blob: 37511d2b2caf7df1c2067019a1decaca4edba00a [file] [log] [blame]
Christoph Hellwigf21e6e12022-04-19 08:33:01 +02001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
Christoph Hellwigf21e6e12022-04-19 08:33:01 +02003 * Copyright 1993 by Theodore Ts'o.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005#include <linux/module.h>
6#include <linux/moduleparam.h>
7#include <linux/sched.h>
8#include <linux/fs.h>
Matthew Wilcox (Oracle)4ee60ec2021-05-06 18:02:27 -07009#include <linux/pagemap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070010#include <linux/file.h>
11#include <linux/stat.h>
12#include <linux/errno.h>
13#include <linux/major.h>
14#include <linux/wait.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/blkpg.h>
16#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/swap.h>
18#include <linux/slab.h>
David Howells863d5b822006-08-29 19:06:14 +010019#include <linux/compat.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/suspend.h>
Rafael J. Wysocki83144182007-07-17 04:03:35 -070021#include <linux/freezer.h>
Arnd Bergmann2a48fc02010-06-02 14:28:52 +020022#include <linux/mutex.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/completion.h>
25#include <linux/highmem.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020026#include <linux/splice.h>
Milan Brozee862732010-08-23 15:16:00 +020027#include <linux/sysfs.h>
Kay Sievers770fe302011-07-31 22:08:04 +020028#include <linux/miscdevice.h>
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +020029#include <linux/falloc.h>
Al Viro283e7e52015-04-03 15:21:59 -040030#include <linux/uio.h>
Adam Manzanaresd9a08a92018-05-22 10:52:19 -070031#include <linux/ioprio.h>
Dennis Zhoudb6638d2018-12-05 12:10:35 -050032#include <linux/blk-cgroup.h>
Dan Schatzbergc74d40e2021-06-28 19:38:21 -070033#include <linux/sched/mm.h>
Ming Lei06582bc2022-01-26 11:58:30 +080034#include <linux/statfs.h>
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080035#include <linux/uaccess.h>
Christoph Hellwig754d9672022-04-19 08:33:00 +020036#include <linux/blk-mq.h>
37#include <linux/spinlock.h>
38#include <uapi/linux/loop.h>
39
40/* Possible states of device */
41enum {
42 Lo_unbound,
43 Lo_bound,
44 Lo_rundown,
45 Lo_deleting,
46};
47
48struct loop_func_table;
49
50struct 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};
80
81struct 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};
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Dan Schatzberg87579e92021-06-28 19:38:15 -070092#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
Bart Van Asschee152a052023-01-30 13:13:47 -080093#define LOOP_DEFAULT_HW_Q_DEPTH 128
Dan Schatzberg87579e92021-06-28 19:38:15 -070094
Kay Sievers34dd82a2011-07-31 22:08:04 +020095static DEFINE_IDR(loop_index_idr);
Tetsuo Handa310ca162018-11-08 14:01:02 +010096static DEFINE_MUTEX(loop_ctl_mutex);
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +090097static 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 */
111static 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 */
132static 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}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
Laurent Vivier476a4812008-03-26 12:11:53 +0100139static int max_part;
140static int part_shift;
141
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100142static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
Guo Chaob7a1da62013-02-21 15:16:50 -0800144 loff_t loopsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146 /* Compute loopsize in bytes */
Guo Chaob7a1da62013-02-21 15:16:50 -0800147 loopsize = i_size_read(file->f_mapping->host);
148 if (offset > 0)
149 loopsize -= offset;
150 /* offset is beyond i_size, weird but possible */
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100151 if (loopsize < 0)
152 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100154 if (sizelimit > 0 && sizelimit < loopsize)
155 loopsize = sizelimit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 /*
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
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +0100165 return get_size(lo->lo_offset, lo->lo_sizelimit, file);
166}
167
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800168static 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
Christoph Hellwig47e96242021-10-19 09:56:39 +0200185 * size of loop is bigger than the backing device's.
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800186 *
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
Masahiro Yamada89d790a2017-02-27 14:29:01 -0800189 * of requests in sane applications should be PAGE_SIZE aligned
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800190 */
191 if (dio) {
192 if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
NeilBrowna2ad63d2022-05-09 18:20:49 -0700193 !(lo->lo_offset & dio_align) &&
194 (file->f_mode & FMODE_CAN_ODIRECT))
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800195 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 */
Martijn Coenen0fbcf5792020-03-10 14:06:54 +0100213 if (lo->lo_state == Lo_bound)
214 blk_mq_freeze_queue(lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800215 lo->use_dio = use_dio;
Shaohua Li40326d82017-08-31 22:09:46 -0700216 if (use_dio) {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800217 blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800218 lo->lo_flags |= LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700219 } else {
Bart Van Assche8b904b52018-03-07 17:10:10 -0800220 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800221 lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
Shaohua Li40326d82017-08-31 22:09:46 -0700222 }
Martijn Coenen0fbcf5792020-03-10 14:06:54 +0100223 if (lo->lo_state == Lo_bound)
224 blk_mq_unfreeze_queue(lo->lo_queue);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800225}
226
Martijn Coenen5795b6f2020-05-13 15:38:37 +0200227/**
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 */
235static void loop_set_size(struct loop_device *lo, loff_t size)
236{
Christoph Hellwig449f4ec2020-11-16 15:56:56 +0100237 if (!set_capacity_and_notify(lo->lo_disk, size))
Christoph Hellwig3b4f85d2020-11-16 15:56:53 +0100238 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200241static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200243 struct iov_iter i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 ssize_t bw;
Al Viro283e7e52015-04-03 15:21:59 -0400245
Al Virode4eda92022-09-15 20:25:47 -0400246 iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Al Viro03d95eb2013-03-20 13:04:20 -0400248 file_start_write(file);
Christoph Hellwigabbb65892017-05-27 11:16:52 +0300249 bw = vfs_iter_write(file, &i, ppos, 0);
Al Viro03d95eb2013-03-20 13:04:20 -0400250 file_end_write(file);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200251
252 if (likely(bw == bvec->bv_len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 return 0;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200254
255 printk_ratelimited(KERN_ERR
256 "loop: Write error at byte offset %llu, length %i.\n",
257 (unsigned long long)*ppos, bvec->bv_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 if (bw >= 0)
259 bw = -EIO;
260 return bw;
261}
262
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200263static int lo_write_simple(struct loop_device *lo, struct request *rq,
264 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265{
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200266 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278}
279
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200280static int lo_read_simple(struct loop_device *lo, struct request *rq,
281 loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Kent Overstreet79886132013-11-23 17:19:00 -0800283 struct bio_vec bvec;
Ming Lei30112012014-12-31 13:22:58 +0000284 struct req_iterator iter;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200285 struct iov_iter i;
286 ssize_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287
Ming Lei30112012014-12-31 13:22:58 +0000288 rq_for_each_segment(bvec, rq, iter) {
Al Virode4eda92022-09-15 20:25:47 -0400289 iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len);
Christoph Hellwig18e97102017-05-27 11:16:51 +0300290 len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200291 if (len < 0)
292 return len;
Dave Young306df072012-02-08 22:07:19 +0100293
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200294 flush_dcache_page(bvec.bv_page);
295
296 if (len != bvec.bv_len) {
Ming Lei30112012014-12-31 13:22:58 +0000297 struct bio *bio;
298
299 __rq_for_each_bio(bio, rq)
300 zero_fill_bio(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 break;
Dave Young306df072012-02-08 22:07:19 +0100302 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200303 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 }
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200305
Dave Young306df072012-02-08 22:07:19 +0100306 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307}
308
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700309static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
310 int mode)
Ming Leicf655d92014-12-31 13:22:59 +0000311{
312 /*
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700313 * We use fallocate to manipulate the space mappings used by the image
Christoph Hellwig47e96242021-10-19 09:56:39 +0200314 * a.k.a. discard/zerorange.
Ming Leicf655d92014-12-31 13:22:59 +0000315 */
316 struct file *file = lo->lo_backing_file;
Ming Leicf655d92014-12-31 13:22:59 +0000317 int ret;
318
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700319 mode |= FALLOC_FL_KEEP_SIZE;
320
Christoph Hellwig70200572022-04-15 06:52:55 +0200321 if (!bdev_max_discard_sectors(lo->lo_device))
322 return -EOPNOTSUPP;
Ming Leicf655d92014-12-31 13:22:59 +0000323
324 ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
325 if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
Christoph Hellwig70200572022-04-15 06:52:55 +0200326 return -EIO;
Ming Leicf655d92014-12-31 13:22:59 +0000327 return ret;
328}
329
330static int lo_req_flush(struct loop_device *lo, struct request *rq)
331{
Chaitanya Kulkarni9c64e382022-02-15 13:33:09 -0800332 int ret = vfs_fsync(lo->lo_backing_file, 0);
Ming Leicf655d92014-12-31 13:22:59 +0000333 if (unlikely(ret && ret != -EINVAL))
334 ret = -EIO;
335
336 return ret;
337}
338
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200339static void lo_complete_rq(struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800340{
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200341 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Jens Axboef9de14b2018-04-13 16:25:57 -0600342 blk_status_t ret = BLK_STS_OK;
Ming Leibc07c102015-08-17 10:31:51 +0800343
Jens Axboef9de14b2018-04-13 16:25:57 -0600344 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)
Evan Green8cd55082020-04-03 16:43:03 +0200347 ret = errno_to_blk_status(cmd->ret);
Jens Axboef9de14b2018-04-13 16:25:57 -0600348 goto end_io;
Ming Leibc07c102015-08-17 10:31:51 +0800349 }
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200350
Jens Axboef9de14b2018-04-13 16:25:57 -0600351 /*
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;
369end_io:
370 blk_mq_end_request(rq, ret);
371 }
Ming Leibc07c102015-08-17 10:31:51 +0800372}
373
Shaohua Li92d77332017-09-01 11:15:17 -0700374static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
375{
Jens Axboe1894e9162018-04-13 16:24:29 -0600376 struct request *rq = blk_mq_rq_from_pdu(cmd);
377
Shaohua Li92d77332017-09-01 11:15:17 -0700378 if (!atomic_dec_and_test(&cmd->ref))
379 return;
380 kfree(cmd->bvec);
381 cmd->bvec = NULL;
Christoph Hellwig15f73f52020-06-11 08:44:47 +0200382 if (likely(!blk_should_fake_timeout(rq->q)))
383 blk_mq_complete_request(rq);
Shaohua Li92d77332017-09-01 11:15:17 -0700384}
385
Jens Axboe6b19b762021-10-21 09:22:35 -0600386static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
Ming Leibc07c102015-08-17 10:31:51 +0800387{
388 struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
Ming Leibc07c102015-08-17 10:31:51 +0800389
Christoph Hellwigfe2cb292017-04-20 16:03:02 +0200390 cmd->ret = ret;
Shaohua Li92d77332017-09-01 11:15:17 -0700391 lo_rw_aio_do_completion(cmd);
Ming Leibc07c102015-08-17 10:31:51 +0800392}
393
394static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
Al Virode4eda92022-09-15 20:25:47 -0400395 loff_t pos, int rw)
Ming Leibc07c102015-08-17 10:31:51 +0800396{
397 struct iov_iter iter;
Ming Lei86af5952019-02-15 19:13:17 +0800398 struct req_iterator rq_iter;
Ming Leibc07c102015-08-17 10:31:51 +0800399 struct bio_vec *bvec;
Jens Axboe1894e9162018-04-13 16:24:29 -0600400 struct request *rq = blk_mq_rq_from_pdu(cmd);
Shaohua Li40326d82017-08-31 22:09:46 -0700401 struct bio *bio = rq->bio;
Ming Leibc07c102015-08-17 10:31:51 +0800402 struct file *file = lo->lo_backing_file;
Ming Lei86af5952019-02-15 19:13:17 +0800403 struct bio_vec tmp;
Shaohua Li40326d82017-08-31 22:09:46 -0700404 unsigned int offset;
Ming Lei86af5952019-02-15 19:13:17 +0800405 int nr_bvec = 0;
Ming Leibc07c102015-08-17 10:31:51 +0800406 int ret;
407
Ming Lei86af5952019-02-15 19:13:17 +0800408 rq_for_each_bvec(tmp, rq, rq_iter)
409 nr_bvec++;
Ming Leibc07c102015-08-17 10:31:51 +0800410
Ming Lei86af5952019-02-15 19:13:17 +0800411 if (rq->bio != rq->biotail) {
412
413 bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
Kees Cook6da2ec52018-06-12 13:55:00 -0700414 GFP_NOIO);
Shaohua Li40326d82017-08-31 22:09:46 -0700415 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
Ming Lei86af5952019-02-15 19:13:17 +0800422 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
Shaohua Li40326d82017-08-31 22:09:46 -0700423 * API will take care of all details for us.
424 */
Ming Lei86af5952019-02-15 19:13:17 +0800425 rq_for_each_bvec(tmp, rq, rq_iter) {
Shaohua Li40326d82017-08-31 22:09:46 -0700426 *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);
Shaohua Li40326d82017-08-31 22:09:46 -0700439 }
Shaohua Li92d77332017-09-01 11:15:17 -0700440 atomic_set(&cmd->ref, 2);
Shaohua Li40326d82017-08-31 22:09:46 -0700441
Christoph Hellwigb6207432019-06-26 15:49:28 +0200442 iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
Shaohua Li40326d82017-08-31 22:09:46 -0700443 iter.iov_offset = offset;
Ming Leibc07c102015-08-17 10:31:51 +0800444
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;
Adam Manzanaresd9a08a92018-05-22 10:52:19 -0700449 cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
Ming Leibc07c102015-08-17 10:31:51 +0800450
Al Virode4eda92022-09-15 20:25:47 -0400451 if (rw == ITER_SOURCE)
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100452 ret = call_write_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800453 else
Miklos Szeredibb7462b2017-02-20 16:51:23 +0100454 ret = call_read_iter(file, &cmd->iocb, &iter);
Ming Leibc07c102015-08-17 10:31:51 +0800455
Shaohua Li92d77332017-09-01 11:15:17 -0700456 lo_rw_aio_do_completion(cmd);
457
Ming Leibc07c102015-08-17 10:31:51 +0800458 if (ret != -EIOCBQUEUED)
Jens Axboe6b19b762021-10-21 09:22:35 -0600459 lo_rw_aio_complete(&cmd->iocb, ret);
Ming Leibc07c102015-08-17 10:31:51 +0800460 return 0;
461}
462
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200463static int do_req_filebacked(struct loop_device *lo, struct request *rq)
Ming Leibc07c102015-08-17 10:31:51 +0800464{
465 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200466 loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
Ming Leibc07c102015-08-17 10:31:51 +0800467
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 */
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200477 switch (req_op(rq)) {
478 case REQ_OP_FLUSH:
479 return lo_req_flush(lo, rq);
Christoph Hellwig19372e22017-04-05 19:21:15 +0200480 case REQ_OP_WRITE_ZEROES:
Darrick J. Wongefcfec572019-10-30 20:29:48 -0700481 /*
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);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200491 case REQ_OP_WRITE:
Christoph Hellwig47e96242021-10-19 09:56:39 +0200492 if (cmd->use_aio)
Al Virode4eda92022-09-15 20:25:47 -0400493 return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200494 else
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200495 return lo_write_simple(lo, rq, pos);
496 case REQ_OP_READ:
Christoph Hellwig47e96242021-10-19 09:56:39 +0200497 if (cmd->use_aio)
Al Virode4eda92022-09-15 20:25:47 -0400498 return lo_rw_aio(lo, cmd, pos, ITER_DEST);
Christoph Hellwigc1c87c22016-08-04 16:10:01 +0200499 else
500 return lo_read_simple(lo, rq, pos);
501 default:
502 WARN_ON_ONCE(1);
503 return -EIO;
Christoph Hellwigaa4d8612015-04-07 18:23:29 +0200504 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505}
506
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800507static inline void loop_update_dio(struct loop_device *lo)
508{
Ira Weinyefbe3c22020-04-30 07:41:33 -0700509 __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
510 lo->use_dio);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800511}
512
Christoph Hellwig03842642021-06-24 14:32:40 +0200513static void loop_reread_partitions(struct loop_device *lo)
Ming Lei06f0e9e2015-05-06 12:26:24 +0800514{
515 int rc;
516
Christoph Hellwig03842642021-06-24 14:32:40 +0200517 mutex_lock(&lo->lo_disk->open_mutex);
518 rc = bdev_disk_changed(lo->lo_disk, false);
519 mutex_unlock(&lo->lo_disk->open_mutex);
Ming Lei06f0e9e2015-05-06 12:26:24 +0800520 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
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400525static inline int is_loop_device(struct file *file)
526{
527 struct inode *i = file->f_mapping->host;
528
Al Viro6f247842021-01-31 19:23:55 -0500529 return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400530}
531
532static 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
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900541 lockdep_assert_held(&loop_validate_mutex);
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100542 if (f->f_mapping->host->i_rdev == bdev->bd_dev)
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400543 return -EBADF;
544
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100545 l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900546 if (l->lo_state != Lo_bound)
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400547 return -EINVAL;
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900548 /* Order wrt setting lo->lo_backing_file in loop_configure(). */
549 rmb();
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400550 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557/*
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 */
Al Virobb214882008-03-02 09:29:48 -0500565static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
566 unsigned int arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567{
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900568 struct file *file = fget(arg);
569 struct file *old_file;
570 int error;
571 bool partscan;
572 bool is_loop;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900574 if (!file)
575 return -EBADF;
Christoph Hellwig498ef5c2022-03-30 07:29:14 +0200576
577 /* suppress uevents while reconfiguring the device */
578 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
579
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900580 is_loop = is_loop_device(file);
581 error = loop_global_lock_killable(lo, is_loop);
Jan Karac3710772018-11-08 14:01:11 +0100582 if (error)
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900583 goto out_putf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 error = -ENXIO;
585 if (lo->lo_state != Lo_bound)
Jan Kara1dded9a2018-11-08 14:01:15 +0100586 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587
588 /* the loop device has to be read-only */
589 error = -EINVAL;
590 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
Jan Kara1dded9a2018-11-08 14:01:15 +0100591 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400593 error = loop_validate_file(file, bdev);
594 if (error)
Jan Kara1dded9a2018-11-08 14:01:15 +0100595 goto out_err;
Theodore Ts'od2ac8382018-05-07 11:37:58 -0400596
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 old_file = lo->lo_backing_file;
598
599 error = -EINVAL;
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 /* size of the new backing store needs to be the same */
602 if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
Jan Kara1dded9a2018-11-08 14:01:15 +0100603 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
605 /* and ... switch */
Matteo Croce9f65c482021-07-13 01:05:30 +0200606 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
Omar Sandoval43cade802017-08-24 00:03:44 -0700607 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);
Jan Kara85b0a542018-11-08 14:01:13 +0100615 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900616 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 }
Jan Kara1dded9a2018-11-08 14:01:15 +0100626 /*
Pavel Tatashin6cc8e742021-01-26 09:46:30 -0500627 * We must drop file reference outside of lo_mutex as dropping
Christoph Hellwiga8698702021-05-25 08:12:56 +0200628 * the file ref can take open_mutex which creates circular locking
Jan Kara1dded9a2018-11-08 14:01:15 +0100629 * dependency.
630 */
631 fput(old_file);
Jan Kara85b0a542018-11-08 14:01:13 +0100632 if (partscan)
Christoph Hellwig03842642021-06-24 14:32:40 +0200633 loop_reread_partitions(lo);
Christoph Hellwig498ef5c2022-03-30 07:29:14 +0200634
635 error = 0;
636done:
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;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Jan Kara1dded9a2018-11-08 14:01:15 +0100641out_err:
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900642 loop_global_unlock(lo, is_loop);
643out_putf:
644 fput(file);
Christoph Hellwig498ef5c2022-03-30 07:29:14 +0200645 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646}
647
Milan Brozee862732010-08-23 15:16:00 +0200648/* loop sysfs attributes */
649
650static ssize_t loop_attr_show(struct device *dev, char *page,
651 ssize_t (*callback)(struct loop_device *, char *))
652{
Kay Sievers34dd82a2011-07-31 22:08:04 +0200653 struct gendisk *disk = dev_to_disk(dev);
654 struct loop_device *lo = disk->private_data;
Milan Brozee862732010-08-23 15:16:00 +0200655
Kay Sievers34dd82a2011-07-31 22:08:04 +0200656 return callback(lo, page);
Milan Brozee862732010-08-23 15:16:00 +0200657}
658
659#define LOOP_ATTR_RO(_name) \
660static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \
661static 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} \
666static struct device_attribute loop_attr_##_name = \
Joe Perches5657a812018-05-24 13:38:59 -0600667 __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
Milan Brozee862732010-08-23 15:16:00 +0200668
669static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
670{
671 ssize_t ret;
672 char *p = NULL;
673
Kay Sievers05eb0f22011-07-31 22:21:35 +0200674 spin_lock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200675 if (lo->lo_backing_file)
Miklos Szeredi9bf39ab2015-06-19 10:29:13 +0200676 p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
Kay Sievers05eb0f22011-07-31 22:21:35 +0200677 spin_unlock_irq(&lo->lo_lock);
Milan Brozee862732010-08-23 15:16:00 +0200678
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
691static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
692{
Chaitanya Kulkarnib27824d2022-02-15 13:33:07 -0800693 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
Milan Brozee862732010-08-23 15:16:00 +0200694}
695
696static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
697{
Chaitanya Kulkarnib27824d2022-02-15 13:33:07 -0800698 return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
Milan Brozee862732010-08-23 15:16:00 +0200699}
700
701static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
702{
703 int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
704
Chaitanya Kulkarnib27824d2022-02-15 13:33:07 -0800705 return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
Milan Brozee862732010-08-23 15:16:00 +0200706}
707
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200708static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
709{
710 int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
711
Chaitanya Kulkarnib27824d2022-02-15 13:33:07 -0800712 return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200713}
714
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800715static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
716{
717 int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
718
Chaitanya Kulkarnib27824d2022-02-15 13:33:07 -0800719 return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800720}
721
Milan Brozee862732010-08-23 15:16:00 +0200722LOOP_ATTR_RO(backing_file);
723LOOP_ATTR_RO(offset);
724LOOP_ATTR_RO(sizelimit);
725LOOP_ATTR_RO(autoclear);
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200726LOOP_ATTR_RO(partscan);
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800727LOOP_ATTR_RO(dio);
Milan Brozee862732010-08-23 15:16:00 +0200728
729static 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,
Kay Sieverse03c8dd2011-08-23 20:12:04 +0200734 &loop_attr_partscan.attr,
Ming Lei2e5ab5f32015-08-17 10:31:49 +0800735 &loop_attr_dio.attr,
Milan Brozee862732010-08-23 15:16:00 +0200736 NULL,
737};
738
739static struct attribute_group loop_attribute_group = {
740 .name = "loop",
741 .attrs= loop_attrs,
742};
743
Tetsuo Handad3349b62018-05-04 10:58:09 -0600744static void loop_sysfs_init(struct loop_device *lo)
Milan Brozee862732010-08-23 15:16:00 +0200745{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600746 lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
747 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200748}
749
750static void loop_sysfs_exit(struct loop_device *lo)
751{
Tetsuo Handad3349b62018-05-04 10:58:09 -0600752 if (lo->sysfs_inited)
753 sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
754 &loop_attribute_group);
Milan Brozee862732010-08-23 15:16:00 +0200755}
756
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200757static 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;
Ming Leibcb21c82020-08-17 18:01:30 +0800762 u32 granularity, max_discard_sectors;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200763
764 /*
Evan Greenc52abf52020-04-03 16:43:04 +0200765 * 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 */
Christoph Hellwig47e96242021-10-19 09:56:39 +0200771 if (S_ISBLK(inode->i_mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +0100772 struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
Evan Greenc52abf52020-04-03 16:43:04 +0200773
Ming Leibcb21c82020-08-17 18:01:30 +0800774 max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
Christoph Hellwig7b47ef52022-04-15 06:52:56 +0200775 granularity = bdev_discard_granularity(I_BDEV(inode)) ?:
Ming Leibcb21c82020-08-17 18:01:30 +0800776 queue_physical_block_size(backingq);
Evan Greenc52abf52020-04-03 16:43:04 +0200777
778 /*
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200779 * We use punch hole to reclaim the free space used by the
Christoph Hellwig47e96242021-10-19 09:56:39 +0200780 * image a.k.a. discard.
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200781 */
Christoph Hellwig47e96242021-10-19 09:56:39 +0200782 } else if (!file->f_op->fallocate) {
Ming Leibcb21c82020-08-17 18:01:30 +0800783 max_discard_sectors = 0;
784 granularity = 0;
Evan Greenc52abf52020-04-03 16:43:04 +0200785
786 } else {
Ming Lei06582bc2022-01-26 11:58:30 +0800787 struct kstatfs sbuf;
788
Ming Leibcb21c82020-08-17 18:01:30 +0800789 max_discard_sectors = UINT_MAX >> 9;
Ming Lei06582bc2022-01-26 11:58:30 +0800790 if (!vfs_statfs(&file->f_path, &sbuf))
791 granularity = sbuf.f_bsize;
792 else
793 max_discard_sectors = 0;
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200794 }
795
Ming Leibcb21c82020-08-17 18:01:30 +0800796 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);
Ming Leibcb21c82020-08-17 18:01:30 +0800800 } else {
801 q->limits.discard_granularity = 0;
802 blk_queue_max_discard_sectors(q, 0);
803 blk_queue_max_write_zeroes_sectors(q, 0);
Ming Leibcb21c82020-08-17 18:01:30 +0800804 }
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +0200805}
806
Dan Schatzberg87579e92021-06-28 19:38:15 -0700807struct 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;
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700813 struct cgroup_subsys_state *blkcg_css;
Dan Schatzberg87579e92021-06-28 19:38:15 -0700814 unsigned long last_ran_at;
815};
Ming Leie03a3d72015-08-17 10:31:48 +0800816
Dan Schatzberg87579e92021-06-28 19:38:15 -0700817static void loop_workfn(struct work_struct *work);
NeilBrownb2ee7d42017-06-16 15:02:09 +1000818
Dan Schatzberg87579e92021-06-28 19:38:15 -0700819#ifdef CONFIG_BLK_CGROUP
820static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
Ming Leie03a3d72015-08-17 10:31:48 +0800821{
Dan Schatzberg87579e92021-06-28 19:38:15 -0700822 return !css || css == blkcg_root_css;
823}
824#else
825static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
826{
827 return !css;
828}
829#endif
830
831static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
832{
Colin Ian King413ec802022-01-13 00:14:32 +0000833 struct rb_node **node, *parent = NULL;
Dan Schatzberg87579e92021-06-28 19:38:15 -0700834 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
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700840 if (queue_on_root_worker(cmd->blkcg_css))
Dan Schatzberg87579e92021-06-28 19:38:15 -0700841 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);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700848 if (cur_worker->blkcg_css == cmd->blkcg_css) {
Dan Schatzberg87579e92021-06-28 19:38:15 -0700849 worker = cur_worker;
850 break;
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700851 } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
Dan Schatzberg87579e92021-06-28 19:38:15 -0700852 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
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700863 * rootcg worker and issue the I/O as the rootcg
Dan Schatzberg87579e92021-06-28 19:38:15 -0700864 */
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700865 if (!worker) {
866 cmd->blkcg_css = NULL;
867 if (cmd->memcg_css)
868 css_put(cmd->memcg_css);
869 cmd->memcg_css = NULL;
Dan Schatzberg87579e92021-06-28 19:38:15 -0700870 goto queue_work;
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700871 }
Dan Schatzberg87579e92021-06-28 19:38:15 -0700872
Dan Schatzbergc74d40e2021-06-28 19:38:21 -0700873 worker->blkcg_css = cmd->blkcg_css;
874 css_get(worker->blkcg_css);
Dan Schatzberg87579e92021-06-28 19:38:15 -0700875 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);
881queue_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);
Ming Leie03a3d72015-08-17 10:31:48 +0800899}
900
Christoph Hellwig2cf429b2022-03-30 07:29:08 +0200901static void loop_set_timer(struct loop_device *lo)
902{
903 timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
904}
905
906static 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
927static 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
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -0700934static 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)
Christoph Hellwig10f0d2a52022-04-15 06:52:42 +0200944 nonrot = bdev_nonrot(file_bdev);
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -0700945
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
Martijn Coenen62ab4662020-05-13 15:38:42 +0200952/**
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 */
960static int
961loop_set_status_from_info(struct loop_device *lo,
962 const struct loop_info64 *info)
963{
Martijn Coenen62ab4662020-05-13 15:38:42 +0200964 if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
965 return -EINVAL;
966
Christoph Hellwig47e96242021-10-19 09:56:39 +0200967 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 }
Martijn Coenen62ab4662020-05-13 15:38:42 +0200979
Zhong Jinghua9f6ad5d2023-02-21 17:50:27 +0800980 /* Avoid assigning overflow values */
981 if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
982 return -EOVERFLOW;
983
Martijn Coenen62ab4662020-05-13 15:38:42 +0200984 lo->lo_offset = info->lo_offset;
985 lo->lo_sizelimit = info->lo_sizelimit;
Siddh Raman Pantc490a0b2022-08-23 21:38:10 +0530986
Martijn Coenen62ab4662020-05-13 15:38:42 +0200987 memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
Martijn Coenen62ab4662020-05-13 15:38:42 +0200988 lo->lo_file_name[LO_NAME_SIZE-1] = 0;
Martijn Coenenfaf1d252020-05-13 15:38:44 +0200989 lo->lo_flags = info->lo_flags;
Martijn Coenen62ab4662020-05-13 15:38:42 +0200990 return 0;
991}
992
Christoph Hellwig05bdb992023-06-08 13:02:55 +0200993static int loop_configure(struct loop_device *lo, blk_mode_t mode,
Martijn Coenen34489142020-05-13 15:38:45 +0200994 struct block_device *bdev,
995 const struct loop_config *config)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +0900997 struct file *file = fget(config->fd);
998 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 struct address_space *mapping;
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +09001000 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 /* This is safe, since we have a reference from open(). */
1011 __module_get(THIS_MODULE);
1012
Jan Kara33ec3e532019-05-16 16:01:27 +02001013 /*
1014 * If we don't hold exclusive handle for the device, upgrade to it
1015 * here to avoid changing device under exclusive owner.
1016 */
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001017 if (!(mode & BLK_OPEN_EXCL)) {
Christoph Hellwig0718afd2023-06-01 11:44:52 +02001018 error = bd_prepare_to_claim(bdev, loop_configure, NULL);
Christoph Hellwigecbe6bc2020-07-16 16:33:09 +02001019 if (error)
Jan Kara33ec3e532019-05-16 16:01:27 +02001020 goto out_putf;
1021 }
1022
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +09001023 error = loop_global_lock_killable(lo, is_loop);
Jan Kara757ecf42018-11-08 14:01:10 +01001024 if (error)
Jan Kara33ec3e532019-05-16 16:01:27 +02001025 goto out_bdev;
Jan Kara757ecf42018-11-08 14:01:10 +01001026
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 error = -EBUSY;
1028 if (lo->lo_state != Lo_unbound)
Jan Kara757ecf42018-11-08 14:01:10 +01001029 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Theodore Ts'od2ac8382018-05-07 11:37:58 -04001031 error = loop_validate_file(file, bdev);
1032 if (error)
Jan Kara757ecf42018-11-08 14:01:10 +01001033 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035 mapping = file->f_mapping;
1036 inode = mapping->host;
1037
Martijn Coenen34489142020-05-13 15:38:45 +02001038 if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1039 error = -EINVAL;
1040 goto out_unlock;
1041 }
1042
1043 if (config->block_size) {
Xie Yongjiaf3c5702021-10-26 22:40:14 +08001044 error = blk_validate_block_size(config->block_size);
Martijn Coenen34489142020-05-13 15:38:45 +02001045 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
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001053 if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
Al Viro283e7e52015-04-03 15:21:59 -04001054 !file->f_op->write_iter)
Martijn Coenen34489142020-05-13 15:38:45 +02001055 lo->lo_flags |= LO_FLAGS_READ_ONLY;
Martijn Coenen083a6a52020-05-13 15:38:36 +02001056
Dan Schatzberg87579e92021-06-28 19:38:15 -07001057 if (!lo->workqueue) {
Christoph Hellwigd292dc82022-03-30 07:29:17 +02001058 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 }
Dan Schatzberg87579e92021-06-28 19:38:15 -07001065 }
Christoph Hellwig456be142011-10-17 12:57:20 +02001066
Alyssa Rossbb430b62023-03-20 13:54:30 +01001067 /* suppress uevents while reconfiguring the device */
1068 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1069
Matteo Croce9f65c482021-07-13 01:05:30 +02001070 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
Christoph Hellwig7a2f0ce2020-11-03 11:00:16 +01001071 set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
Martijn Coenen34489142020-05-13 15:38:45 +02001073 lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 lo->lo_device = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 lo->lo_backing_file = file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 lo->old_gfp_mask = mapping_gfp_mask(mapping);
1077 mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1078
Martijn Coenen34489142020-05-13 15:38:45 +02001079 if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
Jens Axboe21d07272016-03-30 10:09:35 -06001080 blk_queue_write_cache(lo->lo_queue, true, false);
Nikanth Karthikesan68db1962009-03-24 12:29:54 +01001081
Martijn Coenen34489142020-05-13 15:38:45 +02001082 if (config->block_size)
1083 bsize = config->block_size;
Linus Torvalds96ed3202020-06-02 19:45:12 -07001084 else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
Martijn Coenen85560112019-09-04 21:49:01 +02001085 /* In case of direct I/O, match underlying block size */
Martijn Coenen34489142020-05-13 15:38:45 +02001086 bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1087 else
1088 bsize = 512;
Martijn Coenen85560112019-09-04 21:49:01 +02001089
Martijn Coenen34489142020-05-13 15:38:45 +02001090 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);
Martijn Coenen85560112019-09-04 21:49:01 +02001093
Kristian Klausen2b9ac222021-06-18 13:51:57 +02001094 loop_config_discard(lo);
Holger Hoffstätte56a85fd2019-02-12 15:54:24 -07001095 loop_update_rotational(lo);
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001096 loop_update_dio(lo);
Milan Brozee862732010-08-23 15:16:00 +02001097 loop_sysfs_init(lo);
Martijn Coenen79e5dc52020-08-25 09:18:29 +02001098
1099 size = get_loop_size(lo, file);
Martijn Coenen5795b6f2020-05-13 15:38:37 +02001100 loop_set_size(lo, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +09001102 /* Order wrt reading lo_state in loop_validate_file(). */
1103 wmb();
1104
Serge E. Hallyn6c997912006-09-29 01:59:11 -07001105 lo->lo_state = Lo_bound;
Kay Sieverse03c8dd2011-08-23 20:12:04 +02001106 if (part_shift)
1107 lo->lo_flags |= LO_FLAGS_PARTSCAN;
Jan Kara85b0a542018-11-08 14:01:13 +01001108 partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
Lennart Poetteringfe6a8fc2020-08-10 19:16:32 +02001109 if (partscan)
Christoph Hellwigb9684a72022-05-27 07:58:06 +02001110 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
Anatol Pomozovc1681bf2013-04-01 09:47:56 -07001111
Alyssa Rossbb430b62023-03-20 13:54:30 +01001112 /* enable and uncork uevent now that we are done */
1113 dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1114
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +09001115 loop_global_unlock(lo, is_loop);
Jan Kara85b0a542018-11-08 14:01:13 +01001116 if (partscan)
Christoph Hellwig03842642021-06-24 14:32:40 +02001117 loop_reread_partitions(lo);
Alyssa Rossbb430b62023-03-20 13:54:30 +01001118
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001119 if (!(mode & BLK_OPEN_EXCL))
Christoph Hellwig37c3fc92020-11-25 21:20:08 +01001120 bd_abort_claiming(bdev, loop_configure);
Christoph Hellwig498ef5c2022-03-30 07:29:14 +02001121
Alyssa Rossbb430b62023-03-20 13:54:30 +01001122 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123
Jan Kara757ecf42018-11-08 14:01:10 +01001124out_unlock:
Tetsuo Handa3ce6e1f2021-07-06 23:40:34 +09001125 loop_global_unlock(lo, is_loop);
Jan Kara33ec3e532019-05-16 16:01:27 +02001126out_bdev:
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001127 if (!(mode & BLK_OPEN_EXCL))
Christoph Hellwig37c3fc92020-11-25 21:20:08 +01001128 bd_abort_claiming(bdev, loop_configure);
Jan Kara757ecf42018-11-08 14:01:10 +01001129out_putf:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 fput(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 /* This is safe: open() is still holding a reference. */
1132 module_put(THIS_MODULE);
Alyssa Rossbb430b62023-03-20 13:54:30 +01001133 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134}
1135
Tetsuo Handabf237472022-02-11 16:15:54 +09001136static void __loop_clr_fd(struct loop_device *lo, bool release)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001138 struct file *filp;
Al Virob4e3ca12005-10-21 03:22:34 -04001139 gfp_t gfp = lo->old_gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Mauricio Faria de Oliveira4ceddce2021-02-22 12:41:23 -03001141 if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1142 blk_queue_write_cache(lo->lo_queue, false, false);
1143
Christoph Hellwig1fe0b1a2022-03-30 07:29:12 +02001144 /*
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);
Dan Schatzberg87579e92021-06-28 19:38:15 -07001151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 spin_lock_irq(&lo->lo_lock);
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001153 filp = lo->lo_backing_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 lo->lo_backing_file = NULL;
Kay Sievers05eb0f22011-07-31 22:21:35 +02001155 spin_unlock_irq(&lo->lo_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 lo->lo_device = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 lo->lo_offset = 0;
1159 lo->lo_sizelimit = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 memset(lo->lo_file_name, 0, LO_NAME_SIZE);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001161 blk_queue_logical_block_size(lo->lo_queue, 512);
Omar Sandovalbf093752017-09-05 14:24:47 -07001162 blk_queue_physical_block_size(lo->lo_queue, 512);
1163 blk_queue_io_min(lo->lo_queue, 512);
Xie Yongjie515be8f32021-09-22 20:37:09 +08001164 invalidate_disk(lo->lo_disk);
Milan Broz51a0bb02010-10-27 19:51:30 -06001165 loop_sysfs_exit(lo);
Xie Yongji19f553d2021-09-22 20:37:10 +08001166 /* let user-space know about this change */
1167 kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 mapping_set_gfp_mask(filp->f_mapping, gfp);
Tetsuo Handabf237472022-02-11 16:15:54 +09001169 /* This is safe: open() is still holding a reference. */
1170 module_put(THIS_MODULE);
Christoph Hellwig1fe0b1a2022-03-30 07:29:12 +02001171 if (!release)
1172 blk_mq_unfreeze_queue(lo->lo_queue);
Ming Leif8933662015-05-06 12:26:23 +08001173
Matteo Croce9f65c482021-07-13 01:05:30 +02001174 disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001175
1176 if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1177 int err;
1178
Tetsuo Handabf237472022-02-11 16:15:54 +09001179 /*
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);
Christoph Hellwig03842642021-06-24 14:32:40 +02001189 err = bdev_disk_changed(lo->lo_disk, false);
Tetsuo Handabf237472022-02-11 16:15:54 +09001190 if (!release)
1191 mutex_unlock(&lo->lo_disk->open_mutex);
Dongli Zhang40853d62019-02-22 22:10:19 +08001192 if (err)
1193 pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001194 __func__, lo->lo_number, err);
Jan Karad57f3372018-11-08 14:01:12 +01001195 /* Device is gone, no point in returning error */
Jan Karad57f3372018-11-08 14:01:12 +01001196 }
Dongli Zhang758a58d2019-02-22 22:10:20 +08001197
Tetsuo Handabf237472022-02-11 16:15:54 +09001198 /*
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 */
Dongli Zhang758a58d2019-02-22 22:10:20 +08001204 lo->lo_flags = 0;
1205 if (!part_shift)
Christoph Hellwigb9684a72022-05-27 07:58:06 +02001206 set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001207 mutex_lock(&lo->lo_mutex);
Dongli Zhang758a58d2019-02-22 22:10:20 +08001208 lo->lo_state = Lo_unbound;
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001209 mutex_unlock(&lo->lo_mutex);
Dongli Zhang758a58d2019-02-22 22:10:20 +08001210
Tetsuo Handabf237472022-02-11 16:15:54 +09001211 /*
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
Jan Karaa2505b72018-11-08 14:01:06 +01001219static int loop_clr_fd(struct loop_device *lo)
1220{
Jan Kara7ccd0792018-11-08 14:01:07 +01001221 int err;
1222
Tetsuo Handa158eaeb2022-03-30 07:29:15 +02001223 /*
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);
Jan Kara7ccd0792018-11-08 14:01:07 +01001233 if (err)
1234 return err;
1235 if (lo->lo_state != Lo_bound) {
Tetsuo Handa158eaeb2022-03-30 07:29:15 +02001236 loop_global_unlock(lo, true);
Jan Karaa2505b72018-11-08 14:01:06 +01001237 return -ENXIO;
Jan Kara7ccd0792018-11-08 14:01:07 +01001238 }
Jan Karaa2505b72018-11-08 14:01:06 +01001239 /*
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 */
Christoph Hellwiga0e286b2022-03-30 07:29:16 +02001249 if (disk_openers(lo->lo_disk) > 1) {
Jan Karaa2505b72018-11-08 14:01:06 +01001250 lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
Tetsuo Handa158eaeb2022-03-30 07:29:15 +02001251 loop_global_unlock(lo, true);
Jan Karaa2505b72018-11-08 14:01:06 +01001252 return 0;
1253 }
1254 lo->lo_state = Lo_rundown;
Tetsuo Handa158eaeb2022-03-30 07:29:15 +02001255 loop_global_unlock(lo, true);
Jan Karaa2505b72018-11-08 14:01:06 +01001256
Tetsuo Handabf237472022-02-11 16:15:54 +09001257 __loop_clr_fd(lo, false);
Tetsuo Handa6050fa42021-11-24 19:47:40 +09001258 return 0;
Jan Karaa2505b72018-11-08 14:01:06 +01001259}
1260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261static int
1262loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1263{
1264 int err;
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001265 int prev_lo_flags;
Jan Kara85b0a542018-11-08 14:01:13 +01001266 bool partscan = false;
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001267 bool size_changed = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001269 err = mutex_lock_killable(&lo->lo_mutex);
Jan Kara550df5fd2018-11-08 14:01:09 +01001270 if (err)
1271 return err;
Jan Kara550df5fd2018-11-08 14:01:09 +01001272 if (lo->lo_state != Lo_bound) {
1273 err = -ENXIO;
1274 goto out_unlock;
1275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001277 if (lo->lo_offset != info->lo_offset ||
1278 lo->lo_sizelimit != info->lo_sizelimit) {
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001279 size_changed = true;
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001280 sync_blockdev(lo->lo_device);
Zheng Binf4bd34b2020-06-18 12:21:37 +08001281 invalidate_bdev(lo->lo_device);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001282 }
1283
Ming Leiecdd0952017-02-11 11:40:45 +08001284 /* I/O need to be drained during transfer transition */
1285 blk_mq_freeze_queue(lo->lo_queue);
1286
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001287 prev_lo_flags = lo->lo_flags;
1288
Martijn Coenen0c3796c2020-05-13 15:38:41 +02001289 err = loop_set_status_from_info(lo, info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 if (err)
Jan Kara550df5fd2018-11-08 14:01:09 +01001291 goto out_unfreeze;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001293 /* Mask out flags that can't be set using LOOP_SET_STATUS. */
Martijn Coenen6ac92fb2020-06-04 22:25:20 +02001294 lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001295 /* 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Martijn Coenenb0bd1582020-05-13 15:38:39 +02001300 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);
Arnd Bergmannb040ad92017-06-09 12:19:18 +02001304 }
Guo Chao541c7422013-02-21 15:16:46 -08001305
Lukas Czernerdfaa2ef2011-08-19 14:50:46 +02001306 loop_config_discard(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
Ming Lei2e5ab5f32015-08-17 10:31:49 +08001308 /* update dio if lo_offset or transfer is changed */
1309 __loop_update_dio(lo, lo->use_dio);
1310
Jan Kara550df5fd2018-11-08 14:01:09 +01001311out_unfreeze:
Ming Leiecdd0952017-02-11 11:40:45 +08001312 blk_mq_unfreeze_queue(lo->lo_queue);
Omar Sandovale02898b42017-03-01 10:42:38 -08001313
Martijn Coenenfaf1d252020-05-13 15:38:44 +02001314 if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1315 !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
Christoph Hellwigb9684a72022-05-27 07:58:06 +02001316 clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
Jan Kara85b0a542018-11-08 14:01:13 +01001317 partscan = true;
Omar Sandovale02898b42017-03-01 10:42:38 -08001318 }
Jan Kara550df5fd2018-11-08 14:01:09 +01001319out_unlock:
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001320 mutex_unlock(&lo->lo_mutex);
Jan Kara85b0a542018-11-08 14:01:13 +01001321 if (partscan)
Christoph Hellwig03842642021-06-24 14:32:40 +02001322 loop_reread_partitions(lo);
Omar Sandovale02898b42017-03-01 10:42:38 -08001323
Ming Leiecdd0952017-02-11 11:40:45 +08001324 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325}
1326
1327static int
1328loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1329{
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001330 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 struct kstat stat;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001332 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001334 ret = mutex_lock_killable(&lo->lo_mutex);
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001335 if (ret)
1336 return ret;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001337 if (lo->lo_state != Lo_bound) {
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001338 mutex_unlock(&lo->lo_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 return -ENXIO;
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001340 }
1341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 memset(info, 0, sizeof(*info));
1343 info->lo_number = lo->lo_number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 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);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001348
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001349 /* Drop lo_mutex while we call into the filesystem. */
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001350 path = lo->lo_backing_file->f_path;
1351 path_get(&path);
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001352 mutex_unlock(&lo->lo_mutex);
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001353 ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001354 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 }
Tetsuo Handab1ab5fa2018-11-08 14:01:01 +01001359 path_put(&path);
Omar Sandoval2d1d4c12018-03-26 21:39:11 -07001360 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361}
1362
1363static void
1364loop_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 info64->lo_flags = info->lo_flags;
Christoph Hellwig47e96242021-10-19 09:56:39 +02001374 memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375}
1376
1377static int
1378loop_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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 info->lo_flags = info64->lo_flags;
Christoph Hellwig47e96242021-10-19 09:56:39 +02001387 memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388
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
1399static int
1400loop_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
1411static int
1412loop_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
1421static int
1422loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1423 struct loop_info info;
1424 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001425 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001427 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001428 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001429 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 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
1438static int
1439loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1440 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001441 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001443 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001444 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001445 err = loop_get_status(lo, &info64);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1447 err = -EFAULT;
1448
1449 return err;
1450}
1451
Hannes Reinecke51001b72017-06-08 13:46:44 +02001452static int loop_set_capacity(struct loop_device *lo)
J. R. Okajima53d66602009-03-31 15:23:43 -07001453{
Martijn Coenen0a6ed1b2020-05-13 15:38:40 +02001454 loff_t size;
1455
J. R. Okajima53d66602009-03-31 15:23:43 -07001456 if (unlikely(lo->lo_state != Lo_bound))
Guo Chao7b0576a2013-02-21 15:16:47 -08001457 return -ENXIO;
J. R. Okajima53d66602009-03-31 15:23:43 -07001458
Martijn Coenen0a6ed1b2020-05-13 15:38:40 +02001459 size = get_loop_size(lo, lo->lo_backing_file);
1460 loop_set_size(lo, size);
Martijn Coenen083a6a52020-05-13 15:38:36 +02001461
1462 return 0;
J. R. Okajima53d66602009-03-31 15:23:43 -07001463}
1464
Ming Leiab1cb272015-08-17 10:31:50 +08001465static 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
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001479static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1480{
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001481 int err = 0;
1482
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001483 if (lo->lo_state != Lo_bound)
1484 return -ENXIO;
1485
Xie Yongjiaf3c5702021-10-26 22:40:14 +08001486 err = blk_validate_block_size(arg);
Martijn Coenen34489142020-05-13 15:38:45 +02001487 if (err)
1488 return err;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001489
Martijn Coenen7e81f992020-03-10 14:12:30 +01001490 if (lo->lo_queue->limits.logical_block_size == arg)
1491 return 0;
1492
1493 sync_blockdev(lo->lo_device);
Zheng Binf4bd34b2020-06-18 12:21:37 +08001494 invalidate_bdev(lo->lo_device);
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001495
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001496 blk_mq_freeze_queue(lo->lo_queue);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001497 blk_queue_logical_block_size(lo->lo_queue, arg);
Omar Sandovalbf093752017-09-05 14:24:47 -07001498 blk_queue_physical_block_size(lo->lo_queue, arg);
1499 blk_queue_io_min(lo->lo_queue, arg);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001500 loop_update_dio(lo);
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001501 blk_mq_unfreeze_queue(lo->lo_queue);
1502
Jaegeuk Kim5db470e2019-01-09 19:17:14 -08001503 return err;
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001504}
1505
Jan Karaa1316542018-11-08 14:01:05 +01001506static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1507 unsigned long arg)
1508{
1509 int err;
1510
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001511 err = mutex_lock_killable(&lo->lo_mutex);
Jan Karaa1316542018-11-08 14:01:05 +01001512 if (err)
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:
Christoph Hellwig47e96242021-10-19 09:56:39 +02001525 err = -EINVAL;
Jan Karaa1316542018-11-08 14:01:05 +01001526 }
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001527 mutex_unlock(&lo->lo_mutex);
Jan Karaa1316542018-11-08 14:01:05 +01001528 return err;
1529}
1530
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001531static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 unsigned int cmd, unsigned long arg)
1533{
Al Virobb214882008-03-02 09:29:48 -05001534 struct loop_device *lo = bdev->bd_disk->private_data;
Martijn Coenen571fae62020-05-13 15:38:43 +02001535 void __user *argp = (void __user *) arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 int err;
1537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 switch (cmd) {
Martijn Coenen34489142020-05-13 15:38:45 +02001539 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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 case LOOP_CHANGE_FD:
Jan Karac3710772018-11-08 14:01:11 +01001561 return loop_change_fd(lo, bdev, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 case LOOP_CLR_FD:
Jan Kara7ccd0792018-11-08 14:01:07 +01001563 return loop_clr_fd(lo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 case LOOP_SET_STATUS:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001565 err = -EPERM;
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001566 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
Martijn Coenen571fae62020-05-13 15:38:43 +02001567 err = loop_set_status_old(lo, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 break;
1569 case LOOP_GET_STATUS:
Martijn Coenen571fae62020-05-13 15:38:43 +02001570 return loop_get_status_old(lo, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 case LOOP_SET_STATUS64:
Dmitry Monakhov7035b5d2011-11-16 09:21:49 +01001572 err = -EPERM;
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001573 if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
Martijn Coenen571fae62020-05-13 15:38:43 +02001574 err = loop_set_status64(lo, argp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 break;
1576 case LOOP_GET_STATUS64:
Martijn Coenen571fae62020-05-13 15:38:43 +02001577 return loop_get_status64(lo, argp);
J. R. Okajima53d66602009-03-31 15:23:43 -07001578 case LOOP_SET_CAPACITY:
Ming Leiab1cb272015-08-17 10:31:50 +08001579 case LOOP_SET_DIRECT_IO:
Omar Sandoval89e4fde2017-08-24 00:03:43 -07001580 case LOOP_SET_BLOCK_SIZE:
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001581 if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
Jan Karaa1316542018-11-08 14:01:05 +01001582 return -EPERM;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001583 fallthrough;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 default:
Jan Karaa1316542018-11-08 14:01:05 +01001585 err = lo_simple_ioctl(lo, cmd, arg);
1586 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
Nikanth Karthikesanf028f3b2009-03-24 12:33:41 +01001588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 return err;
1590}
1591
David Howells863d5b822006-08-29 19:06:14 +01001592#ifdef CONFIG_COMPAT
1593struct 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;
Carlos Llamasf941c512022-03-29 20:18:15 +00001599 compat_int_t lo_encrypt_type; /* obsolete, ignored */
David Howells863d5b822006-08-29 19:06:14 +01001600 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 */
1612static noinline int
Al Viroba674cf2006-10-10 22:48:27 +01001613loop_info64_from_compat(const struct compat_loop_info __user *arg,
David Howells863d5b822006-08-29 19:06:14 +01001614 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;
David Howells863d5b822006-08-29 19:06:14 +01001628 info64->lo_flags = info.lo_flags;
Christoph Hellwig47e96242021-10-19 09:56:39 +02001629 memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
David Howells863d5b822006-08-29 19:06:14 +01001630 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 */
1637static noinline int
1638loop_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;
David Howells863d5b822006-08-29 19:06:14 +01001649 info.lo_flags = info64->lo_flags;
Christoph Hellwig47e96242021-10-19 09:56:39 +02001650 memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
David Howells863d5b822006-08-29 19:06:14 +01001651
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 ||
Christoph Hellwig47e96242021-10-19 09:56:39 +02001656 info.lo_offset != info64->lo_offset)
David Howells863d5b822006-08-29 19:06:14 +01001657 return -EOVERFLOW;
1658
1659 if (copy_to_user(arg, &info, sizeof(info)))
1660 return -EFAULT;
1661 return 0;
1662}
1663
1664static int
1665loop_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
1677static int
1678loop_get_status_compat(struct loop_device *lo,
1679 struct compat_loop_info __user *arg)
1680{
1681 struct loop_info64 info64;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001682 int err;
David Howells863d5b822006-08-29 19:06:14 +01001683
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001684 if (!arg)
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001685 return -EINVAL;
Omar Sandovalbdac616db2018-04-06 09:57:03 -07001686 err = loop_get_status(lo, &info64);
David Howells863d5b822006-08-29 19:06:14 +01001687 if (!err)
1688 err = loop_info64_to_compat(&info64, arg);
1689 return err;
1690}
1691
Christoph Hellwig05bdb992023-06-08 13:02:55 +02001692static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
Al Virobb214882008-03-02 09:29:48 -05001693 unsigned int cmd, unsigned long arg)
David Howells863d5b822006-08-29 19:06:14 +01001694{
Al Virobb214882008-03-02 09:29:48 -05001695 struct loop_device *lo = bdev->bd_disk->private_data;
David Howells863d5b822006-08-29 19:06:14 +01001696 int err;
1697
David Howells863d5b822006-08-29 19:06:14 +01001698 switch(cmd) {
1699 case LOOP_SET_STATUS:
Jan Kara550df5fd2018-11-08 14:01:09 +01001700 err = loop_set_status_compat(lo,
1701 (const struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001702 break;
1703 case LOOP_GET_STATUS:
Jan Kara4a5ce9b2018-11-08 14:01:08 +01001704 err = loop_get_status_compat(lo,
1705 (struct compat_loop_info __user *)arg);
David Howells863d5b822006-08-29 19:06:14 +01001706 break;
J. R. Okajima53d66602009-03-31 15:23:43 -07001707 case LOOP_SET_CAPACITY:
David Howells863d5b822006-08-29 19:06:14 +01001708 case LOOP_CLR_FD:
1709 case LOOP_GET_STATUS64:
1710 case LOOP_SET_STATUS64:
Martijn Coenen34489142020-05-13 15:38:45 +02001711 case LOOP_CONFIGURE:
David Howells863d5b822006-08-29 19:06:14 +01001712 arg = (unsigned long) compat_ptr(arg);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001713 fallthrough;
David Howells863d5b822006-08-29 19:06:14 +01001714 case LOOP_SET_FD:
1715 case LOOP_CHANGE_FD:
Evan Green9fea4b32018-07-02 16:03:46 -07001716 case LOOP_SET_BLOCK_SIZE:
Alessio Balsinifdbe4ee2019-08-07 01:48:28 +01001717 case LOOP_SET_DIRECT_IO:
Al Virobb214882008-03-02 09:29:48 -05001718 err = lo_ioctl(bdev, mode, cmd, arg);
David Howells863d5b822006-08-29 19:06:14 +01001719 break;
1720 default:
1721 err = -ENOIOCTLCMD;
1722 break;
1723 }
David Howells863d5b822006-08-29 19:06:14 +01001724 return err;
1725}
1726#endif
1727
Christoph Hellwigae220762023-06-08 13:02:37 +02001728static void lo_release(struct gendisk *disk)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729{
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001730 struct loop_device *lo = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Christoph Hellwiga0e286b2022-03-30 07:29:16 +02001732 if (disk_openers(disk) > 0)
1733 return;
Ming Leif8933662015-05-06 12:26:23 +08001734
Christoph Hellwiga0e286b2022-03-30 07:29:16 +02001735 mutex_lock(&lo->lo_mutex);
1736 if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
Jan Karaa2505b72018-11-08 14:01:06 +01001737 lo->lo_state = Lo_rundown;
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001738 mutex_unlock(&lo->lo_mutex);
Milan Broz14f27932008-12-12 14:48:27 +01001739 /*
1740 * In autoclear mode, stop the loop thread
1741 * and remove configuration after last close.
1742 */
Tetsuo Handabf237472022-02-11 16:15:54 +09001743 __loop_clr_fd(lo, true);
Jan Kara7ccd0792018-11-08 14:01:07 +01001744 return;
Milan Broz14f27932008-12-12 14:48:27 +01001745 }
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05001746 mutex_unlock(&lo->lo_mutex);
Linus Torvaldsae665012018-01-05 16:26:00 -08001747}
1748
Christoph Hellwigd2c7f562022-03-30 07:29:13 +02001749static void lo_free_disk(struct gendisk *disk)
1750{
1751 struct loop_device *lo = disk->private_data;
1752
Christoph Hellwigd292dc82022-03-30 07:29:17 +02001753 if (lo->workqueue)
1754 destroy_workqueue(lo->workqueue);
1755 loop_free_idle_workers(lo, true);
Steven Rostedt (Google)292a0892022-12-20 13:45:19 -05001756 timer_shutdown_sync(&lo->timer);
Christoph Hellwigd2c7f562022-03-30 07:29:13 +02001757 mutex_destroy(&lo->lo_mutex);
1758 kfree(lo);
1759}
1760
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07001761static const struct block_device_operations lo_fops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 .owner = THIS_MODULE,
Al Virobb214882008-03-02 09:29:48 -05001763 .release = lo_release,
1764 .ioctl = lo_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001765#ifdef CONFIG_COMPAT
Al Virobb214882008-03-02 09:29:48 -05001766 .compat_ioctl = lo_compat_ioctl,
David Howells863d5b822006-08-29 19:06:14 +01001767#endif
Christoph Hellwigd2c7f562022-03-30 07:29:13 +02001768 .free_disk = lo_free_disk,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769};
1770
1771/*
1772 * And now the modules code and kernel interface.
1773 */
Isaac J. Manjarres85c50192022-12-08 13:29:01 -08001774
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 */
1783static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
Joe Perches5657a812018-05-24 13:38:59 -06001784module_param(max_loop, int, 0444);
Ken Chena47653f2007-06-08 13:46:44 -07001785MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
Joe Perches5657a812018-05-24 13:38:59 -06001786module_param(max_part, int, 0444);
Laurent Vivier476a4812008-03-26 12:11:53 +01001787MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
Chaitanya Kulkarnief44c502022-02-15 13:33:10 -08001788
1789static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1790
1791static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1792{
Bart Van Asschee152a052023-01-30 13:13:47 -08001793 int qd, ret;
Chaitanya Kulkarnief44c502022-02-15 13:33:10 -08001794
Bart Van Asschee152a052023-01-30 13:13:47 -08001795 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;
Chaitanya Kulkarnief44c502022-02-15 13:33:10 -08001802}
1803
1804static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1805 .set = loop_set_hw_queue_depth,
1806 .get = param_get_int,
1807};
1808
1809device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
Bart Van Asschee152a052023-01-30 13:13:47 -08001810MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
Chaitanya Kulkarnief44c502022-02-15 13:33:10 -08001811
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812MODULE_LICENSE("GPL");
1813MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1814
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001815static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
Ming Leib5dd2f62014-12-31 13:22:57 +00001816 const struct blk_mq_queue_data *bd)
1817{
Jens Axboe1894e9162018-04-13 16:24:29 -06001818 struct request *rq = bd->rq;
1819 struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1820 struct loop_device *lo = rq->q->queuedata;
Ming Leib5dd2f62014-12-31 13:22:57 +00001821
Jens Axboe1894e9162018-04-13 16:24:29 -06001822 blk_mq_start_request(rq);
Ming Leib5dd2f62014-12-31 13:22:57 +00001823
Ming Leif4aa4c72015-05-05 19:49:54 +08001824 if (lo->lo_state != Lo_bound)
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001825 return BLK_STS_IOERR;
Ming Leif4aa4c72015-05-05 19:49:54 +08001826
Jens Axboe1894e9162018-04-13 16:24:29 -06001827 switch (req_op(rq)) {
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001828 case REQ_OP_FLUSH:
1829 case REQ_OP_DISCARD:
Christoph Hellwig19372e22017-04-05 19:21:15 +02001830 case REQ_OP_WRITE_ZEROES:
Ming Leibc07c102015-08-17 10:31:51 +08001831 cmd->use_aio = false;
Christoph Hellwigf0225ca2016-08-04 16:10:00 +02001832 break;
1833 default:
1834 cmd->use_aio = lo->use_dio;
1835 break;
1836 }
Ming Leibc07c102015-08-17 10:31:51 +08001837
Shaohua Lid4478e92017-09-25 13:07:22 -06001838 /* always use the first bio's css */
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001839 cmd->blkcg_css = NULL;
1840 cmd->memcg_css = NULL;
Shaohua Li0b508bc2017-09-26 11:02:12 -07001841#ifdef CONFIG_BLK_CGROUP
Christoph Hellwigbbb1ebe2022-04-20 06:27:17 +02001842 if (rq->bio) {
1843 cmd->blkcg_css = bio_blkcg_css(rq->bio);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001844#ifdef CONFIG_MEMCG
Christoph Hellwigbbb1ebe2022-04-20 06:27:17 +02001845 if (cmd->blkcg_css) {
1846 cmd->memcg_css =
1847 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1848 &memory_cgrp_subsys);
1849 }
Shaohua Lid4478e92017-09-25 13:07:22 -06001850#endif
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001851 }
1852#endif
Dan Schatzberg87579e92021-06-28 19:38:15 -07001853 loop_queue_work(lo, cmd);
Ming Leib5dd2f62014-12-31 13:22:57 +00001854
Christoph Hellwigfc17b652017-06-03 09:38:05 +02001855 return BLK_STS_OK;
Ming Leib5dd2f62014-12-31 13:22:57 +00001856}
1857
1858static void loop_handle_cmd(struct loop_cmd *cmd)
1859{
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001860 struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1861 struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
Jens Axboe1894e9162018-04-13 16:24:29 -06001862 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;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001865 int ret = 0;
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001866 struct mem_cgroup *old_memcg = NULL;
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001867 const bool use_aio = cmd->use_aio;
Ming Leib5dd2f62014-12-31 13:22:57 +00001868
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001869 if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1870 ret = -EIO;
Ming Leib5dd2f62014-12-31 13:22:57 +00001871 goto failed;
Christoph Hellwigf4829a92015-09-27 21:01:50 +02001872 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001873
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001874 if (cmd_blkcg_css)
1875 kthread_associate_blkcg(cmd_blkcg_css);
1876 if (cmd_memcg_css)
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001877 old_memcg = set_active_memcg(
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001878 mem_cgroup_from_css(cmd_memcg_css));
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001879
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001880 /*
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 */
Jens Axboe1894e9162018-04-13 16:24:29 -06001886 ret = do_req_filebacked(lo, rq);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001887
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001888 if (cmd_blkcg_css)
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001889 kthread_associate_blkcg(NULL);
1890
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001891 if (cmd_memcg_css) {
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001892 set_active_memcg(old_memcg);
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001893 css_put(cmd_memcg_css);
Dan Schatzbergc74d40e2021-06-28 19:38:21 -07001894 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001895 failed:
Ming Leibc07c102015-08-17 10:31:51 +08001896 /* complete non-aio request */
Bart Van Assche9b0cb772023-03-14 11:21:54 -07001897 if (!use_aio || ret) {
Evan Green8cd55082020-04-03 16:43:03 +02001898 if (ret == -EOPNOTSUPP)
1899 cmd->ret = ret;
1900 else
1901 cmd->ret = ret ? -EIO : 0;
Christoph Hellwig15f73f52020-06-11 08:44:47 +02001902 if (likely(!blk_should_fake_timeout(rq->q)))
1903 blk_mq_complete_request(rq);
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001904 }
Ming Leib5dd2f62014-12-31 13:22:57 +00001905}
1906
Dan Schatzberg87579e92021-06-28 19:38:15 -07001907static void loop_process_work(struct loop_worker *worker,
1908 struct list_head *cmd_list, struct loop_device *lo)
Ming Leib5dd2f62014-12-31 13:22:57 +00001909{
Dan Schatzberg87579e92021-06-28 19:38:15 -07001910 int orig_flags = current->flags;
1911 struct loop_cmd *cmd;
Ming Leib5dd2f62014-12-31 13:22:57 +00001912
Dan Schatzberg87579e92021-06-28 19:38:15 -07001913 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;
1939}
1940
1941static void loop_workfn(struct work_struct *work)
1942{
1943 struct loop_worker *worker =
1944 container_of(work, struct loop_worker, work);
1945 loop_process_work(worker, &worker->cmd_list, worker->lo);
1946}
1947
1948static 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
Eric Biggersf363b082017-03-30 13:39:16 -07001955static const struct blk_mq_ops loop_mq_ops = {
Ming Leib5dd2f62014-12-31 13:22:57 +00001956 .queue_rq = loop_queue_rq,
Christoph Hellwigfe2cb292017-04-20 16:03:02 +02001957 .complete = lo_complete_rq,
Ming Leib5dd2f62014-12-31 13:22:57 +00001958};
1959
Christoph Hellwigd6da83d2021-06-23 16:59:02 +02001960static int loop_add(int i)
Ken Chen73285082007-05-08 00:28:20 -07001961{
1962 struct loop_device *lo;
1963 struct gendisk *disk;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001964 int err;
Ken Chen73285082007-05-08 00:28:20 -07001965
Silva Paulo68d740d2012-07-14 15:39:58 -07001966 err = -ENOMEM;
Ken Chen73285082007-05-08 00:28:20 -07001967 lo = kzalloc(sizeof(*lo), GFP_KERNEL);
Silva Paulo68d740d2012-07-14 15:39:58 -07001968 if (!lo)
Ken Chen73285082007-05-08 00:28:20 -07001969 goto out;
Christoph Hellwigb15ed542022-03-30 07:29:09 +02001970 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);
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06001973 lo->lo_state = Lo_unbound;
1974
Christoph Hellwig18d1f202021-06-23 16:59:05 +02001975 err = mutex_lock_killable(&loop_ctl_mutex);
1976 if (err)
1977 goto out_free_dev;
1978
Tejun Heoc718aa62013-02-27 17:03:58 -08001979 /* allocate id, if @id >= 0, we're requesting that specific id */
Kay Sievers34dd82a2011-07-31 22:08:04 +02001980 if (i >= 0) {
Tejun Heoc718aa62013-02-27 17:03:58 -08001981 err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1982 if (err == -ENOSPC)
Kay Sievers34dd82a2011-07-31 22:08:04 +02001983 err = -EEXIST;
Kay Sievers34dd82a2011-07-31 22:08:04 +02001984 } else {
Tejun Heoc718aa62013-02-27 17:03:58 -08001985 err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001986 }
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09001987 mutex_unlock(&loop_ctl_mutex);
Kay Sievers34dd82a2011-07-31 22:08:04 +02001988 if (err < 0)
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09001989 goto out_free_dev;
Tejun Heoc718aa62013-02-27 17:03:58 -08001990 i = err;
Ken Chen73285082007-05-08 00:28:20 -07001991
Ming Leib5dd2f62014-12-31 13:22:57 +00001992 lo->tag_set.ops = &loop_mq_ops;
1993 lo->tag_set.nr_hw_queues = 1;
Chaitanya Kulkarnief44c502022-02-15 13:33:10 -08001994 lo->tag_set.queue_depth = hw_queue_depth;
Ming Leib5dd2f62014-12-31 13:22:57 +00001995 lo->tag_set.numa_node = NUMA_NO_NODE;
1996 lo->tag_set.cmd_size = sizeof(struct loop_cmd);
Bart Van Assche2112f5c2021-08-05 10:42:00 -07001997 lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
1998 BLK_MQ_F_NO_SCHED_BY_DEFAULT;
Ming Leib5dd2f62014-12-31 13:22:57 +00001999 lo->tag_set.driver_data = lo;
2000
2001 err = blk_mq_alloc_tag_set(&lo->tag_set);
2002 if (err)
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002003 goto out_free_idr;
Ken Chen73285082007-05-08 00:28:20 -07002004
Christoph Hellwig1c995022021-06-02 09:53:33 +03002005 disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
2006 if (IS_ERR(disk)) {
2007 err = PTR_ERR(disk);
Ming Leib5dd2f62014-12-31 13:22:57 +00002008 goto out_cleanup_tags;
2009 }
Christoph Hellwig1c995022021-06-02 09:53:33 +03002010 lo->lo_queue = lo->lo_disk->queue;
Mikulas Patockaef7e7c82013-10-15 14:14:38 -06002011
Shaohua Li54bb0ad2017-08-31 22:09:45 -07002012 blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
Shaohua Li40326d82017-08-31 22:09:46 -07002013
Ming Lei5b5e20f2015-08-17 10:31:47 +08002014 /*
Shaohua Li40326d82017-08-31 22:09:46 -07002015 * 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.
