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> |
306b0c95 | 28 | #include <linux/string.h> |
306b0c95 | 29 | #include <linux/vmalloc.h> |
fcfa8d95 | 30 | #include <linux/err.h> |
85508ec6 | 31 | #include <linux/idr.h> |
6566d1a3 | 32 | #include <linux/sysfs.h> |
306b0c95 | 33 | |
16a4bfb9 | 34 | #include "zram_drv.h" |
306b0c95 | 35 | |
85508ec6 | 36 | static DEFINE_IDR(zram_index_idr); |
6566d1a3 SS |
37 | /* idr index must be protected */ |
38 | static DEFINE_MUTEX(zram_index_mutex); | |
39 | ||
f1e3cfff | 40 | static int zram_major; |
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 | |
8f7d282c SS |
46 | static inline void deprecated_attr_warn(const char *name) |
47 | { | |
48 | pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n", | |
49 | task_pid_nr(current), | |
50 | current->comm, | |
51 | name, | |
52 | "See zram documentation."); | |
53 | } | |
54 | ||
a68eb3b6 | 55 | #define ZRAM_ATTR_RO(name) \ |
3bca3ef7 | 56 | static ssize_t name##_show(struct device *d, \ |
a68eb3b6 SS |
57 | struct device_attribute *attr, char *b) \ |
58 | { \ | |
59 | struct zram *zram = dev_to_zram(d); \ | |
8f7d282c SS |
60 | \ |
61 | deprecated_attr_warn(__stringify(name)); \ | |
56b4e8cb | 62 | return scnprintf(b, PAGE_SIZE, "%llu\n", \ |
a68eb3b6 SS |
63 | (u64)atomic64_read(&zram->stats.name)); \ |
64 | } \ | |
083914ea | 65 | static DEVICE_ATTR_RO(name); |
a68eb3b6 | 66 | |
08eee69f | 67 | static inline bool init_done(struct zram *zram) |
be2d1d56 | 68 | { |
08eee69f | 69 | return zram->disksize; |
be2d1d56 SS |
70 | } |
71 | ||
9b3bb7ab SS |
72 | static inline struct zram *dev_to_zram(struct device *dev) |
73 | { | |
74 | return (struct zram *)dev_to_disk(dev)->private_data; | |
75 | } | |
76 | ||
b31177f2 | 77 | /* flag operations require table entry bit_spin_lock() being held */ |
522698d7 SS |
78 | static int zram_test_flag(struct zram_meta *meta, u32 index, |
79 | enum zram_pageflags flag) | |
99ebbd30 | 80 | { |
522698d7 SS |
81 | return meta->table[index].value & BIT(flag); |
82 | } | |
99ebbd30 | 83 | |
522698d7 SS |
84 | static void zram_set_flag(struct zram_meta *meta, u32 index, |
85 | enum zram_pageflags flag) | |
86 | { | |
87 | meta->table[index].value |= BIT(flag); | |
88 | } | |
99ebbd30 | 89 | |
522698d7 SS |
90 | static void zram_clear_flag(struct zram_meta *meta, u32 index, |
91 | enum zram_pageflags flag) | |
92 | { | |
93 | meta->table[index].value &= ~BIT(flag); | |
94 | } | |
99ebbd30 | 95 | |
522698d7 SS |
96 | static size_t zram_get_obj_size(struct zram_meta *meta, u32 index) |
97 | { | |
98 | return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1); | |
99ebbd30 AM |
99 | } |
100 | ||
522698d7 SS |
101 | static void zram_set_obj_size(struct zram_meta *meta, |
102 | u32 index, size_t size) | |
9b3bb7ab | 103 | { |
522698d7 | 104 | unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT; |
9b3bb7ab | 105 | |
522698d7 SS |
106 | meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size; |
107 | } | |
108 | ||
109 | static inline int is_partial_io(struct bio_vec *bvec) | |
110 | { | |
111 | return bvec->bv_len != PAGE_SIZE; | |
112 | } | |
113 | ||
114 | /* | |
115 | * Check if request is within bounds and aligned on zram logical blocks. | |
116 | */ | |
117 | static inline int valid_io_request(struct zram *zram, | |
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))) | |
124 | return 0; | |
125 | if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1))) | |
126 | return 0; | |
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)) | |
132 | return 0; | |
133 | ||
134 | /* I/O request is valid */ | |
135 | return 1; | |
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 | ||
160 | static int page_zero_filled(void *ptr) | |
161 | { | |
162 | unsigned int pos; | |
163 | unsigned long *page; | |
164 | ||
165 | page = (unsigned long *)ptr; | |
166 | ||
167 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
168 | if (page[pos]) | |
169 | return 0; | |
170 | } | |
171 | ||
172 | return 1; | |
173 | } | |
174 | ||
175 | static void handle_zero_page(struct bio_vec *bvec) | |
176 | { | |
177 | struct page *page = bvec->bv_page; | |
178 | void *user_mem; | |
179 | ||
180 | user_mem = kmap_atomic(page); | |
181 | if (is_partial_io(bvec)) | |
182 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); | |
183 | else | |
184 | clear_page(user_mem); | |
185 | kunmap_atomic(user_mem); | |
186 | ||
187 | flush_dcache_page(page); | |
9b3bb7ab SS |
188 | } |
189 | ||
190 | static ssize_t initstate_show(struct device *dev, | |
191 | struct device_attribute *attr, char *buf) | |
192 | { | |
a68eb3b6 | 193 | u32 val; |
9b3bb7ab SS |
194 | struct zram *zram = dev_to_zram(dev); |
195 | ||
a68eb3b6 SS |
196 | down_read(&zram->init_lock); |
197 | val = init_done(zram); | |
198 | up_read(&zram->init_lock); | |
9b3bb7ab | 199 | |
56b4e8cb | 200 | return scnprintf(buf, PAGE_SIZE, "%u\n", val); |
9b3bb7ab SS |
201 | } |
202 | ||
522698d7 SS |
203 | static ssize_t disksize_show(struct device *dev, |
204 | struct device_attribute *attr, char *buf) | |
205 | { | |
206 | struct zram *zram = dev_to_zram(dev); | |
207 | ||
208 | return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize); | |
209 | } | |
210 | ||
9b3bb7ab SS |
211 | static ssize_t orig_data_size_show(struct device *dev, |
212 | struct device_attribute *attr, char *buf) | |
213 | { | |
214 | struct zram *zram = dev_to_zram(dev); | |
215 | ||
8f7d282c | 216 | deprecated_attr_warn("orig_data_size"); |
56b4e8cb | 217 | return scnprintf(buf, PAGE_SIZE, "%llu\n", |
90a7806e | 218 | (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT); |
9b3bb7ab SS |
219 | } |
220 | ||
9b3bb7ab SS |
221 | static ssize_t mem_used_total_show(struct device *dev, |
222 | struct device_attribute *attr, char *buf) | |
223 | { | |
224 | u64 val = 0; | |
225 | struct zram *zram = dev_to_zram(dev); | |
9b3bb7ab | 226 | |
8f7d282c | 227 | deprecated_attr_warn("mem_used_total"); |
9b3bb7ab | 228 | down_read(&zram->init_lock); |
5a99e95b WY |
229 | if (init_done(zram)) { |
230 | struct zram_meta *meta = zram->meta; | |
722cdc17 | 231 | val = zs_get_total_pages(meta->mem_pool); |
5a99e95b | 232 | } |
9b3bb7ab SS |
233 | up_read(&zram->init_lock); |
234 | ||
722cdc17 | 235 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); |
9b3bb7ab SS |
236 | } |
237 | ||
9ada9da9 MK |
238 | static ssize_t mem_limit_show(struct device *dev, |
