_transfer_size = PAGE_SIZE - offset; |
840 | ||
7988613b | 841 | if (bvec.bv_len > max_transfer_size) { |
924bd88d JM |
842 | /* |
843 | * zram_bvec_rw() can only make operation on a single | |
844 | * zram page. Split the bio vector. | |
845 | */ | |
846 | struct bio_vec bv; | |
847 | ||
7988613b | 848 | bv.bv_page = bvec.bv_page; |
924bd88d | 849 | bv.bv_len = max_transfer_size; |
7988613b | 850 | bv.bv_offset = bvec.bv_offset; |
924bd88d | 851 | |
abf54548 | 852 | if (zram_bvec_rw(zram, &bv, index, offset, |
c11f0c0b | 853 | op_is_write(bio_op(bio))) < 0) |
924bd88d JM |
854 | goto out; |
855 | ||
7988613b | 856 | bv.bv_len = bvec.bv_len - max_transfer_size; |
924bd88d | 857 | bv.bv_offset += max_transfer_size; |
abf54548 | 858 | if (zram_bvec_rw(zram, &bv, index + 1, 0, |
c11f0c0b | 859 | op_is_write(bio_op(bio))) < 0) |
924bd88d JM |
860 | goto out; |
861 | } else | |
abf54548 | 862 | if (zram_bvec_rw(zram, &bvec, index, offset, |
c11f0c0b | 863 | op_is_write(bio_op(bio))) < 0) |
924bd88d JM |
864 | goto out; |
865 | ||
7988613b | 866 | update_position(&index, &offset, &bvec); |
a1dd52af | 867 | } |
306b0c95 | 868 | |
4246a0b6 | 869 | bio_endio(bio); |
7d7854b4 | 870 | return; |
306b0c95 NG |
871 | |
872 | out: | |
306b0c95 | 873 | bio_io_error(bio); |
306b0c95 NG |
874 | } |
875 | ||
306b0c95 | 876 | /* |
f1e3cfff | 877 | * Handler function for all zram I/O requests. |
306b0c95 | 878 | */ |
dece1635 | 879 | static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 880 | { |
f1e3cfff | 881 | struct zram *zram = queue->queuedata; |
306b0c95 | 882 | |
54efd50b KO |
883 | blk_queue_split(queue, &bio, queue->bio_split); |
884 | ||
54850e73 | 885 | if (!valid_io_request(zram, bio->bi_iter.bi_sector, |
886 | bio->bi_iter.bi_size)) { | |
da5cc7d3 | 887 | atomic64_inc(&zram->stats.invalid_io); |
a09759ac | 888 | goto error; |
6642a67c JM |
889 | } |
890 | ||
be257c61 | 891 | __zram_make_request(zram, bio); |
dece1635 | 892 | return BLK_QC_T_NONE; |
a09759ac | 893 | |
0900beae JM |
894 | error: |
895 | bio_io_error(bio); | |
dece1635 | 896 | return BLK_QC_T_NONE; |
306b0c95 NG |
897 | } |
898 | ||
2ccbec05 NG |
899 | static void zram_slot_free_notify(struct block_device *bdev, |
900 | unsigned long index) | |
107c161b | 901 | { |
f1e3cfff | 902 | struct zram *zram; |
f614a9f4 | 903 | struct zram_meta *meta; |
107c161b | 904 | |
f1e3cfff | 905 | zram = bdev->bd_disk->private_data; |
f614a9f4 | 906 | meta = zram->meta; |
a0c516cb | 907 | |
d2d5e762 | 908 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 909 | zram_free_page(zram, index); |
d2d5e762 | 910 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 911 | atomic64_inc(&zram->stats.notify_free); |
107c161b NG |
912 | } |
913 | ||
8c7f0102 | 914 | static int zram_rw_page(struct block_device *bdev, sector_t sector, |
c11f0c0b | 915 | struct page *page, bool is_write) |
8c7f0102 | 916 | { |
08eee69f | 917 | int offset, err = -EIO; |
8c7f0102 | 918 | u32 index; |
919 | struct zram *zram; | |
920 | struct bio_vec bv; | |
921 | ||
922 | zram = bdev->bd_disk->private_data; | |
08eee69f | 923 | |
8c7f0102 | 924 | if (!valid_io_request(zram, sector, PAGE_SIZE)) { |
925 | atomic64_inc(&zram->stats.invalid_io); | |
08eee69f | 926 | err = -EINVAL; |
a09759ac | 927 | goto out; |
8c7f0102 | 928 | } |
929 | ||
930 | index = sector >> SECTORS_PER_PAGE_SHIFT; | |
931 | offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT; | |
932 | ||
933 | bv.bv_page = page; | |
934 | bv.bv_len = PAGE_SIZE; | |
935 | bv.bv_offset = 0; | |
936 | ||
c11f0c0b | 937 | err = zram_bvec_rw(zram, &bv, index, offset, is_write); |
08eee69f | 938 | out: |
8c7f0102 | 939 | /* |
940 | * If I/O fails, just return error(ie, non-zero) without | |
941 | * calling page_endio. | |
942 | * It causes resubmit the I/O with bio request by upper functions | |
943 | * of rw_page(e.g., swap_readpage, __swap_writepage) and | |
944 | * bio->bi_end_io does things to handle the error | |
945 | * (e.g., SetPageError, set_page_dirty and extra works). | |
946 | */ | |
947 | if (err == 0) | |
c11f0c0b | 948 | page_endio(page, is_write, 0); |
8c7f0102 | 949 | return err; |
950 | } | |
951 | ||
522698d7 SS |
952 | static void zram_reset_device(struct zram *zram) |
953 | { | |
954 | struct zram_meta *meta; | |
955 | struct zcomp *comp; | |
956 | u64 disksize; | |
306b0c95 | 957 | |
522698d7 | 958 | down_write(&zram->init_lock); |
9b3bb7ab | 959 | |
522698d7 SS |
960 | zram->limit_pages = 0; |
961 | ||
962 | if (!init_done(zram)) { | |
963 | up_write(&zram->init_lock); | |
964 | return; | |
965 | } | |
966 | ||
967 | meta = zram->meta; | |
968 | comp = zram->comp; | |
969 | disksize = zram->disksize; | |
522698d7 SS |
970 | |
971 | /* Reset stats */ | |
972 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
973 | zram->disksize = 0; | |
522698d7 SS |
974 | |
975 | set_capacity(zram->disk, 0); | |
976 | part_stat_set_all(&zram->disk->part0, 0); | |
977 | ||
978 | up_write(&zram->init_lock); | |
979 | /* I/O operation under all of CPU are done so let's free */ | |
980 | zram_meta_free(meta, disksize); | |
981 | zcomp_destroy(comp); | |
982 | } | |
983 | ||
984 | static ssize_t disksize_store(struct device *dev, | |
985 | struct device_attribute *attr, const char *buf, size_t len) | |
2f6a3bed | 986 | { |
522698d7 SS |
987 | u64 disksize; |
988 | struct zcomp *comp; | |
989 | struct zram_meta *meta; | |
2f6a3bed | 990 | struct zram *zram = dev_to_zram(dev); |
522698d7 | 991 | int err; |
2f6a3bed | 992 | |
522698d7 SS |
993 | disksize = memparse(buf, NULL); |
994 | if (!disksize) | |
995 | return -EINVAL; | |
2f6a3bed | 996 | |
522698d7 | 997 | disksize = PAGE_ALIGN(disksize); |
4ce321f5 | 998 | meta = zram_meta_alloc(zram->disk->disk_name, disksize); |
522698d7 SS |
999 | if (!meta) |
1000 | return -ENOMEM; | |
1001 | ||
da9556a2 | 1002 | comp = zcomp_create(zram->compressor); |
522698d7 | 1003 | if (IS_ERR(comp)) { |
70864969 | 1004 | pr_err("Cannot initialise %s compressing backend\n", |
522698d7 SS |
1005 | zram->compressor); |
1006 | err = PTR_ERR(comp); | |
1007 | goto out_free_meta; | |
1008 | } | |
1009 | ||
1010 | down_write(&zram->init_lock); | |
1011 | if (init_done(zram)) { | |
1012 | pr_info("Cannot change disksize for initialized device\n"); | |
1013 | err = -EBUSY; | |
1014 | goto out_destroy_comp; | |
1015 | } | |
1016 | ||
522698d7 SS |
1017 | zram->meta = meta; |
1018 | zram->comp = comp; | |
1019 | zram->disksize = disksize; | |
1020 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
b09ab054 | 1021 | zram_revalidate_disk(zram); |
e7ccfc4c | 1022 | up_write(&zram->init_lock); |
522698d7 SS |
1023 | |
1024 | return len; | |
1025 | ||
1026 | out_destroy_comp: | |
1027 | up_write(&zram->init_lock); | |
1028 | zcomp_destroy(comp); | |
1029 | out_free_meta: | |
1030 | zram_meta_free(meta, disksize); | |
1031 | return err; | |
2f6a3bed SS |
1032 | } |
1033 | ||
522698d7 SS |
1034 | static ssize_t reset_store(struct device *dev, |
1035 | struct device_attribute *attr, const char *buf, size_t len) | |
4f2109f6 | 1036 | { |
522698d7 SS |
1037 | int ret; |
1038 | unsigned short do_reset; | |
1039 | struct zram *zram; | |
1040 | struct block_device *bdev; | |
4f2109f6 | 1041 | |
f405c445 SS |
1042 | ret = kstrtou16(buf, 10, &do_reset); |
1043 | if (ret) | |
1044 | return ret; | |
1045 | ||
1046 | if (!do_reset) | |
1047 | return -EINVAL; | |
1048 | ||
522698d7 SS |
1049 | zram = dev_to_zram(dev); |
1050 | bdev = bdget_disk(zram->disk, 0); | |
522698d7 SS |
1051 | if (!bdev) |
1052 | return -ENOMEM; | |
4f2109f6 | 1053 | |
522698d7 | 1054 | mutex_lock(&bdev->bd_mutex); |
f405c445 SS |
1055 | /* Do not reset an active device or claimed device */ |
1056 | if (bdev->bd_openers || zram->claim) { | |
1057 | mutex_unlock(&bdev->bd_mutex); | |
1058 | bdput(bdev); | |
1059 | return -EBUSY; | |
522698d7 SS |
1060 | } |
1061 | ||
f405c445 SS |
1062 | /* From now on, anyone can't open /dev/zram[0-9] */ |
1063 | zram->claim = true; | |
1064 | mutex_unlock(&bdev->bd_mutex); | |
522698d7 | 1065 | |
f405c445 | 1066 | /* Make sure all the pending I/O are finished */ |
522698d7 SS |
1067 | fsync_bdev(bdev); |
1068 | zram_reset_device(zram); | |
b09ab054 | 1069 | zram_revalidate_disk(zram); |
522698d7 SS |
1070 | bdput(bdev); |
1071 | ||
f405c445 SS |
1072 | mutex_lock(&bdev->bd_mutex); |
1073 | zram->claim = false; | |
1074 | mutex_unlock(&bdev->bd_mutex); | |
1075 | ||
522698d7 | 1076 | return len; |
f405c445 SS |
1077 | } |
1078 | ||
1079 | static int zram_open(struct block_device *bdev, fmode_t mode) | |
1080 | { | |
1081 | int ret = 0; | |
1082 | struct zram *zram; | |
1083 | ||
1084 | WARN_ON(!mutex_is_locked(&bdev->bd_mutex)); | |
1085 | ||
1086 | zram = bdev->bd_disk->private_data; | |
1087 | /* zram was claimed to reset so open request fails */ | |
1088 | if (zram->claim) | |
1089 | ret = -EBUSY; | |
4f2109f6 SS |
1090 | |
1091 | return ret; | |
1092 | } | |
1093 | ||
522698d7 | 1094 | static const struct block_device_operations zram_devops = { |
f405c445 | 1095 | .open = zram_open, |
522698d7 SS |
1096 | .swap_slot_free_notify = zram_slot_free_notify, |
1097 | .rw_page = zram_rw_page, | |
1098 | .owner = THIS_MODULE | |
1099 | }; | |
1100 | ||
1101 | static DEVICE_ATTR_WO(compact); | |
1102 | static DEVICE_ATTR_RW(disksize); | |
1103 | static DEVICE_ATTR_RO(initstate); | |
1104 | static DEVICE_ATTR_WO(reset); | |
c87d1655 SS |
1105 | static DEVICE_ATTR_WO(mem_limit); |
1106 | static DEVICE_ATTR_WO(mem_used_max); | |
522698d7 SS |
1107 | static DEVICE_ATTR_RW(max_comp_streams); |
1108 | static DEVICE_ATTR_RW(comp_algorithm); | |
a68eb3b6 | 1109 | |
9b3bb7ab SS |
1110 | static struct attribute *zram_disk_attrs[] = { |
1111 | &dev_attr_disksize.attr, | |
1112 | &dev_attr_initstate.attr, | |
1113 | &dev_attr_reset.attr, | |
99ebbd30 | 1114 | &dev_attr_compact.attr, |
9ada9da9 | 1115 | &dev_attr_mem_limit.attr, |
461a8eee | 1116 | &dev_attr_mem_used_max.attr, |
beca3ec7 | 1117 | &dev_attr_max_comp_streams.attr, |
e46b8a03 | 1118 | &dev_attr_comp_algorithm.attr, |
2f6a3bed | 1119 | &dev_attr_io_stat.attr, |
4f2109f6 | 1120 | &dev_attr_mm_stat.attr, |
623e47fc | 1121 | &dev_attr_debug_stat.attr, |
9b3bb7ab SS |
1122 | NULL, |
1123 | }; | |
1124 | ||
1125 | static struct attribute_group zram_disk_attr_group = { | |
1126 | .attrs = zram_disk_attrs, | |
1127 | }; | |
1128 | ||
92ff1528 SS |
1129 | /* |
1130 | * Allocate and initialize new zram device. the function returns | |
1131 | * '>= 0' device_id upon success, and negative value otherwise. | |
1132 | */ | |
1133 | static int zram_add(void) | |
306b0c95 | 1134 | { |
85508ec6 | 1135 | struct zram *zram; |
ee980160 | 1136 | struct request_queue *queue; |
92ff1528 | 1137 | int ret, device_id; |
85508ec6 SS |
1138 | |
1139 | zram = kzalloc(sizeof(struct zram), GFP_KERNEL); | |
1140 | if (!zram) | |
1141 | return -ENOMEM; | |
1142 | ||
92ff1528 | 1143 | ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL); |
85508ec6 SS |
1144 | if (ret < 0) |
1145 | goto out_free_dev; | |
92ff1528 | 1146 | device_id = ret; |
de1a21a0 | 1147 | |
0900beae | 1148 | init_rwsem(&zram->init_lock); |
306b0c95 | 1149 | |
ee980160 SS |
1150 | queue = blk_alloc_queue(GFP_KERNEL); |
1151 | if (!queue) { | |
306b0c95 NG |
1152 | pr_err("Error allocating disk queue for device %d\n", |
1153 | device_id); | |
85508ec6 SS |
1154 | ret = -ENOMEM; |
1155 | goto out_free_idr; | |
306b0c95 NG |
1156 | } |
1157 | ||
ee980160 | 1158 | blk_queue_make_request(queue, zram_make_request); |
306b0c95 | 1159 | |
85508ec6 | 1160 | /* gendisk structure */ |
f1e3cfff NG |
1161 | zram->disk = alloc_disk(1); |
1162 | if (!zram->disk) { | |
70864969 | 1163 | pr_err("Error allocating disk structure for device %d\n", |
306b0c95 | 1164 | device_id); |
201c7b72 | 1165 | ret = -ENOMEM; |
39a9b8ac | 1166 | goto out_free_queue; |
306b0c95 NG |
1167 | } |
1168 | ||
f1e3cfff NG |
1169 | zram->disk->major = zram_major; |
1170 | zram->disk->first_minor = device_id; | |
1171 | zram->disk->fops = &zram_devops; | |
ee980160 SS |
1172 | zram->disk->queue = queue; |
1173 | zram->disk->queue->queuedata = zram; | |
f1e3cfff NG |
1174 | zram->disk->private_data = zram; |
1175 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 1176 | |
33863c21 | 1177 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 1178 | set_capacity(zram->disk, 0); |
b67d1ec1 SS |
1179 | /* zram devices sort of resembles non-rotational disks */ |
1180 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
b277da0a | 1181 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue); |
a1dd52af NG |
1182 | /* |
1183 | * To ensure that we always get PAGE_SIZE aligned | |
1184 | * and n*PAGE_SIZED sized I/O requests. | |
1185 | */ | |
f1e3cfff | 1186 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
1187 | blk_queue_logical_block_size(zram->disk->queue, |
1188 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
1189 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
1190 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
f4659d8e | 1191 | zram->disk->queue->limits.discard_granularity = PAGE_SIZE; |
0bc31538 JT |
1192 | zram->disk->queue->limits.max_sectors = SECTORS_PER_PAGE; |
1193 | zram->disk->queue->limits.chunk_sectors = 0; | |
2bb4cd5c | 1194 | blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX); |
f4659d8e JK |
1195 | /* |
1196 | * zram_bio_discard() will clear all logical blocks if logical block | |
1197 | * size is identical with physical block size(PAGE_SIZE). But if it is | |
1198 | * different, we will skip discarding some parts of logical blocks in | |
1199 | * the part of the request range which isn't aligned to physical block | |
1200 | * size. So we can't ensure that all discarded logical blocks are | |
1201 | * zeroed. | |
1202 | */ | |
1203 | if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE) | |
1204 | zram->disk->queue->limits.discard_zeroes_data = 1; | |
1205 | else | |
1206 | zram->disk->queue->limits.discard_zeroes_data = 0; | |
1207 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue); | |
5d83d5a0 | 1208 | |
f1e3cfff | 1209 | add_disk(zram->disk); |
306b0c95 | 1210 | |
33863c21 NG |
1211 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
1212 | &zram_disk_attr_group); | |
1213 | if (ret < 0) { | |
70864969 SS |
1214 | pr_err("Error creating sysfs group for device %d\n", |
1215 | device_id); | |
39a9b8ac | 1216 | goto out_free_disk; |
33863c21 | 1217 | } |
e46b8a03 | 1218 | strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor)); |
be2d1d56 | 1219 | zram->meta = NULL; |
d12b63c9 SS |
1220 | |
1221 | pr_info("Added device: %s\n", zram->disk->disk_name); | |
92ff1528 | 1222 | return device_id; |
de1a21a0 | 1223 | |
39a9b8ac JL |
1224 | out_free_disk: |
1225 | del_gendisk(zram->disk); | |
1226 | put_disk(zram->disk); | |
1227 | out_free_queue: | |
ee980160 | 1228 | blk_cleanup_queue(queue); |
85508ec6 SS |
1229 | out_free_idr: |
1230 | idr_remove(&zram_index_idr, device_id); | |
1231 | out_free_dev: | |
1232 | kfree(zram); | |
de1a21a0 | 1233 | return ret; |
306b0c95 NG |
1234 | } |
1235 | ||
6566d1a3 | 1236 | static int zram_remove(struct zram *zram) |
306b0c95 | 1237 | { |
6566d1a3 SS |
1238 | struct block_device *bdev; |
1239 | ||
1240 | bdev = bdget_disk(zram->disk, 0); | |
1241 | if (!bdev) | |
1242 | return -ENOMEM; | |
1243 | ||
1244 | mutex_lock(&bdev->bd_mutex); | |
1245 | if (bdev->bd_openers || zram->claim) { | |
1246 | mutex_unlock(&bdev->bd_mutex); | |
1247 | bdput(bdev); | |
1248 | return -EBUSY; | |
1249 | } | |
1250 | ||
1251 | zram->claim = true; | |
1252 | mutex_unlock(&bdev->bd_mutex); | |
1253 | ||
85508ec6 SS |
1254 | /* |
1255 | * Remove sysfs first, so no one will perform a disksize | |
6566d1a3 SS |
1256 | * store while we destroy the devices. This also helps during |
1257 | * hot_remove -- zram_reset_device() is the last holder of | |
1258 | * ->init_lock, no later/concurrent disksize_store() or any | |
1259 | * other sysfs handlers are possible. | |
85508ec6 SS |
1260 | */ |
1261 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, | |
1262 | &zram_disk_attr_group); | |
306b0c95 | 1263 | |
6566d1a3 SS |
1264 | /* Make sure all the pending I/O are finished */ |
1265 | fsync_bdev(bdev); | |
85508ec6 | 1266 | zram_reset_device(zram); |
6566d1a3 SS |
1267 | bdput(bdev); |
1268 | ||
1269 | pr_info("Removed device: %s\n", zram->disk->disk_name); | |
1270 | ||
85508ec6 SS |
1271 | blk_cleanup_queue(zram->disk->queue); |
1272 | del_gendisk(zram->disk); | |
1273 | put_disk(zram->disk); | |
1274 | kfree(zram); | |
6566d1a3 SS |
1275 | return 0; |
1276 | } | |
1277 | ||
1278 | /* zram-control sysfs attributes */ | |
1279 | static ssize_t hot_add_show(struct class *class, | |
1280 | struct class_attribute *attr, | |
1281 | char *buf) | |
1282 | { | |
1283 | int ret; | |
1284 | ||
1285 | mutex_lock(&zram_index_mutex); | |
1286 | ret = zram_add(); | |
1287 | mutex_unlock(&zram_index_mutex); | |
1288 | ||
1289 | if (ret < 0) | |
1290 | return ret; | |
1291 | return scnprintf(buf, PAGE_SIZE, "%d\n", ret); | |
1292 | } | |
1293 | ||
1294 | static ssize_t hot_remove_store(struct class *class, | |
1295 | struct class_attribute *attr, | |
1296 | const char *buf, | |
1297 | size_t count) | |
1298 | { | |
1299 | struct zram *zram; | |
1300 | int ret, dev_id; | |
1301 | ||
1302 | /* dev_id is gendisk->first_minor, which is `int' */ | |
1303 | ret = kstrtoint(buf, 10, &dev_id); | |
1304 | if (ret) | |
1305 | return ret; | |
1306 | if (dev_id < 0) | |
1307 | return -EINVAL; | |
1308 | ||
1309 | mutex_lock(&zram_index_mutex); | |
1310 | ||
1311 | zram = idr_find(&zram_index_idr, dev_id); | |
17ec4cd9 | 1312 | if (zram) { |
6566d1a3 | 1313 | ret = zram_remove(zram); |
529e71e1 TI |
1314 | if (!ret) |
1315 | idr_remove(&zram_index_idr, dev_id); | |
17ec4cd9 | 1316 | } else { |
6566d1a3 | 1317 | ret = -ENODEV; |
17ec4cd9 | 1318 | } |
6566d1a3 SS |
1319 | |
1320 | mutex_unlock(&zram_index_mutex); | |
1321 | return ret ? ret : count; | |
85508ec6 | 1322 | } |
a096cafc | 1323 | |
5c7e9ccd SS |
1324 | /* |
1325 | * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a | |
1326 | * sense that reading from this file does alter the state of your system -- it | |
1327 | * creates a new un-initialized zram device and returns back this device's | |
1328 | * device_id (or an error code if it fails to create a new device). | |
1329 | */ | |
6566d1a3 | 1330 | static struct class_attribute zram_control_class_attrs[] = { |
5c7e9ccd | 1331 | __ATTR(hot_add, 0400, hot_add_show, NULL), |
6566d1a3 SS |
1332 | __ATTR_WO(hot_remove), |
1333 | __ATTR_NULL, | |
1334 | }; | |
1335 | ||
1336 | static struct class zram_control_class = { | |
1337 | .name = "zram-control", | |
1338 | .owner = THIS_MODULE, | |
1339 | .class_attrs = zram_control_class_attrs, | |
1340 | }; | |
1341 | ||
85508ec6 SS |
1342 | static int zram_remove_cb(int id, void *ptr, void *data) |
1343 | { | |
1344 | zram_remove(ptr); | |
1345 | return 0; | |
1346 | } | |
a096cafc | 1347 | |
85508ec6 SS |
1348 | static void destroy_devices(void) |
1349 | { | |
6566d1a3 | 1350 | class_unregister(&zram_control_class); |
85508ec6 SS |
1351 | idr_for_each(&zram_index_idr, &zram_remove_cb, NULL); |
1352 | idr_destroy(&zram_index_idr); | |
a096cafc | 1353 | unregister_blkdev(zram_major, "zram"); |
1dd6c834 | 1354 | cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE); |
306b0c95 NG |
1355 | } |
1356 | ||
f1e3cfff | 1357 | static int __init zram_init(void) |
306b0c95 | 1358 | { |
92ff1528 | 1359 | int ret; |
306b0c95 | 1360 | |
1dd6c834 AMG |
1361 | ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare", |
1362 | zcomp_cpu_up_prepare, zcomp_cpu_dead); | |
1363 | if (ret < 0) | |
1364 | return ret; | |
1365 | ||
6566d1a3 SS |
1366 | ret = class_register(&zram_control_class); |
1367 | if (ret) { | |
70864969 | 1368 | pr_err("Unable to register zram-control class\n"); |
1dd6c834 | 1369 | cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE); |
6566d1a3 SS |
1370 | return ret; |
1371 | } | |
1372 | ||
f1e3cfff NG |
1373 | zram_major = register_blkdev(0, "zram"); |
1374 | if (zram_major <= 0) { | |
70864969 | 1375 | pr_err("Unable to get major number\n"); |
6566d1a3 | 1376 | class_unregister(&zram_control_class); |
1dd6c834 | 1377 | cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE); |
a096cafc | 1378 | return -EBUSY; |
306b0c95 NG |
1379 | } |
1380 | ||
92ff1528 | 1381 | while (num_devices != 0) { |
6566d1a3 | 1382 | mutex_lock(&zram_index_mutex); |
92ff1528 | 1383 | ret = zram_add(); |
6566d1a3 | 1384 | mutex_unlock(&zram_index_mutex); |
92ff1528 | 1385 | if (ret < 0) |
a096cafc | 1386 | goto out_error; |
92ff1528 | 1387 | num_devices--; |
de1a21a0 NG |
1388 | } |
1389 | ||
306b0c95 | 1390 | return 0; |
de1a21a0 | 1391 | |
a096cafc | 1392 | out_error: |
85508ec6 | 1393 | destroy_devices(); |
306b0c95 NG |
1394 | return ret; |
1395 | } | |
1396 | ||
f1e3cfff | 1397 | static void __exit zram_exit(void) |
306b0c95 | 1398 | { |
85508ec6 | 1399 | destroy_devices(); |
306b0c95 NG |
1400 | } |
1401 | ||
f1e3cfff NG |
1402 | module_init(zram_init); |
1403 | module_exit(zram_exit); | |
306b0c95 | 1404 | |
9b3bb7ab | 1405 | module_param(num_devices, uint, 0); |
c3cdb40e | 1406 | MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices"); |
9b3bb7ab | 1407 | |
306b0c95 NG |
1408 | MODULE_LICENSE("Dual BSD/GPL"); |
1409 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 1410 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |