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> 564ee60ec1SMatthew Wilcox (Oracle) #include <linux/pagemap.h> 571da177e4SLinus Torvalds #include <linux/file.h> 581da177e4SLinus Torvalds #include <linux/stat.h> 591da177e4SLinus Torvalds #include <linux/errno.h> 601da177e4SLinus Torvalds #include <linux/major.h> 611da177e4SLinus Torvalds #include <linux/wait.h> 621da177e4SLinus Torvalds #include <linux/blkdev.h> 631da177e4SLinus Torvalds #include <linux/blkpg.h> 641da177e4SLinus Torvalds #include <linux/init.h> 651da177e4SLinus Torvalds #include <linux/swap.h> 661da177e4SLinus Torvalds #include <linux/slab.h> 67863d5b82SDavid Howells #include <linux/compat.h> 681da177e4SLinus Torvalds #include <linux/suspend.h> 6983144186SRafael J. Wysocki #include <linux/freezer.h> 702a48fc0aSArnd Bergmann #include <linux/mutex.h> 711da177e4SLinus Torvalds #include <linux/writeback.h> 721da177e4SLinus Torvalds #include <linux/completion.h> 731da177e4SLinus Torvalds #include <linux/highmem.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> 81c74d40e8SDan Schatzberg #include <linux/sched/mm.h> 8206582bc8SMing Lei #include <linux/statfs.h> 83d9a08a9eSAdam Manzanares 8483a87611SAl Viro #include "loop.h" 851da177e4SLinus Torvalds 867c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 871da177e4SLinus Torvalds 8887579e9bSDan Schatzberg #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ) 89ef44c508SChaitanya Kulkarni #define LOOP_DEFAULT_HW_Q_DEPTH (128) 9087579e9bSDan Schatzberg 9134dd82afSKay Sievers static DEFINE_IDR(loop_index_idr); 92310ca162STetsuo Handa static DEFINE_MUTEX(loop_ctl_mutex); 933ce6e1f6STetsuo Handa static DEFINE_MUTEX(loop_validate_mutex); 943ce6e1f6STetsuo Handa 953ce6e1f6STetsuo Handa /** 963ce6e1f6STetsuo Handa * loop_global_lock_killable() - take locks for safe loop_validate_file() test 973ce6e1f6STetsuo Handa * 983ce6e1f6STetsuo Handa * @lo: struct loop_device 993ce6e1f6STetsuo Handa * @global: true if @lo is about to bind another "struct loop_device", false otherwise 1003ce6e1f6STetsuo Handa * 1013ce6e1f6STetsuo Handa * Returns 0 on success, -EINTR otherwise. 1023ce6e1f6STetsuo Handa * 1033ce6e1f6STetsuo Handa * Since loop_validate_file() traverses on other "struct loop_device" if 1043ce6e1f6STetsuo Handa * is_loop_device() is true, we need a global lock for serializing concurrent 1053ce6e1f6STetsuo Handa * loop_configure()/loop_change_fd()/__loop_clr_fd() calls. 1063ce6e1f6STetsuo Handa */ 1073ce6e1f6STetsuo Handa static int loop_global_lock_killable(struct loop_device *lo, bool global) 1083ce6e1f6STetsuo Handa { 1093ce6e1f6STetsuo Handa int err; 1103ce6e1f6STetsuo Handa 1113ce6e1f6STetsuo Handa if (global) { 1123ce6e1f6STetsuo Handa err = mutex_lock_killable(&loop_validate_mutex); 1133ce6e1f6STetsuo Handa if (err) 1143ce6e1f6STetsuo Handa return err; 1153ce6e1f6STetsuo Handa } 1163ce6e1f6STetsuo Handa err = mutex_lock_killable(&lo->lo_mutex); 1173ce6e1f6STetsuo Handa if (err && global) 1183ce6e1f6STetsuo Handa mutex_unlock(&loop_validate_mutex); 1193ce6e1f6STetsuo Handa return err; 1203ce6e1f6STetsuo Handa } 1213ce6e1f6STetsuo Handa 1223ce6e1f6STetsuo Handa /** 1233ce6e1f6STetsuo Handa * loop_global_unlock() - release locks taken by loop_global_lock_killable() 1243ce6e1f6STetsuo Handa * 1253ce6e1f6STetsuo Handa * @lo: struct loop_device 1263ce6e1f6STetsuo Handa * @global: true if @lo was about to bind another "struct loop_device", false otherwise 1273ce6e1f6STetsuo Handa */ 1283ce6e1f6STetsuo Handa static void loop_global_unlock(struct loop_device *lo, bool global) 1293ce6e1f6STetsuo Handa { 1303ce6e1f6STetsuo Handa mutex_unlock(&lo->lo_mutex); 1313ce6e1f6STetsuo Handa if (global) 1323ce6e1f6STetsuo Handa mutex_unlock(&loop_validate_mutex); 1333ce6e1f6STetsuo Handa } 1341da177e4SLinus Torvalds 135476a4813SLaurent Vivier static int max_part; 136476a4813SLaurent Vivier static int part_shift; 137476a4813SLaurent Vivier 1387035b5dfSDmitry Monakhov static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file) 1391da177e4SLinus Torvalds { 140b7a1da69SGuo Chao loff_t loopsize; 1411da177e4SLinus Torvalds 1421da177e4SLinus Torvalds /* Compute loopsize in bytes */ 143b7a1da69SGuo Chao loopsize = i_size_read(file->f_mapping->host); 144b7a1da69SGuo Chao if (offset > 0) 145b7a1da69SGuo Chao loopsize -= offset; 146b7a1da69SGuo Chao /* offset is beyond i_size, weird but possible */ 1477035b5dfSDmitry Monakhov if (loopsize < 0) 1487035b5dfSDmitry Monakhov return 0; 1491da177e4SLinus Torvalds 1507035b5dfSDmitry Monakhov if (sizelimit > 0 && sizelimit < loopsize) 1517035b5dfSDmitry Monakhov loopsize = sizelimit; 1521da177e4SLinus Torvalds /* 1531da177e4SLinus Torvalds * Unfortunately, if we want to do I/O on the device, 1541da177e4SLinus Torvalds * the number of 512-byte sectors has to fit into a sector_t. 1551da177e4SLinus Torvalds */ 1561da177e4SLinus Torvalds return loopsize >> 9; 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 1597035b5dfSDmitry Monakhov static loff_t get_loop_size(struct loop_device *lo, struct file *file) 1601da177e4SLinus Torvalds { 1617035b5dfSDmitry Monakhov return get_size(lo->lo_offset, lo->lo_sizelimit, file); 1627035b5dfSDmitry Monakhov } 1637035b5dfSDmitry Monakhov 1642e5ab5f3SMing Lei static void __loop_update_dio(struct loop_device *lo, bool dio) 1652e5ab5f3SMing Lei { 1662e5ab5f3SMing Lei struct file *file = lo->lo_backing_file; 1672e5ab5f3SMing Lei struct address_space *mapping = file->f_mapping; 1682e5ab5f3SMing Lei struct inode *inode = mapping->host; 1692e5ab5f3SMing Lei unsigned short sb_bsize = 0; 1702e5ab5f3SMing Lei unsigned dio_align = 0; 1712e5ab5f3SMing Lei bool use_dio; 1722e5ab5f3SMing Lei 1732e5ab5f3SMing Lei if (inode->i_sb->s_bdev) { 1742e5ab5f3SMing Lei sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 1752e5ab5f3SMing Lei dio_align = sb_bsize - 1; 1762e5ab5f3SMing Lei } 1772e5ab5f3SMing Lei 1782e5ab5f3SMing Lei /* 1792e5ab5f3SMing Lei * We support direct I/O only if lo_offset is aligned with the 1802e5ab5f3SMing Lei * logical I/O size of backing device, and the logical block 18147e96246SChristoph Hellwig * size of loop is bigger than the backing device's. 1822e5ab5f3SMing Lei * 1832e5ab5f3SMing Lei * TODO: the above condition may be loosed in the future, and 1842e5ab5f3SMing Lei * direct I/O may be switched runtime at that time because most 18589d790abSMasahiro Yamada * of requests in sane applications should be PAGE_SIZE aligned 1862e5ab5f3SMing Lei */ 1872e5ab5f3SMing Lei if (dio) { 1882e5ab5f3SMing Lei if (queue_logical_block_size(lo->lo_queue) >= sb_bsize && 1892e5ab5f3SMing Lei !(lo->lo_offset & dio_align) && 19047e96246SChristoph Hellwig mapping->a_ops->direct_IO) 1912e5ab5f3SMing Lei use_dio = true; 1922e5ab5f3SMing Lei else 1932e5ab5f3SMing Lei use_dio = false; 1942e5ab5f3SMing Lei } else { 1952e5ab5f3SMing Lei use_dio = false; 1962e5ab5f3SMing Lei } 1972e5ab5f3SMing Lei 1982e5ab5f3SMing Lei if (lo->use_dio == use_dio) 1992e5ab5f3SMing Lei return; 2002e5ab5f3SMing Lei 2012e5ab5f3SMing Lei /* flush dirty pages before changing direct IO */ 2022e5ab5f3SMing Lei vfs_fsync(file, 0); 2032e5ab5f3SMing Lei 2042e5ab5f3SMing Lei /* 2052e5ab5f3SMing Lei * The flag of LO_FLAGS_DIRECT_IO is handled similarly with 2062e5ab5f3SMing Lei * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup 2072e5ab5f3SMing Lei * will get updated by ioctl(LOOP_GET_STATUS) 2082e5ab5f3SMing Lei */ 2090fbcf579SMartijn Coenen if (lo->lo_state == Lo_bound) 2102e5ab5f3SMing Lei blk_mq_freeze_queue(lo->lo_queue); 2112e5ab5f3SMing Lei lo->use_dio = use_dio; 21240326d8aSShaohua Li if (use_dio) { 2138b904b5bSBart Van Assche blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2142e5ab5f3SMing Lei lo->lo_flags |= LO_FLAGS_DIRECT_IO; 21540326d8aSShaohua Li } else { 2168b904b5bSBart Van Assche blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 2172e5ab5f3SMing Lei lo->lo_flags &= ~LO_FLAGS_DIRECT_IO; 21840326d8aSShaohua Li } 2190fbcf579SMartijn Coenen if (lo->lo_state == Lo_bound) 2202e5ab5f3SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 2212e5ab5f3SMing Lei } 2222e5ab5f3SMing Lei 2235795b6f5SMartijn Coenen /** 2245795b6f5SMartijn Coenen * loop_set_size() - sets device size and notifies userspace 2255795b6f5SMartijn Coenen * @lo: struct loop_device to set the size for 2265795b6f5SMartijn Coenen * @size: new size of the loop device 2275795b6f5SMartijn Coenen * 2285795b6f5SMartijn Coenen * Callers must validate that the size passed into this function fits into 2295795b6f5SMartijn Coenen * a sector_t, eg using loop_validate_size() 2305795b6f5SMartijn Coenen */ 2315795b6f5SMartijn Coenen static void loop_set_size(struct loop_device *lo, loff_t size) 2325795b6f5SMartijn Coenen { 233449f4ec9SChristoph Hellwig if (!set_capacity_and_notify(lo->lo_disk, size)) 2343b4f85d0SChristoph Hellwig kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 2351da177e4SLinus Torvalds } 2361da177e4SLinus Torvalds 237aa4d8616SChristoph Hellwig static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos) 2381da177e4SLinus Torvalds { 239aa4d8616SChristoph Hellwig struct iov_iter i; 2401da177e4SLinus Torvalds ssize_t bw; 241283e7e5dSAl Viro 242b6207430SChristoph Hellwig iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len); 2431da177e4SLinus Torvalds 24403d95eb2SAl Viro file_start_write(file); 245abbb6589SChristoph Hellwig bw = vfs_iter_write(file, &i, ppos, 0); 24603d95eb2SAl Viro file_end_write(file); 247aa4d8616SChristoph Hellwig 248aa4d8616SChristoph Hellwig if (likely(bw == bvec->bv_len)) 2491da177e4SLinus Torvalds return 0; 250aa4d8616SChristoph Hellwig 251aa4d8616SChristoph Hellwig printk_ratelimited(KERN_ERR 252aa4d8616SChristoph Hellwig "loop: Write error at byte offset %llu, length %i.