Commit | Line | Data |
---|---|---|
f21e6e18 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 | 2 | /* |
f21e6e18 | 3 | * Copyright 1993 by Theodore Ts'o. |
1da177e4 | 4 | */ |
1da177e4 LT |
5 | #include <linux/module.h> |
6 | #include <linux/moduleparam.h> | |
7 | #include <linux/sched.h> | |
8 | #include <linux/fs.h> | |
4ee60ec1 | 9 | #include <linux/pagemap.h> |
1da177e4 LT |
10 | #include <linux/file.h> |
11 | #include <linux/stat.h> | |
12 | #include <linux/errno.h> | |
13 | #include <linux/major.h> | |
14 | #include <linux/wait.h> | |
1da177e4 LT |
15 | #include <linux/blkpg.h> |
16 | #include <linux/init.h> | |
1da177e4 LT |
17 | #include <linux/swap.h> |
18 | #include <linux/slab.h> | |
863d5b82 | 19 | #include <linux/compat.h> |
1da177e4 | 20 | #include <linux/suspend.h> |
83144186 | 21 | #include <linux/freezer.h> |
2a48fc0a | 22 | #include <linux/mutex.h> |
1da177e4 | 23 | #include <linux/writeback.h> |
1da177e4 LT |
24 | #include <linux/completion.h> |
25 | #include <linux/highmem.h> | |
d6b29d7c | 26 | #include <linux/splice.h> |
ee862730 | 27 | #include <linux/sysfs.h> |
770fe30a | 28 | #include <linux/miscdevice.h> |
dfaa2ef6 | 29 | #include <linux/falloc.h> |
283e7e5d | 30 | #include <linux/uio.h> |
d9a08a9e | 31 | #include <linux/ioprio.h> |
db6638d7 | 32 | #include <linux/blk-cgroup.h> |
c74d40e8 | 33 | #include <linux/sched/mm.h> |
06582bc8 | 34 | #include <linux/statfs.h> |
754d9679 CH |
35 | #include <linux/uaccess.h> |
36 | #include <linux/blk-mq.h> | |
37 | #include <linux/spinlock.h> | |
38 | #include <uapi/linux/loop.h> | |
d9a08a9e | 39 | |
754d9679 CH |
40 | /* Possible states of device */ |
41 | enum { | |
42 | Lo_unbound, | |
43 | Lo_bound, | |
44 | Lo_rundown, | |
45 | Lo_deleting, | |
46 | }; | |
d9a08a9e | 47 | |
754d9679 CH |
48 | struct loop_func_table; |
49 | ||
50 | struct loop_device { | |
51 | int lo_number; | |
52 | loff_t lo_offset; | |
53 | loff_t lo_sizelimit; | |
54 | int lo_flags; | |
55 | char lo_file_name[LO_NAME_SIZE]; | |
56 | ||
57 | struct file * lo_backing_file; | |
58 | struct block_device *lo_device; | |
59 | ||
60 | gfp_t old_gfp_mask; | |
61 | ||
62 | spinlock_t lo_lock; | |
63 | int lo_state; | |
64 | spinlock_t lo_work_lock; | |
65 | struct workqueue_struct *workqueue; | |
66 | struct work_struct rootcg_work; | |
67 | struct list_head rootcg_cmd_list; | |
68 | struct list_head idle_worker_list; | |
69 | struct rb_root worker_tree; | |
70 | struct timer_list timer; | |
71 | bool use_dio; | |
72 | bool sysfs_inited; | |
73 | ||
74 | struct request_queue *lo_queue; | |
75 | struct blk_mq_tag_set tag_set; | |
76 | struct gendisk *lo_disk; | |
77 | struct mutex lo_mutex; | |
78 | bool idr_visible; | |
79 | }; | |
1da177e4 | 80 | |
754d9679 CH |
81 | struct loop_cmd { |
82 | struct list_head list_entry; | |
83 | bool use_aio; /* use AIO interface to handle I/O */ | |
84 | atomic_t ref; /* only for aio */ | |
85 | long ret; | |
86 | struct kiocb iocb; | |
87 | struct bio_vec *bvec; | |
88 | struct cgroup_subsys_state *blkcg_css; | |
89 | struct cgroup_subsys_state *memcg_css; | |
90 | }; | |
1da177e4 | 91 | |
87579e9b | 92 | #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) |
e152a05f | 93 | #define LOOP_DEFAULT_HW_Q_DEPTH 128 |
87579e9b | 94 | |
34dd82af | 95 | static DEFINE_IDR(loop_index_idr); |
310ca162 | 96 | static DEFINE_MUTEX(loop_ctl_mutex); |
3ce6e1f6 TH |
97 | static DEFINE_MUTEX(loop_validate_mutex); |
98 | ||
99 | /** | |
100 | * loop_global_lock_killable() - take locks for safe loop_validate_file() test | |
101 | * | |
102 | * @lo: struct loop_device | |
103 | * @global: true if @lo is about to bind another "struct loop_device", false otherwise | |
104 | * | |
105 | * Returns 0 on success, -EINTR otherwise. | |
106 | * | |
107 | * Since loop_validate_file() traverses on other "struct loop_device" if | |
108 | * is_loop_device() is true, we need a global lock for serializing concurrent | |
109 | * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. | |
110 | */ | |
111 | static int loop_global_lock_killable(struct loop_device *lo, bool global) | |
112 | { | |
113 | int err; | |
114 | ||
115 | if (global) { | |
116 | err = mutex_lock_killable(&loop_validate_mutex); | |
117 | if (err) | |
118 | return err; | |
119 | } | |
120 | err = mutex_lock_killable(&lo->lo_mutex); | |
121 | if (err && global) | |
122 | mutex_unlock(&loop_validate_mutex); | |
123 | return err; | |
124 | } | |
125 | ||
126 | /** | |
127 | * loop_global_unlock() - release locks taken by loop_global_lock_killable() | |
128 | * | |
129 | * @lo: struct loop_device | |
130 | * @global: true if @lo was about to bind another "struct loop_device", false otherwise | |
131 | */ | |
132 | static void loop_global_unlock(struct loop_device *lo, bool global) | |
133 | { | |
134 | mutex_unlock(&lo->lo_mutex); | |
135 | if (global) | |
136 | mutex_unlock(&loop_validate_mutex); | |
137 | } | |
1da177e4 | 138 | |
476a4813 LV |
139 | static int max_part; |
140 | static int part_shift; | |
141 | ||
7035b5df | 142 | static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) |
1da177e4 | 143 | { |
b7a1da69 | 144 | loff_t loopsize; |
1da177e4 LT |
145 | |
146 | /* Compute loopsize in bytes */ | |
b7a1da69 GC |
147 | loopsize = i_size_read(file->f_mapping->host); |
148 | if (offset > 0) | |
149 | loopsize -= offset; | |
150 | /* offset is beyond i_size, weird but possible */ | |
7035b5df DM |
151 | if (loopsize < 0) |
152 | return 0; | |
1da177e4 | 153 | |
7035b5df DM |
154 | if (sizelimit > 0 && sizelimit < loopsize) |
155 | loopsize = sizelimit; | |
1da177e4 LT |
156 | /* |
157 | * Unfortunately, if we want to do I/O on the device, | |
158 | * the number of 512-byte sectors has to fit into a sector_t. | |
159 | */ | |
160 | return loopsize >> 9; | |
161 | } | |
162 | ||
7035b5df DM |
163 | static loff_t get_loop_size(struct loop_device *lo, struct file *file) |
164 | { | |
165 | return get_size(lo->lo_offset, lo->lo_sizelimit, file); | |
166 | } | |
167 | ||
2e5ab5f3 ML |
168 | static void __loop_update_dio(struct loop_device *lo, bool dio) |
169 | { | |
170 | struct file *file = lo->lo_backing_file; | |
171 | struct address_space *mapping = file->f_mapping; | |
172 | struct inode *inode = mapping->host; | |
173 | unsigned short sb_bsize = 0; | |
174 | unsigned dio_align = 0; | |
175 | bool use_dio; | |
176 | ||
177 | if (inode->i_sb->s_bdev) { | |
178 | sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); | |
179 | dio_align = sb_bsize - 1; | |
180 | } | |
181 | ||
182 | /* | |
183 | * We support direct I/O only if lo_offset is aligned with the | |
184 | * logical I/O size of backing device, and the logical block | |
47e96246 | 185 | * size of loop is bigger than the backing device's. |
2e5ab5f3 ML |
186 | * |
187 | * TODO: the above condition may be loosed in the future, and | |
188 | * direct I/O may be switched runtime at that time because most | |
89d790ab | 189 | * of requests in sane applications should be PAGE_SIZE aligned |
2e5ab5f3 ML |
190 | */ |
191 | if (dio) { | |
192 | if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && | |
a2ad63da N |
193 | !(lo->lo_offset & dio_align) && |
194 | (file->f_mode & FMODE_CAN_ODIRECT)) | |
2e5ab5f3 ML |
195 | use_dio = true; |
196 | else | |
197 | use_dio = false; | |
198 | } else { | |
199 | use_dio = false; | |
200 | } | |
201 | ||
202 | if (lo->use_dio == use_dio) | |
203 | return; | |
204 | ||
205 | /* flush dirty pages before changing direct IO */ | |
206 | vfs_fsync(file, 0); | |
207 | ||
208 | /* | |
209 | * The flag of LO_FLAGS_DIRECT_IO is handled similarly with | |
210 | * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup | |
211 | * will get updated by ioctl(LOOP_GET_STATUS) | |
212 | */ | |
0fbcf579 MC |
213 | if (lo->lo_state == Lo_bound) |
214 | blk_mq_freeze_queue(lo->lo_queue); | |
2e5ab5f3 | 215 | lo->use_dio = use_dio; |
40326d8a | 216 | if (use_dio) { |
8b904b5b | 217 | blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 218 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
40326d8a | 219 | } else { |
8b904b5b | 220 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
2e5ab5f3 | 221 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; |
40326d8a | 222 | } |
0fbcf579 MC |
223 | if (lo->lo_state == Lo_bound) |
224 | blk_mq_unfreeze_queue(lo->lo_queue); | |
2e5ab5f3 ML |
225 | } |
226 | ||
5795b6f5 MC |
227 | /** |
228 | * loop_set_size() - sets device size and notifies userspace | |
229 | * @lo: struct loop_device to set the size for | |
230 | * @size: new size of the loop device | |
231 | * | |
232 | * Callers must validate that the size passed into this function fits into | |
233 | * a sector_t, eg using loop_validate_size() | |
234 | */ | |
235 | static void loop_set_size(struct loop_device *lo, loff_t size) | |
236 | { | |
449f4ec9 | 237 | if (!set_capacity_and_notify(lo->lo_disk, size)) |
3b4f85d0 | 238 | kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); |
5795b6f5 MC |
239 | } |
240 | ||
aa4d8616 | 241 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 242 | { |
aa4d8616 | 243 | struct iov_iter i; |
1da177e4 | 244 | ssize_t bw; |
283e7e5d | 245 | |
de4eda9d | 246 | iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len); |
1da177e4 | 247 | |
03d95eb2 | 248 | file_start_write(file); |
abbb6589 | 249 | bw = vfs_iter_write(file, &i, ppos, 0); |
03d95eb2 | 250 | file_end_write(file); |
aa4d8616 CH |
251 | |
252 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 253 | return 0; |
aa4d8616 CH |
254 | |
255 | printk_ratelimited(KERN_ERR | |
256 | "loop: Write error at byte offset %llu, length %i.\n", | |
257 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
258 | if (bw >= 0) |
259 | bw = -EIO; | |
260 | return bw; | |
261 | } | |
262 | ||
aa4d8616 CH |
263 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
264 | loff_t pos) | |
1da177e4 | 265 | { |
aa4d8616 CH |
266 | struct bio_vec bvec; |
267 | struct req_iterator iter; | |
268 | int ret = 0; | |
269 | ||
270 | rq_for_each_segment(bvec, rq, iter) { | |
271 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
272 | if (ret < 0) | |
273 | break; | |
274 | cond_resched(); | |
275 | } | |
276 | ||
277 | return ret; | |
1da177e4 LT |
278 | } |
279 | ||
aa4d8616 CH |
280 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
281 | loff_t pos) | |
1da177e4 | 282 | { |
aa4d8616 CH |
283 | struct bio_vec bvec; |
284 | struct req_iterator iter; | |
285 | struct iov_iter i; | |
286 | ssize_t len; | |
1da177e4 | 287 | |
aa4d8616 | 288 | rq_for_each_segment(bvec, rq, iter) { |
de4eda9d | 289 | iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len); |
18e9710e | 290 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
291 | if (len < 0) |
292 | return len; | |
1da177e4 | 293 | |
aa4d8616 | 294 | flush_dcache_page(bvec.bv_page); |
fd582140 | 295 | |
aa4d8616 CH |
296 | if (len != bvec.bv_len) { |
297 | struct bio *bio; | |
1da177e4 | 298 | |
aa4d8616 CH |
299 | __rq_for_each_bio(bio, rq) |
