Linux Audio

Check our new training course

Loading...
v3.5.6
 
   1/*
   2 *  linux/drivers/block/loop.c
   3 *
   4 *  Written by Theodore Ts'o, 3/29/93
   5 *
   6 * Copyright 1993 by Theodore Ts'o.  Redistribution of this file is
   7 * permitted under the GNU General Public License.
   8 *
   9 * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993
  10 * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996
  11 *
  12 * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994
  13 * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996
  14 *
  15 * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997
  16 *
  17 * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998
  18 *
  19 * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998
  20 *
  21 * Loadable modules and other fixes by AK, 1998
  22 *
  23 * Make real block number available to downstream transfer functions, enables
  24 * CBC (and relatives) mode encryption requiring unique IVs per data block.
  25 * Reed H. Petty, rhp@draper.net
  26 *
  27 * Maximum number of loop devices now dynamic via max_loop module parameter.
  28 * Russell Kroll <rkroll@exploits.org> 19990701
  29 *
  30 * Maximum number of loop devices when compiled-in now selectable by passing
  31 * max_loop=<1-255> to the kernel on boot.
  32 * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999
  33 *
  34 * Completely rewrite request handling to be make_request_fn style and
  35 * non blocking, pushing work to a helper thread. Lots of fixes from
  36 * Al Viro too.
  37 * Jens Axboe <axboe@suse.de>, Nov 2000
  38 *
  39 * Support up to 256 loop devices
  40 * Heinz Mauelshagen <mge@sistina.com>, Feb 2002
  41 *
  42 * Support for falling back on the write file operation when the address space
  43 * operations write_begin is not available on the backing filesystem.
  44 * Anton Altaparmakov, 16 Feb 2005
  45 *
  46 * Still To Fix:
  47 * - Advisory locking is ignored here.
  48 * - Should use an own CAP_* category instead of CAP_SYS_ADMIN
  49 *
  50 */
  51
  52#include <linux/module.h>
  53#include <linux/moduleparam.h>
  54#include <linux/sched.h>
  55#include <linux/fs.h>
 
  56#include <linux/file.h>
  57#include <linux/stat.h>
  58#include <linux/errno.h>
  59#include <linux/major.h>
  60#include <linux/wait.h>
  61#include <linux/blkdev.h>
  62#include <linux/blkpg.h>
  63#include <linux/init.h>
  64#include <linux/swap.h>
  65#include <linux/slab.h>
  66#include <linux/loop.h>
  67#include <linux/compat.h>
  68#include <linux/suspend.h>
  69#include <linux/freezer.h>
  70#include <linux/mutex.h>
  71#include <linux/writeback.h>
  72#include <linux/completion.h>
  73#include <linux/highmem.h>
  74#include <linux/kthread.h>
  75#include <linux/splice.h>
  76#include <linux/sysfs.h>
  77#include <linux/miscdevice.h>
  78#include <linux/falloc.h>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  79
  80#include <asm/uaccess.h>
  81
  82static DEFINE_IDR(loop_index_idr);
  83static DEFINE_MUTEX(loop_index_mutex);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  84
  85static int max_part;
  86static int part_shift;
 
 
 
 
 
 
 
 
  87
  88/*
  89 * Transfer functions
  90 */
  91static int transfer_none(struct loop_device *lo, int cmd,
  92			 struct page *raw_page, unsigned raw_off,
  93			 struct page *loop_page, unsigned loop_off,
  94			 int size, sector_t real_block)
  95{
  96	char *raw_buf = kmap_atomic(raw_page) + raw_off;
  97	char *loop_buf = kmap_atomic(loop_page) + loop_off;
  98
  99	if (cmd == READ)
 100		memcpy(loop_buf, raw_buf, size);
 101	else
 102		memcpy(raw_buf, loop_buf, size);
 103
 104	kunmap_atomic(loop_buf);
 105	kunmap_atomic(raw_buf);
 106	cond_resched();
 107	return 0;
 108}
 
 
 
 
 
 
 
 
 
 
 109
 110static int transfer_xor(struct loop_device *lo, int cmd,
 111			struct page *raw_page, unsigned raw_off,
 112			struct page *loop_page, unsigned loop_off,
 113			int size, sector_t real_block)
 114{
 115	char *raw_buf = kmap_atomic(raw_page) + raw_off;
 116	char *loop_buf = kmap_atomic(loop_page) + loop_off;
 117	char *in, *out, *key;
 118	int i, keysize;
 119
 120	if (cmd == READ) {
 121		in = raw_buf;
 122		out = loop_buf;
 123	} else {
 124		in = loop_buf;
 125		out = raw_buf;
 126	}
 127
 128	key = lo->lo_encrypt_key;
 129	keysize = lo->lo_encrypt_key_size;
 130	for (i = 0; i < size; i++)
 131		*out++ = *in++ ^ key[(i & 511) % keysize];
 132
 133	kunmap_atomic(loop_buf);
 134	kunmap_atomic(raw_buf);
 135	cond_resched();
 136	return 0;
 137}
 138
 139static int xor_init(struct loop_device *lo, const struct loop_info64 *info)
 
 
 
 
 
 
 140{
 141	if (unlikely(info->lo_encrypt_key_size <= 0))
 142		return -EINVAL;
 143	return 0;
 144}
 145
 146static struct loop_func_table none_funcs = {
 147	.number = LO_CRYPT_NONE,
 148	.transfer = transfer_none,
 149}; 	
 150
 151static struct loop_func_table xor_funcs = {
 152	.number = LO_CRYPT_XOR,
 153	.transfer = transfer_xor,
 154	.init = xor_init
 155}; 	
 156
 157/* xfer_funcs[0] is special - its release function is never called */
 158static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = {
 159	&none_funcs,
 160	&xor_funcs
 161};
 162
 163static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
 164{
 165	loff_t size, loopsize;
 166
 167	/* Compute loopsize in bytes */
 168	size = i_size_read(file->f_mapping->host);
 169	loopsize = size - offset;
 170	/* offset is beyond i_size, wierd but possible */
 
 171	if (loopsize < 0)
 172		return 0;
 173
 174	if (sizelimit > 0 && sizelimit < loopsize)
 175		loopsize = sizelimit;
 176	/*
 177	 * Unfortunately, if we want to do I/O on the device,
 178	 * the number of 512-byte sectors has to fit into a sector_t.
 179	 */
 180	return loopsize >> 9;
 181}
 182
 183static loff_t get_loop_size(struct loop_device *lo, struct file *file)
 184{
 185	return get_size(lo->lo_offset, lo->lo_sizelimit, file);
 186}
 187
 188static int
 189figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit)
 190{
 191	loff_t size = get_size(offset, sizelimit, lo->lo_backing_file);
 192	sector_t x = (sector_t)size;
 
 
 
 
 
 
 
 
 
 193
 194	if (unlikely((loff_t)x != size))
 195		return -EFBIG;
 196	if (lo->lo_offset != offset)
 197		lo->lo_offset = offset;
 198	if (lo->lo_sizelimit != sizelimit)
 199		lo->lo_sizelimit = sizelimit;
 200	set_capacity(lo->lo_disk, x);
 201	return 0;
 202}
 
 
 
 
 
 
 
 
 
 
 203
 204static inline int
 205lo_do_transfer(struct loop_device *lo, int cmd,
 206	       struct page *rpage, unsigned roffs,
 207	       struct page *lpage, unsigned loffs,
 208	       int size, sector_t rblock)
 209{
 210	if (unlikely(!lo->transfer))
 211		return 0;
 212
 213	return lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 214}
 215
 216/**
 217 * __do_lo_send_write - helper for writing data to a loop device
 
 
 218 *
 219 * This helper just factors out common code between do_lo_send_direct_write()
 220 * and do_lo_send_write().
 221 */
 222static int __do_lo_send_write(struct file *file,
 223		u8 *buf, const int len, loff_t pos)
 224{
 
 
 
 
 
 
 
 225	ssize_t bw;
 226	mm_segment_t old_fs = get_fs();
 227
 228	set_fs(get_ds());
 229	bw = file->f_op->write(file, buf, len, &pos);
 230	set_fs(old_fs);
 231	if (likely(bw == len))
 
 
 
 232		return 0;
 233	printk(KERN_ERR "loop: Write error at byte offset %llu, length %i.\n",
 234			(unsigned long long)pos, len);
 
 
 235	if (bw >= 0)
 236		bw = -EIO;
 237	return bw;
 238}
 239
 240/**
 241 * do_lo_send_direct_write - helper for writing data to a loop device
 242 *
 243 * This is the fast, non-transforming version that does not need double
 244 * buffering.
 245 */
 246static int do_lo_send_direct_write(struct loop_device *lo,
 247		struct bio_vec *bvec, loff_t pos, struct page *page)
 248{
 249	ssize_t bw = __do_lo_send_write(lo->lo_backing_file,
 250			kmap(bvec->bv_page) + bvec->bv_offset,
 251			bvec->bv_len, pos);
 252	kunmap(bvec->bv_page);
 253	cond_resched();
 254	return bw;
 255}
 256
 257/**
 258 * do_lo_send_write - helper for writing data to a loop device
 259 *
 260 * This is the slow, transforming version that needs to double buffer the
 261 * data as it cannot do the transformations in place without having direct
 262 * access to the destination pages of the backing file.
 263 */
 264static int do_lo_send_write(struct loop_device *lo, struct bio_vec *bvec,
 265		loff_t pos, struct page *page)
 266{
 267	int ret = lo_do_transfer(lo, WRITE, page, 0, bvec->bv_page,
 268			bvec->bv_offset, bvec->bv_len, pos >> 9);
 269	if (likely(!ret))
 270		return __do_lo_send_write(lo->lo_backing_file,
 271				page_address(page), bvec->bv_len,
 272				pos);
 273	printk(KERN_ERR "loop: Transfer error at byte offset %llu, "
 274			"length %i.\n", (unsigned long long)pos, bvec->bv_len);
 275	if (ret > 0)
 276		ret = -EIO;
 277	return ret;
 278}
 279
 280static int lo_send(struct loop_device *lo, struct bio *bio, loff_t pos)
 281{
 282	int (*do_lo_send)(struct loop_device *, struct bio_vec *, loff_t,
 283			struct page *page);
 284	struct bio_vec *bvec;
 285	struct page *page = NULL;
 286	int i, ret = 0;
 287
 288	if (lo->transfer != transfer_none) {
 289		page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
 290		if (unlikely(!page))
 291			goto fail;
 292		kmap(page);
 293		do_lo_send = do_lo_send_write;
 294	} else {
 295		do_lo_send = do_lo_send_direct_write;
 296	}
 297
 298	bio_for_each_segment(bvec, bio, i) {
 299		ret = do_lo_send(lo, bvec, pos, page);
 300		if (ret < 0)
 301			break;
 302		pos += bvec->bv_len;
 303	}
 304	if (page) {
 305		kunmap(page);
 306		__free_page(page);
 307	}
 308out:
 309	return ret;
 310fail:
 311	printk(KERN_ERR "loop: Failed to allocate temporary page for write.\n");
 312	ret = -ENOMEM;
 313	goto out;
 314}
 315
 316struct lo_read_data {
 317	struct loop_device *lo;
 318	struct page *page;
 319	unsigned offset;
 320	int bsize;
 321};
 322
 323static int
 324lo_splice_actor(struct pipe_inode_info *pipe, struct pipe_buffer *buf,
 325		struct splice_desc *sd)
 326{
 327	struct lo_read_data *p = sd->u.data;
 328	struct loop_device *lo = p->lo;
 329	struct page *page = buf->page;
 330	sector_t IV;
 331	int size;
 332
 333	IV = ((sector_t) page->index << (PAGE_CACHE_SHIFT - 9)) +
 334							(buf->offset >> 9);
 335	size = sd->len;
 336	if (size > p->bsize)
 337		size = p->bsize;
 338
 339	if (lo_do_transfer(lo, READ, page, buf->offset, p->page, p->offset, size, IV)) {
 340		printk(KERN_ERR "loop: transfer error block %ld\n",
 341		       page->index);
 342		size = -EINVAL;
 343	}
 344
 345	flush_dcache_page(p->page);
 346
 347	if (size > 0)
 348		p->offset += size;
 349
 350	return size;
 351}
 
 
 
 
 352
 353static int
 354lo_direct_splice_actor(struct pipe_inode_info *pipe, struct splice_desc *sd)
 355{
 356	return __splice_from_pipe(pipe, sd, lo_splice_actor);
 357}
 358
 359static ssize_t
 360do_lo_receive(struct loop_device *lo,
 361	      struct bio_vec *bvec, int bsize, loff_t pos)
 362{
 363	struct lo_read_data cookie;
 364	struct splice_desc sd;
 365	struct file *file;
 366	ssize_t retval;
 367
 368	cookie.lo = lo;
 369	cookie.page = bvec->bv_page;
 370	cookie.offset = bvec->bv_offset;
 371	cookie.bsize = bsize;
 372
 373	sd.len = 0;
 374	sd.total_len = bvec->bv_len;
 375	sd.flags = 0;
 376	sd.pos = pos;
 377	sd.u.data = &cookie;
 378
 379	file = lo->lo_backing_file;
 380	retval = splice_direct_to_actor(file, &sd, lo_direct_splice_actor);
 381
 382	return retval;
 
 
 
 383}
 384
 385static int
 386lo_receive(struct loop_device *lo, struct bio *bio, int bsize, loff_t pos)
 387{
 388	struct bio_vec *bvec;
 389	ssize_t s;
 390	int i;
 391
 392	bio_for_each_segment(bvec, bio, i) {
 393		s = do_lo_receive(lo, bvec, bsize, pos);
 394		if (s < 0)
 395			return s;
 396
 397		if (s != bvec->bv_len) {
 398			zero_fill_bio(bio);
 399			break;
 400		}
 401		pos += bvec->bv_len;
 402	}
 403	return 0;
 404}
 405
 406static int do_bio_filebacked(struct loop_device *lo, struct bio *bio)
 407{
 408	loff_t pos;
 409	int ret;
 410
 411	pos = ((loff_t) bio->bi_sector << 9) + lo->lo_offset;
 412
 413	if (bio_rw(bio) == WRITE) {
 414		struct file *file = lo->lo_backing_file;
 
 
 
 
 415
 416		if (bio->bi_rw & REQ_FLUSH) {
 417			ret = vfs_fsync(file, 0);
 418			if (unlikely(ret && ret != -EINVAL)) {
 419				ret = -EIO;
 420				goto out;
 421			}
 422		}
 
 
 
 
 423
 424		/*
 425		 * We use punch hole to reclaim the free space used by the
 426		 * image a.k.a. discard. However we do not support discard if
 427		 * encryption is enabled, because it may give an attacker
 428		 * useful information.
 429		 */
 430		if (bio->bi_rw & REQ_DISCARD) {
 431			struct file *file = lo->lo_backing_file;
 432			int mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
 433
 434			if ((!file->f_op->fallocate) ||
 435			    lo->lo_encrypt_key_size) {
 436				ret = -EOPNOTSUPP;
 437				goto out;
 438			}
 439			ret = file->f_op->fallocate(file, mode, pos,
 440						    bio->bi_size);
 441			if (unlikely(ret && ret != -EINVAL &&
 442				     ret != -EOPNOTSUPP))
 443				ret = -EIO;
 444			goto out;
 445		}
 446
 447		ret = lo_send(lo, bio, pos);
 448
 449		if ((bio->bi_rw & REQ_FUA) && !ret) {
 450			ret = vfs_fsync(file, 0);
 451			if (unlikely(ret && ret != -EINVAL))
 452				ret = -EIO;
 453		}
 454	} else
 455		ret = lo_receive(lo, bio, lo->lo_blocksize, pos);
 456
 457out:
 458	return ret;
 459}
 460
 461/*
 462 * Add bio to back of pending list
 463 */
 464static void loop_add_bio(struct loop_device *lo, struct bio *bio)
 465{
 466	bio_list_add(&lo->lo_bio_list, bio);
 467}
 468
 469/*
 470 * Grab first pending buffer
 471 */
 472static struct bio *loop_get_bio(struct loop_device *lo)
 473{
 474	return bio_list_pop(&lo->lo_bio_list);
 475}
 476
 477static void loop_make_request(struct request_queue *q, struct bio *old_bio)
 478{
 479	struct loop_device *lo = q->queuedata;
 480	int rw = bio_rw(old_bio);
 481
 482	if (rw == READA)
 483		rw = READ;
 