\n", 253aa4d8616SChristoph Hellwig (unsigned long long)*ppos, bvec->bv_len); 2541da177e4SLinus Torvalds if (bw >= 0) 2551da177e4SLinus Torvalds bw = -EIO; 2561da177e4SLinus Torvalds return bw; 2571da177e4SLinus Torvalds } 2581da177e4SLinus Torvalds 259aa4d8616SChristoph Hellwig static int lo_write_simple(struct loop_device *lo, struct request *rq, 260aa4d8616SChristoph Hellwig loff_t pos) 2611da177e4SLinus Torvalds { 262aa4d8616SChristoph Hellwig struct bio_vec bvec; 263aa4d8616SChristoph Hellwig struct req_iterator iter; 264aa4d8616SChristoph Hellwig int ret = 0; 265aa4d8616SChristoph Hellwig 266aa4d8616SChristoph Hellwig rq_for_each_segment(bvec, rq, iter) { 267aa4d8616SChristoph Hellwig ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos); 268aa4d8616SChristoph Hellwig if (ret < 0) 269aa4d8616SChristoph Hellwig break; 2701da177e4SLinus Torvalds cond_resched(); 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 273aa4d8616SChristoph Hellwig return ret; 274aa4d8616SChristoph Hellwig } 275aa4d8616SChristoph Hellwig 276aa4d8616SChristoph Hellwig static int lo_read_simple(struct loop_device *lo, struct request *rq, 277aa4d8616SChristoph Hellwig loff_t pos) 2781da177e4SLinus Torvalds { 2797988613bSKent Overstreet struct bio_vec bvec; 28030112013SMing Lei struct req_iterator iter; 281aa4d8616SChristoph Hellwig struct iov_iter i; 282aa4d8616SChristoph Hellwig ssize_t len; 2831da177e4SLinus Torvalds 28430112013SMing Lei rq_for_each_segment(bvec, rq, iter) { 285b6207430SChristoph Hellwig iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len); 28618e9710eSChristoph Hellwig len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0); 287aa4d8616SChristoph Hellwig if (len < 0) 288aa4d8616SChristoph Hellwig return len; 289306df071SDave Young 290aa4d8616SChristoph Hellwig flush_dcache_page(bvec.bv_page); 291aa4d8616SChristoph Hellwig 292aa4d8616SChristoph Hellwig if (len != bvec.bv_len) { 29330112013SMing Lei struct bio *bio; 29430112013SMing Lei 29530112013SMing Lei __rq_for_each_bio(bio, rq) 296306df071SDave Young zero_fill_bio(bio); 2971da177e4SLinus Torvalds break; 298306df071SDave Young } 299aa4d8616SChristoph Hellwig cond_resched(); 3001da177e4SLinus Torvalds } 301aa4d8616SChristoph Hellwig 302306df071SDave Young return 0; 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds 305efcfec57SDarrick J. Wong static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos, 306efcfec57SDarrick J. Wong int mode) 307cf655d95SMing Lei { 308cf655d95SMing Lei /* 309efcfec57SDarrick J. Wong * We use fallocate to manipulate the space mappings used by the image 31047e96246SChristoph Hellwig * a.k.a. discard/zerorange. 311cf655d95SMing Lei */ 312cf655d95SMing Lei struct file *file = lo->lo_backing_file; 313cf655d95SMing Lei int ret; 314cf655d95SMing Lei 315efcfec57SDarrick J. Wong mode |= FALLOC_FL_KEEP_SIZE; 316efcfec57SDarrick J. Wong 31770200574SChristoph Hellwig if (!bdev_max_discard_sectors(lo->lo_device)) 31870200574SChristoph Hellwig return -EOPNOTSUPP; 319cf655d95SMing Lei 320cf655d95SMing Lei ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq)); 321cf655d95SMing Lei if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP)) 32270200574SChristoph Hellwig return -EIO; 323cf655d95SMing Lei return ret; 324cf655d95SMing Lei } 325cf655d95SMing Lei 326cf655d95SMing Lei static int lo_req_flush(struct loop_device *lo, struct request *rq) 327cf655d95SMing Lei { 3289c64e38cSChaitanya Kulkarni int ret = vfs_fsync(lo->lo_backing_file, 0); 329cf655d95SMing Lei if (unlikely(ret && ret != -EINVAL)) 330cf655d95SMing Lei ret = -EIO; 331cf655d95SMing Lei 332cf655d95SMing Lei return ret; 333cf655d95SMing Lei } 334cf655d95SMing Lei 335fe2cb290SChristoph Hellwig static void lo_complete_rq(struct request *rq) 336bc07c10aSMing Lei { 337fe2cb290SChristoph Hellwig struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 338f9de14bcSJens Axboe blk_status_t ret = BLK_STS_OK; 339bc07c10aSMing Lei 340f9de14bcSJens Axboe if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) || 341f9de14bcSJens Axboe req_op(rq) != REQ_OP_READ) { 