11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/drivers/block/loop.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Written by Theodore Ts'o, 3/29/93 51da177e4SLinus Torvalds * 61da177e4SLinus Torvalds * Copyright 1993 by Theodore Ts'o. Redistribution of this file is 71da177e4SLinus Torvalds * permitted under the GNU General Public License. 81da177e4SLinus Torvalds * 91da177e4SLinus Torvalds * DES encryption plus some minor changes by Werner Almesberger, 30-MAY-1993 101da177e4SLinus Torvalds * more DES encryption plus IDEA encryption by Nicholas J. Leon, June 20, 1996 111da177e4SLinus Torvalds * 121da177e4SLinus Torvalds * Modularized and updated for 1.1.16 kernel - Mitch Dsouza 28th May 1994 131da177e4SLinus Torvalds * Adapted for 1.3.59 kernel - Andries Brouwer, 1 Feb 1996 141da177e4SLinus Torvalds * 151da177e4SLinus Torvalds * Fixed do_loop_request() re-entrancy - Vincent.Renardias@waw.com Mar 20, 1997 161da177e4SLinus Torvalds * 171da177e4SLinus Torvalds * Added devfs support - Richard Gooch <rgooch@atnf.csiro.au> 16-Jan-1998 181da177e4SLinus Torvalds * 191da177e4SLinus Torvalds * Handle sparse backing files correctly - Kenn Humborg, Jun 28, 1998 201da177e4SLinus Torvalds * 211da177e4SLinus Torvalds * Loadable modules and other fixes by AK, 1998 221da177e4SLinus Torvalds * 231da177e4SLinus Torvalds * Make real block number available to downstream transfer functions, enables 241da177e4SLinus Torvalds * CBC (and relatives) mode encryption requiring unique IVs per data block. 251da177e4SLinus Torvalds * Reed H. Petty, rhp@draper.net 261da177e4SLinus Torvalds * 271da177e4SLinus Torvalds * Maximum number of loop devices now dynamic via max_loop module parameter. 281da177e4SLinus Torvalds * Russell Kroll <rkroll@exploits.org> 19990701 291da177e4SLinus Torvalds * 301da177e4SLinus Torvalds * Maximum number of loop devices when compiled-in now selectable by passing 311da177e4SLinus Torvalds * max_loop=<1-255> to the kernel on boot. 3296de0e25SJan Engelhardt * Erik I. Bolsø, <eriki@himolde.no>, Oct 31, 1999 331da177e4SLinus Torvalds * 341da177e4SLinus Torvalds * Completely rewrite request handling to be make_request_fn style and 351da177e4SLinus Torvalds * non blocking, pushing work to a helper thread. Lots of fixes from 361da177e4SLinus Torvalds * Al Viro too. 371da177e4SLinus Torvalds * Jens Axboe <axboe@suse.de>, Nov 2000 381da177e4SLinus Torvalds * 391da177e4SLinus Torvalds * Support up to 256 loop devices 401da177e4SLinus Torvalds * Heinz Mauelshagen <mge@sistina.com>, Feb 2002 411da177e4SLinus Torvalds * 421da177e4SLinus Torvalds * Support for falling back on the write file operation when the address space 434e02ed4bSNick Piggin * operations write_begin is not available on the backing filesystem. 441da177e4SLinus Torvalds * Anton Altaparmakov, 16 Feb 2005 451da177e4SLinus Torvalds * 461da177e4SLinus Torvalds * Still To Fix: 471da177e4SLinus Torvalds * - Advisory locking is ignored here. 481da177e4SLinus Torvalds * - Should use an own CAP_* category instead of CAP_SYS_ADMIN 491da177e4SLinus Torvalds * 501da177e4SLinus Torvalds */ 511da177e4SLinus Torvalds 521da177e4SLinus Torvalds #include <linux/module.h> 531da177e4SLinus Torvalds #include <linux/moduleparam.h> 541da177e4SLinus Torvalds #include <linux/sched.h> 551da177e4SLinus Torvalds #include <linux/fs.h> 561da177e4SLinus Torvalds #include <linux/file.h> 571da177e4SLinus Torvalds #include <linux/stat.h> 581da177e4SLinus Torvalds #include <linux/errno.h> 591da177e4SLinus Torvalds #include <linux/major.h> 601da177e4SLinus Torvalds #include <linux/wait.h> 611da177e4SLinus Torvalds #include <linux/blkdev.h> 621da177e4SLinus Torvalds #include <linux/blkpg.h> 631da177e4SLinus Torvalds #include <linux/init.h> 641da177e4SLinus Torvalds #include <linux/swap.h> 651da177e4SLinus Torvalds #include <linux/slab.h> 66863d5b82SDavid Howells #include <linux/compat.h> 671da177e4SLinus Torvalds #include <linux/suspend.h> 6883144186SRafael J. Wysocki #include <linux/freezer.h> 692a48fc0aSArnd Bergmann #include <linux/mutex.h> 701da177e4SLinus Torvalds #include <linux/writeback.h> 711da177e4SLinus Torvalds #include <linux/completion.h> 721da177e4SLinus Torvalds #include <linux/highmem.h> 736c997918SSerge E. Hallyn #include <linux/kthread.h> 74d6b29d7cSJens Axboe #include <linux/splice.h> 75ee862730SMilan Broz #include <linux/sysfs.h> 76770fe30aSKay Sievers #include <linux/miscdevice.h> 77dfaa2ef6SLukas Czerner #include <linux/falloc.h> 78283e7e5dSAl Viro #include <linux/uio.h> 79d9a08a9eSAdam Manzanares #include <linux/ioprio.h> 80db6638d7SDennis Zhou #include <linux/blk-cgroup.h> 81d9a08a9eSAdam Manzanares 8283a87611SAl Viro #include "loop.h" 831da177e4SLinus Torvalds 847c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 851da177e4SLinus Torvalds 8634dd82afSKay Sievers static DEFINE_IDR(loop_index_idr); 87310ca162STetsuo Handa static DEFINE_MUTEX(loop_ctl_mutex); 881da177e4SLinus Torvalds 89476a4813SLaurent Vivier static int max_part; 90476a4813SLaurent Vivier static int part_shift; 91476a4813SLaurent Vivier 921da177e4SLinus Torvalds static int transfer_xor(struct loop_device *lo, int cmd, 931da177e4SLinus Torvalds struct page *raw_page, unsigned raw_off, 941da177e4SLinus Torvalds struct page *loop_page, unsigned loop_off, 951da177e4SLinus Torvalds int size, sector_t real_block) 961da177e4SLinus Torvalds { 97cfd8005cSCong Wang char *raw_buf = kmap_atomic(raw_page) + raw_off; 98cfd8005cSCong Wang char *loop_buf = kmap_atomic(loop_page) + loop_off; 991da177e4SLinus Torvalds char *in, *out, *key; 1001da177e4SLinus Torvalds int i, keysize; 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds if (cmd == READ) { 1031da177e4SLinus Torvalds in = raw_buf; 1041da177e4SLinus Torvalds out = loop_buf; 1051da177e4SLinus Torvalds } else { 1061da177e4SLinus Torvalds in = loop_buf; 1071da177e4SLinus Torvalds out = raw_buf; 1081da177e4SLinus Torvalds } 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds key = lo->lo_encrypt_key; 1111da177e4SLinus Torvalds keysize = lo->lo_encrypt_key_size; 1121da177e4SLinus Torvalds for (i = 0; i < size; i++) 1131da177e4SLinus Torvalds *out++ = *in++ ^ key[(i & 511) % keysize]; 1141da177e4SLinus Torvalds 115cfd8005cSCong Wang kunmap_atomic(loop_buf); 116cfd8005cSCong Wang kunmap_atomic(raw_buf); 1171da177e4SLinus Torvalds cond_resched(); 1181da177e4SLinus Torvalds return 0; 1191da177e4SLinus Torvalds } 1201da177e4SLinus Torvalds 1211da177e4SLinus Torvalds static int xor_init(struct loop_device *lo, const struct loop_info64 *info) 1221da177e4SLinus Torvalds { 1231da177e4SLinus Torvalds if (unlikely(info->lo_encrypt_key_size <= 0)) 1241da177e4SLinus Torvalds return -EINVAL; 1251da177e4SLinus Torvalds return 0; 1261da177e4SLinus Torvalds } 1271da177e4SLinus Torvalds 1281da177e4SLinus Torvalds static struct loop_func_table none_funcs = { 1291da177e4SLinus Torvalds .number = LO_CRYPT_NONE, 1301da177e4SLinus Torvalds }; 1311da177e4SLinus Torvalds 1321da177e4SLinus Torvalds static struct loop_func_table xor_funcs = { 1331da177e4SLinus Torvalds .number = LO_CRYPT_XOR, 1341da177e4SLinus Torvalds .transfer = transfer_xor, 1351da177e4SLinus Torvalds .init = xor_init 1361da177e4SLinus Torvalds }; 1371da177e4SLinus Torvalds 1381da177e4SLinus Torvalds /* xfer_funcs[0] is special - its release function is never called */ 1391da177e4SLinus Torvalds static struct loop_func_table *xfer_funcs[MAX_LO_CRYPT] = { 1401da177e4SLinus Torvalds &none_funcs, 1411da177e4SLinus Torvalds &xor_funcs 1421da177e4SLinus Torvalds }; 1431da177e4SLinus Torvalds 1447035b5dfSDmitry Monakhov static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 1451da177e4SLinus Torvalds { 146b7a1da69SGuo Chao loff_t loopsize; 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds /* Compute loopsize in bytes */ 149b7a1da69SGuo Chao loopsize = i_size_read(file->f_mapping->host); 150b7a1da69SGuo Chao if (offset > 0) 151b7a1da69SGuo Chao loopsize -= offset; 152b7a1da69SGuo Chao /* offset is beyond i_size, weird but possible */ 1537035b5dfSDmitry Monakhov if (loopsize < 0) 1547035b5dfSDmitry Monakhov return 0; 1551da177e4SLinus Torvalds 1567035b5dfSDmitry Monakhov if (sizelimit > 0 && sizelimit < loopsize) 1577035b5dfSDmitry Monakhov loopsize = sizelimit; 1581da177e4SLinus Torvalds /* 1591da177e4SLinus Torvalds * Unfortunately, if we want to do I/O on the device, 1601da177e4SLinus Torvalds * the number of 512-byte sectors has to fit into a sector_t. 1611da177e4SLinus Torvalds */ 1621da177e4SLinus Torvalds return loopsize >> 9; 1631da177e4SLinus Torvalds } 1641da177e4SLinus Torvalds 1657035b5dfSDmitry Monakhov static loff_t get_loop_size(struct loop_device *lo, struct file *file) 1661da177e4SLinus Torvalds { 1677035b5dfSDmitry Monakhov return get_size(lo->lo_offset, lo->lo_sizelimit, file); 1687035b5dfSDmitry Monakhov } 1697035b5dfSDmitry Monakhov 1702e5ab5f3SMing Lei static void __loop_update_dio(struct loop_device *lo, bool dio) 1712e5ab5f3SMing Lei { 1722e5ab5f3SMing Lei struct file *file = lo->lo_backing_file; 1732e5ab5f3SMing Lei struct address_space *mapping = file->f_mapping; 1742e5ab5f3SMing Lei struct inode *inode = mapping->host; 1752e5ab5f3SMing Lei unsigned short sb_bsize = 0; 1762e5ab5f3SMing Lei unsigned dio_align = 0; 1772e5ab5f3SMing Lei bool use_dio; 1782e5ab5f3SMing Lei 1792e5ab5f3SMing Lei if (inode->i_sb->s_bdev) { 1802e5ab5f3SMing Lei sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1812e5ab5f3SMing Lei dio_align = sb_bsize - 1; 1822e5ab5f3SMing Lei } 1832e5ab5f3SMing Lei 1842e5ab5f3SMing Lei /* 1852e5ab5f3SMing Lei * We support direct I/O only if lo_offset is aligned with the 1862e5ab5f3SMing Lei * logical I/O size of backing device, and the logical block 1872e5ab5f3SMing Lei * size of loop is bigger than the backing device's and the loop 1882e5ab5f3SMing Lei * needn't transform transfer. 1892e5ab5f3SMing Lei * 1902e5ab5f3SMing Lei * TODO: the above condition may be loosed in the future, and 1912e5ab5f3SMing Lei * direct I/O may be switched runtime at that time because most 19289d790abSMasahiro Yamada * of requests in sane applications should be PAGE_SIZE aligned 1932e5ab5f3SMing Lei */ 1942e5ab5f3SMing Lei if (dio) { 1952e5ab5f3SMing Lei if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 1962e5ab5f3SMing Lei !(lo->lo_offset & dio_align) && 1972e5ab5f3SMing Lei mapping->a_ops->direct_IO && 1982e5ab5f3SMing Lei !lo->transfer) 1992e5ab5f3SMing Lei use_dio = true; 2002e5ab5f3SMing Lei else 2012e5ab5f3SMing Lei use_dio = false; 2022e5ab5f3SMing Lei } else { 2032e5ab5f3SMing Lei use_dio = false; 2042e5ab5f3SMing Lei } 2052e5ab5f3SMing Lei 2062e5ab5f3SMing Lei if (lo->use_dio == use_dio) 2072e5ab5f3SMing Lei return; 2082e5ab5f3SMing Lei 2092e5ab5f3SMing Lei /* flush dirty pages before changing direct IO */ 2102e5ab5f3SMing Lei vfs_fsync(file, 0); 2112e5ab5f3SMing Lei 2122e5ab5f3SMing Lei /* 2132e5ab5f3SMing Lei * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 2142e5ab5f3SMing Lei * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 2152e5ab5f3SMing Lei * will get updated by ioctl(LOOP_GET_STATUS) 2162e5ab5f3SMing Lei */ 2170fbcf579SMartijn Coenen if (lo->lo_state == Lo_bound) 2182e5ab5f3SMing Lei blk_mq_freeze_queue(lo->lo_queue); 2192e5ab5f3SMing Lei lo->use_dio = use_dio; 22040326d8aSShaohua Li if (use_dio) { 2218b904b5bSBart Van Assche blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2222e5ab5f3SMing Lei lo->lo_flags |= LO_FLAGS_DIRECT_IO; 22340326d8aSShaohua Li } else { 2248b904b5bSBart Van Assche blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2252e5ab5f3SMing Lei lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 22640326d8aSShaohua Li } 2270fbcf579SMartijn Coenen if (lo->lo_state == Lo_bound) 2282e5ab5f3SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 2292e5ab5f3SMing Lei } 2302e5ab5f3SMing Lei 2315795b6f5SMartijn Coenen /** 2325795b6f5SMartijn Coenen * loop_set_size() - sets device size and notifies userspace 2335795b6f5SMartijn Coenen * @lo: struct loop_device to set the size for 2345795b6f5SMartijn Coenen * @size: new size of the loop device 2355795b6f5SMartijn Coenen * 2365795b6f5SMartijn Coenen * Callers must validate that the size passed into this function fits into 2375795b6f5SMartijn Coenen * a sector_t, eg using loop_validate_size() 2385795b6f5SMartijn Coenen */ 2395795b6f5SMartijn Coenen static void loop_set_size(struct loop_device *lo, loff_t size) 2405795b6f5SMartijn Coenen { 2415795b6f5SMartijn Coenen struct block_device *bdev = lo->lo_device; 2425795b6f5SMartijn Coenen 2435795b6f5SMartijn Coenen bd_set_size(bdev, size << SECTOR_SHIFT); 244716ad098SMartijn Coenen 245716ad098SMartijn Coenen set_capacity_revalidate_and_notify(lo->lo_disk, size, false); 2465795b6f5SMartijn Coenen } 2475795b6f5SMartijn Coenen 248083a6a50SMartijn Coenen static void 2491e6ec9eaSOmar Sandoval figure_loop_size(struct loop_device *lo, loff_t offset, loff_t sizelimit) 2507035b5dfSDmitry Monakhov { 2517035b5dfSDmitry Monakhov loff_t size = get_size(offset, sizelimit, lo->lo_backing_file); 2521da177e4SLinus Torvalds 2535795b6f5SMartijn Coenen loop_set_size(lo, size); 2541da177e4SLinus Torvalds } 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds static inline int 2571da177e4SLinus Torvalds lo_do_transfer(struct loop_device *lo, int cmd, 2581da177e4SLinus Torvalds struct page *rpage, unsigned roffs, 2591da177e4SLinus Torvalds struct page *lpage, unsigned loffs, 2601da177e4SLinus Torvalds int size, sector_t rblock) 2611da177e4SLinus Torvalds { 262aa4d8616SChristoph Hellwig int ret; 263aa4d8616SChristoph Hellwig 264aa4d8616SChristoph Hellwig ret = lo->transfer(lo, cmd, rpage, roffs, lpage, loffs, size, rblock); 265aa4d8616SChristoph Hellwig if (likely(!ret)) 2661da177e4SLinus Torvalds return 0; 2671da177e4SLinus Torvalds 268aa4d8616SChristoph Hellwig printk_ratelimited(KERN_ERR 269aa4d8616SChristoph Hellwig "loop: Transfer error at byte offset %llu, length %i.\n", 270aa4d8616SChristoph Hellwig (unsigned long long)rblock << 9, size); 271aa4d8616SChristoph Hellwig return ret; 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 274aa4d8616SChristoph Hellwig static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 2751da177e4SLinus Torvalds { 276aa4d8616SChristoph Hellwig struct iov_iter i; 2771da177e4SLinus Torvalds ssize_t bw; 278283e7e5dSAl Viro 279b6207430SChristoph Hellwig iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 2801da177e4SLinus Torvalds 28103d95eb2SAl Viro file_start_write(file); 282abbb6589SChristoph Hellwig bw = vfs_iter_write(file, &i, ppos, 0); 28303d95eb2SAl Viro file_end_write(file); 284aa4d8616SChristoph Hellwig 285aa4d8616SChristoph Hellwig if (likely(bw == bvec->bv_len)) 2861da177e4SLinus Torvalds return 0; 287aa4d8616SChristoph Hellwig 288aa4d8616SChristoph Hellwig printk_ratelimited(KERN_ERR 289aa4d8616SChristoph Hellwig "loop: Write error at byte offset %llu, length %i.