 484
 485	BUG_ON(!lo || (rw != READ && rw != WRITE));
 
 
 
 
 
 
 
 
 
 
 
 
 486
 487	spin_lock_irq(&lo->lo_lock);
 488	if (lo->lo_state != Lo_bound)
 489		goto out;
 490	if (unlikely(rw == WRITE && (lo->lo_flags & LO_FLAGS_READ_ONLY)))
 491		goto out;
 492	loop_add_bio(lo, old_bio);
 493	wake_up(&lo->lo_event);
 494	spin_unlock_irq(&lo->lo_lock);
 495	return;
 496
 497out:
 498	spin_unlock_irq(&lo->lo_lock);
 499	bio_io_error(old_bio);
 500}
 501
 502struct switch_request {
 503	struct file *file;
 504	struct completion wait;
 505};
 
 506
 507static void do_loop_switch(struct loop_device *, struct switch_request *);
 508
 509static inline void loop_handle_bio(struct loop_device *lo, struct bio *bio)
 510{
 511	if (unlikely(!bio->bi_bdev)) {
 512		do_loop_switch(lo, bio->bi_private);
 513		bio_put(bio);
 
 
 
 
 
 514	} else {
 515		int ret = do_bio_filebacked(lo, bio);
 516		bio_endio(bio, ret);
 
 
 
 
 
 517	}
 518}
 519
 520/*
 521 * worker thread that handles reads/writes to file backed loop devices,
 522 * to avoid blocking in our make_request_fn. it also does loop decrypting
 523 * on reads for block backed loop, as that is too heavy to do from
 524 * b_end_io context where irqs may be disabled.
 525 *
 526 * Loop explanation:  loop_clr_fd() sets lo_state to Lo_rundown before
 527 * calling kthread_stop().  Therefore once kthread_should_stop() is
 528 * true, make_request will not place any more requests.  Therefore
 529 * once kthread_should_stop() is true and lo_bio is NULL, we are
 530 * done with the loop.
 531 */
 532static int loop_thread(void *data)
 533{
 534	struct loop_device *lo = data;
 535	struct bio *bio;
 536
 537	set_user_nice(current, -20);
 
 
 
 
 538
 539	while (!kthread_should_stop() || !bio_list_empty(&lo->lo_bio_list)) {
 
 
 
 
 
 540
 541		wait_event_interruptible(lo->lo_event,
 542				!bio_list_empty(&lo->lo_bio_list) ||
 543				kthread_should_stop());
 
 544
 545		if (bio_list_empty(&lo->lo_bio_list))
 546			continue;
 547		spin_lock_irq(&lo->lo_lock);
 548		bio = loop_get_bio(lo);
 549		spin_unlock_irq(&lo->lo_lock);
 550
 551		BUG_ON(!bio);
 552		loop_handle_bio(lo, bio);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 553	}
 
 554
 555	return 0;
 
 
 
 556}
 557
 558/*
 559 * loop_switch performs the hard work of switching a backing store.
 560 * First it needs to flush existing IO, it does this by sending a magic
 561 * BIO down the pipe. The completion of this BIO does the actual switch.
 562 */
 563static int loop_switch(struct loop_device *lo, struct file *file)
 564{
 565	struct switch_request w;
 566	struct bio *bio = bio_alloc(GFP_KERNEL, 0);
 567	if (!bio)
 568		return -ENOMEM;
 569	init_completion(&w.wait);
 570	w.file = file;
 571	bio->bi_private = &w;
 572	bio->bi_bdev = NULL;
 573	loop_make_request(lo->lo_queue, bio);
 574	wait_for_completion(&w.wait);
 575	return 0;
 576}
 577
 578/*
 579 * Helper to flush the IOs in loop, but keeping loop thread running
 580 */
 581static int loop_flush(struct loop_device *lo)
 582{
 583	/* loop not yet configured, no running thread, nothing to flush */
 584	if (!lo->lo_thread)
 585		return 0;
 586
 587	return loop_switch(lo, NULL);
 588}
 589
 590/*
 591 * Do the actual switch; called from the BIO completion routine
 592 */
 593static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
 594{
 595	struct file *file = p->file;
 596	struct file *old_file = lo->lo_backing_file;
 597	struct address_space *mapping;
 598
 599	/* if no new file, only flush of queued bios requested */
 600	if (!file)
 601		goto out;
 602
 603	mapping = file->f_mapping;
 604	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
 605	lo->lo_backing_file = file;
 606	lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
 607		mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
 608	lo->old_gfp_mask = mapping_gfp_mask(mapping);
 609	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
 610out:
 611	complete(&p->wait);
 612}
 613
 
 
 
 
 
 
 
 
 
 
 
 614
 615/*
 616 * loop_change_fd switched the backing store of a loopback device to
 617 * a new file. This is useful for operating system installers to free up
 618 * the original file and in High Availability environments to switch to
 619 * an alternative location for the content in case of server meltdown.
 620 * This can only work if the loop device is used read-only, and if the
 621 * new backing store is the same size and type as the old backing store.
 622 */
 623static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
 624			  unsigned int arg)
 625{
 626	struct file	*file, *old_file;
 627	struct inode	*inode;
 628	int		error;
 
 
 
 
 
 629
 
 
 
 
 
 
 
 630	error = -ENXIO;
 631	if (lo->lo_state != Lo_bound)
 632		goto out;
 633
 634	/* the loop device has to be read-only */
 635	error = -EINVAL;
 636	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
 637		goto out;
 638
 639	error = -EBADF;
 640	file = fget(arg);
 641	if (!file)
 642		goto out;
 643
 644	inode = file->f_mapping->host;
 645	old_file = lo->lo_backing_file;
 646
 647	error = -EINVAL;
 648
 649	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 650		goto out_putf;
 651
 652	/* size of the new backing store needs to be the same */
 653	if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
 654		goto out_putf;
 655
 656	/* and ... switch */
 657	error = loop_switch(lo, file);
 658	if (error)
 659		goto out_putf;
 
 
 
 
 
 
 
 
 660
 
 
 
 
 
 
 
 
 
 
 
 
 
 661	fput(old_file);
 662	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
 663		ioctl_by_bdev(bdev, BLKRRPART, 0);
 664	return 0;
 665
 666 out_putf:
 667	fput(file);
 668 out:
 
 669	return error;
 670}
 671
 672static inline int is_loop_device(struct file *file)
 673{
 674	struct inode *i = file->f_mapping->host;
 675
 676	return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
 
 
 
 
 677}
 678
 679/* loop sysfs attributes */
 680
 681static ssize_t loop_attr_show(struct device *dev, char *page,
 682			      ssize_t (*callback)(struct loop_device *, char *))
 683{
 684	struct gendisk *disk = dev_to_disk(dev);
 685	struct loop_device *lo = disk->private_data;
 686
 687	return callback(lo, page);
 688}
 689
 690#define LOOP_ATTR_RO(_name)						\
 691static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);	\
 692static ssize_t loop_attr_do_show_##_name(struct device *d,		\
 693				struct device_attribute *attr, char *b)	\
 694{									\
 695	return loop_attr_show(d, b, loop_attr_##_name##_show);		\
 696}									\
 697static struct device_attribute loop_attr_##_name =			\
 698	__ATTR(_name, S_IRUGO, loop_attr_do_show_##_name, NULL);
 699
 700static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
 701{
 702	ssize_t ret;
 703	char *p = NULL;
 704
 705	spin_lock_irq(&lo->lo_lock);
 706	if (lo->lo_backing_file)
 707		p = d_path(&lo->lo_backing_file->f_path, buf, PAGE_SIZE - 1);
 708	spin_unlock_irq(&lo->lo_lock);
 709
 710	if (IS_ERR_OR_NULL(p))
 711		ret = PTR_ERR(p);
 712	else {
 713		ret = strlen(p);
 714		memmove(buf, p, ret);
 715		buf[ret++] = '\n';
 716		buf[ret] = 0;
 717	}
 718
 719	return ret;
 720}
 721
 722static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
 723{
 724	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset);
 725}
 726
 727static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
 728{
 729	return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
 730}
 731
 732static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
 733{
 734	int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
 735
 736	return sprintf(buf, "%s\n", autoclear ? "1" : "0");
 737}
 738
 739static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
 740{
 741	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
 742
 743	return sprintf(buf, "%s\n", partscan ? "1" : "0");
 
 
 
 
 
 
 
 744}
 745
 746LOOP_ATTR_RO(backing_file);
 747LOOP_ATTR_RO(offset);
 748LOOP_ATTR_RO(sizelimit);
 749LOOP_ATTR_RO(autoclear);
 750LOOP_ATTR_RO(partscan);
 
 751
 752static struct attribute *loop_attrs[] = {
 753	&loop_attr_backing_file.attr,
 754	&loop_attr_offset.attr,
 755	&loop_attr_sizelimit.attr,
 756	&loop_attr_autoclear.attr,
 757	&loop_attr_partscan.attr,
 
 758	NULL,
 759};
 760
 761static struct attribute_group loop_attribute_group = {
 762	.name = "loop",
 763	.attrs= loop_attrs,
 764};
 765
 766static int loop_sysfs_init(struct loop_device *lo)
 767{
 768	return sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
 769				  &loop_attribute_group);
 770}
 771
 772static void loop_sysfs_exit(struct loop_device *lo)
 773{
 774	sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
 775			   &loop_attribute_group);
 
 776}
 777
 778static void loop_config_discard(struct loop_device *lo)
 779{
 780	struct file *file = lo->lo_backing_file;
 781	struct inode *inode = file->f_mapping->host;
 782	struct request_queue *q = lo->lo_queue;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 783
 784	/*
 785	 * We use punch hole to reclaim the free space used by the
 786	 * image a.k.a. discard. However we do support discard if
 787	 * encryption is enabled, because it may give an attacker
 788	 * useful information.
 789	 */
 790	if ((!file->f_op->fallocate) ||
 791	    lo->lo_encrypt_key_size) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 792		q->limits.discard_granularity = 0;
 793		q->limits.discard_alignment = 0;
 794		q->limits.max_discard_sectors = 0;
 795		q->limits.discard_zeroes_data = 0;
 796		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
 797		return;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 798	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 799
 800	q->limits.discard_granularity = inode->i_sb->s_blocksize;
 801	q->limits.discard_alignment = 0;
 802	q->limits.max_discard_sectors = UINT_MAX >> 9;
 803	q->limits.discard_zeroes_data = 1;
 804	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
 
 
 
 805}
 806
 807static int loop_set_fd(struct loop_device *lo, fmode_t mode,
 808		       struct block_device *bdev, unsigned int arg)
 
 
 
 
 
 
 
 
 
 809{
 810	struct file	*file, *f;
 811	struct inode	*inode;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 812	struct address_space *mapping;
 813	unsigned lo_blocksize;
 814	int		lo_flags = 0;
 815	int		error;
 816	loff_t		size;
 
 
 
 
 
 817
 818	/* This is safe, since we have a reference from open(). */
 819	__module_get(THIS_MODULE);
 820
 821	error = -EBADF;
 822	file = fget(arg);
 823	if (!file)
 824		goto out;
 825
 826	error = -EBUSY;
 827	if (lo->lo_state != Lo_unbound)
 828		goto out_putf;
 
 
 
 
 
 
 829
 830	/* Avoid recursion */
 831	f = file;
 832	while (is_loop_device(f)) {
 833		struct loop_device *l;
 834
 835		if (f->f_mapping->host->i_bdev == bdev)
 836			goto out_putf;
 
 837
 838		l = f->f_mapping->host->i_bdev->bd_disk->private_data;
 839		if (l->lo_state == Lo_unbound) {
 840			error = -EINVAL;
 841			goto out_putf;
 842		}
 843		f = l->lo_backing_file;
 844	}
 845
 846	mapping = file->f_mapping;
 847	inode = mapping->host;
 848
 849	error = -EINVAL;
 850	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 851		goto out_putf;
 
 852
 853	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
 854	    !file->f_op->write)
 855		lo_flags |= LO_FLAGS_READ_ONLY;
 
 
 856
 857	lo_blocksize = S_ISBLK(inode->i_mode) ?
 858		inode->i_bdev->bd_block_size : PAGE_SIZE;
 
 859
 860	error = -EFBIG;
 861	size = get_loop_size(lo, file);
 862	if ((loff_t)(sector_t)size != size)
 863		goto out_putf;
 864
 865	error = 0;
 
 
 
 
 
 
 
 
 866
 867	set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0);
 
 868
 869	lo->lo_blocksize = lo_blocksize;
 870	lo->lo_device = bdev;
 871	lo->lo_flags = lo_flags;
 872	lo->lo_backing_file = file;
 873	lo->transfer = transfer_none;
 874	lo->ioctl = NULL;
 875	lo->lo_sizelimit = 0;
 876	lo->old_gfp_mask = mapping_gfp_mask(mapping);
 877	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
 878
 879	bio_list_init(&lo->lo_bio_list);
 
 880
 881	/*
 882	 * set queue make_request_fn, and add limits based on lower level
 883	 * device
 884	 */
 885	blk_queue_make_request(lo->lo_queue, loop_make_request);
 886	lo->lo_queue->queuedata = lo;
 
 887
 888	if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
 889		blk_queue_flush(lo->lo_queue, REQ_FLUSH);
 
 890
 891	set_capacity(lo->lo_disk, size);
 892	bd_set_size(bdev, size << 9);
 
 893	loop_sysfs_init(lo);
 894	/* let user-space know about the new size */
 895	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
 896
 897	set_blocksize(bdev, lo_blocksize);
 
 
 
 
 898
 899	lo->lo_thread = kthread_create(loop_thread, lo, "loop%d",
 900						lo->lo_number);
 901	if (IS_ERR(lo->lo_thread)) {
 902		error = PTR_ERR(lo->lo_thread);
 903		goto out_clr;
 904	}
 905	lo->lo_state = Lo_bound;
 906	wake_up_process(lo->lo_thread);
 907	if (part_shift)
 908		lo->lo_flags |= LO_FLAGS_PARTSCAN;
 909	if (lo->lo_flags & LO_FLAGS_PARTSCAN)
 910		ioctl_by_bdev(bdev, BLKRRPART, 0);
 911	return 0;
 
 
 
 
 
 
 912
 913out_clr:
 914	loop_sysfs_exit(lo);
 915	lo->lo_thread = NULL;
 916	lo->lo_device = NULL;
 917	lo->lo_backing_file = NULL;
 918	lo->lo_flags = 0;
 919	set_capacity(lo->lo_disk, 0);
 920	invalidate_bdev(bdev);
 921	bd_set_size(bdev, 0);
 922	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
 923	mapping_set_gfp_mask(mapping, lo->old_gfp_mask);
 924	lo->lo_state = Lo_unbound;
 925 out_putf:
 926	fput(file);
 927 out:
 928	/* This is safe: open() is still holding a reference. */
 929	module_put(THIS_MODULE);
 930	return error;
 931}
 932
 933static int
 934loop_release_xfer(struct loop_device *lo)
 935{
 936	int err = 0;
 937	struct loop_func_table *xfer = lo->lo_encryption;
 938
 939	if (xfer) {
 940		if (xfer->release)
 941			err = xfer->release(lo);
 942		lo->transfer = NULL;
 943		lo->lo_encryption = NULL;
 944		module_put(xfer->owner);
 945	}
 946	return err;
 947}
 948
 949static int
 950loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
 951	       const struct loop_info64 *i)
 952{
 953	int err = 0;
 954
 955	if (xfer) {
 956		struct module *owner = xfer->owner;
 957
 958		if (!try_module_get(owner))
 959			return -EINVAL;
 960		if (xfer->init)
 961			err = xfer->init(lo, i);
 962		if (err)
 963			module_put(owner);
 964		else
 965			lo->lo_encryption = xfer;
 966	}
 967	return err;
 968}
 969
 970static int loop_clr_fd(struct loop_device *lo)
 971{
 972	struct file *filp = lo->lo_backing_file;
 973	gfp_t gfp = lo->old_gfp_mask;
 974	struct block_device *bdev = lo->lo_device;
 975
 976	if (lo->lo_state != Lo_bound)
 977		return -ENXIO;
 978
 979	if (lo->lo_refcnt > 1)	/* we needed one fd for the ioctl */
 980		return -EBUSY;
 981
 982	if (filp == NULL)
 983		return -EINVAL;
 984
 985	spin_lock_irq(&lo->lo_lock);
 986	lo->lo_state = Lo_rundown;
 987	spin_unlock_irq(&lo->lo_lock);
 988
 989	kthread_stop(lo->lo_thread);
 
