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 | ||
baa7d536 CH |
168 | /* |
169 | * We support direct I/O only if lo_offset is aligned with the logical I/O size | |
170 | * of backing device, and the logical block size of loop is bigger than that of | |
171 | * the backing device. | |
172 | */ | |
173 | static bool lo_bdev_can_use_dio(struct loop_device *lo, | |
174 | struct block_device *backing_bdev) | |
175 | { | |
176 | unsigned short sb_bsize = bdev_logical_block_size(backing_bdev); | |
177 | ||
178 | if (queue_logical_block_size(lo->lo_queue) < sb_bsize) | |
179 | return false; | |
180 | if (lo->lo_offset & (sb_bsize - 1)) | |
181 | return false; | |
182 | return true; | |
183 | } | |
184 | ||
2e5ab5f3 ML |
185 | static void __loop_update_dio(struct loop_device *lo, bool dio) |
186 | { | |
187 | struct file *file = lo->lo_backing_file; | |
baa7d536 CH |
188 | struct inode *inode = file->f_mapping->host; |
189 | struct block_device *backing_bdev = NULL; | |
2e5ab5f3 ML |
190 | bool use_dio; |
191 | ||
baa7d536 CH |
192 | if (S_ISBLK(inode->i_mode)) |
193 | backing_bdev = I_BDEV(inode); | |
194 | else if (inode->i_sb->s_bdev) | |
195 | backing_bdev = inode->i_sb->s_bdev; | |
2e5ab5f3 | 196 | |
baa7d536 CH |
197 | use_dio = dio && (file->f_mode & FMODE_CAN_ODIRECT) && |
198 | (!backing_bdev || lo_bdev_can_use_dio(lo, backing_bdev)); | |
2e5ab5f3 ML |
199 | |
200 | if (lo->use_dio == use_dio) | |
201 | return; | |
202 | ||
203 | /* flush dirty pages before changing direct IO */ | |
204 | vfs_fsync(file, 0); | |
205 | ||
206 | /* | |
207 | * The flag of LO_FLAGS_DIRECT_IO is handled similarly with | |
208 | * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup | |
209 | * will get updated by ioctl(LOOP_GET_STATUS) | |
210 | */ | |
0fbcf579 MC |
211 | if (lo->lo_state == Lo_bound) |
212 | blk_mq_freeze_queue(lo->lo_queue); | |
2e5ab5f3 | 213 | lo->use_dio = use_dio; |
667ea363 | 214 | if (use_dio) |
2e5ab5f3 | 215 | lo->lo_flags |= LO_FLAGS_DIRECT_IO; |
667ea363 | 216 | else |
2e5ab5f3 | 217 | lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; |
0fbcf579 MC |
218 | if (lo->lo_state == Lo_bound) |
219 | blk_mq_unfreeze_queue(lo->lo_queue); | |
2e5ab5f3 ML |
220 | } |
221 | ||
5795b6f5 MC |
222 | /** |
223 | * loop_set_size() - sets device size and notifies userspace | |
224 | * @lo: struct loop_device to set the size for | |
225 | * @size: new size of the loop device | |
226 | * | |
227 | * Callers must validate that the size passed into this function fits into | |
228 | * a sector_t, eg using loop_validate_size() | |
229 | */ | |
230 | static void loop_set_size(struct loop_device *lo, loff_t size) | |
231 | { | |
449f4ec9 | 232 | if (!set_capacity_and_notify(lo->lo_disk, size)) |
3b4f85d0 | 233 | kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); |
5795b6f5 MC |
234 | } |
235 | ||
aa4d8616 | 236 | static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) |
1da177e4 | 237 | { |
aa4d8616 | 238 | struct iov_iter i; |
1da177e4 | 239 | ssize_t bw; |
283e7e5d | 240 | |
de4eda9d | 241 | iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len); |
1da177e4 | 242 | |
abbb6589 | 243 | bw = vfs_iter_write(file, &i, ppos, 0); |
aa4d8616 CH |
244 | |
245 | if (likely(bw == bvec->bv_len)) | |
1da177e4 | 246 | return 0; |
aa4d8616 CH |
247 | |
248 | printk_ratelimited(KERN_ERR | |
249 | "loop: Write error at byte offset %llu, length %i.\n", | |
250 | (unsigned long long)*ppos, bvec->bv_len); | |
1da177e4 LT |
251 | if (bw >= 0) |
252 | bw = -EIO; | |
253 | return bw; | |
254 | } | |
255 | ||
aa4d8616 CH |
256 | static int lo_write_simple(struct loop_device *lo, struct request *rq, |
257 | loff_t pos) | |
1da177e4 | 258 | { |
aa4d8616 CH |
259 | struct bio_vec bvec; |
260 | struct req_iterator iter; | |
261 | int ret = 0; | |
262 | ||
263 | rq_for_each_segment(bvec, rq, iter) { | |
264 | ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); | |
265 | if (ret < 0) | |
266 | break; | |
267 | cond_resched(); | |
268 | } | |
269 | ||
270 | return ret; | |
1da177e4 LT |
271 | } |
272 | ||
aa4d8616 CH |
273 | static int lo_read_simple(struct loop_device *lo, struct request *rq, |
274 | loff_t pos) | |
1da177e4 | 275 | { |
aa4d8616 CH |
276 | struct bio_vec bvec; |
277 | struct req_iterator iter; | |
278 | struct iov_iter i; | |
279 | ssize_t len; | |
1da177e4 | 280 | |
aa4d8616 | 281 | rq_for_each_segment(bvec, rq, iter) { |
de4eda9d | 282 | iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len); |
18e9710e | 283 | len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); |
aa4d8616 CH |
284 | if (len < 0) |
285 | return len; | |
1da177e4 | 286 | |
aa4d8616 | 287 | flush_dcache_page(bvec.bv_page); |
fd582140 | 288 | |
aa4d8616 CH |
289 | if (len != bvec.bv_len) { |
290 | struct bio *bio; | |
1da177e4 | 291 | |
aa4d8616 CH |
292 | __rq_for_each_bio(bio, rq) |
293 | zero_fill_bio(bio); | |
294 | break; | |
295 | } | |
296 | cond_resched(); | |
297 | } | |
298 | ||
299 | return 0; | |
fd582140 JA |
300 | } |
301 | ||
5f75e081 CH |
302 | static void loop_clear_limits(struct loop_device *lo, int mode) |
303 | { | |
304 | struct queue_limits lim = queue_limits_start_update(lo->lo_queue); | |
305 | ||
306 | if (mode & FALLOC_FL_ZERO_RANGE) | |
307 | lim.max_write_zeroes_sectors = 0; | |
308 | ||
309 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
310 | lim.max_hw_discard_sectors = 0; | |
311 | lim.discard_granularity = 0; | |
312 | } | |
313 | ||
314 | queue_limits_commit_update(lo->lo_queue, &lim); | |
315 | } | |
316 | ||
efcfec57 DW |
317 | static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, |
318 | int mode) | |
cf655d95 ML |
319 | { |
320 | /* | |
efcfec57 | 321 | * We use fallocate to manipulate the space mappings used by the image |
47e96246 | 322 | * a.k.a. discard/zerorange. |
cf655d95 ML |
323 | */ |
324 | struct file *file = lo->lo_backing_file; | |
cf655d95 ML |
325 | int ret; |
326 | ||
efcfec57 DW |
327 | mode |= FALLOC_FL_KEEP_SIZE; |
328 | ||
70200574 CH |
329 | if (!bdev_max_discard_sectors(lo->lo_device)) |
330 | return -EOPNOTSUPP; | |
cf655d95 ML |
331 | |
332 | ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); | |
333 | if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) | |
70200574 | 334 | return -EIO; |
5f75e081 CH |
335 | |
336 | /* | |
337 | * We initially configure the limits in a hope that fallocate is | |
338 | * supported and clear them here if that turns out not to be true. | |
339 | */ | |
340 | if (unlikely(ret == -EOPNOTSUPP)) | |
341 | loop_clear_limits(lo, mode); | |
342 | ||
cf655d95 ML |
343 | return ret; |
344 | } | |
345 | ||
346 | static int lo_req_flush(struct loop_device *lo, struct request *rq) | |
347 | { | |
9c64e38c | 348 | int ret = vfs_fsync(lo->lo_backing_file, 0); |
cf655d95 ML |
349 | if (unlikely(ret && ret != -EINVAL)) |
350 | ret = -EIO; | |
351 | ||
352 | return ret; | |
353 | } | |
354 | ||
fe2cb290 | 355 | static void lo_complete_rq(struct request *rq) |
bc07c10a | 356 | { |
fe2cb290 | 357 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); |
f9de14bc | 358 | blk_status_t ret = BLK_STS_OK; |
bc07c10a | 359 | |
f9de14bc JA |
360 | if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || |
361 | req_op(rq) != REQ_OP_READ) { | |
362 | if (cmd->ret < 0) | |
8cd55087 | 363 | ret = errno_to_blk_status(cmd->ret); |
f9de14bc | 364 | goto end_io; |
bc07c10a | 365 | } |
fe2cb290 | 366 | |
f9de14bc JA |
367 | /* |
368 | * Short READ - if we got some data, advance our request and | |
369 | * retry it. If we got no data, end the rest with EIO. | |
370 | */ | |
371 | if (cmd->ret) { | |
372 | blk_update_request(rq, BLK_STS_OK, cmd->ret); | |
373 | cmd->ret = 0; | |
374 | blk_mq_requeue_request(rq, true); | |
375 | } else { | |
376 | if (cmd->use_aio) { | |
377 | struct bio *bio = rq->bio; | |
378 | ||
379 | while (bio) { | |
380 | zero_fill_bio(bio); | |
381 | bio = bio->bi_next; | |
382 | } | |
383 | } | |
384 | ret = BLK_STS_IOERR; | |
385 | end_io: | |
386 | blk_mq_end_request(rq, ret); | |
387 | } | |
bc07c10a ML |
388 | } |
389 | ||
92d77332 SL |
390 | static void lo_rw_aio_do_completion(struct loop_cmd *cmd) |
391 | { | |
1894e916 JA |
392 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
393 | ||
92d77332 SL |
394 | if (!atomic_dec_and_test(&cmd->ref)) |
395 | return; | |
396 | kfree(cmd->bvec); | |
397 | cmd->bvec = NULL; | |
15f73f5b CH |
398 | if (likely(!blk_should_fake_timeout(rq->q))) |
399 | blk_mq_complete_request(rq); | |
92d77332 SL |
400 | } |
401 | ||
6b19b766 | 402 | static void lo_rw_aio_complete(struct kiocb *iocb, long ret) |
bc07c10a ML |
403 | { |
404 | struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); | |
bc07c10a | 405 | |
fe2cb290 | 406 | cmd->ret = ret; |
92d77332 | 407 | lo_rw_aio_do_completion(cmd); |
bc07c10a ML |
408 | } |
409 | ||
410 | static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, | |
de4eda9d | 411 | loff_t pos, int rw) |
bc07c10a ML |
412 | { |
413 | struct iov_iter iter; | |
86af5952 | 414 | struct req_iterator rq_iter; |
bc07c10a | 415 | struct bio_vec *bvec; |
1894e916 | 416 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
40326d8a | 417 | struct bio *bio = rq->bio; |
bc07c10a | 418 | struct file *file = lo->lo_backing_file; |
86af5952 | 419 | struct bio_vec tmp; |
40326d8a | 420 | unsigned int offset; |
86af5952 | 421 | int nr_bvec = 0; |
bc07c10a ML |
422 | int ret; |
423 | ||
86af5952 ML |
424 | rq_for_each_bvec(tmp, rq, rq_iter) |
425 | nr_bvec++; | |
426 | ||
40326d8a | 427 | if (rq->bio != rq->biotail) { |
40326d8a | 428 | |
86af5952 | 429 | bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), |
6da2ec56 | 430 | GFP_NOIO); |
40326d8a SL |
431 | if (!bvec) |
432 | return -EIO; | |
433 | cmd->bvec = bvec; | |
434 | ||
435 | /* | |
436 | * The bios of the request may be started from the middle of | |
437 | * the 'bvec' because of bio splitting, so we can't directly | |
86af5952 | 438 | * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec |
40326d8a SL |
439 | * API will take care of all details for us. |
440 | */ | |
86af5952 | 441 | rq_for_each_bvec(tmp, rq, rq_iter) { |
40326d8a SL |
442 | *bvec = tmp; |
443 | bvec++; | |
444 | } | |
445 | bvec = cmd->bvec; | |
446 | offset = 0; | |
447 | } else { | |
448 | /* | |
449 | * Same here, this bio may be started from the middle of the | |
450 | * 'bvec' because of bio splitting, so offset from the bvec | |
451 | * must be passed to iov iterator | |