\n", 290aa4d8616SChristoph Hellwig (unsigned long long)*ppos, bvec->bv_len); 2911da177e4SLinus Torvalds if (bw >= 0) 2921da177e4SLinus Torvalds bw = -EIO; 2931da177e4SLinus Torvalds return bw; 2941da177e4SLinus Torvalds } 2951da177e4SLinus Torvalds 296aa4d8616SChristoph Hellwig static int lo_write_simple(struct loop_device *lo, struct request *rq, 297aa4d8616SChristoph Hellwig loff_t pos) 2981da177e4SLinus Torvalds { 299aa4d8616SChristoph Hellwig struct bio_vec bvec; 300aa4d8616SChristoph Hellwig struct req_iterator iter; 301aa4d8616SChristoph Hellwig int ret = 0; 302aa4d8616SChristoph Hellwig 303aa4d8616SChristoph Hellwig rq_for_each_segment(bvec, rq, iter) { 304aa4d8616SChristoph Hellwig ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 305aa4d8616SChristoph Hellwig if (ret < 0) 306aa4d8616SChristoph Hellwig break; 3071da177e4SLinus Torvalds cond_resched(); 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds 310aa4d8616SChristoph Hellwig return ret; 311aa4d8616SChristoph Hellwig } 312aa4d8616SChristoph Hellwig 313aa4d8616SChristoph Hellwig /* 314456be148SChristoph Hellwig * This is the slow, transforming version that needs to double buffer the 315456be148SChristoph Hellwig * data as it cannot do the transformations in place without having direct 316456be148SChristoph Hellwig * access to the destination pages of the backing file. 3171da177e4SLinus Torvalds */ 318aa4d8616SChristoph Hellwig static int lo_write_transfer(struct loop_device *lo, struct request *rq, 319aa4d8616SChristoph Hellwig loff_t pos) 3201da177e4SLinus Torvalds { 321aa4d8616SChristoph Hellwig struct bio_vec bvec, b; 32230112013SMing Lei struct req_iterator iter; 323aa4d8616SChristoph Hellwig struct page *page; 3247988613bSKent Overstreet int ret = 0; 3251da177e4SLinus Torvalds 326aa4d8616SChristoph Hellwig page = alloc_page(GFP_NOIO); 3271da177e4SLinus Torvalds if (unlikely(!page)) 328aa4d8616SChristoph Hellwig return -ENOMEM; 329456be148SChristoph Hellwig 33030112013SMing Lei rq_for_each_segment(bvec, rq, iter) { 331aa4d8616SChristoph Hellwig ret = lo_do_transfer(lo, WRITE, page, 0, bvec.bv_page, 332aa4d8616SChristoph Hellwig bvec.bv_offset, bvec.bv_len, pos >> 9); 333aa4d8616SChristoph Hellwig if (unlikely(ret)) 334aa4d8616SChristoph Hellwig break; 335aa4d8616SChristoph Hellwig 336aa4d8616SChristoph Hellwig b.bv_page = page; 337aa4d8616SChristoph Hellwig b.bv_offset = 0; 338aa4d8616SChristoph Hellwig b.bv_len = bvec.bv_len; 339aa4d8616SChristoph Hellwig ret = lo_write_bvec(lo->lo_backing_file, &b, &pos); 3401da177e4SLinus Torvalds if (ret < 0) 3411da177e4SLinus Torvalds break; 3421da177e4SLinus Torvalds } 343aa4d8616SChristoph Hellwig 3441da177e4SLinus Torvalds __free_page(page); 3451da177e4SLinus Torvalds return ret; 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds 348aa4d8616SChristoph Hellwig static int lo_read_simple(struct loop_device *lo, struct request *rq, 349aa4d8616SChristoph Hellwig loff_t pos) 3501da177e4SLinus Torvalds { 3517988613bSKent Overstreet struct bio_vec bvec; 35230112013SMing Lei struct req_iterator iter; 353aa4d8616SChristoph Hellwig struct iov_iter i; 354aa4d8616SChristoph Hellwig ssize_t len; 3551da177e4SLinus Torvalds 35630112013SMing Lei rq_for_each_segment(bvec, rq, iter) { 357b6207430SChristoph Hellwig iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 35818e9710eSChristoph Hellwig len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 359aa4d8616SChristoph Hellwig if (len < 0) 360aa4d8616SChristoph Hellwig return len; 361306df071SDave Young 362aa4d8616SChristoph Hellwig flush_dcache_page(bvec.bv_page); 363aa4d8616SChristoph Hellwig 364aa4d8616SChristoph Hellwig if (len != bvec.bv_len) { 36530112013SMing Lei struct bio *bio; 36630112013SMing Lei 36730112013SMing Lei __rq_for_each_bio(bio, rq) 368306df071SDave Young zero_fill_bio(bio); 3691da177e4SLinus Torvalds break; 370306df071SDave Young } 371aa4d8616SChristoph Hellwig cond_resched(); 3721da177e4SLinus Torvalds } 373aa4d8616SChristoph Hellwig 374306df071SDave Young return 0; 3751da177e4SLinus Torvalds } 3761da177e4SLinus Torvalds 377aa4d8616SChristoph Hellwig static int lo_read_transfer(struct loop_device *lo, struct request *rq, 378aa4d8616SChristoph Hellwig loff_t pos) 379aa4d8616SChristoph Hellwig { 380aa4d8616SChristoph Hellwig struct bio_vec bvec, b; 381aa4d8616SChristoph Hellwig struct req_iterator iter; 382aa4d8616SChristoph Hellwig struct iov_iter i; 383aa4d8616SChristoph Hellwig struct page *page; 384aa4d8616SChristoph Hellwig ssize_t len; 385aa4d8616SChristoph Hellwig int ret = 0; 386aa4d8616SChristoph Hellwig 387aa4d8616SChristoph Hellwig page = alloc_page(GFP_NOIO); 388aa4d8616SChristoph Hellwig if (unlikely(!page)) 389aa4d8616SChristoph Hellwig return -ENOMEM; 390aa4d8616SChristoph Hellwig 391aa4d8616SChristoph Hellwig rq_for_each_segment(bvec, rq, iter) { 392aa4d8616SChristoph Hellwig loff_t offset = pos; 393aa4d8616SChristoph Hellwig 394aa4d8616SChristoph Hellwig b.bv_page = page; 395aa4d8616SChristoph Hellwig b.bv_offset = 0; 396aa4d8616SChristoph Hellwig b.bv_len = bvec.bv_len; 397aa4d8616SChristoph Hellwig 398b6207430SChristoph Hellwig iov_iter_bvec(&i, READ, &b, 1, b.bv_len); 39918e9710eSChristoph Hellwig len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 400aa4d8616SChristoph Hellwig if (len < 0) { 401aa4d8616SChristoph Hellwig ret = len; 402aa4d8616SChristoph Hellwig goto out_free_page; 403aa4d8616SChristoph Hellwig } 404aa4d8616SChristoph Hellwig 405aa4d8616SChristoph Hellwig ret = lo_do_transfer(lo, READ, page, 0, bvec.bv_page, 406aa4d8616SChristoph Hellwig bvec.bv_offset, len, offset >> 9); 407aa4d8616SChristoph Hellwig if (ret) 408aa4d8616SChristoph Hellwig goto out_free_page; 409aa4d8616SChristoph Hellwig 410aa4d8616SChristoph Hellwig flush_dcache_page(bvec.bv_page); 411aa4d8616SChristoph Hellwig 412aa4d8616SChristoph Hellwig if (len != bvec.bv_len) { 413aa4d8616SChristoph Hellwig struct bio *bio; 414aa4d8616SChristoph Hellwig 415aa4d8616SChristoph Hellwig __rq_for_each_bio(bio, rq) 416aa4d8616SChristoph Hellwig zero_fill_bio(bio); 417aa4d8616SChristoph Hellwig break; 418aa4d8616SChristoph Hellwig } 419aa4d8616SChristoph Hellwig } 420aa4d8616SChristoph Hellwig 421aa4d8616SChristoph Hellwig ret = 0; 422aa4d8616SChristoph Hellwig out_free_page: 423aa4d8616SChristoph Hellwig __free_page(page); 424aa4d8616SChristoph Hellwig return ret; 425aa4d8616SChristoph Hellwig } 426aa4d8616SChristoph Hellwig 427efcfec57SDarrick J. Wong static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 428efcfec57SDarrick J. Wong int mode) 429cf655d95SMing Lei { 430cf655d95SMing Lei /* 431efcfec57SDarrick J. Wong * We use fallocate to manipulate the space mappings used by the image 432efcfec57SDarrick J. Wong * a.k.a. discard/zerorange. However we do not support this if 433efcfec57SDarrick J. Wong * encryption is enabled, because it may give an attacker useful 434efcfec57SDarrick J. Wong * information. 435cf655d95SMing Lei */ 436cf655d95SMing Lei struct file *file = lo->lo_backing_file; 437c52abf56SEvan Green struct request_queue *q = lo->lo_queue; 438cf655d95SMing Lei int ret; 439cf655d95SMing Lei 440efcfec57SDarrick J. Wong mode |= FALLOC_FL_KEEP_SIZE; 441efcfec57SDarrick J. Wong 442c52abf56SEvan Green if (!blk_queue_discard(q)) { 443cf655d95SMing Lei ret = -EOPNOTSUPP; 444cf655d95SMing Lei goto out; 445cf655d95SMing Lei } 446cf655d95SMing Lei 447cf655d95SMing Lei ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 448cf655d95SMing Lei if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 449cf655d95SMing Lei ret = -EIO; 450cf655d95SMing Lei out: 451cf655d95SMing Lei return ret; 452cf655d95SMing Lei } 453cf655d95SMing Lei 454cf655d95SMing Lei static int lo_req_flush(struct loop_device *lo, struct request *rq) 455cf655d95SMing Lei { 456cf655d95SMing Lei struct file *file = lo->lo_backing_file; 457cf655d95SMing Lei int ret = vfs_fsync(file, 0); 458cf655d95SMing Lei if (unlikely(ret && ret != -EINVAL)) 459cf655d95SMing Lei ret = -EIO; 460cf655d95SMing Lei 461cf655d95SMing Lei return ret; 462cf655d95SMing Lei } 463cf655d95SMing Lei 464fe2cb290SChristoph Hellwig static void lo_complete_rq(struct request *rq) 465bc07c10aSMing Lei { 466fe2cb290SChristoph Hellwig struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 467f9de14bcSJens Axboe blk_status_t ret = BLK_STS_OK; 468bc07c10aSMing Lei 469f9de14bcSJens Axboe if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 470f9de14bcSJens Axboe req_op(rq) != REQ_OP_READ) { 471f9de14bcSJens Axboe if (cmd->ret < 0) 4728cd55087SEvan Green ret = errno_to_blk_status(cmd->ret); 473f9de14bcSJens Axboe goto end_io; 474bc07c10aSMing Lei } 475fe2cb290SChristoph Hellwig 476f9de14bcSJens Axboe /* 477f9de14bcSJens Axboe * Short READ - if we got some data, advance our request and 478f9de14bcSJens Axboe * retry it. If we got no data, end the rest with EIO. 479f9de14bcSJens Axboe */ 480f9de14bcSJens Axboe if (cmd->ret) { 481f9de14bcSJens Axboe blk_update_request(rq, BLK_STS_OK, cmd->ret); 482f9de14bcSJens Axboe cmd->ret = 0; 483f9de14bcSJens Axboe blk_mq_requeue_request(rq, true); 484f9de14bcSJens Axboe } else { 485f9de14bcSJens Axboe if (cmd->use_aio) { 486f9de14bcSJens Axboe struct bio *bio = rq->bio; 487f9de14bcSJens Axboe 488f9de14bcSJens Axboe while (bio) { 489f9de14bcSJens Axboe zero_fill_bio(bio); 490f9de14bcSJens Axboe bio = bio->bi_next; 491f9de14bcSJens Axboe } 492f9de14bcSJens Axboe } 493f9de14bcSJens Axboe ret = BLK_STS_IOERR; 494f9de14bcSJens Axboe end_io: 495f9de14bcSJens Axboe blk_mq_end_request(rq, ret); 496f9de14bcSJens Axboe } 497bc07c10aSMing Lei } 498bc07c10aSMing Lei 49992d77332SShaohua Li static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 50092d77332SShaohua Li { 5011894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 5021894e916SJens Axboe 50392d77332SShaohua Li if (!atomic_dec_and_test(&cmd->ref)) 50492d77332SShaohua Li return; 50592d77332SShaohua Li kfree(cmd->bvec); 50692d77332SShaohua Li cmd->bvec = NULL; 5071894e916SJens Axboe blk_mq_complete_request(rq); 50892d77332SShaohua Li } 50992d77332SShaohua Li 510bc07c10aSMing Lei static void lo_rw_aio_complete(struct kiocb *iocb, long ret, long ret2) 511bc07c10aSMing Lei { 512bc07c10aSMing Lei struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 513bc07c10aSMing Lei 514d4478e92SShaohua Li if (cmd->css) 515d4478e92SShaohua Li css_put(cmd->css); 516fe2cb290SChristoph Hellwig cmd->ret = ret; 51792d77332SShaohua Li lo_rw_aio_do_completion(cmd); 518bc07c10aSMing Lei } 519bc07c10aSMing Lei 520bc07c10aSMing Lei static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 521bc07c10aSMing Lei loff_t pos, bool rw) 522bc07c10aSMing Lei { 523bc07c10aSMing Lei struct iov_iter iter; 52486af5952SMing Lei struct req_iterator rq_iter; 525bc07c10aSMing Lei struct bio_vec *bvec; 5261894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 52740326d8aSShaohua Li struct bio *bio = rq->bio; 528bc07c10aSMing Lei struct file *file = lo->lo_backing_file; 52986af5952SMing Lei struct bio_vec tmp; 53040326d8aSShaohua Li unsigned int offset; 53186af5952SMing Lei int nr_bvec = 0; 532bc07c10aSMing Lei int ret; 533bc07c10aSMing Lei 53486af5952SMing Lei rq_for_each_bvec(tmp, rq, rq_iter) 53586af5952SMing Lei nr_bvec++; 536bc07c10aSMing Lei 53786af5952SMing Lei if (rq->bio != rq->biotail) { 53886af5952SMing Lei 53986af5952SMing Lei bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 5406da2ec56SKees Cook GFP_NOIO); 54140326d8aSShaohua Li if (!bvec) 54240326d8aSShaohua Li return -EIO; 54340326d8aSShaohua Li cmd->bvec = bvec; 54440326d8aSShaohua Li 545a7297a6aSMing Lei /* 54640326d8aSShaohua Li * The bios of the request may be started from the middle of 54740326d8aSShaohua Li * the 'bvec' because of bio splitting, so we can't directly 54886af5952SMing Lei * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 54940326d8aSShaohua Li * API will take care of all details for us. 550a7297a6aSMing Lei */ 55186af5952SMing Lei rq_for_each_bvec(tmp, rq, rq_iter) { 55240326d8aSShaohua Li *bvec = tmp; 55340326d8aSShaohua Li bvec++; 55440326d8aSShaohua Li } 55540326d8aSShaohua Li bvec = cmd->bvec; 55640326d8aSShaohua Li offset = 0; 55740326d8aSShaohua Li } else { 55840326d8aSShaohua Li /* 55940326d8aSShaohua Li * Same here, this bio may be started from the middle of the 56040326d8aSShaohua Li * 'bvec' because of bio splitting, so offset from the bvec 56140326d8aSShaohua Li * must be passed to iov iterator 56240326d8aSShaohua Li */ 56340326d8aSShaohua Li offset = bio->bi_iter.bi_bvec_done; 56440326d8aSShaohua Li bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 56540326d8aSShaohua Li } 56692d77332SShaohua Li atomic_set(&cmd->ref, 2); 56740326d8aSShaohua Li 568b6207430SChristoph Hellwig iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 56940326d8aSShaohua Li iter.iov_offset = offset; 570bc07c10aSMing Lei 571bc07c10aSMing Lei cmd->iocb.ki_pos = pos; 572bc07c10aSMing Lei cmd->iocb.ki_filp = file; 573bc07c10aSMing Lei cmd->iocb.ki_complete = lo_rw_aio_complete; 574bc07c10aSMing Lei cmd->iocb.ki_flags = IOCB_DIRECT; 575d9a08a9eSAdam Manzanares cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 576d4478e92SShaohua Li if (cmd->css) 577d4478e92SShaohua Li kthread_associate_blkcg(cmd->css); 578bc07c10aSMing Lei 579bc07c10aSMing Lei if (rw == WRITE) 580bb7462b6SMiklos Szeredi ret = call_write_iter(file, &cmd->iocb, &iter); 581bc07c10aSMing Lei else 582bb7462b6SMiklos Szeredi ret = call_read_iter(file, &cmd->iocb, &iter); 583bc07c10aSMing Lei 58492d77332SShaohua Li lo_rw_aio_do_completion(cmd); 585d4478e92SShaohua Li kthread_associate_blkcg(NULL); 58692d77332SShaohua Li 587bc07c10aSMing Lei if (ret != -EIOCBQUEUED) 588bc07c10aSMing Lei cmd->iocb.ki_complete(&cmd->iocb, ret, 0); 589bc07c10aSMing Lei return 0; 590bc07c10aSMing Lei } 591bc07c10aSMing Lei 592c1c87c2bSChristoph Hellwig static int do_req_filebacked(struct loop_device *lo, struct request *rq) 593bc07c10aSMing Lei { 594bc07c10aSMing Lei struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 595c1c87c2bSChristoph Hellwig loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 596bc07c10aSMing Lei 597bc07c10aSMing Lei /* 598bc07c10aSMing Lei * lo_write_simple and lo_read_simple should have been covered 599bc07c10aSMing Lei * by io submit style function like lo_rw_aio(), one blocker 600bc07c10aSMing Lei * is that lo_read_simple() need to call flush_dcache_page after 601bc07c10aSMing Lei * the page is written from kernel, and it isn't easy to handle 602bc07c10aSMing Lei * this in io submit style function which submits all segments 603bc07c10aSMing Lei * of the req at one time. And direct read IO doesn't need to 604bc07c10aSMing Lei * run flush_dcache_page(). 605bc07c10aSMing Lei */ 606c1c87c2bSChristoph Hellwig switch (req_op(rq)) { 607c1c87c2bSChristoph Hellwig case REQ_OP_FLUSH: 608c1c87c2bSChristoph Hellwig return lo_req_flush(lo, rq); 60919372e27SChristoph Hellwig case REQ_OP_WRITE_ZEROES: 610efcfec57SDarrick J. Wong /* 611efcfec57SDarrick J. Wong * If the caller doesn't want deallocation, call zeroout to 612efcfec57SDarrick J. Wong * write zeroes the range. Otherwise, punch them out. 613efcfec57SDarrick J. Wong */ 614efcfec57SDarrick J. Wong return lo_fallocate(lo, rq, pos, 615efcfec57SDarrick J. Wong (rq->cmd_flags & REQ_NOUNMAP) ? 616efcfec57SDarrick J. Wong FALLOC_FL_ZERO_RANGE : 617efcfec57SDarrick J. Wong FALLOC_FL_PUNCH_HOLE); 618efcfec57SDarrick J. Wong case REQ_OP_DISCARD: 619efcfec57SDarrick J. Wong return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 620c1c87c2bSChristoph Hellwig case REQ_OP_WRITE: 621c1c87c2bSChristoph Hellwig if (lo->transfer) 622c1c87c2bSChristoph Hellwig return lo_write_transfer(lo, rq, pos); 623c1c87c2bSChristoph Hellwig else if (cmd->use_aio) 624c1c87c2bSChristoph Hellwig return lo_rw_aio(lo, cmd, pos, WRITE); 625c1c87c2bSChristoph Hellwig else 626bc07c10aSMing Lei return lo_write_simple(lo, rq, pos); 627c1c87c2bSChristoph Hellwig case REQ_OP_READ: 628c1c87c2bSChristoph Hellwig if (lo->transfer) 629c1c87c2bSChristoph Hellwig return lo_read_transfer(lo, rq, pos); 630c1c87c2bSChristoph Hellwig else if (cmd->use_aio) 631c1c87c2bSChristoph Hellwig return lo_rw_aio(lo, cmd, pos, READ); 632bc07c10aSMing Lei else 633bc07c10aSMing Lei return lo_read_simple(lo, rq, pos); 634c1c87c2bSChristoph Hellwig default: 635c1c87c2bSChristoph Hellwig WARN_ON_ONCE(1); 636c1c87c2bSChristoph Hellwig return -EIO; 637bc07c10aSMing Lei } 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds 6402e5ab5f3SMing Lei static inline void loop_update_dio(struct loop_device *lo) 6412e5ab5f3SMing Lei { 6422e5ab5f3SMing Lei __loop_update_dio(lo, io_is_direct(lo->lo_backing_file) | 6432e5ab5f3SMing Lei lo->use_dio); 6442e5ab5f3SMing Lei } 6452e5ab5f3SMing Lei 64606f0e9e6SMing Lei static void loop_reread_partitions(struct loop_device *lo, 64706f0e9e6SMing Lei struct block_device *bdev) 64806f0e9e6SMing Lei { 64906f0e9e6SMing Lei int rc; 65006f0e9e6SMing Lei 651f0b870dfSChristoph Hellwig mutex_lock(&bdev->bd_mutex); 652f0b870dfSChristoph Hellwig rc = bdev_disk_changed(bdev, false); 653f0b870dfSChristoph Hellwig mutex_unlock(&bdev->bd_mutex); 65406f0e9e6SMing Lei if (rc) 65506f0e9e6SMing Lei pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 65606f0e9e6SMing Lei __func__, lo->lo_number, lo->lo_file_name, rc); 65706f0e9e6SMing Lei } 65806f0e9e6SMing Lei 659d2ac838eSTheodore Ts'o static inline int is_loop_device(struct file *file) 660d2ac838eSTheodore Ts'o { 661d2ac838eSTheodore Ts'o struct inode *i = file->f_mapping->host; 662d2ac838eSTheodore Ts'o 663d2ac838eSTheodore Ts'o return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR; 664d2ac838eSTheodore Ts'o } 665d2ac838eSTheodore Ts'o 666d2ac838eSTheodore Ts'o static int loop_validate_file(struct file *file, struct block_device *bdev) 667d2ac838eSTheodore Ts'o { 668d2ac838eSTheodore Ts'o struct inode *inode = file->f_mapping->host; 669d2ac838eSTheodore Ts'o struct file *f = file; 670d2ac838eSTheodore Ts'o 671d2ac838eSTheodore Ts'o /* Avoid recursion */ 672d2ac838eSTheodore Ts'o while (is_loop_device(f)) { 673d2ac838eSTheodore Ts'o struct loop_device *l; 674d2ac838eSTheodore Ts'o 675d2ac838eSTheodore Ts'o if (f->f_mapping->host->i_bdev == bdev) 676d2ac838eSTheodore Ts'o return -EBADF; 677d2ac838eSTheodore Ts'o 678d2ac838eSTheodore Ts'o l = f->f_mapping->host->i_bdev->bd_disk->private_data; 679f7c8a412SDongli Zhang if (l->lo_state != Lo_bound) { 680d2ac838eSTheodore Ts'o return -EINVAL; 681d2ac838eSTheodore Ts'o } 682d2ac838eSTheodore Ts'o f = l->lo_backing_file; 683d2ac838eSTheodore Ts'o } 684d2ac838eSTheodore Ts'o if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 685d2ac838eSTheodore Ts'o return -EINVAL; 686d2ac838eSTheodore Ts'o return 0; 687d2ac838eSTheodore Ts'o } 688d2ac838eSTheodore Ts'o 6891da177e4SLinus Torvalds /* 6901da177e4SLinus Torvalds * loop_change_fd switched the backing store of a loopback device to 6911da177e4SLinus Torvalds * a new file. This is useful for operating system installers to free up 6921da177e4SLinus Torvalds * the original file and in High Availability environments to switch to 6931da177e4SLinus Torvalds * an alternative location for the content in case of server meltdown. 6941da177e4SLinus Torvalds * This can only work if the loop device is used read-only, and if the 6951da177e4SLinus Torvalds * new backing store is the same size and type as the old backing store. 6961da177e4SLinus Torvalds */ 697bb214884SAl Viro static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 698bb214884SAl Viro unsigned int arg) 6991da177e4SLinus Torvalds { 7001dded9acSJan Kara struct file *file = NULL, *old_file; 7011da177e4SLinus Torvalds int error; 70285b0a54aSJan Kara bool partscan; 7031da177e4SLinus Torvalds 704c28445faSJan Kara error = mutex_lock_killable(&loop_ctl_mutex); 705c3710770SJan Kara if (error) 706c3710770SJan Kara return error; 7071da177e4SLinus Torvalds error = -ENXIO; 7081da177e4SLinus Torvalds if (lo->lo_state != Lo_bound) 7091dded9acSJan Kara goto out_err; 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds /* the loop device has to be read-only */ 7121da177e4SLinus Torvalds error = -EINVAL; 7131da177e4SLinus Torvalds if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 7141dded9acSJan Kara goto out_err; 7151da177e4SLinus Torvalds 7161da177e4SLinus Torvalds error = -EBADF; 7171da177e4SLinus Torvalds file = fget(arg); 7181da177e4SLinus Torvalds if (!file) 7191dded9acSJan Kara goto out_err; 7201da177e4SLinus Torvalds 721d2ac838eSTheodore Ts'o error = loop_validate_file(file, bdev); 722d2ac838eSTheodore Ts'o if (error) 7231dded9acSJan Kara goto out_err; 724d2ac838eSTheodore Ts'o 7251da177e4SLinus Torvalds old_file = lo->lo_backing_file; 7261da177e4SLinus Torvalds 7271da177e4SLinus Torvalds error = -EINVAL; 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds /* size of the new backing store needs to be the same */ 7301da177e4SLinus Torvalds if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 7311dded9acSJan Kara goto out_err; 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds /* and ... switch */ 73443cade80SOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 73543cade80SOmar Sandoval mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 73643cade80SOmar Sandoval lo->lo_backing_file = file; 73743cade80SOmar Sandoval lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 73843cade80SOmar Sandoval mapping_set_gfp_mask(file->f_mapping, 73943cade80SOmar Sandoval lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 74043cade80SOmar Sandoval loop_update_dio(lo); 74143cade80SOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 74285b0a54aSJan Kara partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 743c3710770SJan Kara mutex_unlock(&loop_ctl_mutex); 7441dded9acSJan Kara /* 7451dded9acSJan Kara * We must drop file reference outside of loop_ctl_mutex as dropping 7461dded9acSJan Kara * the file ref can take bd_mutex which creates circular locking 7471dded9acSJan Kara * dependency. 7481dded9acSJan Kara */ 7491dded9acSJan Kara fput(old_file); 75085b0a54aSJan Kara if (partscan) 75185b0a54aSJan Kara loop_reread_partitions(lo, bdev); 7521da177e4SLinus Torvalds return 0; 7531da177e4SLinus Torvalds 7541dded9acSJan Kara out_err: 755c3710770SJan Kara mutex_unlock(&loop_ctl_mutex); 7561dded9acSJan Kara if (file) 7571dded9acSJan Kara fput(file); 7581da177e4SLinus Torvalds return error; 7591da177e4SLinus Torvalds } 7601da177e4SLinus Torvalds 761ee862730SMilan Broz /* loop sysfs attributes */ 762ee862730SMilan Broz 763ee862730SMilan Broz static ssize_t loop_attr_show(struct device *dev, char *page, 764ee862730SMilan Broz ssize_t (*callback)(struct loop_device *, char *)) 765ee862730SMilan Broz { 76634dd82afSKay Sievers struct gendisk *disk = dev_to_disk(dev); 76734dd82afSKay Sievers struct loop_device *lo = disk->private_data; 768ee862730SMilan Broz 76934dd82afSKay Sievers return callback(lo, page); 770ee862730SMilan Broz } 771ee862730SMilan Broz 772ee862730SMilan Broz #define LOOP_ATTR_RO(_name) \ 773ee862730SMilan Broz static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 774ee862730SMilan Broz static ssize_t loop_attr_do_show_##_name(struct device *d, \ 775ee862730SMilan Broz struct device_attribute *attr, char *b) \ 776ee862730SMilan Broz { \ 777ee862730SMilan Broz return loop_attr_show(d, b, loop_attr_##_name##_show); \ 778ee862730SMilan Broz } \ 779ee862730SMilan Broz static struct device_attribute loop_attr_##_name = \ 7805657a819SJoe Perches __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 781ee862730SMilan Broz 782ee862730SMilan Broz static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 783ee862730SMilan Broz { 784ee862730SMilan Broz ssize_t ret; 785ee862730SMilan Broz char *p = NULL; 786ee862730SMilan Broz 78705eb0f25SKay Sievers spin_lock_irq(&lo->lo_lock); 788ee862730SMilan Broz if (lo->lo_backing_file) 7899bf39ab2SMiklos Szeredi p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 79005eb0f25SKay Sievers spin_unlock_irq(&lo->lo_lock); 791ee862730SMilan Broz 792ee862730SMilan Broz if (IS_ERR_OR_NULL(p)) 793ee862730SMilan Broz ret = PTR_ERR(p); 794ee862730SMilan Broz else { 795ee862730SMilan Broz ret = strlen(p); 796ee862730SMilan Broz memmove(buf, p, ret); 797ee862730SMilan Broz buf[ret++] = '\n'; 798ee862730SMilan Broz buf[ret] = 0; 799ee862730SMilan Broz } 800ee862730SMilan Broz 801ee862730SMilan Broz return ret; 802ee862730SMilan Broz } 803ee862730SMilan Broz 804ee862730SMilan Broz static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 805ee862730SMilan Broz { 806ee862730SMilan Broz return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_offset); 807ee862730SMilan Broz } 808ee862730SMilan Broz 809ee862730SMilan Broz static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 810ee862730SMilan Broz { 811ee862730SMilan Broz return sprintf(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 812ee862730SMilan Broz } 813ee862730SMilan Broz 814ee862730SMilan Broz static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 815ee862730SMilan Broz { 816ee862730SMilan Broz int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 817ee862730SMilan Broz 818ee862730SMilan Broz return sprintf(buf, "%s\n", autoclear ? "1" : "0"); 819ee862730SMilan Broz } 820ee862730SMilan Broz 821e03c8dd1SKay Sievers static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 822e03c8dd1SKay Sievers { 823e03c8dd1SKay Sievers int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 824e03c8dd1SKay Sievers 825e03c8dd1SKay Sievers return sprintf(buf, "%s\n", partscan ? "1" : "0"); 826e03c8dd1SKay Sievers } 827e03c8dd1SKay Sievers 8282e5ab5f3SMing Lei static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 8292e5ab5f3SMing Lei { 8302e5ab5f3SMing Lei int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 8312e5ab5f3SMing Lei 8322e5ab5f3SMing Lei return sprintf(buf, "%s\n", dio ? "1" : "0"); 8332e5ab5f3SMing Lei } 8342e5ab5f3SMing Lei 835ee862730SMilan Broz LOOP_ATTR_RO(backing_file); 836ee862730SMilan Broz LOOP_ATTR_RO(offset); 837ee862730SMilan Broz LOOP_ATTR_RO(sizelimit); 838ee862730SMilan Broz LOOP_ATTR_RO(autoclear); 839e03c8dd1SKay Sievers LOOP_ATTR_RO(partscan); 8402e5ab5f3SMing Lei LOOP_ATTR_RO(dio); 841ee862730SMilan Broz 842ee862730SMilan Broz static struct attribute *loop_attrs[] = { 843ee862730SMilan Broz &loop_attr_backing_file.attr, 844ee862730SMilan Broz &loop_attr_offset.attr, 845ee862730SMilan Broz &loop_attr_sizelimit.attr, 846ee862730SMilan Broz &loop_attr_autoclear.attr, 847e03c8dd1SKay Sievers &loop_attr_partscan.attr, 8482e5ab5f3SMing Lei &loop_attr_dio.attr, 849ee862730SMilan Broz NULL, 850ee862730SMilan Broz }; 851ee862730SMilan Broz 852ee862730SMilan Broz static struct attribute_group loop_attribute_group = { 853ee862730SMilan Broz .name = "loop", 854ee862730SMilan Broz .attrs= loop_attrs, 855ee862730SMilan Broz }; 856ee862730SMilan Broz 857d3349b6bSTetsuo Handa static void loop_sysfs_init(struct loop_device *lo) 858ee862730SMilan Broz { 859d3349b6bSTetsuo Handa lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 860ee862730SMilan Broz &loop_attribute_group); 861ee862730SMilan Broz } 862ee862730SMilan Broz 863ee862730SMilan Broz static void loop_sysfs_exit(struct loop_device *lo) 864ee862730SMilan Broz { 865d3349b6bSTetsuo Handa if (lo->sysfs_inited) 866ee862730SMilan Broz sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 867ee862730SMilan Broz &loop_attribute_group); 868ee862730SMilan Broz } 869ee862730SMilan Broz 870dfaa2ef6SLukas Czerner static void loop_config_discard(struct loop_device *lo) 871dfaa2ef6SLukas Czerner { 872dfaa2ef6SLukas Czerner struct file *file = lo->lo_backing_file; 873dfaa2ef6SLukas Czerner struct inode *inode = file->f_mapping->host; 874dfaa2ef6SLukas Czerner struct request_queue *q = lo->lo_queue; 875dfaa2ef6SLukas Czerner 876dfaa2ef6SLukas Czerner /* 877c52abf56SEvan Green * If the backing device is a block device, mirror its zeroing 878c52abf56SEvan Green * capability. Set the discard sectors to the block device's zeroing 879c52abf56SEvan Green * capabilities because loop discards result in blkdev_issue_zeroout(), 880c52abf56SEvan Green * not blkdev_issue_discard(). This maintains consistent behavior with 881c52abf56SEvan Green * file-backed loop devices: discarded regions read back as zero. 882c52abf56SEvan Green */ 883c52abf56SEvan Green if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) { 884c52abf56SEvan Green struct request_queue *backingq; 885c52abf56SEvan Green 886c52abf56SEvan Green backingq = bdev_get_queue(inode->i_bdev); 887c52abf56SEvan Green blk_queue_max_discard_sectors(q, 888c52abf56SEvan Green backingq->limits.max_write_zeroes_sectors); 889c52abf56SEvan Green 890c52abf56SEvan Green blk_queue_max_write_zeroes_sectors(q, 891c52abf56SEvan Green backingq->limits.max_write_zeroes_sectors); 892c52abf56SEvan Green 893c52abf56SEvan Green /* 894dfaa2ef6SLukas Czerner * We use punch hole to reclaim the free space used by the 89512a64d2fSOlaf Hering * image a.k.a. discard. However we do not support discard if 896dfaa2ef6SLukas Czerner * encryption is enabled, because it may give an attacker 897dfaa2ef6SLukas Czerner * useful information. 898dfaa2ef6SLukas Czerner */ 899c52abf56SEvan Green } else if (!file->f_op->fallocate || lo->lo_encrypt_key_size) { 900dfaa2ef6SLukas Czerner q->limits.discard_granularity = 0; 901dfaa2ef6SLukas Czerner q->limits.discard_alignment = 0; 9022bb4cd5cSJens Axboe blk_queue_max_discard_sectors(q, 0); 90319372e27SChristoph Hellwig blk_queue_max_write_zeroes_sectors(q, 0); 904dfaa2ef6SLukas Czerner 905c52abf56SEvan Green } else { 906dfaa2ef6SLukas Czerner q->limits.discard_granularity = inode->i_sb->s_blocksize; 907dfaf3c03SLukas Czerner q->limits.discard_alignment = 0; 908f2c6df7dSHannes Reinecke 9091e6ec9eaSOmar Sandoval blk_queue_max_discard_sectors(q, UINT_MAX >> 9); 9101e6ec9eaSOmar Sandoval blk_queue_max_write_zeroes_sectors(q, UINT_MAX >> 9); 911c52abf56SEvan Green } 912c52abf56SEvan Green 913c52abf56SEvan Green if (q->limits.max_write_zeroes_sectors) 9148b904b5bSBart Van Assche blk_queue_flag_set(QUEUE_FLAG_DISCARD, q); 915c52abf56SEvan Green else 916c52abf56SEvan Green blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q); 917dfaa2ef6SLukas Czerner } 918dfaa2ef6SLukas Czerner 919e03a3d7aSMing Lei static void loop_unprepare_queue(struct loop_device *lo) 920e03a3d7aSMing Lei { 9213989144fSPetr Mladek kthread_flush_worker(&lo->worker); 922e03a3d7aSMing Lei kthread_stop(lo->worker_task); 923e03a3d7aSMing Lei } 924e03a3d7aSMing Lei 925b2ee7d46SNeilBrown static int loop_kthread_worker_fn(void *worker_ptr) 926b2ee7d46SNeilBrown { 927d0a255e7SMikulas Patocka current->flags |= PF_LESS_THROTTLE | PF_MEMALLOC_NOIO; 928b2ee7d46SNeilBrown return kthread_worker_fn(worker_ptr); 929b2ee7d46SNeilBrown } 930b2ee7d46SNeilBrown 931e03a3d7aSMing Lei static int loop_prepare_queue(struct loop_device *lo) 932e03a3d7aSMing Lei { 9333989144fSPetr Mladek kthread_init_worker(&lo->worker); 934b2ee7d46SNeilBrown lo->worker_task = kthread_run(loop_kthread_worker_fn, 935e03a3d7aSMing Lei &lo->worker, "loop%d", lo->lo_number); 936e03a3d7aSMing Lei if (IS_ERR(lo->worker_task)) 937e03a3d7aSMing Lei return -ENOMEM; 938e03a3d7aSMing Lei set_user_nice(lo->worker_task, MIN_NICE); 939e03a3d7aSMing Lei return 0; 940e03a3d7aSMing Lei } 941e03a3d7aSMing Lei 94256a85fd8SHolger Hoffstätte static void loop_update_rotational(struct loop_device *lo) 94356a85fd8SHolger Hoffstätte { 94456a85fd8SHolger Hoffstätte struct file *file = lo->lo_backing_file; 94556a85fd8SHolger Hoffstätte struct inode *file_inode = file->f_mapping->host; 94656a85fd8SHolger Hoffstätte struct block_device *file_bdev = file_inode->i_sb->s_bdev; 94756a85fd8SHolger Hoffstätte struct request_queue *q = lo->lo_queue; 94856a85fd8SHolger Hoffstätte bool nonrot = true; 94956a85fd8SHolger Hoffstätte 95056a85fd8SHolger Hoffstätte /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 95156a85fd8SHolger Hoffstätte if (file_bdev) 95256a85fd8SHolger Hoffstätte nonrot = blk_queue_nonrot(bdev_get_queue(file_bdev)); 95356a85fd8SHolger Hoffstätte 95456a85fd8SHolger Hoffstätte if (nonrot) 95556a85fd8SHolger Hoffstätte blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 95656a85fd8SHolger Hoffstätte else 95756a85fd8SHolger Hoffstätte blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 95856a85fd8SHolger Hoffstätte } 95956a85fd8SHolger Hoffstätte 960bb214884SAl Viro static int loop_set_fd(struct loop_device *lo, fmode_t mode, 9611da177e4SLinus Torvalds struct block_device *bdev, unsigned int arg) 9621da177e4SLinus Torvalds { 963d2ac838eSTheodore Ts'o struct file *file; 9641da177e4SLinus Torvalds struct inode *inode; 9651da177e4SLinus Torvalds struct address_space *mapping; 96689e524c0SJan Kara struct block_device *claimed_bdev = NULL; 9671da177e4SLinus Torvalds int lo_flags = 0; 9681da177e4SLinus Torvalds int error; 9691da177e4SLinus Torvalds loff_t size; 97085b0a54aSJan Kara bool partscan; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds /* This is safe, since we have a reference from open(). */ 9731da177e4SLinus Torvalds __module_get(THIS_MODULE); 9741da177e4SLinus Torvalds 9751da177e4SLinus Torvalds error = -EBADF; 9761da177e4SLinus Torvalds file = fget(arg); 9771da177e4SLinus Torvalds if (!file) 9781da177e4SLinus Torvalds goto out; 9791da177e4SLinus Torvalds 98033ec3e53SJan Kara /* 98133ec3e53SJan Kara * If we don't hold exclusive handle for the device, upgrade to it 98233ec3e53SJan Kara * here to avoid changing device under exclusive owner. 98333ec3e53SJan Kara */ 98433ec3e53SJan Kara if (!(mode & FMODE_EXCL)) { 98589e524c0SJan Kara claimed_bdev = bd_start_claiming(bdev, loop_set_fd); 98689e524c0SJan Kara if (IS_ERR(claimed_bdev)) { 98789e524c0SJan Kara error = PTR_ERR(claimed_bdev); 988757ecf40SJan Kara goto out_putf; 98933ec3e53SJan Kara } 99089e524c0SJan Kara } 99133ec3e53SJan Kara 99233ec3e53SJan Kara error = mutex_lock_killable(&loop_ctl_mutex); 99333ec3e53SJan Kara if (error) 99433ec3e53SJan Kara goto out_bdev; 995757ecf40SJan Kara 9961da177e4SLinus Torvalds error = -EBUSY; 9971da177e4SLinus Torvalds if (lo->lo_state != Lo_unbound) 998757ecf40SJan Kara goto out_unlock; 9991da177e4SLinus Torvalds 1000d2ac838eSTheodore Ts'o error = loop_validate_file(file, bdev); 1001d2ac838eSTheodore Ts'o if (error) 1002757ecf40SJan Kara goto out_unlock; 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds mapping = file->f_mapping; 10051da177e4SLinus Torvalds inode = mapping->host; 10061da177e4SLinus Torvalds 1007456be148SChristoph Hellwig if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1008283e7e5dSAl Viro !file->f_op->write_iter) 10091da177e4SLinus Torvalds lo_flags |= LO_FLAGS_READ_ONLY; 10101da177e4SLinus Torvalds 1011456be148SChristoph Hellwig size = get_loop_size(lo, file); 1012083a6a50SMartijn Coenen 1013e03a3d7aSMing Lei error = loop_prepare_queue(lo); 1014e03a3d7aSMing Lei if (error) 1015757ecf40SJan Kara goto out_unlock; 10161da177e4SLinus Torvalds 1017456be148SChristoph Hellwig error = 0; 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds set_device_ro(bdev, (lo_flags & LO_FLAGS_READ_ONLY) != 0); 10201da177e4SLinus Torvalds 10212e5ab5f3SMing Lei lo->use_dio = false; 10221da177e4SLinus Torvalds lo->lo_device = bdev; 10231da177e4SLinus Torvalds lo->lo_flags = lo_flags; 10241da177e4SLinus Torvalds lo->lo_backing_file = file; 1025aa4d8616SChristoph Hellwig lo->transfer = NULL; 10261da177e4SLinus Torvalds lo->ioctl = NULL; 10271da177e4SLinus Torvalds lo->lo_sizelimit = 0; 10281da177e4SLinus Torvalds lo->old_gfp_mask = mapping_gfp_mask(mapping); 10291da177e4SLinus Torvalds mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 10301da177e4SLinus Torvalds 103168db1961SNikanth Karthikesan if (!(lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 103221d0727fSJens Axboe blk_queue_write_cache(lo->lo_queue, true, false); 103368db1961SNikanth Karthikesan 103485560117SMartijn Coenen if (io_is_direct(lo->lo_backing_file) && inode->i_sb->s_bdev) { 103585560117SMartijn Coenen /* In case of direct I/O, match underlying block size */ 103685560117SMartijn Coenen unsigned short bsize = bdev_logical_block_size( 103785560117SMartijn Coenen inode->i_sb->s_bdev); 103885560117SMartijn Coenen 103985560117SMartijn Coenen blk_queue_logical_block_size(lo->lo_queue, bsize); 104085560117SMartijn Coenen blk_queue_physical_block_size(lo->lo_queue, bsize); 104185560117SMartijn Coenen blk_queue_io_min(lo->lo_queue, bsize); 104285560117SMartijn Coenen } 104385560117SMartijn Coenen 104456a85fd8SHolger Hoffstätte loop_update_rotational(lo); 10452e5ab5f3SMing Lei loop_update_dio(lo); 1046ee862730SMilan Broz loop_sysfs_init(lo); 10475795b6f5SMartijn Coenen loop_set_size(lo, size); 10481da177e4SLinus Torvalds 10498a0740c4SOmar Sandoval set_blocksize(bdev, S_ISBLK(inode->i_mode) ? 10508a0740c4SOmar Sandoval block_size(inode->i_bdev) : PAGE_SIZE); 10511da177e4SLinus Torvalds 10526c997918SSerge E. Hallyn lo->lo_state = Lo_bound; 1053e03c8dd1SKay Sievers if (part_shift) 1054e03c8dd1SKay Sievers lo->lo_flags |= LO_FLAGS_PARTSCAN; 105585b0a54aSJan Kara partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1056c1681bf8SAnatol Pomozov 1057c1681bf8SAnatol Pomozov /* Grab the block_device to prevent its destruction after we 1058a2505b79SJan Kara * put /dev/loopXX inode. Later in __loop_clr_fd() we bdput(bdev). 1059c1681bf8SAnatol Pomozov */ 1060c1681bf8SAnatol Pomozov bdgrab(bdev); 1061757ecf40SJan Kara mutex_unlock(&loop_ctl_mutex); 106285b0a54aSJan Kara if (partscan) 106385b0a54aSJan Kara loop_reread_partitions(lo, bdev); 106489e524c0SJan Kara if (claimed_bdev) 106589e524c0SJan Kara bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); 10661da177e4SLinus Torvalds return 0; 10671da177e4SLinus Torvalds 1068757ecf40SJan Kara out_unlock: 1069757ecf40SJan Kara mutex_unlock(&loop_ctl_mutex); 107033ec3e53SJan Kara out_bdev: 107189e524c0SJan Kara if (claimed_bdev) 107289e524c0SJan Kara bd_abort_claiming(bdev, claimed_bdev, loop_set_fd); 10731da177e4SLinus Torvalds out_putf: 10741da177e4SLinus Torvalds fput(file); 10751da177e4SLinus Torvalds out: 10761da177e4SLinus Torvalds /* This is safe: open() is still holding a reference. */ 10771da177e4SLinus Torvalds module_put(THIS_MODULE); 10781da177e4SLinus Torvalds return error; 10791da177e4SLinus Torvalds } 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds static int 10821da177e4SLinus Torvalds loop_release_xfer(struct loop_device *lo) 10831da177e4SLinus Torvalds { 10841da177e4SLinus Torvalds int err = 0; 10851da177e4SLinus Torvalds struct loop_func_table *xfer = lo->lo_encryption; 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds if (xfer) { 10881da177e4SLinus Torvalds if (xfer->release) 10891da177e4SLinus Torvalds err = xfer->release(lo); 10901da177e4SLinus Torvalds lo->transfer = NULL; 10911da177e4SLinus Torvalds lo->lo_encryption = NULL; 10921da177e4SLinus Torvalds module_put(xfer->owner); 10931da177e4SLinus Torvalds } 10941da177e4SLinus Torvalds return err; 10951da177e4SLinus Torvalds } 10961da177e4SLinus Torvalds 10971da177e4SLinus Torvalds static int 10981da177e4SLinus Torvalds loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer, 10991da177e4SLinus Torvalds const struct loop_info64 *i) 11001da177e4SLinus Torvalds { 11011da177e4SLinus Torvalds int err = 0; 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds if (xfer) { 11041da177e4SLinus Torvalds struct module *owner = xfer->owner; 11051da177e4SLinus Torvalds 11061da177e4SLinus Torvalds if (!try_module_get(owner)) 11071da177e4SLinus Torvalds return -EINVAL; 11081da177e4SLinus Torvalds if (xfer->init) 11091da177e4SLinus Torvalds err = xfer->init(lo, i); 11101da177e4SLinus Torvalds if (err) 11111da177e4SLinus Torvalds module_put(owner); 11121da177e4SLinus Torvalds else 11131da177e4SLinus Torvalds lo->lo_encryption = xfer; 11141da177e4SLinus Torvalds } 11151da177e4SLinus Torvalds return err; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds 11180da03cabSJan Kara static int __loop_clr_fd(struct loop_device *lo, bool release) 11191da177e4SLinus Torvalds { 11207ccd0791SJan Kara struct file *filp = NULL; 1121b4e3ca1aSAl Viro gfp_t gfp = lo->old_gfp_mask; 11224c823cc3SAyan George struct block_device *bdev = lo->lo_device; 11237ccd0791SJan Kara int err = 0; 11240da03cabSJan Kara bool partscan = false; 11250da03cabSJan Kara int lo_number; 11261da177e4SLinus Torvalds 11277ccd0791SJan Kara mutex_lock(&loop_ctl_mutex); 11287ccd0791SJan Kara if (WARN_ON_ONCE(lo->lo_state != Lo_rundown)) { 11297ccd0791SJan Kara err = -ENXIO; 11307ccd0791SJan Kara goto out_unlock; 11317ccd0791SJan Kara } 11321da177e4SLinus Torvalds 11337ccd0791SJan Kara filp = lo->lo_backing_file; 11347ccd0791SJan Kara if (filp == NULL) { 11357ccd0791SJan Kara err = -EINVAL; 11367ccd0791SJan Kara goto out_unlock; 11377ccd0791SJan Kara } 11381da177e4SLinus Torvalds 1139f8933667SMing Lei /* freeze request queue during the transition */ 1140f8933667SMing Lei blk_mq_freeze_queue(lo->lo_queue); 1141f8933667SMing Lei 11421da177e4SLinus Torvalds spin_lock_irq(&lo->lo_lock); 11431da177e4SLinus Torvalds lo->lo_backing_file = NULL; 114405eb0f25SKay Sievers spin_unlock_irq(&lo->lo_lock); 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds loop_release_xfer(lo); 11471da177e4SLinus Torvalds lo->transfer = NULL; 11481da177e4SLinus Torvalds lo->ioctl = NULL; 11491da177e4SLinus Torvalds lo->lo_device = NULL; 11501da177e4SLinus Torvalds lo->lo_encryption = NULL; 11511da177e4SLinus Torvalds lo->lo_offset = 0; 11521da177e4SLinus Torvalds lo->lo_sizelimit = 0; 11531da177e4SLinus Torvalds lo->lo_encrypt_key_size = 0; 11541da177e4SLinus Torvalds memset(lo->lo_encrypt_key, 0, LO_KEY_SIZE); 11551da177e4SLinus Torvalds memset(lo->lo_crypt_name, 0, LO_NAME_SIZE); 11561da177e4SLinus Torvalds memset(lo->lo_file_name, 0, LO_NAME_SIZE); 115789e4fdecSOmar Sandoval blk_queue_logical_block_size(lo->lo_queue, 512); 1158bf093753SOmar Sandoval blk_queue_physical_block_size(lo->lo_queue, 512); 1159bf093753SOmar Sandoval blk_queue_io_min(lo->lo_queue, 512); 1160c1681bf8SAnatol Pomozov if (bdev) { 1161c1681bf8SAnatol Pomozov bdput(bdev); 1162f98393a6SPeter Zijlstra invalidate_bdev(bdev); 1163eedffa28SJeff Layton bdev->bd_inode->i_mapping->wb_err = 0; 1164c1681bf8SAnatol Pomozov } 116573285082SKen Chen set_capacity(lo->lo_disk, 0); 116651a0bb0cSMilan Broz loop_sysfs_exit(lo); 1167c3473c63SDavid Zeuthen if (bdev) { 11681da177e4SLinus Torvalds bd_set_size(bdev, 0); 1169c3473c63SDavid Zeuthen /* let user-space know about this change */ 1170c3473c63SDavid Zeuthen kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE); 1171c3473c63SDavid Zeuthen } 11721da177e4SLinus Torvalds mapping_set_gfp_mask(filp->f_mapping, gfp); 11731da177e4SLinus Torvalds /* This is safe: open() is still holding a reference. */ 11741da177e4SLinus Torvalds module_put(THIS_MODULE); 1175f8933667SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 1176f8933667SMing Lei 11770da03cabSJan Kara partscan = lo->lo_flags & LO_FLAGS_PARTSCAN && bdev; 11780da03cabSJan Kara lo_number = lo->lo_number; 11790da03cabSJan Kara loop_unprepare_queue(lo); 11800da03cabSJan Kara out_unlock: 11810da03cabSJan Kara mutex_unlock(&loop_ctl_mutex); 11820da03cabSJan Kara if (partscan) { 1183d57f3374SJan Kara /* 1184d57f3374SJan Kara * bd_mutex has been held already in release path, so don't 1185d57f3374SJan Kara * acquire it if this function is called in such case. 1186d57f3374SJan Kara * 1187d57f3374SJan Kara * If the reread partition isn't from release path, lo_refcnt 1188d57f3374SJan Kara * must be at least one and it can only become zero when the 1189d57f3374SJan Kara * current holder is released. 1190d57f3374SJan Kara */ 1191f0b870dfSChristoph Hellwig if (!release) 1192f0b870dfSChristoph Hellwig mutex_lock(&bdev->bd_mutex); 1193f0b870dfSChristoph Hellwig err = bdev_disk_changed(bdev, false); 1194f0b870dfSChristoph Hellwig if (!release) 1195f0b870dfSChristoph Hellwig mutex_unlock(&bdev->bd_mutex); 119640853d6fSDongli Zhang if (err) 1197d57f3374SJan Kara pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 11980da03cabSJan Kara __func__, lo_number, err); 1199d57f3374SJan Kara /* Device is gone, no point in returning error */ 1200d57f3374SJan Kara err = 0; 1201d57f3374SJan Kara } 1202758a58d0SDongli Zhang 1203758a58d0SDongli Zhang /* 1204758a58d0SDongli Zhang * lo->lo_state is set to Lo_unbound here after above partscan has 1205758a58d0SDongli Zhang * finished. 1206758a58d0SDongli Zhang * 1207758a58d0SDongli Zhang * There cannot be anybody else entering __loop_clr_fd() as 1208758a58d0SDongli Zhang * lo->lo_backing_file is already cleared and Lo_rundown state 1209758a58d0SDongli Zhang * protects us from all the other places trying to change the 'lo' 1210758a58d0SDongli Zhang * device. 1211758a58d0SDongli Zhang */ 1212758a58d0SDongli Zhang mutex_lock(&loop_ctl_mutex); 1213758a58d0SDongli Zhang lo->lo_flags = 0; 1214758a58d0SDongli Zhang if (!part_shift) 1215758a58d0SDongli Zhang lo->lo_disk->flags |= GENHD_FL_NO_PART_SCAN; 1216758a58d0SDongli Zhang lo->lo_state = Lo_unbound; 1217758a58d0SDongli Zhang mutex_unlock(&loop_ctl_mutex); 1218758a58d0SDongli Zhang 1219f028f3b2SNikanth Karthikesan /* 1220310ca162STetsuo Handa * Need not hold loop_ctl_mutex to fput backing file. 1221310ca162STetsuo Handa * Calling fput holding loop_ctl_mutex triggers a circular 1222f028f3b2SNikanth Karthikesan * lock dependency possibility warning as fput can take 1223310ca162STetsuo Handa * bd_mutex which is usually taken before loop_ctl_mutex. 1224f028f3b2SNikanth Karthikesan */ 12257ccd0791SJan Kara if (filp) 1226f028f3b2SNikanth Karthikesan fput(filp); 12277ccd0791SJan Kara return err; 12281da177e4SLinus Torvalds } 12291da177e4SLinus Torvalds 1230a2505b79SJan Kara static int loop_clr_fd(struct loop_device *lo) 1231a2505b79SJan Kara { 12327ccd0791SJan Kara int err; 12337ccd0791SJan Kara 1234c28445faSJan Kara err = mutex_lock_killable(&loop_ctl_mutex); 12357ccd0791SJan Kara if (err) 12367ccd0791SJan Kara return err; 12377ccd0791SJan Kara if (lo->lo_state != Lo_bound) { 12387ccd0791SJan Kara mutex_unlock(&loop_ctl_mutex); 1239a2505b79SJan Kara return -ENXIO; 12407ccd0791SJan Kara } 1241a2505b79SJan Kara /* 1242a2505b79SJan Kara * If we've explicitly asked to tear down the loop device, 1243a2505b79SJan Kara * and it has an elevated reference count, set it for auto-teardown when 1244a2505b79SJan Kara * the last reference goes away. This stops $!~#$@ udev from 1245a2505b79SJan Kara * preventing teardown because it decided that it needs to run blkid on 1246a2505b79SJan Kara * the loopback device whenever they appear. xfstests is notorious for 1247a2505b79SJan Kara * failing tests because blkid via udev races with a losetup 1248a2505b79SJan Kara * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1249a2505b79SJan Kara * command to fail with EBUSY. 1250a2505b79SJan Kara */ 1251a2505b79SJan Kara if (atomic_read(&lo->lo_refcnt) > 1) { 1252a2505b79SJan Kara lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 1253a2505b79SJan Kara mutex_unlock(&loop_ctl_mutex); 1254a2505b79SJan Kara return 0; 1255a2505b79SJan Kara } 1256a2505b79SJan Kara lo->lo_state = Lo_rundown; 12577ccd0791SJan Kara mutex_unlock(&loop_ctl_mutex); 1258a2505b79SJan Kara 12590da03cabSJan Kara return __loop_clr_fd(lo, false); 1260a2505b79SJan Kara } 1261a2505b79SJan Kara 12621da177e4SLinus Torvalds static int 12631da177e4SLinus Torvalds loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 12641da177e4SLinus Torvalds { 12651da177e4SLinus Torvalds int err; 12661da177e4SLinus Torvalds struct loop_func_table *xfer; 1267e4849737SEric W. Biederman kuid_t uid = current_uid(); 126885b0a54aSJan Kara struct block_device *bdev; 126985b0a54aSJan Kara bool partscan = false; 1270*b0bd158dSMartijn Coenen bool size_changed = false; 12711da177e4SLinus Torvalds 1272c28445faSJan Kara err = mutex_lock_killable(&loop_ctl_mutex); 1273550df5fdSJan Kara if (err) 1274550df5fdSJan Kara return err; 1275b0fafa81SDavid Howells if (lo->lo_encrypt_key_size && 1276e4849737SEric W. Biederman !uid_eq(lo->lo_key_owner, uid) && 1277550df5fdSJan Kara !capable(CAP_SYS_ADMIN)) { 1278550df5fdSJan Kara err = -EPERM; 1279550df5fdSJan Kara goto out_unlock; 1280550df5fdSJan Kara } 1281550df5fdSJan Kara if (lo->lo_state != Lo_bound) { 1282550df5fdSJan Kara err = -ENXIO; 1283550df5fdSJan Kara goto out_unlock; 1284550df5fdSJan Kara } 1285550df5fdSJan Kara if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) { 1286550df5fdSJan Kara err = -EINVAL; 1287550df5fdSJan Kara goto out_unlock; 1288550df5fdSJan Kara } 12891da177e4SLinus Torvalds 12905db470e2SJaegeuk Kim if (lo->lo_offset != info->lo_offset || 12915db470e2SJaegeuk Kim lo->lo_sizelimit != info->lo_sizelimit) { 1292*b0bd158dSMartijn Coenen size_changed = true; 12935db470e2SJaegeuk Kim sync_blockdev(lo->lo_device); 12945db470e2SJaegeuk Kim kill_bdev(lo->lo_device); 12955db470e2SJaegeuk Kim } 12965db470e2SJaegeuk Kim 1297ecdd0959SMing Lei /* I/O need to be drained during transfer transition */ 1298ecdd0959SMing Lei blk_mq_freeze_queue(lo->lo_queue); 1299ecdd0959SMing Lei 1300*b0bd158dSMartijn Coenen if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1301*b0bd158dSMartijn Coenen /* If any pages were dirtied after kill_bdev(), try again */ 1302*b0bd158dSMartijn Coenen err = -EAGAIN; 1303*b0bd158dSMartijn Coenen pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 1304*b0bd158dSMartijn Coenen __func__, lo->lo_number, lo->lo_file_name, 1305*b0bd158dSMartijn Coenen lo->lo_device->bd_inode->i_mapping->nrpages); 1306*b0bd158dSMartijn Coenen goto out_unfreeze; 1307*b0bd158dSMartijn Coenen } 1308*b0bd158dSMartijn Coenen 13091da177e4SLinus Torvalds err = loop_release_xfer(lo); 13101da177e4SLinus Torvalds if (err) 1311550df5fdSJan Kara goto out_unfreeze; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds if (info->lo_encrypt_type) { 13141da177e4SLinus Torvalds unsigned int type = info->lo_encrypt_type; 13151da177e4SLinus Torvalds 13161e047eaaSTetsuo Handa if (type >= MAX_LO_CRYPT) { 13171e047eaaSTetsuo Handa err = -EINVAL; 1318550df5fdSJan Kara goto out_unfreeze; 13191e047eaaSTetsuo Handa } 13201da177e4SLinus Torvalds xfer = xfer_funcs[type]; 13211e047eaaSTetsuo Handa if (xfer == NULL) { 13221e047eaaSTetsuo Handa err = -EINVAL; 1323550df5fdSJan Kara goto out_unfreeze; 13241e047eaaSTetsuo Handa } 13251da177e4SLinus Torvalds } else 13261da177e4SLinus Torvalds xfer = NULL; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds err = loop_init_xfer(lo, xfer, info); 13291da177e4SLinus Torvalds if (err) 1330550df5fdSJan Kara goto out_unfreeze; 13311da177e4SLinus Torvalds 1332*b0bd158dSMartijn Coenen lo->lo_offset = info->lo_offset; 1333*b0bd158dSMartijn Coenen lo->lo_sizelimit = info->lo_sizelimit; 13341da177e4SLinus Torvalds memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 13351da177e4SLinus Torvalds memcpy(lo->lo_crypt_name, info->lo_crypt_name, LO_NAME_SIZE); 13361da177e4SLinus Torvalds lo->lo_file_name[LO_NAME_SIZE-1] = 0; 13371da177e4SLinus Torvalds lo->lo_crypt_name[LO_NAME_SIZE-1] = 0; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds if (!xfer) 13401da177e4SLinus Torvalds xfer = &none_funcs; 13411da177e4SLinus Torvalds lo->transfer = xfer->transfer; 13421da177e4SLinus Torvalds lo->ioctl = xfer->ioctl; 13431da177e4SLinus Torvalds 134496c58655SDavid Woodhouse if ((lo->lo_flags & LO_FLAGS_AUTOCLEAR) != 134596c58655SDavid Woodhouse (info->lo_flags & LO_FLAGS_AUTOCLEAR)) 134696c58655SDavid Woodhouse lo->lo_flags ^= LO_FLAGS_AUTOCLEAR; 134796c58655SDavid Woodhouse 13481da177e4SLinus Torvalds lo->lo_encrypt_key_size = info->lo_encrypt_key_size; 13491da177e4SLinus Torvalds lo->lo_init[0] = info->lo_init[0]; 13501da177e4SLinus Torvalds lo->lo_init[1] = info->lo_init[1]; 13511da177e4SLinus Torvalds if (info->lo_encrypt_key_size) { 13521da177e4SLinus Torvalds memcpy(lo->lo_encrypt_key, info->lo_encrypt_key, 13531da177e4SLinus Torvalds info->lo_encrypt_key_size); 1354b0fafa81SDavid Howells lo->lo_key_owner = uid; 13551da177e4SLinus Torvalds } 13561da177e4SLinus Torvalds 1357*b0bd158dSMartijn Coenen if (size_changed) { 1358*b0bd158dSMartijn Coenen loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1359*b0bd158dSMartijn Coenen lo->lo_backing_file); 1360*b0bd158dSMartijn Coenen loop_set_size(lo, new_size); 1361*b0bd158dSMartijn Coenen } 1362*b0bd158dSMartijn Coenen 13637c5014b0SMartijn Coenen loop_config_discard(lo); 13647c5014b0SMartijn Coenen 13652e5ab5f3SMing Lei /* update dio if lo_offset or transfer is changed */ 13662e5ab5f3SMing Lei __loop_update_dio(lo, lo->use_dio); 13672e5ab5f3SMing Lei 1368550df5fdSJan Kara out_unfreeze: 1369ecdd0959SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 1370e02898b4SOmar Sandoval 1371e02898b4SOmar Sandoval if (!err && (info->lo_flags & LO_FLAGS_PARTSCAN) && 1372e02898b4SOmar Sandoval !(lo->lo_flags & LO_FLAGS_PARTSCAN)) { 1373e02898b4SOmar Sandoval lo->lo_flags |= LO_FLAGS_PARTSCAN; 1374e02898b4SOmar Sandoval lo->lo_disk->flags &= ~GENHD_FL_NO_PART_SCAN; 137585b0a54aSJan Kara bdev = lo->lo_device; 137685b0a54aSJan Kara partscan = true; 1377e02898b4SOmar Sandoval } 1378550df5fdSJan Kara out_unlock: 1379550df5fdSJan Kara mutex_unlock(&loop_ctl_mutex); 138085b0a54aSJan Kara if (partscan) 138185b0a54aSJan Kara loop_reread_partitions(lo, bdev); 1382e02898b4SOmar Sandoval 1383ecdd0959SMing Lei return err; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds static int 13871da177e4SLinus Torvalds loop_get_status(struct loop_device *lo, struct loop_info64 *info) 13881da177e4SLinus Torvalds { 1389b1ab5fa3STetsuo Handa struct path path; 13901da177e4SLinus Torvalds struct kstat stat; 13912d1d4c1eSOmar Sandoval int ret; 13921da177e4SLinus Torvalds 1393c28445faSJan Kara ret = mutex_lock_killable(&loop_ctl_mutex); 13944a5ce9baSJan Kara if (ret) 13954a5ce9baSJan Kara return ret; 13962d1d4c1eSOmar Sandoval if (lo->lo_state != Lo_bound) { 1397310ca162STetsuo Handa mutex_unlock(&loop_ctl_mutex); 13981da177e4SLinus Torvalds return -ENXIO; 13992d1d4c1eSOmar Sandoval } 14002d1d4c1eSOmar Sandoval 14011da177e4SLinus Torvalds memset(info, 0, sizeof(*info)); 14021da177e4SLinus Torvalds info->lo_number = lo->lo_number; 14031da177e4SLinus Torvalds info->lo_offset = lo->lo_offset; 14041da177e4SLinus Torvalds info->lo_sizelimit = lo->lo_sizelimit; 14051da177e4SLinus Torvalds info->lo_flags = lo->lo_flags; 14061da177e4SLinus Torvalds memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 14071da177e4SLinus Torvalds memcpy(info->lo_crypt_name, lo->lo_crypt_name, LO_NAME_SIZE); 14081da177e4SLinus Torvalds info->lo_encrypt_type = 14091da177e4SLinus Torvalds lo->lo_encryption ? lo->lo_encryption->number : 0; 14101da177e4SLinus Torvalds if (lo->lo_encrypt_key_size && capable(CAP_SYS_ADMIN)) { 14111da177e4SLinus Torvalds info->lo_encrypt_key_size = lo->lo_encrypt_key_size; 14121da177e4SLinus Torvalds memcpy(info->lo_encrypt_key, lo->lo_encrypt_key, 14131da177e4SLinus Torvalds lo->lo_encrypt_key_size); 14141da177e4SLinus Torvalds } 14152d1d4c1eSOmar Sandoval 1416310ca162STetsuo Handa /* Drop loop_ctl_mutex while we call into the filesystem. */ 1417b1ab5fa3STetsuo Handa path = lo->lo_backing_file->f_path; 1418b1ab5fa3STetsuo Handa path_get(&path); 1419310ca162STetsuo Handa mutex_unlock(&loop_ctl_mutex); 1420b1ab5fa3STetsuo Handa ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 14212d1d4c1eSOmar Sandoval if (!ret) { 14222d1d4c1eSOmar Sandoval info->lo_device = huge_encode_dev(stat.dev); 14232d1d4c1eSOmar Sandoval info->lo_inode = stat.ino; 14242d1d4c1eSOmar Sandoval info->lo_rdevice = huge_encode_dev(stat.rdev); 14252d1d4c1eSOmar Sandoval } 1426b1ab5fa3STetsuo Handa path_put(&path); 14272d1d4c1eSOmar Sandoval return ret; 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds static void 14311da177e4SLinus Torvalds loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 14321da177e4SLinus Torvalds { 14331da177e4SLinus Torvalds memset(info64, 0, sizeof(*info64)); 14341da177e4SLinus Torvalds info64->lo_number = info->lo_number; 14351da177e4SLinus Torvalds info64->lo_device = info->lo_device; 14361da177e4SLinus Torvalds info64->lo_inode = info->lo_inode; 14371da177e4SLinus Torvalds info64->lo_rdevice = info->lo_rdevice; 14381da177e4SLinus Torvalds info64->lo_offset = info->lo_offset; 14391da177e4SLinus Torvalds info64->lo_sizelimit = 0; 14401da177e4SLinus Torvalds info64->lo_encrypt_type = info->lo_encrypt_type; 14411da177e4SLinus Torvalds info64->lo_encrypt_key_size = info->lo_encrypt_key_size; 14421da177e4SLinus Torvalds info64->lo_flags = info->lo_flags; 14431da177e4SLinus Torvalds info64->lo_init[0] = info->lo_init[0]; 14441da177e4SLinus Torvalds info64->lo_init[1] = info->lo_init[1]; 14451da177e4SLinus Torvalds if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 14461da177e4SLinus Torvalds memcpy(info64->lo_crypt_name, info->lo_name, LO_NAME_SIZE); 14471da177e4SLinus Torvalds else 14481da177e4SLinus Torvalds memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 14491da177e4SLinus Torvalds memcpy(info64->lo_encrypt_key, info->lo_encrypt_key, LO_KEY_SIZE); 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds 14521da177e4SLinus Torvalds static int 14531da177e4SLinus Torvalds loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 14541da177e4SLinus Torvalds { 14551da177e4SLinus Torvalds memset(info, 0, sizeof(*info)); 14561da177e4SLinus Torvalds info->lo_number = info64->lo_number; 14571da177e4SLinus Torvalds info->lo_device = info64->lo_device; 14581da177e4SLinus Torvalds info->lo_inode = info64->lo_inode; 14591da177e4SLinus Torvalds info->lo_rdevice = info64->lo_rdevice; 14601da177e4SLinus Torvalds info->lo_offset = info64->lo_offset; 14611da177e4SLinus Torvalds info->lo_encrypt_type = info64->lo_encrypt_type; 14621da177e4SLinus Torvalds info->lo_encrypt_key_size = info64->lo_encrypt_key_size; 14631da177e4SLinus Torvalds info->lo_flags = info64->lo_flags; 14641da177e4SLinus Torvalds info->lo_init[0] = info64->lo_init[0]; 14651da177e4SLinus Torvalds info->lo_init[1] = info64->lo_init[1]; 14661da177e4SLinus Torvalds if (info->lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 14671da177e4SLinus Torvalds memcpy(info->lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 14681da177e4SLinus Torvalds else 14691da177e4SLinus Torvalds memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 14701da177e4SLinus Torvalds memcpy(info->lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds /* error in case values were truncated */ 14731da177e4SLinus Torvalds if (info->lo_device != info64->lo_device || 14741da177e4SLinus Torvalds info->lo_rdevice != info64->lo_rdevice || 14751da177e4SLinus Torvalds info->lo_inode != info64->lo_inode || 14761da177e4SLinus Torvalds info->lo_offset != info64->lo_offset) 14771da177e4SLinus Torvalds return -EOVERFLOW; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds return 0; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds static int 14831da177e4SLinus Torvalds loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 14841da177e4SLinus Torvalds { 14851da177e4SLinus Torvalds struct loop_info info; 14861da177e4SLinus Torvalds struct loop_info64 info64; 14871da177e4SLinus Torvalds 14881da177e4SLinus Torvalds if (copy_from_user(&info, arg, sizeof (struct loop_info))) 14891da177e4SLinus Torvalds return -EFAULT; 14901da177e4SLinus Torvalds loop_info64_from_old(&info, &info64); 14911da177e4SLinus Torvalds return loop_set_status(lo, &info64); 14921da177e4SLinus Torvalds } 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds static int 14951da177e4SLinus Torvalds loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 14961da177e4SLinus Torvalds { 14971da177e4SLinus Torvalds struct loop_info64 info64; 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 15001da177e4SLinus Torvalds return -EFAULT; 15011da177e4SLinus Torvalds return loop_set_status(lo, &info64); 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds static int 15051da177e4SLinus Torvalds loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 15061da177e4SLinus Torvalds struct loop_info info; 15071da177e4SLinus Torvalds struct loop_info64 info64; 1508bdac616dSOmar Sandoval int err; 15091da177e4SLinus Torvalds 15104a5ce9baSJan Kara if (!arg) 1511bdac616dSOmar Sandoval return -EINVAL; 15121da177e4SLinus Torvalds err = loop_get_status(lo, &info64); 15131da177e4SLinus Torvalds if (!err) 15141da177e4SLinus Torvalds err = loop_info64_to_old(&info64, &info); 15151da177e4SLinus Torvalds if (!err && copy_to_user(arg, &info, sizeof(info))) 15161da177e4SLinus Torvalds err = -EFAULT; 15171da177e4SLinus Torvalds 15181da177e4SLinus Torvalds return err; 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds 15211da177e4SLinus Torvalds static int 15221da177e4SLinus Torvalds loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 15231da177e4SLinus Torvalds struct loop_info64 info64; 1524bdac616dSOmar Sandoval int err; 15251da177e4SLinus Torvalds 15264a5ce9baSJan Kara if (!arg) 1527bdac616dSOmar Sandoval return -EINVAL; 15281da177e4SLinus Torvalds err = loop_get_status(lo, &info64); 15291da177e4SLinus Torvalds if (!err && copy_to_user(arg, &info64, sizeof(info64))) 15301da177e4SLinus Torvalds err = -EFAULT; 15311da177e4SLinus Torvalds 15321da177e4SLinus Torvalds return err; 15331da177e4SLinus Torvalds } 15341da177e4SLinus Torvalds 153551001b7dSHannes Reinecke static int loop_set_capacity(struct loop_device *lo) 153653d66608SJ. R. Okajima { 153753d66608SJ. R. Okajima if (unlikely(lo->lo_state != Lo_bound)) 15387b0576a3SGuo Chao return -ENXIO; 153953d66608SJ. R. Okajima 1540083a6a50SMartijn Coenen figure_loop_size(lo, lo->lo_offset, lo->lo_sizelimit); 1541083a6a50SMartijn Coenen 1542083a6a50SMartijn Coenen return 0; 154353d66608SJ. R. Okajima } 154453d66608SJ. R. Okajima 1545ab1cb278SMing Lei static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1546ab1cb278SMing Lei { 1547ab1cb278SMing Lei int error = -ENXIO; 1548ab1cb278SMing Lei if (lo->lo_state != Lo_bound) 1549ab1cb278SMing Lei goto out; 1550ab1cb278SMing Lei 1551ab1cb278SMing Lei __loop_update_dio(lo, !!arg); 1552ab1cb278SMing Lei if (lo->use_dio == !!arg) 1553ab1cb278SMing Lei return 0; 1554ab1cb278SMing Lei error = -EINVAL; 1555ab1cb278SMing Lei out: 1556ab1cb278SMing Lei return error; 1557ab1cb278SMing Lei } 1558ab1cb278SMing Lei 155989e4fdecSOmar Sandoval static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 156089e4fdecSOmar Sandoval { 15615db470e2SJaegeuk Kim int err = 0; 15625db470e2SJaegeuk Kim 156389e4fdecSOmar Sandoval if (lo->lo_state != Lo_bound) 156489e4fdecSOmar Sandoval return -ENXIO; 156589e4fdecSOmar Sandoval 156689e4fdecSOmar Sandoval if (arg < 512 || arg > PAGE_SIZE || !is_power_of_2(arg)) 156789e4fdecSOmar Sandoval return -EINVAL; 156889e4fdecSOmar Sandoval 15697e81f99aSMartijn Coenen if (lo->lo_queue->limits.logical_block_size == arg) 15707e81f99aSMartijn Coenen return 0; 15717e81f99aSMartijn Coenen 15725db470e2SJaegeuk Kim sync_blockdev(lo->lo_device); 15735db470e2SJaegeuk Kim kill_bdev(lo->lo_device); 15745db470e2SJaegeuk Kim 157589e4fdecSOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 157689e4fdecSOmar Sandoval 15775db470e2SJaegeuk Kim /* kill_bdev should have truncated all the pages */ 15787e81f99aSMartijn Coenen if (lo->lo_device->bd_inode->i_mapping->nrpages) { 15795db470e2SJaegeuk Kim err = -EAGAIN; 15805db470e2SJaegeuk Kim pr_warn("%s: loop%d (%s) has still dirty pages (nrpages=%lu)\n", 15815db470e2SJaegeuk Kim __func__, lo->lo_number, lo->lo_file_name, 15825db470e2SJaegeuk Kim lo->lo_device->bd_inode->i_mapping->nrpages); 15835db470e2SJaegeuk Kim goto out_unfreeze; 15845db470e2SJaegeuk Kim } 15855db470e2SJaegeuk Kim 158689e4fdecSOmar Sandoval blk_queue_logical_block_size(lo->lo_queue, arg); 1587bf093753SOmar Sandoval blk_queue_physical_block_size(lo->lo_queue, arg); 1588bf093753SOmar Sandoval blk_queue_io_min(lo->lo_queue, arg); 158989e4fdecSOmar Sandoval loop_update_dio(lo); 15905db470e2SJaegeuk Kim out_unfreeze: 159189e4fdecSOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 159289e4fdecSOmar Sandoval 15935db470e2SJaegeuk Kim return err; 159489e4fdecSOmar Sandoval } 159589e4fdecSOmar Sandoval 1596a1316544SJan Kara static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1597a1316544SJan Kara unsigned long arg) 15981da177e4SLinus Torvalds { 15991da177e4SLinus Torvalds int err; 16001da177e4SLinus Torvalds 1601c28445faSJan Kara err = mutex_lock_killable(&loop_ctl_mutex); 16023148ffbdSOmar Sandoval if (err) 1603a1316544SJan Kara return err; 16041da177e4SLinus Torvalds switch (cmd) { 160553d66608SJ. R. Okajima case LOOP_SET_CAPACITY: 160651001b7dSHannes Reinecke err = loop_set_capacity(lo); 160753d66608SJ. R. Okajima break; 1608ab1cb278SMing Lei case LOOP_SET_DIRECT_IO: 1609ab1cb278SMing Lei err = loop_set_dio(lo, arg); 1610ab1cb278SMing Lei break; 161189e4fdecSOmar Sandoval case LOOP_SET_BLOCK_SIZE: 161289e4fdecSOmar Sandoval err = loop_set_block_size(lo, arg); 161389e4fdecSOmar Sandoval break; 16141da177e4SLinus Torvalds default: 16151da177e4SLinus Torvalds err = lo->ioctl ? lo->ioctl(lo, cmd, arg) : -EINVAL; 16161da177e4SLinus Torvalds } 1617310ca162STetsuo Handa mutex_unlock(&loop_ctl_mutex); 1618a1316544SJan Kara return err; 1619a1316544SJan Kara } 1620f028f3b2SNikanth Karthikesan 1621a1316544SJan Kara static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1622a1316544SJan Kara unsigned int cmd, unsigned long arg) 1623a1316544SJan Kara { 1624a1316544SJan Kara struct loop_device *lo = bdev->bd_disk->private_data; 1625a1316544SJan Kara int err; 1626a1316544SJan Kara 1627a1316544SJan Kara switch (cmd) { 1628a1316544SJan Kara case LOOP_SET_FD: 1629757ecf40SJan Kara return loop_set_fd(lo, mode, bdev, arg); 1630a1316544SJan Kara case LOOP_CHANGE_FD: 1631c3710770SJan Kara return loop_change_fd(lo, bdev, arg); 1632a1316544SJan Kara case LOOP_CLR_FD: 16337ccd0791SJan Kara return loop_clr_fd(lo); 1634a1316544SJan Kara case LOOP_SET_STATUS: 1635a1316544SJan Kara err = -EPERM; 1636a1316544SJan Kara if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1637a1316544SJan Kara err = loop_set_status_old(lo, 1638a1316544SJan Kara (struct loop_info __user *)arg); 1639a1316544SJan Kara } 1640a1316544SJan Kara break; 1641a1316544SJan Kara case LOOP_GET_STATUS: 16424a5ce9baSJan Kara return loop_get_status_old(lo, (struct loop_info __user *) arg); 1643a1316544SJan Kara case LOOP_SET_STATUS64: 1644a1316544SJan Kara err = -EPERM; 1645a1316544SJan Kara if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1646a1316544SJan Kara err = loop_set_status64(lo, 1647a1316544SJan Kara (struct loop_info64 __user *) arg); 1648a1316544SJan Kara } 1649a1316544SJan Kara break; 1650a1316544SJan Kara case LOOP_GET_STATUS64: 16514a5ce9baSJan Kara return loop_get_status64(lo, (struct loop_info64 __user *) arg); 1652a1316544SJan Kara case LOOP_SET_CAPACITY: 1653a1316544SJan Kara case LOOP_SET_DIRECT_IO: 1654a1316544SJan Kara case LOOP_SET_BLOCK_SIZE: 1655a1316544SJan Kara if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1656a1316544SJan Kara return -EPERM; 1657a1316544SJan Kara /* Fall through */ 1658a1316544SJan Kara default: 1659a1316544SJan Kara err = lo_simple_ioctl(lo, cmd, arg); 1660a1316544SJan Kara break; 1661a1316544SJan Kara } 1662a1316544SJan Kara 16631da177e4SLinus Torvalds return err; 16641da177e4SLinus Torvalds } 16651da177e4SLinus Torvalds 1666863d5b82SDavid Howells #ifdef CONFIG_COMPAT 1667863d5b82SDavid Howells struct compat_loop_info { 1668863d5b82SDavid Howells compat_int_t lo_number; /* ioctl r/o */ 1669863d5b82SDavid Howells compat_dev_t lo_device; /* ioctl r/o */ 1670863d5b82SDavid Howells compat_ulong_t lo_inode; /* ioctl r/o */ 1671863d5b82SDavid Howells compat_dev_t lo_rdevice; /* ioctl r/o */ 1672863d5b82SDavid Howells compat_int_t lo_offset; 1673863d5b82SDavid Howells compat_int_t lo_encrypt_type; 1674863d5b82SDavid Howells compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1675863d5b82SDavid Howells compat_int_t lo_flags; /* ioctl r/o */ 1676863d5b82SDavid Howells char lo_name[LO_NAME_SIZE]; 1677863d5b82SDavid Howells unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1678863d5b82SDavid Howells compat_ulong_t lo_init[2]; 1679863d5b82SDavid Howells char reserved[4]; 1680863d5b82SDavid Howells }; 1681863d5b82SDavid Howells 1682863d5b82SDavid Howells /* 1683863d5b82SDavid Howells * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1684863d5b82SDavid Howells * - noinlined to reduce stack space usage in main part of driver 1685863d5b82SDavid Howells */ 1686863d5b82SDavid Howells static noinline int 1687ba674cfcSAl Viro loop_info64_from_compat(const struct compat_loop_info __user *arg, 1688863d5b82SDavid Howells struct loop_info64 *info64) 1689863d5b82SDavid Howells { 1690863d5b82SDavid Howells struct compat_loop_info info; 1691863d5b82SDavid Howells 1692863d5b82SDavid Howells if (copy_from_user(&info, arg, sizeof(info))) 1693863d5b82SDavid Howells return -EFAULT; 1694863d5b82SDavid Howells 1695863d5b82SDavid Howells memset(info64, 0, sizeof(*info64)); 1696863d5b82SDavid Howells info64->lo_number = info.lo_number; 1697863d5b82SDavid Howells info64->lo_device = info.lo_device; 1698863d5b82SDavid Howells info64->lo_inode = info.lo_inode; 1699863d5b82SDavid Howells info64->lo_rdevice = info.lo_rdevice; 1700863d5b82SDavid Howells info64->lo_offset = info.lo_offset; 1701863d5b82SDavid Howells info64->lo_sizelimit = 0; 1702863d5b82SDavid Howells info64->lo_encrypt_type = info.lo_encrypt_type; 1703863d5b82SDavid Howells info64->lo_encrypt_key_size = info.lo_encrypt_key_size; 1704863d5b82SDavid Howells info64->lo_flags = info.lo_flags; 1705863d5b82SDavid Howells info64->lo_init[0] = info.lo_init[0]; 1706863d5b82SDavid Howells info64->lo_init[1] = info.lo_init[1]; 1707863d5b82SDavid Howells if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1708863d5b82SDavid Howells memcpy(info64->lo_crypt_name, info.lo_name, LO_NAME_SIZE); 1709863d5b82SDavid Howells else 1710863d5b82SDavid Howells memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1711863d5b82SDavid Howells memcpy(info64->lo_encrypt_key, info.lo_encrypt_key, LO_KEY_SIZE); 1712863d5b82SDavid Howells return 0; 1713863d5b82SDavid Howells } 1714863d5b82SDavid Howells 1715863d5b82SDavid Howells /* 1716863d5b82SDavid Howells * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1717863d5b82SDavid Howells * - noinlined to reduce stack space usage in main part of driver 1718863d5b82SDavid Howells */ 1719863d5b82SDavid Howells static noinline int 1720863d5b82SDavid Howells loop_info64_to_compat(const struct loop_info64 *info64, 1721863d5b82SDavid Howells struct compat_loop_info __user *arg) 1722863d5b82SDavid Howells { 1723863d5b82SDavid Howells struct compat_loop_info info; 1724863d5b82SDavid Howells 1725863d5b82SDavid Howells memset(&info, 0, sizeof(info)); 1726863d5b82SDavid Howells info.lo_number = info64->lo_number; 1727863d5b82SDavid Howells info.lo_device = info64->lo_device; 1728863d5b82SDavid Howells info.lo_inode = info64->lo_inode; 1729863d5b82SDavid Howells info.lo_rdevice = info64->lo_rdevice; 1730863d5b82SDavid Howells info.lo_offset = info64->lo_offset; 1731863d5b82SDavid Howells info.lo_encrypt_type = info64->lo_encrypt_type; 1732863d5b82SDavid Howells info.lo_encrypt_key_size = info64->lo_encrypt_key_size; 1733863d5b82SDavid Howells info.lo_flags = info64->lo_flags; 1734863d5b82SDavid Howells info.lo_init[0] = info64->lo_init[0]; 1735863d5b82SDavid Howells info.lo_init[1] = info64->lo_init[1]; 1736863d5b82SDavid Howells if (info.lo_encrypt_type == LO_CRYPT_CRYPTOAPI) 1737863d5b82SDavid Howells memcpy(info.lo_name, info64->lo_crypt_name, LO_NAME_SIZE); 1738863d5b82SDavid Howells else 1739863d5b82SDavid Howells memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1740863d5b82SDavid Howells memcpy(info.lo_encrypt_key, info64->lo_encrypt_key, LO_KEY_SIZE); 1741863d5b82SDavid Howells 1742863d5b82SDavid Howells /* error in case values were truncated */ 1743863d5b82SDavid Howells if (info.lo_device != info64->lo_device || 1744863d5b82SDavid Howells info.lo_rdevice != info64->lo_rdevice || 1745863d5b82SDavid Howells info.lo_inode != info64->lo_inode || 1746863d5b82SDavid Howells info.lo_offset != info64->lo_offset || 1747863d5b82SDavid Howells info.lo_init[0] != info64->lo_init[0] || 1748863d5b82SDavid Howells info.lo_init[1] != info64->lo_init[1]) 1749863d5b82SDavid Howells return -EOVERFLOW; 1750863d5b82SDavid Howells 1751863d5b82SDavid Howells if (copy_to_user(arg, &info, sizeof(info))) 1752863d5b82SDavid Howells return -EFAULT; 1753863d5b82SDavid Howells return 0; 1754863d5b82SDavid Howells } 1755863d5b82SDavid Howells 1756863d5b82SDavid Howells static int 1757863d5b82SDavid Howells loop_set_status_compat(struct loop_device *lo, 1758863d5b82SDavid Howells const struct compat_loop_info __user *arg) 1759863d5b82SDavid Howells { 1760863d5b82SDavid Howells struct loop_info64 info64; 1761863d5b82SDavid Howells int ret; 1762863d5b82SDavid Howells 1763863d5b82SDavid Howells ret = loop_info64_from_compat(arg, &info64); 1764863d5b82SDavid Howells if (ret < 0) 1765863d5b82SDavid Howells return ret; 1766863d5b82SDavid Howells return loop_set_status(lo, &info64); 1767863d5b82SDavid Howells } 1768863d5b82SDavid Howells 1769863d5b82SDavid Howells static int 1770863d5b82SDavid Howells loop_get_status_compat(struct loop_device *lo, 1771863d5b82SDavid Howells struct compat_loop_info __user *arg) 1772863d5b82SDavid Howells { 1773863d5b82SDavid Howells struct loop_info64 info64; 1774bdac616dSOmar Sandoval int err; 1775863d5b82SDavid Howells 17764a5ce9baSJan Kara if (!arg) 1777bdac616dSOmar Sandoval return -EINVAL; 1778863d5b82SDavid Howells err = loop_get_status(lo, &info64); 1779863d5b82SDavid Howells if (!err) 1780863d5b82SDavid Howells err = loop_info64_to_compat(&info64, arg); 1781863d5b82SDavid Howells return err; 1782863d5b82SDavid Howells } 1783863d5b82SDavid Howells 1784bb214884SAl Viro static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1785bb214884SAl Viro unsigned int cmd, unsigned long arg) 1786863d5b82SDavid Howells { 1787bb214884SAl Viro struct loop_device *lo = bdev->bd_disk->private_data; 1788863d5b82SDavid Howells int err; 1789863d5b82SDavid Howells 1790863d5b82SDavid Howells switch(cmd) { 1791863d5b82SDavid Howells case LOOP_SET_STATUS: 17923148ffbdSOmar Sandoval err = loop_set_status_compat(lo, 17933148ffbdSOmar Sandoval (const struct compat_loop_info __user *)arg); 1794863d5b82SDavid Howells break; 1795863d5b82SDavid Howells case LOOP_GET_STATUS: 17963148ffbdSOmar Sandoval err = loop_get_status_compat(lo, 17973148ffbdSOmar Sandoval (struct compat_loop_info __user *)arg); 1798863d5b82SDavid Howells break; 179953d66608SJ. R. Okajima case LOOP_SET_CAPACITY: 1800863d5b82SDavid Howells case LOOP_CLR_FD: 1801863d5b82SDavid Howells case LOOP_GET_STATUS64: 1802863d5b82SDavid Howells case LOOP_SET_STATUS64: 1803863d5b82SDavid Howells arg = (unsigned long) compat_ptr(arg); 1804d893ff86SGustavo A. R. Silva /* fall through */ 1805863d5b82SDavid Howells case LOOP_SET_FD: 1806863d5b82SDavid Howells case LOOP_CHANGE_FD: 18079fea4b39SEvan Green case LOOP_SET_BLOCK_SIZE: 1808fdbe4eeeSAlessio Balsini case LOOP_SET_DIRECT_IO: 1809bb214884SAl Viro err = lo_ioctl(bdev, mode, cmd, arg); 1810863d5b82SDavid Howells break; 1811863d5b82SDavid Howells default: 1812863d5b82SDavid Howells err = -ENOIOCTLCMD; 1813863d5b82SDavid Howells break; 1814863d5b82SDavid Howells } 1815863d5b82SDavid Howells return err; 1816863d5b82SDavid Howells } 1817863d5b82SDavid Howells #endif 1818863d5b82SDavid Howells 1819bb214884SAl Viro static int lo_open(struct block_device *bdev, fmode_t mode) 18201da177e4SLinus Torvalds { 1821770fe30aSKay Sievers struct loop_device *lo; 18220a42e99bSJan Kara int err; 1823770fe30aSKay Sievers 18240a42e99bSJan Kara err = mutex_lock_killable(&loop_ctl_mutex); 18250a42e99bSJan Kara if (err) 18260a42e99bSJan Kara return err; 1827770fe30aSKay Sievers lo = bdev->bd_disk->private_data; 1828770fe30aSKay Sievers if (!lo) { 1829770fe30aSKay Sievers err = -ENXIO; 1830770fe30aSKay Sievers goto out; 1831770fe30aSKay Sievers } 18321da177e4SLinus Torvalds 1833f8933667SMing Lei atomic_inc(&lo->lo_refcnt); 1834770fe30aSKay Sievers out: 18350a42e99bSJan Kara mutex_unlock(&loop_ctl_mutex); 1836770fe30aSKay Sievers return err; 18371da177e4SLinus Torvalds } 18381da177e4SLinus Torvalds 1839967d1dc1SJan Kara static void lo_release(struct gendisk *disk, fmode_t mode) 18401da177e4SLinus Torvalds { 1841967d1dc1SJan Kara struct loop_device *lo; 18421da177e4SLinus Torvalds 18430a42e99bSJan Kara mutex_lock(&loop_ctl_mutex); 1844967d1dc1SJan Kara lo = disk->private_data; 1845f8933667SMing Lei if (atomic_dec_return(&lo->lo_refcnt)) 18460a42e99bSJan Kara goto out_unlock; 1847f8933667SMing Lei 184814f27939SMilan Broz if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1849a2505b79SJan Kara if (lo->lo_state != Lo_bound) 1850a2505b79SJan Kara goto out_unlock; 1851a2505b79SJan Kara lo->lo_state = Lo_rundown; 18527ccd0791SJan Kara mutex_unlock(&loop_ctl_mutex); 185314f27939SMilan Broz /* 185414f27939SMilan Broz * In autoclear mode, stop the loop thread 185514f27939SMilan Broz * and remove configuration after last close. 185614f27939SMilan Broz */ 18570da03cabSJan Kara __loop_clr_fd(lo, true); 18580a42e99bSJan Kara return; 185943cade80SOmar Sandoval } else if (lo->lo_state == Lo_bound) { 186014f27939SMilan Broz /* 186114f27939SMilan Broz * Otherwise keep thread (if running) and config, 186214f27939SMilan Broz * but flush possible ongoing bios in thread. 186314f27939SMilan Broz */ 186443cade80SOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 186543cade80SOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 186614f27939SMilan Broz } 186796c58655SDavid Woodhouse 18680a42e99bSJan Kara out_unlock: 1869310ca162STetsuo Handa mutex_unlock(&loop_ctl_mutex); 1870ae665016SLinus Torvalds } 1871ae665016SLinus Torvalds 187283d5cde4SAlexey Dobriyan static const struct block_device_operations lo_fops = { 18731da177e4SLinus Torvalds .owner = THIS_MODULE, 1874bb214884SAl Viro .open = lo_open, 1875bb214884SAl Viro .release = lo_release, 1876bb214884SAl Viro .ioctl = lo_ioctl, 1877863d5b82SDavid Howells #ifdef CONFIG_COMPAT 1878bb214884SAl Viro .compat_ioctl = lo_compat_ioctl, 1879863d5b82SDavid Howells #endif 18801da177e4SLinus Torvalds }; 18811da177e4SLinus Torvalds 18821da177e4SLinus Torvalds /* 18831da177e4SLinus Torvalds * And now the modules code and kernel interface. 18841da177e4SLinus Torvalds */ 188573285082SKen Chen static int max_loop; 18865657a819SJoe Perches module_param(max_loop, int, 0444); 1887a47653fcSKen Chen MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 18885657a819SJoe Perches module_param(max_part, int, 0444); 1889476a4813SLaurent Vivier MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 18901da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 18911da177e4SLinus Torvalds MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 18921da177e4SLinus Torvalds 18931da177e4SLinus Torvalds int loop_register_transfer(struct loop_func_table *funcs) 18941da177e4SLinus Torvalds { 18951da177e4SLinus Torvalds unsigned int n = funcs->number; 18961da177e4SLinus Torvalds 18971da177e4SLinus Torvalds if (n >= MAX_LO_CRYPT || xfer_funcs[n]) 18981da177e4SLinus Torvalds return -EINVAL; 18991da177e4SLinus Torvalds xfer_funcs[n] = funcs; 19001da177e4SLinus Torvalds return 0; 19011da177e4SLinus Torvalds } 19021da177e4SLinus Torvalds 190334dd82afSKay Sievers static int unregister_transfer_cb(int id, void *ptr, void *data) 190434dd82afSKay Sievers { 190534dd82afSKay Sievers struct loop_device *lo = ptr; 190634dd82afSKay Sievers struct loop_func_table *xfer = data; 190734dd82afSKay Sievers 1908310ca162STetsuo Handa mutex_lock(&loop_ctl_mutex); 190934dd82afSKay Sievers if (lo->lo_encryption == xfer) 191034dd82afSKay Sievers loop_release_xfer(lo); 1911310ca162STetsuo Handa mutex_unlock(&loop_ctl_mutex); 191234dd82afSKay Sievers return 0; 191334dd82afSKay Sievers } 191434dd82afSKay Sievers 19151da177e4SLinus Torvalds int loop_unregister_transfer(int number) 19161da177e4SLinus Torvalds { 19171da177e4SLinus Torvalds unsigned int n = number; 19181da177e4SLinus Torvalds struct loop_func_table *xfer; 19191da177e4SLinus Torvalds 19201da177e4SLinus Torvalds if (n == 0 || n >= MAX_LO_CRYPT || (xfer = xfer_funcs[n]) == NULL) 19211da177e4SLinus Torvalds return -EINVAL; 19221da177e4SLinus Torvalds 19231da177e4SLinus Torvalds xfer_funcs[n] = NULL; 192434dd82afSKay Sievers idr_for_each(&loop_index_idr, &unregister_transfer_cb, xfer); 19251da177e4SLinus Torvalds return 0; 19261da177e4SLinus Torvalds } 19271da177e4SLinus Torvalds 19281da177e4SLinus Torvalds EXPORT_SYMBOL(loop_register_transfer); 19291da177e4SLinus Torvalds EXPORT_SYMBOL(loop_unregister_transfer); 19301da177e4SLinus Torvalds 1931fc17b653SChristoph Hellwig static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1932b5dd2f60SMing Lei const struct blk_mq_queue_data *bd) 1933b5dd2f60SMing Lei { 19341894e916SJens Axboe struct request *rq = bd->rq; 19351894e916SJens Axboe struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 19361894e916SJens Axboe struct loop_device *lo = rq->q->queuedata; 1937b5dd2f60SMing Lei 19381894e916SJens Axboe blk_mq_start_request(rq); 1939b5dd2f60SMing Lei 1940f4aa4c7bSMing Lei if (lo->lo_state != Lo_bound) 1941fc17b653SChristoph Hellwig return BLK_STS_IOERR; 1942f4aa4c7bSMing Lei 19431894e916SJens Axboe switch (req_op(rq)) { 1944f0225cacSChristoph Hellwig case REQ_OP_FLUSH: 1945f0225cacSChristoph Hellwig case REQ_OP_DISCARD: 194619372e27SChristoph Hellwig case REQ_OP_WRITE_ZEROES: 1947bc07c10aSMing Lei cmd->use_aio = false; 1948f0225cacSChristoph Hellwig break; 1949f0225cacSChristoph Hellwig default: 1950f0225cacSChristoph Hellwig cmd->use_aio = lo->use_dio; 1951f0225cacSChristoph Hellwig break; 1952f0225cacSChristoph Hellwig } 1953bc07c10aSMing Lei 1954d4478e92SShaohua Li /* always use the first bio's css */ 19550b508bc9SShaohua Li #ifdef CONFIG_BLK_CGROUP 1956db6638d7SDennis Zhou if (cmd->use_aio && rq->bio && rq->bio->bi_blkg) { 1957db6638d7SDennis Zhou cmd->css = &bio_blkcg(rq->bio)->css; 1958d4478e92SShaohua Li css_get(cmd->css); 1959d4478e92SShaohua Li } else 1960d4478e92SShaohua Li #endif 1961d4478e92SShaohua Li cmd->css = NULL; 19623989144fSPetr Mladek kthread_queue_work(&lo->worker, &cmd->work); 1963b5dd2f60SMing Lei 1964fc17b653SChristoph Hellwig return BLK_STS_OK; 1965b5dd2f60SMing Lei } 1966b5dd2f60SMing Lei 1967b5dd2f60SMing Lei static void loop_handle_cmd(struct loop_cmd *cmd) 1968b5dd2f60SMing Lei { 19691894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 19701894e916SJens Axboe const bool write = op_is_write(req_op(rq)); 19711894e916SJens Axboe struct loop_device *lo = rq->q->queuedata; 1972f4829a9bSChristoph Hellwig int ret = 0; 1973b5dd2f60SMing Lei 1974f4829a9bSChristoph Hellwig if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1975f4829a9bSChristoph Hellwig ret = -EIO; 1976b5dd2f60SMing Lei goto failed; 1977f4829a9bSChristoph Hellwig } 1978b5dd2f60SMing Lei 19791894e916SJens Axboe ret = do_req_filebacked(lo, rq); 1980b5dd2f60SMing Lei failed: 1981bc07c10aSMing Lei /* complete non-aio request */ 1982fe2cb290SChristoph Hellwig if (!cmd->use_aio || ret) { 19838cd55087SEvan Green if (ret == -EOPNOTSUPP) 19848cd55087SEvan Green cmd->ret = ret; 19858cd55087SEvan Green else 1986fe2cb290SChristoph Hellwig cmd->ret = ret ? -EIO : 0; 19871894e916SJens Axboe blk_mq_complete_request(rq); 1988fe2cb290SChristoph Hellwig } 1989b5dd2f60SMing Lei } 1990b5dd2f60SMing Lei 1991e03a3d7aSMing Lei static void loop_queue_work(struct kthread_work *work) 1992b5dd2f60SMing Lei { 1993b5dd2f60SMing Lei struct loop_cmd *cmd = 1994e03a3d7aSMing Lei container_of(work, struct loop_cmd, work); 1995b5dd2f60SMing Lei 1996b5dd2f60SMing Lei loop_handle_cmd(cmd); 1997b5dd2f60SMing Lei } 1998b5dd2f60SMing Lei 1999d6296d39SChristoph Hellwig static int loop_init_request(struct blk_mq_tag_set *set, struct request *rq, 2000d6296d39SChristoph Hellwig unsigned int hctx_idx, unsigned int numa_node) 2001b5dd2f60SMing Lei { 2002b5dd2f60SMing Lei struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 2003b5dd2f60SMing Lei 20043989144fSPetr Mladek kthread_init_work(&cmd->work, loop_queue_work); 2005b5dd2f60SMing Lei return 0; 2006b5dd2f60SMing Lei } 2007b5dd2f60SMing Lei 2008f363b089SEric Biggers static const struct blk_mq_ops loop_mq_ops = { 2009b5dd2f60SMing Lei .queue_rq = loop_queue_rq, 2010b5dd2f60SMing Lei .init_request = loop_init_request, 2011fe2cb290SChristoph Hellwig .complete = lo_complete_rq, 2012b5dd2f60SMing Lei }; 2013b5dd2f60SMing Lei 201434dd82afSKay Sievers static int loop_add(struct loop_device **l, int i) 20151da177e4SLinus Torvalds { 201673285082SKen Chen struct loop_device *lo; 201773285082SKen Chen struct gendisk *disk; 201834dd82afSKay Sievers int err; 20191da177e4SLinus Torvalds 202034dd82afSKay Sievers err = -ENOMEM; 202168d740d7SSilva Paulo lo = kzalloc(sizeof(*lo), GFP_KERNEL); 202268d740d7SSilva Paulo if (!lo) 202373285082SKen Chen goto out; 202434dd82afSKay Sievers 2025ef7e7c82SMikulas Patocka lo->lo_state = Lo_unbound; 2026ef7e7c82SMikulas Patocka 2027c718aa65STejun Heo /* allocate id, if @id >= 0, we're requesting that specific id */ 202834dd82afSKay Sievers if (i >= 0) { 2029c718aa65STejun Heo err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 2030c718aa65STejun Heo if (err == -ENOSPC) 203134dd82afSKay Sievers err = -EEXIST; 203234dd82afSKay Sievers } else { 2033c718aa65STejun Heo err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 203434dd82afSKay Sievers } 203534dd82afSKay Sievers if (err < 0) 203634dd82afSKay Sievers goto out_free_dev; 2037c718aa65STejun Heo i = err; 20381da177e4SLinus Torvalds 2039183cfb57SWei Yongjun err = -ENOMEM; 2040b5dd2f60SMing Lei lo->tag_set.ops = &loop_mq_ops; 2041b5dd2f60SMing Lei lo->tag_set.nr_hw_queues = 1; 2042b5dd2f60SMing Lei lo->tag_set.queue_depth = 128; 2043b5dd2f60SMing Lei lo->tag_set.numa_node = NUMA_NO_NODE; 2044b5dd2f60SMing Lei lo->tag_set.cmd_size = sizeof(struct loop_cmd); 204556d18f62SMing Lei lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE; 2046b5dd2f60SMing Lei lo->tag_set.driver_data = lo; 2047b5dd2f60SMing Lei 2048b5dd2f60SMing Lei err = blk_mq_alloc_tag_set(&lo->tag_set); 2049b5dd2f60SMing Lei if (err) 20503ec981e3SMikulas Patocka goto out_free_idr; 205173285082SKen Chen 2052b5dd2f60SMing Lei lo->lo_queue = blk_mq_init_queue(&lo->tag_set); 205338a3499fSChengguang Xu if (IS_ERR(lo->lo_queue)) { 2054b5dd2f60SMing Lei err = PTR_ERR(lo->lo_queue); 2055b5dd2f60SMing Lei goto out_cleanup_tags; 2056b5dd2f60SMing Lei } 2057ef7e7c82SMikulas Patocka lo->lo_queue->queuedata = lo; 2058ef7e7c82SMikulas Patocka 205954bb0adeSShaohua Li blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 206040326d8aSShaohua Li 20615b5e20f4SMing Lei /* 206240326d8aSShaohua Li * By default, we do buffer IO, so it doesn't make sense to enable 206340326d8aSShaohua Li * merge because the I/O submitted to backing file is handled page by 206440326d8aSShaohua Li * page. For directio mode, merge does help to dispatch bigger request 206540326d8aSShaohua Li * to underlayer disk. We will enable merge once directio is enabled. 