 
 990
 991	spin_lock_irq(&lo->lo_lock);
 
 992	lo->lo_backing_file = NULL;
 993	spin_unlock_irq(&lo->lo_lock);
 994
 995	loop_release_xfer(lo);
 996	lo->transfer = NULL;
 997	lo->ioctl = NULL;
 998	lo->lo_device = NULL;
 999	lo->lo_encryption = NULL;
1000	lo->lo_offset = 0;
1001	lo->lo_sizelimit = 0;
1002	lo->lo_encrypt_key_size = 0;
1003	lo->lo_thread = NULL;
1004	memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE);
1005	memset(lo->lo_crypt_name, 0, LO_NAME_SIZE);
1006	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1007	if (bdev)
1008		invalidate_bdev(bdev);
1009	set_capacity(lo->lo_disk, 0);
 
1010	loop_sysfs_exit(lo);
1011	if (bdev) {
1012		bd_set_size(bdev, 0);
1013		/* let user-space know about this change */
1014		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1015	}
1016	mapping_set_gfp_mask(filp->f_mapping, gfp);
1017	lo->lo_state = Lo_unbound;
1018	/* This is safe: open() is still holding a reference. */
1019	module_put(THIS_MODULE);
1020	if (lo->lo_flags & LO_FLAGS_PARTSCAN && bdev)
1021		ioctl_by_bdev(bdev, BLKRRPART, 0);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1022	lo->lo_flags = 0;
1023	if (!part_shift)
1024		lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN;
1025	mutex_unlock(&lo->lo_ctl_mutex);
 
 
 
1026	/*
1027	 * Need not hold lo_ctl_mutex to fput backing file.
1028	 * Calling fput holding lo_ctl_mutex triggers a circular
1029	 * lock dependency possibility warning as fput can take
1030	 * bd_mutex which is usually taken before lo_ctl_mutex.
1031	 */
1032	fput(filp);
1033	return 0;
1034}
1035
1036static int
1037loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1038{
1039	int err;
1040	struct loop_func_table *xfer;
1041	uid_t uid = current_uid();
1042
1043	if (lo->lo_encrypt_key_size &&
1044	    lo->lo_key_owner != uid &&
1045	    !capable(CAP_SYS_ADMIN))
1046		return -EPERM;
1047	if (lo->lo_state != Lo_bound)
1048		return -ENXIO;
1049	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
1050		return -EINVAL;
1051
1052	err = loop_release_xfer(lo);
 
 
 
 
 
 
 
 
 
1053	if (err)
1054		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1055
1056	if (info->lo_encrypt_type) {
1057		unsigned int type = info->lo_encrypt_type;
 
1058
1059		if (type >= MAX_LO_CRYPT)
1060			return -EINVAL;
1061		xfer = xfer_funcs[type];
1062		if (xfer == NULL)
1063			return -EINVAL;
1064	} else
1065		xfer = NULL;
1066
1067	err = loop_init_xfer(lo, xfer, info);
1068	if (err)
1069		return err;
 
 
 
 
1070
1071	if (lo->lo_offset != info->lo_offset ||
1072	    lo->lo_sizelimit != info->lo_sizelimit) {
1073		if (figure_loop_size(lo, info->lo_offset, info->lo_sizelimit))
1074			return -EFBIG;
 
1075	}
1076	loop_config_discard(lo);
1077
1078	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
1079	memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE);
1080	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
1081	lo->lo_crypt_name[LO_NAME_SIZE-1] = 0;
1082
1083	if (!xfer)
1084		xfer = &none_funcs;
1085	lo->transfer = xfer->transfer;
1086	lo->ioctl = xfer->ioctl;
1087
1088	if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) !=
1089	     (info->lo_flags & LO_FLAGS_AUTOCLEAR))
1090		lo->lo_flags ^= LO_FLAGS_AUTOCLEAR;
1091
1092	if ((info->lo_flags & LO_FLAGS_PARTSCAN) &&
1093	     !(lo->lo_flags & LO_FLAGS_PARTSCAN)) {
1094		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1095		lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN;
1096		ioctl_by_bdev(lo->lo_device, BLKRRPART, 0);
 
 
 
 
 
 
 
 
 
 
1097	}
1098
1099	lo->lo_encrypt_key_size = info->lo_encrypt_key_size;
1100	lo->lo_init[0] = info->lo_init[0];
1101	lo->lo_init[1] = info->lo_init[1];
1102	if (info->lo_encrypt_key_size) {
1103		memcpy(lo->lo_encrypt_key, info->lo_encrypt_key,
1104		       info->lo_encrypt_key_size);
1105		lo->lo_key_owner = uid;
1106	}	
1107
1108	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1109}
1110
1111static int
1112loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1113{
1114	struct file *file = lo->lo_backing_file;
1115	struct kstat stat;
1116	int error;
1117
1118	if (lo->lo_state != Lo_bound)
 
 
 
 
1119		return -ENXIO;
1120	error = vfs_getattr(file->f_path.mnt, file->f_path.dentry, &stat);
1121	if (error)
1122		return error;
1123	memset(info, 0, sizeof(*info));
1124	info->lo_number = lo->lo_number;
1125	info->lo_device = huge_encode_dev(stat.dev);
1126	info->lo_inode = stat.ino;
1127	info->lo_rdevice = huge_encode_dev(lo->lo_device ? stat.rdev : stat.dev);
1128	info->lo_offset = lo->lo_offset;
1129	info->lo_sizelimit = lo->lo_sizelimit;
1130	info->lo_flags = lo->lo_flags;
1131	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1132	memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE);
1133	info->lo_encrypt_type =
1134		lo->lo_encryption ? lo->lo_encryption->number : 0;
1135	if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) {
1136		info->lo_encrypt_key_size = lo->lo_encrypt_key_size;
1137		memcpy(info->lo_encrypt_key, lo->lo_encrypt_key,
1138		       lo->lo_encrypt_key_size);
 
 
 
1139	}
1140	return 0;
 
1141}
1142
1143static void
1144loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1145{
1146	memset(info64, 0, sizeof(*info64));
1147	info64->lo_number = info->lo_number;
1148	info64->lo_device = info->lo_device;
1149	info64->lo_inode = info->lo_inode;
1150	info64->lo_rdevice = info->lo_rdevice;
1151	info64->lo_offset = info->lo_offset;
1152	info64->lo_sizelimit = 0;
1153	info64->lo_encrypt_type = info->lo_encrypt_type;
1154	info64->lo_encrypt_key_size = info->lo_encrypt_key_size;
1155	info64->lo_flags = info->lo_flags;
1156	info64->lo_init[0] = info->lo_init[0];
1157	info64->lo_init[1] = info->lo_init[1];
1158	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1159		memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE);
1160	else
1161		memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1162	memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE);
1163}
1164
1165static int
1166loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1167{
1168	memset(info, 0, sizeof(*info));
1169	info->lo_number = info64->lo_number;
1170	info->lo_device = info64->lo_device;
1171	info->lo_inode = info64->lo_inode;
1172	info->lo_rdevice = info64->lo_rdevice;
1173	info->lo_offset = info64->lo_offset;
1174	info->lo_encrypt_type = info64->lo_encrypt_type;
1175	info->lo_encrypt_key_size = info64->lo_encrypt_key_size;
1176	info->lo_flags = info64->lo_flags;
1177	info->lo_init[0] = info64->lo_init[0];
1178	info->lo_init[1] = info64->lo_init[1];
1179	if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1180		memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1181	else
1182		memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1183	memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1184
1185	/* error in case values were truncated */
1186	if (info->lo_device != info64->lo_device ||
1187	    info->lo_rdevice != info64->lo_rdevice ||
1188	    info->lo_inode != info64->lo_inode ||
1189	    info->lo_offset != info64->lo_offset)
1190		return -EOVERFLOW;
1191
1192	return 0;
1193}
1194
1195static int
1196loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1197{
1198	struct loop_info info;
1199	struct loop_info64 info64;
1200
1201	if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1202		return -EFAULT;
1203	loop_info64_from_old(&info, &info64);
1204	return loop_set_status(lo, &info64);
1205}
1206
1207static int
1208loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1209{
1210	struct loop_info64 info64;
1211
1212	if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1213		return -EFAULT;
1214	return loop_set_status(lo, &info64);
1215}
1216
1217static int
1218loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1219	struct loop_info info;
1220	struct loop_info64 info64;
1221	int err = 0;
1222
1223	if (!arg)
1224		err = -EINVAL;
1225	if (!err)
1226		err = loop_get_status(lo, &info64);
1227	if (!err)
1228		err = loop_info64_to_old(&info64, &info);
1229	if (!err && copy_to_user(arg, &info, sizeof(info)))
1230		err = -EFAULT;
1231
1232	return err;
1233}
1234
1235static int
1236loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1237	struct loop_info64 info64;
1238	int err = 0;
1239
1240	if (!arg)
1241		err = -EINVAL;
1242	if (!err)
1243		err = loop_get_status(lo, &info64);
1244	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1245		err = -EFAULT;
1246
1247	return err;
1248}
1249
1250static int loop_set_capacity(struct loop_device *lo, struct block_device *bdev)
1251{
1252	int err;
1253	sector_t sec;
1254	loff_t sz;
1255
1256	err = -ENXIO;
1257	if (unlikely(lo->lo_state != Lo_bound))
 
 
 
 
 
 
 
 
 
 
 
 
1258		goto out;
1259	err = figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit);
1260	if (unlikely(err))
1261		goto out;
1262	sec = get_capacity(lo->lo_disk);
1263	/* the width of sector_t may be narrow for bit-shift */
1264	sz = sec;
1265	sz <<= 9;
1266	mutex_lock(&bdev->bd_mutex);
1267	bd_set_size(bdev, sz);
1268	/* let user-space know about the new size */
1269	kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
1270	mutex_unlock(&bdev->bd_mutex);
1271
 
 
 
 
1272 out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1273	return err;
1274}
1275
1276static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1277	unsigned int cmd, unsigned long arg)
1278{
1279	struct loop_device *lo = bdev->bd_disk->private_data;
 
1280	int err;
1281
1282	mutex_lock_nested(&lo->lo_ctl_mutex, 1);
1283	switch (cmd) {
1284	case LOOP_SET_FD:
1285		err = loop_set_fd(lo, mode, bdev, arg);
1286		break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1287	case LOOP_CHANGE_FD:
1288		err = loop_change_fd(lo, bdev, arg);
1289		break;
1290	case LOOP_CLR_FD:
1291		/* loop_clr_fd would have unlocked lo_ctl_mutex on success */
1292		err = loop_clr_fd(lo);
1293		if (!err)
1294			goto out_unlocked;
1295		break;
1296	case LOOP_SET_STATUS:
1297		err = -EPERM;
1298		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1299			err = loop_set_status_old(lo,
1300					(struct loop_info __user *)arg);
1301		break;
1302	case LOOP_GET_STATUS:
1303		err = loop_get_status_old(lo, (struct loop_info __user *) arg);
1304		break;
1305	case LOOP_SET_STATUS64:
1306		err = -EPERM;
1307		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1308			err = loop_set_status64(lo,
1309					(struct loop_info64 __user *) arg);
1310		break;
1311	case LOOP_GET_STATUS64:
1312		err = loop_get_status64(lo, (struct loop_info64 __user *) arg);
1313		break;
1314	case LOOP_SET_CAPACITY:
1315		err = -EPERM;
1316		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN))
1317			err = loop_set_capacity(lo, bdev);
1318		break;
 
1319	default:
1320		err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL;
 
1321	}
1322	mutex_unlock(&lo->lo_ctl_mutex);
1323
1324out_unlocked:
1325	return err;
1326}
1327
1328#ifdef CONFIG_COMPAT
1329struct compat_loop_info {
1330	compat_int_t	lo_number;      /* ioctl r/o */
1331	compat_dev_t	lo_device;      /* ioctl r/o */
1332	compat_ulong_t	lo_inode;       /* ioctl r/o */
1333	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1334	compat_int_t	lo_offset;
1335	compat_int_t	lo_encrypt_type;
1336	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1337	compat_int_t	lo_flags;       /* ioctl r/o */
1338	char		lo_name[LO_NAME_SIZE];
1339	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1340	compat_ulong_t	lo_init[2];
1341	char		reserved[4];
1342};
1343
1344/*
1345 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1346 * - noinlined to reduce stack space usage in main part of driver
1347 */
1348static noinline int
1349loop_info64_from_compat(const struct compat_loop_info __user *arg,
1350			struct loop_info64 *info64)
1351{
1352	struct compat_loop_info info;
1353
1354	if (copy_from_user(&info, arg, sizeof(info)))
1355		return -EFAULT;
1356
1357	memset(info64, 0, sizeof(*info64));
1358	info64->lo_number = info.lo_number;
1359	info64->lo_device = info.lo_device;
1360	info64->lo_inode = info.lo_inode;
1361	info64->lo_rdevice = info.lo_rdevice;
1362	info64->lo_offset = info.lo_offset;
1363	info64->lo_sizelimit = 0;
1364	info64->lo_encrypt_type = info.lo_encrypt_type;
1365	info64->lo_encrypt_key_size = info.lo_encrypt_key_size;
1366	info64->lo_flags = info.lo_flags;
1367	info64->lo_init[0] = info.lo_init[0];
1368	info64->lo_init[1] = info.lo_init[1];
1369	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1370		memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE);
1371	else
1372		memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1373	memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE);
1374	return 0;
1375}
1376
1377/*
1378 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1379 * - noinlined to reduce stack space usage in main part of driver
1380 */
1381static noinline int
1382loop_info64_to_compat(const struct loop_info64 *info64,
1383		      struct compat_loop_info __user *arg)
1384{
1385	struct compat_loop_info info;
1386
1387	memset(&info, 0, sizeof(info));
1388	info.lo_number = info64->lo_number;
1389	info.lo_device = info64->lo_device;
1390	info.lo_inode = info64->lo_inode;
1391	info.lo_rdevice = info64->lo_rdevice;
1392	info.lo_offset = info64->lo_offset;
1393	info.lo_encrypt_type = info64->lo_encrypt_type;
1394	info.lo_encrypt_key_size = info64->lo_encrypt_key_size;
1395	info.lo_flags = info64->lo_flags;
1396	info.lo_init[0] = info64->lo_init[0];
1397	info.lo_init[1] = info64->lo_init[1];
1398	if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI)
1399		memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE);
1400	else
1401		memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1402	memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE);
1403
1404	/* error in case values were truncated */
1405	if (info.lo_device != info64->lo_device ||
1406	    info.lo_rdevice != info64->lo_rdevice ||
1407	    info.lo_inode != info64->lo_inode ||
1408	    info.lo_offset != info64->lo_offset ||
1409	    info.lo_init[0] != info64->lo_init[0] ||
1410	    info.lo_init[1] != info64->lo_init[1])
1411		return -EOVERFLOW;
1412
1413	if (copy_to_user(arg, &info, sizeof(info)))
1414		return -EFAULT;
1415	return 0;
1416}
1417
1418static int
1419loop_set_status_compat(struct loop_device *lo,
1420		       const struct compat_loop_info __user *arg)
1421{
1422	struct loop_info64 info64;
1423	int ret;
1424
1425	ret = loop_info64_from_compat(arg, &info64);
1426	if (ret < 0)
1427		return ret;
1428	return loop_set_status(lo, &info64);
1429}
1430
1431static int
1432loop_get_status_compat(struct loop_device *lo,
1433		       struct compat_loop_info __user *arg)
1434{
1435	struct loop_info64 info64;
1436	int err = 0;
1437
1438	if (!arg)
1439		err = -EINVAL;
1440	if (!err)
1441		err = loop_get_status(lo, &info64);
1442	if (!err)
1443		err = loop_info64_to_compat(&info64, arg);
1444	return err;
1445}
1446
1447static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1448			   unsigned int cmd, unsigned long arg)
1449{
1450	struct loop_device *lo = bdev->bd_disk->private_data;
1451	int err;
1452
1453	switch(cmd) {
1454	case LOOP_SET_STATUS:
1455		mutex_lock(&lo->lo_ctl_mutex);
1456		err = loop_set_status_compat(
1457			lo, (const struct compat_loop_info __user *) arg);
1458		mutex_unlock(&lo->lo_ctl_mutex);
1459		break;
1460	case LOOP_GET_STATUS:
1461		mutex_lock(&lo->lo_ctl_mutex);
1462		err = loop_get_status_compat(
1463			lo, (struct compat_loop_info __user *) arg);
1464		mutex_unlock(&lo->lo_ctl_mutex);
1465		break;
1466	case LOOP_SET_CAPACITY:
1467	case LOOP_CLR_FD:
1468	case LOOP_GET_STATUS64:
1469	case LOOP_SET_STATUS64:
 
1470		arg = (unsigned long) compat_ptr(arg);
 
1471	case LOOP_SET_FD:
1472	case LOOP_CHANGE_FD:
 
 
1473		err = lo_ioctl(bdev, mode, cmd, arg);
1474		break;
1475	default:
1476		err = -ENOIOCTLCMD;
1477		break;
1478	}
1479	return err;
1480}
1481#endif
1482
1483static int lo_open(struct block_device *bdev, fmode_t mode)
1484{
1485	struct loop_device *lo;
1486	int err = 0;
1487
1488	mutex_lock(&loop_index_mutex);
1489	lo = bdev->bd_disk->private_data;
1490	if (!lo) {
1491		err = -ENXIO;
1492		goto out;
1493	}
1494
1495	mutex_lock(&lo->lo_ctl_mutex);
1496	lo->lo_refcnt++;
1497	mutex_unlock(&lo->lo_ctl_mutex);
1498out:
1499	mutex_unlock(&loop_index_mutex);
1500	return err;
1501}
1502
1503static int lo_release(struct gendisk *disk, fmode_t mode)
1504{
1505	struct loop_device *lo = disk->private_data;
1506	int err;
1507
1508	mutex_lock(&lo->lo_ctl_mutex);
1509
1510	if (--lo->lo_refcnt)
1511		goto out;
1512
1513	if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) {
 
 
 
1514		/*
1515		 * In autoclear mode, stop the loop thread
1516		 * and remove configuration after last close.
1517		 */
1518		err = loop_clr_fd(lo);
1519		if (!err)
1520			goto out_unlocked;
1521	} else {
1522		/*
1523		 * Otherwise keep thread (if running) and config,
1524		 * but flush possible ongoing bios in thread.
1525		 */
1526		loop_flush(lo);
1527	}
 
 
1528
1529out:
1530	mutex_unlock(&lo->lo_ctl_mutex);
1531out_unlocked:
1532	return 0;
 
 
 
 
 
 
1533}
1534
1535static const struct block_device_operations lo_fops = {
1536	.owner =	THIS_MODULE,
1537	.open =		lo_open,
1538	.release =	lo_release,
1539	.ioctl =	lo_ioctl,
1540#ifdef CONFIG_COMPAT
1541	.compat_ioctl =	lo_compat_ioctl,
1542#endif
 
1543};
1544
1545/*
1546 * And now the modules code and kernel interface.
1547 */
1548static int max_loop;
1549module_param(max_loop, int, S_IRUGO);
 
 
 
 
 
 
 
 
 
1550MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1551module_param(max_part, int, S_IRUGO);
1552MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1553MODULE_LICENSE("GPL");
1554MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1555
1556int loop_register_transfer(struct loop_func_table *funcs)
 
1557{
1558	unsigned int n = funcs->number;
 
 
1559
1560	if (n >= MAX_LO_CRYPT || xfer_funcs[n])
1561		return -EINVAL;
1562	xfer_funcs[n] = funcs;
1563	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1564}
1565
1566static int unregister_transfer_cb(int id, void *ptr, void *data)
1567{
1568	struct loop_device *lo = ptr;
1569	struct loop_func_table *xfer = data;
 
 
 
1570
1571	mutex_lock(&lo->lo_ctl_mutex);
1572	if (lo->lo_encryption == xfer)
1573		loop_release_xfer(lo);
1574	mutex_unlock(&lo->lo_ctl_mutex);
1575	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1576}
1577
1578int loop_unregister_transfer(int number)
 
1579{
1580	unsigned int n = number;
1581	struct loop_func_table *xfer;
1582
1583	if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL)
1584		return -EINVAL;
 
 
 
 
 
1585
1586	xfer_funcs[n] = NULL;
1587	idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer);
1588	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1589}
1590
1591EXPORT_SYMBOL(loop_register_transfer);
1592EXPORT_SYMBOL(loop_unregister_transfer);
 
 
 
 
1593
1594static int loop_add(struct loop_device **l, int i)
 
 
 
 
 
 
 
 
 
 
 
 
1595{
1596	struct loop_device *lo;
1597	struct gendisk *disk;
1598	int err;
1599
1600	err = -ENOMEM;
1601	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1602	if (!lo)
1603		goto out;
 
 
 
 
1604
1605	if (!idr_pre_get(&loop_index_idr, GFP_KERNEL))
 
1606		goto out_free_dev;
1607
 
1608	if (i >= 0) {
1609		int m;
1610
1611		/* create specific i in the index */
1612		err = idr_get_new_above(&loop_index_idr, lo, i, &m);
1613		if (err >= 0 && i != m) {
1614			idr_remove(&loop_index_idr, m);
1615			err = -EEXIST;
1616		}
1617	} else if (i == -1) {
1618		int m;
1619
1620		/* get next free nr */
1621		err = idr_get_new(&loop_index_idr, lo, &m);
1622		if (err >= 0)
1623			i = m;
1624	} else {
1625		err = -EINVAL;
1626	}
 
1627	if (err < 0)
1628		goto out_free_dev;
 
1629
1630	lo->lo_queue = blk_alloc_queue(GFP_KERNEL);
1631	if (!lo->lo_queue)
1632		goto out_free_dev;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1633
1634	disk = lo->lo_disk = alloc_disk(1 << part_shift);
1635	if (!disk)
1636		goto out_free_queue;
 
 
 
 
 
 
1637
1638	/*
1639	 * Disable partition scanning by default. The in-kernel partition
1640	 * scanning can be requested individually per-device during its
1641	 * setup. Userspace can always add and remove partitions from all
1642	 * devices. The needed partition minors are allocated from the
1643	 * extended minor space, the main loop device numbers will continue
1644	 * to match the loop minors, regardless of the number of partitions
1645	 * used.
1646	 *
1647	 * If max_part is given, partition scanning is globally enabled for
1648	 * all loop devices. The minors for the main loop devices will be
1649	 * multiples of max_part.
1650	 *
1651	 * Note: Global-for-all-devices, set-only-at-init, read-only module
1652	 * parameteters like 'max_loop' and 'max_part' make things needlessly
1653	 * complicated, are too static, inflexible and may surprise
1654	 * userspace tools. Parameters like this in general should be avoided.
1655	 */
1656	if (!part_shift)
1657		disk->flags |= GENHD_FL_NO_PART_SCAN;
1658	disk->flags |= GENHD_FL_EXT_DEVT;
1659	mutex_init(&lo->lo_ctl_mutex);
1660	lo->lo_number		= i;
1661	lo->lo_thread		= NULL;
1662	init_waitqueue_head(&lo->lo_event);
1663	spin_lock_init(&lo->lo_lock);
 
 
 
1664	disk->major		= LOOP_MAJOR;
1665	disk->first_minor	= i << part_shift;
 
1666	disk->fops		= &lo_fops;
1667	disk->private_data	= lo;
1668	disk->queue		= lo->lo_queue;
 
 
1669	sprintf(disk->disk_name, "loop%d", i);
1670	add_disk(disk);
1671	*l = lo;
1672	return lo->lo_number;
 
1673
1674out_free_queue:
1675	blk_cleanup_queue(lo->lo_queue);
 
 
 
 
 
 
 
 
 
 
 
 
 
1676out_free_dev:
1677	kfree(lo);
1678out:
1679	return err;
1680}
1681
1682static void loop_remove(struct loop_device *lo)
1683{
 
1684	del_gendisk(lo->lo_disk);
1685	blk_cleanup_queue(lo->lo_queue);
 
 
 
 
 
1686	put_disk(lo->lo_disk);
1687	kfree(lo);
1688}
1689
1690static int find_free_cb(int id, void *ptr, void *data)
1691{
1692	struct loop_device *lo = ptr;
1693	struct loop_device **l = data;
1694
1695	if (lo->lo_state == Lo_unbound) {
1696		*l = lo;
1697		return 1;
1698	}
1699	return 0;
1700}
1701
1702static int loop_lookup(struct loop_device **l, int i)
1703{
1704	struct loop_device *lo;
1705	int ret = -ENODEV;
1706
1707	if (i < 0) {
1708		int err;
1709
1710		err = idr_for_each(&loop_index_idr, &find_free_cb, &lo);
1711		if (err == 1) {
1712			*l = lo;
1713			ret = lo->lo_number;
1714		}
1715		goto out;
1716	}
 
 
 
 
 
 
 
 
 
 
 
 
 
1717
1718	/* lookup and return a specific i */
1719	lo = idr_find(&loop_index_idr, i);
1720	if (lo) {
1721		*l = lo;
1722		ret = lo->lo_number;
1723	}
1724out:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1725	return ret;
1726}
1727
1728static struct kobject *loop_probe(dev_t dev, int *part, void *data)
1729{
1730	struct loop_device *lo;
1731	struct kobject *kobj;
1732	int err;
1733
1734	mutex_lock(&loop_index_mutex);
1735	err = loop_lookup(&lo, MINOR(dev) >> part_shift);
1736	if (err < 0)
1737		err = loop_add(&lo, MINOR(dev) >> part_shift);
1738	if (err < 0)
1739		kobj = ERR_PTR(err);
1740	else
1741		kobj = get_disk(lo->lo_disk);
1742	mutex_unlock(&loop_index_mutex);
1743
1744	*part = 0;
1745	return kobj;
 
1746}
1747
1748static long loop_control_ioctl(struct file *file, unsigned int cmd,
1749			       unsigned long parm)
1750{
1751	struct loop_device *lo;
1752	int ret = -ENOSYS;
1753
1754	mutex_lock(&loop_index_mutex);
1755	switch (cmd) {
1756	case LOOP_CTL_ADD:
1757		ret = loop_lookup(&lo, parm);
1758		if (ret >= 0) {
1759			ret = -EEXIST;
1760			break;
1761		}
1762		ret = loop_add(&lo, parm);
1763		break;
1764	case LOOP_CTL_REMOVE:
1765		ret = loop_lookup(&lo, parm);
1766		if (ret < 0)
1767			break;
1768		mutex_lock(&lo->lo_ctl_mutex);
1769		if (lo->lo_state != Lo_unbound) {
1770			ret = -EBUSY;
1771			mutex_unlock(&lo->lo_ctl_mutex);
1772			break;
1773		}
1774		if (lo->lo_refcnt > 0) {
1775			ret = -EBUSY;
1776			mutex_unlock(&lo->lo_ctl_mutex);
1777			break;
1778		}
1779		lo->lo_disk->private_data = NULL;
1780		mutex_unlock(&lo->lo_ctl_mutex);
1781		idr_remove(&loop_index_idr, lo->lo_number);
1782		loop_remove(lo);
1783		break;
1784	case LOOP_CTL_GET_FREE:
1785		ret = loop_lookup(&lo, -1);
1786		if (ret >= 0)
1787			break;
1788		ret = loop_add(&lo, -1);
1789	}
1790	mutex_unlock(&loop_index_mutex);
1791
1792	return ret;
1793}
1794
1795static const struct file_operations loop_ctl_fops = {
1796	.open		= nonseekable_open,
1797	.unlocked_ioctl	= loop_control_ioctl,
1798	.compat_ioctl	= loop_control_ioctl,
1799	.owner		= THIS_MODULE,
1800	.llseek		= noop_llseek,
1801};
1802
1803static struct miscdevice loop_misc = {
1804	.minor		= LOOP_CTRL_MINOR,
1805	.name		= "loop-control",
1806	.fops		= &loop_ctl_fops,
1807};
1808
1809MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
1810MODULE_ALIAS("devname:loop-control");
1811
1812static int __init loop_init(void)
1813{
1814	int i, nr;
1815	unsigned long range;
1816	struct loop_device *lo;
1817	int err;
1818
1819	err = misc_register(&loop_misc);
1820	if (err < 0)
1821		return err;
1822
1823	part_shift = 0;
1824	if (max_part > 0) {
1825		part_shift = fls(max_part);
1826
1827		/*
1828		 * Adjust max_part according to part_shift as it is exported
1829		 * to user space so that user can decide correct minor number
1830		 * if [s]he want to create more devices.
1831		 *
1832		 * Note that -1 is required because partition 0 is reserved
1833		 * for the whole disk.
1834		 */
1835		max_part = (1UL << part_shift) - 1;
1836	}
1837
1838	if ((1UL << part_shift) > DISK_MAX_PARTS)
1839		return -EINVAL;
1840
1841	if (max_loop > 1UL << (MINORBITS - part_shift))
1842		return -EINVAL;
1843
1844	/*
1845	 * If max_loop is specified, create that many devices upfront.
1846	 * This also becomes a hard limit. If max_loop is not specified,
1847	 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1848	 * init time. Loop devices can be requested on-demand with the
1849	 * /dev/loop-control interface, or be instantiated by accessing
1850	 * a 'dead' device node.
1851	 */
1852	if (max_loop) {
1853		nr = max_loop;
1854		range = max_loop << part_shift;
1855	} else {
1856		nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1857		range = 1UL << MINORBITS;
1858	}
1859
1860	if (register_blkdev(LOOP_MAJOR, "loop"))
1861		return -EIO;
 
1862
1863	blk_register_region(MKDEV(LOOP_MAJOR, 0), range,
1864				  THIS_MODULE, loop_probe, NULL, NULL);
 
 
 
1865
1866	/* pre-create number of devices given by config or max_loop */
1867	mutex_lock(&loop_index_mutex);
1868	for (i = 0; i < nr; i++)
1869		loop_add(&lo, i);
1870	mutex_unlock(&loop_index_mutex);
1871
1872	printk(KERN_INFO "loop: module loaded\n");
1873	return 0;
1874}
1875
1876static int loop_exit_cb(int id, void *ptr, void *data)
1877{
1878	struct loop_device *lo = ptr;
1879
1880	loop_remove(lo);
1881	return 0;
1882}
1883
1884static void __exit loop_exit(void)
1885{
1886	unsigned long range;
1887
1888	range = max_loop ? max_loop << part_shift : 1UL << MINORBITS;
1889
1890	idr_for_each(&loop_index_idr, &loop_exit_cb, NULL);
1891	idr_remove_all(&loop_index_idr);
1892	idr_destroy(&loop_index_idr);
1893
1894	blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range);
1895	unregister_blkdev(LOOP_MAJOR, "loop");
1896
1897	misc_deregister(&loop_misc);
 