239 | struct device_attribute *attr, char *buf) | |
240 | { | |
241 | u64 val; | |
242 | struct zram *zram = dev_to_zram(dev); | |
243 | ||
8f7d282c | 244 | deprecated_attr_warn("mem_limit"); |
9ada9da9 MK |
245 | down_read(&zram->init_lock); |
246 | val = zram->limit_pages; | |
247 | up_read(&zram->init_lock); | |
248 | ||
249 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
250 | } | |
251 | ||
252 | static ssize_t mem_limit_store(struct device *dev, | |
253 | struct device_attribute *attr, const char *buf, size_t len) | |
254 | { | |
255 | u64 limit; | |
256 | char *tmp; | |
257 | struct zram *zram = dev_to_zram(dev); | |
258 | ||
259 | limit = memparse(buf, &tmp); | |
260 | if (buf == tmp) /* no chars parsed, invalid input */ | |
261 | return -EINVAL; | |
262 | ||
263 | down_write(&zram->init_lock); | |
264 | zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT; | |
265 | up_write(&zram->init_lock); | |
266 | ||
267 | return len; | |
268 | } | |
269 | ||
461a8eee MK |
270 | static ssize_t mem_used_max_show(struct device *dev, |
271 | struct device_attribute *attr, char *buf) | |
272 | { | |
273 | u64 val = 0; | |
274 | struct zram *zram = dev_to_zram(dev); | |
275 | ||
8f7d282c | 276 | deprecated_attr_warn("mem_used_max"); |
461a8eee MK |
277 | down_read(&zram->init_lock); |
278 | if (init_done(zram)) | |
279 | val = atomic_long_read(&zram->stats.max_used_pages); | |
280 | up_read(&zram->init_lock); | |
281 | ||
282 | return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT); | |
283 | } | |
284 | ||
285 | static ssize_t mem_used_max_store(struct device *dev, | |
286 | struct device_attribute *attr, const char *buf, size_t len) | |
287 | { | |
288 | int err; | |
289 | unsigned long val; | |
290 | struct zram *zram = dev_to_zram(dev); | |
461a8eee MK |
291 | |
292 | err = kstrtoul(buf, 10, &val); | |
293 | if (err || val != 0) | |
294 | return -EINVAL; | |
295 | ||
296 | down_read(&zram->init_lock); | |
5a99e95b WY |
297 | if (init_done(zram)) { |
298 | struct zram_meta *meta = zram->meta; | |
461a8eee MK |
299 | atomic_long_set(&zram->stats.max_used_pages, |
300 | zs_get_total_pages(meta->mem_pool)); | |
5a99e95b | 301 | } |
461a8eee MK |
302 | up_read(&zram->init_lock); |
303 | ||
304 | return len; | |
305 | } | |
306 | ||
522698d7 SS |
307 | static ssize_t max_comp_streams_show(struct device *dev, |
308 | struct device_attribute *attr, char *buf) | |
309 | { | |
310 | int val; | |
311 | struct zram *zram = dev_to_zram(dev); | |
312 | ||
313 | down_read(&zram->init_lock); | |
314 | val = zram->max_comp_streams; | |
315 | up_read(&zram->init_lock); | |
316 | ||
317 | return scnprintf(buf, PAGE_SIZE, "%d\n", val); | |
318 | } | |
319 | ||
beca3ec7 SS |
320 | static ssize_t max_comp_streams_store(struct device *dev, |
321 | struct device_attribute *attr, const char *buf, size_t len) | |
322 | { | |
323 | int num; | |
324 | struct zram *zram = dev_to_zram(dev); | |
60a726e3 | 325 | int ret; |
beca3ec7 | 326 | |
60a726e3 MK |
327 | ret = kstrtoint(buf, 0, &num); |
328 | if (ret < 0) | |
329 | return ret; | |
beca3ec7 SS |
330 | if (num < 1) |
331 | return -EINVAL; | |
60a726e3 | 332 | |
beca3ec7 SS |
333 | down_write(&zram->init_lock); |
334 | if (init_done(zram)) { | |
60a726e3 | 335 | if (!zcomp_set_max_streams(zram->comp, num)) { |
fe8eb122 | 336 | pr_info("Cannot change max compression streams\n"); |
60a726e3 MK |
337 | ret = -EINVAL; |
338 | goto out; | |
339 | } | |
beca3ec7 | 340 | } |
60a726e3 | 341 | |
beca3ec7 | 342 | zram->max_comp_streams = num; |
60a726e3 MK |
343 | ret = len; |
344 | out: | |
beca3ec7 | 345 | up_write(&zram->init_lock); |
60a726e3 | 346 | return ret; |
beca3ec7 SS |
347 | } |
348 | ||
e46b8a03 SS |
349 | static ssize_t comp_algorithm_show(struct device *dev, |
350 | struct device_attribute *attr, char *buf) | |
351 | { | |
352 | size_t sz; | |
353 | struct zram *zram = dev_to_zram(dev); | |
354 | ||
355 | down_read(&zram->init_lock); | |
356 | sz = zcomp_available_show(zram->compressor, buf); | |
357 | up_read(&zram->init_lock); | |
358 | ||
359 | return sz; | |
360 | } | |
361 | ||
362 | static ssize_t comp_algorithm_store(struct device *dev, | |
363 | struct device_attribute *attr, const char *buf, size_t len) | |
364 | { | |
365 | struct zram *zram = dev_to_zram(dev); | |
4bbacd51 SS |
366 | size_t sz; |
367 | ||
e46b8a03 SS |
368 | down_write(&zram->init_lock); |
369 | if (init_done(zram)) { | |
370 | up_write(&zram->init_lock); | |
371 | pr_info("Can't change algorithm for initialized device\n"); | |
372 | return -EBUSY; | |
373 | } | |
374 | strlcpy(zram->compressor, buf, sizeof(zram->compressor)); | |
4bbacd51 SS |
375 | |
376 | /* ignore trailing newline */ | |
377 | sz = strlen(zram->compressor); | |
378 | if (sz > 0 && zram->compressor[sz - 1] == '\n') | |
379 | zram->compressor[sz - 1] = 0x00; | |
380 | ||
e46b8a03 SS |
381 | up_write(&zram->init_lock); |
382 | return len; | |
383 | } | |
384 | ||
522698d7 SS |
385 | static ssize_t compact_store(struct device *dev, |
386 | struct device_attribute *attr, const char *buf, size_t len) | |
306b0c95 | 387 | { |
522698d7 SS |
388 | unsigned long nr_migrated; |
389 | struct zram *zram = dev_to_zram(dev); | |
390 | struct zram_meta *meta; | |
306b0c95 | 391 | |
522698d7 SS |
392 | down_read(&zram->init_lock); |
393 | if (!init_done(zram)) { | |
394 | up_read(&zram->init_lock); | |
395 | return -EINVAL; | |
396 | } | |
306b0c95 | 397 | |
522698d7 SS |
398 | meta = zram->meta; |
399 | nr_migrated = zs_compact(meta->mem_pool); | |
400 | atomic64_add(nr_migrated, &zram->stats.num_migrated); | |
401 | up_read(&zram->init_lock); | |
d2d5e762 | 402 | |
522698d7 | 403 | return len; |
d2d5e762 WY |
404 | } |
405 | ||
522698d7 SS |
406 | static ssize_t io_stat_show(struct device *dev, |
407 | struct device_attribute *attr, char *buf) | |
d2d5e762 | 408 | { |
522698d7 SS |
409 | struct zram *zram = dev_to_zram(dev); |
410 | ssize_t ret; | |
d2d5e762 | 411 | |
522698d7 SS |
412 | down_read(&zram->init_lock); |
413 | ret = scnprintf(buf, PAGE_SIZE, | |
414 | "%8llu %8llu %8llu %8llu\n", | |
415 | (u64)atomic64_read(&zram->stats.failed_reads), | |
416 | (u64)atomic64_read(&zram->stats.failed_writes), | |
417 | (u64)atomic64_read(&zram->stats.invalid_io), | |
418 | (u64)atomic64_read(&zram->stats.notify_free)); | |
419 | up_read(&zram->init_lock); | |
306b0c95 | 420 | |