20665b5e20f4SMing Lei */ 20678b904b5bSBart Van Assche blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 20685b5e20f4SMing Lei 20697a649737SMinfei Huang err = -ENOMEM; 2070476a4813SLaurent Vivier disk = lo->lo_disk = alloc_disk(1 << part_shift); 207173285082SKen Chen if (!disk) 207273285082SKen Chen goto out_free_queue; 207373285082SKen Chen 2074e03c8dd1SKay Sievers /* 2075e03c8dd1SKay Sievers * Disable partition scanning by default. The in-kernel partition 2076e03c8dd1SKay Sievers * scanning can be requested individually per-device during its 2077e03c8dd1SKay Sievers * setup. Userspace can always add and remove partitions from all 2078e03c8dd1SKay Sievers * devices. The needed partition minors are allocated from the 2079e03c8dd1SKay Sievers * extended minor space, the main loop device numbers will continue 2080e03c8dd1SKay Sievers * to match the loop minors, regardless of the number of partitions 2081e03c8dd1SKay Sievers * used. 2082e03c8dd1SKay Sievers * 2083e03c8dd1SKay Sievers * If max_part is given, partition scanning is globally enabled for 2084e03c8dd1SKay Sievers * all loop devices. The minors for the main loop devices will be 2085e03c8dd1SKay Sievers * multiples of max_part. 2086e03c8dd1SKay Sievers * 2087e03c8dd1SKay Sievers * Note: Global-for-all-devices, set-only-at-init, read-only module 2088e03c8dd1SKay Sievers * parameteters like 'max_loop' and 'max_part' make things needlessly 2089e03c8dd1SKay Sievers * complicated, are too static, inflexible and may surprise 2090e03c8dd1SKay Sievers * userspace tools. Parameters like this in general should be avoided. 2091e03c8dd1SKay Sievers */ 2092e03c8dd1SKay Sievers if (!part_shift) 2093e03c8dd1SKay Sievers disk->flags |= GENHD_FL_NO_PART_SCAN; 2094e03c8dd1SKay Sievers disk->flags |= GENHD_FL_EXT_DEVT; 2095f8933667SMing Lei atomic_set(&lo->lo_refcnt, 0); 20961da177e4SLinus Torvalds lo->lo_number = i; 20971da177e4SLinus Torvalds spin_lock_init(&lo->lo_lock); 20981da177e4SLinus Torvalds disk->major = LOOP_MAJOR; 2099476a4813SLaurent Vivier disk->first_minor = i << part_shift; 21001da177e4SLinus Torvalds disk->fops = &lo_fops; 21011da177e4SLinus Torvalds disk->private_data = lo; 21021da177e4SLinus Torvalds disk->queue = lo->lo_queue; 210373285082SKen Chen sprintf(disk->disk_name, "loop%d", i); 210434dd82afSKay Sievers add_disk(disk); 210534dd82afSKay Sievers *l = lo; 210634dd82afSKay Sievers return lo->lo_number; 210773285082SKen Chen 210873285082SKen Chen out_free_queue: 210973285082SKen Chen blk_cleanup_queue(lo->lo_queue); 2110b5dd2f60SMing Lei out_cleanup_tags: 2111b5dd2f60SMing Lei blk_mq_free_tag_set(&lo->tag_set); 21123ec981e3SMikulas Patocka out_free_idr: 21133ec981e3SMikulas Patocka idr_remove(&loop_index_idr, i); 211473285082SKen Chen out_free_dev: 211573285082SKen Chen kfree(lo); 211673285082SKen Chen out: 211734dd82afSKay Sievers return err; 21181da177e4SLinus Torvalds } 21191da177e4SLinus Torvalds 212034dd82afSKay Sievers static void loop_remove(struct loop_device *lo) 212173285082SKen Chen { 21226cd18e71SNeilBrown del_gendisk(lo->lo_disk); 21230fa8ebddSBart Van Assche blk_cleanup_queue(lo->lo_queue); 2124b5dd2f60SMing Lei blk_mq_free_tag_set(&lo->tag_set); 212573285082SKen Chen put_disk(lo->lo_disk); 212673285082SKen Chen kfree(lo); 212773285082SKen Chen } 212873285082SKen Chen 2129770fe30aSKay Sievers static int find_free_cb(int id, void *ptr, void *data) 2130770fe30aSKay Sievers { 2131770fe30aSKay Sievers struct loop_device *lo = ptr; 2132770fe30aSKay Sievers struct loop_device **l = data; 2133770fe30aSKay Sievers 2134770fe30aSKay Sievers if (lo->lo_state == Lo_unbound) { 2135770fe30aSKay Sievers *l = lo; 2136770fe30aSKay Sievers return 1; 2137770fe30aSKay Sievers } 2138770fe30aSKay Sievers return 0; 2139770fe30aSKay Sievers } 2140770fe30aSKay Sievers 214134dd82afSKay Sievers static int loop_lookup(struct loop_device **l, int i) 2142a47653fcSKen Chen { 2143a47653fcSKen Chen struct loop_device *lo; 214434dd82afSKay Sievers int ret = -ENODEV; 2145a47653fcSKen Chen 2146770fe30aSKay Sievers if (i < 0) { 2147770fe30aSKay Sievers int err; 2148770fe30aSKay Sievers 2149770fe30aSKay Sievers err = idr_for_each(&loop_index_idr, &find_free_cb, &lo); 2150770fe30aSKay Sievers if (err == 1) { 2151770fe30aSKay Sievers *l = lo; 2152770fe30aSKay Sievers ret = lo->lo_number; 2153770fe30aSKay Sievers } 2154770fe30aSKay Sievers goto out; 2155a47653fcSKen Chen } 2156a47653fcSKen Chen 2157770fe30aSKay Sievers /* lookup and return a specific i */ 215834dd82afSKay Sievers lo = idr_find(&loop_index_idr, i); 2159a47653fcSKen Chen if (lo) { 216034dd82afSKay Sievers *l = lo; 216134dd82afSKay Sievers ret = lo->lo_number; 2162a47653fcSKen Chen } 2163770fe30aSKay Sievers out: 216434dd82afSKay Sievers return ret; 2165a47653fcSKen Chen } 2166a47653fcSKen Chen 216773285082SKen Chen static struct kobject *loop_probe(dev_t dev, int *part, void *data) 216873285082SKen Chen { 2169705962ccSAl Viro struct loop_device *lo; 217007002e99SAl Viro struct kobject *kobj; 217134dd82afSKay Sievers int err; 217273285082SKen Chen 21730a42e99bSJan Kara mutex_lock(&loop_ctl_mutex); 217434dd82afSKay Sievers err = loop_lookup(&lo, MINOR(dev) >> part_shift); 217534dd82afSKay Sievers if (err < 0) 217634dd82afSKay Sievers err = loop_add(&lo, MINOR(dev) >> part_shift); 217734dd82afSKay Sievers if (err < 0) 2178a207f593SMikulas Patocka kobj = NULL; 217934dd82afSKay Sievers else 21803079c22eSJan Kara kobj = get_disk_and_module(lo->lo_disk); 21810a42e99bSJan Kara mutex_unlock(&loop_ctl_mutex); 218273285082SKen Chen 218373285082SKen Chen *part = 0; 218407002e99SAl Viro return kobj; 218573285082SKen Chen } 218673285082SKen Chen 2187770fe30aSKay Sievers static long loop_control_ioctl(struct file *file, unsigned int cmd, 2188770fe30aSKay Sievers unsigned long parm) 2189770fe30aSKay Sievers { 2190770fe30aSKay Sievers struct loop_device *lo; 21910a42e99bSJan Kara int ret; 2192770fe30aSKay Sievers 21930a42e99bSJan Kara ret = mutex_lock_killable(&loop_ctl_mutex); 21940a42e99bSJan Kara if (ret) 21950a42e99bSJan Kara return ret; 21960a42e99bSJan Kara 21970a42e99bSJan Kara ret = -ENOSYS; 2198770fe30aSKay Sievers switch (cmd) { 2199770fe30aSKay Sievers case LOOP_CTL_ADD: 2200770fe30aSKay Sievers ret = loop_lookup(&lo, parm); 2201770fe30aSKay Sievers if (ret >= 0) { 2202770fe30aSKay Sievers ret = -EEXIST; 2203770fe30aSKay Sievers break; 2204770fe30aSKay Sievers } 2205770fe30aSKay Sievers ret = loop_add(&lo, parm); 2206770fe30aSKay Sievers break; 2207770fe30aSKay Sievers case LOOP_CTL_REMOVE: 2208770fe30aSKay Sievers ret = loop_lookup(&lo, parm); 2209770fe30aSKay Sievers if (ret < 0) 2210770fe30aSKay Sievers break; 2211770fe30aSKay Sievers if (lo->lo_state != Lo_unbound) { 2212770fe30aSKay Sievers ret = -EBUSY; 2213770fe30aSKay Sievers break; 2214770fe30aSKay Sievers } 2215f8933667SMing Lei if (atomic_read(&lo->lo_refcnt) > 0) { 2216770fe30aSKay Sievers ret = -EBUSY; 2217770fe30aSKay Sievers break; 2218770fe30aSKay Sievers } 2219770fe30aSKay Sievers lo->lo_disk->private_data = NULL; 2220770fe30aSKay Sievers idr_remove(&loop_index_idr, lo->lo_number); 2221770fe30aSKay Sievers loop_remove(lo); 2222770fe30aSKay Sievers break; 2223770fe30aSKay Sievers case LOOP_CTL_GET_FREE: 2224770fe30aSKay Sievers ret = loop_lookup(&lo, -1); 2225770fe30aSKay Sievers if (ret >= 0) 2226770fe30aSKay Sievers break; 2227770fe30aSKay Sievers ret = loop_add(&lo, -1); 2228770fe30aSKay Sievers } 22290a42e99bSJan Kara mutex_unlock(&loop_ctl_mutex); 2230770fe30aSKay Sievers 2231770fe30aSKay Sievers return ret; 2232770fe30aSKay Sievers } 2233770fe30aSKay Sievers 2234770fe30aSKay Sievers static const struct file_operations loop_ctl_fops = { 2235770fe30aSKay Sievers .open = nonseekable_open, 2236770fe30aSKay Sievers .unlocked_ioctl = loop_control_ioctl, 2237770fe30aSKay Sievers .compat_ioctl = loop_control_ioctl, 2238770fe30aSKay Sievers .owner = THIS_MODULE, 2239770fe30aSKay Sievers .llseek = noop_llseek, 2240770fe30aSKay Sievers }; 2241770fe30aSKay Sievers 2242770fe30aSKay Sievers static struct miscdevice loop_misc = { 2243770fe30aSKay Sievers .minor = LOOP_CTRL_MINOR, 2244770fe30aSKay Sievers .name = "loop-control", 2245770fe30aSKay Sievers .fops = &loop_ctl_fops, 2246770fe30aSKay Sievers }; 2247770fe30aSKay Sievers 2248770fe30aSKay Sievers MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2249770fe30aSKay Sievers MODULE_ALIAS("devname:loop-control"); 2250770fe30aSKay Sievers 225173285082SKen Chen static int __init loop_init(void) 225273285082SKen Chen { 2253a47653fcSKen Chen int i, nr; 2254a47653fcSKen Chen unsigned long range; 225534dd82afSKay Sievers struct loop_device *lo; 2256770fe30aSKay Sievers int err; 2257a47653fcSKen Chen 2258476a4813SLaurent Vivier part_shift = 0; 2259ac04fee0SNamhyung Kim if (max_part > 0) { 2260476a4813SLaurent Vivier part_shift = fls(max_part); 2261476a4813SLaurent Vivier 2262ac04fee0SNamhyung Kim /* 2263ac04fee0SNamhyung Kim * Adjust max_part according to part_shift as it is exported 2264ac04fee0SNamhyung Kim * to user space so that user can decide correct minor number 2265ac04fee0SNamhyung Kim * if [s]he want to create more devices. 2266ac04fee0SNamhyung Kim * 2267ac04fee0SNamhyung Kim * Note that -1 is required because partition 0 is reserved 2268ac04fee0SNamhyung Kim * for the whole disk. 2269ac04fee0SNamhyung Kim */ 2270ac04fee0SNamhyung Kim max_part = (1UL << part_shift) - 1; 2271ac04fee0SNamhyung Kim } 2272ac04fee0SNamhyung Kim 2273b1a66504SGuo Chao if ((1UL << part_shift) > DISK_MAX_PARTS) { 2274b1a66504SGuo Chao err = -EINVAL; 2275a8c1d064SAnton Volkov goto err_out; 2276b1a66504SGuo Chao } 227778f4bb36SNamhyung Kim 2278b1a66504SGuo Chao if (max_loop > 1UL << (MINORBITS - part_shift)) { 2279b1a66504SGuo Chao err = -EINVAL; 2280a8c1d064SAnton Volkov goto err_out; 2281b1a66504SGuo Chao } 228273285082SKen Chen 2283d134b00bSKay Sievers /* 2284d134b00bSKay Sievers * If max_loop is specified, create that many devices upfront. 2285d134b00bSKay Sievers * This also becomes a hard limit. If max_loop is not specified, 2286d134b00bSKay Sievers * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2287d134b00bSKay Sievers * init time. Loop devices can be requested on-demand with the 2288d134b00bSKay Sievers * /dev/loop-control interface, or be instantiated by accessing 2289d134b00bSKay Sievers * a 'dead' device node. 2290d134b00bSKay Sievers */ 229173285082SKen Chen if (max_loop) { 2292a47653fcSKen Chen nr = max_loop; 2293a1c15c59SNamhyung Kim range = max_loop << part_shift; 2294a47653fcSKen Chen } else { 2295d134b00bSKay Sievers nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2296a1c15c59SNamhyung Kim range = 1UL << MINORBITS; 229773285082SKen Chen } 2298a47653fcSKen Chen 2299a8c1d064SAnton Volkov err = misc_register(&loop_misc); 2300a8c1d064SAnton Volkov if (err < 0) 2301a8c1d064SAnton Volkov goto err_out; 2302a8c1d064SAnton Volkov 2303a8c1d064SAnton Volkov 2304b1a66504SGuo Chao if (register_blkdev(LOOP_MAJOR, "loop")) { 2305b1a66504SGuo Chao err = -EIO; 2306b1a66504SGuo Chao goto misc_out; 2307b1a66504SGuo Chao } 2308a47653fcSKen Chen 2309a47653fcSKen Chen blk_register_region(MKDEV(LOOP_MAJOR, 0), range, 2310a47653fcSKen Chen THIS_MODULE, loop_probe, NULL, NULL); 2311a47653fcSKen Chen 2312d134b00bSKay Sievers /* pre-create number of devices given by config or max_loop */ 23130a42e99bSJan Kara mutex_lock(&loop_ctl_mutex); 231434dd82afSKay Sievers for (i = 0; i < nr; i++) 231534dd82afSKay Sievers loop_add(&lo, i); 23160a42e99bSJan Kara mutex_unlock(&loop_ctl_mutex); 231734dd82afSKay Sievers 231873285082SKen Chen printk(KERN_INFO "loop: module loaded\n"); 23191da177e4SLinus Torvalds return 0; 2320b1a66504SGuo Chao 2321b1a66504SGuo Chao misc_out: 2322b1a66504SGuo Chao misc_deregister(&loop_misc); 2323a8c1d064SAnton Volkov err_out: 2324b1a66504SGuo Chao return err; 232534dd82afSKay Sievers } 2326a47653fcSKen Chen 232734dd82afSKay Sievers static int loop_exit_cb(int id, void *ptr, void *data) 232834dd82afSKay Sievers { 232934dd82afSKay Sievers struct loop_device *lo = ptr; 2330a47653fcSKen Chen 233134dd82afSKay Sievers loop_remove(lo); 233234dd82afSKay Sievers return 0; 23331da177e4SLinus Torvalds } 23341da177e4SLinus Torvalds 233573285082SKen Chen static void __exit loop_exit(void) 23361da177e4SLinus Torvalds { 2337a47653fcSKen Chen unsigned long range; 23381da177e4SLinus Torvalds 2339a1c15c59SNamhyung Kim range = max_loop ? max_loop << part_shift : 1UL << MINORBITS; 2340a47653fcSKen Chen 234134dd82afSKay Sievers idr_for_each(&loop_index_idr, &loop_exit_cb, NULL); 234234dd82afSKay Sievers idr_destroy(&loop_index_idr); 234373285082SKen Chen 2344a47653fcSKen Chen blk_unregister_region(MKDEV(LOOP_MAJOR, 0), range); 234500d59405SAkinobu Mita unregister_blkdev(LOOP_MAJOR, "loop"); 2346770fe30aSKay Sievers 2347770fe30aSKay Sievers misc_deregister(&loop_misc); 23481da177e4SLinus Torvalds } 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds module_init(loop_init); 23511da177e4SLinus Torvalds module_exit(loop_exit); 23521da177e4SLinus Torvalds 23531da177e4SLinus Torvalds #ifndef MODULE 23541da177e4SLinus Torvalds static int __init max_loop_setup(char *str) 23551da177e4SLinus Torvalds { 23561da177e4SLinus Torvalds max_loop = simple_strtol(str, NULL, 0); 23571da177e4SLinus Torvalds return 1; 23581da177e4SLinus Torvalds } 23591da177e4SLinus Torvalds 23601da177e4SLinus Torvalds __setup("max_loop=", max_loop_setup); 23611da177e4SLinus Torvalds #endif 2362