 
 
 
 
 
 
 
 
 
 
1898}
1899
1900module_init(loop_init);
1901module_exit(loop_exit);
1902
1903#ifndef MODULE
1904static int __init max_loop_setup(char *str)
1905{
1906	max_loop = simple_strtol(str, NULL, 0);
1907	return 1;
1908}
1909
1910__setup("max_loop=", max_loop_setup);
1911#endif
v6.2
   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_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};
  91
  92#define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
  93#define LOOP_DEFAULT_HW_Q_DEPTH (128)
 
 
 
 
 
 
 
 
  94
  95static DEFINE_IDR(loop_index_idr);
  96static DEFINE_MUTEX(loop_ctl_mutex);
  97static 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}
 138
 139static int max_part;
 140static int part_shift;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 141
 142static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
 143{
 144	loff_t loopsize;
 145
 146	/* Compute loopsize in bytes */
 147	loopsize = i_size_read(file->f_mapping->host);
 148	if (offset > 0)
 149		loopsize -= offset;
 150	/* offset is beyond i_size, weird but possible */
 151	if (loopsize < 0)
 152		return 0;
 153
 154	if (sizelimit > 0 && sizelimit < loopsize)
 155		loopsize = sizelimit;
 156	/*
 157	 * Unfortunately, if we want to do I/O on the device,
 158	 * the number of 512-byte sectors has to fit into a sector_t.
 159	 */
 160	return loopsize >> 9;
 161}
 162
 163static loff_t get_loop_size(struct loop_device *lo, struct file *file)
 164{
 165	return get_size(lo->lo_offset, lo->lo_sizelimit, file);
 166}
 167
 168static 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
 185	 * size of loop is bigger than the backing device's.
 186	 *
 187	 * TODO: the above condition may be loosed in the future, and
 188	 * direct I/O may be switched runtime at that time because most
 189	 * of requests in sane applications should be PAGE_SIZE aligned
 190	 */
 191	if (dio) {
 192		if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
 193		    !(lo->lo_offset & dio_align) &&
 194		    (file->f_mode & FMODE_CAN_ODIRECT))
 195			use_dio = true;
 196		else
 197			use_dio = false;
 198	} else {
 199		use_dio = false;
 200	}
 201
 202	if (lo->use_dio == use_dio)
 203		return;
 204
 205	/* flush dirty pages before changing direct IO */
 206	vfs_fsync(file, 0);
 
 
 
 207
 208	/*
 209	 * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
 210	 * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
 211	 * will get updated by ioctl(LOOP_GET_STATUS)
 212	 */
 213	if (lo->lo_state == Lo_bound)
 214		blk_mq_freeze_queue(lo->lo_queue);
 215	lo->use_dio = use_dio;
 216	if (use_dio) {
 217		blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
 218		lo->lo_flags |= LO_FLAGS_DIRECT_IO;
 219	} else {
 220		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
 221		lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
 222	}
 223	if (lo->lo_state == Lo_bound)
 224		blk_mq_unfreeze_queue(lo->lo_queue);
 225}
 226
 227/**
 228 * loop_set_size() - sets device size and notifies userspace
 229 * @lo: struct loop_device to set the size for
 230 * @size: new size of the loop device
 231 *
 232 * Callers must validate that the size passed into this function fits into
 233 * a sector_t, eg using loop_validate_size()
 234 */
 235static void loop_set_size(struct loop_device *lo, loff_t size)
 
 236{
 237	if (!set_capacity_and_notify(lo->lo_disk, size))
 238		kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
 239}
 240
 241static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
 242{
 243	struct iov_iter i;
 244	ssize_t bw;
 
 245
 246	iov_iter_bvec(&i, ITER_SOURCE, bvec, 1, bvec->bv_len);
 247
 248	file_start_write(file);
 249	bw = vfs_iter_write(file, &i, ppos, 0);
 250	file_end_write(file);
 251
 252	if (likely(bw ==  bvec->bv_len))
 253		return 0;
 254
 255	printk_ratelimited(KERN_ERR
 256		"loop: Write error at byte offset %llu, length %i.\n",
 257		(unsigned long long)*ppos, bvec->bv_len);
 258	if (bw >= 0)
 259		bw = -EIO;
 260	return bw;
 261}
 262
 263static int lo_write_simple(struct loop_device *lo, struct request *rq,
 264		loff_t pos)
 
 
 
 
 
 
 265{
 266	struct bio_vec bvec;
 267	struct req_iterator iter;
 268	int ret = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 269
 270	rq_for_each_segment(bvec, rq, iter) {
 271		ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
 272		if (ret < 0)
 273			break;
 274		cond_resched();
 275	}
 276
 
 
 
 
 277	return ret;
 
 
 
 
 278}
 279
 280static int lo_read_simple(struct loop_device *lo, struct request *rq,
 281		loff_t pos)
 
 
 
 
 
 
 
 
 282{
 283	struct bio_vec bvec;
 284	struct req_iterator iter;
 285	struct iov_iter i;
 286	ssize_t len;
 287
 288	rq_for_each_segment(bvec, rq, iter) {
 289		iov_iter_bvec(&i, ITER_DEST, &bvec, 1, bvec.bv_len);
 290		len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
 291		if (len < 0)
 292			return len;
 
 
 
 
 
 
 
 293
 294		flush_dcache_page(bvec.bv_page);
 295
 296		if (len != bvec.bv_len) {
 297			struct bio *bio;
 298
 299			__rq_for_each_bio(bio, rq)
 300				zero_fill_bio(bio);
 301			break;
 302		}
 303		cond_resched();
 304	}
 305
 306	return 0;
 
 
 
 307}
 308
 309static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
 310			int mode)
 
 311{
 312	/*
 313	 * We use fallocate to manipulate the space mappings used by the image
 314	 * a.k.a. discard/zerorange.
 315	 */
 316	struct file *file = lo->lo_backing_file;
 317	int ret;
 
 
 
 318
 319	mode |= FALLOC_FL_KEEP_SIZE;
 
 
 
 
 320
 321	if (!bdev_max_discard_sectors(lo->lo_device))
 322		return -EOPNOTSUPP;
 323
 324	ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
 325	if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
 326		return -EIO;
 327	return ret;
 328}
 329
 330static int lo_req_flush(struct loop_device *lo, struct request *rq)
 
 331{
 332	int ret = vfs_fsync(lo->lo_backing_file, 0);
 333	if (unlikely(ret && ret != -EINVAL))
 334		ret = -EIO;
 
 
 
 
 
 335
 336	return ret;
 
 
 
 
 
 
 337}
 338
 339static void lo_complete_rq(struct request *rq)
 340{
 341	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
 342	blk_status_t ret = BLK_STS_OK;
 
 
 343
 344	if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
 345	    req_op(rq) != REQ_OP_READ) {
 346		if (cmd->ret < 0)
 347			ret = errno_to_blk_status(cmd->ret);
 348		goto end_io;
 349	}
 350
 351	/*
 352	 * Short READ - if we got some data, advance our request and
 353	 * retry it. If we got no data, end the rest with EIO.
 354	 */
 355	if (cmd->ret) {
 356		blk_update_request(rq, BLK_STS_OK, cmd->ret);
 357		cmd->ret = 0;
 358		blk_mq_requeue_request(rq, true);
 359	} else {
 360		if (cmd->use_aio) {
 361			struct bio *bio = rq->bio;
 362
 363			while (bio) {
 364				zero_fill_bio(bio);
 365				bio = bio->bi_next;
 
 
 
 
 
 
 
 
 
 
 
 366			}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 367		}
 368		ret = BLK_STS_IOERR;
 369end_io:
 370		blk_mq_end_request(rq, ret);
 371	}
 
 372}
 373
 374static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
 
 
 
 375{
 376	struct request *rq = blk_mq_rq_from_pdu(cmd);
 
 377
 378	if (!atomic_dec_and_test(&cmd->ref))
 379		return;
 380	kfree(cmd->bvec);
 381	cmd->bvec = NULL;
 382	if (likely(!blk_should_fake_timeout(rq->q)))
 383		blk_mq_complete_request(rq);
 384}
 385
 386static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
 387{
 388	struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
 
 389
 390	cmd->ret = ret;
 391	lo_rw_aio_do_completion(cmd);
 392}
 393
 394static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 395		     loff_t pos, int rw)
 396{
 397	struct iov_iter iter;
 398	struct req_iterator rq_iter;
 399	struct bio_vec *bvec;
 400	struct request *rq = blk_mq_rq_from_pdu(cmd);
 401	struct bio *bio = rq->bio;
 402	struct file *file = lo->lo_backing_file;
 403	struct bio_vec tmp;
 404	unsigned int offset;
 405	int nr_bvec = 0;
 406	int ret;
 407
 408	rq_for_each_bvec(tmp, rq, rq_iter)
 409		nr_bvec++;
 
 
 
 
 
 
 
 410
 411	if (rq->bio != rq->biotail) {
 
 
 
 412
 413		bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
 414				     GFP_NOIO);
 415		if (!bvec)
 416			return -EIO;
 417		cmd->bvec = bvec;
 418
 419		/*
 420		 * The bios of the request may be started from the middle of
 421		 * the 'bvec' because of bio splitting, so we can't directly
 422		 * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
 423		 * API will take care of all details for us.
 424		 */
 425		rq_for_each_bvec(tmp, rq, rq_iter) {
 426			*bvec = tmp;
 427			bvec++;
 428		}
 429		bvec = cmd->bvec;
 430		offset = 0;
 431	} else {
 432		/*
 433		 * Same here, this bio may be started from the middle of the
 434		 * 'bvec' because of bio splitting, so offset from the bvec
 435		 * must be passed to iov iterator
 436		 */
 437		offset = bio->bi_iter.bi_bvec_done;
 438		bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
 439	}
 440	atomic_set(&cmd->ref, 2);
 441
 442	iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
 443	iter.iov_offset = offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 444
 445	cmd->iocb.ki_pos = pos;
 446	cmd->iocb.ki_filp = file;
 447	cmd->iocb.ki_complete = lo_rw_aio_complete;
 448	cmd->iocb.ki_flags = IOCB_DIRECT;
 449	cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
 450
 451	if (rw == ITER_SOURCE)
 452		ret = call_write_iter(file, &cmd->iocb, &iter);
 453	else
 454		ret = call_read_iter(file, &cmd->iocb, &iter);
 455
 456	lo_rw_aio_do_completion(cmd);
 457
 458	if (ret != -EIOCBQUEUED)
 459		lo_rw_aio_complete(&cmd->iocb, ret);
 460	return 0;
 461}
 462
 463static int do_req_filebacked(struct loop_device *lo, struct request *rq)
 464{
 465	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
 466	loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
 
 467
 468	/*
 469	 * lo_write_simple and lo_read_simple should have been covered
 470	 * by io submit style function like lo_rw_aio(), one blocker
 471	 * is that lo_read_simple() need to call flush_dcache_page after
 472	 * the page is written from kernel, and it isn't easy to handle
 473	 * this in io submit style function which submits all segments
 474	 * of the req at one time. And direct read IO doesn't need to
 475	 * run flush_dcache_page().
 476	 */
 477	switch (req_op(rq)) {
 478	case REQ_OP_FLUSH:
 479		return lo_req_flush(lo, rq);
 480	case REQ_OP_WRITE_ZEROES:
 481		/*
 482		 * If the caller doesn't want deallocation, call zeroout to
 483		 * write zeroes the range.  Otherwise, punch them out.
 484		 */
 485		return lo_fallocate(lo, rq, pos,
 486			(rq->cmd_flags & REQ_NOUNMAP) ?
 487				FALLOC_FL_ZERO_RANGE :
 488				FALLOC_FL_PUNCH_HOLE);
 489	case REQ_OP_DISCARD:
 490		return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
 491	case REQ_OP_WRITE:
 492		if (cmd->use_aio)
 493			return lo_rw_aio(lo, cmd, pos, ITER_SOURCE);
 494		else
 495			return lo_write_simple(lo, rq, pos);
 496	case REQ_OP_READ:
 497		if (cmd->use_aio)
 498			return lo_rw_aio(lo, cmd, pos, ITER_DEST);
 499		else
 500			return lo_read_simple(lo, rq, pos);
 501	default:
 502		WARN_ON_ONCE(1);
 503		return -EIO;
 504	}
 505}
 506
 507static inline void loop_update_dio(struct loop_device *lo)
 508{
 509	__loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
 510				lo->use_dio);
 511}
 512
 513static void loop_reread_partitions(struct loop_device *lo)
 
 
 
 
 
 514{
 515	int rc;
 516
 517	mutex_lock(&lo->lo_disk->open_mutex);
 518	rc = bdev_disk_changed(lo->lo_disk, false);
 519	mutex_unlock(&lo->lo_disk->open_mutex);
 520	if (rc)
 521		pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
 522			__func__, lo->lo_number, lo->lo_file_name, rc);
 
 
 
 523}
 524
 525static inline int is_loop_device(struct file *file)
 
 
 
 526{
 527	struct inode *i = file->f_mapping->host;
 
 
 528
 529	return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
 530}
 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
 541		lockdep_assert_held(&loop_validate_mutex);
 542		if (f->f_mapping->host->i_rdev == bdev->bd_dev)
 543			return -EBADF;
 
 
 
 
 
 
 
 544
 545		l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
 546		if (l->lo_state != Lo_bound)
 547			return -EINVAL;
 548		/* Order wrt setting lo->lo_backing_file in loop_configure(). */
 549		rmb();
 550		f = l->lo_backing_file;
 551	}
 552	if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
 553		return -EINVAL;
 554	return 0;
 555}
 556
 557/*
 558 * loop_change_fd switched the backing store of a loopback device to
 559 * a new file. This is useful for operating system installers to free up
 560 * the original file and in High Availability environments to switch to
 561 * an alternative location for the content in case of server meltdown.
 562 * This can only work if the loop device is used read-only, and if the
 563 * new backing store is the same size and type as the old backing store.
 564 */
 565static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
 566			  unsigned int arg)
 567{
 568	struct file *file = fget(arg);
 569	struct file *old_file;
 570	int error;
 571	bool partscan;
 572	bool is_loop;
 573
 574	if (!file)
 575		return -EBADF;
 576
 577	/* suppress uevents while reconfiguring the device */
 578	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
 579
 580	is_loop = is_loop_device(file);
 581	error = loop_global_lock_killable(lo, is_loop);
 582	if (error)
 583		goto out_putf;
 584	error = -ENXIO;
 585	if (lo->lo_state != Lo_bound)
 586		goto out_err;
 587
 588	/* the loop device has to be read-only */
 589	error = -EINVAL;
 590	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
 591		goto out_err;
 592
 593	error = loop_validate_file(file, bdev);
 594	if (error)
 595		goto out_err;
 
 596
 
 597	old_file = lo->lo_backing_file;
 598
 599	error = -EINVAL;
 600
 
 
 
 601	/* size of the new backing store needs to be the same */
 602	if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
 603		goto out_err;
 604
 605	/* and ... switch */
 606	disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
 607	blk_mq_freeze_queue(lo->lo_queue);
 608	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
 609	lo->lo_backing_file = file;
 610	lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
 611	mapping_set_gfp_mask(file->f_mapping,
 612			     lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
 613	loop_update_dio(lo);
 614	blk_mq_unfreeze_queue(lo->lo_queue);
 615	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
 616	loop_global_unlock(lo, is_loop);
 617
 618	/*
 619	 * Flush loop_validate_file() before fput(), for l->lo_backing_file
 620	 * might be pointing at old_file which might be the last reference.
 621	 */
 622	if (!is_loop) {
 623		mutex_lock(&loop_validate_mutex);
 624		mutex_unlock(&loop_validate_mutex);
 625	}
 626	/*
 627	 * We must drop file reference outside of lo_mutex as dropping
 628	 * the file ref can take open_mutex which creates circular locking
 629	 * dependency.
 630	 */
 631	fput(old_file);
 632	if (partscan)
 633		loop_reread_partitions(lo);
 
 634
 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;
 
 
 
 
 
 640
 641out_err:
 642	loop_global_unlock(lo, is_loop);
 643out_putf:
 644	fput(file);
 645	goto done;
 646}
 647
 648/* loop sysfs attributes */
 649
 650static ssize_t loop_attr_show(struct device *dev, char *page,
 651			      ssize_t (*callback)(struct loop_device *, char *))
 652{
 653	struct gendisk *disk = dev_to_disk(dev);
 654	struct loop_device *lo = disk->private_data;
 655
 656	return callback(lo, page);
 657}
 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 =			\
 667	__ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
 668
 669static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
 670{
 671	ssize_t ret;
 672	char *p = NULL;
 673
 674	spin_lock_irq(&lo->lo_lock);
 675	if (lo->lo_backing_file)
 676		p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
 677	spin_unlock_irq(&lo->lo_lock);
 678
 679	if (IS_ERR_OR_NULL(p))
 680		ret = PTR_ERR(p);
 681	else {
 682		ret = strlen(p);
 683		memmove(buf, p, ret);
 684		buf[ret++] = '\n';
 685		buf[ret] = 0;
 686	}
 687
 688	return ret;
 689}
 690
 691static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
 692{
 693	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
 694}
 695
 696static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
 697{
 698	return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
 699}
 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
 705	return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
 706}
 707
 708static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
 709{
 710	int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
 711
 712	return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
 713}
 714
 715static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
 716{
 717	int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
 718
 719	return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
 720}
 721
 722LOOP_ATTR_RO(backing_file);
 723LOOP_ATTR_RO(offset);
 724LOOP_ATTR_RO(sizelimit);
 725LOOP_ATTR_RO(autoclear);
 726LOOP_ATTR_RO(partscan);
 727LOOP_ATTR_RO(dio);
 728
 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,
 734	&loop_attr_partscan.attr,
 735	&loop_attr_dio.attr,
 736	NULL,
 737};
 738
 739static struct attribute_group loop_attribute_group = {
 740	.name = "loop",
 741	.attrs= loop_attrs,
 742};
 743
 744static void loop_sysfs_init(struct loop_device *lo)
 745{
 746	lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
 747						&loop_attribute_group);
 748}
 749
 750static void loop_sysfs_exit(struct loop_device *lo)
 751{
 752	if (lo->sysfs_inited)
 753		sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
 754				   &loop_attribute_group);
 755}
 756
 757static 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;
 762	u32 granularity, max_discard_sectors;
 763
 764	/*
 765	 * If the backing device is a block device, mirror its zeroing
 766	 * capability. Set the discard sectors to the block device's zeroing
 767	 * capabilities because loop discards result in blkdev_issue_zeroout(),
 768	 * not blkdev_issue_discard(). This maintains consistent behavior with
 769	 * file-backed loop devices: discarded regions read back as zero.
 770	 */
 771	if (S_ISBLK(inode->i_mode)) {
 772		struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
 773
 774		max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
 775		granularity = bdev_discard_granularity(I_BDEV(inode)) ?:
 776			queue_physical_block_size(backingq);
 777
 778	/*
 779	 * We use punch hole to reclaim the free space used by the
 780	 * image a.k.a. discard.
 
 
 781	 */
 782	} else if (!file->f_op->fallocate) {
 783		max_discard_sectors = 0;
 784		granularity = 0;
 785
 786	} else {
 787		struct kstatfs sbuf;
 788
 789		max_discard_sectors = UINT_MAX >> 9;
 790		if (!vfs_statfs(&file->f_path, &sbuf))
 791			granularity = sbuf.f_bsize;
 792		else
 793			max_discard_sectors = 0;
 794	}
 795
 796	if (max_discard_sectors) {
 797		q->limits.discard_granularity = granularity;
 798		blk_queue_max_discard_sectors(q, max_discard_sectors);
 799		blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
 800	} else {
 801		q->limits.discard_granularity = 0;
 802		blk_queue_max_discard_sectors(q, 0);
 803		blk_queue_max_write_zeroes_sectors(q, 0);
 804	}
 805}
 806
 807struct 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;
 813	struct cgroup_subsys_state *blkcg_css;
 814	unsigned long last_ran_at;
 815};
 816
 817static void loop_workfn(struct work_struct *work);
 818
 819#ifdef CONFIG_BLK_CGROUP
 820static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
 821{
 822	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{
 833	struct rb_node **node, *parent = NULL;
 834	struct loop_worker *cur_worker, *worker = NULL;
 835	struct work_struct *work;
 836	struct list_head *cmd_list;
 837
 838	spin_lock_irq(&lo->lo_work_lock);
 839
 840	if (queue_on_root_worker(cmd->blkcg_css))
 841		goto queue_work;
 842
 843	node = &lo->worker_tree.rb_node;
 844
 845	while (*node) {
 846		parent = *node;
 847		cur_worker = container_of(*node, struct loop_worker, rb_node);
 848		if (cur_worker->blkcg_css == cmd->blkcg_css) {
 849			worker = cur_worker;
 850			break;
 851		} else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
 852			node = &(*node)->rb_left;
 853		} else {
 854			node = &(*node)->rb_right;
 855		}
 856	}
 857	if (worker)
 858		goto queue_work;
 859
 860	worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
 861	/*
 862	 * In the event we cannot allocate a worker, just queue on the
 863	 * rootcg worker and issue the I/O as the rootcg
 864	 */
 865	if (!worker) {
 866		cmd->blkcg_css = NULL;
 867		if (cmd->memcg_css)
 868			css_put(cmd->memcg_css);
 869		cmd->memcg_css = NULL;
 870		goto queue_work;
 871	}
 872
 873	worker->blkcg_css = cmd->blkcg_css;
 874	css_get(worker->blkcg_css);
 875	INIT_WORK(&worker->work, loop_workfn);
 876	INIT_LIST_HEAD(&worker->cmd_list);
 877	INIT_LIST_HEAD(&worker->idle_list);
 878	worker->lo = lo;
 879	rb_link_node(&worker->rb_node, parent, node);
 880	rb_insert_color(&worker->rb_node, &lo->worker_tree);
 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);
 899}
 900
 901static 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
 934static 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)
 944		nonrot = bdev_nonrot(file_bdev);
 945
 946	if (nonrot)
 947		blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
 948	else
 949		blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
 950}
 951
 952/**
 953 * loop_set_status_from_info - configure device from loop_info
 954 * @lo: struct loop_device to configure
 955 * @info: struct loop_info64 to configure the device with
 956 *
 957 * Configures the loop device parameters according to the passed
 958 * in loop_info64 configuration.
 959 */
 960static int
 961loop_set_status_from_info(struct loop_device *lo,
 962			  const struct loop_info64 *info)
 963{
 964	if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
 965		return -EINVAL;
 966
 967	switch (info->lo_encrypt_type) {
 968	case LO_CRYPT_NONE:
 969		break;
 970	case LO_CRYPT_XOR:
 971		pr_warn("support for the xor transformation has been removed.\n");
 972		return -EINVAL;
 973	case LO_CRYPT_CRYPTOAPI:
 974		pr_warn("support for cryptoloop has been removed.  Use dm-crypt instead.\n");
 975		return -EINVAL;
 976	default:
 977		return -EINVAL;
 978	}
 979
 980	lo->lo_offset = info->lo_offset;
 981	lo->lo_sizelimit = info->lo_sizelimit;
 982
 983	/* loff_t vars have been assigned __u64 */
 984	if (lo->lo_offset < 0 || lo->lo_sizelimit < 0)
 985		return -EOVERFLOW;
 986
 987	memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
 988	lo->lo_file_name[LO_NAME_SIZE-1] = 0;
 989	lo->lo_flags = info->lo_flags;
 990	return 0;
 991}
 992
 993static int loop_configure(struct loop_device *lo, fmode_t mode,
 994			  struct block_device *bdev,
 995			  const struct loop_config *config)
 996{
 997	struct file *file = fget(config->fd);
 998	struct inode *inode;
 999	struct address_space *mapping;
1000	int error;
1001	loff_t size;
1002	bool partscan;
1003	unsigned short bsize;
1004	bool is_loop;
1005
1006	if (!file)
1007		return -EBADF;
1008	is_loop = is_loop_device(file);
1009
1010	/* This is safe, since we have a reference from open(). */
1011	__module_get(THIS_MODULE);
1012
1013	/* suppress uevents while reconfiguring the device */
1014	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
 
 
1015
1016	/*
1017	 * If we don't hold exclusive handle for the device, upgrade to it
1018	 * here to avoid changing device under exclusive owner.
1019	 */
1020	if (!(mode & FMODE_EXCL)) {
1021		error = bd_prepare_to_claim(bdev, loop_configure);
1022		if (error)
1023			goto out_putf;
1024	}
1025
1026	error = loop_global_lock_killable(lo, is_loop);
1027	if (error)
1028		goto out_bdev;
 
1029
1030	error = -EBUSY;
1031	if (lo->lo_state != Lo_unbound)
1032		goto out_unlock;
1033
1034	error = loop_validate_file(file, bdev);
1035	if (error)
1036		goto out_unlock;
 
 
 
 
1037
1038	mapping = file->f_mapping;
1039	inode = mapping->host;
1040
1041	if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1042		error = -EINVAL;
1043		goto out_unlock;
1044	}
1045
1046	if (config->block_size) {
1047		error = blk_validate_block_size(config->block_size);
1048		if (error)
1049			goto out_unlock;
1050	}
1051
1052	error = loop_set_status_from_info(lo, &config->info);
1053	if (error)
1054		goto out_unlock;
1055
1056	if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1057	    !file->f_op->write_iter)
1058		lo->lo_flags |= LO_FLAGS_READ_ONLY;
 
1059
1060	if (!lo->workqueue) {
1061		lo->workqueue = alloc_workqueue("loop%d",
1062						WQ_UNBOUND | WQ_FREEZABLE,
1063						0, lo->lo_number);
1064		if (!lo->workqueue) {
1065			error = -ENOMEM;
1066			goto out_unlock;
1067		}
1068	}
1069
1070	disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1071	set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1072
1073	lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1074	lo->lo_device = bdev;
 
1075	lo->lo_backing_file = file;
 
 
 
1076	lo->old_gfp_mask = mapping_gfp_mask(mapping);
1077	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1078
1079	if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1080		blk_queue_write_cache(lo->lo_queue, true, false);
1081
1082	if (config->block_size)
1083		bsize = config->block_size;
1084	else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1085		/* In case of direct I/O, match underlying block size */
1086		bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1087	else
1088		bsize = 512;
1089
1090	blk_queue_logical_block_size(lo->lo_queue, bsize);
1091	blk_queue_physical_block_size(lo->lo_queue, bsize);
1092	blk_queue_io_min(lo->lo_queue, bsize);
1093
1094	loop_config_discard(lo);
1095	loop_update_rotational(lo);
1096	loop_update_dio(lo);
1097	loop_sysfs_init(lo);
 
 
1098
1099	size = get_loop_size(lo, file);
1100	loop_set_size(lo, size);
1101
1102	/* Order wrt reading lo_state in loop_validate_file(). */
1103	wmb();
1104
 
 
 
 
 
 
1105	lo->lo_state = Lo_bound;
 
1106	if (part_shift)
1107		lo->lo_flags |= LO_FLAGS_PARTSCAN;
1108	partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1109	if (partscan)
1110		clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1111
1112	loop_global_unlock(lo, is_loop);
1113	if (partscan)
1114		loop_reread_partitions(lo);
1115	if (!(mode & FMODE_EXCL))
1116		bd_abort_claiming(bdev, loop_configure);
1117
1118	error = 0;
1119done:
1120	/* enable and uncork uevent now that we are done */
1121	dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1122	return error;
1123
1124out_unlock:
1125	loop_global_unlock(lo, is_loop);
1126out_bdev:
1127	if (!(mode & FMODE_EXCL))
1128		bd_abort_claiming(bdev, loop_configure);
1129out_putf:
 
1130	fput(file);
 
1131	/* This is safe: open() is still holding a reference. */
1132	module_put(THIS_MODULE);
1133	goto done;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1134}
1135
1136static void __loop_clr_fd(struct loop_device *lo, bool release)
1137{
1138	struct file *filp;
1139	gfp_t gfp = lo->old_gfp_mask;
 
 
 
 
 
 
 
1140
1141	if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1142		blk_queue_write_cache(lo->lo_queue, false, false);
1143
1144	/*
1145	 * Freeze the request queue when unbinding on a live file descriptor and
1146	 * thus an open device.  When called from ->release we are guaranteed
1147	 * that there is no I/O in progress already.
1148	 */
1149	if (!release)
1150		blk_mq_freeze_queue(lo->lo_queue);
1151
1152	spin_lock_irq(&lo->lo_lock);
1153	filp = lo->lo_backing_file;
1154	lo->lo_backing_file = NULL;
1155	spin_unlock_irq(&lo->lo_lock);
1156
 
 
 
1157	lo->lo_device = NULL;
 
1158	lo->lo_offset = 0;
1159	lo->lo_sizelimit = 0;
 
 
 
 
1160	memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1161	blk_queue_logical_block_size(lo->lo_queue, 512);
1162	blk_queue_physical_block_size(lo->lo_queue, 512);
1163	blk_queue_io_min(lo->lo_queue, 512);
1164	invalidate_disk(lo->lo_disk);
1165	loop_sysfs_exit(lo);
1166	/* let user-space know about this change */
1167	kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
 
 
 
1168	mapping_set_gfp_mask(filp->f_mapping, gfp);
 
1169	/* This is safe: open() is still holding a reference. */
1170	module_put(THIS_MODULE);
1171	if (!release)
1172		blk_mq_unfreeze_queue(lo->lo_queue);
1173
1174	disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1175
1176	if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1177		int err;
1178
1179		/*
1180		 * open_mutex has been held already in release path, so don't
1181		 * acquire it if this function is called in such case.
1182		 *
1183		 * If the reread partition isn't from release path, lo_refcnt
1184		 * must be at least one and it can only become zero when the
1185		 * current holder is released.
1186		 */
1187		if (!release)
1188			mutex_lock(&lo->lo_disk->open_mutex);
1189		err = bdev_disk_changed(lo->lo_disk, false);
1190		if (!release)
1191			mutex_unlock(&lo->lo_disk->open_mutex);
1192		if (err)
1193			pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1194				__func__, lo->lo_number, err);
1195		/* Device is gone, no point in returning error */
1196	}
1197
1198	/*
1199	 * lo->lo_state is set to Lo_unbound here after above partscan has
1200	 * finished. There cannot be anybody else entering __loop_clr_fd() as
1201	 * Lo_rundown state protects us from all the other places trying to
1202	 * change the 'lo' device.
1203	 */
1204	lo->lo_flags = 0;
1205	if (!part_shift)
1206		set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1207	mutex_lock(&lo->lo_mutex);
1208	lo->lo_state = Lo_unbound;
1209	mutex_unlock(&lo->lo_mutex);
1210
1211	/*
1212	 * Need not hold lo_mutex to fput backing file. Calling fput holding
1213	 * lo_mutex triggers a circular lock dependency possibility warning as
1214	 * fput can take open_mutex which is usually taken before lo_mutex.
 
1215	 */
1216	fput(filp);
 
1217}
1218
1219static int loop_clr_fd(struct loop_device *lo)
 
1220{
1221	int err;
 
 
 
 
 
 
 
 
 
 
 
1222
1223	/*
1224	 * Since lo_ioctl() is called without locks held, it is possible that
1225	 * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1226	 *
1227	 * Therefore, use global lock when setting Lo_rundown state in order to
1228	 * make sure that loop_validate_file() will fail if the "struct file"
1229	 * which loop_configure()/loop_change_fd() found via fget() was this
1230	 * loop device.
1231	 */
1232	err = loop_global_lock_killable(lo, true);
1233	if (err)
1234		return err;
1235	if (lo->lo_state != Lo_bound) {
1236		loop_global_unlock(lo, true);
1237		return -ENXIO;
1238	}
1239	/*
1240	 * If we've explicitly asked to tear down the loop device,
1241	 * and it has an elevated reference count, set it for auto-teardown when
1242	 * the last reference goes away. This stops $!~#$@ udev from
1243	 * preventing teardown because it decided that it needs to run blkid on
1244	 * the loopback device whenever they appear. xfstests is notorious for
1245	 * failing tests because blkid via udev races with a losetup
1246	 * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1247	 * command to fail with EBUSY.
1248	 */
1249	if (disk_openers(lo->lo_disk) > 1) {
1250		lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1251		loop_global_unlock(lo, true);
1252		return 0;
1253	}
1254	lo->lo_state = Lo_rundown;
1255	loop_global_unlock(lo, true);
1256
1257	__loop_clr_fd(lo, false);
1258	return 0;
1259}
1260
1261static int
1262loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1263{
1264	int err;
1265	int prev_lo_flags;
1266	bool partscan = false;
1267	bool size_changed = false;
1268
1269	err = mutex_lock_killable(&lo->lo_mutex);
1270	if (err)
1271		return err;
1272	if (lo->lo_state != Lo_bound) {
1273		err = -ENXIO;
1274		goto out_unlock;
1275	}
1276
1277	if (lo->lo_offset != info->lo_offset ||
1278	    lo->lo_sizelimit != info->lo_sizelimit) {
1279		size_changed = true;
1280		sync_blockdev(lo->lo_device);
1281		invalidate_bdev(lo->lo_device);
1282	}
 
1283
1284	/* I/O need to be drained during transfer transition */
1285	blk_mq_freeze_queue(lo->lo_queue);
 
 
1286
1287	prev_lo_flags = lo->lo_flags;
 
 
 
 
 
 
 
1288
1289	err = loop_set_status_from_info(lo, info);
1290	if (err)
1291		goto out_unfreeze;
1292
1293	/* Mask out flags that can't be set using LOOP_SET_STATUS. */
1294	lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1295	/* For those flags, use the previous values instead */
1296	lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1297	/* For flags that can't be cleared, use previous values too */
1298	lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1299
1300	if (size_changed) {
1301		loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1302					   lo->lo_backing_file);
1303		loop_set_size(lo, new_size);
1304	}
1305
1306	loop_config_discard(lo);
 
 
 
 
 
 
 
1307
1308	/* update dio if lo_offset or transfer is changed */
1309	__loop_update_dio(lo, lo->use_dio);
1310
1311out_unfreeze:
1312	blk_mq_unfreeze_queue(lo->lo_queue);
1313
1314	if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1315	     !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1316		clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1317		partscan = true;
1318	}
1319out_unlock:
1320	mutex_unlock(&lo->lo_mutex);
1321	if (partscan)
1322		loop_reread_partitions(lo);
1323
1324	return err;
1325}
1326
1327static int
1328loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1329{
1330	struct path path;
1331	struct kstat stat;
1332	int ret;
1333
1334	ret = mutex_lock_killable(&lo->lo_mutex);
1335	if (ret)
1336		return ret;
1337	if (lo->lo_state != Lo_bound) {
1338		mutex_unlock(&lo->lo_mutex);
1339		return -ENXIO;
1340	}
1341
 
1342	memset(info, 0, sizeof(*info));
1343	info->lo_number = lo->lo_number;
 
 
 
1344	info->lo_offset = lo->lo_offset;
1345	info->lo_sizelimit = lo->lo_sizelimit;
1346	info->lo_flags = lo->lo_flags;
1347	memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1348
1349	/* Drop lo_mutex while we call into the filesystem. */
1350	path = lo->lo_backing_file->f_path;
1351	path_get(&path);
1352	mutex_unlock(&lo->lo_mutex);
1353	ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1354	if (!ret) {
1355		info->lo_device = huge_encode_dev(stat.dev);
1356		info->lo_inode = stat.ino;
1357		info->lo_rdevice = huge_encode_dev(stat.rdev);
1358	}
1359	path_put(&path);
1360	return ret;
1361}
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;
 
 
1373	info64->lo_flags = info->lo_flags;
1374	memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
 
 
 
 
 
 
1375}
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;
 
 
1386	info->lo_flags = info64->lo_flags;
1387	memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
 
 
 
 
 
 
1388
1389	/* error in case values were truncated */
1390	if (info->lo_device != info64->lo_device ||
1391	    info->lo_rdevice != info64->lo_rdevice ||
1392	    info->lo_inode != info64->lo_inode ||
1393	    info->lo_offset != info64->lo_offset)
1394		return -EOVERFLOW;
1395
1396	return 0;
1397}
1398
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;
1425	int err;
1426
1427	if (!arg)
1428		return -EINVAL;
1429	err = loop_get_status(lo, &info64);
 
1430	if (!err)
1431		err = loop_info64_to_old(&info64, &info);
1432	if (!err && copy_to_user(arg, &info, sizeof(info)))
1433		err = -EFAULT;
1434
1435	return err;
1436}
1437
1438static int
1439loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1440	struct loop_info64 info64;
1441	int err;
1442
1443	if (!arg)
1444		return -EINVAL;
1445	err = loop_get_status(lo, &info64);
 
1446	if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1447		err = -EFAULT;
1448
1449	return err;
1450}
1451
1452static int loop_set_capacity(struct loop_device *lo)
1453{
1454	loff_t size;
 
 
1455
 
1456	if (unlikely(lo->lo_state != Lo_bound))
1457		return -ENXIO;
1458
1459	size = get_loop_size(lo, lo->lo_backing_file);
1460	loop_set_size(lo, size);
1461
1462	return 0;
1463}
1464
1465static 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
1479static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1480{
1481	int err = 0;
1482
1483	if (lo->lo_state != Lo_bound)
1484		return -ENXIO;
1485
1486	err = blk_validate_block_size(arg);
1487	if (err)
1488		return err;
1489
1490	if (lo->lo_queue->limits.logical_block_size == arg)
1491		return 0;
1492
1493	sync_blockdev(lo->lo_device);
1494	invalidate_bdev(lo->lo_device);
1495
1496	blk_mq_freeze_queue(lo->lo_queue);
1497	blk_queue_logical_block_size(lo->lo_queue, arg);
1498	blk_queue_physical_block_size(lo->lo_queue, arg);
1499	blk_queue_io_min(lo->lo_queue, arg);
1500	loop_update_dio(lo);
1501	blk_mq_unfreeze_queue(lo->lo_queue);
1502
1503	return err;
1504}
1505
1506static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1507			   unsigned long arg)
1508{
1509	int err;
1510
1511	err = mutex_lock_killable(&lo->lo_mutex);
1512	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:
1525		err = -EINVAL;
1526	}
1527	mutex_unlock(&lo->lo_mutex);
1528	return err;
1529}
1530
1531static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1532	unsigned int cmd, unsigned long arg)
1533{
1534	struct loop_device *lo = bdev->bd_disk->private_data;
1535	void __user *argp = (void __user *) arg;
1536	int err;
1537
 
1538	switch (cmd) {
1539	case LOOP_SET_FD: {
1540		/*
1541		 * Legacy case - pass in a zeroed out struct loop_config with
1542		 * only the file descriptor set , which corresponds with the
1543		 * default parameters we'd have used otherwise.
1544		 */
1545		struct loop_config config;
1546
1547		memset(&config, 0, sizeof(config));
1548		config.fd = arg;
1549
1550		return loop_configure(lo, mode, bdev, &config);
1551	}
1552	case LOOP_CONFIGURE: {
1553		struct loop_config config;
1554
1555		if (copy_from_user(&config, argp, sizeof(config)))
1556			return -EFAULT;
1557
1558		return loop_configure(lo, mode, bdev, &config);
1559	}
1560	case LOOP_CHANGE_FD:
1561		return loop_change_fd(lo, bdev, arg);
 
1562	case LOOP_CLR_FD:
1563		return loop_clr_fd(lo);
 
 
 
 
1564	case LOOP_SET_STATUS:
1565		err = -EPERM;
1566		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1567			err = loop_set_status_old(lo, argp);
1568		}
1569		break;
1570	case LOOP_GET_STATUS:
1571		return loop_get_status_old(lo, argp);
 
1572	case LOOP_SET_STATUS64:
1573		err = -EPERM;
1574		if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1575			err = loop_set_status64(lo, argp);
1576		}
1577		break;
1578	case LOOP_GET_STATUS64:
1579		return loop_get_status64(lo, argp);
 
1580	case LOOP_SET_CAPACITY:
1581	case LOOP_SET_DIRECT_IO:
1582	case LOOP_SET_BLOCK_SIZE:
1583		if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1584			return -EPERM;
1585		fallthrough;
1586	default:
1587		err = lo_simple_ioctl(lo, cmd, arg);
1588		break;
1589	}
 
1590
 
1591	return err;
1592}
1593
1594#ifdef CONFIG_COMPAT
1595struct compat_loop_info {
1596	compat_int_t	lo_number;      /* ioctl r/o */
1597	compat_dev_t	lo_device;      /* ioctl r/o */
1598	compat_ulong_t	lo_inode;       /* ioctl r/o */
1599	compat_dev_t	lo_rdevice;     /* ioctl r/o */
1600	compat_int_t	lo_offset;
1601	compat_int_t	lo_encrypt_type;        /* obsolete, ignored */
1602	compat_int_t	lo_encrypt_key_size;    /* ioctl w/o */
1603	compat_int_t	lo_flags;       /* ioctl r/o */
1604	char		lo_name[LO_NAME_SIZE];
1605	unsigned char	lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1606	compat_ulong_t	lo_init[2];
1607	char		reserved[4];
1608};
1609
1610/*
1611 * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1612 * - noinlined to reduce stack space usage in main part of driver
1613 */
1614static noinline int
1615loop_info64_from_compat(const struct compat_loop_info __user *arg,
1616			struct loop_info64 *info64)
1617{
1618	struct compat_loop_info info;
1619
1620	if (copy_from_user(&info, arg, sizeof(info)))
1621		return -EFAULT;
1622
1623	memset(info64, 0, sizeof(*info64));
1624	info64->lo_number = info.lo_number;
1625	info64->lo_device = info.lo_device;
1626	info64->lo_inode = info.lo_inode;
1627	info64->lo_rdevice = info.lo_rdevice;
1628	info64->lo_offset = info.lo_offset;
1629	info64->lo_sizelimit = 0;
 
 
1630	info64->lo_flags = info.lo_flags;
1631	memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
 
 
 
 
 
 
1632	return 0;
1633}
1634
1635/*
1636 * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1637 * - noinlined to reduce stack space usage in main part of driver
1638 */
1639static noinline int
1640loop_info64_to_compat(const struct loop_info64 *info64,
1641		      struct compat_loop_info __user *arg)
1642{
1643	struct compat_loop_info info;
1644
1645	memset(&info, 0, sizeof(info));
1646	info.lo_number = info64->lo_number;
1647	info.lo_device = info64->lo_device;
1648	info.lo_inode = info64->lo_inode;
1649	info.lo_rdevice = info64->lo_rdevice;
1650	info.lo_offset = info64->lo_offset;
 
 
1651	info.lo_flags = info64->lo_flags;
1652	memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
 
 
 
 
 
 
1653
1654	/* error in case values were truncated */
1655	if (info.lo_device != info64->lo_device ||
1656	    info.lo_rdevice != info64->lo_rdevice ||
1657	    info.lo_inode != info64->lo_inode ||
1658	    info.lo_offset != info64->lo_offset)
 
 
1659		return -EOVERFLOW;
1660
1661	if (copy_to_user(arg, &info, sizeof(info)))
1662		return -EFAULT;
1663	return 0;
1664}
1665
1666static int
1667loop_set_status_compat(struct loop_device *lo,
1668		       const struct compat_loop_info __user *arg)
1669{
1670	struct loop_info64 info64;
1671	int ret;
1672
1673	ret = loop_info64_from_compat(arg, &info64);
1674	if (ret < 0)
1675		return ret;
1676	return loop_set_status(lo, &info64);
1677}
1678
1679static int
1680loop_get_status_compat(struct loop_device *lo,
1681		       struct compat_loop_info __user *arg)
1682{
1683	struct loop_info64 info64;
1684	int err;
1685
1686	if (!arg)
1687		return -EINVAL;
1688	err = loop_get_status(lo, &info64);
 
1689	if (!err)
1690		err = loop_info64_to_compat(&info64, arg);
1691	return err;
1692}
1693
1694static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1695			   unsigned int cmd, unsigned long arg)
1696{
1697	struct loop_device *lo = bdev->bd_disk->private_data;
1698	int err;
1699
1700	switch(cmd) {
1701	case LOOP_SET_STATUS:
1702		err = loop_set_status_compat(lo,
1703			     (const struct compat_loop_info __user *)arg);
 
 
1704		break;
1705	case LOOP_GET_STATUS:
1706		err = loop_get_status_compat(lo,
1707				     (struct compat_loop_info __user *)arg);
 
 
1708		break;
1709	case LOOP_SET_CAPACITY:
1710	case LOOP_CLR_FD:
1711	case LOOP_GET_STATUS64:
1712	case LOOP_SET_STATUS64:
1713	case LOOP_CONFIGURE:
1714		arg = (unsigned long) compat_ptr(arg);
1715		fallthrough;
1716	case LOOP_SET_FD:
1717	case LOOP_CHANGE_FD:
1718	case LOOP_SET_BLOCK_SIZE:
1719	case LOOP_SET_DIRECT_IO:
1720		err = lo_ioctl(bdev, mode, cmd, arg);
1721		break;
1722	default:
1723		err = -ENOIOCTLCMD;
1724		break;
1725	}
1726	return err;
1727}
1728#endif
1729
1730static void lo_release(struct gendisk *disk, fmode_t mode)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1731{
1732	struct loop_device *lo = disk->private_data;
 
1733
1734	if (disk_openers(disk) > 0)
1735		return;
 
 
1736
1737	mutex_lock(&lo->lo_mutex);
1738	if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
1739		lo->lo_state = Lo_rundown;
1740		mutex_unlock(&lo->lo_mutex);
1741		/*
1742		 * In autoclear mode, stop the loop thread
1743		 * and remove configuration after last close.
1744		 */
1745		__loop_clr_fd(lo, true);
1746		return;
 
 
 
 
 
 
 
1747	}
1748	mutex_unlock(&lo->lo_mutex);
1749}
1750
1751static void lo_free_disk(struct gendisk *disk)
1752{
1753	struct loop_device *lo = disk->private_data;
1754
1755	if (lo->workqueue)
1756		destroy_workqueue(lo->workqueue);
1757	loop_free_idle_workers(lo, true);
1758	timer_shutdown_sync(&lo->timer);
1759	mutex_destroy(&lo->lo_mutex);
1760	kfree(lo);
1761}
1762
1763static const struct block_device_operations lo_fops = {
1764	.owner =	THIS_MODULE,
 