452 | */ | |
453 | offset = bio->bi_iter.bi_bvec_done; | |
454 | bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); | |
40326d8a | 455 | } |
92d77332 | 456 | atomic_set(&cmd->ref, 2); |
bc07c10a | 457 | |
b6207430 | 458 | iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); |
40326d8a | 459 | iter.iov_offset = offset; |
bc07c10a ML |
460 | |
461 | cmd->iocb.ki_pos = pos; | |
462 | cmd->iocb.ki_filp = file; | |
463 | cmd->iocb.ki_complete = lo_rw_aio_complete; | |
464 | cmd->iocb.ki_flags = IOCB_DIRECT; | |
d9a08a9e | 465 | cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); |
bc07c10a | 466 | |
de4eda9d | 467 | if (rw == ITER_SOURCE) |
7c98f7cb | 468 | ret = file->f_op->write_iter(&cmd->iocb, &iter); |
bc07c10a | 469 | else |
7c98f7cb | 470 | ret = file->f_op->read_iter(&cmd->iocb, &iter); |
bc07c10a | 471 | |
92d77332 SL |
472 | lo_rw_aio_do_completion(cmd); |
473 | ||
bc07c10a | 474 | if (ret != -EIOCBQUEUED) |
6b19b766 | 475 | lo_rw_aio_complete(&cmd->iocb, ret); |
bc07c10a ML |
476 | return 0; |
477 | } | |
478 | ||
c1c87c2b | 479 | static int do_req_filebacked(struct loop_device *lo, struct request *rq) |
bc07c10a ML |
480 | { |
481 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
c1c87c2b | 482 | loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; |
bc07c10a ML |
483 | |
484 | /* | |
485 | * lo_write_simple and lo_read_simple should have been covered | |
486 | * by io submit style function like lo_rw_aio(), one blocker | |
487 | * is that lo_read_simple() need to call flush_dcache_page after | |
488 | * the page is written from kernel, and it isn't easy to handle | |
489 | * this in io submit style function which submits all segments | |
490 | * of the req at one time. And direct read IO doesn't need to | |
491 | * run flush_dcache_page(). | |
492 | */ | |
c1c87c2b CH |
493 | switch (req_op(rq)) { |
494 | case REQ_OP_FLUSH: | |
495 | return lo_req_flush(lo, rq); | |
19372e27 | 496 | case REQ_OP_WRITE_ZEROES: |
efcfec57 DW |
497 | /* |
498 | * If the caller doesn't want deallocation, call zeroout to | |
499 | * write zeroes the range. Otherwise, punch them out. | |
500 | */ | |
501 | return lo_fallocate(lo, rq, pos, | |
502 | (rq->cmd_flags & REQ_NOUNMAP) ? | |
503 | FALLOC_FL_ZERO_RANGE : | |
504 | FALLOC_FL_PUNCH_HOLE); | |
505 | case REQ_OP_DISCARD: | |
506 | return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); | |
c1c87c2b | 507 | case REQ_OP_WRITE: |
47e96246 | 508 | if (cmd->use_aio) |
de4eda9d | 509 | return lo_rw_aio(lo, cmd, pos, ITER_SOURCE); |
af65aa8e | 510 | else |
c1c87c2b CH |
511 | return lo_write_simple(lo, rq, pos); |
512 | case REQ_OP_READ: | |
47e96246 | 513 | if (cmd->use_aio) |
de4eda9d | 514 | return lo_rw_aio(lo, cmd, pos, ITER_DEST); |
aa4d8616 | 515 | else |
c1c87c2b CH |
516 | return lo_read_simple(lo, rq, pos); |
517 | default: | |
518 | WARN_ON_ONCE(1); | |
519 | return -EIO; | |
aa4d8616 | 520 | } |
1da177e4 LT |
521 | } |
522 | ||
2e5ab5f3 ML |
523 | static inline void loop_update_dio(struct loop_device *lo) |
524 | { | |
efbe3c24 IW |
525 | __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | |
526 | lo->use_dio); | |
2e5ab5f3 ML |
527 | } |
528 | ||
0384264e | 529 | static void loop_reread_partitions(struct loop_device *lo) |
06f0e9e6 ML |
530 | { |
531 | int rc; | |
532 | ||
0384264e CH |
533 | mutex_lock(&lo->lo_disk->open_mutex); |
534 | rc = bdev_disk_changed(lo->lo_disk, false); | |
535 | mutex_unlock(&lo->lo_disk->open_mutex); | |
06f0e9e6 ML |
536 | if (rc) |
537 | pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", | |
538 | __func__, lo->lo_number, lo->lo_file_name, rc); | |
539 | } | |
540 | ||
d2ac838e TT |
541 | static inline int is_loop_device(struct file *file) |
542 | { | |
543 | struct inode *i = file->f_mapping->host; | |
544 | ||
6f24784f | 545 | return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; |
d2ac838e TT |
546 | } |
547 | ||
548 | static int loop_validate_file(struct file *file, struct block_device *bdev) | |
549 | { | |
550 | struct inode *inode = file->f_mapping->host; | |
551 | struct file *f = file; | |
552 | ||
553 | /* Avoid recursion */ | |
554 | while (is_loop_device(f)) { | |
555 | struct loop_device *l; | |
556 | ||
3ce6e1f6 | 557 | lockdep_assert_held(&loop_validate_mutex); |
4e7b5671 | 558 | if (f->f_mapping->host->i_rdev == bdev->bd_dev) |
d2ac838e TT |
559 | return -EBADF; |
560 | ||
4e7b5671 | 561 | l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; |
3ce6e1f6 | 562 | if (l->lo_state != Lo_bound) |
d2ac838e | 563 | return -EINVAL; |
3ce6e1f6 TH |
564 | /* Order wrt setting lo->lo_backing_file in loop_configure(). */ |
565 | rmb(); | |
d2ac838e TT |
566 | f = l->lo_backing_file; |
567 | } | |
568 | if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) | |
569 | return -EINVAL; | |
570 | return 0; | |
571 | } | |
572 | ||
1da177e4 LT |
573 | /* |
574 | * loop_change_fd switched the backing store of a loopback device to | |
575 | * a new file. This is useful for operating system installers to free up | |
576 | * the original file and in High Availability environments to switch to | |
577 | * an alternative location for the content in case of server meltdown. | |
578 | * This can only work if the loop device is used read-only, and if the | |
579 | * new backing store is the same size and type as the old backing store. | |
580 | */ | |
bb214884 AV |
581 | static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, |
582 | unsigned int arg) | |
1da177e4 | 583 | { |
3ce6e1f6 TH |
584 | struct file *file = fget(arg); |
585 | struct file *old_file; | |
586 | int error; | |
587 | bool partscan; | |
588 | bool is_loop; | |
1da177e4 | 589 | |
3ce6e1f6 TH |
590 | if (!file) |
591 | return -EBADF; | |
498ef5c7 CH |
592 | |
593 | /* suppress uevents while reconfiguring the device */ | |
594 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); | |
595 | ||
3ce6e1f6 TH |
596 | is_loop = is_loop_device(file); |
597 | error = loop_global_lock_killable(lo, is_loop); | |
c3710770 | 598 | if (error) |
3ce6e1f6 | 599 | goto out_putf; |
1da177e4 LT |
600 | error = -ENXIO; |
601 | if (lo->lo_state != Lo_bound) | |
1dded9ac | 602 | goto out_err; |
1da177e4 LT |
603 | |
604 | /* the loop device has to be read-only */ | |
605 | error = -EINVAL; | |
606 | if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) | |
1dded9ac | 607 | goto out_err; |
1da177e4 | 608 | |
d2ac838e TT |
609 | error = loop_validate_file(file, bdev); |
610 | if (error) | |
1dded9ac | 611 | goto out_err; |
d2ac838e | 612 | |
1da177e4 LT |
613 | old_file = lo->lo_backing_file; |
614 | ||
615 | error = -EINVAL; | |
616 | ||
1da177e4 LT |
617 | /* size of the new backing store needs to be the same */ |
618 | if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) | |
1dded9ac | 619 | goto out_err; |
1da177e4 LT |
620 | |
621 | /* and ... switch */ | |
ab6860f6 | 622 | disk_force_media_change(lo->lo_disk); |
43cade80 OS |
623 | blk_mq_freeze_queue(lo->lo_queue); |
624 | mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); | |
625 | lo->lo_backing_file = file; | |
626 | lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); | |
627 | mapping_set_gfp_mask(file->f_mapping, | |
628 | lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
629 | loop_update_dio(lo); | |
630 | blk_mq_unfreeze_queue(lo->lo_queue); | |
85b0a54a | 631 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
3ce6e1f6 TH |
632 | loop_global_unlock(lo, is_loop); |
633 | ||
634 | /* | |
635 | * Flush loop_validate_file() before fput(), for l->lo_backing_file | |
636 | * might be pointing at old_file which might be the last reference. | |
637 | */ | |
638 | if (!is_loop) { | |
639 | mutex_lock(&loop_validate_mutex); | |
640 | mutex_unlock(&loop_validate_mutex); | |
641 | } | |
1dded9ac | 642 | /* |
6cc8e743 | 643 | * We must drop file reference outside of lo_mutex as dropping |
a8698707 | 644 | * the file ref can take open_mutex which creates circular locking |
1dded9ac JK |
645 | * dependency. |
646 | */ | |
647 | fput(old_file); | |
85b0a54a | 648 | if (partscan) |
0384264e | 649 | loop_reread_partitions(lo); |
498ef5c7 CH |
650 | |
651 | error = 0; | |
652 | done: | |
653 | /* enable and uncork uevent now that we are done */ | |
654 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); | |
655 | return error; | |
1da177e4 | 656 | |
1dded9ac | 657 | out_err: |
3ce6e1f6 TH |
658 | loop_global_unlock(lo, is_loop); |
659 | out_putf: | |
660 | fput(file); | |
498ef5c7 | 661 | goto done; |
1da177e4 LT |
662 | } |
663 | ||
ee862730 MB |
664 | /* loop sysfs attributes */ |
665 | ||
666 | static ssize_t loop_attr_show(struct device *dev, char *page, | |
667 | ssize_t (*callback)(struct loop_device *, char *)) | |
668 | { | |
34dd82af KS |
669 | struct gendisk *disk = dev_to_disk(dev); |
670 | struct loop_device *lo = disk->private_data; | |
ee862730 | 671 | |
34dd82af | 672 | return callback(lo, page); |
ee862730 MB |
673 | } |
674 | ||
675 | #define LOOP_ATTR_RO(_name) \ | |
676 | static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ | |
677 | static ssize_t loop_attr_do_show_##_name(struct device *d, \ | |
678 | struct device_attribute *attr, char *b) \ | |
679 | { \ | |
680 | return loop_attr_show(d, b, loop_attr_##_name##_show); \ | |
681 | } \ | |
682 | static struct device_attribute loop_attr_##_name = \ | |
5657a819 | 683 | __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); |
ee862730 MB |
684 | |
685 | static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) | |
686 | { | |
687 | ssize_t ret; | |
688 | char *p = NULL; | |
689 | ||
05eb0f25 | 690 | spin_lock_irq(&lo->lo_lock); |
ee862730 | 691 | if (lo->lo_backing_file) |
9bf39ab2 | 692 | p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); |
05eb0f25 | 693 | spin_unlock_irq(&lo->lo_lock); |
ee862730 MB |
694 | |
695 | if (IS_ERR_OR_NULL(p)) | |
696 | ret = PTR_ERR(p); | |
697 | else { | |
698 | ret = strlen(p); | |
699 | memmove(buf, p, ret); | |
700 | buf[ret++] = '\n'; | |
701 | buf[ret] = 0; | |
702 | } | |
703 | ||
704 | return ret; | |
705 | } | |
706 | ||
707 | static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) | |
708 | { | |
b27824d3 | 709 | return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset); |
ee862730 MB |
710 | } |
711 | ||
712 | static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) | |
713 | { | |
b27824d3 | 714 | return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); |
ee862730 MB |
715 | } |
716 | ||
717 | static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) | |
718 | { | |
719 | int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); | |
720 | ||
b27824d3 | 721 | return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0"); |
ee862730 MB |
722 | } |
723 | ||
e03c8dd1 KS |
724 | static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) |
725 | { | |
726 | int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); | |
727 | ||
b27824d3 | 728 | return sysfs_emit(buf, "%s\n", partscan ? "1" : "0"); |
e03c8dd1 KS |
729 | } |
730 | ||
2e5ab5f3 ML |
731 | static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) |
732 | { | |
733 | int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); | |
734 | ||
b27824d3 | 735 | return sysfs_emit(buf, "%s\n", dio ? "1" : "0"); |
2e5ab5f3 ML |
736 | } |
737 | ||
ee862730 MB |
738 | LOOP_ATTR_RO(backing_file); |
739 | LOOP_ATTR_RO(offset); | |
740 | LOOP_ATTR_RO(sizelimit); | |
741 | LOOP_ATTR_RO(autoclear); | |
e03c8dd1 | 742 | LOOP_ATTR_RO(partscan); |
2e5ab5f3 | 743 | LOOP_ATTR_RO(dio); |
ee862730 MB |
744 | |
745 | static struct attribute *loop_attrs[] = { | |
746 | &loop_attr_backing_file.attr, | |
747 | &loop_attr_offset.attr, | |
748 | &loop_attr_sizelimit.attr, | |
749 | &loop_attr_autoclear.attr, | |
e03c8dd1 | 750 | &loop_attr_partscan.attr, |
2e5ab5f3 | 751 | &loop_attr_dio.attr, |
ee862730 MB |
752 | NULL, |
753 | }; | |
754 | ||
755 | static struct attribute_group loop_attribute_group = { | |
756 | .name = "loop", | |
757 | .attrs= loop_attrs, | |
758 | }; | |
759 | ||
d3349b6b | 760 | static void loop_sysfs_init(struct loop_device *lo) |
ee862730 | 761 | { |
d3349b6b TH |
762 | lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, |
763 | &loop_attribute_group); | |
ee862730 MB |
764 | } |
765 | ||
766 | static void loop_sysfs_exit(struct loop_device *lo) | |
767 | { | |
d3349b6b TH |
768 | if (lo->sysfs_inited) |
769 | sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, | |
770 | &loop_attribute_group); | |
ee862730 MB |
771 | } |
772 | ||
473516b3 CH |
773 | static void loop_config_discard(struct loop_device *lo, |
774 | struct queue_limits *lim) | |
dfaa2ef6 LC |
775 | { |
776 | struct file *file = lo->lo_backing_file; | |
777 | struct inode *inode = file->f_mapping->host; | |
65bdd16f CH |
778 | u32 granularity = 0, max_discard_sectors = 0; |
779 | struct kstatfs sbuf; | |
dfaa2ef6 | 780 | |
c52abf56 EG |
781 | /* |
782 | * If the backing device is a block device, mirror its zeroing | |
783 | * capability. Set the discard sectors to the block device's zeroing | |
784 | * capabilities because loop discards result in blkdev_issue_zeroout(), | |
785 | * not blkdev_issue_discard(). This maintains consistent behavior with | |
786 | * file-backed loop devices: discarded regions read back as zero. | |
787 | */ | |
47e96246 | 788 | if (S_ISBLK(inode->i_mode)) { |
4e7b5671 | 789 | struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); |
c52abf56 | 790 | |
bcb21c8c | 791 | max_discard_sectors = backingq->limits.max_write_zeroes_sectors; |
7b47ef52 | 792 | granularity = bdev_discard_granularity(I_BDEV(inode)) ?: |
bcb21c8c | 793 | queue_physical_block_size(backingq); |
c52abf56 | 794 | |
dfaa2ef6 LC |
795 | /* |
796 | * We use punch hole to reclaim the free space used by the | |
47e96246 | 797 | * image a.k.a. discard. |
dfaa2ef6 | 798 | */ |
65bdd16f | 799 | } else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) { |
bcb21c8c | 800 | max_discard_sectors = UINT_MAX >> 9; |
65bdd16f | 801 | granularity = sbuf.f_bsize; |
c52abf56 EG |
802 | } |
803 | ||
473516b3 CH |
804 | lim->max_hw_discard_sectors = max_discard_sectors; |
805 | lim->max_write_zeroes_sectors = max_discard_sectors; | |
65bdd16f | 806 | if (max_discard_sectors) |
473516b3 | 807 | lim->discard_granularity = granularity; |
65bdd16f | 808 | else |
473516b3 | 809 | lim->discard_granularity = 0; |
dfaa2ef6 LC |
810 | } |
811 | ||
87579e9b DS |
812 | struct loop_worker { |
813 | struct rb_node rb_node; | |
814 | struct work_struct work; | |
815 | struct list_head cmd_list; | |
816 | struct list_head idle_list; | |
817 | struct loop_device *lo; | |
c74d40e8 | 818 | struct cgroup_subsys_state *blkcg_css; |
87579e9b DS |
819 | unsigned long last_ran_at; |
820 | }; | |
821 | ||
822 | static void loop_workfn(struct work_struct *work); | |
87579e9b DS |
823 | |
824 | #ifdef CONFIG_BLK_CGROUP | |
825 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
e03a3d7a | 826 | { |
87579e9b | 827 | return !css || css == blkcg_root_css; |
e03a3d7a | 828 | } |
87579e9b DS |
829 | #else |
830 | static inline int queue_on_root_worker(struct cgroup_subsys_state *css) | |
b2ee7d46 | 831 | { |
87579e9b | 832 | return !css; |
b2ee7d46 | 833 | } |
87579e9b | 834 | #endif |
b2ee7d46 | 835 | |
87579e9b | 836 | static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) |
e03a3d7a | 837 | { |
413ec805 | 838 | struct rb_node **node, *parent = NULL; |
87579e9b DS |
839 | struct loop_worker *cur_worker, *worker = NULL; |
840 | struct work_struct *work; | |
841 | struct list_head *cmd_list; | |
842 | ||
843 | spin_lock_irq(&lo->lo_work_lock); | |
844 | ||
c74d40e8 | 845 | if (queue_on_root_worker(cmd->blkcg_css)) |
87579e9b DS |
846 | goto queue_work; |
847 | ||
848 | node = &lo->worker_tree.rb_node; | |
849 | ||
850 | while (*node) { | |
851 | parent = *node; | |
852 | cur_worker = container_of(*node, struct loop_worker, rb_node); | |
c74d40e8 | 853 | if (cur_worker->blkcg_css == cmd->blkcg_css) { |
87579e9b DS |
854 | worker = cur_worker; |
855 | break; | |
c74d40e8 | 856 | } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { |
87579e9b DS |
857 | node = &(*node)->rb_left; |
858 | } else { | |
859 | node = &(*node)->rb_right; | |
860 | } | |
861 | } | |
862 | if (worker) | |
863 | goto queue_work; | |
864 | ||
865 | worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); | |
866 | /* | |
867 | * In the event we cannot allocate a worker, just queue on the | |
c74d40e8 | 868 | * rootcg worker and issue the I/O as the rootcg |
87579e9b | 869 | */ |
c74d40e8 DS |
870 | if (!worker) { |
871 | cmd->blkcg_css = NULL; | |
872 | if (cmd->memcg_css) | |
873 | css_put(cmd->memcg_css); | |
874 | cmd->memcg_css = NULL; | |
87579e9b | 875 | goto queue_work; |
c74d40e8 | 876 | } |
87579e9b | 877 | |
c74d40e8 DS |
878 | worker->blkcg_css = cmd->blkcg_css; |
879 | css_get(worker->blkcg_css); | |
87579e9b DS |
880 | INIT_WORK(&worker->work, loop_workfn); |
881 | INIT_LIST_HEAD(&worker->cmd_list); | |
882 | INIT_LIST_HEAD(&worker->idle_list); | |
883 | worker->lo = lo; | |
884 | rb_link_node(&worker->rb_node, parent, node); | |
885 | rb_insert_color(&worker->rb_node, &lo->worker_tree); | |
886 | queue_work: | |
887 | if (worker) { | |
888 | /* | |
889 | * We need to remove from the idle list here while | |
890 | * holding the lock so that the idle timer doesn't | |
891 | * free the worker | |
892 | */ | |
893 | if (!list_empty(&worker->idle_list)) | |
894 | list_del_init(&worker->idle_list); | |
895 | work = &worker->work; | |
896 | cmd_list = &worker->cmd_list; | |
897 | } else { | |
898 | work = &lo->rootcg_work; | |
899 | cmd_list = &lo->rootcg_cmd_list; | |
900 | } | |
901 | list_add_tail(&cmd->list_entry, cmd_list); | |
902 | queue_work(lo->workqueue, work); | |
903 | spin_unlock_irq(&lo->lo_work_lock); | |
e03a3d7a ML |
904 | } |
905 | ||
2cf429b5 CH |
906 | static void loop_set_timer(struct loop_device *lo) |
907 | { | |
908 | timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); | |
909 | } | |
910 | ||
911 | static void loop_free_idle_workers(struct loop_device *lo, bool delete_all) | |
912 | { | |
913 | struct loop_worker *pos, *worker; | |
914 | ||
915 | spin_lock_irq(&lo->lo_work_lock); | |
916 | list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, | |
917 | idle_list) { | |
918 | if (!delete_all && | |
919 | time_is_after_jiffies(worker->last_ran_at + | |
920 | LOOP_IDLE_WORKER_TIMEOUT)) | |
921 | break; | |
922 | list_del(&worker->idle_list); | |
923 | rb_erase(&worker->rb_node, &lo->worker_tree); | |
924 | css_put(worker->blkcg_css); | |
925 | kfree(worker); | |
926 | } | |
927 | if (!list_empty(&lo->idle_worker_list)) | |
928 | loop_set_timer(lo); | |
929 | spin_unlock_irq(&lo->lo_work_lock); | |
930 | } | |
931 | ||
932 | static void loop_free_idle_workers_timer(struct timer_list *timer) | |
933 | { | |
934 | struct loop_device *lo = container_of(timer, struct loop_device, timer); | |
935 | ||
936 | return loop_free_idle_workers(lo, false); | |
937 | } | |
938 | ||
62ab466c MC |
939 | /** |
940 | * loop_set_status_from_info - configure device from loop_info | |
941 | * @lo: struct loop_device to configure | |
942 | * @info: struct loop_info64 to configure the device with | |
943 | * | |
944 | * Configures the loop device parameters according to the passed | |
945 | * in loop_info64 configuration. | |
946 | */ | |
947 | static int | |
948 | loop_set_status_from_info(struct loop_device *lo, | |
949 | const struct loop_info64 *info) | |
950 | { | |
62ab466c MC |
951 | if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) |
952 | return -EINVAL; | |
953 | ||
47e96246 CH |
954 | switch (info->lo_encrypt_type) { |
955 | case LO_CRYPT_NONE: | |
956 | break; | |
957 | case LO_CRYPT_XOR: | |
958 | pr_warn("support for the xor transformation has been removed.\n"); | |
959 | return -EINVAL; | |
960 | case LO_CRYPT_CRYPTOAPI: | |
961 | pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); | |
962 | return -EINVAL; | |
963 | default: | |
964 | return -EINVAL; | |
965 | } | |
62ab466c | 966 | |
9f6ad5d5 ZJ |
967 | /* Avoid assigning overflow values */ |
968 | if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX) | |
969 | return -EOVERFLOW; | |
970 | ||
62ab466c MC |
971 | lo->lo_offset = info->lo_offset; |
972 | lo->lo_sizelimit = info->lo_sizelimit; | |
c490a0b5 | 973 | |
62ab466c | 974 | memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); |
62ab466c | 975 | lo->lo_file_name[LO_NAME_SIZE-1] = 0; |
faf1d254 | 976 | lo->lo_flags = info->lo_flags; |
62ab466c MC |
977 | return 0; |
978 | } | |
979 | ||
a17ece76 CH |
980 | static unsigned short loop_default_blocksize(struct loop_device *lo, |
981 | struct block_device *backing_bdev) | |
473516b3 | 982 | { |
a17ece76 CH |
983 | /* In case of direct I/O, match underlying block size */ |
984 | if ((lo->lo_backing_file->f_flags & O_DIRECT) && backing_bdev) | |
985 | return bdev_logical_block_size(backing_bdev); | |
986 | return SECTOR_SIZE; | |
987 | } | |
988 | ||
ae0d40ff | 989 | static int loop_reconfigure_limits(struct loop_device *lo, unsigned short bsize) |
473516b3 | 990 | { |
a17ece76 CH |
991 | struct file *file = lo->lo_backing_file; |
992 | struct inode *inode = file->f_mapping->host; | |
4ce37fe0 | 993 | struct block_device *backing_bdev = NULL; |
473516b3 CH |
994 | struct queue_limits lim; |
995 | ||
4ce37fe0 CH |
996 | if (S_ISBLK(inode->i_mode)) |
997 | backing_bdev = I_BDEV(inode); | |
998 | else if (inode->i_sb->s_bdev) | |
999 | backing_bdev = inode->i_sb->s_bdev; | |
1000 | ||
a17ece76 | 1001 | if (!bsize) |
4ce37fe0 | 1002 | bsize = loop_default_blocksize(lo, backing_bdev); |
a17ece76 | 1003 | |
473516b3 CH |
1004 | lim = queue_limits_start_update(lo->lo_queue); |
1005 | lim.logical_block_size = bsize; | |
1006 | lim.physical_block_size = bsize; | |
1007 | lim.io_min = bsize; | |
bd4a633b | 1008 | lim.features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL); |
1122c0c1 CH |
1009 | if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY)) |
1010 | lim.features |= BLK_FEAT_WRITE_CACHE; | |
bd4a633b CH |
1011 | if (backing_bdev && !bdev_nonrot(backing_bdev)) |
1012 | lim.features |= BLK_FEAT_ROTATIONAL; | |
ae0d40ff | 1013 | loop_config_discard(lo, &lim); |
473516b3 CH |
1014 | return queue_limits_commit_update(lo->lo_queue, &lim); |
1015 | } | |
1016 | ||
05bdb996 | 1017 | static int loop_configure(struct loop_device *lo, blk_mode_t mode, |
3448914e MC |
1018 | struct block_device *bdev, |
1019 | const struct loop_config *config) | |
1da177e4 | 1020 | { |
3ce6e1f6 | 1021 | struct file *file = fget(config->fd); |
1da177e4 | 1022 | struct address_space *mapping; |
3ce6e1f6 TH |
1023 | int error; |
1024 | loff_t size; | |
1025 | bool partscan; | |
3ce6e1f6 TH |
1026 | bool is_loop; |
1027 | ||
1028 | if (!file) | |
1029 | return -EBADF; | |
1030 | is_loop = is_loop_device(file); | |
1da177e4 LT |
1031 | |
1032 | /* This is safe, since we have a reference from open(). */ | |
1033 | __module_get(THIS_MODULE); | |
1034 | ||
33ec3e53 JK |
1035 | /* |
1036 | * If we don't hold exclusive handle for the device, upgrade to it | |
1037 | * here to avoid changing device under exclusive owner. | |
1038 | */ | |
05bdb996 | 1039 | if (!(mode & BLK_OPEN_EXCL)) { |
0718afd4 | 1040 | error = bd_prepare_to_claim(bdev, loop_configure, NULL); |
ecbe6bc0 | 1041 | if (error) |
33ec3e53 JK |
1042 | goto out_putf; |
1043 | } | |
1044 | ||
3ce6e1f6 | 1045 | error = loop_global_lock_killable(lo, is_loop); |
757ecf40 | 1046 | if (error) |
33ec3e53 | 1047 | goto out_bdev; |
757ecf40 | 1048 | |
1da177e4 LT |
1049 | error = -EBUSY; |
1050 | if (lo->lo_state != Lo_unbound) | |
757ecf40 | 1051 | goto out_unlock; |
1da177e4 | 1052 | |
d2ac838e TT |
1053 | error = loop_validate_file(file, bdev); |
1054 | if (error) | |
757ecf40 | 1055 | goto out_unlock; |
1da177e4 LT |
1056 | |
1057 | mapping = file->f_mapping; | |
1da177e4 | 1058 | |
3448914e MC |
1059 | if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { |
1060 | error = -EINVAL; | |
1061 | goto out_unlock; | |
1062 | } | |
1063 | ||
3448914e MC |
1064 | error = loop_set_status_from_info(lo, &config->info); |
1065 | if (error) | |
1066 | goto out_unlock; | |
1067 | ||
05bdb996 | 1068 | if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) || |
283e7e5d | 1069 | !file->f_op->write_iter) |
3448914e | 1070 | lo->lo_flags |= LO_FLAGS_READ_ONLY; |
ba52de12 | 1071 | |
87579e9b | 1072 | if (!lo->workqueue) { |
d292dc80 CH |
1073 | lo->workqueue = alloc_workqueue("loop%d", |
1074 | WQ_UNBOUND | WQ_FREEZABLE, | |
1075 | 0, lo->lo_number); | |
1076 | if (!lo->workqueue) { | |
1077 | error = -ENOMEM; | |
1078 | goto out_unlock; | |
1079 | } | |
87579e9b | 1080 | } |
1da177e4 | 1081 | |
bb430b69 AR |
1082 | /* suppress uevents while reconfiguring the device */ |
1083 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1); | |
1084 | ||
ab6860f6 | 1085 | disk_force_media_change(lo->lo_disk); |
7a2f0ce1 | 1086 | set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); |
1da177e4 | 1087 | |
3448914e | 1088 | lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; |
1da177e4 | 1089 | lo->lo_device = bdev; |
1da177e4 | 1090 | lo->lo_backing_file = file; |
1da177e4 LT |
1091 | lo->old_gfp_mask = mapping_gfp_mask(mapping); |
1092 | mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); | |
1093 | ||
a17ece76 | 1094 | error = loop_reconfigure_limits(lo, config->block_size); |
9423c653 | 1095 | if (error) |
473516b3 | 1096 | goto out_unlock; |
85560117 | 1097 | |
2e5ab5f3 | 1098 | loop_update_dio(lo); |
ee862730 | 1099 | loop_sysfs_init(lo); |
79e5dc59 MC |
1100 | |
1101 | size = get_loop_size(lo, file); | |
5795b6f5 | 1102 | loop_set_size(lo, size); |
1da177e4 | 1103 | |
3ce6e1f6 TH |
1104 | /* Order wrt reading lo_state in loop_validate_file(). */ |
1105 | wmb(); | |
1106 | ||
6c997918 | 1107 | lo->lo_state = Lo_bound; |
e03c8dd1 KS |
1108 | if (part_shift) |
1109 | lo->lo_flags |= LO_FLAGS_PARTSCAN; | |
85b0a54a | 1110 | partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; |
fe6a8fc5 | 1111 | if (partscan) |
b9684a71 | 1112 | clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
c1681bf8 | 1113 | |
bb430b69 AR |
1114 | /* enable and uncork uevent now that we are done */ |
1115 | dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0); | |
1116 | ||
3ce6e1f6 | 1117 | loop_global_unlock(lo, is_loop); |
85b0a54a | 1118 | if (partscan) |
0384264e | 1119 | loop_reread_partitions(lo); |
bb430b69 | 1120 | |
05bdb996 | 1121 | if (!(mode & BLK_OPEN_EXCL)) |
37c3fc9a | 1122 | bd_abort_claiming(bdev, loop_configure); |
498ef5c7 | 1123 | |
bb430b69 | 1124 | return 0; |
1da177e4 | 1125 | |
757ecf40 | 1126 | out_unlock: |
3ce6e1f6 | 1127 | loop_global_unlock(lo, is_loop); |
33ec3e53 | 1128 | out_bdev: |
05bdb996 | 1129 | if (!(mode & BLK_OPEN_EXCL)) |
37c3fc9a | 1130 | bd_abort_claiming(bdev, loop_configure); |
757ecf40 | 1131 | out_putf: |
1da177e4 | 1132 | fput(file); |
1da177e4 LT |
1133 | /* This is safe: open() is still holding a reference. */ |
1134 | module_put(THIS_MODULE); | |
bb430b69 | 1135 | return error; |
1da177e4 LT |
1136 | } |
1137 | ||
18048c1a | 1138 | static void __loop_clr_fd(struct loop_device *lo) |
1da177e4 | 1139 | { |
c9055b44 | 1140 | struct queue_limits lim; |
6050fa4c | 1141 | struct file *filp; |
b4e3ca1a | 1142 | gfp_t gfp = lo->old_gfp_mask; |
1da177e4 LT |
1143 | |
1144 | spin_lock_irq(&lo->lo_lock); | |
6050fa4c | 1145 | filp = lo->lo_backing_file; |
1da177e4 | 1146 | lo->lo_backing_file = NULL; |
05eb0f25 | 1147 | spin_unlock_irq(&lo->lo_lock); |
1da177e4 | 1148 | |
1da177e4 | 1149 | lo->lo_device = NULL; |
1da177e4 LT |
1150 | lo->lo_offset = 0; |
1151 | lo->lo_sizelimit = 0; | |
1da177e4 | 1152 | memset(lo->lo_file_name, 0, LO_NAME_SIZE); |
c9055b44 CH |
1153 | |
1154 | /* reset the block size to the default */ | |
1155 | lim = queue_limits_start_update(lo->lo_queue); | |
1156 | lim.logical_block_size = SECTOR_SIZE; | |
1157 | lim.physical_block_size = SECTOR_SIZE; | |
1158 | lim.io_min = SECTOR_SIZE; | |
1159 | queue_limits_commit_update(lo->lo_queue, &lim); | |
1160 | ||
e515be8f | 1161 | invalidate_disk(lo->lo_disk); |
51a0bb0c | 1162 | loop_sysfs_exit(lo); |
19f553db XY |
1163 | /* let user-space know about this change */ |
1164 | kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); | |
1da177e4 | 1165 | mapping_set_gfp_mask(filp->f_mapping, gfp); |
bf23747e TH |
1166 | /* This is safe: open() is still holding a reference. */ |
1167 | module_put(THIS_MODULE); | |
f8933667 | 1168 | |
ab6860f6 | 1169 | disk_force_media_change(lo->lo_disk); |
6050fa4c TH |
1170 | |
1171 | if (lo->lo_flags & LO_FLAGS_PARTSCAN) { | |
1172 | int err; | |
1173 | ||
bf23747e TH |
1174 | /* |
1175 | * open_mutex has been held already in release path, so don't | |
1176 | * acquire it if this function is called in such case. | |
1177 | * | |
1178 | * If the reread partition isn't from release path, lo_refcnt | |
1179 | * must be at least one and it can only become zero when the | |
1180 | * current holder is released. | |
1181 | */ | |
0384264e | 1182 | err = bdev_disk_changed(lo->lo_disk, false); |
40853d6f DZ |
1183 | if (err) |
1184 | pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", | |
6050fa4c | 1185 | __func__, lo->lo_number, err); |
d57f3374 | 1186 | /* Device is gone, no point in returning error */ |
d57f3374 | 1187 | } |
758a58d0 | 1188 | |
bf23747e TH |
1189 | /* |
1190 | * lo->lo_state is set to Lo_unbound here after above partscan has | |
1191 | * finished. There cannot be anybody else entering __loop_clr_fd() as | |
1192 | * Lo_rundown state protects us from all the other places trying to | |
1193 | * change the 'lo' device. | |
1194 | */ | |
758a58d0 DZ |
1195 | lo->lo_flags = 0; |
1196 | if (!part_shift) | |
b9684a71 | 1197 | set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
6050fa4c | 1198 | mutex_lock(&lo->lo_mutex); |
758a58d0 | 1199 | lo->lo_state = Lo_unbound; |
6cc8e743 | 1200 | mutex_unlock(&lo->lo_mutex); |
322c4293 | 1201 | |
bf23747e TH |
1202 | /* |
1203 | * Need not hold lo_mutex to fput backing file. Calling fput holding | |
1204 | * lo_mutex triggers a circular lock dependency possibility warning as | |
1205 | * fput can take open_mutex which is usually taken before lo_mutex. | |
1206 | */ | |
1207 | fput(filp); | |
1da177e4 LT |
1208 | } |
1209 | ||
a2505b79 JK |
1210 | static int loop_clr_fd(struct loop_device *lo) |
1211 | { | |
7ccd0791 JK |
1212 | int err; |
1213 | ||
158eaeba TH |
1214 | /* |
1215 | * Since lo_ioctl() is called without locks held, it is possible that | |
1216 | * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel. | |
1217 | * | |
1218 | * Therefore, use global lock when setting Lo_rundown state in order to | |
1219 | * make sure that loop_validate_file() will fail if the "struct file" | |
1220 | * which loop_configure()/loop_change_fd() found via fget() was this | |
1221 | * loop device. | |
1222 | */ | |
1223 | err = loop_global_lock_killable(lo, true); | |
7ccd0791 JK |
1224 | if (err) |
1225 | return err; | |
1226 | if (lo->lo_state != Lo_bound) { | |
158eaeba | 1227 | loop_global_unlock(lo, true); |
a2505b79 | 1228 | return -ENXIO; |
7ccd0791 | 1229 | } |
a2505b79 | 1230 | /* |
18048c1a GM |
1231 | * Mark the device for removing the backing device on last close. |
1232 | * If we are the only opener, also switch the state to roundown here to | |
1233 | * prevent new openers from coming in. | |
a2505b79 | 1234 | */ |
18048c1a GM |
1235 | |
1236 | lo->lo_flags |= LO_FLAGS_AUTOCLEAR; | |
1237 | if (disk_openers(lo->lo_disk) == 1) | |
1238 | lo->lo_state = Lo_rundown; | |
158eaeba | 1239 | loop_global_unlock(lo, true); |
a2505b79 | 1240 | |
6050fa4c | 1241 | return 0; |
a2505b79 JK |
1242 | } |
1243 | ||
1da177e4 LT |
1244 | static int |
1245 | loop_set_status(struct loop_device *lo, const struct loop_info64 *info) | |
1246 | { | |
1247 | int err; | |
faf1d254 | 1248 | int prev_lo_flags; |
85b0a54a | 1249 | bool partscan = false; |
0c3796c2 | 1250 | bool size_changed = false; |
1da177e4 | 1251 | |
6cc8e743 | 1252 | err = mutex_lock_killable(&lo->lo_mutex); |
550df5fd JK |
1253 | if (err) |
1254 | return err; | |
550df5fd JK |
1255 | if (lo->lo_state != Lo_bound) { |
1256 | err = -ENXIO; | |
1257 | goto out_unlock; | |
1258 | } | |
1da177e4 | 1259 | |
5db470e2 JK |
1260 | if (lo->lo_offset != info->lo_offset || |
1261 | lo->lo_sizelimit != info->lo_sizelimit) { | |
0c3796c2 | 1262 | size_changed = true; |
5db470e2 | 1263 | sync_blockdev(lo->lo_device); |
f4bd34b1 | 1264 | invalidate_bdev(lo->lo_device); |
5db470e2 JK |
1265 | } |
1266 | ||
ecdd0959 ML |
1267 | /* I/O need to be drained during transfer transition */ |
1268 | blk_mq_freeze_queue(lo->lo_queue); | |
1269 | ||
faf1d254 | 1270 | prev_lo_flags = lo->lo_flags; |
1da177e4 | 1271 | |
0c3796c2 | 1272 | err = loop_set_status_from_info(lo, info); |
1da177e4 | 1273 | if (err) |
550df5fd | 1274 | goto out_unfreeze; |
1da177e4 | 1275 | |
faf1d254 | 1276 | /* Mask out flags that can't be set using LOOP_SET_STATUS. */ |
6ac92fb5 | 1277 | lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; |
faf1d254 MC |
1278 | /* For those flags, use the previous values instead */ |
1279 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; | |
1280 | /* For flags that can't be cleared, use previous values too */ | |
1281 | lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; | |
1282 | ||
b0bd158d MC |
1283 | if (size_changed) { |
1284 | loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, | |
1285 | lo->lo_backing_file); | |
1286 | loop_set_size(lo, new_size); | |
b040ad9c | 1287 | } |
541c742a | 1288 | |
2e5ab5f3 ML |
1289 | /* update dio if lo_offset or transfer is changed */ |
1290 | __loop_update_dio(lo, lo->use_dio); | |
1291 | ||
550df5fd | 1292 | out_unfreeze: |
ecdd0959 | 1293 | blk_mq_unfreeze_queue(lo->lo_queue); |
e02898b4 | 1294 | |
faf1d254 MC |
1295 | if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && |
1296 | !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { | |
b9684a71 | 1297 | clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state); |
85b0a54a | 1298 | partscan = true; |
e02898b4 | 1299 | } |
550df5fd | 1300 | out_unlock: |
6cc8e743 | 1301 | mutex_unlock(&lo->lo_mutex); |
85b0a54a | 1302 | if (partscan) |
0384264e | 1303 | loop_reread_partitions(lo); |
e02898b4 | 1304 | |
ecdd0959 | 1305 | return err; |
1da177e4 LT |
1306 | } |
1307 | ||
1308 | static int | |
1309 | loop_get_status(struct loop_device *lo, struct loop_info64 *info) | |
1310 | { | |
b1ab5fa3 | 1311 | struct path path; |
1da177e4 | 1312 | struct kstat stat; |
2d1d4c1e | 1313 | int ret; |
1da177e4 | 1314 | |
6cc8e743 | 1315 | ret = mutex_lock_killable(&lo->lo_mutex); |
4a5ce9ba JK |
1316 | if (ret) |
1317 | return ret; | |
2d1d4c1e | 1318 | if (lo->lo_state != Lo_bound) { |
6cc8e743 | 1319 | mutex_unlock(&lo->lo_mutex); |
1da177e4 | 1320 | return -ENXIO; |
2d1d4c1e OS |
1321 | } |
1322 | ||
1da177e4 LT |
1323 | memset(info, 0, sizeof(*info)); |
1324 | info->lo_number = lo->lo_number; | |
1da177e4 LT |
1325 | info->lo_offset = lo->lo_offset; |
1326 | info->lo_sizelimit = lo->lo_sizelimit; | |
1327 | info->lo_flags = lo->lo_flags; | |
1328 | memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); | |
2d1d4c1e | 1329 | |
6cc8e743 | 1330 | /* Drop lo_mutex while we call into the filesystem. */ |
b1ab5fa3 TH |
1331 | path = lo->lo_backing_file->f_path; |
1332 | path_get(&path); | |
6cc8e743 | 1333 | mutex_unlock(&lo->lo_mutex); |
b1ab5fa3 | 1334 | ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); |
2d1d4c1e OS |
1335 | if (!ret) { |
1336 | info->lo_device = huge_encode_dev(stat.dev); | |
1337 | info->lo_inode = stat.ino; | |
1338 | info->lo_rdevice = huge_encode_dev(stat.rdev); | |
1339 | } | |
b1ab5fa3 | 1340 | path_put(&path); |
2d1d4c1e | 1341 | return ret; |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | static void | |
1345 | loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) | |
1346 | { | |
1347 | memset(info64, 0, sizeof(*info64)); | |
1348 | info64->lo_number = info->lo_number; | |
1349 | info64->lo_device = info->lo_device; | |
1350 | info64->lo_inode = info->lo_inode; | |
1351 | info64->lo_rdevice = info->lo_rdevice; | |
1352 | info64->lo_offset = info->lo_offset; | |
1353 | info64->lo_sizelimit = 0; | |
1da177e4 | 1354 | info64->lo_flags = info->lo_flags; |
47e96246 | 1355 | memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); |
1da177e4 LT |
1356 | } |
1357 | ||
1358 | static int | |
1359 | loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) | |
1360 | { | |
1361 | memset(info, 0, sizeof(*info)); | |
1362 | info->lo_number = info64->lo_number; | |
1363 | info->lo_device = info64->lo_device; | |
1364 | info->lo_inode = info64->lo_inode; | |
1365 | info->lo_rdevice = info64->lo_rdevice; | |
1366 | info->lo_offset = info64->lo_offset; | |
1da177e4 | 1367 | info->lo_flags = info64->lo_flags; |
47e96246 | 1368 | memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); |
1da177e4 LT |
1369 | |
1370 | /* error in case values were truncated */ | |
1371 | if (info->lo_device != info64->lo_device || | |
1372 | info->lo_rdevice != info64->lo_rdevice || | |
1373 | info->lo_inode != info64->lo_inode || | |
1374 | info->lo_offset != info64->lo_offset) | |
1375 | return -EOVERFLOW; | |
1376 | ||
1377 | return 0; | |
1378 | } | |
1379 | ||
1380 | static int | |
1381 | loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) | |
1382 | { | |
1383 | struct loop_info info; | |
1384 | struct loop_info64 info64; | |
1385 | ||
1386 | if (copy_from_user(&info, arg, sizeof (struct loop_info))) | |
1387 | return -EFAULT; | |
1388 | loop_info64_from_old(&info, &info64); | |
1389 | return loop_set_status(lo, &info64); | |
1390 | } | |
1391 | ||
1392 | static int | |
1393 | loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) | |
1394 | { | |
1395 | struct loop_info64 info64; | |
1396 | ||
1397 | if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) | |
1398 | return -EFAULT; | |
1399 | return loop_set_status(lo, &info64); | |
1400 | } | |
1401 | ||
1402 | static int | |
1403 | loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { | |
1404 | struct loop_info info; | |
1405 | struct loop_info64 info64; | |
bdac616d | 1406 | int err; |
1da177e4 | 1407 | |
4a5ce9ba | 1408 | if (!arg) |
bdac616d | 1409 | return -EINVAL; |
bdac616d | 1410 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1411 | if (!err) |
1412 | err = loop_info64_to_old(&info64, &info); | |
1413 | if (!err && copy_to_user(arg, &info, sizeof(info))) | |
1414 | err = -EFAULT; | |
1415 | ||
1416 | return err; | |
1417 | } | |
1418 | ||
1419 | static int | |
1420 | loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { | |
1421 | struct loop_info64 info64; | |
bdac616d | 1422 | int err; |
1da177e4 | 1423 | |
4a5ce9ba | 1424 | if (!arg) |
bdac616d | 1425 | return -EINVAL; |
bdac616d | 1426 | err = loop_get_status(lo, &info64); |
1da177e4 LT |
1427 | if (!err && copy_to_user(arg, &info64, sizeof(info64))) |
1428 | err = -EFAULT; | |
1429 | ||
1430 | return err; | |
1431 | } | |
1432 | ||
51001b7d | 1433 | static int loop_set_capacity(struct loop_device *lo) |
53d66608 | 1434 | { |
0a6ed1b5 MC |
1435 | loff_t size; |
1436 | ||
53d66608 | 1437 | if (unlikely(lo->lo_state != Lo_bound)) |
7b0576a3 | 1438 | return -ENXIO; |
53d66608 | 1439 | |
0a6ed1b5 MC |
1440 | size = get_loop_size(lo, lo->lo_backing_file); |
1441 | loop_set_size(lo, size); | |
083a6a50 MC |
1442 | |
1443 | return 0; | |
53d66608 O |
1444 | } |
1445 | ||
ab1cb278 ML |
1446 | static int loop_set_dio(struct loop_device *lo, unsigned long arg) |
1447 | { | |
1448 | int error = -ENXIO; | |
1449 | if (lo->lo_state != Lo_bound) | |
1450 | goto out; | |
1451 | ||
1452 | __loop_update_dio(lo, !!arg); | |
1453 | if (lo->use_dio == !!arg) | |
1454 | return 0; | |
1455 | error = -EINVAL; | |
1456 | out: | |
1457 | return error; | |
1458 | } | |
1459 | ||
89e4fdec OS |
1460 | static int loop_set_block_size(struct loop_device *lo, unsigned long arg) |
1461 | { | |
5db470e2 JK |
1462 | int err = 0; |
1463 | ||
89e4fdec OS |
1464 | if (lo->lo_state != Lo_bound) |
1465 | return -ENXIO; | |
1466 | ||
7e81f99a MC |
1467 | if (lo->lo_queue->limits.logical_block_size == arg) |
1468 | return 0; | |
1469 | ||
1470 | sync_blockdev(lo->lo_device); | |
f4bd34b1 | 1471 | invalidate_bdev(lo->lo_device); |
5db470e2 | 1472 | |
89e4fdec | 1473 | blk_mq_freeze_queue(lo->lo_queue); |
ae0d40ff | 1474 | err = loop_reconfigure_limits(lo, arg); |
89e4fdec | 1475 | loop_update_dio(lo); |
89e4fdec OS |
1476 | blk_mq_unfreeze_queue(lo->lo_queue); |
1477 | ||
5db470e2 | 1478 | return err; |
89e4fdec OS |
1479 | } |
1480 | ||
a1316544 JK |
1481 | static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, |
1482 | unsigned long arg) | |
1da177e4 | 1483 | { |
1da177e4 LT |
1484 | int err; |
1485 | ||
6cc8e743 | 1486 | err = mutex_lock_killable(&lo->lo_mutex); |
3148ffbd | 1487 | if (err) |
a1316544 JK |
1488 | return err; |
1489 | switch (cmd) { | |
1490 | case LOOP_SET_CAPACITY: | |
1491 | err = loop_set_capacity(lo); | |
1492 | break; | |
1493 | case LOOP_SET_DIRECT_IO: | |
1494 | err = loop_set_dio(lo, arg); | |
1495 | break; | |
1496 | case LOOP_SET_BLOCK_SIZE: | |
1497 | err = loop_set_block_size(lo, arg); | |
1498 | break; | |
1499 | default: | |
47e96246 | 1500 | err = -EINVAL; |
a1316544 | 1501 | } |
6cc8e743 | 1502 | mutex_unlock(&lo->lo_mutex); |
a1316544 JK |
1503 | return err; |
1504 | } | |
1505 | ||
05bdb996 | 1506 | static int lo_ioctl(struct block_device *bdev, blk_mode_t mode, |
a1316544 JK |
1507 | unsigned int cmd, unsigned long arg) |
1508 | { | |
1509 | struct loop_device *lo = bdev->bd_disk->private_data; | |
571fae6e | 1510 | void __user *argp = (void __user *) arg; |
a1316544 | 1511 | int err; |
3148ffbd | 1512 | |
1da177e4 | 1513 | switch (cmd) { |
3448914e MC |
1514 | case LOOP_SET_FD: { |
1515 | /* | |
1516 | * Legacy case - pass in a zeroed out struct loop_config with | |
1517 | * only the file descriptor set , which corresponds with the | |
1518 | * default parameters we'd have used otherwise. | |
1519 | */ | |
1520 | struct loop_config config; | |
1521 | ||
1522 | memset(&config, 0, sizeof(config)); | |
1523 | config.fd = arg; | |
1524 | ||
1525 | return loop_configure(lo, mode, bdev, &config); | |
1526 | } | |
1527 | case LOOP_CONFIGURE: { | |
1528 | struct loop_config config; | |
1529 | ||
1530 | if (copy_from_user(&config, argp, sizeof(config))) | |
1531 | return -EFAULT; | |
1532 | ||
1533 | return loop_configure(lo, mode, bdev, &config); | |
1534 | } | |
1da177e4 | 1535 | case LOOP_CHANGE_FD: |
c3710770 | 1536 | return loop_change_fd(lo, bdev, arg); |
1da177e4 | 1537 | case LOOP_CLR_FD: |
7ccd0791 | 1538 | return loop_clr_fd(lo); |
1da177e4 | 1539 | case LOOP_SET_STATUS: |
7035b5df | 1540 | err = -EPERM; |
05bdb996 | 1541 | if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN)) |
571fae6e | 1542 | err = loop_set_status_old(lo, argp); |
1da177e4 LT |
1543 | break; |
1544 | case LOOP_GET_STATUS: | |
571fae6e | 1545 | return loop_get_status_old(lo, argp); |
1da177e4 | 1546 | case LOOP_SET_STATUS64: |
7035b5df | 1547 | err = -EPERM; |
05bdb996 | 1548 | if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN)) |
571fae6e | 1549 | err = loop_set_status64(lo, argp); |
1da177e4 LT |
1550 | break; |
1551 | case LOOP_GET_STATUS64: | |
571fae6e | 1552 | return loop_get_status64(lo, argp); |
a1316544 | 1553 | case LOOP_SET_CAPACITY: |
ab1cb278 | 1554 | case LOOP_SET_DIRECT_IO: |
89e4fdec | 1555 | case LOOP_SET_BLOCK_SIZE: |
05bdb996 | 1556 | if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN)) |
a1316544 | 1557 | return -EPERM; |
df561f66 | 1558 | fallthrough; |
1da177e4 | 1559 | default: |
a1316544 JK |
1560 | err = lo_simple_ioctl(lo, cmd, arg); |
1561 | break; | |
1da177e4 | 1562 | } |
f028f3b2 | 1563 | |
1da177e4 LT |
1564 | return err; |
1565 | } | |
1566 | ||
863d5b82 DH |
1567 | #ifdef CONFIG_COMPAT |
1568 | struct compat_loop_info { | |
1569 | compat_int_t lo_number; /* ioctl r/o */ | |
1570 | compat_dev_t lo_device; /* ioctl r/o */ | |
1571 | compat_ulong_t lo_inode; /* ioctl r/o */ | |
1572 | compat_dev_t lo_rdevice; /* ioctl r/o */ | |
1573 | compat_int_t lo_offset; | |
f941c51e | 1574 | compat_int_t lo_encrypt_type; /* obsolete, ignored */ |
863d5b82 DH |
1575 | compat_int_t lo_encrypt_key_size; /* ioctl w/o */ |
1576 | compat_int_t lo_flags; /* ioctl r/o */ | |
1577 | char lo_name[LO_NAME_SIZE]; | |
1578 | unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ | |
1579 | compat_ulong_t lo_init[2]; | |
1580 | char reserved[4]; | |
1581 | }; | |
1582 | ||
1583 | /* | |
1584 | * Transfer 32-bit compatibility structure in userspace to 64-bit loop info | |
1585 | * - noinlined to reduce stack space usage in main part of driver | |
1586 | */ | |
1587 | static noinline int | |
ba674cfc | 1588 | loop_info64_from_compat(const struct compat_loop_info __user *arg, |
863d5b82 DH |
1589 | struct loop_info64 *info64) |
1590 | { | |
1591 | struct compat_loop_info info; | |
1592 | ||
1593 | if (copy_from_user(&info, arg, sizeof(info))) | |
1594 | return -EFAULT; | |
1595 | ||
1596 | memset(info64, 0, sizeof(*info64)); | |
1597 | info64->lo_number = info.lo_number; | |
1598 | info64->lo_device = info.lo_device; | |
1599 | info64->lo_inode = info.lo_inode; | |
1600 | info64->lo_rdevice = info.lo_rdevice; | |
1601 | info64->lo_offset = info.lo_offset; | |
1602 | info64->lo_sizelimit = 0; | |
863d5b82 | 1603 | info64->lo_flags = info.lo_flags; |
47e96246 | 1604 | memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); |
863d5b82 DH |
1605 | return 0; |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Transfer 64-bit loop info to 32-bit compatibility structure in userspace | |
1610 | * - noinlined to reduce stack space usage in main part of driver | |
1611 | */ | |
1612 | static noinline int | |
1613 | loop_info64_to_compat(const struct loop_info64 *info64, | |
1614 | struct compat_loop_info __user *arg) | |
1615 | { | |
1616 | struct compat_loop_info info; | |
1617 | ||
1618 | memset(&info, 0, sizeof(info)); | |
1619 | info.lo_number = info64->lo_number; | |
1620 | info.lo_device = info64->lo_device; | |
1621 | info.lo_inode = info64->lo_inode; | |
1622 | info.lo_rdevice = info64->lo_rdevice; | |
1623 | info.lo_offset = info64->lo_offset; | |
863d5b82 | 1624 | info.lo_flags = info64->lo_flags; |
47e96246 | 1625 | memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); |
863d5b82 DH |
1626 | |
1627 | /* error in case values were truncated */ | |
1628 | if (info.lo_device != info64->lo_device || | |
1629 | info.lo_rdevice != info64->lo_rdevice || | |
1630 | info.lo_inode != info64->lo_inode || | |
47e96246 | 1631 | info.lo_offset != info64->lo_offset) |
863d5b82 DH |
1632 | return -EOVERFLOW; |
1633 | ||
1634 | if (copy_to_user(arg, &info, sizeof(info))) | |
1635 | return -EFAULT; | |
1636 | return 0; | |
1637 | } | |
1638 | ||
1639 | static int | |
1640 | loop_set_status_compat(struct loop_device *lo, | |
1641 | const struct compat_loop_info __user *arg) | |
1642 | { | |
1643 | struct loop_info64 info64; | |
1644 | int ret; | |
1645 | ||
1646 | ret = loop_info64_from_compat(arg, &info64); | |
1647 | if (ret < 0) | |
1648 | return ret; | |
1649 | return loop_set_status(lo, &info64); | |
1650 | } | |
1651 | ||
1652 | static int | |
1653 | loop_get_status_compat(struct loop_device *lo, | |
1654 | struct compat_loop_info __user *arg) | |
1655 | { | |
1656 | struct loop_info64 info64; | |
bdac616d | 1657 | int err; |
863d5b82 | 1658 | |
4a5ce9ba | 1659 | if (!arg) |
bdac616d | 1660 | return -EINVAL; |
bdac616d | 1661 | err = loop_get_status(lo, &info64); |
863d5b82 DH |
1662 | if (!err) |
1663 | err = loop_info64_to_compat(&info64, arg); | |
1664 | return err; | |
1665 | } | |
1666 | ||
05bdb996 | 1667 | static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode, |
bb214884 | 1668 | unsigned int cmd, unsigned long arg) |
863d5b82 | 1669 | { |
bb214884 | 1670 | struct loop_device *lo = bdev->bd_disk->private_data; |
863d5b82 DH |
1671 | int err; |
1672 | ||
863d5b82 DH |
1673 | switch(cmd) { |
1674 | case LOOP_SET_STATUS: | |
550df5fd JK |
1675 | err = loop_set_status_compat(lo, |
1676 | (const struct compat_loop_info __user *)arg); | |
863d5b82 DH |
1677 | break; |
1678 | case LOOP_GET_STATUS: | |
4a5ce9ba JK |
1679 | err = loop_get_status_compat(lo, |
1680 | (struct compat_loop_info __user *)arg); | |
863d5b82 | 1681 | break; |
53d66608 | 1682 | case LOOP_SET_CAPACITY: |
863d5b82 DH |
1683 | case LOOP_CLR_FD: |
1684 | case LOOP_GET_STATUS64: | |
1685 | case LOOP_SET_STATUS64: | |
3448914e | 1686 | case LOOP_CONFIGURE: |
863d5b82 | 1687 | arg = (unsigned long) compat_ptr(arg); |
df561f66 | 1688 | fallthrough; |
863d5b82 DH |
1689 | case LOOP_SET_FD: |
1690 | case LOOP_CHANGE_FD: | |
9fea4b39 | 1691 | case LOOP_SET_BLOCK_SIZE: |
fdbe4eee | 1692 | case LOOP_SET_DIRECT_IO: |
bb214884 | 1693 | err = lo_ioctl(bdev, mode, cmd, arg); |
863d5b82 DH |
1694 | break; |
1695 | default: | |
1696 | err = -ENOIOCTLCMD; | |
1697 | break; | |
1698 | } | |
863d5b82 DH |
1699 | return err; |
1700 | } | |
1701 | #endif | |
1702 | ||
18048c1a GM |
1703 | static int lo_open(struct gendisk *disk, blk_mode_t mode) |
1704 | { | |
1705 | struct loop_device *lo = disk->private_data; | |
1706 | int err; | |
1707 | ||
1708 | err = mutex_lock_killable(&lo->lo_mutex); | |
1709 | if (err) | |
1710 | return err; | |
1711 | ||
1712 | if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown) | |
1713 | err = -ENXIO; | |
1714 | mutex_unlock(&lo->lo_mutex); | |
1715 | return err; | |
1716 | } | |
1717 | ||
ae220766 | 1718 | static void lo_release(struct gendisk *disk) |
1da177e4 | 1719 | { |
6cc8e743 | 1720 | struct loop_device *lo = disk->private_data; |
18048c1a | 1721 | bool need_clear = false; |
1da177e4 | 1722 | |
a0e286b6 CH |
1723 | if (disk_openers(disk) > 0) |
1724 | return; | |
18048c1a GM |
1725 | /* |
1726 | * Clear the backing device information if this is the last close of | |
1727 | * a device that's been marked for auto clear, or on which LOOP_CLR_FD | |
1728 | * has been called. | |
1729 | */ | |
14f27939 | 1730 | |
a0e286b6 | 1731 | mutex_lock(&lo->lo_mutex); |
18048c1a | 1732 | if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) |
a2505b79 | 1733 | lo->lo_state = Lo_rundown; |
18048c1a GM |
1734 | |
1735 | need_clear = (lo->lo_state == Lo_rundown); | |
6cc8e743 | 1736 | mutex_unlock(&lo->lo_mutex); |
18048c1a GM |
1737 | |
1738 | if (need_clear) | |
1739 | __loop_clr_fd(lo); | |
ae665016 LT |
1740 | } |
1741 | ||
d2c7f56f CH |
1742 | static void lo_free_disk(struct gendisk *disk) |
1743 | { | |
1744 | struct loop_device *lo = disk->private_data; | |
1745 | ||
d292dc80 CH |
1746 | if (lo->workqueue) |
1747 | destroy_workqueue(lo->workqueue); | |
1748 | loop_free_idle_workers(lo, true); | |
292a089d | 1749 | timer_shutdown_sync(&lo->timer); |
d2c7f56f CH |
1750 | mutex_destroy(&lo->lo_mutex); |
1751 | kfree(lo); | |
1752 | } | |
1753 | ||
83d5cde4 | 1754 | static const struct block_device_operations lo_fops = { |
1da177e4 | 1755 | .owner = THIS_MODULE, |
18048c1a | 1756 | .open = lo_open, |
bb214884 AV |
1757 | .release = lo_release, |
1758 | .ioctl = lo_ioctl, | |
863d5b82 | 1759 | #ifdef CONFIG_COMPAT |
bb214884 | 1760 | .compat_ioctl = lo_compat_ioctl, |
863d5b82 | 1761 | #endif |
d2c7f56f | 1762 | .free_disk = lo_free_disk, |
1da177e4 LT |
1763 | }; |
1764 | ||
1765 | /* | |
1766 | * And now the modules code and kernel interface. | |
1767 | */ | |
85c50197 IM |
1768 | |
1769 | /* | |
1770 | * If max_loop is specified, create that many devices upfront. | |
1771 | * This also becomes a hard limit. If max_loop is not specified, | |
bb5faa99 MFO |
1772 | * the default isn't a hard limit (as before commit 85c50197716c |
1773 | * changed the default value from 0 for max_loop=0 reasons), just | |
85c50197 IM |
1774 | * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module |
1775 | * init time. Loop devices can be requested on-demand with the | |
1776 | * /dev/loop-control interface, or be instantiated by accessing | |
1777 | * a 'dead' device node. | |
1778 | */ | |
1779 | static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT; | |
bb5faa99 MFO |
1780 | |
1781 | #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD | |
1782 | static bool max_loop_specified; | |
1783 | ||
1784 | static int max_loop_param_set_int(const char *val, | |
1785 | const struct kernel_param *kp) | |
1786 | { | |
1787 | int ret; | |
1788 | ||
1789 | ret = param_set_int(val, kp); | |
1790 | if (ret < 0) | |
1791 | return ret; | |
1792 | ||
1793 | max_loop_specified = true; | |
1794 | return 0; | |
1795 | } | |
1796 | ||
1797 | static const struct kernel_param_ops max_loop_param_ops = { | |
1798 | .set = max_loop_param_set_int, | |
1799 | .get = param_get_int, | |
1800 | }; | |
1801 | ||
1802 | module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444); | |
a47653fc | 1803 | MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); |
bb5faa99 MFO |
1804 | #else |
1805 | module_param(max_loop, int, 0444); | |
1806 | MODULE_PARM_DESC(max_loop, "Initial number of loop devices"); | |
1807 | #endif | |
1808 | ||
5657a819 | 1809 | module_param(max_part, int, 0444); |
476a4813 | 1810 | MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); |
ef44c508 CK |
1811 | |
1812 | static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; | |
1813 | ||
1814 | static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) | |
1815 | { | |
e152a05f | 1816 | int qd, ret; |
ef44c508 | 1817 | |
e152a05f BVA |
1818 | ret = kstrtoint(s, 0, &qd); |
1819 | if (ret < 0) | |
1820 | return ret; | |
1821 | if (qd < 1) | |
1822 | return -EINVAL; | |
1823 | hw_queue_depth = qd; | |
1824 | return 0; | |
ef44c508 CK |
1825 | } |
1826 | ||
1827 | static const struct kernel_param_ops loop_hw_qdepth_param_ops = { | |
1828 | .set = loop_set_hw_queue_depth, | |
1829 | .get = param_get_int, | |
1830 | }; | |
1831 | ||
1832 | device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); | |
e152a05f | 1833 | MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH)); |
ef44c508 | 1834 | |
7d4425d2 | 1835 | MODULE_DESCRIPTION("Loopback device support"); |
1da177e4 LT |
1836 | MODULE_LICENSE("GPL"); |
1837 | MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); | |
1838 | ||
fc17b653 | 1839 | static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, |
b5dd2f60 ML |
1840 | const struct blk_mq_queue_data *bd) |
1841 | { | |
1894e916 JA |
1842 | struct request *rq = bd->rq; |
1843 | struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); | |
1844 | struct loop_device *lo = rq->q->queuedata; | |
b5dd2f60 | 1845 | |
1894e916 | 1846 | blk_mq_start_request(rq); |
b5dd2f60 | 1847 | |
f4aa4c7b | 1848 | if (lo->lo_state != Lo_bound) |
fc17b653 | 1849 | return BLK_STS_IOERR; |
f4aa4c7b | 1850 | |
1894e916 | 1851 | switch (req_op(rq)) { |
f0225cac CH |
1852 | case REQ_OP_FLUSH: |
1853 | case REQ_OP_DISCARD: | |
19372e27 | 1854 | case REQ_OP_WRITE_ZEROES: |
bc07c10a | 1855 | cmd->use_aio = false; |
f0225cac CH |
1856 | break; |
1857 | default: | |
1858 | cmd->use_aio = lo->use_dio; | |
1859 | break; | |
1860 | } | |
bc07c10a | 1861 | |
d4478e92 | 1862 | /* always use the first bio's css */ |
c74d40e8 DS |
1863 | cmd->blkcg_css = NULL; |
1864 | cmd->memcg_css = NULL; | |
0b508bc9 | 1865 | #ifdef CONFIG_BLK_CGROUP |
bbb1ebe7 CH |
1866 | if (rq->bio) { |
1867 | cmd->blkcg_css = bio_blkcg_css(rq->bio); | |
c74d40e8 | 1868 | #ifdef CONFIG_MEMCG |
bbb1ebe7 CH |
1869 | if (cmd->blkcg_css) { |
1870 | cmd->memcg_css = | |
1871 | cgroup_get_e_css(cmd->blkcg_css->cgroup, | |
1872 | &memory_cgrp_subsys); | |
1873 | } | |
c74d40e8 DS |
1874 | #endif |
1875 | } | |
d4478e92 | 1876 | #endif |
87579e9b | 1877 | loop_queue_work(lo, cmd); |
b5dd2f60 | 1878 | |
fc17b653 | 1879 | return BLK_STS_OK; |
b5dd2f60 ML |
1880 | } |
1881 | ||
1882 | static void loop_handle_cmd(struct loop_cmd *cmd) | |
1883 | { | |
9b0cb770 BVA |
1884 | struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css; |
1885 | struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css; | |
1894e916 JA |
1886 | struct request *rq = blk_mq_rq_from_pdu(cmd); |
1887 | const bool write = op_is_write(req_op(rq)); | |
1888 | struct loop_device *lo = rq->q->queuedata; | |
f4829a9b | 1889 | int ret = 0; |
c74d40e8 | 1890 | struct mem_cgroup *old_memcg = NULL; |
9b0cb770 | 1891 | const bool use_aio = cmd->use_aio; |
b5dd2f60 | 1892 | |
f4829a9b CH |
1893 | if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { |
1894 | ret = -EIO; | |
b5dd2f60 | 1895 | goto failed; |
f4829a9b | 1896 | } |
b5dd2f60 | 1897 | |
9b0cb770 BVA |
1898 | if (cmd_blkcg_css) |
1899 | kthread_associate_blkcg(cmd_blkcg_css); | |
1900 | if (cmd_memcg_css) | |
c74d40e8 | 1901 | old_memcg = set_active_memcg( |
9b0cb770 | 1902 | mem_cgroup_from_css(cmd_memcg_css)); |
c74d40e8 | 1903 | |
9b0cb770 BVA |
1904 | /* |
1905 | * do_req_filebacked() may call blk_mq_complete_request() synchronously | |
1906 | * or asynchronously if using aio. Hence, do not touch 'cmd' after | |
1907 | * do_req_filebacked() has returned unless we are sure that 'cmd' has | |
1908 | * not yet been completed. | |
1909 | */ | |
1894e916 | 1910 | ret = do_req_filebacked(lo, rq); |
c74d40e8 | 1911 | |
9b0cb770 | 1912 | if (cmd_blkcg_css) |
c74d40e8 DS |
1913 | kthread_associate_blkcg(NULL); |
1914 | ||
9b0cb770 | 1915 | if (cmd_memcg_css) { |
c74d40e8 | 1916 | set_active_memcg(old_memcg); |
9b0cb770 | 1917 | css_put(cmd_memcg_css); |
c74d40e8 | 1918 | } |
b5dd2f60 | 1919 | failed: |
bc07c10a | 1920 | /* complete non-aio request */ |
9b0cb770 | 1921 | if (!use_aio || ret) { |
8cd55087 EG |
1922 | if (ret == -EOPNOTSUPP) |
1923 | cmd->ret = ret; | |
1924 | else | |
1925 | cmd->ret = ret ? -EIO : 0; | |
15f73f5b CH |
1926 | if (likely(!blk_should_fake_timeout(rq->q))) |
1927 | blk_mq_complete_request(rq); | |
fe2cb290 | 1928 | } |
b5dd2f60 ML |
1929 | } |
1930 | ||
87579e9b DS |
1931 | static void loop_process_work(struct loop_worker *worker, |
1932 | struct list_head *cmd_list, struct loop_device *lo) | |
b5dd2f60 | 1933 | { |
87579e9b DS |
1934 | int orig_flags = current->flags; |
1935 | struct loop_cmd *cmd; | |
b5dd2f60 | 1936 | |
87579e9b DS |
1937 | current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; |
1938 | spin_lock_irq(&lo->lo_work_lock); | |
1939 | while (!list_empty(cmd_list)) { | |
1940 | cmd = container_of( | |
1941 | cmd_list->next, struct loop_cmd, list_entry); | |
1942 | list_del(cmd_list->next); | |
1943 | spin_unlock_irq(&lo->lo_work_lock); | |
1944 | ||
1945 | loop_handle_cmd(cmd); | |
1946 | cond_resched(); | |
1947 | ||
1948 | spin_lock_irq(&lo->lo_work_lock); | |
1949 | } | |
1950 | ||
1951 | /* | |
1952 | * We only add to the idle list if there are no pending cmds | |
1953 | * *and* the worker will not run again which ensures that it | |
1954 | * is safe to free any worker on the idle list | |
1955 | */ | |
1956 | if (worker && !work_pending(&worker->work)) { | |
1957 | worker->last_ran_at = jiffies; | |
1958 | list_add_tail(&worker->idle_list, &lo->idle_worker_list); | |
1959 | loop_set_timer(lo); | |
1960 | } | |
1961 | spin_unlock_irq(&lo->lo_work_lock); | |
1962 | current->flags = orig_flags; | |
b5dd2f60 ML |
1963 | } |
1964 | ||
87579e9b | 1965 | static void loop_workfn(struct work_struct *work) |
b5dd2f60 | 1966 | { |
87579e9b DS |
1967 | struct loop_worker *worker = |
1968 | container_of(work, struct loop_worker, work); | |
1969 | loop_process_work(worker, &worker->cmd_list, worker->lo); | |
1970 | } | |
b5dd2f60 | 1971 | |
87579e9b DS |
1972 | static void loop_rootcg_workfn(struct work_struct *work) |
1973 | { | |
1974 | struct loop_device *lo = | |
1975 | container_of(work, struct loop_device, rootcg_work); | |
1976 | loop_process_work(NULL, &lo->rootcg_cmd_list, lo); | |
1977 | } | |
1978 | ||
f363b089 | 1979 | static const struct blk_mq_ops loop_mq_ops = { |
b5dd2f60 | 1980 | .queue_rq = loop_queue_rq, |
fe2cb290 | 1981 | .complete = lo_complete_rq, |
b5dd2f60 ML |
1982 | }; |
1983 | ||
d6da83d0 | 1984 | static int loop_add(int i) |
73285082 | 1985 | { |
02aed4a1 CH |
1986 | struct queue_limits lim = { |
1987 | /* | |
1988 | * Random number picked from the historic block max_sectors cap. | |
1989 | */ | |
1990 | .max_hw_sectors = 2560u, | |
1991 | }; | |
73285082 KC |
1992 | struct loop_device *lo; |
1993 | struct gendisk *disk; | |
34dd82af | 1994 | int err; |
73285082 | 1995 | |
68d740d7 | 1996 | err = -ENOMEM; |
73285082 | 1997 | lo = kzalloc(sizeof(*lo), GFP_KERNEL); |
68d740d7 | 1998 | if (!lo) |
73285082 | 1999 | goto out; |
b15ed546 CH |
2000 | lo->worker_tree = RB_ROOT; |
2001 | INIT_LIST_HEAD(&lo->idle_worker_list); | |
2002 | timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE); | |
ef7e7c82 MP |
2003 | lo->lo_state = Lo_unbound; |
2004 | ||
18d1f200 CH |
2005 | err = mutex_lock_killable(&loop_ctl_mutex); |
2006 | if (err) | |
2007 | goto out_free_dev; | |
2008 | ||
c718aa65 | 2009 | /* allocate id, if @id >= 0, we're requesting that specific id */ |
34dd82af | 2010 | if (i >= 0) { |
c718aa65 TH |
2011 | err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); |
2012 | if (err == -ENOSPC) | |
34dd82af | 2013 | err = -EEXIST; |
34dd82af | 2014 | } else { |
c718aa65 | 2015 | err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); |
34dd82af | 2016 | } |
1c500ad7 | 2017 | mutex_unlock(&loop_ctl_mutex); |
34dd82af | 2018 | if (err < 0) |
1c500ad7 | 2019 | goto out_free_dev; |
c718aa65 | 2020 | i = err; |
73285082 | 2021 | |
b5dd2f60 ML |
2022 | lo->tag_set.ops = &loop_mq_ops; |
2023 | lo->tag_set.nr_hw_queues = 1; | |
ef44c508 | 2024 | lo->tag_set.queue_depth = hw_queue_depth; |
b5dd2f60 ML |
2025 | lo->tag_set.numa_node = NUMA_NO_NODE; |
2026 | lo->tag_set.cmd_size = sizeof(struct loop_cmd); | |
2112f5c1 BVA |
2027 | lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | |
2028 | BLK_MQ_F_NO_SCHED_BY_DEFAULT; | |
b5dd2f60 ML |
2029 | lo->tag_set.driver_data = lo; |
2030 | ||
2031 | err = blk_mq_alloc_tag_set(&lo->tag_set); | |
2032 | if (err) | |
3ec981e3 | 2033 | goto out_free_idr; |
73285082 | 2034 | |
02aed4a1 | 2035 | disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo); |
1c99502f CH |
2036 | if (IS_ERR(disk)) { |
2037 | err = PTR_ERR(disk); | |
b5dd2f60 ML |
2038 | goto out_cleanup_tags; |
2039 | } | |
1c99502f | 2040 | lo->lo_queue = lo->lo_disk->queue; |
ef7e7c82 | 2041 | |
e03c8dd1 KS |
2042 | /* |
2043 | * Disable partition scanning by default. The in-kernel partition | |
2044 | * scanning can be requested individually per-device during its | |
2045 | * setup. Userspace can always add and remove partitions from all | |
2046 | * devices. The needed partition minors are allocated from the | |
2047 | * extended minor space, the main loop device numbers will continue | |
2048 | * to match the loop minors, regardless of the number of partitions | |
2049 | * used. | |
2050 | * | |
2051 | * If max_part is given, partition scanning is globally enabled for | |
2052 | * all loop devices. The minors for the main loop devices will be | |
2053 | * multiples of max_part. | |
2054 | * | |
2055 | * Note: Global-for-all-devices, set-only-at-init, read-only module | |
2056 | * parameteters like 'max_loop' and 'max_part' make things needlessly | |
2057 | * complicated, are too static, inflexible and may surprise | |
2058 | * userspace tools. Parameters like this in general should be avoided. | |
2059 | */ | |
2060 | if (!part_shift) | |
b9684a71 | 2061 | set_bit(GD_SUPPRESS_PART_SCAN, &disk->state); |
6cc8e743 | 2062 | mutex_init(&lo->lo_mutex); |
73285082 | 2063 | lo->lo_number = i; |
73285082 | 2064 | spin_lock_init(&lo->lo_lock); |
87579e9b | 2065 | spin_lock_init(&lo->lo_work_lock); |
d292dc80 CH |
2066 | INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); |
2067 | INIT_LIST_HEAD(&lo->rootcg_cmd_list); | |
73285082 | 2068 | disk->major = LOOP_MAJOR; |
476a4813 | 2069 | disk->first_minor = i << part_shift; |
1c99502f | 2070 | disk->minors = 1 << part_shift; |
73285082 KC |
2071 | disk->fops = &lo_fops; |
2072 | disk->private_data = lo; | |
2073 | disk->queue = lo->lo_queue; | |
9f65c489 MC |
2074 | disk->events = DISK_EVENT_MEDIA_CHANGE; |
2075 | disk->event_flags = DISK_EVENT_FLAG_UEVENT; | |
73285082 | 2076 | sprintf(disk->disk_name, "loop%d", i); |
1c500ad7 | 2077 | /* Make this loop device reachable from pathname. */ |
905705f0 LC |
2078 | err = add_disk(disk); |
2079 | if (err) | |
2080 | goto out_cleanup_disk; | |
2081 | ||
1c500ad7 TH |
2082 | /* Show this loop device. */ |
2083 | mutex_lock(&loop_ctl_mutex); | |
2084 | lo->idr_visible = true; | |
18d1f200 | 2085 | mutex_unlock(&loop_ctl_mutex); |
905705f0 | 2086 | |
18d1f200 | 2087 | return i; |
73285082 | 2088 | |
905705f0 | 2089 | out_cleanup_disk: |
8b9ab626 | 2090 | put_disk(disk); |
b5dd2f60 ML |
2091 | out_cleanup_tags: |
2092 | blk_mq_free_tag_set(&lo->tag_set); | |
3ec981e3 | 2093 | out_free_idr: |
1c500ad7 | 2094 | mutex_lock(&loop_ctl_mutex); |
3ec981e3 | 2095 | idr_remove(&loop_index_idr, i); |
18d1f200 | 2096 | mutex_unlock(&loop_ctl_mutex); |
73285082 KC |
2097 | out_free_dev: |
2098 | kfree(lo); | |
2099 | out: | |
34dd82af | 2100 | return err; |
73285082 KC |
2101 | } |
2102 | ||
34dd82af | 2103 | static void loop_remove(struct loop_device *lo) |
1da177e4 | 2104 | { |
1c500ad7 | 2105 | /* Make this loop device unreachable from pathname. */ |
6cd18e71 | 2106 | del_gendisk(lo->lo_disk); |
b5dd2f60 | 2107 | blk_mq_free_tag_set(&lo->tag_set); |
ce8d7861 | 2108 | |
1c500ad7 TH |
2109 | mutex_lock(&loop_ctl_mutex); |
2110 | idr_remove(&loop_index_idr, lo->lo_number); | |
2111 | mutex_unlock(&loop_ctl_mutex); | |
d2c7f56f CH |
2112 | |
2113 | put_disk(lo->lo_disk); | |
73285082 | 2114 | } |
1da177e4 | 2115 | |
23881aec | 2116 | #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD |
8410d38c | 2117 | static void loop_probe(dev_t dev) |
73285082 | 2118 | { |
8410d38c | 2119 | int idx = MINOR(dev) >> part_shift; |
8410d38c | 2120 | |
bb5faa99 | 2121 | if (max_loop_specified && max_loop && idx >= max_loop) |
8410d38c | 2122 | return; |
4157fe0b | 2123 | loop_add(idx); |
f9d10764 | 2124 | } |
23881aec MFO |
2125 | #else |
2126 | #define loop_probe NULL | |
2127 | #endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */ | |
f9d10764 CH |
2128 | |
2129 | static int loop_control_remove(int idx) | |
2130 | { | |
2131 | struct loop_device *lo; | |
2132 | int ret; | |
e5d66a10 CH |
2133 | |
2134 | if (idx < 0) { | |
e3f9387a | 2135 | pr_warn_once("deleting an unspecified loop device is not supported.\n"); |
e5d66a10 CH |
2136 | return -EINVAL; |
2137 | } | |
f9d10764 | 2138 | |
1c500ad7 | 2139 | /* Hide this loop device for serialization. */ |
f9d10764 CH |
2140 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2141 | if (ret) | |
2142 | return ret; | |
b9848081 | 2143 | lo = idr_find(&loop_index_idr, idx); |
1c500ad7 | 2144 | if (!lo || !lo->idr_visible) |
b9848081 | 2145 | ret = -ENODEV; |
1c500ad7 TH |
2146 | else |
2147 | lo->idr_visible = false; | |
2148 | mutex_unlock(&loop_ctl_mutex); | |
2149 | if (ret) | |
2150 | return ret; | |
f9d10764 | 2151 | |
1c500ad7 | 2152 | /* Check whether this loop device can be removed. */ |
f9d10764 CH |
2153 | ret = mutex_lock_killable(&lo->lo_mutex); |
2154 | if (ret) | |
1c500ad7 | 2155 | goto mark_visible; |
a0e286b6 | 2156 | if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) { |
f9d10764 CH |
2157 | mutex_unlock(&lo->lo_mutex); |
2158 | ret = -EBUSY; | |
1c500ad7 | 2159 | goto mark_visible; |
f9d10764 | 2160 | } |
a0e286b6 | 2161 | /* Mark this loop device as no more bound, but not quite unbound yet */ |
f9d10764 CH |
2162 | lo->lo_state = Lo_deleting; |
2163 | mutex_unlock(&lo->lo_mutex); | |
2164 | ||
f9d10764 | 2165 | loop_remove(lo); |
1c500ad7 TH |
2166 | return 0; |
2167 | ||
2168 | mark_visible: | |
2169 | /* Show this loop device again. */ | |
2170 | mutex_lock(&loop_ctl_mutex); | |
2171 | lo->idr_visible = true; | |
f9d10764 CH |
2172 | mutex_unlock(&loop_ctl_mutex); |
2173 | return ret; | |
2174 | } | |
2175 | ||
2176 | static int loop_control_get_free(int idx) | |
770fe30a KS |
2177 | { |
2178 | struct loop_device *lo; | |
b9848081 | 2179 | int id, ret; |
770fe30a | 2180 | |
0a42e99b JK |
2181 | ret = mutex_lock_killable(&loop_ctl_mutex); |
2182 | if (ret) | |
2183 | return ret; | |
b9848081 | 2184 | idr_for_each_entry(&loop_index_idr, lo, id) { |
1c500ad7 TH |
2185 | /* Hitting a race results in creating a new loop device which is harmless. */ |
2186 | if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) | |
b9848081 CH |
2187 | goto found; |
2188 | } | |
f9d10764 | 2189 | mutex_unlock(&loop_ctl_mutex); |
18d1f200 | 2190 | return loop_add(-1); |
b9848081 CH |
2191 | found: |
2192 | mutex_unlock(&loop_ctl_mutex); | |
2193 | return id; | |
f9d10764 | 2194 | } |
0a42e99b | 2195 | |
f9d10764 CH |
2196 | static long loop_control_ioctl(struct file *file, unsigned int cmd, |
2197 | unsigned long parm) | |
2198 | { | |
770fe30a KS |
2199 | switch (cmd) { |
2200 | case LOOP_CTL_ADD: | |
18d1f200 | 2201 | return loop_add(parm); |
770fe30a | 2202 | case LOOP_CTL_REMOVE: |
f9d10764 | 2203 | return loop_control_remove(parm); |
770fe30a | 2204 | case LOOP_CTL_GET_FREE: |
f9d10764 CH |
2205 | return loop_control_get_free(parm); |
2206 | default: | |
2207 | return -ENOSYS; | |
770fe30a | 2208 | } |
770fe30a KS |
2209 | } |
2210 | ||
2211 | static const struct file_operations loop_ctl_fops = { | |
2212 | .open = nonseekable_open, | |
2213 | .unlocked_ioctl = loop_control_ioctl, | |
2214 | .compat_ioctl = loop_control_ioctl, | |
2215 | .owner = THIS_MODULE, | |
2216 | .llseek = noop_llseek, | |
2217 | }; | |
2218 | ||
2219 | static struct miscdevice loop_misc = { | |
2220 | .minor = LOOP_CTRL_MINOR, | |
2221 | .name = "loop-control", | |
2222 | .fops = &loop_ctl_fops, | |
2223 | }; | |
2224 | ||
2225 | MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); | |
2226 | MODULE_ALIAS("devname:loop-control"); | |
2227 | ||
73285082 KC |
2228 | static int __init loop_init(void) |
2229 | { | |
85c50197 | 2230 | int i; |
770fe30a | 2231 | int err; |
a47653fc | 2232 | |
476a4813 | 2233 | part_shift = 0; |
ac04fee0 | 2234 | if (max_part > 0) { |
476a4813 LV |
2235 | part_shift = fls(max_part); |
2236 | ||
ac04fee0 NK |
2237 | /* |
2238 | * Adjust max_part according to part_shift as it is exported | |
2239 | * to user space so that user can decide correct minor number | |
2240 | * if [s]he want to create more devices. | |
2241 | * | |
2242 | * Note that -1 is required because partition 0 is reserved | |
2243 | * for the whole disk. | |
2244 | */ | |
2245 | max_part = (1UL << part_shift) - 1; | |
2246 | } | |
2247 | ||
b1a66504 GC |
2248 | if ((1UL << part_shift) > DISK_MAX_PARTS) { |
2249 | err = -EINVAL; | |
a8c1d064 | 2250 | goto err_out; |
b1a66504 | 2251 | } |
78f4bb36 | 2252 | |
b1a66504 GC |
2253 | if (max_loop > 1UL << (MINORBITS - part_shift)) { |
2254 | err = -EINVAL; | |
a8c1d064 | 2255 | goto err_out; |
b1a66504 | 2256 | } |
1da177e4 | 2257 | |
a8c1d064 AV |
2258 | err = misc_register(&loop_misc); |
2259 | if (err < 0) | |
2260 | goto err_out; | |
2261 | ||
2262 | ||
8410d38c | 2263 | if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { |
b1a66504 GC |
2264 | err = -EIO; |
2265 | goto misc_out; | |
2266 | } | |
1da177e4 | 2267 | |
d134b00b | 2268 | /* pre-create number of devices given by config or max_loop */ |
85c50197 | 2269 | for (i = 0; i < max_loop; i++) |
d6da83d0 | 2270 | loop_add(i); |
34dd82af | 2271 | |
73285082 | 2272 | printk(KERN_INFO "loop: module loaded\n"); |
1da177e4 | 2273 | return 0; |
b1a66504 GC |
2274 | |
2275 | misc_out: | |
2276 | misc_deregister(&loop_misc); | |
a8c1d064 | 2277 | err_out: |
b1a66504 | 2278 | return err; |
34dd82af | 2279 | } |
a47653fc | 2280 | |
73285082 | 2281 | static void __exit loop_exit(void) |
1da177e4 | 2282 | { |
8e60947d CH |
2283 | struct loop_device *lo; |
2284 | int id; | |
2285 | ||
00d59405 | 2286 | unregister_blkdev(LOOP_MAJOR, "loop"); |
770fe30a | 2287 | misc_deregister(&loop_misc); |
200f9337 | 2288 | |
1c500ad7 TH |
2289 | /* |
2290 | * There is no need to use loop_ctl_mutex here, for nobody else can | |
2291 | * access loop_index_idr when this module is unloading (unless forced | |
2292 | * module unloading is requested). If this is not a clean unloading, | |
2293 | * we have no means to avoid kernel crash. | |
2294 | */ | |
8e60947d CH |
2295 | idr_for_each_entry(&loop_index_idr, lo, id) |
2296 | loop_remove(lo); | |
bd5c39ed CH |
2297 | |
2298 | idr_destroy(&loop_index_idr); | |
1da177e4 LT |
2299 | } |
2300 | ||
2301 | module_init(loop_init); | |
2302 | module_exit(loop_exit); | |
2303 | ||
2304 | #ifndef MODULE | |
2305 | static int __init max_loop_setup(char *str) | |
2306 | { | |
2307 | max_loop = simple_strtol(str, NULL, 0); | |
bb5faa99 MFO |
2308 | #ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD |
2309 | max_loop_specified = true; | |
2310 | #endif | |
1da177e4 LT |
2311 | return 1; |
2312 | } | |
2313 | ||
2314 | __setup("max_loop=", max_loop_setup); | |
2315 | #endif |