342f9de14bcSJens Axboe if (cmd->ret < 0) 3438cd55087SEvan Green ret = errno_to_blk_status(cmd->ret); 344f9de14bcSJens Axboe goto end_io; 345bc07c10aSMing Lei } 346fe2cb290SChristoph Hellwig 347f9de14bcSJens Axboe /* 348f9de14bcSJens Axboe * Short READ - if we got some data, advance our request and 349f9de14bcSJens Axboe * retry it. If we got no data, end the rest with EIO. 350f9de14bcSJens Axboe */ 351f9de14bcSJens Axboe if (cmd->ret) { 352f9de14bcSJens Axboe blk_update_request(rq, BLK_STS_OK, cmd->ret); 353f9de14bcSJens Axboe cmd->ret = 0; 354f9de14bcSJens Axboe blk_mq_requeue_request(rq, true); 355f9de14bcSJens Axboe } else { 356f9de14bcSJens Axboe if (cmd->use_aio) { 357f9de14bcSJens Axboe struct bio *bio = rq->bio; 358f9de14bcSJens Axboe 359f9de14bcSJens Axboe while (bio) { 360f9de14bcSJens Axboe zero_fill_bio(bio); 361f9de14bcSJens Axboe bio = bio->bi_next; 362f9de14bcSJens Axboe } 363f9de14bcSJens Axboe } 364f9de14bcSJens Axboe ret = BLK_STS_IOERR; 365f9de14bcSJens Axboe end_io: 366f9de14bcSJens Axboe blk_mq_end_request(rq, ret); 367f9de14bcSJens Axboe } 368bc07c10aSMing Lei } 369bc07c10aSMing Lei 37092d77332SShaohua Li static void lo_rw_aio_do_completion(struct loop_cmd *cmd) 37192d77332SShaohua Li { 3721894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 3731894e916SJens Axboe 37492d77332SShaohua Li if (!atomic_dec_and_test(&cmd->ref)) 37592d77332SShaohua Li return; 37692d77332SShaohua Li kfree(cmd->bvec); 37792d77332SShaohua Li cmd->bvec = NULL; 37815f73f5bSChristoph Hellwig if (likely(!blk_should_fake_timeout(rq->q))) 3791894e916SJens Axboe blk_mq_complete_request(rq); 38092d77332SShaohua Li } 38192d77332SShaohua Li 3826b19b766SJens Axboe static void lo_rw_aio_complete(struct kiocb *iocb, long ret) 383bc07c10aSMing Lei { 384bc07c10aSMing Lei struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb); 385bc07c10aSMing Lei 386fe2cb290SChristoph Hellwig cmd->ret = ret; 38792d77332SShaohua Li lo_rw_aio_do_completion(cmd); 388bc07c10aSMing Lei } 389bc07c10aSMing Lei 390bc07c10aSMing Lei static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd, 391bc07c10aSMing Lei loff_t pos, bool rw) 392bc07c10aSMing Lei { 393bc07c10aSMing Lei struct iov_iter iter; 39486af5952SMing Lei struct req_iterator rq_iter; 395bc07c10aSMing Lei struct bio_vec *bvec; 3961894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 39740326d8aSShaohua Li struct bio *bio = rq->bio; 398bc07c10aSMing Lei struct file *file = lo->lo_backing_file; 39986af5952SMing Lei struct bio_vec tmp; 40040326d8aSShaohua Li unsigned int offset; 40186af5952SMing Lei int nr_bvec = 0; 402bc07c10aSMing Lei int ret; 403bc07c10aSMing Lei 40486af5952SMing Lei rq_for_each_bvec(tmp, rq, rq_iter) 40586af5952SMing Lei nr_bvec++; 406bc07c10aSMing Lei 40786af5952SMing Lei if (rq->bio != rq->biotail) { 40886af5952SMing Lei 40986af5952SMing Lei bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec), 4106da2ec56SKees Cook GFP_NOIO); 41140326d8aSShaohua Li if (!bvec) 41240326d8aSShaohua Li return -EIO; 41340326d8aSShaohua Li cmd->bvec = bvec; 41440326d8aSShaohua Li 415a7297a6aSMing Lei /* 41640326d8aSShaohua Li * The bios of the request may be started from the middle of 41740326d8aSShaohua Li * the 'bvec' because of bio splitting, so we can't directly 41886af5952SMing Lei * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec 41940326d8aSShaohua Li * API will take care of all details for us. 420a7297a6aSMing Lei */ 42186af5952SMing Lei rq_for_each_bvec(tmp, rq, rq_iter) { 42240326d8aSShaohua Li *bvec = tmp; 42340326d8aSShaohua Li bvec++; 42440326d8aSShaohua Li } 42540326d8aSShaohua Li bvec = cmd->bvec; 42640326d8aSShaohua Li offset = 0; 42740326d8aSShaohua Li } else { 42840326d8aSShaohua Li /* 42940326d8aSShaohua Li * Same here, this bio may be started from the middle of the 43040326d8aSShaohua Li * 'bvec' because of bio splitting, so offset from the bvec 43140326d8aSShaohua Li * must be passed to iov iterator 43240326d8aSShaohua Li */ 43340326d8aSShaohua Li offset = bio->bi_iter.bi_bvec_done; 43440326d8aSShaohua Li bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter); 43540326d8aSShaohua Li } 43692d77332SShaohua Li atomic_set(&cmd->ref, 2); 43740326d8aSShaohua Li 438b6207430SChristoph Hellwig iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq)); 43940326d8aSShaohua Li iter.iov_offset = offset; 440bc07c10aSMing Lei 441bc07c10aSMing Lei cmd->iocb.ki_pos = pos; 442bc07c10aSMing Lei cmd->iocb.ki_filp = file; 443bc07c10aSMing Lei cmd->iocb.ki_complete = lo_rw_aio_complete; 444bc07c10aSMing Lei cmd->iocb.ki_flags = IOCB_DIRECT; 445d9a08a9eSAdam Manzanares cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0); 446bc07c10aSMing Lei 447bc07c10aSMing Lei if (rw == WRITE) 448bb7462b6SMiklos Szeredi ret = call_write_iter(file, &cmd->iocb, &iter); 449bc07c10aSMing Lei else 450bb7462b6SMiklos Szeredi ret = call_read_iter(file, &cmd->iocb, &iter); 451bc07c10aSMing Lei 45292d77332SShaohua Li lo_rw_aio_do_completion(cmd); 45392d77332SShaohua Li 454bc07c10aSMing Lei if (ret != -EIOCBQUEUED) 4556b19b766SJens Axboe lo_rw_aio_complete(&cmd->iocb, ret); 456bc07c10aSMing Lei return 0; 457bc07c10aSMing Lei } 458bc07c10aSMing Lei 459c1c87c2bSChristoph Hellwig static int do_req_filebacked(struct loop_device *lo, struct request *rq) 460bc07c10aSMing Lei { 461bc07c10aSMing Lei struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 462c1c87c2bSChristoph Hellwig loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset; 463bc07c10aSMing Lei 464bc07c10aSMing Lei /* 465bc07c10aSMing Lei * lo_write_simple and lo_read_simple should have been covered 466bc07c10aSMing Lei * by io submit style function like lo_rw_aio(), one blocker 467bc07c10aSMing Lei * is that lo_read_simple() need to call flush_dcache_page after 468bc07c10aSMing Lei * the page is written from kernel, and it isn't easy to handle 469bc07c10aSMing Lei * this in io submit style function which submits all segments 470bc07c10aSMing Lei * of the req at one time. And direct read IO doesn't need to 471bc07c10aSMing Lei * run flush_dcache_page(). 472bc07c10aSMing Lei */ 473c1c87c2bSChristoph Hellwig switch (req_op(rq)) { 474c1c87c2bSChristoph Hellwig case REQ_OP_FLUSH: 475c1c87c2bSChristoph Hellwig return lo_req_flush(lo, rq); 47619372e27SChristoph Hellwig case REQ_OP_WRITE_ZEROES: 477efcfec57SDarrick J. Wong /* 478efcfec57SDarrick J. Wong * If the caller doesn't want deallocation, call zeroout to 479efcfec57SDarrick J. Wong * write zeroes the range. Otherwise, punch them out. 480efcfec57SDarrick J. Wong */ 481efcfec57SDarrick J. Wong return lo_fallocate(lo, rq, pos, 482efcfec57SDarrick J. Wong (rq->cmd_flags & REQ_NOUNMAP) ? 483efcfec57SDarrick J. Wong FALLOC_FL_ZERO_RANGE : 484efcfec57SDarrick J. Wong FALLOC_FL_PUNCH_HOLE); 485efcfec57SDarrick J. Wong case REQ_OP_DISCARD: 486efcfec57SDarrick J. Wong return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE); 487c1c87c2bSChristoph Hellwig case REQ_OP_WRITE: 48847e96246SChristoph Hellwig if (cmd->use_aio) 489c1c87c2bSChristoph Hellwig return lo_rw_aio(lo, cmd, pos, WRITE); 490c1c87c2bSChristoph Hellwig else 491bc07c10aSMing Lei return lo_write_simple(lo, rq, pos); 492c1c87c2bSChristoph Hellwig case REQ_OP_READ: 49347e96246SChristoph Hellwig if (cmd->use_aio) 494c1c87c2bSChristoph Hellwig return lo_rw_aio(lo, cmd, pos, READ); 495bc07c10aSMing Lei else 496bc07c10aSMing Lei return lo_read_simple(lo, rq, pos); 497c1c87c2bSChristoph Hellwig default: 498c1c87c2bSChristoph Hellwig WARN_ON_ONCE(1); 499c1c87c2bSChristoph Hellwig return -EIO; 500bc07c10aSMing Lei } 5011da177e4SLinus Torvalds } 5021da177e4SLinus Torvalds 5032e5ab5f3SMing Lei static inline void loop_update_dio(struct loop_device *lo) 5042e5ab5f3SMing Lei { 505efbe3c24SIra Weiny __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) | 5062e5ab5f3SMing Lei lo->use_dio); 5072e5ab5f3SMing Lei } 5082e5ab5f3SMing Lei 5090384264eSChristoph Hellwig static void loop_reread_partitions(struct loop_device *lo) 51006f0e9e6SMing Lei { 51106f0e9e6SMing Lei int rc; 51206f0e9e6SMing Lei 5130384264eSChristoph Hellwig mutex_lock(&lo->lo_disk->open_mutex); 5140384264eSChristoph Hellwig rc = bdev_disk_changed(lo->lo_disk, false); 5150384264eSChristoph Hellwig mutex_unlock(&lo->lo_disk->open_mutex); 51606f0e9e6SMing Lei if (rc) 51706f0e9e6SMing Lei pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n", 51806f0e9e6SMing Lei __func__, lo->lo_number, lo->lo_file_name, rc); 51906f0e9e6SMing Lei } 52006f0e9e6SMing Lei 521d2ac838eSTheodore Ts'o static inline int is_loop_device(struct file *file) 522d2ac838eSTheodore Ts'o { 523d2ac838eSTheodore Ts'o struct inode *i = file->f_mapping->host; 524d2ac838eSTheodore Ts'o 5256f24784fSAl Viro return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR; 526d2ac838eSTheodore Ts'o } 527d2ac838eSTheodore Ts'o 528d2ac838eSTheodore Ts'o static int loop_validate_file(struct file *file, struct block_device *bdev) 529d2ac838eSTheodore Ts'o { 530d2ac838eSTheodore Ts'o struct inode *inode = file->f_mapping->host; 531d2ac838eSTheodore Ts'o struct file *f = file; 532d2ac838eSTheodore Ts'o 533d2ac838eSTheodore Ts'o /* Avoid recursion */ 534d2ac838eSTheodore Ts'o while (is_loop_device(f)) { 535d2ac838eSTheodore Ts'o struct loop_device *l; 536d2ac838eSTheodore Ts'o 5373ce6e1f6STetsuo Handa lockdep_assert_held(&loop_validate_mutex); 5384e7b5671SChristoph Hellwig if (f->f_mapping->host->i_rdev == bdev->bd_dev) 539d2ac838eSTheodore Ts'o return -EBADF; 540d2ac838eSTheodore Ts'o 5414e7b5671SChristoph Hellwig l = I_BDEV(f->f_mapping->host)->bd_disk->private_data; 5423ce6e1f6STetsuo Handa if (l->lo_state != Lo_bound) 543d2ac838eSTheodore Ts'o return -EINVAL; 5443ce6e1f6STetsuo Handa /* Order wrt setting lo->lo_backing_file in loop_configure(). */ 5453ce6e1f6STetsuo Handa rmb(); 546d2ac838eSTheodore Ts'o f = l->lo_backing_file; 547d2ac838eSTheodore Ts'o } 548d2ac838eSTheodore Ts'o if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode)) 549d2ac838eSTheodore Ts'o return -EINVAL; 550d2ac838eSTheodore Ts'o return 0; 551d2ac838eSTheodore Ts'o } 552d2ac838eSTheodore Ts'o 5531da177e4SLinus Torvalds /* 5541da177e4SLinus Torvalds * loop_change_fd switched the backing store of a loopback device to 5551da177e4SLinus Torvalds * a new file. This is useful for operating system installers to free up 5561da177e4SLinus Torvalds * the original file and in High Availability environments to switch to 5571da177e4SLinus Torvalds * an alternative location for the content in case of server meltdown. 5581da177e4SLinus Torvalds * This can only work if the loop device is used read-only, and if the 5591da177e4SLinus Torvalds * new backing store is the same size and type as the old backing store. 5601da177e4SLinus Torvalds */ 561bb214884SAl Viro static int loop_change_fd(struct loop_device *lo, struct block_device *bdev, 562bb214884SAl Viro unsigned int arg) 5631da177e4SLinus Torvalds { 5643ce6e1f6STetsuo Handa struct file *file = fget(arg); 5653ce6e1f6STetsuo Handa struct file *old_file; 5661da177e4SLinus Torvalds int error; 56785b0a54aSJan Kara bool partscan; 5683ce6e1f6STetsuo Handa bool is_loop; 5691da177e4SLinus Torvalds 5703ce6e1f6STetsuo Handa if (!file) 5713ce6e1f6STetsuo Handa return -EBADF; 5723ce6e1f6STetsuo Handa is_loop = is_loop_device(file); 5733ce6e1f6STetsuo Handa error = loop_global_lock_killable(lo, is_loop); 574c3710770SJan Kara if (error) 5753ce6e1f6STetsuo Handa goto out_putf; 5761da177e4SLinus Torvalds error = -ENXIO; 5771da177e4SLinus Torvalds if (lo->lo_state != Lo_bound) 5781dded9acSJan Kara goto out_err; 5791da177e4SLinus Torvalds 5801da177e4SLinus Torvalds /* the loop device has to be read-only */ 5811da177e4SLinus Torvalds error = -EINVAL; 5821da177e4SLinus Torvalds if (!(lo->lo_flags & LO_FLAGS_READ_ONLY)) 5831dded9acSJan Kara goto out_err; 5841da177e4SLinus Torvalds 585d2ac838eSTheodore Ts'o error = loop_validate_file(file, bdev); 586d2ac838eSTheodore Ts'o if (error) 5871dded9acSJan Kara goto out_err; 588d2ac838eSTheodore Ts'o 5891da177e4SLinus Torvalds old_file = lo->lo_backing_file; 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds error = -EINVAL; 5921da177e4SLinus Torvalds 5931da177e4SLinus Torvalds /* size of the new backing store needs to be the same */ 5941da177e4SLinus Torvalds if (get_loop_size(lo, file) != get_loop_size(lo, old_file)) 5951dded9acSJan Kara goto out_err; 5961da177e4SLinus Torvalds 5971da177e4SLinus Torvalds /* and ... switch */ 5989f65c489SMatteo Croce disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 59943cade80SOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 60043cade80SOmar Sandoval mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask); 60143cade80SOmar Sandoval lo->lo_backing_file = file; 60243cade80SOmar Sandoval lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping); 60343cade80SOmar Sandoval mapping_set_gfp_mask(file->f_mapping, 60443cade80SOmar Sandoval lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 60543cade80SOmar Sandoval loop_update_dio(lo); 60643cade80SOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 60785b0a54aSJan Kara partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 6083ce6e1f6STetsuo Handa loop_global_unlock(lo, is_loop); 6093ce6e1f6STetsuo Handa 6103ce6e1f6STetsuo Handa /* 6113ce6e1f6STetsuo Handa * Flush loop_validate_file() before fput(), for l->lo_backing_file 6123ce6e1f6STetsuo Handa * might be pointing at old_file which might be the last reference. 6133ce6e1f6STetsuo Handa */ 6143ce6e1f6STetsuo Handa if (!is_loop) { 6153ce6e1f6STetsuo Handa mutex_lock(&loop_validate_mutex); 6163ce6e1f6STetsuo Handa mutex_unlock(&loop_validate_mutex); 6173ce6e1f6STetsuo Handa } 6181dded9acSJan Kara /* 6196cc8e743SPavel Tatashin * We must drop file reference outside of lo_mutex as dropping 620a8698707SChristoph Hellwig * the file ref can take open_mutex which creates circular locking 6211dded9acSJan Kara * dependency. 