300 | zero_fill_bio(bio); | |
301 | break; | |
302 | } | |
303 | cond_resched(); | |
304 | } | |
305 | ||
306 | return 0; | |
fd582140 JA |
307 | } |
308 | ||
efcfec57 DW |
309 | static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, |
310 | int mode) | |
cf655d95 ML |
311 | { |
312 | /* | |
efcfec57 | 313 | * We use fallocate to manipulate the space mappings used by the image |
47e96246 | 314 | * a.k.a. discard/zerorange. |
cf655d95 ML |
315 | */ |
316 | struct file *file = lo->lo_backing_file; | |
cf655d95 ML |
317 | int ret; |
318 | ||
efcfec57 DW |
319 | mode |= FALLOC_FL_KEEP_SIZE; |
320 | ||
70200574 CH |
321 | if (!bdev_max_discard_sectors(lo->lo_device)) |
322 | return -EOPNOTSUPP; | |
cf655d95 ML |
323 | |
324 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
325 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
70200574 | 326 | return -EIO; |
cf655d95 ML |
327 | return ret; |
328 | } | |
329 | ||
330 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
331 | { | |
9c64e38c | 332 | int ret = vfs_fsync(lo->lo_backing_file, 0); |
cf655d95 ML |
333 | if (unlikely(ret && ret != -EINVAL)) |
334 | ret = -EIO; | |
335 | ||
336 | return ret; | |
337 | } | |
338 | ||
fe2cb290 | 339 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 340 | { |
fe2cb290 | 341 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 342 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 343 | |
f9de14bc JA |
344 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
345 | req_op(rq) != REQ_OP_READ) { | |
346 | if (cmd->ret < 0) | |
8cd55087 | 347 | ret = errno_to_blk_status(cmd->ret); |
f9de14bc | 348 | goto end_io; |
bc07c10a | 349 | } |
fe2cb290 | 350 | |
f9de14bc JA |
351 | /* |
352 | * Short READ - if we got some data, advance our request and | |
353 | * retry it. If we got no data, end the rest with EIO. | |
354 | */ | |
355 | if (cmd->ret) { | |
356 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
357 | cmd->ret = 0; | |
358 | blk_mq_requeue_request(rq, true); | |
359 | } else { | |
360 | if (cmd->use_aio) { | |
361 | struct bio *bio = rq->bio; | |
362 | ||
363 | while (bio) { | |
364 | zero_fill_bio(bio); | |
365 | bio = bio->bi_next; | |
366 | } | |
367 | } | |
368 | ret = BLK_STS_IOERR; | |
369 | end_io: | |
370 | blk_mq_end_request(rq, ret); | |
371 | } | |
bc07c10a ML |
372 | } |
373 | ||
92d77332 SL |
374 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
375 | { | |
1894e916 JA |
376 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
377 | ||
92d77332 SL |
378 | if (!atomic_dec_and_test(&cmd->ref)) |
379 | return; | |
380 | kfree(cmd->bvec); | |
381 | cmd->bvec = NULL; | |
15f73f5b CH |
382 | if (likely(!blk_should_fake_timeout(rq->q))) |
383 | blk_mq_complete_request(rq); | |
92d77332 SL |
384 | } |
385 | ||
6b19b766 | 386 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret) |
bc07c10a ML |
387 | { |
388 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 389 | |
fe2cb290 | 390 | cmd->ret = ret; |
92d77332 | 391 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
392 | } |
393 | ||
394 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
de4eda9d | 395 | loff_t pos, int rw) |
bc07c10a ML |
396 | { |
397 | struct iov_iter iter; | |
86af5952 | 398 | struct req_iterator rq_iter; |
bc07c10a | 399 | struct bio_vec *bvec; |
1894e916 | 400 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 401 | struct bio *bio = rq->bio; |
bc07c10a | 402 | struct file *file = lo->lo_backing_file; |
86af5952 | 403 | struct bio_vec tmp; |
40326d8a | 404 | unsigned int offset; |
86af5952 | 405 | int nr_bvec = 0; |
bc07c10a ML |
406 | int ret; |
407 | ||
86af5952 ML |
408 | rq_for_each_bvec(tmp, rq, rq_iter) |
409 | nr_bvec++; | |
410 | ||
40326d8a | 411 | if (rq->bio != rq->biotail) { |
40326d8a | 412 | |
86af5952 | 413 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 414 | GFP_NOIO); |
40326d8a SL |
415 | if (!bvec) |
416 | return -EIO; | |
417 | cmd->bvec = bvec; | |
418 | ||
419 | /* | |
420 | * The bios of the request may be started from the middle of | |
421 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 422 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
423 | * API will take care of all details for us. |
424 | */ | |
86af5952 | 425 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
426 | *bvec = tmp; |
427 | bvec++; | |
428 | } | |
429 | bvec = cmd->bvec; | |
430 | offset = 0; | |
431 | } else { | |
432 | /* | |
433 | * Same here, this bio may be started from the middle of the | |
434 | * 'bvec' because of bio splitting, so offset from the bvec | |
435 | * must be passed to iov iterator | |
436 | */ | |
437 | offset = bio->bi_iter.bi_bvec_done; | |
438 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 439 | } |
92d77332 | 440 | atomic_set(&cmd->ref, 2); |
bc07c10a | 441 | |
b6207430 | 442 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 443 | iter.iov_offset = offset; |
bc07c10a ML |
444 | |
445 | cmd->iocb.ki_pos = pos; | |
446 | cmd->iocb.ki_filp = file; | |
447 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
448 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 449 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
bc07c10a | 450 | |
de4eda9d | 451 | if (rw == ITER_SOURCE) |
bb7462b6 | 452 | ret = call_write_iter(file, &cmd->iocb, &iter); |
bc07c10a | 453 | else |
bb7462b6 | 454 | ret = call_read_iter(file, &cmd->iocb, &iter); |
bc07c10a | 455 | |
92d77332 SL |
456 | lo_rw_aio_do_completion(cmd); |
457 | ||
bc07c10a | 458 | if (ret != -EIOCBQUEUED) |
6b19b766 | 459 | lo_rw_aio_complete(&cmd->iocb, ret); |
bc07c10a ML |
460 | return 0; |
461 | } | |
462 | ||
c1c87c2b | 463 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
464 | { |
465 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 466 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
467 | |
468 | /* | |
469 | * lo_write_simple and lo_read_simple should have been covered | |
470 | * by io submit style function like lo_rw_aio(), one blocker | |
471 | * is that lo_read_simple() need to call flush_dcache_page after | |
472 | * the page is written from kernel, and it isn't easy to handle | |
473 | * this in io submit style function which submits all segments | |
474 | * of the req at one time. And direct read IO doesn't need to | |
475 | * run flush_dcache_page(). | |
476 | */ | |
c1c87c2b CH |
477 | switch (req_op(rq)) { |
478 | case REQ_OP_FLUSH: | |
479 | return lo_req_flush(lo, rq); | |
19372e27 | 480 | case REQ_OP_WRITE_ZEROES: |
efcfec57 DW |
481 | /* |
482 | * If the caller doesn't want deallocation, call zeroout to | |
483 | * write zeroes the range. Otherwise, punch them out. | |
484 | */ | |
485 | return lo_fallocate(lo, rq, pos, | |
486 | (rq->cmd_flags & REQ_NOUNMAP) ? | |
487 | FALLOC_FL_ZERO_RANGE : | |
488 | FALLOC_FL_PUNCH_HOLE); | |
489 | case REQ_OP_DISCARD: | |
490 | return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); | |
c1c87c2b | 491 | case REQ_OP_WRITE: |
47e96246 | 492 | if (cmd->use_aio) |
de4eda9d | 493 | return lo_rw_aio(lo, cmd, pos, ITER_SOURCE); |
af65aa8e | 494 | else |
c1c87c2b CH |
495 | return lo_write_simple(lo, rq, pos); |
496 | case REQ_OP_READ: | |
47e96246 | 497 | if (cmd->use_aio) |
de4eda9d | 498 | return lo_rw_aio(lo, cmd, pos, ITER_DEST); |
aa4d8616 | 499 | else |
c1c87c2b CH |
500 | return lo_read_simple(lo, rq, pos); |
501 | default: | |
502 | WARN_ON_ONCE(1); | |
503 | return -EIO; | |
aa4d8616 | 504 | } |
1da177e4 LT |
505 | } |
506 | ||
2e5ab5f3 ML |
507 | static inline void loop_update_dio(struct loop_device *lo) |
508 | { | |
efbe3c24 IW |
509 | __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | |
510 | lo->use_dio); | |
2e5ab5f3 ML |
511 | } |
512 | ||
0384264e | 513 | static void loop_reread_partitions(struct loop_device *lo) |
06f0e9e6 ML |
514 | { |
515 | int rc; | |
516 | ||
0384264e CH |
517 | mutex_lock(&lo->lo_disk->open_mutex); |
518 | rc = bdev_disk_changed(lo->lo_disk, false); | |
519 | mutex_unlock(&lo->lo_disk->open_mutex); | |
06f0e9e6 ML |
520 | if (rc) |
521 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
522 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
523 | } | |
524 | ||
d2ac838e TT |
525 | static inline int is_loop_device(struct file *file) |
526 | { | |
527 | struct inode *i = file->f_mapping->host; | |
528 | ||
6f24784f | 529 | return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; |
d2ac838e TT |
530 | } |
531 | ||
532 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
533 | { | |
534 | struct inode *inode = file->f_mapping->host; | |
535 | struct file *f = file; | |
536 | ||
537 | /* Avoid recursion */ | |
538 | while (is_loop_device(f)) { | |
539 | struct loop_device *l; | |
540 | ||
3ce6e1f6 | 541 | lockdep_assert_held(&loop_validate_mutex); |
4e7b5671 | 542 | if (f->f_mapping->host->i_rdev == bdev->bd_dev) |
d2ac838e TT |
543 | return -EBADF; |
544 | ||
4e7b5671 | 545 | l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; |
3ce6e1f6 | 546 | if (l->lo_state != Lo_bound) |
d2ac838e | 547 | return -EINVAL; |
3ce6e1f6 TH |
548 | /* Order wrt setting lo->lo_backing_file in loop_configure(). */ |
549 | rmb(); | |
d2ac838e TT |
550 | f = l->lo_backing_file; |
551 | } | |
552 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
553 | return -EINVAL; | |
554 | return 0; | |
555 | } | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * loop_change_fd switched the backing store of a loopback device to | |
559 | * a new file. This is useful for operating system installers to free up | |
560 | * the original file and in High Availability environments to switch to | |
561 | * an alternative location for the content in case of server meltdown. | |
562 | * This can only work if the loop device is used read-only, and if the | |
563 | * new backing store is the same size and type as the old backing store. | |
564 | */ | |
bb214884 AV |
565 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
566 | unsigned int arg) | |
1da177e4 | 567 | { |
3ce6e1f6 TH |
568 | struct file *file = fget(arg); |
569 | struct file *old_file; | |
570 | int error; | |
571 | bool partscan; | |
572 | bool is_loop; | |
1da177e4 | 573 | |
3ce6e1f6 TH |
574 | if (!file) |
575 | return -EBADF; | |
498ef5c7 CH |
576 | |
577 | /* suppress uevents while reconfiguring the device */ | |
578 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); | |
579 | ||
3ce6e1f6 TH |
580 | is_loop = is_loop_device(file); |
581 | error = loop_global_lock_killable(lo, is_loop); | |
c3710770 | 582 | if (error) |
3ce6e1f6 | 583 | goto out_putf; |
1da177e4 LT |
584 | error = -ENXIO; |
585 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 586 | goto out_err; |
1da177e4 LT |
587 | |
588 | /* the loop device has to be read-only */ | |
589 | error = -EINVAL; | |
590 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 591 | goto out_err; |
1da177e4 | 592 | |
d2ac838e TT |
593 | error = loop_validate_file(file, bdev); |
594 | if (error) | |
1dded9ac | 595 | goto out_err; |
d2ac838e | 596 | |
1da177e4 LT |
597 | old_file = lo->lo_backing_file; |
598 | ||
599 | error = -EINVAL; | |
600 | ||
1da177e4 LT |
601 | /* size of the new backing store needs to be the same */ |
602 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 603 | goto out_err; |
1da177e4 LT |
604 | |
605 | /* and ... switch */ | |
9f65c489 | 606 | disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); |
43cade80 OS |
607 | blk_mq_freeze_queue(lo->lo_queue); |
608 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
609 | lo->lo_backing_file = file; | |
610 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
611 | mapping_set_gfp_mask(file->f_mapping, | |
612 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
613 | loop_update_dio(lo); | |
614 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 615 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
3ce6e1f6 TH |
616 | loop_global_unlock(lo, is_loop); |
617 | ||
618 | /* | |
619 | * Flush loop_validate_file() before fput(), for l->lo_backing_file | |
620 | * might be pointing at old_file which might be the last reference. | |
621 | */ | |
622 | if (!is_loop) { | |
623 | mutex_lock(&loop_validate_mutex); | |
624 | mutex_unlock(&loop_validate_mutex); | |
625 | } | |
1dded9ac | 626 | /* |
6cc8e743 | 627 | * We must drop file reference outside of lo_mutex as dropping |
a8698707 | 628 | * the file ref can take open_mutex which creates circular locking |
1dded9ac JK |
629 | * dependency. |
630 | */ | |
631 | fput(old_file); | |
85b0a54a | 632 | if (partscan) |
0384264e | 633 | loop_reread_partitions(lo); |
498ef5c7 CH |
634 | |
635 | error = 0; | |
636 | done: | |
637 | /* enable and uncork uevent now that we are done */ | |
638 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); | |
639 | return error; | |
1da177e4 | 640 | |
1dded9ac | 641 | out_err: |
3ce6e1f6 TH |
642 | loop_global_unlock(lo, is_loop); |
643 | out_putf: | |
644 | fput(file); | |
498ef5c7 | 645 | goto done; |
1da177e4 LT |
646 | } |
647 | ||
ee862730 MB |
648 | /* loop sysfs attributes */ |
649 | ||
650 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
651 | ssize_t (*callback)(struct loop_device *, char *)) | |
652 | { | |
34dd82af KS |
653 | struct gendisk *disk = dev_to_disk(dev); |
654 | struct loop_device *lo = disk->private_data; | |
ee862730 | 655 | |
34dd82af | 656 | return callback(lo, page); |
ee862730 MB |
657 | } |
658 | ||
659 | #define LOOP_ATTR_RO(_name) \ | |
660 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
661 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
662 | struct device_attribute *attr, char *b) \ | |
663 | { \ | |
664 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
665 | } \ | |
666 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 667 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
668 | |
669 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
670 | { | |
671 | ssize_t ret; | |
672 | char *p = NULL; | |
673 | ||
05eb0f25 | 674 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 675 | if (lo->lo_backing_file) |
9bf39ab2 | 676 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 677 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
678 | |
679 | if (IS_ERR_OR_NULL(p)) | |
680 | ret = PTR_ERR(p); | |
681 | else { | |
682 | ret = strlen(p); | |
683 | memmove(buf, p, ret); | |
684 | buf[ret++] = '\n'; | |
685 | buf[ret] = 0; | |
686 | } | |
687 | ||
688 | return ret; | |
689 | } | |
690 | ||
691 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
692 | { | |
b27824d3 | 693 | return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset); |
ee862730 MB |
694 | } |
695 | ||
696 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
697 | { | |
b27824d3 | 698 | return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); |
ee862730 MB |
699 | } |
700 | ||
701 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
702 | { | |
703 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
704 | ||
b27824d3 | 705 | return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0"); |
ee862730 MB |
706 | } |
707 | ||
e03c8dd1 KS |
708 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
709 | { | |
710 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
711 | ||
b27824d3 | 712 | return sysfs_emit(buf, "%s\n", partscan ? "1" : "0"); |
e03c8dd1 KS |
713 | } |
714 | ||
2e5ab5f3 ML |
715 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
716 | { | |
717 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
718 | ||
b27824d3 | 719 | return sysfs_emit(buf, "%s\n", dio ? "1" : "0"); |
2e5ab5f3 ML |
720 | } |
721 | ||
ee862730 MB |
722 | LOOP_ATTR_RO(backing_file); |
723 | LOOP_ATTR_RO(offset); | |
724 | LOOP_ATTR_RO(sizelimit); | |
725 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 726 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 727 | LOOP_ATTR_RO(dio); |
ee862730 MB |
728 | |
729 | static struct attribute *loop_attrs[] = { | |
730 | &loop_attr_backing_file.attr, | |
731 | &loop_attr_offset.attr, | |
732 | &loop_attr_sizelimit.attr, | |
733 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 734 | &loop_attr_partscan.attr, |
2e5ab5f3 | 735 | &loop_attr_dio.attr, |
ee862730 MB |
736 | NULL, |
737 | }; | |
738 | ||
739 | static struct attribute_group loop_attribute_group = { | |
740 | .name = "loop", | |
741 | .attrs= loop_attrs, | |
742 | }; | |
743 | ||
d3349b6b | 744 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 745 | { |
d3349b6b TH |
746 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
747 | &loop_attribute_group); | |
ee862730 MB |
748 | } |
749 | ||
750 | static void loop_sysfs_exit(struct loop_device *lo) | |
751 | { | |
d3349b6b TH |
752 | if (lo->sysfs_inited) |
753 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
754 | &loop_attribute_group); | |
ee862730 MB |
755 | } |
756 | ||
dfaa2ef6 LC |
757 | static void loop_config_discard(struct loop_device *lo) |
758 | { | |
759 | struct file *file = lo->lo_backing_file; | |
760 | struct inode *inode = file->f_mapping->host; | |
761 | struct request_queue *q = lo->lo_queue; | |
bcb21c8c | 762 | u32 granularity, max_discard_sectors; |
dfaa2ef6 | 763 | |
c52abf56 EG |
764 | /* |
765 | * If the backing device is a block device, mirror its zeroing | |
766 | * capability. Set the discard sectors to the block device's zeroing | |
767 | * capabilities because loop discards result in blkdev_issue_zeroout(), | |
768 | * not blkdev_issue_discard(). This maintains consistent behavior with | |
769 | * file-backed loop devices: discarded regions read back as zero. | |
770 | */ | |
47e96246 | 771 | if (S_ISBLK(inode->i_mode)) { |
4e7b5671 | 772 | struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); |
c52abf56 | 773 | |
bcb21c8c | 774 | max_discard_sectors = backingq->limits.max_write_zeroes_sectors; |
7b47ef52 | 775 | granularity = bdev_discard_granularity(I_BDEV(inode)) ?: |
bcb21c8c | 776 | queue_physical_block_size(backingq); |
c52abf56 | 777 | |
dfaa2ef6 LC |
778 | /* |
779 | * We use punch hole to reclaim the free space used by the | |
47e96246 | 780 | * image a.k.a. discard. |
dfaa2ef6 | 781 | */ |
47e96246 | 782 | } else if (!file->f_op->fallocate) { |
bcb21c8c ML |
783 | max_discard_sectors = 0; |
784 | granularity = 0; | |
dfaa2ef6 | 785 | |
c52abf56 | 786 | } else { |
06582bc8 ML |
787 | struct kstatfs sbuf; |
788 | ||
bcb21c8c | 789 | max_discard_sectors = UINT_MAX >> 9; |
06582bc8 ML |
790 | if (!vfs_statfs(&file->f_path, &sbuf)) |
791 | granularity = sbuf.f_bsize; | |
792 | else | |
793 | max_discard_sectors = 0; | |
c52abf56 EG |
794 | } |
795 | ||
bcb21c8c ML |
796 | if (max_discard_sectors) { |
797 | q->limits.discard_granularity = granularity; | |
798 | blk_queue_max_discard_sectors(q, max_discard_sectors); | |
799 | blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); | |
bcb21c8c ML |
800 | } else { |
801 | q->limits.discard_granularity = 0; | |
802 | blk_queue_max_discard_sectors(q, 0); | |
803 | blk_queue_max_write_zeroes_sectors(q, 0); | |
bcb21c8c | 804 | } |
dfaa2ef6 LC |
805 | } |
806 | ||
87579e9b DS |
807 | struct loop_worker { |
808 | struct rb_node rb_node; | |
809 | struct work_struct work; | |
810 | struct list_head cmd_list; | |
811 | struct list_head idle_list; | |
812 | struct loop_device *lo; | |
c74d40e8 | 813 | struct cgroup_subsys_state *blkcg_css; |
87579e9b DS |
814 | unsigned long last_ran_at; |
815 | }; | |
816 | ||
817 | static void loop_workfn(struct work_struct *work); | |
87579e9b DS |
818 | |
819 | #ifdef CONFIG_BLK_CGROUP | |
820 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
e03a3d7a | 821 | { |
87579e9b | 822 | return !css || css == blkcg_root_css; |
e03a3d7a | 823 | } |
87579e9b DS |
824 | #else |
825 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
b2ee7d46 | 826 | { |
87579e9b | 827 | return !css; |
b2ee7d46 | 828 | } |
87579e9b | 829 | #endif |
b2ee7d46 | 830 | |
87579e9b | 831 | static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) |
e03a3d7a | 832 | { |
413ec805 | 833 | struct rb_node **node, *parent = NULL; |
87579e9b DS |
834 | struct loop_worker *cur_worker, *worker = NULL; |
835 | struct work_struct *work; | |
836 | struct list_head *cmd_list; | |
837 | ||
838 | spin_lock_irq(&lo->lo_work_lock); | |
839 | ||
c74d40e8 | 840 | if (queue_on_root_worker(cmd->blkcg_css)) |
87579e9b DS |
841 | goto queue_work; |
842 | ||
843 | node = &lo->worker_tree.rb_node; | |
844 | ||
845 | while (*node) { | |
846 | parent = *node; | |
847 | cur_worker = container_of(*node, struct loop_worker, rb_node); | |
c74d40e8 | 848 | if (cur_worker->blkcg_css == cmd->blkcg_css) { |
87579e9b DS |
849 | worker = cur_worker; |
850 | break; | |
c74d40e8 | 851 | } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { |
87579e9b DS |
852 | node = &(*node)->rb_left; |
853 | } else { | |
854 | node = &(*node)->rb_right; | |
855 | } | |
856 | } | |
857 | if (worker) | |
858 | goto queue_work; | |
859 | ||
860 | worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); | |
861 | /* | |
862 | * In the event we cannot allocate a worker, just queue on the | |
c74d40e8 | 863 | * rootcg worker and issue the I/O as the rootcg |
87579e9b | 864 | */ |
c74d40e8 DS |
865 | if (!worker) { |
866 | cmd->blkcg_css = NULL; | |
867 | if (cmd->memcg_css) | |
868 | css_put(cmd->memcg_css); | |
869 | cmd->memcg_css = NULL; | |
87579e9b | 870 | goto queue_work; |
c74d40e8 | 871 | } |
87579e9b | 872 | |
c74d40e8 DS |
873 | worker->blkcg_css = cmd->blkcg_css; |
874 | css_get(worker->blkcg_css); | |
87579e9b DS |
875 | INIT_WORK(&worker->work, loop_workfn); |
876 | INIT_LIST_HEAD(&worker->cmd_list); | |
877 | INIT_LIST_HEAD(&worker->idle_list); | |
878 | worker->lo = lo; | |
879 | rb_link_node(&worker->rb_node, parent, node); | |
880 | rb_insert_color(&worker->rb_node, &lo->worker_tree); | |
881 | queue_work: | |
882 | if (worker) { | |
883 | /* | |
884 | * We need to remove from the idle list here while | |
885 | * holding the lock so that the idle timer doesn't | |
886 | * free the worker | |
887 | */ | |
888 | if (!list_empty(&worker->idle_list)) | |
889 | list_del_init(&worker->idle_list); | |
890 | work = &worker->work; | |
891 | cmd_list = &worker->cmd_list; | |
892 | } else { | |
893 | work = &lo->rootcg_work; | |
894 | cmd_list = &lo->rootcg_cmd_list; | |
895 | } | |
896 | list_add_tail(&cmd->list_entry, cmd_list); | |
897 | queue_work(lo->workqueue, work); | |
898 | spin_unlock_irq(&lo->lo_work_lock); | |
e03a3d7a ML |
899 | } |
900 | ||
2cf429b5 CH |
901 | static void loop_set_timer(struct loop_device *lo) |
902 | { | |
903 | timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); | |
904 | } | |
905 | ||
906 | static void loop_free_idle_workers(struct loop_device *lo, bool delete_all) | |
907 | { | |
908 | struct loop_worker *pos, *worker; | |
909 | ||
910 | spin_lock_irq(&lo->lo_work_lock); | |
911 | list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, | |
912 | idle_list) { | |
913 | if (!delete_all && | |
914 | time_is_after_jiffies(worker->last_ran_at + | |
915 | LOOP_IDLE_WORKER_TIMEOUT)) | |
916 | break; | |
917 | list_del(&worker->idle_list); | |
918 | rb_erase(&worker->rb_node, &lo->worker_tree); | |
919 | css_put(worker->blkcg_css); | |
920 | kfree(worker); | |
921 | } | |
922 | if (!list_empty(&lo->idle_worker_list)) | |
923 | loop_set_timer(lo); | |
924 | spin_unlock_irq(&lo->lo_work_lock); | |
925 | } | |
926 | ||
927 | static void loop_free_idle_workers_timer(struct timer_list *timer) | |
928 | { | |
929 | struct loop_device *lo = container_of(timer, struct loop_device, timer); | |
930 | ||
931 | return loop_free_idle_workers(lo, false); | |
932 | } | |
933 | ||
56a85fd8 HH |
934 | static void loop_update_rotational(struct loop_device *lo) |
935 | { | |
936 | struct file *file = lo->lo_backing_file; | |
937 | struct inode *file_inode = file->f_mapping->host; | |
938 | struct block_device *file_bdev = file_inode->i_sb->s_bdev; | |
939 | struct request_queue *q = lo->lo_queue; | |
940 | bool nonrot = true; | |
941 | ||
942 | /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ | |
943 | if (file_bdev) | |
10f0d2a5 | 944 | nonrot = bdev_nonrot(file_bdev); |
56a85fd8 HH |
945 | |
946 | if (nonrot) | |
947 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q); | |
948 | else | |
949 | blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); | |
950 | } | |
951 | ||
62ab466c MC |
952 | /** |
953 | * loop_set_status_from_info - configure device from loop_info | |
954 | * @lo: struct loop_device to configure | |
955 | * @info: struct loop_info64 to configure the device with | |
956 | * | |
957 | * Configures the loop device parameters according to the passed | |
958 | * in loop_info64 configuration. | |
959 | */ | |
960 | static int | |
961 | loop_set_status_from_info(struct loop_device *lo, | |
962 | const struct loop_info64 *info) | |
963 | { | |
62ab466c MC |
964 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) |
965 | return -EINVAL; | |
966 | ||
47e96246 CH |
967 | switch (info->lo_encrypt_type) { |
968 | case LO_CRYPT_NONE: | |
969 | break; | |
970 | case LO_CRYPT_XOR: | |
971 | pr_warn("support for the xor transformation has been removed.\n"); | |
972 | return -EINVAL; | |
973 | case LO_CRYPT_CRYPTOAPI: | |
974 | pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); | |
975 | return -EINVAL; | |
976 | default: | |
977 | return -EINVAL; | |
978 | } | |
62ab466c | 979 | |
9f6ad5d5 ZJ |
980 | /* Avoid assigning overflow values */ |
981 | if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX) | |
982 | return -EOVERFLOW; | |
983 | ||
62ab466c MC |
984 | lo->lo_offset = info->lo_offset; |
985 | lo->lo_sizelimit = info->lo_sizelimit; | |
c490a0b5 | 986 | |
62ab466c | 987 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); |
62ab466c | 988 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; |
faf1d254 | 989 | lo->lo_flags = info->lo_flags; |
62ab466c MC |
990 | return 0; |
991 | } | |
992 | ||
05bdb996 | 993 | static int loop_configure(struct loop_device *lo, blk_mode_t mode, |
3448914e MC |
994 | struct block_device *bdev, |
995 | const struct loop_config *config) | |
1da177e4 | 996 | { |
3ce6e1f6 TH |
997 | struct file *file = fget(config->fd); |
998 | struct inode *inode; | |
1da177e4 | 999 | struct address_space *mapping; |
3ce6e1f6 TH |
1000 | int error; |
1001 | loff_t size; | |
1002 | bool partscan; | |
1003 | unsigned short bsize; | |
1004 | bool is_loop; | |
1005 | ||
1006 | if (!file) | |
1007 | return -EBADF; | |
1008 | is_loop = is_loop_device(file); | |
1da177e4 LT |
1009 | |
1010 | /* This is safe, since we have a reference from open(). */ | |
1011 | __module_get(THIS_MODULE); | |
1012 | ||
33ec3e53 JK |
1013 | /* |
1014 | * If we don't hold exclusive handle for the device, upgrade to it | |
1015 | * here to avoid changing device under exclusive owner. | |
1016 | */ | |
05bdb996 | 1017 | if (!(mode & BLK_OPEN_EXCL)) { |
0718afd4 | 1018 | error = bd_prepare_to_claim(bdev, loop_configure, NULL); |
ecbe6bc0 | 1019 | if (error) |
33ec3e53 JK |
1020 | goto out_putf; |
1021 | } | |
1022 | ||
3ce6e1f6 | 1023 | error = loop_global_lock_killable(lo, is_loop); |
757ecf40 | 1024 | if (error) |
33ec3e53 | 1025 | goto out_bdev; |
757ecf40 | 1026 | |
1da177e4 LT |
1027 | error = -EBUSY; |
1028 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 1029 | goto out_unlock; |
1da177e4 | 1030 | |
d2ac838e TT |
1031 | error = loop_validate_file(file, bdev); |
1032 | if (error) | |
757ecf40 | 1033 | goto out_unlock; |
1da177e4 LT |
1034 | |
1035 | mapping = file->f_mapping; | |
1036 | inode = mapping->host; | |
1037 | ||
3448914e MC |
1038 | if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { |
1039 | error = -EINVAL; | |
1040 | goto out_unlock; | |
1041 | } | |
1042 | ||
1043 | if (config->block_size) { | |
af3c570f | 1044 | error = blk_validate_block_size(config->block_size); |
3448914e MC |
1045 | if (error) |
1046 | goto out_unlock; | |
1047 | } | |
1048 | ||
1049 | error = loop_set_status_from_info(lo, &config->info); | |
1050 | if (error) | |
1051 | goto out_unlock; | |
1052 | ||
05bdb996 | 1053 | if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) || |
283e7e5d | 1054 | !file->f_op->write_iter) |
3448914e | 1055 | lo->lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 1056 | |
87579e9b | 1057 | if (!lo->workqueue) { |
d292dc80 CH |
1058 | lo->workqueue = alloc_workqueue("loop%d", |
1059 | WQ_UNBOUND | WQ_FREEZABLE, | |
1060 | 0, lo->lo_number); | |
1061 | if (!lo->workqueue) { | |
1062 | error = -ENOMEM; | |
1063 | goto out_unlock; | |
1064 | } | |
87579e9b | 1065 | } |
1da177e4 | 1066 | |
bb430b69 AR |
1067 | /* suppress uevents while reconfiguring the device */ |
1068 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); | |
1069 | ||
9f65c489 | 1070 | disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); |
7a2f0ce1 | 1071 | set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); |
1da177e4 | 1072 | |
3448914e | 1073 | lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; |
1da177e4 | 1074 | lo->lo_device = bdev; |
1da177e4 | 1075 | lo->lo_backing_file = file; |
1da177e4 LT |
1076 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
1077 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1078 | ||
3448914e | 1079 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) |
21d0727f | 1080 | blk_queue_write_cache(lo->lo_queue, true, false); |
68db1961 | 1081 | |
3448914e MC |
1082 | if (config->block_size) |
1083 | bsize = config->block_size; | |
96ed320d | 1084 | else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) |
85560117 | 1085 | /* In case of direct I/O, match underlying block size */ |
3448914e MC |
1086 | bsize = bdev_logical_block_size(inode->i_sb->s_bdev); |
1087 | else | |
1088 | bsize = 512; | |
85560117 | 1089 | |
3448914e MC |
1090 | blk_queue_logical_block_size(lo->lo_queue, bsize); |
1091 | blk_queue_physical_block_size(lo->lo_queue, bsize); | |
1092 | blk_queue_io_min(lo->lo_queue, bsize); | |
85560117 | 1093 | |
2b9ac22b | 1094 | loop_config_discard(lo); |
56a85fd8 | 1095 | loop_update_rotational(lo); |
2e5ab5f3 | 1096 | loop_update_dio(lo); |
ee862730 | 1097 | loop_sysfs_init(lo); |
79e5dc59 MC |
1098 | |
1099 | size = get_loop_size(lo, file); | |
5795b6f5 | 1100 | loop_set_size(lo, size); |
1da177e4 | 1101 | |
3ce6e1f6 TH |
1102 | /* Order wrt reading lo_state in loop_validate_file(). */ |
1103 | wmb(); | |
1104 | ||
6c997918 | 1105 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1106 | if (part_shift) |
1107 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 1108 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
fe6a8fc5 | 1109 | if (partscan) |
b9684a71 | 1110 | clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
c1681bf8 | 1111 | |
bb430b69 AR |
1112 | /* enable and uncork uevent now that we are done */ |
1113 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); | |
1114 | ||
3ce6e1f6 | 1115 | loop_global_unlock(lo, is_loop); |
85b0a54a | 1116 | if (partscan) |
0384264e | 1117 | loop_reread_partitions(lo); |
bb430b69 | 1118 | |
05bdb996 | 1119 | if (!(mode & BLK_OPEN_EXCL)) |
37c3fc9a | 1120 | bd_abort_claiming(bdev, loop_configure); |
498ef5c7 | 1121 | |
bb430b69 | 1122 | return 0; |
1da177e4 | 1123 | |
757ecf40 | 1124 | out_unlock: |
3ce6e1f6 | 1125 | loop_global_unlock(lo, is_loop); |
33ec3e53 | 1126 | out_bdev: |
05bdb996 | 1127 | if (!(mode & BLK_OPEN_EXCL)) |
37c3fc9a | 1128 | bd_abort_claiming(bdev, loop_configure); |
757ecf40 | 1129 | out_putf: |
1da177e4 | 1130 | fput(file); |
1da177e4 LT |
1131 | /* This is safe: open() is still holding a reference. */ |
1132 | module_put(THIS_MODULE); | |
bb430b69 | 1133 | return error; |
1da177e4 LT |
1134 | } |
1135 | ||
bf23747e | 1136 | static void __loop_clr_fd(struct loop_device *lo, bool release) |
1da177e4 | 1137 | { |
6050fa4c | 1138 | struct file *filp; |
b4e3ca1a | 1139 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 | 1140 | |
4ceddce5 MFO |
1141 | if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) |
1142 | blk_queue_write_cache(lo->lo_queue, false, false); | |
1143 | ||
1fe0b1ac CH |
1144 | /* |
1145 | * Freeze the request queue when unbinding on a live file descriptor and | |
1146 | * thus an open device. When called from ->release we are guaranteed | |
1147 | * that there is no I/O in progress already. | |
1148 | */ | |
1149 | if (!release) | |
1150 | blk_mq_freeze_queue(lo->lo_queue); | |
87579e9b | 1151 | |
1da177e4 | 1152 | spin_lock_irq(&lo->lo_lock); |
6050fa4c | 1153 | filp = lo->lo_backing_file; |
1da177e4 | 1154 | lo->lo_backing_file = NULL; |
05eb0f25 | 1155 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 1156 | |
1da177e4 | 1157 | lo->lo_device = NULL; |
1da177e4 LT |
1158 | lo->lo_offset = 0; |
1159 | lo->lo_sizelimit = 0; | |
1da177e4 | 1160 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); |
89e4fdec | 1161 | blk_queue_logical_block_size(lo->lo_queue, 512); |
bf093753 OS |
1162 | blk_queue_physical_block_size(lo->lo_queue, 512); |
1163 | blk_queue_io_min(lo->lo_queue, 512); | |
e515be8f | 1164 | invalidate_disk(lo->lo_disk); |
51a0bb0c | 1165 | loop_sysfs_exit(lo); |
19f553db XY |
1166 | /* let user-space know about this change */ |
1167 | kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 1168 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
bf23747e TH |
1169 | /* This is safe: open() is still holding a reference. */ |
1170 | module_put(THIS_MODULE); | |
1fe0b1ac CH |
1171 | if (!release) |
1172 | blk_mq_unfreeze_queue(lo->lo_queue); | |
f8933667 | 1173 | |
9f65c489 | 1174 | disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); |
6050fa4c TH |
1175 | |
1176 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) { | |
1177 | int err; | |
1178 | ||
bf23747e TH |
1179 | /* |
1180 | * open_mutex has been held already in release path, so don't | |
1181 | * acquire it if this function is called in such case. | |
1182 | * | |
1183 | * If the reread partition isn't from release path, lo_refcnt | |
1184 | * must be at least one and it can only become zero when the | |
1185 | * current holder is released. | |
1186 | */ | |
1187 | if (!release) | |
1188 | mutex_lock(&lo->lo_disk->open_mutex); | |
0384264e | 1189 | err = bdev_disk_changed(lo->lo_disk, false); |
bf23747e TH |
1190 | if (!release) |
1191 | mutex_unlock(&lo->lo_disk->open_mutex); | |
40853d6f DZ |
1192 | if (err) |
1193 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
6050fa4c | 1194 | __func__, lo->lo_number, err); |
d57f3374 | 1195 | /* Device is gone, no point in returning error */ |
d57f3374 | 1196 | } |
758a58d0 | 1197 | |
bf23747e TH |
1198 | /* |
1199 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1200 | * finished. There cannot be anybody else entering __loop_clr_fd() as | |
1201 | * Lo_rundown state protects us from all the other places trying to | |
1202 | * change the 'lo' device. | |
1203 | */ | |
758a58d0 DZ |
1204 | lo->lo_flags = 0; |
1205 | if (!part_shift) | |
b9684a71 | 1206 | set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
6050fa4c | 1207 | mutex_lock(&lo->lo_mutex); |
758a58d0 | 1208 | lo->lo_state = Lo_unbound; |
6cc8e743 | 1209 | mutex_unlock(&lo->lo_mutex); |
322c4293 | 1210 | |
bf23747e TH |
1211 | /* |
1212 | * Need not hold lo_mutex to fput backing file. Calling fput holding | |
1213 | * lo_mutex triggers a circular lock dependency possibility warning as | |
1214 | * fput can take open_mutex which is usually taken before lo_mutex. | |
1215 | */ | |
1216 | fput(filp); | |
1da177e4 LT |
1217 | } |
1218 | ||
a2505b79 JK |
1219 | static int loop_clr_fd(struct loop_device *lo) |
1220 | { | |
7ccd0791 JK |
1221 | int err; |
1222 | ||
158eaeba TH |
1223 | /* |
1224 | * Since lo_ioctl() is called without locks held, it is possible that | |
1225 | * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel. | |
1226 | * | |
1227 | * Therefore, use global lock when setting Lo_rundown state in order to | |
1228 | * make sure that loop_validate_file() will fail if the "struct file" | |
1229 | * which loop_configure()/loop_change_fd() found via fget() was this | |
1230 | * loop device. | |
1231 | */ | |
1232 | err = loop_global_lock_killable(lo, true); | |
7ccd0791 JK |
1233 | if (err) |
1234 | return err; | |
1235 | if (lo->lo_state != Lo_bound) { | |
158eaeba | 1236 | loop_global_unlock(lo, true); |
a2505b79 | 1237 | return -ENXIO; |
7ccd0791 | 1238 | } |
a2505b79 JK |
1239 | /* |
1240 | * If we've explicitly asked to tear down the loop device, | |
1241 | * and it has an elevated reference count, set it for auto-teardown when | |
1242 | * the last reference goes away. This stops $!~#$@ udev from | |
1243 | * preventing teardown because it decided that it needs to run blkid on | |
1244 | * the loopback device whenever they appear. xfstests is notorious for | |
1245 | * failing tests because blkid via udev races with a losetup | |
1246 | * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d | |
1247 | * command to fail with EBUSY. | |
1248 | */ | |
a0e286b6 | 1249 | if (disk_openers(lo->lo_disk) > 1) { |
a2505b79 | 1250 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; |
158eaeba | 1251 | loop_global_unlock(lo, true); |
a2505b79 JK |
1252 | return 0; |
1253 | } | |
1254 | lo->lo_state = Lo_rundown; | |
158eaeba | 1255 | loop_global_unlock(lo, true); |
a2505b79 | 1256 | |
bf23747e | 1257 | __loop_clr_fd(lo, false); |
6050fa4c | 1258 | return 0; |
a2505b79 JK |
1259 | } |
1260 | ||
1da177e4 LT |
1261 | static int |
1262 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1263 | { | |
1264 | int err; | |
faf1d254 | 1265 | int prev_lo_flags; |
85b0a54a | 1266 | bool partscan = false; |
0c3796c2 | 1267 | bool size_changed = false; |
1da177e4 | 1268 | |
6cc8e743 | 1269 | err = mutex_lock_killable(&lo->lo_mutex); |
550df5fd JK |
1270 | if (err) |
1271 | return err; | |
550df5fd JK |
1272 | if (lo->lo_state != Lo_bound) { |
1273 | err = -ENXIO; | |
1274 | goto out_unlock; | |
1275 | } | |
1da177e4 | 1276 | |
5db470e2 JK |
1277 | if (lo->lo_offset != info->lo_offset || |
1278 | lo->lo_sizelimit != info->lo_sizelimit) { | |
0c3796c2 | 1279 | size_changed = true; |
5db470e2 | 1280 | sync_blockdev(lo->lo_device); |
f4bd34b1 | 1281 | invalidate_bdev(lo->lo_device); |
5db470e2 JK |
1282 | } |
1283 | ||
ecdd0959 ML |
1284 | /* I/O need to be drained during transfer transition */ |
1285 | blk_mq_freeze_queue(lo->lo_queue); | |
1286 | ||
faf1d254 | 1287 | prev_lo_flags = lo->lo_flags; |
1da177e4 | 1288 | |
0c3796c2 | 1289 | err = loop_set_status_from_info(lo, info); |
1da177e4 | 1290 | if (err) |
550df5fd | 1291 | goto out_unfreeze; |
1da177e4 | 1292 | |
faf1d254 | 1293 | /* Mask out flags that can't be set using LOOP_SET_STATUS. */ |
6ac92fb5 | 1294 | lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; |
faf1d254 MC |
1295 | /* For those flags, use the previous values instead */ |
1296 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; | |
1297 | /* For flags that can't be cleared, use previous values too */ | |
1298 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; | |
1299 | ||
b0bd158d MC |
1300 | if (size_changed) { |
1301 | loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, | |
1302 | lo->lo_backing_file); | |
1303 | loop_set_size(lo, new_size); | |
b040ad9c | 1304 | } |
541c742a | 1305 | |
dfaa2ef6 | 1306 | loop_config_discard(lo); |
1da177e4 | 1307 | |
2e5ab5f3 ML |
1308 | /* update dio if lo_offset or transfer is changed */ |
1309 | __loop_update_dio(lo, lo->use_dio); | |
1310 | ||
550df5fd | 1311 | out_unfreeze: |
ecdd0959 | 1312 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 | 1313 | |
faf1d254 MC |
1314 | if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && |
1315 | !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { | |
b9684a71 | 1316 | clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
85b0a54a | 1317 | partscan = true; |
e02898b4 | 1318 | } |
550df5fd | 1319 | out_unlock: |
6cc8e743 | 1320 | mutex_unlock(&lo->lo_mutex); |
85b0a54a | 1321 | if (partscan) |
0384264e | 1322 | loop_reread_partitions(lo); |
e02898b4 | 1323 | |
ecdd0959 | 1324 | return err; |
1da177e4 LT |
1325 | } |
1326 | ||
1327 | static int | |
1328 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1329 | { | |
b1ab5fa3 | 1330 | struct path path; |
1da177e4 | 1331 | struct kstat stat; |
2d1d4c1e | 1332 | int ret; |
1da177e4 | 1333 | |
6cc8e743 | 1334 | ret = mutex_lock_killable(&lo->lo_mutex); |
4a5ce9ba JK |
1335 | if (ret) |
1336 | return ret; | |
2d1d4c1e | 1337 | if (lo->lo_state != Lo_bound) { |
6cc8e743 | 1338 | mutex_unlock(&lo->lo_mutex); |
1da177e4 | 1339 | return -ENXIO; |
2d1d4c1e OS |
1340 | } |
1341 | ||
1da177e4 LT |
1342 | memset(info, 0, sizeof(*info)); |
1343 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1344 | info->lo_offset = lo->lo_offset; |
1345 | info->lo_sizelimit = lo->lo_sizelimit; | |
1346 | info->lo_flags = lo->lo_flags; | |
1347 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
2d1d4c1e | 1348 | |
6cc8e743 | 1349 | /* Drop lo_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1350 | path = lo->lo_backing_file->f_path; |
1351 | path_get(&path); | |
6cc8e743 | 1352 | mutex_unlock(&lo->lo_mutex); |
b1ab5fa3 | 1353 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1354 | if (!ret) { |
1355 | info->lo_device = huge_encode_dev(stat.dev); | |
1356 | info->lo_inode = stat.ino; | |
1357 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1358 | } | |
b1ab5fa3 | 1359 | path_put(&path); |
2d1d4c1e | 1360 | return ret; |
1da177e4 LT |
1361 | } |
1362 | ||
1363 | static void | |
1364 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1365 | { | |
1366 | memset(info64, 0, sizeof(*info64)); | |
1367 | info64->lo_number = info->lo_number; | |
1368 | info64->lo_device = info->lo_device; | |
1369 | info64->lo_inode = info->lo_inode; | |
1370 | info64->lo_rdevice = info->lo_rdevice; | |
1371 | info64->lo_offset = info->lo_offset; | |
1372 | info64->lo_sizelimit = 0; | |
1da177e4 | 1373 | info64->lo_flags = info->lo_flags; |
47e96246 | 1374 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); |
1da177e4 LT |
1375 | } |
1376 | ||
1377 | static int | |
1378 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1379 | { | |
1380 | memset(info, 0, sizeof(*info)); | |
1381 | info->lo_number = info64->lo_number; | |
1382 | info->lo_device = info64->lo_device; | |
1383 | info->lo_inode = info64->lo_inode; | |
1384 | info->lo_rdevice = info64->lo_rdevice; | |
1385 | info->lo_offset = info64->lo_offset; | |
1da177e4 | 1386 | info->lo_flags = info64->lo_flags; |
47e96246 | 1387 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); |
1da177e4 LT |
1388 | |
1389 | /* error in case values were truncated */ | |
1390 | if (info->lo_device != info64->lo_device || | |
1391 | info->lo_rdevice != info64->lo_rdevice || | |
1392 | info->lo_inode != info64->lo_inode || | |
1393 | info->lo_offset != info64->lo_offset) | |
1394 | return -EOVERFLOW; | |
1395 | ||
1396 | return 0; | |
1397 | } | |
1398 | ||
1399 | static int | |
1400 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1401 | { | |
1402 | struct loop_info info; | |
1403 | struct loop_info64 info64; | |
1404 | ||
1405 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1406 | return -EFAULT; | |
1407 | loop_info64_from_old(&info, &info64); | |
1408 | return loop_set_status(lo, &info64); | |
1409 | } | |
1410 | ||
1411 | static int | |
1412 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1413 | { | |
1414 | struct loop_info64 info64; | |
1415 | ||
1416 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1417 | return -EFAULT; | |
1418 | return loop_set_status(lo, &info64); | |
1419 | } | |
1420 | ||
1421 | static int | |
1422 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1423 | struct loop_info info; | |
1424 | struct loop_info64 info64; | |
bdac616d | 1425 | int err; |
1da177e4 | 1426 | |
4a5ce9ba | 1427 | if (!arg) |
bdac616d | 1428 | return -EINVAL; |
bdac616d | 1429 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1430 | if (!err) |
1431 | err = loop_info64_to_old(&info64, &info); | |
1432 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1433 | err = -EFAULT; | |
1434 | ||
1435 | return err; | |
1436 | } | |
1437 | ||
1438 | static int | |
1439 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1440 | struct loop_info64 info64; | |
bdac616d | 1441 | int err; |
1da177e4 | 1442 | |
4a5ce9ba | 1443 | if (!arg) |
bdac616d | 1444 | return -EINVAL; |
bdac616d | 1445 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1446 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1447 | err = -EFAULT; | |
1448 | ||
1449 | return err; | |
1450 | } | |
1451 | ||
51001b7d | 1452 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1453 | { |
0a6ed1b5 MC |
1454 | loff_t size; |
1455 | ||
53d66608 | 1456 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1457 | return -ENXIO; |
53d66608 | 1458 | |
0a6ed1b5 MC |
1459 | size = get_loop_size(lo, lo->lo_backing_file); |
1460 | loop_set_size(lo, size); | |
083a6a50 MC |
1461 | |
1462 | return 0; | |
53d66608 O |
1463 | } |
1464 | ||
ab1cb278 ML |
1465 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1466 | { | |
1467 | int error = -ENXIO; | |
1468 | if (lo->lo_state != Lo_bound) | |
1469 | goto out; | |
1470 | ||
1471 | __loop_update_dio(lo, !!arg); | |
1472 | if (lo->use_dio == !!arg) | |
1473 | return 0; | |
1474 | error = -EINVAL; | |
1475 | out: | |
1476 | return error; | |
1477 | } | |
1478 | ||
89e4fdec OS |
1479 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1480 | { | |
5db470e2 JK |
1481 | int err = 0; |
1482 | ||
89e4fdec OS |
1483 | if (lo->lo_state != Lo_bound) |
1484 | return -ENXIO; | |
1485 | ||
af3c570f | 1486 | err = blk_validate_block_size(arg); |
3448914e MC |
1487 | if (err) |
1488 | return err; | |
89e4fdec | 1489 | |
7e81f99a MC |
1490 | if (lo->lo_queue->limits.logical_block_size == arg) |
1491 | return 0; | |
1492 | ||
1493 | sync_blockdev(lo->lo_device); | |
f4bd34b1 | 1494 | invalidate_bdev(lo->lo_device); |
5db470e2 | 1495 | |
89e4fdec | 1496 | blk_mq_freeze_queue(lo->lo_queue); |
89e4fdec | 1497 | blk_queue_logical_block_size(lo->lo_queue, arg); |
bf093753 OS |
1498 | blk_queue_physical_block_size(lo->lo_queue, arg); |
1499 | blk_queue_io_min(lo->lo_queue, arg); | |
89e4fdec | 1500 | loop_update_dio(lo); |
89e4fdec OS |
1501 | blk_mq_unfreeze_queue(lo->lo_queue); |
1502 | ||
5db470e2 | 1503 | return err; |
89e4fdec OS |
1504 | } |
1505 | ||
a1316544 JK |
1506 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1507 | unsigned long arg) | |
1da177e4 | 1508 | { |
1da177e4 LT |
1509 | int err; |
1510 | ||
6cc8e743 | 1511 | err = mutex_lock_killable(&lo->lo_mutex); |
3148ffbd | 1512 | if (err) |
a1316544 JK |
1513 | return err; |
1514 | switch (cmd) { | |
1515 | case LOOP_SET_CAPACITY: | |
1516 | err = loop_set_capacity(lo); | |
1517 | break; | |
1518 | case LOOP_SET_DIRECT_IO: | |
1519 | err = loop_set_dio(lo, arg); | |
1520 | break; | |
1521 | case LOOP_SET_BLOCK_SIZE: | |
1522 | err = loop_set_block_size(lo, arg); | |
1523 | break; | |
1524 | default: | |
47e96246 | 1525 | err = -EINVAL; |
a1316544 | 1526 | } |
6cc8e743 | 1527 | mutex_unlock(&lo->lo_mutex); |
a1316544 JK |
1528 | return err; |
1529 | } | |
1530 | ||
05bdb996 | 1531 | static int lo_ioctl(struct block_device *bdev, blk_mode_t mode, |
a1316544 JK |
1532 | unsigned int cmd, unsigned long arg) |
1533 | { | |
1534 | struct loop_device *lo = bdev->bd_disk->private_data; | |
571fae6e | 1535 | void __user *argp = (void __user *) arg; |
a1316544 | 1536 | int err; |
3148ffbd | 1537 | |
1da177e4 | 1538 | switch (cmd) { |
3448914e MC |
1539 | case LOOP_SET_FD: { |
1540 | /* | |
1541 | * Legacy case - pass in a zeroed out struct loop_config with | |
1542 | * only the file descriptor set , which corresponds with the | |
1543 | * default parameters we'd have used otherwise. | |
1544 | */ | |
1545 | struct loop_config config; | |
1546 | ||
1547 | memset(&config, 0, sizeof(config)); | |
1548 | config.fd = arg; | |
1549 | ||
1550 | return loop_configure(lo, mode, bdev, &config); | |
1551 | } | |
1552 | case LOOP_CONFIGURE: { | |
1553 | struct loop_config config; | |
1554 | ||
1555 | if (copy_from_user(&config, argp, sizeof(config))) | |
1556 | return -EFAULT; | |
1557 | ||
1558 | return loop_configure(lo, mode, bdev, &config); | |
1559 | } | |
1da177e4 | 1560 | case LOOP_CHANGE_FD: |
c3710770 | 1561 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1562 | case LOOP_CLR_FD: |
7ccd0791 | 1563 | return loop_clr_fd(lo); |
1da177e4 | 1564 | case LOOP_SET_STATUS: |
7035b5df | 1565 | err = -EPERM; |
05bdb996 | 1566 | if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN)) |
571fae6e | 1567 | err = loop_set_status_old(lo, argp); |
1da177e4 LT |
1568 | break; |
1569 | case LOOP_GET_STATUS: | |
571fae6e | 1570 | return loop_get_status_old(lo, argp); |
1da177e4 | 1571 | case LOOP_SET_STATUS64: |
7035b5df | 1572 | err = -EPERM; |
05bdb996 | 1573 | if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN)) |
571fae6e | 1574 | err = loop_set_status64(lo, argp); |
1da177e4 LT |
1575 | break; |
1576 | case LOOP_GET_STATUS64: | |
571fae6e | 1577 | return loop_get_status64(lo, argp); |
a1316544 | 1578 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1579 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1580 | case LOOP_SET_BLOCK_SIZE: |
05bdb996 | 1581 | if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN)) |
a1316544 | 1582 | return -EPERM; |
df561f66 | 1583 | fallthrough; |
1da177e4 | 1584 | default: |
a1316544 JK |
1585 | err = lo_simple_ioctl(lo, cmd, arg); |
1586 | break; | |
1da177e4 | 1587 | } |
f028f3b2 | 1588 | |
1da177e4 LT |
1589 | return err; |
1590 | } | |
1591 | ||
863d5b82 DH |
1592 | #ifdef CONFIG_COMPAT |
1593 | struct compat_loop_info { | |
1594 | compat_int_t lo_number; /* ioctl r/o */ | |
1595 | compat_dev_t lo_device; /* ioctl r/o */ | |
1596 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1597 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1598 | compat_int_t lo_offset; | |
f941c51e | 1599 | compat_int_t lo_encrypt_type; /* obsolete, ignored */ |
863d5b82 DH |
1600 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ |
1601 | compat_int_t lo_flags; /* ioctl r/o */ | |
1602 | char lo_name[LO_NAME_SIZE]; | |
1603 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1604 | compat_ulong_t lo_init[2]; | |
1605 | char reserved[4]; | |
1606 | }; | |
1607 | ||
1608 | /* | |
1609 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1610 | * - noinlined to reduce stack space usage in main part of driver | |
1611 | */ | |
1612 | static noinline int | |
ba674cfc | 1613 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1614 | struct loop_info64 *info64) |
1615 | { | |
1616 | struct compat_loop_info info; | |
1617 | ||
1618 | if (copy_from_user(&info, arg, sizeof(info))) | |
1619 | return -EFAULT; | |
1620 | ||
1621 | memset(info64, 0, sizeof(*info64)); | |
1622 | info64->lo_number = info.lo_number; | |
1623 | info64->lo_device = info.lo_device; | |
1624 | info64->lo_inode = info.lo_inode; | |
1625 | info64->lo_rdevice = info.lo_rdevice; | |
1626 | info64->lo_offset = info.lo_offset; | |
1627 | info64->lo_sizelimit = 0; | |
863d5b82 | 1628 | info64->lo_flags = info.lo_flags; |
47e96246 | 1629 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); |
863d5b82 DH |
1630 | return 0; |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1635 | * - noinlined to reduce stack space usage in main part of driver | |
1636 | */ | |
1637 | static noinline int | |
1638 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1639 | struct compat_loop_info __user *arg) | |
1640 | { | |
1641 | struct compat_loop_info info; | |
1642 | ||
1643 | memset(&info, 0, sizeof(info)); | |
1644 | info.lo_number = info64->lo_number; | |
1645 | info.lo_device = info64->lo_device; | |
1646 | info.lo_inode = info64->lo_inode; | |
1647 | info.lo_rdevice = info64->lo_rdevice; | |
1648 | info.lo_offset = info64->lo_offset; | |
863d5b82 | 1649 | info.lo_flags = info64->lo_flags; |
47e96246 | 1650 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); |
863d5b82 DH |
1651 | |
1652 | /* error in case values were truncated */ | |
1653 | if (info.lo_device != info64->lo_device || | |
1654 | info.lo_rdevice != info64->lo_rdevice || | |
1655 | info.lo_inode != info64->lo_inode || | |
47e96246 | 1656 | info.lo_offset != info64->lo_offset) |
863d5b82 DH |
1657 | return -EOVERFLOW; |
1658 | ||
1659 | if (copy_to_user(arg, &info, sizeof(info))) | |
1660 | return -EFAULT; | |
1661 | return 0; | |
1662 | } | |
1663 | ||
1664 | static int | |
1665 | loop_set_status_compat(struct loop_device *lo, | |
1666 | const struct compat_loop_info __user *arg) | |
1667 | { | |
1668 | struct loop_info64 info64; | |
1669 | int ret; | |
1670 | ||
1671 | ret = loop_info64_from_compat(arg, &info64); | |
1672 | if (ret < 0) | |
1673 | return ret; | |
1674 | return loop_set_status(lo, &info64); | |
1675 | } | |
1676 | ||
1677 | static int | |
1678 | loop_get_status_compat(struct loop_device *lo, | |
1679 | struct compat_loop_info __user *arg) | |
1680 | { | |
1681 | struct loop_info64 info64; | |
bdac616d | 1682 | int err; |
863d5b82 | 1683 | |
4a5ce9ba | 1684 | if (!arg) |
bdac616d | 1685 | return -EINVAL; |
bdac616d | 1686 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1687 | if (!err) |
1688 | err = loop_info64_to_compat(&info64, arg); | |
1689 | return err; | |
1690 | } | |
1691 | ||
05bdb996 | 1692 | static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode, |
bb214884 | 1693 | unsigned int cmd, unsigned long arg) |
863d5b82 | 1694 | { |
bb214884 | 1695 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1696 | int err; |
1697 | ||
863d5b82 DH |
1698 | switch(cmd) { |
1699 | case LOOP_SET_STATUS: | |
550df5fd JK |
1700 | err = loop_set_status_compat(lo, |
1701 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1702 | break; |
1703 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1704 | err = loop_get_status_compat(lo, |
1705 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1706 | break; |
53d66608 | 1707 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1708 | case LOOP_CLR_FD: |
1709 | case LOOP_GET_STATUS64: | |
1710 | case LOOP_SET_STATUS64: | |
3448914e | 1711 | case LOOP_CONFIGURE: |
863d5b82 | 1712 | arg = (unsigned long) compat_ptr(arg); |
df561f66 | 1713 | fallthrough; |
863d5b82 DH |
1714 | case LOOP_SET_FD: |
1715 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1716 | case LOOP_SET_BLOCK_SIZE: |
fdbe4eee | 1717 | case LOOP_SET_DIRECT_IO: |
bb214884 | 1718 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1719 | break; |
1720 | default: | |
1721 | err = -ENOIOCTLCMD; | |
1722 | break; | |
1723 | } | |
863d5b82 DH |
1724 | return err; |
1725 | } | |
1726 | #endif | |
1727 | ||
ae220766 | 1728 | static void lo_release(struct gendisk *disk) |
1da177e4 | 1729 | { |
6cc8e743 | 1730 | struct loop_device *lo = disk->private_data; |
1da177e4 | 1731 | |
a0e286b6 CH |
1732 | if (disk_openers(disk) > 0) |
1733 | return; | |
14f27939 | 1734 | |
a0e286b6 CH |
1735 | mutex_lock(&lo->lo_mutex); |
1736 | if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) { | |
a2505b79 | 1737 | lo->lo_state = Lo_rundown; |
6cc8e743 | 1738 | mutex_unlock(&lo->lo_mutex); |
14f27939 MB |
1739 | /* |
1740 | * In autoclear mode, stop the loop thread | |
1741 | * and remove configuration after last close. | |
1742 | */ | |
bf23747e | 1743 | __loop_clr_fd(lo, true); |
7ccd0791 | 1744 | return; |
14f27939 | 1745 | } |
6cc8e743 | 1746 | mutex_unlock(&lo->lo_mutex); |
ae665016 LT |
1747 | } |
1748 | ||
d2c7f56f CH |
1749 | static void lo_free_disk(struct gendisk *disk) |
1750 | { | |
1751 | struct loop_device *lo = disk->private_data; | |
1752 | ||
d292dc80 CH |
1753 | if (lo->workqueue) |
1754 | destroy_workqueue(lo->workqueue); | |
1755 | loop_free_idle_workers(lo, true); | |
292a089d | 1756 | timer_shutdown_sync(&lo->timer); |
d2c7f56f CH |
1757 | mutex_destroy(&lo->lo_mutex); |
1758 | kfree(lo); | |
1759 | } | |
1760 | ||
83d5cde4 | 1761 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1762 | .owner = THIS_MODULE, |
bb214884 AV |
1763 | .release = lo_release, |
1764 | .ioctl = lo_ioctl, | |
863d5b82 | 1765 | #ifdef CONFIG_COMPAT |
bb214884 | 1766 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1767 | #endif |
d2c7f56f | 1768 | .free_disk = lo_free_disk, |
1da177e4 LT |
1769 | }; |
1770 | ||
1771 | /* | |
1772 | * And now the modules code and kernel interface. | |
1773 | */ | |
85c50197 IM |
1774 | |
1775 | /* | |
1776 | * If max_loop is specified, create that many devices upfront. | |
1777 | * This also becomes a hard limit. If max_loop is not specified, | |
1778 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module | |
1779 | * init time. Loop devices can be requested on-demand with the | |
1780 | * /dev/loop-control interface, or be instantiated by accessing | |
1781 | * a 'dead' device node. | |
1782 | */ | |
1783 | static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT; | |
5657a819 | 1784 | module_param(max_loop, int, 0444); |
a47653fc | 1785 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
5657a819 | 1786 | module_param(max_part, int, 0444); |
476a4813 | 1787 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
ef44c508 CK |
1788 | |
1789 | static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; | |
1790 | ||
1791 | static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) | |
1792 | { | |
e152a05f | 1793 | int qd, ret; |
ef44c508 | 1794 | |
e152a05f BVA |
1795 | ret = kstrtoint(s, 0, &qd); |
1796 | if (ret < 0) | |
1797 | return ret; | |
1798 | if (qd < 1) | |
1799 | return -EINVAL; | |
1800 | hw_queue_depth = qd; | |
1801 | return 0; | |
ef44c508 CK |
1802 | } |
1803 | ||
1804 | static const struct kernel_param_ops loop_hw_qdepth_param_ops = { | |
1805 | .set = loop_set_hw_queue_depth, | |
1806 | .get = param_get_int, | |
1807 | }; | |
1808 | ||
1809 | device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); | |
e152a05f | 1810 | MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH)); |
ef44c508 | 1811 | |
1da177e4 LT |
1812 | MODULE_LICENSE("GPL"); |
1813 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1814 | ||
fc17b653 | 1815 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1816 | const struct blk_mq_queue_data *bd) |
1817 | { | |
1894e916 JA |
1818 | struct request *rq = bd->rq; |
1819 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1820 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 1821 | |
1894e916 | 1822 | blk_mq_start_request(rq); |
b5dd2f60 | 1823 | |
f4aa4c7b | 1824 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1825 | return BLK_STS_IOERR; |
f4aa4c7b | 1826 | |
1894e916 | 1827 | switch (req_op(rq)) { |
f0225cac CH |
1828 | case REQ_OP_FLUSH: |
1829 | case REQ_OP_DISCARD: | |
19372e27 | 1830 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1831 | cmd->use_aio = false; |
f0225cac CH |
1832 | break; |
1833 | default: | |
1834 | cmd->use_aio = lo->use_dio; | |
1835 | break; | |
1836 | } | |
bc07c10a | 1837 | |
d4478e92 | 1838 | /* always use the first bio's css */ |
c74d40e8 DS |
1839 | cmd->blkcg_css = NULL; |
1840 | cmd->memcg_css = NULL; | |
0b508bc9 | 1841 | #ifdef CONFIG_BLK_CGROUP |
bbb1ebe7 CH |
1842 | if (rq->bio) { |
1843 | cmd->blkcg_css = bio_blkcg_css(rq->bio); | |
c74d40e8 | 1844 | #ifdef CONFIG_MEMCG |
bbb1ebe7 CH |
1845 | if (cmd->blkcg_css) { |
1846 | cmd->memcg_css = | |
1847 | cgroup_get_e_css(cmd->blkcg_css->cgroup, | |
1848 | &memory_cgrp_subsys); | |
1849 | } | |
c74d40e8 DS |
1850 | #endif |
1851 | } | |
d4478e92 | 1852 | #endif |
87579e9b | 1853 | loop_queue_work(lo, cmd); |
b5dd2f60 | 1854 | |
fc17b653 | 1855 | return BLK_STS_OK; |
b5dd2f60 ML |
1856 | } |
1857 | ||
1858 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1859 | { | |
9b0cb770 BVA |
1860 | struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css; |
1861 | struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css; | |
1894e916 JA |
1862 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
1863 | const bool write = op_is_write(req_op(rq)); | |
1864 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 1865 | int ret = 0; |
c74d40e8 | 1866 | struct mem_cgroup *old_memcg = NULL; |
9b0cb770 | 1867 | const bool use_aio = cmd->use_aio; |
b5dd2f60 | 1868 | |
f4829a9b CH |
1869 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1870 | ret = -EIO; | |
b5dd2f60 | 1871 | goto failed; |
f4829a9b | 1872 | } |
b5dd2f60 | 1873 | |
9b0cb770 BVA |
1874 | if (cmd_blkcg_css) |
1875 | kthread_associate_blkcg(cmd_blkcg_css); | |
1876 | if (cmd_memcg_css) | |
c74d40e8 | 1877 | old_memcg = set_active_memcg( |
9b0cb770 | 1878 | mem_cgroup_from_css(cmd_memcg_css)); |
c74d40e8 | 1879 | |
9b0cb770 BVA |
1880 | /* |
1881 | * do_req_filebacked() may call blk_mq_complete_request() synchronously | |
1882 | * or asynchronously if using aio. Hence, do not touch 'cmd' after | |
1883 | * do_req_filebacked() has returned unless we are sure that 'cmd' has | |
1884 | * not yet been completed. | |
1885 | */ | |
1894e916 | 1886 | ret = do_req_filebacked(lo, rq); |
c74d40e8 | 1887 | |
9b0cb770 | 1888 | if (cmd_blkcg_css) |
c74d40e8 DS |
1889 | kthread_associate_blkcg(NULL); |
1890 | ||
9b0cb770 | 1891 | if (cmd_memcg_css) { |
c74d40e8 | 1892 | set_active_memcg(old_memcg); |
9b0cb770 | 1893 | css_put(cmd_memcg_css); |
c74d40e8 | 1894 | } |
b5dd2f60 | 1895 | failed: |
bc07c10a | 1896 | /* complete non-aio request */ |
9b0cb770 | 1897 | if (!use_aio || ret) { |
8cd55087 EG |
1898 | if (ret == -EOPNOTSUPP) |
1899 | cmd->ret = ret; | |
1900 | else | |
1901 | cmd->ret = ret ? -EIO : 0; | |
15f73f5b CH |
1902 | if (likely(!blk_should_fake_timeout(rq->q))) |
1903 | blk_mq_complete_request(rq); | |
fe2cb290 | 1904 | } |
b5dd2f60 ML |
1905 | } |
1906 | ||
87579e9b DS |
1907 | static void loop_process_work(struct loop_worker *worker, |
1908 | struct list_head *cmd_list, struct loop_device *lo) | |
b5dd2f60 | 1909 | { |
87579e9b DS |
1910 | int orig_flags = current->flags; |
1911 | struct loop_cmd *cmd; | |
b5dd2f60 | 1912 | |
87579e9b DS |
1913 | current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; |
1914 | spin_lock_irq(&lo->lo_work_lock); | |
1915 | while (!list_empty(cmd_list)) { | |
1916 | cmd = container_of( | |
1917 | cmd_list->next, struct loop_cmd, list_entry); | |
1918 | list_del(cmd_list->next); | |
1919 | spin_unlock_irq(&lo->lo_work_lock); | |
1920 | ||
1921 | loop_handle_cmd(cmd); | |
1922 | cond_resched(); | |
1923 | ||
1924 | spin_lock_irq(&lo->lo_work_lock); | |
1925 | } | |
1926 | ||
1927 | /* | |
1928 | * We only add to the idle list if there are no pending cmds | |
1929 | * *and* the worker will not run again which ensures that it | |
1930 | * is safe to free any worker on the idle list | |
1931 | */ | |
1932 | if (worker && !work_pending(&worker->work)) { | |
1933 | worker->last_ran_at = jiffies; | |
1934 | list_add_tail(&worker->idle_list, &lo->idle_worker_list); | |
1935 | loop_set_timer(lo); | |
1936 | } | |
1937 | spin_unlock_irq(&lo->lo_work_lock); | |
1938 | current->flags = orig_flags; | |
b5dd2f60 ML |
1939 | } |
1940 | ||
87579e9b | 1941 | static void loop_workfn(struct work_struct *work) |
b5dd2f60 | 1942 | { |
87579e9b DS |
1943 | struct loop_worker *worker = |
1944 | container_of(work, struct loop_worker, work); | |
1945 | loop_process_work(worker, &worker->cmd_list, worker->lo); | |
1946 | } | |
b5dd2f60 | 1947 | |
87579e9b DS |
1948 | static void loop_rootcg_workfn(struct work_struct *work) |
1949 | { | |
1950 | struct loop_device *lo = | |
1951 | container_of(work, struct loop_device, rootcg_work); | |
1952 | loop_process_work(NULL, &lo->rootcg_cmd_list, lo); | |
1953 | } | |
1954 | ||
f363b089 | 1955 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1956 | .queue_rq = loop_queue_rq, |
fe2cb290 | 1957 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1958 | }; |
1959 | ||
d6da83d0 | 1960 | static int loop_add(int i) |
73285082 KC |
1961 | { |
1962 | struct loop_device *lo; | |
1963 | struct gendisk *disk; | |
34dd82af | 1964 | int err; |
73285082 | 1965 | |
68d740d7 | 1966 | err = -ENOMEM; |
73285082 | 1967 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1968 | if (!lo) |
73285082 | 1969 | goto out; |
b15ed546 CH |
1970 | lo->worker_tree = RB_ROOT; |
1971 | INIT_LIST_HEAD(&lo->idle_worker_list); | |
1972 | timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE); | |
ef7e7c82 MP |
1973 | lo->lo_state = Lo_unbound; |
1974 | ||
18d1f200 CH |
1975 | err = mutex_lock_killable(&loop_ctl_mutex); |
1976 | if (err) | |
1977 | goto out_free_dev; | |
1978 | ||
c718aa65 | 1979 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 1980 | if (i >= 0) { |
c718aa65 TH |
1981 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
1982 | if (err == -ENOSPC) | |
34dd82af | 1983 | err = -EEXIST; |
34dd82af | 1984 | } else { |
c718aa65 | 1985 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af | 1986 | } |
1c500ad7 | 1987 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 1988 | if (err < 0) |
1c500ad7 | 1989 | goto out_free_dev; |
c718aa65 | 1990 | i = err; |
73285082 | 1991 | |
b5dd2f60 ML |
1992 | lo->tag_set.ops = &loop_mq_ops; |
1993 | lo->tag_set.nr_hw_queues = 1; | |
ef44c508 | 1994 | lo->tag_set.queue_depth = hw_queue_depth; |
b5dd2f60 ML |
1995 | lo->tag_set.numa_node = NUMA_NO_NODE; |
1996 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
2112f5c1 BVA |
1997 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | |
1998 | BLK_MQ_F_NO_SCHED_BY_DEFAULT; | |
b5dd2f60 ML |
1999 | lo->tag_set.driver_data = lo; |
2000 | ||
2001 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
2002 | if (err) | |
3ec981e3 | 2003 | goto out_free_idr; |
73285082 | 2004 | |
1c99502f CH |
2005 | disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); |
2006 | if (IS_ERR(disk)) { | |
2007 | err = PTR_ERR(disk); | |
b5dd2f60 ML |
2008 | goto out_cleanup_tags; |
2009 | } | |
1c99502f | 2010 | lo->lo_queue = lo->lo_disk->queue; |
ef7e7c82 | 2011 | |
54bb0ade | 2012 | blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); |
40326d8a | 2013 | |
5b5e20f4 | 2014 | /* |
40326d8a SL |
2015 | * By default, we do buffer IO, so it doesn't make sense to enable |
2016 | * merge because the I/O submitted to backing file is handled page by | |
2017 | * page. For directio mode, merge does help to dispatch bigger request | |
2018 | * to underlayer disk. We will enable merge once directio is enabled. | |
5b5e20f4 | 2019 | */ |
8b904b5b | 2020 | blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); |
5b5e20f4 | 2021 | |
e03c8dd1 KS |
2022 | /* |
2023 | * Disable partition scanning by default. The in-kernel partition | |
2024 | * scanning can be requested individually per-device during its | |
2025 | * setup. Userspace can always add and remove partitions from all | |
2026 | * devices. The needed partition minors are allocated from the | |
2027 | * extended minor space, the main loop device numbers will continue | |
2028 | * to match the loop minors, regardless of the number of partitions | |
2029 | * used. | |
2030 | * | |
2031 | * If max_part is given, partition scanning is globally enabled for | |
2032 | * all loop devices. The minors for the main loop devices will be | |
2033 | * multiples of max_part. | |
2034 | * | |
2035 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
2036 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
2037 | * complicated, are too static, inflexible and may surprise | |
2038 | * userspace tools. Parameters like this in general should be avoided. | |
2039 | */ | |
2040 | if (!part_shift) | |
b9684a71 | 2041 | set_bit(GD_SUPPRESS_PART_SCAN, &disk->state); |
6cc8e743 | 2042 | mutex_init(&lo->lo_mutex); |
73285082 | 2043 | lo->lo_number = i; |
73285082 | 2044 | spin_lock_init(&lo->lo_lock); |
87579e9b | 2045 | spin_lock_init(&lo->lo_work_lock); |
d292dc80 CH |
2046 | INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); |
2047 | INIT_LIST_HEAD(&lo->rootcg_cmd_list); | |
73285082 | 2048 | disk->major = LOOP_MAJOR; |
476a4813 | 2049 | disk->first_minor = i << part_shift; |
1c99502f | 2050 | disk->minors = 1 << part_shift; |
73285082 KC |
2051 | disk->fops = &lo_fops; |
2052 | disk->private_data = lo; | |
2053 | disk->queue = lo->lo_queue; | |
9f65c489 MC |
2054 | disk->events = DISK_EVENT_MEDIA_CHANGE; |
2055 | disk->event_flags = DISK_EVENT_FLAG_UEVENT; | |
73285082 | 2056 | sprintf(disk->disk_name, "loop%d", i); |
1c500ad7 | 2057 | /* Make this loop device reachable from pathname. */ |
905705f0 LC |
2058 | err = add_disk(disk); |
2059 | if (err) | |
2060 | goto out_cleanup_disk; | |
2061 | ||
1c500ad7 TH |
2062 | /* Show this loop device. */ |
2063 | mutex_lock(&loop_ctl_mutex); | |
2064 | lo->idr_visible = true; | |
18d1f200 | 2065 | mutex_unlock(&loop_ctl_mutex); |
905705f0 | 2066 | |
18d1f200 | 2067 | return i; |
73285082 | 2068 | |
905705f0 | 2069 | out_cleanup_disk: |
8b9ab626 | 2070 | put_disk(disk); |
b5dd2f60 ML |
2071 | out_cleanup_tags: |
2072 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 | 2073 | out_free_idr: |
1c500ad7 | 2074 | mutex_lock(&loop_ctl_mutex); |
3ec981e3 | 2075 | idr_remove(&loop_index_idr, i); |
18d1f200 | 2076 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2077 | out_free_dev: |
2078 | kfree(lo); | |
2079 | out: | |
34dd82af | 2080 | return err; |
73285082 KC |
2081 | } |
2082 | ||
34dd82af | 2083 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2084 | { |
1c500ad7 | 2085 | /* Make this loop device unreachable from pathname. */ |
6cd18e71 | 2086 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 2087 | blk_mq_free_tag_set(&lo->tag_set); |
ce8d7861 | 2088 | |
1c500ad7 TH |
2089 | mutex_lock(&loop_ctl_mutex); |
2090 | idr_remove(&loop_index_idr, lo->lo_number); | |
2091 | mutex_unlock(&loop_ctl_mutex); | |
d2c7f56f CH |
2092 | |
2093 | put_disk(lo->lo_disk); | |
73285082 | 2094 | } |
1da177e4 | 2095 | |
8410d38c | 2096 | static void loop_probe(dev_t dev) |
73285082 | 2097 | { |
8410d38c | 2098 | int idx = MINOR(dev) >> part_shift; |
8410d38c CH |
2099 | |
2100 | if (max_loop && idx >= max_loop) | |
2101 | return; | |
4157fe0b | 2102 | loop_add(idx); |
f9d10764 CH |
2103 | } |
2104 | ||
2105 | static int loop_control_remove(int idx) | |
2106 | { | |
2107 | struct loop_device *lo; | |
2108 | int ret; | |
e5d66a10 CH |
2109 | |
2110 | if (idx < 0) { | |
e3f9387a | 2111 | pr_warn_once("deleting an unspecified loop device is not supported.\n"); |
e5d66a10 CH |
2112 | return -EINVAL; |
2113 | } | |
f9d10764 | 2114 | |
1c500ad7 | 2115 | /* Hide this loop device for serialization. */ |
f9d10764 CH |
2116 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2117 | if (ret) | |
2118 | return ret; | |
b9848081 | 2119 | lo = idr_find(&loop_index_idr, idx); |
1c500ad7 | 2120 | if (!lo || !lo->idr_visible) |
b9848081 | 2121 | ret = -ENODEV; |
1c500ad7 TH |
2122 | else |
2123 | lo->idr_visible = false; | |
2124 | mutex_unlock(&loop_ctl_mutex); | |
2125 | if (ret) | |
2126 | return ret; | |
f9d10764 | 2127 | |
1c500ad7 | 2128 | /* Check whether this loop device can be removed. */ |
f9d10764 CH |
2129 | ret = mutex_lock_killable(&lo->lo_mutex); |
2130 | if (ret) | |
1c500ad7 | 2131 | goto mark_visible; |
a0e286b6 | 2132 | if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) { |
f9d10764 CH |
2133 | mutex_unlock(&lo->lo_mutex); |
2134 | ret = -EBUSY; | |
1c500ad7 | 2135 | goto mark_visible; |
f9d10764 | 2136 | } |
a0e286b6 | 2137 | /* Mark this loop device as no more bound, but not quite unbound yet */ |
f9d10764 CH |
2138 | lo->lo_state = Lo_deleting; |
2139 | mutex_unlock(&lo->lo_mutex); | |
2140 | ||
f9d10764 | 2141 | loop_remove(lo); |
1c500ad7 TH |
2142 | return 0; |
2143 | ||
2144 | mark_visible: | |
2145 | /* Show this loop device again. */ | |
2146 | mutex_lock(&loop_ctl_mutex); | |
2147 | lo->idr_visible = true; | |
f9d10764 CH |
2148 | mutex_unlock(&loop_ctl_mutex); |
2149 | return ret; | |
2150 | } | |
2151 | ||
2152 | static int loop_control_get_free(int idx) | |
770fe30a KS |
2153 | { |
2154 | struct loop_device *lo; | |
b9848081 | 2155 | int id, ret; |
770fe30a | 2156 | |
0a42e99b JK |
2157 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2158 | if (ret) | |
2159 | return ret; | |
b9848081 | 2160 | idr_for_each_entry(&loop_index_idr, lo, id) { |
1c500ad7 TH |
2161 | /* Hitting a race results in creating a new loop device which is harmless. */ |
2162 | if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) | |
b9848081 CH |
2163 | goto found; |
2164 | } | |
f9d10764 | 2165 | mutex_unlock(&loop_ctl_mutex); |
18d1f200 | 2166 | return loop_add(-1); |
b9848081 CH |
2167 | found: |
2168 | mutex_unlock(&loop_ctl_mutex); | |
2169 | return id; | |
f9d10764 | 2170 | } |
0a42e99b | 2171 | |
f9d10764 CH |
2172 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2173 | unsigned long parm) | |
2174 | { | |
770fe30a KS |
2175 | switch (cmd) { |
2176 | case LOOP_CTL_ADD: | |
18d1f200 | 2177 | return loop_add(parm); |
770fe30a | 2178 | case LOOP_CTL_REMOVE: |
f9d10764 | 2179 | return loop_control_remove(parm); |
770fe30a | 2180 | case LOOP_CTL_GET_FREE: |
f9d10764 CH |
2181 | return loop_control_get_free(parm); |
2182 | default: | |
2183 | return -ENOSYS; | |
770fe30a | 2184 | } |
770fe30a KS |
2185 | } |
2186 | ||
2187 | static const struct file_operations loop_ctl_fops = { | |
2188 | .open = nonseekable_open, | |
2189 | .unlocked_ioctl = loop_control_ioctl, | |
2190 | .compat_ioctl = loop_control_ioctl, | |
2191 | .owner = THIS_MODULE, | |
2192 | .llseek = noop_llseek, | |
2193 | }; | |
2194 | ||
2195 | static struct miscdevice loop_misc = { | |
2196 | .minor = LOOP_CTRL_MINOR, | |
2197 | .name = "loop-control", | |
2198 | .fops = &loop_ctl_fops, | |
2199 | }; | |
2200 | ||
2201 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2202 | MODULE_ALIAS("devname:loop-control"); | |
2203 | ||
73285082 KC |
2204 | static int __init loop_init(void) |
2205 | { | |
85c50197 | 2206 | int i; |
770fe30a | 2207 | int err; |
a47653fc | 2208 | |
476a4813 | 2209 | part_shift = 0; |
ac04fee0 | 2210 | if (max_part > 0) { |
476a4813 LV |
2211 | part_shift = fls(max_part); |
2212 | ||
ac04fee0 NK |
2213 | /* |
2214 | * Adjust max_part according to part_shift as it is exported | |
2215 | * to user space so that user can decide correct minor number | |
2216 | * if [s]he want to create more devices. | |
2217 | * | |
2218 | * Note that -1 is required because partition 0 is reserved | |
2219 | * for the whole disk. | |
2220 | */ | |
2221 | max_part = (1UL << part_shift) - 1; | |
2222 | } | |
2223 | ||
b1a66504 GC |
2224 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2225 | err = -EINVAL; | |
a8c1d064 | 2226 | goto err_out; |
b1a66504 | 2227 | } |
78f4bb36 | 2228 | |
b1a66504 GC |
2229 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2230 | err = -EINVAL; | |
a8c1d064 | 2231 | goto err_out; |
b1a66504 | 2232 | } |
1da177e4 | 2233 | |
a8c1d064 AV |
2234 | err = misc_register(&loop_misc); |
2235 | if (err < 0) | |
2236 | goto err_out; | |
2237 | ||
2238 | ||
8410d38c | 2239 | if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { |
b1a66504 GC |
2240 | err = -EIO; |
2241 | goto misc_out; | |
2242 | } | |
1da177e4 | 2243 | |
d134b00b | 2244 | /* pre-create number of devices given by config or max_loop */ |
85c50197 | 2245 | for (i = 0; i < max_loop; i++) |
d6da83d0 | 2246 | loop_add(i); |
34dd82af | 2247 | |
73285082 | 2248 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2249 | return 0; |
b1a66504 GC |
2250 | |
2251 | misc_out: | |
2252 | misc_deregister(&loop_misc); | |
a8c1d064 | 2253 | err_out: |
b1a66504 | 2254 | return err; |
34dd82af | 2255 | } |
a47653fc | 2256 | |
73285082 | 2257 | static void __exit loop_exit(void) |
1da177e4 | 2258 | { |
8e60947d CH |
2259 | struct loop_device *lo; |
2260 | int id; | |
2261 | ||
00d59405 | 2262 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a | 2263 | misc_deregister(&loop_misc); |
200f9337 | 2264 | |
1c500ad7 TH |
2265 | /* |
2266 | * There is no need to use loop_ctl_mutex here, for nobody else can | |
2267 | * access loop_index_idr when this module is unloading (unless forced | |
2268 | * module unloading is requested). If this is not a clean unloading, | |
2269 | * we have no means to avoid kernel crash. | |
2270 | */ | |
8e60947d CH |
2271 | idr_for_each_entry(&loop_index_idr, lo, id) |
2272 | loop_remove(lo); | |
bd5c39ed CH |
2273 | |
2274 | idr_destroy(&loop_index_idr); | |
1da177e4 LT |
2275 | } |
2276 | ||
2277 | module_init(loop_init); | |
2278 | module_exit(loop_exit); | |
2279 | ||
2280 | #ifndef MODULE | |
2281 | static int __init max_loop_setup(char *str) | |
2282 | { | |
2283 | max_loop = simple_strtol(str, NULL, 0); | |
2284 | return 1; | |
2285 | } | |
2286 | ||
2287 | __setup("max_loop=", max_loop_setup); | |
2288 | #endif |