522698d7 | 421 | return ret; |
9b3bb7ab SS |
422 | } |
423 | ||
522698d7 SS |
424 | static ssize_t mm_stat_show(struct device *dev, |
425 | struct device_attribute *attr, char *buf) | |
9b3bb7ab | 426 | { |
522698d7 SS |
427 | struct zram *zram = dev_to_zram(dev); |
428 | u64 orig_size, mem_used = 0; | |
429 | long max_used; | |
430 | ssize_t ret; | |
a539c72a | 431 | |
522698d7 SS |
432 | down_read(&zram->init_lock); |
433 | if (init_done(zram)) | |
434 | mem_used = zs_get_total_pages(zram->meta->mem_pool); | |
9b3bb7ab | 435 | |
522698d7 SS |
436 | orig_size = atomic64_read(&zram->stats.pages_stored); |
437 | max_used = atomic_long_read(&zram->stats.max_used_pages); | |
9b3bb7ab | 438 | |
522698d7 SS |
439 | ret = scnprintf(buf, PAGE_SIZE, |
440 | "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n", | |
441 | orig_size << PAGE_SHIFT, | |
442 | (u64)atomic64_read(&zram->stats.compr_data_size), | |
443 | mem_used << PAGE_SHIFT, | |
444 | zram->limit_pages << PAGE_SHIFT, | |
445 | max_used << PAGE_SHIFT, | |
446 | (u64)atomic64_read(&zram->stats.zero_pages), | |
447 | (u64)atomic64_read(&zram->stats.num_migrated)); | |
448 | up_read(&zram->init_lock); | |
9b3bb7ab | 449 | |
522698d7 SS |
450 | return ret; |
451 | } | |
452 | ||
453 | static DEVICE_ATTR_RO(io_stat); | |
454 | static DEVICE_ATTR_RO(mm_stat); | |
455 | ZRAM_ATTR_RO(num_reads); | |
456 | ZRAM_ATTR_RO(num_writes); | |
457 | ZRAM_ATTR_RO(failed_reads); | |
458 | ZRAM_ATTR_RO(failed_writes); | |
459 | ZRAM_ATTR_RO(invalid_io); | |
460 | ZRAM_ATTR_RO(notify_free); | |
461 | ZRAM_ATTR_RO(zero_pages); | |
462 | ZRAM_ATTR_RO(compr_data_size); | |
463 | ||
464 | static inline bool zram_meta_get(struct zram *zram) | |
465 | { | |
466 | if (atomic_inc_not_zero(&zram->refcount)) | |
467 | return true; | |
468 | return false; | |
469 | } | |
470 | ||
471 | static inline void zram_meta_put(struct zram *zram) | |
472 | { | |
473 | atomic_dec(&zram->refcount); | |
474 | } | |
475 | ||
476 | static void zram_meta_free(struct zram_meta *meta, u64 disksize) | |
477 | { | |
478 | size_t num_pages = disksize >> PAGE_SHIFT; | |
479 | size_t index; | |
1fec1172 GM |
480 | |
481 | /* Free all pages that are still in this zram device */ | |
482 | for (index = 0; index < num_pages; index++) { | |
483 | unsigned long handle = meta->table[index].handle; | |
484 | ||
485 | if (!handle) | |
486 | continue; | |
487 | ||
488 | zs_free(meta->mem_pool, handle); | |
489 | } | |
490 | ||
9b3bb7ab | 491 | zs_destroy_pool(meta->mem_pool); |
9b3bb7ab SS |
492 | vfree(meta->table); |
493 | kfree(meta); | |
494 | } | |
495 | ||
3eba0c6a | 496 | static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize) |
9b3bb7ab SS |
497 | { |
498 | size_t num_pages; | |
3eba0c6a | 499 | char pool_name[8]; |
9b3bb7ab | 500 | struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL); |
b8179958 | 501 | |
9b3bb7ab | 502 | if (!meta) |
b8179958 | 503 | return NULL; |
9b3bb7ab | 504 | |
9b3bb7ab SS |
505 | num_pages = disksize >> PAGE_SHIFT; |
506 | meta->table = vzalloc(num_pages * sizeof(*meta->table)); | |
507 | if (!meta->table) { | |
508 | pr_err("Error allocating zram address table\n"); | |
b8179958 | 509 | goto out_error; |
9b3bb7ab SS |
510 | } |
511 | ||
3eba0c6a GM |
512 | snprintf(pool_name, sizeof(pool_name), "zram%d", device_id); |
513 | meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM); | |
9b3bb7ab SS |
514 | if (!meta->mem_pool) { |
515 | pr_err("Error creating memory pool\n"); | |
b8179958 | 516 | goto out_error; |
9b3bb7ab SS |
517 | } |
518 | ||
519 | return meta; | |
520 | ||
b8179958 | 521 | out_error: |
9b3bb7ab | 522 | vfree(meta->table); |
9b3bb7ab | 523 | kfree(meta); |
b8179958 | 524 | return NULL; |
9b3bb7ab SS |
525 | } |
526 | ||
d2d5e762 WY |
527 | /* |
528 | * To protect concurrent access to the same index entry, | |
529 | * caller should hold this table index entry's bit_spinlock to | |
530 | * indicate this index entry is accessing. | |
531 | */ | |
f1e3cfff | 532 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 | 533 | { |
8b3cc3ed MK |
534 | struct zram_meta *meta = zram->meta; |
535 | unsigned long handle = meta->table[index].handle; | |
306b0c95 | 536 | |
fd1a30de | 537 | if (unlikely(!handle)) { |
2e882281 NG |
538 | /* |
539 | * No memory is allocated for zero filled pages. | |
540 | * Simply clear zero page flag. | |
541 | */ | |
8b3cc3ed MK |
542 | if (zram_test_flag(meta, index, ZRAM_ZERO)) { |
543 | zram_clear_flag(meta, index, ZRAM_ZERO); | |
90a7806e | 544 | atomic64_dec(&zram->stats.zero_pages); |
306b0c95 NG |
545 | } |
546 | return; | |
547 | } | |
548 | ||
8b3cc3ed | 549 | zs_free(meta->mem_pool, handle); |
306b0c95 | 550 | |
d2d5e762 WY |
551 | atomic64_sub(zram_get_obj_size(meta, index), |
552 | &zram->stats.compr_data_size); | |
90a7806e | 553 | atomic64_dec(&zram->stats.pages_stored); |
306b0c95 | 554 | |
8b3cc3ed | 555 | meta->table[index].handle = 0; |
d2d5e762 | 556 | zram_set_obj_size(meta, index, 0); |
306b0c95 NG |
557 | } |
558 | ||
37b51fdd | 559 | static int zram_decompress_page(struct zram *zram, char *mem, u32 index) |
306b0c95 | 560 | { |
b7ca232e | 561 | int ret = 0; |
37b51fdd | 562 | unsigned char *cmem; |
8b3cc3ed | 563 | struct zram_meta *meta = zram->meta; |
92967471 | 564 | unsigned long handle; |
023b409f | 565 | size_t size; |
92967471 | 566 | |
d2d5e762 | 567 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
92967471 | 568 | handle = meta->table[index].handle; |
d2d5e762 | 569 | size = zram_get_obj_size(meta, index); |
306b0c95 | 570 | |
8b3cc3ed | 571 | if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) { |
d2d5e762 | 572 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
42e99bd9 | 573 | clear_page(mem); |
8c921b2b JM |
574 | return 0; |
575 | } | |
306b0c95 | 576 | |
8b3cc3ed | 577 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO); |
92967471 | 578 | if (size == PAGE_SIZE) |
42e99bd9 | 579 | copy_page(mem, cmem); |
37b51fdd | 580 | else |
b7ca232e | 581 | ret = zcomp_decompress(zram->comp, cmem, size, mem); |
8b3cc3ed | 582 | zs_unmap_object(meta->mem_pool, handle); |
d2d5e762 | 583 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
a1dd52af | 584 | |
8c921b2b | 585 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 586 | if (unlikely(ret)) { |
8c921b2b | 587 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); |
8c921b2b | 588 | return ret; |
a1dd52af | 589 | } |
306b0c95 | 590 | |
8c921b2b | 591 | return 0; |
306b0c95 NG |
592 | } |
593 | ||
37b51fdd | 594 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
b627cff3 | 595 | u32 index, int offset) |
924bd88d JM |
596 | { |
597 | int ret; | |
37b51fdd SS |
598 | struct page *page; |
599 | unsigned char *user_mem, *uncmem = NULL; | |
8b3cc3ed | 600 | struct zram_meta *meta = zram->meta; |
37b51fdd SS |
601 | page = bvec->bv_page; |
602 | ||
d2d5e762 | 603 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
8b3cc3ed MK |
604 | if (unlikely(!meta->table[index].handle) || |
605 | zram_test_flag(meta, index, ZRAM_ZERO)) { | |
d2d5e762 | 606 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
37b51fdd | 607 | handle_zero_page(bvec); |
924bd88d JM |
608 | return 0; |
609 | } | |
d2d5e762 | 610 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
924bd88d | 611 | |
37b51fdd SS |
612 | if (is_partial_io(bvec)) |
613 | /* Use a temporary buffer to decompress the page */ | |
7e5a5104 MK |
614 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
615 | ||
616 | user_mem = kmap_atomic(page); | |
617 | if (!is_partial_io(bvec)) | |
37b51fdd SS |
618 | uncmem = user_mem; |
619 | ||
620 | if (!uncmem) { | |
621 | pr_info("Unable to allocate temp memory\n"); | |
622 | ret = -ENOMEM; | |
623 | goto out_cleanup; | |
624 | } | |
924bd88d | 625 | |
37b51fdd | 626 | ret = zram_decompress_page(zram, uncmem, index); |
924bd88d | 627 | /* Should NEVER happen. Return bio error if it does. */ |
b7ca232e | 628 | if (unlikely(ret)) |
37b51fdd | 629 | goto out_cleanup; |
924bd88d | 630 | |
37b51fdd SS |
631 | if (is_partial_io(bvec)) |
632 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
633 | bvec->bv_len); | |
634 | ||
635 | flush_dcache_page(page); | |
636 | ret = 0; | |
637 | out_cleanup: | |
638 | kunmap_atomic(user_mem); | |
639 | if (is_partial_io(bvec)) | |
640 | kfree(uncmem); | |
641 | return ret; | |
924bd88d JM |
642 | } |
643 | ||
644 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
645 | int offset) | |
306b0c95 | 646 | { |
397c6066 | 647 | int ret = 0; |
8c921b2b | 648 | size_t clen; |
c2344348 | 649 | unsigned long handle; |
130f315a | 650 | struct page *page; |
924bd88d | 651 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
8b3cc3ed | 652 | struct zram_meta *meta = zram->meta; |
17162f41 | 653 | struct zcomp_strm *zstrm = NULL; |
461a8eee | 654 | unsigned long alloced_pages; |
306b0c95 | 655 | |
8c921b2b | 656 | page = bvec->bv_page; |
924bd88d JM |
657 | if (is_partial_io(bvec)) { |
658 | /* | |
659 | * This is a partial IO. We need to read the full page | |
660 | * before to write the changes. | |
661 | */ | |
7e5a5104 | 662 | uncmem = kmalloc(PAGE_SIZE, GFP_NOIO); |
924bd88d | 663 | if (!uncmem) { |
924bd88d JM |
664 | ret = -ENOMEM; |
665 | goto out; | |
666 | } | |
37b51fdd | 667 | ret = zram_decompress_page(zram, uncmem, index); |
397c6066 | 668 | if (ret) |
924bd88d | 669 | goto out; |
924bd88d JM |
670 | } |
671 | ||
b7ca232e | 672 | zstrm = zcomp_strm_find(zram->comp); |
ba82fe2e | 673 | user_mem = kmap_atomic(page); |
924bd88d | 674 | |
397c6066 | 675 | if (is_partial_io(bvec)) { |
924bd88d JM |
676 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, |
677 | bvec->bv_len); | |
397c6066 NG |
678 | kunmap_atomic(user_mem); |
679 | user_mem = NULL; | |
680 | } else { | |
924bd88d | 681 | uncmem = user_mem; |
397c6066 | 682 | } |
924bd88d JM |
683 | |
684 | if (page_zero_filled(uncmem)) { | |
c4065152 WY |
685 | if (user_mem) |
686 | kunmap_atomic(user_mem); | |
f40ac2ae | 687 | /* Free memory associated with this sector now. */ |
d2d5e762 | 688 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 689 | zram_free_page(zram, index); |
92967471 | 690 | zram_set_flag(meta, index, ZRAM_ZERO); |
d2d5e762 | 691 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae | 692 | |
90a7806e | 693 | atomic64_inc(&zram->stats.zero_pages); |
924bd88d JM |
694 | ret = 0; |
695 | goto out; | |
8c921b2b | 696 | } |
306b0c95 | 697 | |
b7ca232e | 698 | ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen); |
397c6066 NG |
699 | if (!is_partial_io(bvec)) { |
700 | kunmap_atomic(user_mem); | |
701 | user_mem = NULL; | |
702 | uncmem = NULL; | |
703 | } | |
306b0c95 | 704 | |
b7ca232e | 705 | if (unlikely(ret)) { |
8c921b2b | 706 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 707 | goto out; |
8c921b2b | 708 | } |
b7ca232e | 709 | src = zstrm->buffer; |
c8f2f0db | 710 | if (unlikely(clen > max_zpage_size)) { |
c8f2f0db | 711 | clen = PAGE_SIZE; |
397c6066 NG |
712 | if (is_partial_io(bvec)) |
713 | src = uncmem; | |
c8f2f0db | 714 | } |
a1dd52af | 715 | |
8b3cc3ed | 716 | handle = zs_malloc(meta->mem_pool, clen); |
fd1a30de | 717 | if (!handle) { |
596b3dd4 MR |
718 | pr_info("Error allocating memory for compressed page: %u, size=%zu\n", |
719 | index, clen); | |
924bd88d JM |
720 | ret = -ENOMEM; |
721 | goto out; | |
8c921b2b | 722 | } |
9ada9da9 | 723 | |
461a8eee MK |
724 | alloced_pages = zs_get_total_pages(meta->mem_pool); |
725 | if (zram->limit_pages && alloced_pages > zram->limit_pages) { | |
9ada9da9 MK |
726 | zs_free(meta->mem_pool, handle); |
727 | ret = -ENOMEM; | |
728 | goto out; | |
729 | } | |
730 | ||
461a8eee MK |
731 | update_used_max(zram, alloced_pages); |
732 | ||
8b3cc3ed | 733 | cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO); |
306b0c95 | 734 | |
42e99bd9 | 735 | if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) { |
397c6066 | 736 | src = kmap_atomic(page); |
42e99bd9 | 737 | copy_page(cmem, src); |
397c6066 | 738 | kunmap_atomic(src); |
42e99bd9 JL |
739 | } else { |
740 | memcpy(cmem, src, clen); | |
741 | } | |
306b0c95 | 742 | |
b7ca232e | 743 | zcomp_strm_release(zram->comp, zstrm); |
17162f41 | 744 | zstrm = NULL; |
8b3cc3ed | 745 | zs_unmap_object(meta->mem_pool, handle); |
fd1a30de | 746 | |
f40ac2ae SS |
747 | /* |
748 | * Free memory associated with this sector | |
749 | * before overwriting unused sectors. | |
750 | */ | |
d2d5e762 | 751 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f40ac2ae SS |
752 | zram_free_page(zram, index); |
753 | ||
8b3cc3ed | 754 | meta->table[index].handle = handle; |
d2d5e762 WY |
755 | zram_set_obj_size(meta, index, clen); |
756 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); | |
306b0c95 | 757 | |
8c921b2b | 758 | /* Update stats */ |
90a7806e SS |
759 | atomic64_add(clen, &zram->stats.compr_data_size); |
760 | atomic64_inc(&zram->stats.pages_stored); | |
924bd88d | 761 | out: |
17162f41 | 762 | if (zstrm) |
b7ca232e | 763 | zcomp_strm_release(zram->comp, zstrm); |
397c6066 NG |
764 | if (is_partial_io(bvec)) |
765 | kfree(uncmem); | |
924bd88d | 766 | return ret; |
8c921b2b JM |
767 | } |
768 | ||
f4659d8e JK |
769 | /* |
770 | * zram_bio_discard - handler on discard request | |
771 | * @index: physical block index in PAGE_SIZE units | |
772 | * @offset: byte offset within physical block | |
773 | */ | |
774 | static void zram_bio_discard(struct zram *zram, u32 index, | |
775 | int offset, struct bio *bio) | |
776 | { | |
777 | size_t n = bio->bi_iter.bi_size; | |
d2d5e762 | 778 | struct zram_meta *meta = zram->meta; |
f4659d8e JK |
779 | |
780 | /* | |
781 | * zram manages data in physical block size units. Because logical block | |
782 | * size isn't identical with physical block size on some arch, we | |
783 | * could get a discard request pointing to a specific offset within a | |
784 | * certain physical block. Although we can handle this request by | |
785 | * reading that physiclal block and decompressing and partially zeroing | |
786 | * and re-compressing and then re-storing it, this isn't reasonable | |
787 | * because our intent with a discard request is to save memory. So | |
788 | * skipping this logical block is appropriate here. | |
789 | */ | |
790 | if (offset) { | |
38515c73 | 791 | if (n <= (PAGE_SIZE - offset)) |
f4659d8e JK |
792 | return; |
793 | ||
38515c73 | 794 | n -= (PAGE_SIZE - offset); |
f4659d8e JK |
795 | index++; |
796 | } | |
797 | ||
798 | while (n >= PAGE_SIZE) { | |
d2d5e762 | 799 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f4659d8e | 800 | zram_free_page(zram, index); |
d2d5e762 | 801 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
015254da | 802 | atomic64_inc(&zram->stats.notify_free); |
f4659d8e JK |
803 | index++; |
804 | n -= PAGE_SIZE; | |
805 | } | |
806 | } | |
807 | ||
522698d7 SS |
808 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, |
809 | int offset, int rw) | |
9b3bb7ab | 810 | { |
522698d7 | 811 | unsigned long start_time = jiffies; |
9b3bb7ab | 812 | int ret; |
9b3bb7ab | 813 | |
522698d7 SS |
814 | generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT, |
815 | &zram->disk->part0); | |
46a51c80 | 816 | |
522698d7 SS |
817 | if (rw == READ) { |
818 | atomic64_inc(&zram->stats.num_reads); | |
819 | ret = zram_bvec_read(zram, bvec, index, offset); | |
820 | } else { | |
821 | atomic64_inc(&zram->stats.num_writes); | |
822 | ret = zram_bvec_write(zram, bvec, index, offset); | |
1b672224 | 823 | } |
9b3bb7ab | 824 | |
522698d7 | 825 | generic_end_io_acct(rw, &zram->disk->part0, start_time); |
9b3bb7ab | 826 | |
522698d7 SS |
827 | if (unlikely(ret)) { |
828 | if (rw == READ) | |
829 | atomic64_inc(&zram->stats.failed_reads); | |
830 | else | |
831 | atomic64_inc(&zram->stats.failed_writes); | |
1b672224 | 832 | } |
9b3bb7ab | 833 | |
1b672224 | 834 | return ret; |
8c921b2b JM |
835 | } |
836 | ||
be257c61 | 837 | static void __zram_make_request(struct zram *zram, struct bio *bio) |
8c921b2b | 838 | { |
b627cff3 | 839 | int offset, rw; |
8c921b2b | 840 | u32 index; |
7988613b KO |
841 | struct bio_vec bvec; |
842 | struct bvec_iter iter; | |
8c921b2b | 843 | |
4f024f37 KO |
844 | index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT; |
845 | offset = (bio->bi_iter.bi_sector & | |
846 | (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; | |
8c921b2b | 847 | |
f4659d8e JK |
848 | if (unlikely(bio->bi_rw & REQ_DISCARD)) { |
849 | zram_bio_discard(zram, index, offset, bio); | |
850 | bio_endio(bio, 0); | |
851 | return; | |
852 | } | |
853 | ||
b627cff3 | 854 | rw = bio_data_dir(bio); |
7988613b | 855 | bio_for_each_segment(bvec, bio, iter) { |
924bd88d JM |
856 | int max_transfer_size = PAGE_SIZE - offset; |
857 | ||
7988613b | 858 | if (bvec.bv_len > max_transfer_size) { |
924bd88d JM |
859 | /* |
860 | * zram_bvec_rw() can only make operation on a single | |
861 | * zram page. Split the bio vector. | |
862 | */ | |
863 | struct bio_vec bv; | |
864 | ||
7988613b | 865 | bv.bv_page = bvec.bv_page; |
924bd88d | 866 | bv.bv_len = max_transfer_size; |
7988613b | 867 | bv.bv_offset = bvec.bv_offset; |
924bd88d | 868 | |
b627cff3 | 869 | if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0) |
924bd88d JM |
870 | goto out; |
871 | ||
7988613b | 872 | bv.bv_len = bvec.bv_len - max_transfer_size; |
924bd88d | 873 | bv.bv_offset += max_transfer_size; |
b627cff3 | 874 | if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0) |
924bd88d JM |
875 | goto out; |
876 | } else | |
b627cff3 | 877 | if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0) |
924bd88d JM |
878 | goto out; |
879 | ||
7988613b | 880 | update_position(&index, &offset, &bvec); |
a1dd52af | 881 | } |
306b0c95 NG |
882 | |
883 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
884 | bio_endio(bio, 0); | |
7d7854b4 | 885 | return; |
306b0c95 NG |
886 | |
887 | out: | |
306b0c95 | 888 | bio_io_error(bio); |
306b0c95 NG |
889 | } |
890 | ||
306b0c95 | 891 | /* |
f1e3cfff | 892 | * Handler function for all zram I/O requests. |
306b0c95 | 893 | */ |
5a7bbad2 | 894 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 895 | { |
f1e3cfff | 896 | struct zram *zram = queue->queuedata; |
306b0c95 | 897 | |
08eee69f | 898 | if (unlikely(!zram_meta_get(zram))) |
3de738cd | 899 | goto error; |
0900beae | 900 | |
54850e73 | 901 | if (!valid_io_request(zram, bio->bi_iter.bi_sector, |
902 | bio->bi_iter.bi_size)) { | |
da5cc7d3 | 903 | atomic64_inc(&zram->stats.invalid_io); |
08eee69f | 904 | goto put_zram; |
6642a67c JM |
905 | } |
906 | ||
be257c61 | 907 | __zram_make_request(zram, bio); |
08eee69f | 908 | zram_meta_put(zram); |
b4fdcb02 | 909 | return; |
08eee69f MK |
910 | put_zram: |
911 | zram_meta_put(zram); | |
0900beae JM |
912 | error: |
913 | bio_io_error(bio); | |
306b0c95 NG |
914 | } |
915 | ||
2ccbec05 NG |
916 | static void zram_slot_free_notify(struct block_device *bdev, |
917 | unsigned long index) | |
107c161b | 918 | { |
f1e3cfff | 919 | struct zram *zram; |
f614a9f4 | 920 | struct zram_meta *meta; |
107c161b | 921 | |
f1e3cfff | 922 | zram = bdev->bd_disk->private_data; |
f614a9f4 | 923 | meta = zram->meta; |
a0c516cb | 924 | |
d2d5e762 | 925 | bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 926 | zram_free_page(zram, index); |
d2d5e762 | 927 | bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value); |
f614a9f4 | 928 | atomic64_inc(&zram->stats.notify_free); |
107c161b NG |
929 | } |
930 | ||
8c7f0102 | 931 | static int zram_rw_page(struct block_device *bdev, sector_t sector, |
932 | struct page *page, int rw) | |
933 | { | |
08eee69f | 934 | int offset, err = -EIO; |
8c7f0102 | 935 | u32 index; |
936 | struct zram *zram; | |
937 | struct bio_vec bv; | |
938 | ||
939 | zram = bdev->bd_disk->private_data; | |
08eee69f MK |
940 | if (unlikely(!zram_meta_get(zram))) |
941 | goto out; | |
942 | ||
8c7f0102 | 943 | if (!valid_io_request(zram, sector, PAGE_SIZE)) { |
944 | atomic64_inc(&zram->stats.invalid_io); | |
08eee69f MK |
945 | err = -EINVAL; |
946 | goto put_zram; | |
8c7f0102 | 947 | } |
948 | ||
949 | index = sector >> SECTORS_PER_PAGE_SHIFT; | |
950 | offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT; | |
951 | ||
952 | bv.bv_page = page; | |
953 | bv.bv_len = PAGE_SIZE; | |
954 | bv.bv_offset = 0; | |
955 | ||
956 | err = zram_bvec_rw(zram, &bv, index, offset, rw); | |
08eee69f MK |
957 | put_zram: |
958 | zram_meta_put(zram); | |
959 | out: | |
8c7f0102 | 960 | /* |
961 | * If I/O fails, just return error(ie, non-zero) without | |
962 | * calling page_endio. | |
963 | * It causes resubmit the I/O with bio request by upper functions | |
964 | * of rw_page(e.g., swap_readpage, __swap_writepage) and | |
965 | * bio->bi_end_io does things to handle the error | |
966 | * (e.g., SetPageError, set_page_dirty and extra works). | |
967 | */ | |
968 | if (err == 0) | |
969 | page_endio(page, rw, 0); | |
970 | return err; | |
971 | } | |
972 | ||
522698d7 SS |
973 | static void zram_reset_device(struct zram *zram) |
974 | { | |
975 | struct zram_meta *meta; | |
976 | struct zcomp *comp; | |
977 | u64 disksize; | |
306b0c95 | 978 | |
522698d7 | 979 | down_write(&zram->init_lock); |
9b3bb7ab | 980 | |
522698d7 SS |
981 | zram->limit_pages = 0; |
982 | ||
983 | if (!init_done(zram)) { | |
984 | up_write(&zram->init_lock); | |
985 | return; | |
986 | } | |
987 | ||
988 | meta = zram->meta; | |
989 | comp = zram->comp; | |
990 | disksize = zram->disksize; | |
991 | /* | |
992 | * Refcount will go down to 0 eventually and r/w handler | |
993 | * cannot handle further I/O so it will bail out by | |
994 | * check zram_meta_get. | |
995 | */ | |
996 | zram_meta_put(zram); | |
997 | /* | |
998 | * We want to free zram_meta in process context to avoid | |
999 | * deadlock between reclaim path and any other locks. | |
1000 | */ | |
1001 | wait_event(zram->io_done, atomic_read(&zram->refcount) == 0); | |
1002 | ||
1003 | /* Reset stats */ | |
1004 | memset(&zram->stats, 0, sizeof(zram->stats)); | |
1005 | zram->disksize = 0; | |
1006 | zram->max_comp_streams = 1; | |
1007 | ||
1008 | set_capacity(zram->disk, 0); | |
1009 | part_stat_set_all(&zram->disk->part0, 0); | |
1010 | ||
1011 | up_write(&zram->init_lock); | |
1012 | /* I/O operation under all of CPU are done so let's free */ | |
1013 | zram_meta_free(meta, disksize); | |
1014 | zcomp_destroy(comp); | |
1015 | } | |
1016 | ||
1017 | static ssize_t disksize_store(struct device *dev, | |
1018 | struct device_attribute *attr, const char *buf, size_t len) | |
2f6a3bed | 1019 | { |
522698d7 SS |
1020 | u64 disksize; |
1021 | struct zcomp *comp; | |
1022 | struct zram_meta *meta; | |
2f6a3bed | 1023 | struct zram *zram = dev_to_zram(dev); |
522698d7 | 1024 | int err; |
2f6a3bed | 1025 | |
522698d7 SS |
1026 | disksize = memparse(buf, NULL); |
1027 | if (!disksize) | |
1028 | return -EINVAL; | |
2f6a3bed | 1029 | |
522698d7 SS |
1030 | disksize = PAGE_ALIGN(disksize); |
1031 | meta = zram_meta_alloc(zram->disk->first_minor, disksize); | |
1032 | if (!meta) | |
1033 | return -ENOMEM; | |
1034 | ||
1035 | comp = zcomp_create(zram->compressor, zram->max_comp_streams); | |
1036 | if (IS_ERR(comp)) { | |
1037 | pr_info("Cannot initialise %s compressing backend\n", | |
1038 | zram->compressor); | |
1039 | err = PTR_ERR(comp); | |
1040 | goto out_free_meta; | |
1041 | } | |
1042 | ||
1043 | down_write(&zram->init_lock); | |
1044 | if (init_done(zram)) { | |
1045 | pr_info("Cannot change disksize for initialized device\n"); | |
1046 | err = -EBUSY; | |
1047 | goto out_destroy_comp; | |
1048 | } | |
1049 | ||
1050 | init_waitqueue_head(&zram->io_done); | |
1051 | atomic_set(&zram->refcount, 1); | |
1052 | zram->meta = meta; | |
1053 | zram->comp = comp; | |
1054 | zram->disksize = disksize; | |
1055 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); | |
1056 | up_write(&zram->init_lock); | |
1057 | ||
1058 | /* | |
1059 | * Revalidate disk out of the init_lock to avoid lockdep splat. | |
1060 | * It's okay because disk's capacity is protected by init_lock | |
1061 | * so that revalidate_disk always sees up-to-date capacity. | |
1062 | */ | |
1063 | revalidate_disk(zram->disk); | |
1064 | ||
1065 | return len; | |
1066 | ||
1067 | out_destroy_comp: | |
1068 | up_write(&zram->init_lock); | |
1069 | zcomp_destroy(comp); | |
1070 | out_free_meta: | |
1071 | zram_meta_free(meta, disksize); | |
1072 | return err; | |
2f6a3bed SS |
1073 | } |
1074 | ||
522698d7 SS |
1075 | static ssize_t reset_store(struct device *dev, |
1076 | struct device_attribute *attr, const char *buf, size_t len) | |
4f2109f6 | 1077 | { |
522698d7 SS |
1078 | int ret; |
1079 | unsigned short do_reset; | |
1080 | struct zram *zram; | |
1081 | struct block_device *bdev; | |
4f2109f6 | 1082 | |
f405c445 SS |
1083 | ret = kstrtou16(buf, 10, &do_reset); |
1084 | if (ret) | |
1085 | return ret; | |
1086 | ||
1087 | if (!do_reset) | |
1088 | return -EINVAL; | |
1089 | ||
522698d7 SS |
1090 | zram = dev_to_zram(dev); |
1091 | bdev = bdget_disk(zram->disk, 0); | |
522698d7 SS |
1092 | if (!bdev) |
1093 | return -ENOMEM; | |
4f2109f6 | 1094 | |
522698d7 | 1095 | mutex_lock(&bdev->bd_mutex); |
f405c445 SS |
1096 | /* Do not reset an active device or claimed device */ |
1097 | if (bdev->bd_openers || zram->claim) { | |
1098 | mutex_unlock(&bdev->bd_mutex); | |
1099 | bdput(bdev); | |
1100 | return -EBUSY; | |
522698d7 SS |
1101 | } |
1102 | ||
f405c445 SS |
1103 | /* From now on, anyone can't open /dev/zram[0-9] */ |
1104 | zram->claim = true; | |
1105 | mutex_unlock(&bdev->bd_mutex); | |
522698d7 | 1106 | |
f405c445 | 1107 | /* Make sure all the pending I/O are finished */ |
522698d7 SS |
1108 | fsync_bdev(bdev); |
1109 | zram_reset_device(zram); | |
522698d7 SS |
1110 | revalidate_disk(zram->disk); |
1111 | bdput(bdev); | |
1112 | ||
f405c445 SS |
1113 | mutex_lock(&bdev->bd_mutex); |
1114 | zram->claim = false; | |
1115 | mutex_unlock(&bdev->bd_mutex); | |
1116 | ||
522698d7 | 1117 | return len; |
f405c445 SS |
1118 | } |
1119 | ||
1120 | static int zram_open(struct block_device *bdev, fmode_t mode) | |
1121 | { | |
1122 | int ret = 0; | |
1123 | struct zram *zram; | |
1124 | ||
1125 | WARN_ON(!mutex_is_locked(&bdev->bd_mutex)); | |
1126 | ||
1127 | zram = bdev->bd_disk->private_data; | |
1128 | /* zram was claimed to reset so open request fails */ | |
1129 | if (zram->claim) | |
1130 | ret = -EBUSY; | |
4f2109f6 SS |
1131 | |
1132 | return ret; | |
1133 | } | |
1134 | ||
522698d7 | 1135 | static const struct block_device_operations zram_devops = { |
f405c445 | 1136 | .open = zram_open, |
522698d7 SS |
1137 | .swap_slot_free_notify = zram_slot_free_notify, |
1138 | .rw_page = zram_rw_page, | |
1139 | .owner = THIS_MODULE | |
1140 | }; | |
1141 | ||
1142 | static DEVICE_ATTR_WO(compact); | |
1143 | static DEVICE_ATTR_RW(disksize); | |
1144 | static DEVICE_ATTR_RO(initstate); | |
1145 | static DEVICE_ATTR_WO(reset); | |
1146 | static DEVICE_ATTR_RO(orig_data_size); | |
1147 | static DEVICE_ATTR_RO(mem_used_total); | |
1148 | static DEVICE_ATTR_RW(mem_limit); | |
1149 | static DEVICE_ATTR_RW(mem_used_max); | |
1150 | static DEVICE_ATTR_RW(max_comp_streams); | |
1151 | static DEVICE_ATTR_RW(comp_algorithm); | |
a68eb3b6 | 1152 | |
9b3bb7ab SS |
1153 | static struct attribute *zram_disk_attrs[] = { |
1154 | &dev_attr_disksize.attr, | |
1155 | &dev_attr_initstate.attr, | |
1156 | &dev_attr_reset.attr, | |
1157 | &dev_attr_num_reads.attr, | |
1158 | &dev_attr_num_writes.attr, | |
64447249 SS |
1159 | &dev_attr_failed_reads.attr, |
1160 | &dev_attr_failed_writes.attr, | |
99ebbd30 | 1161 | &dev_attr_compact.attr, |
9b3bb7ab SS |
1162 | &dev_attr_invalid_io.attr, |
1163 | &dev_attr_notify_free.attr, | |
1164 | &dev_attr_zero_pages.attr, | |
1165 | &dev_attr_orig_data_size.attr, | |
1166 | &dev_attr_compr_data_size.attr, | |
1167 | &dev_attr_mem_used_total.attr, | |
9ada9da9 | 1168 | &dev_attr_mem_limit.attr, |
461a8eee | 1169 | &dev_attr_mem_used_max.attr, |
beca3ec7 | 1170 | &dev_attr_max_comp_streams.attr, |
e46b8a03 | 1171 | &dev_attr_comp_algorithm.attr, |
2f6a3bed | 1172 | &dev_attr_io_stat.attr, |
4f2109f6 | 1173 | &dev_attr_mm_stat.attr, |
9b3bb7ab SS |
1174 | NULL, |
1175 | }; | |
1176 | ||
1177 | static struct attribute_group zram_disk_attr_group = { | |
1178 | .attrs = zram_disk_attrs, | |
1179 | }; | |
1180 | ||
92ff1528 SS |
1181 | /* |
1182 | * Allocate and initialize new zram device. the function returns | |
1183 | * '>= 0' device_id upon success, and negative value otherwise. | |
1184 | */ | |
1185 | static int zram_add(void) | |
306b0c95 | 1186 | { |
85508ec6 | 1187 | struct zram *zram; |
ee980160 | 1188 | struct request_queue *queue; |
92ff1528 | 1189 | int ret, device_id; |
85508ec6 SS |
1190 | |
1191 | zram = kzalloc(sizeof(struct zram), GFP_KERNEL); | |
1192 | if (!zram) | |
1193 | return -ENOMEM; | |
1194 | ||
92ff1528 | 1195 | ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL); |
85508ec6 SS |
1196 | if (ret < 0) |
1197 | goto out_free_dev; | |
92ff1528 | 1198 | device_id = ret; |
de1a21a0 | 1199 | |
0900beae | 1200 | init_rwsem(&zram->init_lock); |
306b0c95 | 1201 | |
ee980160 SS |
1202 | queue = blk_alloc_queue(GFP_KERNEL); |
1203 | if (!queue) { | |
306b0c95 NG |
1204 | pr_err("Error allocating disk queue for device %d\n", |
1205 | device_id); | |
85508ec6 SS |
1206 | ret = -ENOMEM; |
1207 | goto out_free_idr; | |
306b0c95 NG |
1208 | } |
1209 | ||
ee980160 | 1210 | blk_queue_make_request(queue, zram_make_request); |
306b0c95 | 1211 | |
85508ec6 | 1212 | /* gendisk structure */ |
f1e3cfff NG |
1213 | zram->disk = alloc_disk(1); |
1214 | if (!zram->disk) { | |
94b8435f | 1215 | pr_warn("Error allocating disk structure for device %d\n", |
306b0c95 | 1216 | device_id); |
201c7b72 | 1217 | ret = -ENOMEM; |
39a9b8ac | 1218 | goto out_free_queue; |
306b0c95 NG |
1219 | } |
1220 | ||
f1e3cfff NG |
1221 | zram->disk->major = zram_major; |
1222 | zram->disk->first_minor = device_id; | |
1223 | zram->disk->fops = &zram_devops; | |
ee980160 SS |
1224 | zram->disk->queue = queue; |
1225 | zram->disk->queue->queuedata = zram; | |
f1e3cfff NG |
1226 | zram->disk->private_data = zram; |
1227 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 1228 | |
33863c21 | 1229 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 1230 | set_capacity(zram->disk, 0); |
b67d1ec1 SS |
1231 | /* zram devices sort of resembles non-rotational disks */ |
1232 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
b277da0a | 1233 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue); |
a1dd52af NG |
1234 | /* |
1235 | * To ensure that we always get PAGE_SIZE aligned | |
1236 | * and n*PAGE_SIZED sized I/O requests. | |
1237 | */ | |
f1e3cfff | 1238 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
1239 | blk_queue_logical_block_size(zram->disk->queue, |
1240 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
1241 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
1242 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
f4659d8e JK |
1243 | zram->disk->queue->limits.discard_granularity = PAGE_SIZE; |
1244 | zram->disk->queue->limits.max_discard_sectors = UINT_MAX; | |
1245 | /* | |
1246 | * zram_bio_discard() will clear all logical blocks if logical block | |
1247 | * size is identical with physical block size(PAGE_SIZE). But if it is | |
1248 | * different, we will skip discarding some parts of logical blocks in | |
1249 | * the part of the request range which isn't aligned to physical block | |
1250 | * size. So we can't ensure that all discarded logical blocks are | |
1251 | * zeroed. | |
1252 | */ | |
1253 | if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE) | |
1254 | zram->disk->queue->limits.discard_zeroes_data = 1; | |
1255 | else | |
1256 | zram->disk->queue->limits.discard_zeroes_data = 0; | |
1257 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue); | |
5d83d5a0 | 1258 | |
f1e3cfff | 1259 | add_disk(zram->disk); |
306b0c95 | 1260 | |
33863c21 NG |
1261 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
1262 | &zram_disk_attr_group); | |
1263 | if (ret < 0) { | |
94b8435f | 1264 | pr_warn("Error creating sysfs group"); |
39a9b8ac | 1265 | goto out_free_disk; |
33863c21 | 1266 | } |
e46b8a03 | 1267 | strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor)); |
be2d1d56 | 1268 | zram->meta = NULL; |
beca3ec7 | 1269 | zram->max_comp_streams = 1; |
d12b63c9 SS |
1270 | |
1271 | pr_info("Added device: %s\n", zram->disk->disk_name); | |
92ff1528 | 1272 | return device_id; |
de1a21a0 | 1273 | |
39a9b8ac JL |
1274 | out_free_disk: |
1275 | del_gendisk(zram->disk); | |
1276 | put_disk(zram->disk); | |
1277 | out_free_queue: | |
ee980160 | 1278 | blk_cleanup_queue(queue); |
85508ec6 SS |
1279 | out_free_idr: |
1280 | idr_remove(&zram_index_idr, device_id); | |
1281 | out_free_dev: | |
1282 | kfree(zram); | |
de1a21a0 | 1283 | return ret; |
306b0c95 NG |
1284 | } |
1285 | ||
6566d1a3 | 1286 | static int zram_remove(struct zram *zram) |
306b0c95 | 1287 | { |
6566d1a3 SS |
1288 | struct block_device *bdev; |
1289 | ||
1290 | bdev = bdget_disk(zram->disk, 0); | |
1291 | if (!bdev) | |
1292 | return -ENOMEM; | |
1293 | ||
1294 | mutex_lock(&bdev->bd_mutex); | |
1295 | if (bdev->bd_openers || zram->claim) { | |
1296 | mutex_unlock(&bdev->bd_mutex); | |
1297 | bdput(bdev); | |
1298 | return -EBUSY; | |
1299 | } | |
1300 | ||
1301 | zram->claim = true; | |
1302 | mutex_unlock(&bdev->bd_mutex); | |
1303 | ||
85508ec6 SS |
1304 | /* |
1305 | * Remove sysfs first, so no one will perform a disksize | |
6566d1a3 SS |
1306 | * store while we destroy the devices. This also helps during |
1307 | * hot_remove -- zram_reset_device() is the last holder of | |
1308 | * ->init_lock, no later/concurrent disksize_store() or any | |
1309 | * other sysfs handlers are possible. | |
85508ec6 SS |
1310 | */ |
1311 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, | |
1312 | &zram_disk_attr_group); | |
306b0c95 | 1313 | |
6566d1a3 SS |
1314 | /* Make sure all the pending I/O are finished */ |
1315 | fsync_bdev(bdev); | |
85508ec6 | 1316 | zram_reset_device(zram); |
6566d1a3 SS |
1317 | bdput(bdev); |
1318 | ||
1319 | pr_info("Removed device: %s\n", zram->disk->disk_name); | |
1320 | ||
85508ec6 SS |
1321 | idr_remove(&zram_index_idr, zram->disk->first_minor); |
1322 | blk_cleanup_queue(zram->disk->queue); | |
1323 | del_gendisk(zram->disk); | |
1324 | put_disk(zram->disk); | |
1325 | kfree(zram); | |
6566d1a3 SS |
1326 | return 0; |
1327 | } | |
1328 | ||
1329 | /* zram-control sysfs attributes */ | |
1330 | static ssize_t hot_add_show(struct class *class, | |
1331 | struct class_attribute *attr, | |
1332 | char *buf) | |
1333 | { | |
1334 | int ret; | |
1335 | ||
1336 | mutex_lock(&zram_index_mutex); | |
1337 | ret = zram_add(); | |
1338 | mutex_unlock(&zram_index_mutex); | |
1339 | ||
1340 | if (ret < 0) | |
1341 | return ret; | |
1342 | return scnprintf(buf, PAGE_SIZE, "%d\n", ret); | |
1343 | } | |
1344 | ||
1345 | static ssize_t hot_remove_store(struct class *class, | |
1346 | struct class_attribute *attr, | |
1347 | const char *buf, | |
1348 | size_t count) | |
1349 | { | |
1350 | struct zram *zram; | |
1351 | int ret, dev_id; | |
1352 | ||
1353 | /* dev_id is gendisk->first_minor, which is `int' */ | |
1354 | ret = kstrtoint(buf, 10, &dev_id); | |
1355 | if (ret) | |
1356 | return ret; | |
1357 | if (dev_id < 0) | |
1358 | return -EINVAL; | |
1359 | ||
1360 | mutex_lock(&zram_index_mutex); | |
1361 | ||
1362 | zram = idr_find(&zram_index_idr, dev_id); | |
1363 | if (zram) | |
1364 | ret = zram_remove(zram); | |
1365 | else | |
1366 | ret = -ENODEV; | |
1367 | ||
1368 | mutex_unlock(&zram_index_mutex); | |
1369 | return ret ? ret : count; | |
85508ec6 | 1370 | } |
a096cafc | 1371 | |
6566d1a3 SS |
1372 | static struct class_attribute zram_control_class_attrs[] = { |
1373 | __ATTR_RO(hot_add), | |
1374 | __ATTR_WO(hot_remove), | |
1375 | __ATTR_NULL, | |
1376 | }; | |
1377 | ||
1378 | static struct class zram_control_class = { | |
1379 | .name = "zram-control", | |
1380 | .owner = THIS_MODULE, | |
1381 | .class_attrs = zram_control_class_attrs, | |
1382 | }; | |
1383 | ||
85508ec6 SS |
1384 | static int zram_remove_cb(int id, void *ptr, void *data) |
1385 | { | |
1386 | zram_remove(ptr); | |
1387 | return 0; | |
1388 | } | |
a096cafc | 1389 | |
85508ec6 SS |
1390 | static void destroy_devices(void) |
1391 | { | |
6566d1a3 | 1392 | class_unregister(&zram_control_class); |
85508ec6 SS |
1393 | idr_for_each(&zram_index_idr, &zram_remove_cb, NULL); |
1394 | idr_destroy(&zram_index_idr); | |
a096cafc | 1395 | unregister_blkdev(zram_major, "zram"); |
306b0c95 NG |
1396 | } |
1397 | ||
f1e3cfff | 1398 | static int __init zram_init(void) |
306b0c95 | 1399 | { |
92ff1528 | 1400 | int ret; |
306b0c95 | 1401 | |
6566d1a3 SS |
1402 | ret = class_register(&zram_control_class); |
1403 | if (ret) { | |
1404 | pr_warn("Unable to register zram-control class\n"); | |
1405 | return ret; | |
1406 | } | |
1407 | ||
f1e3cfff NG |
1408 | zram_major = register_blkdev(0, "zram"); |
1409 | if (zram_major <= 0) { | |
94b8435f | 1410 | pr_warn("Unable to get major number\n"); |
6566d1a3 | 1411 | class_unregister(&zram_control_class); |
a096cafc | 1412 | return -EBUSY; |
306b0c95 NG |
1413 | } |
1414 | ||
92ff1528 | 1415 | while (num_devices != 0) { |
6566d1a3 | 1416 | mutex_lock(&zram_index_mutex); |
92ff1528 | 1417 | ret = zram_add(); |
6566d1a3 | 1418 | mutex_unlock(&zram_index_mutex); |
92ff1528 | 1419 | if (ret < 0) |
a096cafc | 1420 | goto out_error; |
92ff1528 | 1421 | num_devices--; |
de1a21a0 NG |
1422 | } |
1423 | ||
306b0c95 | 1424 | return 0; |
de1a21a0 | 1425 | |
a096cafc | 1426 | out_error: |
85508ec6 | 1427 | destroy_devices(); |
306b0c95 NG |
1428 | return ret; |
1429 | } | |
1430 | ||
f1e3cfff | 1431 | static void __exit zram_exit(void) |
306b0c95 | 1432 | { |
85508ec6 | 1433 | destroy_devices(); |
306b0c95 NG |
1434 | } |
1435 | ||
f1e3cfff NG |
1436 | module_init(zram_init); |
1437 | module_exit(zram_exit); | |
306b0c95 | 1438 | |
9b3bb7ab | 1439 | module_param(num_devices, uint, 0); |
c3cdb40e | 1440 | MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices"); |
9b3bb7ab | 1441 | |
306b0c95 NG |
1442 | MODULE_LICENSE("Dual BSD/GPL"); |
1443 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 1444 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |