loop.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright 1993 by Theodore Ts'o.
   4 */
   5#include <linux/module.h>
   6#include <linux/moduleparam.h>
   7#include <linux/sched.h>
   8#include <linux/fs.h>
   9#include <linux/pagemap.h>
  10#include <linux/file.h>
  11#include <linux/stat.h>
  12#include <linux/errno.h>
  13#include <linux/major.h>
  14#include <linux/wait.h>
  15#include <linux/blkpg.h>
  16#include <linux/init.h>
  17#include <linux/swap.h>
  18#include <linux/slab.h>
  19#include <linux/compat.h>
  20#include <linux/suspend.h>
  21#include <linux/freezer.h>
  22#include <linux/mutex.h>
  23#include <linux/writeback.h>
  24#include <linux/completion.h>
  25#include <linux/highmem.h>
  26#include <linux/splice.h>
  27#include <linux/sysfs.h>
  28#include <linux/miscdevice.h>
  29#include <linux/falloc.h>
  30#include <linux/uio.h>
  31#include <linux/ioprio.h>
  32#include <linux/blk-cgroup.h>
  33#include <linux/sched/mm.h>
  34#include <linux/statfs.h>
  35#include <linux/uaccess.h>
  36#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_device {
  49	int		lo_number;
  50	loff_t		lo_offset;
  51	loff_t		lo_sizelimit;
  52	int		lo_flags;
  53	char		lo_file_name[LO_NAME_SIZE];
  54
  55	struct file	*lo_backing_file;
  56	unsigned int	lo_min_dio_size;
  57	struct block_device *lo_device;
  58
  59	gfp_t		old_gfp_mask;
  60
  61	spinlock_t		lo_lock;
  62	int			lo_state;
  63	spinlock_t              lo_work_lock;
  64	struct workqueue_struct *workqueue;
  65	struct work_struct      rootcg_work;
  66	struct list_head        rootcg_cmd_list;
  67	struct list_head        idle_worker_list;
  68	struct rb_root          worker_tree;
  69	struct timer_list       timer;
  70	bool			sysfs_inited;
  71
  72	struct request_queue	*lo_queue;
  73	struct blk_mq_tag_set	tag_set;
  74	struct gendisk		*lo_disk;
  75	struct mutex		lo_mutex;
  76	bool			idr_visible;
  77};
  78
  79struct loop_cmd {
  80	struct list_head list_entry;
  81	bool use_aio; /* use AIO interface to handle I/O */
  82	atomic_t ref; /* only for aio */
  83	long ret;
  84	struct kiocb iocb;
  85	struct bio_vec *bvec;
  86	struct cgroup_subsys_state *blkcg_css;
  87	struct cgroup_subsys_state *memcg_css;
  88};
  89
  90#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
  91#define LOOP_DEFAULT_HW_Q_DEPTH 128
  92
  93static DEFINE_IDR(loop_index_idr);
  94static DEFINE_MUTEX(loop_ctl_mutex);
  95static DEFINE_MUTEX(loop_validate_mutex);
  96
  97/**
  98 * loop_global_lock_killable() - take locks for safe loop_validate_file() test
  99 *
 100 * @lo: struct loop_device
 101 * @global: true if @lo is about to bind another "struct loop_device", false otherwise
 102 *
 103 * Returns 0 on success, -EINTR otherwise.
 104 *
 105 * Since loop_validate_file() traverses on other "struct loop_device" if
 106 * is_loop_device() is true, we need a global lock for serializing concurrent
 107 * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
 108 */
 109static int loop_global_lock_killable(struct loop_device *lo, bool global)
 110{
 111	int err;
 112
 113	if (global) {
 114		err = mutex_lock_killable(&loop_validate_mutex);
 115		if (err)
 116			return err;
 117	}
 118	err = mutex_lock_killable(&lo->lo_mutex);
 119	if (err && global)
 120		mutex_unlock(&loop_validate_mutex);
 121	return err;
 122}
 123
 124/**
 125 * loop_global_unlock() - release locks taken by loop_global_lock_killable()
 126 *
 127 * @lo: struct loop_device
 128 * @global: true if @lo was about to bind another "struct loop_device", false otherwise
 129 */
 130static void loop_global_unlock(struct loop_device *lo, bool global)
 131{
 132	mutex_unlock(&lo->lo_mutex);
 133	if (global)
 134		mutex_unlock(&loop_validate_mutex);
 135}
 136
 137static int max_part;
 138static int part_shift;
 139
 140static loff_t lo_calculate_size(struct loop_device *lo, struct file *file)
 141{
 142	loff_t loopsize;
 143	int ret;
 144
 145	if (S_ISBLK(file_inode(file)->i_mode)) {
 146		loopsize = i_size_read(file->f_mapping->host);
 147	} else {
 148		struct kstat stat;
 149
 150		/*
 151		 * Get the accurate file size. This provides better results than
 152		 * cached inode data, particularly for network filesystems where
 153		 * metadata may be stale.
 154		 */
 155		ret = vfs_getattr_nosec(&file->f_path, &stat, STATX_SIZE, 0);
 156		if (ret)
 157			return 0;
 158
 159		loopsize = stat.size;
 160	}
 161
 162	if (lo->lo_offset > 0)
 163		loopsize -= lo->lo_offset;
 164	/* offset is beyond i_size, weird but possible */
 165	if (loopsize < 0)
 166		return 0;
 167	if (lo->lo_sizelimit > 0 && lo->lo_sizelimit < loopsize)
 168		loopsize = lo->lo_sizelimit;
 169	/*
 170	 * Unfortunately, if we want to do I/O on the device,
 171	 * the number of 512-byte sectors has to fit into a sector_t.
 172	 */
 173	return loopsize >> 9;
 174}
 175
 176/*
 177 * We support direct I/O only if lo_offset is aligned with the logical I/O size
 178 * of backing device, and the logical block size of loop is bigger than that of
 179 * the backing device.
 180 */
 181static bool lo_can_use_dio(struct loop_device *lo)
 182{
 183	if (!(lo->lo_backing_file->f_mode & FMODE_CAN_ODIRECT))
 184		return false;
 185	if (queue_logical_block_size(lo->lo_queue) < lo->lo_min_dio_size)
 186		return false;
 187	if (lo->lo_offset & (lo->lo_min_dio_size - 1))
 188		return false;
 189	return true;
 190}
 191
 192/*
 193 * Direct I/O can be enabled either by using an O_DIRECT file descriptor, or by
 194 * passing in the LO_FLAGS_DIRECT_IO flag from userspace.  It will be silently
 195 * disabled when the device block size is too small or the offset is unaligned.
 196 *
 197 * loop_get_status will always report the effective LO_FLAGS_DIRECT_IO flag and
 198 * not the originally passed in one.
 199 */
 200static inline void loop_update_dio(struct loop_device *lo)
 201{
 202	lockdep_assert_held(&lo->lo_mutex);
 203	WARN_ON_ONCE(lo->lo_state == Lo_bound &&
 204		     lo->lo_queue->mq_freeze_depth == 0);
 205
 206	if ((lo->lo_flags & LO_FLAGS_DIRECT_IO) && !lo_can_use_dio(lo))
 207		lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
 208}
 209
 210/**
 211 * loop_set_size() - sets device size and notifies userspace
 212 * @lo: struct loop_device to set the size for
 213 * @size: new size of the loop device
 214 *
 215 * Callers must validate that the size passed into this function fits into
 216 * a sector_t, eg using loop_validate_size()
 217 */
 218static void loop_set_size(struct loop_device *lo, loff_t size)
 219{
 220	if (!set_capacity_and_notify(lo->lo_disk, size))
 221		kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
 222}
 223
 224static void loop_clear_limits(struct loop_device *lo, int mode)
 225{
 226	struct queue_limits lim = queue_limits_start_update(lo->lo_queue);
 227
 228	if (mode & FALLOC_FL_ZERO_RANGE)
 229		lim.max_write_zeroes_sectors = 0;
 230
 231	if (mode & FALLOC_FL_PUNCH_HOLE) {
 232		lim.max_hw_discard_sectors = 0;
 233		lim.discard_granularity = 0;
 234	}
 235
 236	/*
 237	 * XXX: this updates the queue limits without freezing the queue, which
 238	 * is against the locking protocol and dangerous.  But we can't just
 239	 * freeze the queue as we're inside the ->queue_rq method here.  So this
 240	 * should move out into a workqueue unless we get the file operations to
 241	 * advertise if they support specific fallocate operations.
 242	 */
 243	queue_limits_commit_update(lo->lo_queue, &lim);
 244}
 245
 246static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
 247			int mode)
 248{
 249	/*
 250	 * We use fallocate to manipulate the space mappings used by the image
 251	 * a.k.a. discard/zerorange.
 252	 */
 253	struct file *file = lo->lo_backing_file;
 254	int ret;
 255
 256	mode |= FALLOC_FL_KEEP_SIZE;
 257
 258	if (!bdev_max_discard_sectors(lo->lo_device))
 259		return -EOPNOTSUPP;
 260
 261	ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
 262	if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
 263		return -EIO;
 264
 265	/*
 266	 * We initially configure the limits in a hope that fallocate is
 267	 * supported and clear them here if that turns out not to be true.
 268	 */
 269	if (unlikely(ret == -EOPNOTSUPP))
 270		loop_clear_limits(lo, mode);
 271
 272	return ret;
 273}
 274
 275static int lo_req_flush(struct loop_device *lo, struct request *rq)
 276{
 277	int ret = vfs_fsync(lo->lo_backing_file, 0);
 278	if (unlikely(ret && ret != -EINVAL))
 279		ret = -EIO;
 280
 281	return ret;
 282}
 283
 284static void lo_complete_rq(struct request *rq)
 285{
 286	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
 287	blk_status_t ret = BLK_STS_OK;
 288
 289	if (cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
 290	    req_op(rq) != REQ_OP_READ) {
 291		if (cmd->ret < 0)
 292			ret = errno_to_blk_status(cmd->ret);
 293		goto end_io;
 294	}
 295
 296	/*
 297	 * Short READ - if we got some data, advance our request and
 298	 * retry it. If we got no data, end the rest with EIO.
 299	 */
 300	if (cmd->ret) {
 301		blk_update_request(rq, BLK_STS_OK, cmd->ret);
 302		cmd->ret = 0;
 303		blk_mq_requeue_request(rq, true);
 304	} else {
 305		struct bio *bio = rq->bio;
 306
 307		while (bio) {
 308			zero_fill_bio(bio);
 309			bio = bio->bi_next;
 310		}
 311
 312		ret = BLK_STS_IOERR;
 313end_io:
 314		blk_mq_end_request(rq, ret);
 315	}
 316}
 317
 318static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
 319{
 320	struct request *rq = blk_mq_rq_from_pdu(cmd);
 321
 322	if (!atomic_dec_and_test(&cmd->ref))
 323		return;
 324	kfree(cmd->bvec);
 325	cmd->bvec = NULL;
 326	if (req_op(rq) == REQ_OP_WRITE)
 327		kiocb_end_write(&cmd->iocb);
 328	if (likely(!blk_should_fake_timeout(rq->q)))
 329		blk_mq_complete_request(rq);
 330}
 331
 332static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
 333{
 334	struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
 335
 336	cmd->ret = ret;
 337	lo_rw_aio_do_completion(cmd);
 338}
 339
 340static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 341		     loff_t pos, int rw)
 342{
 343	struct iov_iter iter;
 344	struct req_iterator rq_iter;
 345	struct bio_vec *bvec;
 346	struct request *rq = blk_mq_rq_from_pdu(cmd);
 347	struct bio *bio = rq->bio;
 348	struct file *file = lo->lo_backing_file;
 349	struct bio_vec tmp;
 350	unsigned int offset;
 351	unsigned int nr_bvec;
 352	int ret;
 353
 354	nr_bvec = blk_rq_nr_bvec(rq);
 355
 356	if (rq->bio != rq->biotail) {
 357
 358		bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
 359				     GFP_NOIO);
 360		if (!bvec)
 361			return -EIO;
 362		cmd->bvec = bvec;
 363
 364		/*
 365		 * The bios of the request may be started from the middle of
 366		 * the 'bvec' because of bio splitting, so we can't directly
 367		 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
 368		 * API will take care of all details for us.
 369		 */
 370		rq_for_each_bvec(tmp, rq, rq_iter) {
 371			*bvec = tmp;
 372			bvec++;
 373		}
 374		bvec = cmd->bvec;
 375		offset = 0;
 376	} else {
 377		/*
 378		 * Same here, this bio may be started from the middle of the
 379		 * 'bvec' because of bio splitting, so offset from the bvec
 380		 * must be passed to iov iterator
 381		 */
 382		offset = bio->bi_iter.bi_bvec_done;
 383		bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
 384	}
 385	atomic_set(&cmd->ref, 2);
 386
 387	iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
 388	iter.iov_offset = offset;
 389
 390	cmd->iocb.ki_pos = pos;
 391	cmd->iocb.ki_filp = file;
 392	cmd->iocb.ki_ioprio = req_get_ioprio(rq);
 393	if (cmd->use_aio) {
 394		cmd->iocb.ki_complete = lo_rw_aio_complete;
 395		cmd->iocb.ki_flags = IOCB_DIRECT;
 396	} else {
 397		cmd->iocb.ki_complete = NULL;
 398		cmd->iocb.ki_flags = 0;
 399	}
 400
 401	if (rw == ITER_SOURCE) {
 402		kiocb_start_write(&cmd->iocb);
 403		ret = file->f_op->write_iter(&cmd->iocb, &iter);
 404	} else
 405		ret = file->f_op->read_iter(&cmd->iocb, &iter);
 406
 407	lo_rw_aio_do_completion(cmd);
 408
 409	if (ret != -EIOCBQUEUED)
 410		lo_rw_aio_complete(&cmd->iocb, ret);
 411	return -EIOCBQUEUED;
 412}
 413
 414static int do_req_filebacked(struct loop_device *lo, struct request *rq)
 415{
 416	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
 417	loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
 418
 419	switch (req_op(rq)) {
 420	case REQ_OP_FLUSH:
 421		return lo_req_flush(lo, rq);
 422	case REQ_OP_WRITE_ZEROES:
 423		/*
 424		 * If the caller doesn't want deallocation, call zeroout to
 425		 * write zeroes the range.  Otherwise, punch them out.
 426		 */
 427		return lo_fallocate(lo, rq, pos,
 428			(rq->cmd_flags & REQ_NOUNMAP) ?
 429				FALLOC_FL_ZERO_RANGE :
 430				FALLOC_FL_PUNCH_HOLE);
 431	case REQ_OP_DISCARD:
 432		return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
 433	case REQ_OP_WRITE:
 434		return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
 435	case REQ_OP_READ:
 436		return lo_rw_aio(lo, cmd, pos, ITER_DEST);
 437	default:
 438		WARN_ON_ONCE(1);
 439		return -EIO;
 440	}
 441}
 442
 443static void loop_reread_partitions(struct loop_device *lo)
 444{
 445	int rc;
 446
 447	mutex_lock(&lo->lo_disk->open_mutex);
 448	rc = bdev_disk_changed(lo->lo_disk, false);
 449	mutex_unlock(&lo->lo_disk->open_mutex);
 450	if (rc)
 451		pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
 452			__func__, lo->lo_number, lo->lo_file_name, rc);
 453}
 454
 455static unsigned int loop_query_min_dio_size(struct loop_device *lo)
 456{
 457	struct file *file = lo->lo_backing_file;
 458	struct block_device *sb_bdev = file->f_mapping->host->i_sb->s_bdev;
 459	struct kstat st;
 460
 461	/*
 462	 * Use the minimal dio alignment of the file system if provided.
 463	 */
 464	if (!vfs_getattr(&file->f_path, &st, STATX_DIOALIGN, 0) &&
 465	    (st.result_mask & STATX_DIOALIGN))
 466		return st.dio_offset_align;
 467
 468	/*
 469	 * In a perfect world this wouldn't be needed, but as of Linux 6.13 only
 470	 * a handful of file systems support the STATX_DIOALIGN flag.
 471	 */
 472	if (sb_bdev)
 473		return bdev_logical_block_size(sb_bdev);
 474	return SECTOR_SIZE;
 475}
 476
 477static inline int is_loop_device(struct file *file)
 478{
 479	struct inode *i = file->f_mapping->host;
 480
 481	return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
 482}
 483
 484static int loop_validate_file(struct file *file, struct block_device *bdev)
 485{
 486	struct inode	*inode = file->f_mapping->host;
 487	struct file	*f = file;
 488
 489	/* Avoid recursion */
 490	while (is_loop_device(f)) {
 491		struct loop_device *l;
 492
 493		lockdep_assert_held(&loop_validate_mutex);
 494		if (f->f_mapping->host->i_rdev == bdev->bd_dev)
 495			return -EBADF;
 496
 497		l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
 498		if (l->lo_state != Lo_bound)
 499			return -EINVAL;
 500		/* Order wrt setting lo->lo_backing_file in loop_configure(). */
 501		rmb();
 502		f = l->lo_backing_file;
 503	}
 504	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 505		return -EINVAL;
 506	return 0;
 507}
 508
 509static void loop_assign_backing_file(struct loop_device *lo, struct file *file)
 510{
 511	lo->lo_backing_file = file;
 512	lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
 513	mapping_set_gfp_mask(file->f_mapping,
 514			lo->old_gfp_mask & ~(__GFP_IO | __GFP_FS));
 515	if (lo->lo_backing_file->f_flags & O_DIRECT)
 516		lo->lo_flags |= LO_FLAGS_DIRECT_IO;
 517	lo->lo_min_dio_size = loop_query_min_dio_size(lo);
 518}
 519
 520static int loop_check_backing_file(struct file *file)
 521{
 522	if (!file->f_op->read_iter)
 523		return -EINVAL;
 524
 525	if ((file->f_mode & FMODE_WRITE) && !file->f_op->write_iter)
 526		return -EINVAL;
 527
 528	return 0;
 529}
 530
 531/*
 532 * loop_change_fd switched the backing store of a loopback device to
 533 * a new file. This is useful for operating system installers to free up
 534 * the original file and in High Availability environments to switch to
 535 * an alternative location for the content in case of server meltdown.
 536 * This can only work if the loop device is used read-only, and if the
 537 * new backing store is the same size and type as the old backing store.
 538 */
 539static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
 540			  unsigned int arg)
 541{
 542	struct file *file = fget(arg);
 543	struct file *old_file;
 544	unsigned int memflags;
 545	int error;
 546	bool partscan;
 547	bool is_loop;
 548
 549	if (!file)
 550		return -EBADF;
 551
 552	error = loop_check_backing_file(file);
 553	if (error) {
 554		fput(file);
 555		return error;
 556	}
 557
 558	/* suppress uevents while reconfiguring the device */
 559	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
 560
 561	is_loop = is_loop_device(file);
 562	error = loop_global_lock_killable(lo, is_loop);
 563	if (error)
 564		goto out_putf;
 565	error = -ENXIO;
 566	if (lo->lo_state != Lo_bound)
 567		goto out_err;
 568
 569	/* the loop device has to be read-only */
 570	error = -EINVAL;
 571	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
 572		goto out_err;
 573
 574	error = loop_validate_file(file, bdev);
 575	if (error)
 576		goto out_err;
 577
 578	old_file = lo->lo_backing_file;
 579
 580	error = -EINVAL;
 581
 582	/* size of the new backing store needs to be the same */
 583	if (lo_calculate_size(lo, file) != lo_calculate_size(lo, old_file))
 584		goto out_err;
 585
 586	/*
 587	 * We might switch to direct I/O mode for the loop device, write back
 588	 * all dirty data the page cache now that so that the individual I/O
 589	 * operations don't have to do that.
 590	 */
 591	vfs_fsync(file, 0);
 592
 593	/* and ... switch */
 594	disk_force_media_change(lo->lo_disk);
 595	memflags = blk_mq_freeze_queue(lo->lo_queue);
 596	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
 597	loop_assign_backing_file(lo, file);
 598	loop_update_dio(lo);
 599	blk_mq_unfreeze_queue(lo->lo_queue, memflags);
 600	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
 601	loop_global_unlock(lo, is_loop);
 602
 603	/*
 604	 * Flush loop_validate_file() before fput(), for l->lo_backing_file
 605	 * might be pointing at old_file which might be the last reference.
 606	 */
 607	if (!is_loop) {
 608		mutex_lock(&loop_validate_mutex);
 609		mutex_unlock(&loop_validate_mutex);
 610	}
 611	/*
 612	 * We must drop file reference outside of lo_mutex as dropping
 613	 * the file ref can take open_mutex which creates circular locking
 614	 * dependency.
 615	 */
 616	fput(old_file);
 617	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
 618	if (partscan)
 619		loop_reread_partitions(lo);
 620
 621	error = 0;
 622done:
 623	kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
 624	return error;
 625
 626out_err:
 627	loop_global_unlock(lo, is_loop);
 628out_putf:
 629	fput(file);
 630	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
 631	goto done;
 632}
 633
 634/* loop sysfs attributes */
 635
 636static ssize_t loop_attr_show(struct device *dev, char *page,
 637			      ssize_t (*callback)(struct loop_device *, char *))
 638{
 639	struct gendisk *disk = dev_to_disk(dev);
 640	struct loop_device *lo = disk->private_data;
 641
 642	return callback(lo, page);
 643}
 644
 645#define LOOP_ATTR_RO(_name)						\
 646static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);	\
 647static ssize_t loop_attr_do_show_##_name(struct device *d,		\
 648				struct device_attribute *attr, char *b)	\
 649{									\
 650	return loop_attr_show(d, b, loop_attr_##_name##_show);		\
 651}									\
 652static struct device_attribute loop_attr_##_name =			\
 653	__ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
 654
 655static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
 656{
 657	ssize_t ret;
 658	char *p = NULL;
 659
 660	spin_lock_irq(&lo->lo_lock);
 661	if (lo->lo_backing_file)
 662		p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
 663	spin_unlock_irq(&lo->lo_lock);
 664
 665	if (IS_ERR_OR_NULL(p))
 666		ret = PTR_ERR(p);
 667	else {
 668		ret = strlen(p);
 669		memmove(buf, p, ret);
 670		buf[ret++] = '\n';
 671		buf[ret] = 0;
 672	}
 673
 674	return ret;
 675}
 676
 677static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
 678{
 679	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
 680}
 681
 682static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
 683{
 684	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
 685}
 686
 687static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
 688{
 689	int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
 690
 691	return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
 692}
 693
 694static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
 695{
 696	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
 697
 698	return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
 699}
 700
 701static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
 702{
 703	int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
 704
 705	return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
 706}
 707
 708LOOP_ATTR_RO(backing_file);
 709LOOP_ATTR_RO(offset);
 710LOOP_ATTR_RO(sizelimit);
 711LOOP_ATTR_RO(autoclear);
 712LOOP_ATTR_RO(partscan);
 713LOOP_ATTR_RO(dio);
 714
 715static struct attribute *loop_attrs[] = {
 716	&loop_attr_backing_file.attr,
 717	&loop_attr_offset.attr,
 718	&loop_attr_sizelimit.attr,
 719	&loop_attr_autoclear.attr,
 720	&loop_attr_partscan.attr,
 721	&loop_attr_dio.attr,
 722	NULL,
 723};
 724
 725static struct attribute_group loop_attribute_group = {
 726	.name = "loop",
 727	.attrs= loop_attrs,
 728};
 729
 730static void loop_sysfs_init(struct loop_device *lo)
 731{
 732	lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
 733						&loop_attribute_group);
 734}
 735
 736static void loop_sysfs_exit(struct loop_device *lo)
 737{
 738	if (lo->sysfs_inited)
 739		sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
 740				   &loop_attribute_group);
 741}
 742
 743static void loop_get_discard_config(struct loop_device *lo,
 744				    u32 *granularity, u32 *max_discard_sectors)
 745{
 746	struct file *file = lo->lo_backing_file;
 747	struct inode *inode = file->f_mapping->host;
 748	struct kstatfs sbuf;
 749
 750	/*
 751	 * If the backing device is a block device, mirror its zeroing
 752	 * capability. Set the discard sectors to the block device's zeroing
 753	 * capabilities because loop discards result in blkdev_issue_zeroout(),
 754	 * not blkdev_issue_discard(). This maintains consistent behavior with
 755	 * file-backed loop devices: discarded regions read back as zero.
 756	 */
 757	if (S_ISBLK(inode->i_mode)) {
 758		struct block_device *bdev = I_BDEV(inode);
 759
 760		*max_discard_sectors = bdev_write_zeroes_sectors(bdev);
 761		*granularity = bdev_discard_granularity(bdev);
 762
 763	/*
 764	 * We use punch hole to reclaim the free space used by the
 765	 * image a.k.a. discard.
 766	 */
 767	} else if (file->f_op->fallocate && !vfs_statfs(&file->f_path, &sbuf)) {
 768		*max_discard_sectors = UINT_MAX >> 9;
 769		*granularity = sbuf.f_bsize;
 770	}
 771}
 772
 773struct loop_worker {
 774	struct rb_node rb_node;
 775	struct work_struct work;
 776	struct list_head cmd_list;
 777	struct list_head idle_list;
 778	struct loop_device *lo;
 779	struct cgroup_subsys_state *blkcg_css;
 780	unsigned long last_ran_at;
 781};
 782
 783static void loop_workfn(struct work_struct *work);
 784
 785#ifdef CONFIG_BLK_CGROUP
 786static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
 787{
 788	return !css || css == blkcg_root_css;
 789}
 790#else
 791static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
 792{
 793	return !css;
 794}
 795#endif
 796
 797static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
 798{
 799	struct rb_node **node, *parent = NULL;
 800	struct loop_worker *cur_worker, *worker = NULL;
 801	struct work_struct *work;
 802	struct list_head *cmd_list;
 803
 804	spin_lock_irq(&lo->lo_work_lock);
 805
 806	if (queue_on_root_worker(cmd->blkcg_css))
 807		goto queue_work;
 808
 809	node = &lo->worker_tree.rb_node;
 810
 811	while (*node) {
 812		parent = *node;
 813		cur_worker = container_of(*node, struct loop_worker, rb_node);
 814		if (cur_worker->blkcg_css == cmd->blkcg_css) {
 815			worker = cur_worker;
 816			break;
 817		} else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
 818			node = &(*node)->rb_left;
 819		} else {
 820			node = &(*node)->rb_right;
 821		}
 822	}
 823	if (worker)
 824		goto queue_work;
 825
 826	worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT);
 827	/*
 828	 * In the event we cannot allocate a worker, just queue on the
 829	 * rootcg worker and issue the I/O as the rootcg
 830	 */
 831	if (!worker) {
 832		cmd->blkcg_css = NULL;
 833		if (cmd->memcg_css)
 834			css_put(cmd->memcg_css);
 835		cmd->memcg_css = NULL;
 836		goto queue_work;
 837	}
 838
 839	worker->blkcg_css = cmd->blkcg_css;
 840	css_get(worker->blkcg_css);
 841	INIT_WORK(&worker->work, loop_workfn);
 842	INIT_LIST_HEAD(&worker->cmd_list);
 843	INIT_LIST_HEAD(&worker->idle_list);
 844	worker->lo = lo;
 845	rb_link_node(&worker->rb_node, parent, node);
 846	rb_insert_color(&worker->rb_node, &lo->worker_tree);
 847queue_work:
 848	if (worker) {
 849		/*
 850		 * We need to remove from the idle list here while
 851		 * holding the lock so that the idle timer doesn't
 852		 * free the worker
 853		 */
 854		if (!list_empty(&worker->idle_list))
 855			list_del_init(&worker->idle_list);
 856		work = &worker->work;
 857		cmd_list = &worker->cmd_list;
 858	} else {
 859		work = &lo->rootcg_work;
 860		cmd_list = &lo->rootcg_cmd_list;
 861	}
 862	list_add_tail(&cmd->list_entry, cmd_list);
 863	queue_work(lo->workqueue, work);
 864	spin_unlock_irq(&lo->lo_work_lock);
 865}
 866
 867static void loop_set_timer(struct loop_device *lo)
 868{
 869	timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
 870}
 871
 872static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
 873{
 874	struct loop_worker *pos, *worker;
 875
 876	spin_lock_irq(&lo->lo_work_lock);
 877	list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
 878				idle_list) {
 879		if (!delete_all &&
 880		    time_is_after_jiffies(worker->last_ran_at +
 881					  LOOP_IDLE_WORKER_TIMEOUT))
 882			break;
 883		list_del(&worker->idle_list);
 884		rb_erase(&worker->rb_node, &lo->worker_tree);
 885		css_put(worker->blkcg_css);
 886		kfree(worker);
 887	}
 888	if (!list_empty(&lo->idle_worker_list))
 889		loop_set_timer(lo);
 890	spin_unlock_irq(&lo->lo_work_lock);
 891}
 892
 893static void loop_free_idle_workers_timer(struct timer_list *timer)
 894{
 895	struct loop_device *lo = container_of(timer, struct loop_device, timer);
 896
 897	return loop_free_idle_workers(lo, false);
 898}
 899
 900/**
 901 * loop_set_status_from_info - configure device from loop_info
 902 * @lo: struct loop_device to configure
 903 * @info: struct loop_info64 to configure the device with
 904 *
 905 * Configures the loop device parameters according to the passed
 906 * in loop_info64 configuration.
 907 */
 908static int
 909loop_set_status_from_info(struct loop_device *lo,
 910			  const struct loop_info64 *info)
 911{
 912	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
 913		return -EINVAL;
 914
 915	switch (info->lo_encrypt_type) {
 916	case LO_CRYPT_NONE:
 917		break;
 918	case LO_CRYPT_XOR:
 919		pr_warn("support for the xor transformation has been removed.\n");
 920		return -EINVAL;
 921	case LO_CRYPT_CRYPTOAPI:
 922		pr_warn("support for cryptoloop has been removed.  Use dm-crypt instead.\n");
 923		return -EINVAL;
 924	default:
 925		return -EINVAL;
 926	}
 927
 928	/* Avoid assigning overflow values */
 929	if (info->lo_offset > LLONG_MAX || info->lo_sizelimit > LLONG_MAX)
 930		return -EOVERFLOW;
 931
 932	lo->lo_offset = info->lo_offset;
 933	lo->lo_sizelimit = info->lo_sizelimit;
 934
 935	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
 936	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
 937	return 0;
 938}
 939
 940static unsigned int loop_default_blocksize(struct loop_device *lo)
 941{
 942	/* In case of direct I/O, match underlying minimum I/O size */
 943	if (lo->lo_flags & LO_FLAGS_DIRECT_IO)
 944		return lo->lo_min_dio_size;
 945	return SECTOR_SIZE;
 946}
 947
 948static void loop_update_limits(struct loop_device *lo, struct queue_limits *lim,
 949		unsigned int bsize)
 950{
 951	struct file *file = lo->lo_backing_file;
 952	struct inode *inode = file->f_mapping->host;
 953	struct block_device *backing_bdev = NULL;
 954	u32 granularity = 0, max_discard_sectors = 0;
 955
 956	if (S_ISBLK(inode->i_mode))
 957		backing_bdev = I_BDEV(inode);
 958	else if (inode->i_sb->s_bdev)
 959		backing_bdev = inode->i_sb->s_bdev;
 960
 961	if (!bsize)
 962		bsize = loop_default_blocksize(lo);
 963
 964	loop_get_discard_config(lo, &granularity, &max_discard_sectors);
 965
 966	lim->logical_block_size = bsize;
 967	lim->physical_block_size = bsize;
 968	lim->io_min = bsize;
 969	lim->features &= ~(BLK_FEAT_WRITE_CACHE | BLK_FEAT_ROTATIONAL);
 970	if (file->f_op->fsync && !(lo->lo_flags & LO_FLAGS_READ_ONLY))
 971		lim->features |= BLK_FEAT_WRITE_CACHE;
 972	if (backing_bdev && !bdev_nonrot(backing_bdev))
 973		lim->features |= BLK_FEAT_ROTATIONAL;
 974	lim->max_hw_discard_sectors = max_discard_sectors;
 975	lim->max_write_zeroes_sectors = max_discard_sectors;
 976	if (max_discard_sectors)
 977		lim->discard_granularity = granularity;
 978	else
 979		lim->discard_granularity = 0;
 980}
 981
 982static int loop_configure(struct loop_device *lo, blk_mode_t mode,
 983			  struct block_device *bdev,
 984			  const struct loop_config *config)
 985{
 986	struct file *file = fget(config->fd);
 987	struct queue_limits lim;
 988	int error;
 989	loff_t size;
 990	bool partscan;
 991	bool is_loop;
 992
 993	if (!file)
 994		return -EBADF;
 995
 996	error = loop_check_backing_file(file);
 997	if (error) {
 998		fput(file);
 999		return error;
1000	}
1001
1002	is_loop = is_loop_device(file);
1003
1004	/* This is safe, since we have a reference from open(). */
1005	__module_get(THIS_MODULE);
1006
1007	/*
1008	 * If we don't hold exclusive handle for the device, upgrade to it
1009	 * here to avoid changing device under exclusive owner.
1010	 */
1011	if (!(mode & BLK_OPEN_EXCL)) {
1012		error = bd_prepare_to_claim(bdev, loop_configure, NULL);
1013		if (error)
1014			goto out_putf;
1015	}
1016
1017	error = loop_global_lock_killable(lo, is_loop);
1018	if (error)
1019		goto out_bdev;
1020
1021	error = -EBUSY;
1022	if (lo->lo_state != Lo_unbound)
1023		goto out_unlock;
1024
1025	error = loop_validate_file(file, bdev);
1026	if (error)
1027		goto out_unlock;
1028
1029	if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1030		error = -EINVAL;
1031		goto out_unlock;
1032	}
1033
1034	error = loop_set_status_from_info(lo, &config->info);
1035	if (error)
1036		goto out_unlock;
1037	lo->lo_flags = config->info.lo_flags;
1038
1039	if (!(file->f_mode & FMODE_WRITE) || !(mode & BLK_OPEN_WRITE) ||
1040	    !file->f_op->write_iter)
1041		lo->lo_flags |= LO_FLAGS_READ_ONLY;
1042
1043	if (!lo->workqueue) {
1044		lo->workqueue = alloc_workqueue("loop%d",
1045						WQ_UNBOUND | WQ_FREEZABLE,
1046						0, lo->lo_number);
1047		if (!lo->workqueue) {
1048			error = -ENOMEM;
1049			goto out_unlock;
1050		}
1051	}
1052
1053	/* suppress uevents while reconfiguring the device */
1054	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1055
1056	disk_force_media_change(lo->lo_disk);
1057	set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1058
1059	lo->lo_device = bdev;
1060	loop_assign_backing_file(lo, file);
1061
1062	lim = queue_limits_start_update(lo->lo_queue);
1063	loop_update_limits(lo, &lim, config->block_size);
1064	/* No need to freeze the queue as the device isn't bound yet. */
1065	error = queue_limits_commit_update(lo->lo_queue, &lim);
1066	if (error)
1067		goto out_unlock;
1068
1069	/*
1070	 * We might switch to direct I/O mode for the loop device, write back
1071	 * all dirty data the page cache now that so that the individual I/O
1072	 * operations don't have to do that.
1073	 */
1074	vfs_fsync(file, 0);
1075
1076	loop_update_dio(lo);
1077	loop_sysfs_init(lo);
1078
1079	size = lo_calculate_size(lo, file);
1080	loop_set_size(lo, size);
1081
1082	/* Order wrt reading lo_state in loop_validate_file(). */
1083	wmb();
1084
1085	WRITE_ONCE(lo->lo_state, Lo_bound);
1086	if (part_shift)
1087		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1088	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1089	if (partscan)
1090		clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1091
1092	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1093	kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1094
1095	loop_global_unlock(lo, is_loop);
1096	if (partscan)
1097		loop_reread_partitions(lo);
1098
1099	if (!(mode & BLK_OPEN_EXCL))
1100		bd_abort_claiming(bdev, loop_configure);
1101
1102	return 0;
1103
1104out_unlock:
1105	loop_global_unlock(lo, is_loop);
1106out_bdev:
1107	if (!(mode & BLK_OPEN_EXCL))
1108		bd_abort_claiming(bdev, loop_configure);
1109out_putf:
1110	fput(file);
1111	/* This is safe: open() is still holding a reference. */
1112	module_put(THIS_MODULE);
1113	return error;
1114}
1115
1116static void __loop_clr_fd(struct loop_device *lo)
1117{
1118	struct queue_limits lim;
1119	struct file *filp;
1120	gfp_t gfp = lo->old_gfp_mask;
1121
1122	spin_lock_irq(&lo->lo_lock);
1123	filp = lo->lo_backing_file;
1124	lo->lo_backing_file = NULL;
1125	spin_unlock_irq(&lo->lo_lock);
1126
1127	lo->lo_device = NULL;
1128	lo->lo_offset = 0;
1129	lo->lo_sizelimit = 0;
1130	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1131
1132	/*
1133	 * Reset the block size to the default.
1134	 *
1135	 * No queue freezing needed because this is called from the final
1136	 * ->release call only, so there can't be any outstanding I/O.
1137	 */
1138	lim = queue_limits_start_update(lo->lo_queue);
1139	lim.logical_block_size = SECTOR_SIZE;
1140	lim.physical_block_size = SECTOR_SIZE;
1141	lim.io_min = SECTOR_SIZE;
1142	queue_limits_commit_update(lo->lo_queue, &lim);
1143
1144	invalidate_disk(lo->lo_disk);
1145	loop_sysfs_exit(lo);
1146	/* let user-space know about this change */
1147	kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1148	mapping_set_gfp_mask(filp->f_mapping, gfp);
1149	/* This is safe: open() is still holding a reference. */
1150	module_put(THIS_MODULE);
1151
1152	disk_force_media_change(lo->lo_disk);
1153
1154	if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1155		int err;
1156
1157		/*
1158		 * open_mutex has been held already in release path, so don't
1159		 * acquire it if this function is called in such case.
1160		 *
1161		 * If the reread partition isn't from release path, lo_refcnt
1162		 * must be at least one and it can only become zero when the
1163		 * current holder is released.