1765	.release =	lo_release,
1766	.ioctl =	lo_ioctl,
1767#ifdef CONFIG_COMPAT
1768	.compat_ioctl =	lo_compat_ioctl,
1769#endif
1770	.free_disk =	lo_free_disk,
1771};
1772
1773/*
1774 * And now the modules code and kernel interface.
1775 */
1776
1777/*
1778 * If max_loop is specified, create that many devices upfront.
1779 * This also becomes a hard limit. If max_loop is not specified,
1780 * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
1781 * init time. Loop devices can be requested on-demand with the
1782 * /dev/loop-control interface, or be instantiated by accessing
1783 * a 'dead' device node.
1784 */
1785static int max_loop = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
1786module_param(max_loop, int, 0444);
1787MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1788module_param(max_part, int, 0444);
1789MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1790
1791static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1792
1793static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1794{
1795	int ret = kstrtoint(s, 10, &hw_queue_depth);
1796
1797	return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0;
1798}
1799
1800static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1801	.set	= loop_set_hw_queue_depth,
1802	.get	= param_get_int,
1803};
1804
1805device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1806MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128");
1807
1808MODULE_LICENSE("GPL");
1809MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1810
1811static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1812		const struct blk_mq_queue_data *bd)
1813{
1814	struct request *rq = bd->rq;
1815	struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1816	struct loop_device *lo = rq->q->queuedata;
1817
1818	blk_mq_start_request(rq);
1819
1820	if (lo->lo_state != Lo_bound)
1821		return BLK_STS_IOERR;
1822
1823	switch (req_op(rq)) {
1824	case REQ_OP_FLUSH:
1825	case REQ_OP_DISCARD:
1826	case REQ_OP_WRITE_ZEROES:
1827		cmd->use_aio = false;
1828		break;
1829	default:
1830		cmd->use_aio = lo->use_dio;
1831		break;
1832	}
1833
1834	/* always use the first bio's css */
1835	cmd->blkcg_css = NULL;
1836	cmd->memcg_css = NULL;
1837#ifdef CONFIG_BLK_CGROUP
1838	if (rq->bio) {
1839		cmd->blkcg_css = bio_blkcg_css(rq->bio);
1840#ifdef CONFIG_MEMCG
1841		if (cmd->blkcg_css) {
1842			cmd->memcg_css =
1843				cgroup_get_e_css(cmd->blkcg_css->cgroup,
1844						&memory_cgrp_subsys);
1845		}
1846#endif
1847	}
1848#endif
1849	loop_queue_work(lo, cmd);
1850
1851	return BLK_STS_OK;
1852}
1853
1854static void loop_handle_cmd(struct loop_cmd *cmd)
1855{
1856	struct request *rq = blk_mq_rq_from_pdu(cmd);
1857	const bool write = op_is_write(req_op(rq));
1858	struct loop_device *lo = rq->q->queuedata;
1859	int ret = 0;
1860	struct mem_cgroup *old_memcg = NULL;
1861
1862	if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1863		ret = -EIO;
1864		goto failed;
1865	}
1866
1867	if (cmd->blkcg_css)
1868		kthread_associate_blkcg(cmd->blkcg_css);
1869	if (cmd->memcg_css)
1870		old_memcg = set_active_memcg(
1871			mem_cgroup_from_css(cmd->memcg_css));
1872
1873	ret = do_req_filebacked(lo, rq);
1874
1875	if (cmd->blkcg_css)
1876		kthread_associate_blkcg(NULL);
1877
1878	if (cmd->memcg_css) {
1879		set_active_memcg(old_memcg);
1880		css_put(cmd->memcg_css);
1881	}
1882 failed:
1883	/* complete non-aio request */
1884	if (!cmd->use_aio || ret) {
1885		if (ret == -EOPNOTSUPP)
1886			cmd->ret = ret;
1887		else
1888			cmd->ret = ret ? -EIO : 0;
1889		if (likely(!blk_should_fake_timeout(rq->q)))
1890			blk_mq_complete_request(rq);
1891	}
1892}
1893
1894static void loop_process_work(struct loop_worker *worker,
1895			struct list_head *cmd_list, struct loop_device *lo)
1896{
1897	int orig_flags = current->flags;
1898	struct loop_cmd *cmd;
1899
1900	current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1901	spin_lock_irq(&lo->lo_work_lock);
1902	while (!list_empty(cmd_list)) {
1903		cmd = container_of(
1904			cmd_list->next, struct loop_cmd, list_entry);
1905		list_del(cmd_list->next);
1906		spin_unlock_irq(&lo->lo_work_lock);
1907
1908		loop_handle_cmd(cmd);
1909		cond_resched();
1910
1911		spin_lock_irq(&lo->lo_work_lock);
1912	}
1913
1914	/*
1915	 * We only add to the idle list if there are no pending cmds
1916	 * *and* the worker will not run again which ensures that it
1917	 * is safe to free any worker on the idle list
1918	 */
1919	if (worker && !work_pending(&worker->work)) {
1920		worker->last_ran_at = jiffies;
1921		list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1922		loop_set_timer(lo);
1923	}
1924	spin_unlock_irq(&lo->lo_work_lock);
1925	current->flags = orig_flags;
1926}
1927
1928static void loop_workfn(struct work_struct *work)
1929{
1930	struct loop_worker *worker =
1931		container_of(work, struct loop_worker, work);
1932	loop_process_work(worker, &worker->cmd_list, worker->lo);
1933}
1934
1935static void loop_rootcg_workfn(struct work_struct *work)
1936{
1937	struct loop_device *lo =
1938		container_of(work, struct loop_device, rootcg_work);
1939	loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1940}
1941
1942static const struct blk_mq_ops loop_mq_ops = {
1943	.queue_rq       = loop_queue_rq,
1944	.complete	= lo_complete_rq,
1945};
1946
1947static int loop_add(int i)
1948{
1949	struct loop_device *lo;
1950	struct gendisk *disk;
1951	int err;
1952
1953	err = -ENOMEM;
1954	lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1955	if (!lo)
1956		goto out;
1957	lo->worker_tree = RB_ROOT;
1958	INIT_LIST_HEAD(&lo->idle_worker_list);
1959	timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
1960	lo->lo_state = Lo_unbound;
1961
1962	err = mutex_lock_killable(&loop_ctl_mutex);
1963	if (err)
1964		goto out_free_dev;
1965
1966	/* allocate id, if @id >= 0, we're requesting that specific id */
1967	if (i >= 0) {
1968		err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1969		if (err == -ENOSPC)
 
 
 
 
1970			err = -EEXIST;
 
 
 
 
 
 
 
 
1971	} else {
1972		err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1973	}
1974	mutex_unlock(&loop_ctl_mutex);
1975	if (err < 0)
1976		goto out_free_dev;
1977	i = err;
1978
1979	lo->tag_set.ops = &loop_mq_ops;
1980	lo->tag_set.nr_hw_queues = 1;
1981	lo->tag_set.queue_depth = hw_queue_depth;
1982	lo->tag_set.numa_node = NUMA_NO_NODE;
1983	lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1984	lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
1985		BLK_MQ_F_NO_SCHED_BY_DEFAULT;
1986	lo->tag_set.driver_data = lo;
1987
1988	err = blk_mq_alloc_tag_set(&lo->tag_set);
1989	if (err)
1990		goto out_free_idr;
1991
1992	disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
1993	if (IS_ERR(disk)) {
1994		err = PTR_ERR(disk);
1995		goto out_cleanup_tags;
1996	}
1997	lo->lo_queue = lo->lo_disk->queue;
1998
1999	blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
2000
2001	/*
2002	 * By default, we do buffer IO, so it doesn't make sense to enable
2003	 * merge because the I/O submitted to backing file is handled page by
2004	 * page. For directio mode, merge does help to dispatch bigger request
2005	 * to underlayer disk. We will enable merge once directio is enabled.
2006	 */
2007	blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
2008
2009	/*
2010	 * Disable partition scanning by default. The in-kernel partition
2011	 * scanning can be requested individually per-device during its
2012	 * setup. Userspace can always add and remove partitions from all
2013	 * devices. The needed partition minors are allocated from the
2014	 * extended minor space, the main loop device numbers will continue
2015	 * to match the loop minors, regardless of the number of partitions
2016	 * used.
2017	 *
2018	 * If max_part is given, partition scanning is globally enabled for
2019	 * all loop devices. The minors for the main loop devices will be
2020	 * multiples of max_part.
2021	 *
2022	 * Note: Global-for-all-devices, set-only-at-init, read-only module
2023	 * parameteters like 'max_loop' and 'max_part' make things needlessly
2024	 * complicated, are too static, inflexible and may surprise
2025	 * userspace tools. Parameters like this in general should be avoided.
2026	 */
2027	if (!part_shift)
2028		set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2029	mutex_init(&lo->lo_mutex);
 
2030	lo->lo_number		= i;
 
 
2031	spin_lock_init(&lo->lo_lock);
2032	spin_lock_init(&lo->lo_work_lock);
2033	INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2034	INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2035	disk->major		= LOOP_MAJOR;
2036	disk->first_minor	= i << part_shift;
2037	disk->minors		= 1 << part_shift;
2038	disk->fops		= &lo_fops;
2039	disk->private_data	= lo;
2040	disk->queue		= lo->lo_queue;
2041	disk->events		= DISK_EVENT_MEDIA_CHANGE;
2042	disk->event_flags	= DISK_EVENT_FLAG_UEVENT;
2043	sprintf(disk->disk_name, "loop%d", i);
2044	/* Make this loop device reachable from pathname. */
2045	err = add_disk(disk);
2046	if (err)
2047		goto out_cleanup_disk;
2048
2049	/* Show this loop device. */
2050	mutex_lock(&loop_ctl_mutex);
2051	lo->idr_visible = true;
2052	mutex_unlock(&loop_ctl_mutex);
2053
2054	return i;
2055
2056out_cleanup_disk:
2057	put_disk(disk);
2058out_cleanup_tags:
2059	blk_mq_free_tag_set(&lo->tag_set);
2060out_free_idr:
2061	mutex_lock(&loop_ctl_mutex);
2062	idr_remove(&loop_index_idr, i);
2063	mutex_unlock(&loop_ctl_mutex);
2064out_free_dev:
2065	kfree(lo);
2066out:
2067	return err;
2068}
2069
2070static void loop_remove(struct loop_device *lo)
2071{
2072	/* Make this loop device unreachable from pathname. */
2073	del_gendisk(lo->lo_disk);
2074	blk_mq_free_tag_set(&lo->tag_set);
2075
2076	mutex_lock(&loop_ctl_mutex);
2077	idr_remove(&loop_index_idr, lo->lo_number);
2078	mutex_unlock(&loop_ctl_mutex);
2079
2080	put_disk(lo->lo_disk);
 
2081}
2082
2083static void loop_probe(dev_t dev)
2084{
2085	int idx = MINOR(dev) >> part_shift;
 
2086
2087	if (max_loop && idx >= max_loop)
2088		return;
2089	loop_add(idx);
 
 
2090}
2091
2092static int loop_control_remove(int idx)
2093{
2094	struct loop_device *lo;
2095	int ret;
 
 
 
2096
2097	if (idx < 0) {
2098		pr_warn_once("deleting an unspecified loop device is not supported.\n");
2099		return -EINVAL;
 
 
 
2100	}
2101		
2102	/* Hide this loop device for serialization. */
2103	ret = mutex_lock_killable(&loop_ctl_mutex);
2104	if (ret)
2105		return ret;
2106	lo = idr_find(&loop_index_idr, idx);
2107	if (!lo || !lo->idr_visible)
2108		ret = -ENODEV;
2109	else
2110		lo->idr_visible = false;
2111	mutex_unlock(&loop_ctl_mutex);
2112	if (ret)
2113		return ret;
2114
2115	/* Check whether this loop device can be removed. */
2116	ret = mutex_lock_killable(&lo->lo_mutex);
2117	if (ret)
2118		goto mark_visible;
2119	if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2120		mutex_unlock(&lo->lo_mutex);
2121		ret = -EBUSY;
2122		goto mark_visible;
2123	}
2124	/* Mark this loop device as no more bound, but not quite unbound yet */
2125	lo->lo_state = Lo_deleting;
2126	mutex_unlock(&lo->lo_mutex);
2127
2128	loop_remove(lo);
2129	return 0;
2130
2131mark_visible:
2132	/* Show this loop device again. */
2133	mutex_lock(&loop_ctl_mutex);
2134	lo->idr_visible = true;
2135	mutex_unlock(&loop_ctl_mutex);
2136	return ret;
2137}
2138
2139static int loop_control_get_free(int idx)
2140{
2141	struct loop_device *lo;
2142	int id, ret;
 
2143
2144	ret = mutex_lock_killable(&loop_ctl_mutex);
2145	if (ret)
2146		return ret;
2147	idr_for_each_entry(&loop_index_idr, lo, id) {
2148		/* Hitting a race results in creating a new loop device which is harmless. */
2149		if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2150			goto found;
2151	}
2152	mutex_unlock(&loop_ctl_mutex);
2153	return loop_add(-1);
2154found:
2155	mutex_unlock(&loop_ctl_mutex);
2156	return id;
2157}
2158
2159static long loop_control_ioctl(struct file *file, unsigned int cmd,
2160			       unsigned long parm)
2161{
 
 
 
 
2162	switch (cmd) {
2163	case LOOP_CTL_ADD:
2164		return loop_add(parm);
 
 
 
 
 
 
2165	case LOOP_CTL_REMOVE:
2166		return loop_control_remove(parm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2167	case LOOP_CTL_GET_FREE:
2168		return loop_control_get_free(parm);
2169	default:
2170		return -ENOSYS;
 
2171	}
 
 
 
2172}
2173
2174static const struct file_operations loop_ctl_fops = {
2175	.open		= nonseekable_open,
2176	.unlocked_ioctl	= loop_control_ioctl,
2177	.compat_ioctl	= loop_control_ioctl,
2178	.owner		= THIS_MODULE,
2179	.llseek		= noop_llseek,
2180};
2181
2182static struct miscdevice loop_misc = {
2183	.minor		= LOOP_CTRL_MINOR,
2184	.name		= "loop-control",
2185	.fops		= &loop_ctl_fops,
2186};
2187
2188MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2189MODULE_ALIAS("devname:loop-control");
2190
2191static int __init loop_init(void)
2192{
2193	int i;
 
 
2194	int err;
2195
 
 
 
 
2196	part_shift = 0;
2197	if (max_part > 0) {
2198		part_shift = fls(max_part);
2199
2200		/*
2201		 * Adjust max_part according to part_shift as it is exported
2202		 * to user space so that user can decide correct minor number
2203		 * if [s]he want to create more devices.
2204		 *
2205		 * Note that -1 is required because partition 0 is reserved
2206		 * for the whole disk.
2207		 */
2208		max_part = (1UL << part_shift) - 1;
2209	}
2210
2211	if ((1UL << part_shift) > DISK_MAX_PARTS) {
2212		err = -EINVAL;
2213		goto err_out;
2214	}
 
2215
2216	if (max_loop > 1UL << (MINORBITS - part_shift)) {
2217		err = -EINVAL;
2218		goto err_out;
 
 
 
 
 
 
 
 
 
 
 
2219	}
2220
2221	err = misc_register(&loop_misc);
2222	if (err < 0)
2223		goto err_out;
2224
2225
2226	if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2227		err = -EIO;
2228		goto misc_out;
2229	}
2230
2231	/* pre-create number of devices given by config or max_loop */
2232	for (i = 0; i < max_loop; i++)
2233		loop_add(i);
 
 
2234
2235	printk(KERN_INFO "loop: module loaded\n");
2236	return 0;
 
2237
2238misc_out:
2239	misc_deregister(&loop_misc);
2240err_out:
2241	return err;
 
 
2242}
2243
2244static void __exit loop_exit(void)
2245{
2246	struct loop_device *lo;
2247	int id;
 
 
 
 
 
2248
 
2249	unregister_blkdev(LOOP_MAJOR, "loop");
 
2250	misc_deregister(&loop_misc);
2251
2252	/*
2253	 * There is no need to use loop_ctl_mutex here, for nobody else can
2254	 * access loop_index_idr when this module is unloading (unless forced
2255	 * module unloading is requested). If this is not a clean unloading,
2256	 * we have no means to avoid kernel crash.
2257	 */
2258	idr_for_each_entry(&loop_index_idr, lo, id)
2259		loop_remove(lo);
2260
2261	idr_destroy(&loop_index_idr);
2262}
2263
2264module_init(loop_init);
2265module_exit(loop_exit);
2266
2267#ifndef MODULE
2268static int __init max_loop_setup(char *str)
2269{
2270	max_loop = simple_strtol(str, NULL, 0);
2271	return 1;
2272}
2273
2274__setup("max_loop=", max_loop_setup);
2275#endif