Ming Lei5b5e20f2015-08-17 10:31:47 +08002019 */
Bart Van Assche8b904b52018-03-07 17:10:10 -08002020 blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
Ming Lei5b5e20f2015-08-17 10:31:47 +08002021
Kay Sieverse03c8dd2011-08-23 20:12:04 +02002022 /*
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)
Christoph Hellwigb9684a72022-05-27 07:58:06 +02002041 set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
Pavel Tatashin6cc8e742021-01-26 09:46:30 -05002042 mutex_init(&lo->lo_mutex);
Ken Chen73285082007-05-08 00:28:20 -07002043 lo->lo_number = i;
Ken Chen73285082007-05-08 00:28:20 -07002044 spin_lock_init(&lo->lo_lock);
Dan Schatzberg87579e92021-06-28 19:38:15 -07002045 spin_lock_init(&lo->lo_work_lock);
Christoph Hellwigd292dc82022-03-30 07:29:17 +02002046 INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2047 INIT_LIST_HEAD(&lo->rootcg_cmd_list);
Ken Chen73285082007-05-08 00:28:20 -07002048 disk->major = LOOP_MAJOR;
Laurent Vivier476a4812008-03-26 12:11:53 +01002049 disk->first_minor = i << part_shift;
Christoph Hellwig1c995022021-06-02 09:53:33 +03002050 disk->minors = 1 << part_shift;
Ken Chen73285082007-05-08 00:28:20 -07002051 disk->fops = &lo_fops;
2052 disk->private_data = lo;
2053 disk->queue = lo->lo_queue;
Matteo Croce9f65c482021-07-13 01:05:30 +02002054 disk->events = DISK_EVENT_MEDIA_CHANGE;
2055 disk->event_flags = DISK_EVENT_FLAG_UEVENT;
Ken Chen73285082007-05-08 00:28:20 -07002056 sprintf(disk->disk_name, "loop%d", i);
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002057 /* Make this loop device reachable from pathname. */
Luis Chamberlain905705f2021-09-27 14:59:57 -07002058 err = add_disk(disk);
2059 if (err)
2060 goto out_cleanup_disk;
2061
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002062 /* Show this loop device. */
2063 mutex_lock(&loop_ctl_mutex);
2064 lo->idr_visible = true;
Christoph Hellwig18d1f202021-06-23 16:59:05 +02002065 mutex_unlock(&loop_ctl_mutex);
Luis Chamberlain905705f2021-09-27 14:59:57 -07002066
Christoph Hellwig18d1f202021-06-23 16:59:05 +02002067 return i;
Ken Chen73285082007-05-08 00:28:20 -07002068
Luis Chamberlain905705f2021-09-27 14:59:57 -07002069out_cleanup_disk:
Christoph Hellwig8b9ab622022-06-19 08:05:52 +02002070 put_disk(disk);
Ming Leib5dd2f62014-12-31 13:22:57 +00002071out_cleanup_tags:
2072 blk_mq_free_tag_set(&lo->tag_set);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002073out_free_idr:
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002074 mutex_lock(&loop_ctl_mutex);
Mikulas Patocka3ec981e2013-10-14 12:12:24 -04002075 idr_remove(&loop_index_idr, i);
Christoph Hellwig18d1f202021-06-23 16:59:05 +02002076 mutex_unlock(&loop_ctl_mutex);
Ken Chen73285082007-05-08 00:28:20 -07002077out_free_dev:
2078 kfree(lo);
2079out:
Kay Sievers34dd82a2011-07-31 22:08:04 +02002080 return err;
Ken Chen73285082007-05-08 00:28:20 -07002081}
2082
Kay Sievers34dd82a2011-07-31 22:08:04 +02002083static void loop_remove(struct loop_device *lo)
Ken Chen73285082007-05-08 00:28:20 -07002084{
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002085 /* Make this loop device unreachable from pathname. */
NeilBrown6cd18e72015-04-27 14:12:22 +10002086 del_gendisk(lo->lo_disk);
Ming Leib5dd2f62014-12-31 13:22:57 +00002087 blk_mq_free_tag_set(&lo->tag_set);
Christoph Hellwigce8d7862022-03-15 13:27:07 +01002088
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002089 mutex_lock(&loop_ctl_mutex);
2090 idr_remove(&loop_index_idr, lo->lo_number);
2091 mutex_unlock(&loop_ctl_mutex);
Christoph Hellwigd2c7f562022-03-30 07:29:13 +02002092
2093 put_disk(lo->lo_disk);
Ken Chen73285082007-05-08 00:28:20 -07002094}
2095
Christoph Hellwig8410d382020-10-29 15:58:33 +01002096static void loop_probe(dev_t dev)
Ken Chen73285082007-05-08 00:28:20 -07002097{
Christoph Hellwig8410d382020-10-29 15:58:33 +01002098 int idx = MINOR(dev) >> part_shift;
Christoph Hellwig8410d382020-10-29 15:58:33 +01002099
2100 if (max_loop && idx >= max_loop)
2101 return;
Christoph Hellwig4157fe02021-06-23 16:59:03 +02002102 loop_add(idx);
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002103}
2104
2105static int loop_control_remove(int idx)
2106{
2107 struct loop_device *lo;
2108 int ret;
Christoph Hellwige5d66a102021-06-23 16:59:06 +02002109
2110 if (idx < 0) {
Tetsuo Handae3f93872021-11-29 19:00:43 +09002111 pr_warn_once("deleting an unspecified loop device is not supported.\n");
Christoph Hellwige5d66a102021-06-23 16:59:06 +02002112 return -EINVAL;
2113 }
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002114
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002115 /* Hide this loop device for serialization. */
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002116 ret = mutex_lock_killable(&loop_ctl_mutex);
2117 if (ret)
2118 return ret;
Christoph Hellwigb9848082021-06-23 16:59:07 +02002119 lo = idr_find(&loop_index_idr, idx);
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002120 if (!lo || !lo->idr_visible)
Christoph Hellwigb9848082021-06-23 16:59:07 +02002121 ret = -ENODEV;
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002122 else
2123 lo->idr_visible = false;
2124 mutex_unlock(&loop_ctl_mutex);
2125 if (ret)
2126 return ret;
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002127
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002128 /* Check whether this loop device can be removed. */
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002129 ret = mutex_lock_killable(&lo->lo_mutex);
2130 if (ret)
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002131 goto mark_visible;
Christoph Hellwiga0e286b2022-03-30 07:29:16 +02002132 if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002133 mutex_unlock(&lo->lo_mutex);
2134 ret = -EBUSY;
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002135 goto mark_visible;
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002136 }
Christoph Hellwiga0e286b2022-03-30 07:29:16 +02002137 /* Mark this loop device as no more bound, but not quite unbound yet */
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002138 lo->lo_state = Lo_deleting;
2139 mutex_unlock(&lo->lo_mutex);
2140
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002141 loop_remove(lo);
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002142 return 0;
2143
2144mark_visible:
2145 /* Show this loop device again. */
2146 mutex_lock(&loop_ctl_mutex);
2147 lo->idr_visible = true;
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002148 mutex_unlock(&loop_ctl_mutex);
2149 return ret;
2150}
2151
2152static int loop_control_get_free(int idx)
Kay Sievers770fe302011-07-31 22:08:04 +02002153{
2154 struct loop_device *lo;
Christoph Hellwigb9848082021-06-23 16:59:07 +02002155 int id, ret;
Kay Sievers770fe302011-07-31 22:08:04 +02002156
Jan Kara0a42e992018-11-08 14:01:04 +01002157 ret = mutex_lock_killable(&loop_ctl_mutex);
2158 if (ret)
2159 return ret;
Christoph Hellwigb9848082021-06-23 16:59:07 +02002160 idr_for_each_entry(&loop_index_idr, lo, id) {
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002161 /* 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)
Christoph Hellwigb9848082021-06-23 16:59:07 +02002163 goto found;
2164 }
Jan Kara0a42e992018-11-08 14:01:04 +01002165 mutex_unlock(&loop_ctl_mutex);
Christoph Hellwig18d1f202021-06-23 16:59:05 +02002166 return loop_add(-1);
Christoph Hellwigb9848082021-06-23 16:59:07 +02002167found:
2168 mutex_unlock(&loop_ctl_mutex);
2169 return id;
Kay Sievers770fe302011-07-31 22:08:04 +02002170}
2171
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002172static long loop_control_ioctl(struct file *file, unsigned int cmd,
2173 unsigned long parm)
2174{
2175 switch (cmd) {
2176 case LOOP_CTL_ADD:
Christoph Hellwig18d1f202021-06-23 16:59:05 +02002177 return loop_add(parm);
Christoph Hellwigf9d10762021-06-23 16:59:04 +02002178 case LOOP_CTL_REMOVE:
2179 return loop_control_remove(parm);
2180 case LOOP_CTL_GET_FREE:
2181 return loop_control_get_free(parm);
2182 default:
2183 return -ENOSYS;
2184 }
2185}
2186
Kay Sievers770fe302011-07-31 22:08:04 +02002187static 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
2195static struct miscdevice loop_misc = {
2196 .minor = LOOP_CTRL_MINOR,
2197 .name = "loop-control",
2198 .fops = &loop_ctl_fops,
2199};
2200
2201MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2202MODULE_ALIAS("devname:loop-control");
2203
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204static int __init loop_init(void)
2205{
Isaac J. Manjarres85c50192022-12-08 13:29:01 -08002206 int i;
Kay Sievers770fe302011-07-31 22:08:04 +02002207 int err;
Ken Chena47653f2007-06-08 13:46:44 -07002208
Laurent Vivier476a4812008-03-26 12:11:53 +01002209 part_shift = 0;
Namhyung Kimac04fee2011-05-27 07:59:25 +02002210 if (max_part > 0) {
Laurent Vivier476a4812008-03-26 12:11:53 +01002211 part_shift = fls(max_part);
2212
Namhyung Kimac04fee2011-05-27 07:59:25 +02002213 /*
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
Guo Chaob1a66502013-02-21 15:16:49 -08002224 if ((1UL << part_shift) > DISK_MAX_PARTS) {
2225 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002226 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002227 }
Namhyung Kim78f4bb32011-05-24 16:48:54 +02002228
Guo Chaob1a66502013-02-21 15:16:49 -08002229 if (max_loop > 1UL << (MINORBITS - part_shift)) {
2230 err = -EINVAL;
Anton Volkova8c1d062017-08-07 15:37:50 +03002231 goto err_out;
Guo Chaob1a66502013-02-21 15:16:49 -08002232 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233
Anton Volkova8c1d062017-08-07 15:37:50 +03002234 err = misc_register(&loop_misc);
2235 if (err < 0)
2236 goto err_out;
2237
2238
Christoph Hellwig8410d382020-10-29 15:58:33 +01002239 if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
Guo Chaob1a66502013-02-21 15:16:49 -08002240 err = -EIO;
2241 goto misc_out;
2242 }
Ken Chena47653f2007-06-08 13:46:44 -07002243
Kay Sieversd134b002011-07-31 22:08:04 +02002244 /* pre-create number of devices given by config or max_loop */
Isaac J. Manjarres85c50192022-12-08 13:29:01 -08002245 for (i = 0; i < max_loop; i++)
Christoph Hellwigd6da83d2021-06-23 16:59:02 +02002246 loop_add(i);
Kay Sievers34dd82a2011-07-31 22:08:04 +02002247
Ken Chen73285082007-05-08 00:28:20 -07002248 printk(KERN_INFO "loop: module loaded\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 return 0;
Guo Chaob1a66502013-02-21 15:16:49 -08002250
2251misc_out:
2252 misc_deregister(&loop_misc);
Anton Volkova8c1d062017-08-07 15:37:50 +03002253err_out:
Guo Chaob1a66502013-02-21 15:16:49 -08002254 return err;
Kay Sievers34dd82a2011-07-31 22:08:04 +02002255}
Ken Chena47653f2007-06-08 13:46:44 -07002256
Ken Chen73285082007-05-08 00:28:20 -07002257static void __exit loop_exit(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258{
Christoph Hellwig8e609472021-06-23 16:59:08 +02002259 struct loop_device *lo;
2260 int id;
2261
Christoph Hellwig8b52d8b2021-06-23 16:59:00 +02002262 unregister_blkdev(LOOP_MAJOR, "loop");
2263 misc_deregister(&loop_misc);
Luis Chamberlain200f9332020-06-19 20:47:27 +00002264
Tetsuo Handa1c500ad2021-09-02 09:07:35 +09002265 /*
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 */
Christoph Hellwig8e609472021-06-23 16:59:08 +02002271 idr_for_each_entry(&loop_index_idr, lo, id)
2272 loop_remove(lo);
Christoph Hellwigbd5c39e2021-06-23 16:59:01 +02002273
2274 idr_destroy(&loop_index_idr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275}
2276
2277module_init(loop_init);
2278module_exit(loop_exit);
2279
2280#ifndef MODULE
2281static 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