1164		 */
1165		err = bdev_disk_changed(lo->lo_disk, false);
1166		if (err)
1167			pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1168				__func__, lo->lo_number, err);
1169		/* Device is gone, no point in returning error */
1170	}
1171
1172	/*
1173	 * lo->lo_state is set to Lo_unbound here after above partscan has
1174	 * finished. There cannot be anybody else entering __loop_clr_fd() as
1175	 * Lo_rundown state protects us from all the other places trying to
1176	 * change the 'lo' device.
1177	 */
1178	lo->lo_flags = 0;
1179	if (!part_shift)
1180		set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1181	mutex_lock(&lo->lo_mutex);
1182	WRITE_ONCE(lo->lo_state, Lo_unbound);
1183	mutex_unlock(&lo->lo_mutex);
1184
1185	/*
1186	 * Need not hold lo_mutex to fput backing file. Calling fput holding
1187	 * lo_mutex triggers a circular lock dependency possibility warning as
1188	 * fput can take open_mutex which is usually taken before lo_mutex.
1189	 */
1190	fput(filp);
1191}
1192
1193static int loop_clr_fd(struct loop_device *lo)
1194{
1195	int err;
1196
1197	/*
1198	 * Since lo_ioctl() is called without locks held, it is possible that
1199	 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1200	 *
1201	 * Therefore, use global lock when setting Lo_rundown state in order to
1202	 * make sure that loop_validate_file() will fail if the "struct file"
1203	 * which loop_configure()/loop_change_fd() found via fget() was this
1204	 * loop device.
1205	 */
1206	err = loop_global_lock_killable(lo, true);
1207	if (err)
1208		return err;
1209	if (lo->lo_state != Lo_bound) {
1210		loop_global_unlock(lo, true);
1211		return -ENXIO;
1212	}
1213	/*
1214	 * Mark the device for removing the backing device on last close.
1215	 * If we are the only opener, also switch the state to roundown here to
1216	 * prevent new openers from coming in.
1217	 */
1218
1219	lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1220	if (disk_openers(lo->lo_disk) == 1)
1221		WRITE_ONCE(lo->lo_state, Lo_rundown);
1222	loop_global_unlock(lo, true);
1223
1224	return 0;
1225}
1226
1227static int
1228loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1229{
1230	int err;
1231	bool partscan = false;
1232	bool size_changed = false;
1233	unsigned int memflags;
1234
1235	err = mutex_lock_killable(&lo->lo_mutex);
1236	if (err)
1237		return err;
1238	if (lo->lo_state != Lo_bound) {
1239		err = -ENXIO;
1240		goto out_unlock;
1241	}
1242
1243	if (lo->lo_offset != info->lo_offset ||
1244	    lo->lo_sizelimit != info->lo_sizelimit) {
1245		size_changed = true;
1246		sync_blockdev(lo->lo_device);
1247		invalidate_bdev(lo->lo_device);
1248	}
1249
1250	/* I/O needs to be drained before changing lo_offset or lo_sizelimit */
1251	memflags = blk_mq_freeze_queue(lo->lo_queue);
1252
1253	err = loop_set_status_from_info(lo, info);
1254	if (err)
1255		goto out_unfreeze;
1256
1257	partscan = !(lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1258		(info->lo_flags & LO_FLAGS_PARTSCAN);
1259
1260	lo->lo_flags &= ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1261	lo->lo_flags |= (info->lo_flags & LOOP_SET_STATUS_SETTABLE_FLAGS);
1262
1263	/* update the direct I/O flag if lo_offset changed */
1264	loop_update_dio(lo);
1265
1266out_unfreeze:
1267	blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1268	if (partscan)
1269		clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1270	if (!err && size_changed) {
1271		loff_t new_size = lo_calculate_size(lo, lo->lo_backing_file);
1272		loop_set_size(lo, new_size);
1273	}
1274out_unlock:
1275	mutex_unlock(&lo->lo_mutex);
1276	if (partscan)
1277		loop_reread_partitions(lo);
1278
1279	return err;
1280}
1281
1282static int
1283loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1284{
1285	struct path path;
1286	struct kstat stat;
1287	int ret;
1288
1289	ret = mutex_lock_killable(&lo->lo_mutex);
1290	if (ret)
1291		return ret;
1292	if (lo->lo_state != Lo_bound) {
1293		mutex_unlock(&lo->lo_mutex);
1294		return -ENXIO;
1295	}
1296
1297	memset(info, 0, sizeof(*info));
1298	info->lo_number = lo->lo_number;
1299	info->lo_offset = lo->lo_offset;
1300	info->lo_sizelimit = lo->lo_sizelimit;
1301	info->lo_flags = lo->lo_flags;
1302	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1303
1304	/* Drop lo_mutex while we call into the filesystem. */
1305	path = lo->lo_backing_file->f_path;
1306	path_get(&path);
1307	mutex_unlock(&lo->lo_mutex);
1308	ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1309	if (!ret) {
1310		info->lo_device = huge_encode_dev(stat.dev);
1311		info->lo_inode = stat.ino;
1312		info->lo_rdevice = huge_encode_dev(stat.rdev);
1313	}
1314	path_put(&path);
1315	return ret;
1316}
1317
1318static void
1319loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1320{
1321	memset(info64, 0, sizeof(*info64));
1322	info64->lo_number = info->lo_number;
1323	info64->lo_device = info->lo_device;
1324	info64->lo_inode = info->lo_inode;
1325	info64->lo_rdevice = info->lo_rdevice;
1326	info64->lo_offset = info->lo_offset;
1327	info64->lo_sizelimit = 0;
1328	info64->lo_flags = info->lo_flags;
1329	memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1330}
1331
1332static int
1333loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1334{
1335	memset(info, 0, sizeof(*info));
1336	info->lo_number = info64->lo_number;
1337	info->lo_device = info64->lo_device;
1338	info->lo_inode = info64->lo_inode;
1339	info->lo_rdevice = info64->lo_rdevice;
1340	info->lo_offset = info64->lo_offset;
1341	info->lo_flags = info64->lo_flags;
1342	memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1343
1344	/* error in case values were truncated */
1345	if (info->lo_device != info64->lo_device ||
1346	    info->lo_rdevice != info64->lo_rdevice ||
1347	    info->lo_inode != info64->lo_inode ||
1348	    info->lo_offset != info64->lo_offset)
1349		return -EOVERFLOW;
1350
1351	return 0;
1352}
1353
1354static int
1355loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1356{
1357	struct loop_info info;
1358	struct loop_info64 info64;
1359
1360	if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1361		return -EFAULT;
1362	loop_info64_from_old(&info, &info64);
1363	return loop_set_status(lo, &info64);
1364}
1365
1366static int
1367loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1368{
1369	struct loop_info64 info64;
1370
1371	if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1372		return -EFAULT;
1373	return loop_set_status(lo, &info64);
1374}
1375
1376static int
1377loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1378	struct loop_info info;
1379	struct loop_info64 info64;
1380	int err;
1381
1382	if (!arg)
1383		return -EINVAL;
1384	err = loop_get_status(lo, &info64);
1385	if (!err)
1386		err = loop_info64_to_old(&info64, &info);
1387	if (!err && copy_to_user(arg, &info, sizeof(info)))
1388		err = -EFAULT;
1389
1390	return err;
1391}
1392
1393static int
1394loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1395	struct loop_info64 info64;
1396	int err;
1397
1398	if (!arg)
1399		return -EINVAL;
1400	err = loop_get_status(lo, &info64);
1401	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1402		err = -EFAULT;
1403
1404	return err;
1405}
1406
1407static int loop_set_capacity(struct loop_device *lo)
1408{
1409	loff_t size;
1410
1411	if (unlikely(lo->lo_state != Lo_bound))
1412		return -ENXIO;
1413
1414	size = lo_calculate_size(lo, lo->lo_backing_file);
1415	loop_set_size(lo, size);
1416
1417	return 0;
1418}
1419
1420static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1421{
1422	bool use_dio = !!arg;
1423	unsigned int memflags;
1424
1425	if (lo->lo_state != Lo_bound)
1426		return -ENXIO;
1427	if (use_dio == !!(lo->lo_flags & LO_FLAGS_DIRECT_IO))
1428		return 0;
1429
1430	if (use_dio) {
1431		if (!lo_can_use_dio(lo))
1432			return -EINVAL;
1433		/* flush dirty pages before starting to use direct I/O */
1434		vfs_fsync(lo->lo_backing_file, 0);
1435	}
1436
1437	memflags = blk_mq_freeze_queue(lo->lo_queue);
1438	if (use_dio)
1439		lo->lo_flags |= LO_FLAGS_DIRECT_IO;
1440	else
1441		lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
1442	blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1443	return 0;
1444}
1445
1446static int loop_set_block_size(struct loop_device *lo, blk_mode_t mode,
1447			       struct block_device *bdev, unsigned long arg)
1448{
1449	struct queue_limits lim;
1450	unsigned int memflags;
1451	int err = 0;
1452
1453	/*
1454	 * If we don't hold exclusive handle for the device, upgrade to it
1455	 * here to avoid changing device under exclusive owner.
1456	 */
1457	if (!(mode & BLK_OPEN_EXCL)) {
1458		err = bd_prepare_to_claim(bdev, loop_set_block_size, NULL);
1459		if (err)
1460			return err;
1461	}
1462
1463	err = mutex_lock_killable(&lo->lo_mutex);
1464	if (err)
1465		goto abort_claim;
1466
1467	if (lo->lo_state != Lo_bound) {
1468		err = -ENXIO;
1469		goto unlock;
1470	}
1471
1472	if (lo->lo_queue->limits.logical_block_size == arg)
1473		goto unlock;
1474
1475	sync_blockdev(lo->lo_device);
1476	invalidate_bdev(lo->lo_device);
1477
1478	lim = queue_limits_start_update(lo->lo_queue);
1479	loop_update_limits(lo, &lim, arg);
1480
1481	memflags = blk_mq_freeze_queue(lo->lo_queue);
1482	err = queue_limits_commit_update(lo->lo_queue, &lim);
1483	loop_update_dio(lo);
1484	blk_mq_unfreeze_queue(lo->lo_queue, memflags);
1485
1486unlock:
1487	mutex_unlock(&lo->lo_mutex);
1488abort_claim:
1489	if (!(mode & BLK_OPEN_EXCL))
1490		bd_abort_claiming(bdev, loop_set_block_size);
1491	return err;
1492}
1493
1494static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1495			   unsigned long arg)
1496{
1497	int err;
1498
1499	err = mutex_lock_killable(&lo->lo_mutex);
1500	if (err)
1501		return err;
1502	switch (cmd) {
1503	case LOOP_SET_CAPACITY:
1504		err = loop_set_capacity(lo);
1505		break;
1506	case LOOP_SET_DIRECT_IO:
1507		err = loop_set_dio(lo, arg);
1508		break;
1509	default:
1510		err = -EINVAL;
1511	}
1512	mutex_unlock(&lo->lo_mutex);
1513	return err;
1514}
1515
1516static int lo_ioctl(struct block_device *bdev, blk_mode_t mode,
1517	unsigned int cmd, unsigned long arg)
1518{
1519	struct loop_device *lo = bdev->bd_disk->private_data;
1520	void __user *argp = (void __user *) arg;
1521	int err;
1522
1523	switch (cmd) {
1524	case LOOP_SET_FD: {
1525		/*
1526		 * Legacy case - pass in a zeroed out struct loop_config with
1527		 * only the file descriptor set , which corresponds with the
1528		 * default parameters we'd have used otherwise.
1529		 */
1530		struct loop_config config;
1531
1532		memset(&config, 0, sizeof(config));
1533		config.fd = arg;
1534
1535		return loop_configure(lo, mode, bdev, &config);
1536	}
1537	case LOOP_CONFIGURE: {
1538		struct loop_config config;
1539
1540		if (copy_from_user(&config, argp, sizeof(config)))
1541			return -EFAULT;
1542
1543		return loop_configure(lo, mode, bdev, &config);
1544	}
1545	case LOOP_CHANGE_FD:
1546		return loop_change_fd(lo, bdev, arg);
1547	case LOOP_CLR_FD:
1548		return loop_clr_fd(lo);
1549	case LOOP_SET_STATUS:
1550		err = -EPERM;
1551		if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1552			err = loop_set_status_old(lo, argp);
1553		break;
1554	case LOOP_GET_STATUS:
1555		return loop_get_status_old(lo, argp);
1556	case LOOP_SET_STATUS64:
1557		err = -EPERM;
1558		if ((mode & BLK_OPEN_WRITE) || capable(CAP_SYS_ADMIN))
1559			err = loop_set_status64(lo, argp);
1560		break;
1561	case LOOP_GET_STATUS64:
1562		return loop_get_status64(lo, argp);
1563	case LOOP_SET_BLOCK_SIZE:
1564		if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1565			return -EPERM;
1566		return loop_set_block_size(lo, mode, bdev, arg);
1567	case LOOP_SET_CAPACITY:
1568	case LOOP_SET_DIRECT_IO:
1569		if (!(mode & BLK_OPEN_WRITE) && !capable(CAP_SYS_ADMIN))
1570			return -EPERM;
1571		fallthrough;
1572	default:
1573		err = lo_simple_ioctl(lo, cmd, arg);
1574		break;
1575	}
1576
1577	return err;
1578}
1579
1580#ifdef CONFIG_COMPAT
1581struct compat_loop_info {
1582	compat_int_t	lo_number;      /* ioctl r/o */
1583	compat_dev_t	lo_device;      /* ioctl r/o */
1584	compat_ulong_t	lo_inode;       /* ioctl r/o */
1585	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1586	compat_int_t	lo_offset;
1587	compat_int_t	lo_encrypt_type;        /* obsolete, ignored */
1588	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1589	compat_int_t	lo_flags;       /* ioctl r/o */
1590	char		lo_name[LO_NAME_SIZE];
1591	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1592	compat_ulong_t	lo_init[2];
1593	char		reserved[4];
1594};
1595
1596/*
1597 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1598 * - noinlined to reduce stack space usage in main part of driver
1599 */
1600static noinline int
1601loop_info64_from_compat(const struct compat_loop_info __user *arg,
1602			struct loop_info64 *info64)
1603{
1604	struct compat_loop_info info;
1605
1606	if (copy_from_user(&info, arg, sizeof(info)))
1607		return -EFAULT;
1608
1609	memset(info64, 0, sizeof(*info64));
1610	info64->lo_number = info.lo_number;
1611	info64->lo_device = info.lo_device;
1612	info64->lo_inode = info.lo_inode;
1613	info64->lo_rdevice = info.lo_rdevice;
1614	info64->lo_offset = info.lo_offset;
1615	info64->lo_sizelimit = 0;
1616	info64->lo_flags = info.lo_flags;
1617	memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1618	return 0;
1619}
1620
1621/*
1622 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1623 * - noinlined to reduce stack space usage in main part of driver
1624 */
1625static noinline int
1626loop_info64_to_compat(const struct loop_info64 *info64,
1627		      struct compat_loop_info __user *arg)
1628{
1629	struct compat_loop_info info;
1630
1631	memset(&info, 0, sizeof(info));
1632	info.lo_number = info64->lo_number;
1633	info.lo_device = info64->lo_device;
1634	info.lo_inode = info64->lo_inode;
1635	info.lo_rdevice = info64->lo_rdevice;
1636	info.lo_offset = info64->lo_offset;
1637	info.lo_flags = info64->lo_flags;
1638	memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1639
1640	/* error in case values were truncated */
1641	if (info.lo_device != info64->lo_device ||
1642	    info.lo_rdevice != info64->lo_rdevice ||
1643	    info.lo_inode != info64->lo_inode ||
1644	    info.lo_offset != info64->lo_offset)
1645		return -EOVERFLOW;
1646
1647	if (copy_to_user(arg, &info, sizeof(info)))
1648		return -EFAULT;
1649	return 0;
1650}
1651
1652static int
1653loop_set_status_compat(struct loop_device *lo,
1654		       const struct compat_loop_info __user *arg)
1655{
1656	struct loop_info64 info64;
1657	int ret;
1658
1659	ret = loop_info64_from_compat(arg, &info64);
1660	if (ret < 0)
1661		return ret;
1662	return loop_set_status(lo, &info64);
1663}
1664
1665static int
1666loop_get_status_compat(struct loop_device *lo,
1667		       struct compat_loop_info __user *arg)
1668{
1669	struct loop_info64 info64;
1670	int err;
1671
1672	if (!arg)
1673		return -EINVAL;
1674	err = loop_get_status(lo, &info64);
1675	if (!err)
1676		err = loop_info64_to_compat(&info64, arg);
1677	return err;
1678}
1679
1680static int lo_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
1681			   unsigned int cmd, unsigned long arg)
1682{
1683	struct loop_device *lo = bdev->bd_disk->private_data;
1684	int err;
1685
1686	switch(cmd) {
1687	case LOOP_SET_STATUS:
1688		err = loop_set_status_compat(lo,
1689			     (const struct compat_loop_info __user *)arg);
1690		break;
1691	case LOOP_GET_STATUS:
1692		err = loop_get_status_compat(lo,
1693				     (struct compat_loop_info __user *)arg);
1694		break;
1695	case LOOP_SET_CAPACITY:
1696	case LOOP_CLR_FD:
1697	case LOOP_GET_STATUS64:
1698	case LOOP_SET_STATUS64:
1699	case LOOP_CONFIGURE:
1700		arg = (unsigned long) compat_ptr(arg);
1701		fallthrough;
1702	case LOOP_SET_FD:
1703	case LOOP_CHANGE_FD:
1704	case LOOP_SET_BLOCK_SIZE:
1705	case LOOP_SET_DIRECT_IO:
1706		err = lo_ioctl(bdev, mode, cmd, arg);
1707		break;
1708	default:
1709		err = -ENOIOCTLCMD;
1710		break;
1711	}
1712	return err;
1713}
1714#endif
1715
1716static int lo_open(struct gendisk *disk, blk_mode_t mode)
1717{
1718	struct loop_device *lo = disk->private_data;
1719	int err;
1720
1721	err = mutex_lock_killable(&lo->lo_mutex);
1722	if (err)
1723		return err;
1724
1725	if (lo->lo_state == Lo_deleting || lo->lo_state == Lo_rundown)
1726		err = -ENXIO;
1727	mutex_unlock(&lo->lo_mutex);
1728	return err;
1729}
1730
1731static void lo_release(struct gendisk *disk)
1732{
1733	struct loop_device *lo = disk->private_data;
1734	bool need_clear = false;
1735
1736	if (disk_openers(disk) > 0)
1737		return;
1738	/*
1739	 * Clear the backing device information if this is the last close of
1740	 * a device that's been marked for auto clear, or on which LOOP_CLR_FD
1741	 * has been called.
1742	 */
1743
1744	mutex_lock(&lo->lo_mutex);
1745	if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR))
1746		WRITE_ONCE(lo->lo_state, Lo_rundown);
1747
1748	need_clear = (lo->lo_state == Lo_rundown);
1749	mutex_unlock(&lo->lo_mutex);
1750
1751	if (need_clear)
1752		__loop_clr_fd(lo);
1753}
1754
1755static void lo_free_disk(struct gendisk *disk)
1756{
1757	struct loop_device *lo = disk->private_data;
1758
1759	if (lo->workqueue)
1760		destroy_workqueue(lo->workqueue);
1761	loop_free_idle_workers(lo, true);
1762	timer_shutdown_sync(&lo->timer);
1763	mutex_destroy(&lo->lo_mutex);
1764	kfree(lo);
1765}
1766
1767static const struct block_device_operations lo_fops = {
1768	.owner =	THIS_MODULE,
1769	.open =         lo_open,
1770	.release =	lo_release,
1771	.ioctl =	lo_ioctl,
1772#ifdef CONFIG_COMPAT
1773	.compat_ioctl =	lo_compat_ioctl,
1774#endif
1775	.free_disk =	lo_free_disk,
1776};
1777
1778/*
1779 * And now the modules code and kernel interface.
1780 */
1781
1782/*
1783 * If max_loop is specified, create that many devices upfront.
1784 * This also becomes a hard limit. If max_loop is not specified,
1785 * the default isn't a hard limit (as before commit 85c50197716c
1786 * changed the default value from 0 for max_loop=0 reasons), just
1787 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1788 * init time. Loop devices can be requested on-demand with the
1789 * /dev/loop-control interface, or be instantiated by accessing
1790 * a 'dead' device node.
1791 */
1792static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1793
1794#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
1795static bool max_loop_specified;
1796
1797static int max_loop_param_set_int(const char *val,
1798				  const struct kernel_param *kp)
1799{
1800	int ret;
1801
1802	ret = param_set_int(val, kp);
1803	if (ret < 0)
1804		return ret;
1805
1806	max_loop_specified = true;
1807	return 0;
1808}
1809
1810static const struct kernel_param_ops max_loop_param_ops = {
1811	.set = max_loop_param_set_int,
1812	.get = param_get_int,
1813};
1814
1815module_param_cb(max_loop, &max_loop_param_ops, &max_loop, 0444);
1816MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1817#else
1818module_param(max_loop, int, 0444);
1819MODULE_PARM_DESC(max_loop, "Initial number of loop devices");
1820#endif
1821
1822module_param(max_part, int, 0444);
1823MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1824
1825static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1826
1827static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1828{
1829	int qd, ret;
1830
1831	ret = kstrtoint(s, 0, &qd);
1832	if (ret < 0)
1833		return ret;
1834	if (qd < 1)
1835		return -EINVAL;
1836	hw_queue_depth = qd;
1837	return 0;
1838}
1839
1840static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1841	.set	= loop_set_hw_queue_depth,
1842	.get	= param_get_int,
1843};
1844
1845device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1846MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: " __stringify(LOOP_DEFAULT_HW_Q_DEPTH));
1847
1848MODULE_DESCRIPTION("Loopback device support");
1849MODULE_LICENSE("GPL");
1850MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1851
1852static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1853		const struct blk_mq_queue_data *bd)
1854{
1855	struct request *rq = bd->rq;
1856	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1857	struct loop_device *lo = rq->q->queuedata;
1858
1859	blk_mq_start_request(rq);
1860
1861	if (data_race(READ_ONCE(lo->lo_state)) != Lo_bound)
1862		return BLK_STS_IOERR;
1863
1864	switch (req_op(rq)) {
1865	case REQ_OP_FLUSH:
1866	case REQ_OP_DISCARD:
1867	case REQ_OP_WRITE_ZEROES:
1868		cmd->use_aio = false;
1869		break;
1870	default:
1871		cmd->use_aio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1872		break;
1873	}
1874
1875	/* always use the first bio's css */
1876	cmd->blkcg_css = NULL;
1877	cmd->memcg_css = NULL;
1878#ifdef CONFIG_BLK_CGROUP
1879	if (rq->bio) {
1880		cmd->blkcg_css = bio_blkcg_css(rq->bio);
1881#ifdef CONFIG_MEMCG
1882		if (cmd->blkcg_css) {
1883			cmd->memcg_css =
1884				cgroup_get_e_css(cmd->blkcg_css->cgroup,
1885						&memory_cgrp_subsys);
1886		}
1887#endif
1888	}
1889#endif
1890	loop_queue_work(lo, cmd);
1891
1892	return BLK_STS_OK;
1893}
1894
1895static void loop_handle_cmd(struct loop_cmd *cmd)
1896{
1897	struct cgroup_subsys_state *cmd_blkcg_css = cmd->blkcg_css;
1898	struct cgroup_subsys_state *cmd_memcg_css = cmd->memcg_css;
1899	struct request *rq = blk_mq_rq_from_pdu(cmd);
1900	const bool write = op_is_write(req_op(rq));
1901	struct loop_device *lo = rq->q->queuedata;
1902	int ret = 0;
1903	struct mem_cgroup *old_memcg = NULL;
1904
1905	if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1906		ret = -EIO;
1907		goto failed;
1908	}
1909
1910	/* We can block in this context, so ignore REQ_NOWAIT. */
1911	if (rq->cmd_flags & REQ_NOWAIT)
1912		rq->cmd_flags &= ~REQ_NOWAIT;
1913
1914	if (cmd_blkcg_css)
1915		kthread_associate_blkcg(cmd_blkcg_css);
1916	if (cmd_memcg_css)
1917		old_memcg = set_active_memcg(
1918			mem_cgroup_from_css(cmd_memcg_css));
1919
1920	/*
1921	 * do_req_filebacked() may call blk_mq_complete_request() synchronously
1922	 * or asynchronously if using aio. Hence, do not touch 'cmd' after
1923	 * do_req_filebacked() has returned unless we are sure that 'cmd' has
1924	 * not yet been completed.
1925	 */
1926	ret = do_req_filebacked(lo, rq);
1927
1928	if (cmd_blkcg_css)
1929		kthread_associate_blkcg(NULL);
1930
1931	if (cmd_memcg_css) {
1932		set_active_memcg(old_memcg);
1933		css_put(cmd_memcg_css);
1934	}
1935 failed:
1936	/* complete non-aio request */
1937	if (ret != -EIOCBQUEUED) {
1938		if (ret == -EOPNOTSUPP)
1939			cmd->ret = ret;
1940		else
1941			cmd->ret = ret ? -EIO : 0;
1942		if (likely(!blk_should_fake_timeout(rq->q)))
1943			blk_mq_complete_request(rq);
1944	}
1945}
1946
1947static void loop_process_work(struct loop_worker *worker,
1948			struct list_head *cmd_list, struct loop_device *lo)
1949{
1950	int orig_flags = current->flags;
1951	struct loop_cmd *cmd;
1952
1953	current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1954	spin_lock_irq(&lo->lo_work_lock);
1955	while (!list_empty(cmd_list)) {
1956		cmd = container_of(
1957			cmd_list->next, struct loop_cmd, list_entry);
1958		list_del(cmd_list->next);
1959		spin_unlock_irq(&lo->lo_work_lock);
1960
1961		loop_handle_cmd(cmd);
1962		cond_resched();
1963
1964		spin_lock_irq(&lo->lo_work_lock);
1965	}
1966
1967	/*
1968	 * We only add to the idle list if there are no pending cmds
1969	 * *and* the worker will not run again which ensures that it
1970	 * is safe to free any worker on the idle list
1971	 */
1972	if (worker && !work_pending(&worker->work)) {
1973		worker->last_ran_at = jiffies;
1974		list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1975		loop_set_timer(lo);
1976	}
1977	spin_unlock_irq(&lo->lo_work_lock);
1978	current->flags = orig_flags;
1979}
1980
1981static void loop_workfn(struct work_struct *work)
1982{
1983	struct loop_worker *worker =
1984		container_of(work, struct loop_worker, work);
1985	loop_process_work(worker, &worker->cmd_list, worker->lo);
1986}
1987
1988static void loop_rootcg_workfn(struct work_struct *work)
1989{
1990	struct loop_device *lo =
1991		container_of(work, struct loop_device, rootcg_work);
1992	loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1993}
1994
1995static const struct blk_mq_ops loop_mq_ops = {
1996	.queue_rq       = loop_queue_rq,
1997	.complete	= lo_complete_rq,
1998};
1999
2000static int loop_add(int i)
2001{
2002	struct queue_limits lim = {
2003		/*
2004		 * Random number picked from the historic block max_sectors cap.
2005		 */
2006		.max_hw_sectors		= 2560u,
2007	};
2008	struct loop_device *lo;
2009	struct gendisk *disk;
2010	int err;
2011
2012	err = -ENOMEM;
2013	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
2014	if (!lo)
2015		goto out;
2016	lo->worker_tree = RB_ROOT;
2017	INIT_LIST_HEAD(&lo->idle_worker_list);
2018	timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
2019	WRITE_ONCE(lo->lo_state, Lo_unbound);
2020
2021	err = mutex_lock_killable(&loop_ctl_mutex);
2022	if (err)
2023		goto out_free_dev;
2024
2025	/* allocate id, if @id >= 0, we're requesting that specific id */
2026	if (i >= 0) {
2027		err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
2028		if (err == -ENOSPC)
2029			err = -EEXIST;
2030	} else {
2031		err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
2032	}
2033	mutex_unlock(&loop_ctl_mutex);
2034	if (err < 0)
2035		goto out_free_dev;
2036	i = err;
2037
2038	lo->tag_set.ops = &loop_mq_ops;
2039	lo->tag_set.nr_hw_queues = 1;
2040	lo->tag_set.queue_depth = hw_queue_depth;
2041	lo->tag_set.numa_node = NUMA_NO_NODE;
2042	lo->tag_set.cmd_size = sizeof(struct loop_cmd);
2043	lo->tag_set.flags = BLK_MQ_F_STACKING | BLK_MQ_F_NO_SCHED_BY_DEFAULT;
2044	lo->tag_set.driver_data = lo;
2045
2046	err = blk_mq_alloc_tag_set(&lo->tag_set);
2047	if (err)
2048		goto out_free_idr;
2049
2050	disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, &lim, lo);
2051	if (IS_ERR(disk)) {
2052		err = PTR_ERR(disk);
2053		goto out_cleanup_tags;
2054	}
2055	lo->lo_queue = lo->lo_disk->queue;
2056
2057	/*
2058	 * Disable partition scanning by default. The in-kernel partition
2059	 * scanning can be requested individually per-device during its
2060	 * setup. Userspace can always add and remove partitions from all
2061	 * devices. The needed partition minors are allocated from the
2062	 * extended minor space, the main loop device numbers will continue
2063	 * to match the loop minors, regardless of the number of partitions
2064	 * used.
2065	 *
2066	 * If max_part is given, partition scanning is globally enabled for
2067	 * all loop devices. The minors for the main loop devices will be
2068	 * multiples of max_part.
2069	 *
2070	 * Note: Global-for-all-devices, set-only-at-init, read-only module
2071	 * parameteters like 'max_loop' and 'max_part' make things needlessly
2072	 * complicated, are too static, inflexible and may surprise
2073	 * userspace tools. Parameters like this in general should be avoided.
2074	 */
2075	if (!part_shift)
2076		set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2077	mutex_init(&lo->lo_mutex);
2078	lo->lo_number		= i;
2079	spin_lock_init(&lo->lo_lock);
2080	spin_lock_init(&lo->lo_work_lock);
2081	INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2082	INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2083	disk->major		= LOOP_MAJOR;
2084	disk->first_minor	= i << part_shift;
2085	disk->minors		= 1 << part_shift;
2086	disk->fops		= &lo_fops;
2087	disk->private_data	= lo;
2088	disk->queue		= lo->lo_queue;
2089	disk->events		= DISK_EVENT_MEDIA_CHANGE;
2090	disk->event_flags	= DISK_EVENT_FLAG_UEVENT;
2091	sprintf(disk->disk_name, "loop%d", i);
2092	/* Make this loop device reachable from pathname. */
2093	err = add_disk(disk);
2094	if (err)
2095		goto out_cleanup_disk;
2096
2097	/* Show this loop device. */
2098	mutex_lock(&loop_ctl_mutex);
2099	lo->idr_visible = true;
2100	mutex_unlock(&loop_ctl_mutex);
2101
2102	return i;
2103
2104out_cleanup_disk:
2105	put_disk(disk);
2106out_cleanup_tags:
2107	blk_mq_free_tag_set(&lo->tag_set);
2108out_free_idr:
2109	mutex_lock(&loop_ctl_mutex);
2110	idr_remove(&loop_index_idr, i);
2111	mutex_unlock(&loop_ctl_mutex);
2112out_free_dev:
2113	kfree(lo);
2114out:
2115	return err;
2116}
2117
2118static void loop_remove(struct loop_device *lo)
2119{
2120	/* Make this loop device unreachable from pathname. */
2121	del_gendisk(lo->lo_disk);
2122	blk_mq_free_tag_set(&lo->tag_set);
2123
2124	mutex_lock(&loop_ctl_mutex);
2125	idr_remove(&loop_index_idr, lo->lo_number);
2126	mutex_unlock(&loop_ctl_mutex);
2127
2128	put_disk(lo->lo_disk);
2129}
2130
2131#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2132static void loop_probe(dev_t dev)
2133{
2134	int idx = MINOR(dev) >> part_shift;
2135
2136	if (max_loop_specified && max_loop && idx >= max_loop)
2137		return;
2138	loop_add(idx);
2139}
2140#else
2141#define loop_probe NULL
2142#endif /* !CONFIG_BLOCK_LEGACY_AUTOLOAD */
2143
2144static int loop_control_remove(int idx)
2145{
2146	struct loop_device *lo;
2147	int ret;
2148
2149	if (idx < 0) {
2150		pr_warn_once("deleting an unspecified loop device is not supported.\n");
2151		return -EINVAL;
2152	}
2153		
2154	/* Hide this loop device for serialization. */
2155	ret = mutex_lock_killable(&loop_ctl_mutex);
2156	if (ret)
2157		return ret;
2158	lo = idr_find(&loop_index_idr, idx);
2159	if (!lo || !lo->idr_visible)
2160		ret = -ENODEV;
2161	else
2162		lo->idr_visible = false;
2163	mutex_unlock(&loop_ctl_mutex);
2164	if (ret)
2165		return ret;
2166
2167	/* Check whether this loop device can be removed. */
2168	ret = mutex_lock_killable(&lo->lo_mutex);
2169	if (ret)
2170		goto mark_visible;
2171	if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2172		mutex_unlock(&lo->lo_mutex);
2173		ret = -EBUSY;
2174		goto mark_visible;
2175	}
2176	/* Mark this loop device as no more bound, but not quite unbound yet */
2177	WRITE_ONCE(lo->lo_state, Lo_deleting);
2178	mutex_unlock(&lo->lo_mutex);
2179
2180	loop_remove(lo);
2181	return 0;
2182
2183mark_visible:
2184	/* Show this loop device again. */
2185	mutex_lock(&loop_ctl_mutex);
2186	lo->idr_visible = true;
2187	mutex_unlock(&loop_ctl_mutex);
2188	return ret;
2189}
2190
2191static int loop_control_get_free(int idx)
2192{
2193	struct loop_device *lo;
2194	int id, ret;
2195
2196	ret = mutex_lock_killable(&loop_ctl_mutex);
2197	if (ret)
2198		return ret;
2199	idr_for_each_entry(&loop_index_idr, lo, id) {
2200		/*
2201		 * Hitting a race results in creating a new loop device
2202		 * which is harmless.
2203		 */
2204		if (lo->idr_visible &&
2205		    data_race(READ_ONCE(lo->lo_state)) == Lo_unbound)
2206			goto found;
2207	}
2208	mutex_unlock(&loop_ctl_mutex);
2209	return loop_add(-1);
2210found:
2211	mutex_unlock(&loop_ctl_mutex);
2212	return id;
2213}
2214
2215static long loop_control_ioctl(struct file *file, unsigned int cmd,
2216			       unsigned long parm)
2217{
2218	switch (cmd) {
2219	case LOOP_CTL_ADD:
2220		return loop_add(parm);
2221	case LOOP_CTL_REMOVE:
2222		return loop_control_remove(parm);
2223	case LOOP_CTL_GET_FREE:
2224		return loop_control_get_free(parm);
2225	default:
2226		return -ENOSYS;
2227	}
2228}
2229
2230static const struct file_operations loop_ctl_fops = {
2231	.open		= nonseekable_open,
2232	.unlocked_ioctl	= loop_control_ioctl,
2233	.compat_ioctl	= loop_control_ioctl,
2234	.owner		= THIS_MODULE,
2235	.llseek		= noop_llseek,
2236};
2237
2238static struct miscdevice loop_misc = {
2239	.minor		= LOOP_CTRL_MINOR,
2240	.name		= "loop-control",
2241	.fops		= &loop_ctl_fops,
2242};
2243
2244MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2245MODULE_ALIAS("devname:loop-control");
2246
2247static int __init loop_init(void)
2248{
2249	int i;
2250	int err;
2251
2252	part_shift = 0;
2253	if (max_part > 0) {
2254		part_shift = fls(max_part);
2255
2256		/*
2257		 * Adjust max_part according to part_shift as it is exported
2258		 * to user space so that user can decide correct minor number
2259		 * if [s]he want to create more devices.
2260		 *
2261		 * Note that -1 is required because partition 0 is reserved
2262		 * for the whole disk.
2263		 */
2264		max_part = (1UL << part_shift) - 1;
2265	}
2266
2267	if ((1UL << part_shift) > DISK_MAX_PARTS) {
2268		err = -EINVAL;
2269		goto err_out;
2270	}
2271
2272	if (max_loop > 1UL << (MINORBITS - part_shift)) {
2273		err = -EINVAL;
2274		goto err_out;
2275	}
2276
2277	err = misc_register(&loop_misc);
2278	if (err < 0)
2279		goto err_out;
2280
2281
2282	if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2283		err = -EIO;
2284		goto misc_out;
2285	}
2286
2287	/* pre-create number of devices given by config or max_loop */
2288	for (i = 0; i < max_loop; i++)
2289		loop_add(i);
2290
2291	printk(KERN_INFO "loop: module loaded\n");
2292	return 0;
2293
2294misc_out:
2295	misc_deregister(&loop_misc);
2296err_out:
2297	return err;
2298}
2299
2300static void __exit loop_exit(void)
2301{
2302	struct loop_device *lo;
2303	int id;
2304
2305	unregister_blkdev(LOOP_MAJOR, "loop");
2306	misc_deregister(&loop_misc);
2307
2308	/*
2309	 * There is no need to use loop_ctl_mutex here, for nobody else can
2310	 * access loop_index_idr when this module is unloading (unless forced
2311	 * module unloading is requested). If this is not a clean unloading,
2312	 * we have no means to avoid kernel crash.
2313	 */
2314	idr_for_each_entry(&loop_index_idr, lo, id)
2315		loop_remove(lo);
2316
2317	idr_destroy(&loop_index_idr);
2318}
2319
2320module_init(loop_init);
2321module_exit(loop_exit);
2322
2323#ifndef MODULE
2324static int __init max_loop_setup(char *str)
2325{
2326	max_loop = simple_strtol(str, NULL, 0);
2327#ifdef CONFIG_BLOCK_LEGACY_AUTOLOAD
2328	max_loop_specified = true;
2329#endif
2330	return 1;
2331}
2332
2333__setup("max_loop=", max_loop_setup);
2334#endif