6221dded9acSJan Kara */ 6231dded9acSJan Kara fput(old_file); 62485b0a54aSJan Kara if (partscan) 6250384264eSChristoph Hellwig loop_reread_partitions(lo); 6261da177e4SLinus Torvalds return 0; 6271da177e4SLinus Torvalds 6281dded9acSJan Kara out_err: 6293ce6e1f6STetsuo Handa loop_global_unlock(lo, is_loop); 6303ce6e1f6STetsuo Handa out_putf: 6311dded9acSJan Kara fput(file); 6321da177e4SLinus Torvalds return error; 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds 635ee862730SMilan Broz /* loop sysfs attributes */ 636ee862730SMilan Broz 637ee862730SMilan Broz static ssize_t loop_attr_show(struct device *dev, char *page, 638ee862730SMilan Broz ssize_t (*callback)(struct loop_device *, char *)) 639ee862730SMilan Broz { 64034dd82afSKay Sievers struct gendisk *disk = dev_to_disk(dev); 64134dd82afSKay Sievers struct loop_device *lo = disk->private_data; 642ee862730SMilan Broz 64334dd82afSKay Sievers return callback(lo, page); 644ee862730SMilan Broz } 645ee862730SMilan Broz 646ee862730SMilan Broz #define LOOP_ATTR_RO(_name) \ 647ee862730SMilan Broz static ssize_t loop_attr_##_name##_show(struct loop_device *, char *); \ 648ee862730SMilan Broz static ssize_t loop_attr_do_show_##_name(struct device *d, \ 649ee862730SMilan Broz struct device_attribute *attr, char *b) \ 650ee862730SMilan Broz { \ 651ee862730SMilan Broz return loop_attr_show(d, b, loop_attr_##_name##_show); \ 652ee862730SMilan Broz } \ 653ee862730SMilan Broz static struct device_attribute loop_attr_##_name = \ 6545657a819SJoe Perches __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL); 655ee862730SMilan Broz 656ee862730SMilan Broz static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf) 657ee862730SMilan Broz { 658ee862730SMilan Broz ssize_t ret; 659ee862730SMilan Broz char *p = NULL; 660ee862730SMilan Broz 66105eb0f25SKay Sievers spin_lock_irq(&lo->lo_lock); 662ee862730SMilan Broz if (lo->lo_backing_file) 6639bf39ab2SMiklos Szeredi p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1); 66405eb0f25SKay Sievers spin_unlock_irq(&lo->lo_lock); 665ee862730SMilan Broz 666ee862730SMilan Broz if (IS_ERR_OR_NULL(p)) 667ee862730SMilan Broz ret = PTR_ERR(p); 668ee862730SMilan Broz else { 669ee862730SMilan Broz ret = strlen(p); 670ee862730SMilan Broz memmove(buf, p, ret); 671ee862730SMilan Broz buf[ret++] = '\n'; 672ee862730SMilan Broz buf[ret] = 0; 673ee862730SMilan Broz } 674ee862730SMilan Broz 675ee862730SMilan Broz return ret; 676ee862730SMilan Broz } 677ee862730SMilan Broz 678ee862730SMilan Broz static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf) 679ee862730SMilan Broz { 680b27824d3SChaitanya Kulkarni return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset); 681ee862730SMilan Broz } 682ee862730SMilan Broz 683ee862730SMilan Broz static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf) 684ee862730SMilan Broz { 685b27824d3SChaitanya Kulkarni return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit); 686ee862730SMilan Broz } 687ee862730SMilan Broz 688ee862730SMilan Broz static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf) 689ee862730SMilan Broz { 690ee862730SMilan Broz int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR); 691ee862730SMilan Broz 692b27824d3SChaitanya Kulkarni return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0"); 693ee862730SMilan Broz } 694ee862730SMilan Broz 695e03c8dd1SKay Sievers static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf) 696e03c8dd1SKay Sievers { 697e03c8dd1SKay Sievers int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN); 698e03c8dd1SKay Sievers 699b27824d3SChaitanya Kulkarni return sysfs_emit(buf, "%s\n", partscan ? "1" : "0"); 700e03c8dd1SKay Sievers } 701e03c8dd1SKay Sievers 7022e5ab5f3SMing Lei static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf) 7032e5ab5f3SMing Lei { 7042e5ab5f3SMing Lei int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO); 7052e5ab5f3SMing Lei 706b27824d3SChaitanya Kulkarni return sysfs_emit(buf, "%s\n", dio ? "1" : "0"); 7072e5ab5f3SMing Lei } 7082e5ab5f3SMing Lei 709ee862730SMilan Broz LOOP_ATTR_RO(backing_file); 710ee862730SMilan Broz LOOP_ATTR_RO(offset); 711ee862730SMilan Broz LOOP_ATTR_RO(sizelimit); 712ee862730SMilan Broz LOOP_ATTR_RO(autoclear); 713e03c8dd1SKay Sievers LOOP_ATTR_RO(partscan); 7142e5ab5f3SMing Lei LOOP_ATTR_RO(dio); 715ee862730SMilan Broz 716ee862730SMilan Broz static struct attribute *loop_attrs[] = { 717ee862730SMilan Broz &loop_attr_backing_file.attr, 718ee862730SMilan Broz &loop_attr_offset.attr, 719ee862730SMilan Broz &loop_attr_sizelimit.attr, 720ee862730SMilan Broz &loop_attr_autoclear.attr, 721e03c8dd1SKay Sievers &loop_attr_partscan.attr, 7222e5ab5f3SMing Lei &loop_attr_dio.attr, 723ee862730SMilan Broz NULL, 724ee862730SMilan Broz }; 725ee862730SMilan Broz 726ee862730SMilan Broz static struct attribute_group loop_attribute_group = { 727ee862730SMilan Broz .name = "loop", 728ee862730SMilan Broz .attrs= loop_attrs, 729ee862730SMilan Broz }; 730ee862730SMilan Broz 731d3349b6bSTetsuo Handa static void loop_sysfs_init(struct loop_device *lo) 732ee862730SMilan Broz { 733d3349b6bSTetsuo Handa lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj, 734ee862730SMilan Broz &loop_attribute_group); 735ee862730SMilan Broz } 736ee862730SMilan Broz 737ee862730SMilan Broz static void loop_sysfs_exit(struct loop_device *lo) 738ee862730SMilan Broz { 739d3349b6bSTetsuo Handa if (lo->sysfs_inited) 740ee862730SMilan Broz sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj, 741ee862730SMilan Broz &loop_attribute_group); 742ee862730SMilan Broz } 743ee862730SMilan Broz 744dfaa2ef6SLukas Czerner static void loop_config_discard(struct loop_device *lo) 745dfaa2ef6SLukas Czerner { 746dfaa2ef6SLukas Czerner struct file *file = lo->lo_backing_file; 747dfaa2ef6SLukas Czerner struct inode *inode = file->f_mapping->host; 748dfaa2ef6SLukas Czerner struct request_queue *q = lo->lo_queue; 749bcb21c8cSMing Lei u32 granularity, max_discard_sectors; 750dfaa2ef6SLukas Czerner 751dfaa2ef6SLukas Czerner /* 752c52abf56SEvan Green * If the backing device is a block device, mirror its zeroing 753c52abf56SEvan Green * capability. Set the discard sectors to the block device's zeroing 754c52abf56SEvan Green * capabilities because loop discards result in blkdev_issue_zeroout(), 755c52abf56SEvan Green * not blkdev_issue_discard(). This maintains consistent behavior with 756c52abf56SEvan Green * file-backed loop devices: discarded regions read back as zero. 757c52abf56SEvan Green */ 75847e96246SChristoph Hellwig if (S_ISBLK(inode->i_mode)) { 7594e7b5671SChristoph Hellwig struct request_queue *backingq = bdev_get_queue(I_BDEV(inode)); 760c52abf56SEvan Green 761bcb21c8cSMing Lei max_discard_sectors = backingq->limits.max_write_zeroes_sectors; 762*7b47ef52SChristoph Hellwig granularity = bdev_discard_granularity(I_BDEV(inode)) ?: 763bcb21c8cSMing Lei queue_physical_block_size(backingq); 764c52abf56SEvan Green 765c52abf56SEvan Green /* 766dfaa2ef6SLukas Czerner * We use punch hole to reclaim the free space used by the 76747e96246SChristoph Hellwig * image a.k.a. discard. 768dfaa2ef6SLukas Czerner */ 76947e96246SChristoph Hellwig } else if (!file->f_op->fallocate) { 770bcb21c8cSMing Lei max_discard_sectors = 0; 771bcb21c8cSMing Lei granularity = 0; 772dfaa2ef6SLukas Czerner 773c52abf56SEvan Green } else { 77406582bc8SMing Lei struct kstatfs sbuf; 77506582bc8SMing Lei 776bcb21c8cSMing Lei max_discard_sectors = UINT_MAX >> 9; 77706582bc8SMing Lei if (!vfs_statfs(&file->f_path, &sbuf)) 77806582bc8SMing Lei granularity = sbuf.f_bsize; 77906582bc8SMing Lei else 78006582bc8SMing Lei max_discard_sectors = 0; 781c52abf56SEvan Green } 782c52abf56SEvan Green 783bcb21c8cSMing Lei if (max_discard_sectors) { 784bcb21c8cSMing Lei q->limits.discard_granularity = granularity; 785bcb21c8cSMing Lei blk_queue_max_discard_sectors(q, max_discard_sectors); 786bcb21c8cSMing Lei blk_queue_max_write_zeroes_sectors(q, max_discard_sectors); 787bcb21c8cSMing Lei } else { 788bcb21c8cSMing Lei q->limits.discard_granularity = 0; 789bcb21c8cSMing Lei blk_queue_max_discard_sectors(q, 0); 790bcb21c8cSMing Lei blk_queue_max_write_zeroes_sectors(q, 0); 791dfaa2ef6SLukas Czerner } 792bcb21c8cSMing Lei q->limits.discard_alignment = 0; 793bcb21c8cSMing Lei } 794dfaa2ef6SLukas Czerner 79587579e9bSDan Schatzberg struct loop_worker { 79687579e9bSDan Schatzberg struct rb_node rb_node; 79787579e9bSDan Schatzberg struct work_struct work; 79887579e9bSDan Schatzberg struct list_head cmd_list; 79987579e9bSDan Schatzberg struct list_head idle_list; 80087579e9bSDan Schatzberg struct loop_device *lo; 801c74d40e8SDan Schatzberg struct cgroup_subsys_state *blkcg_css; 80287579e9bSDan Schatzberg unsigned long last_ran_at; 80387579e9bSDan Schatzberg }; 804e03a3d7aSMing Lei 80587579e9bSDan Schatzberg static void loop_workfn(struct work_struct *work); 80687579e9bSDan Schatzberg static void loop_rootcg_workfn(struct work_struct *work); 80787579e9bSDan Schatzberg static void loop_free_idle_workers(struct timer_list *timer); 808b2ee7d46SNeilBrown 80987579e9bSDan Schatzberg #ifdef CONFIG_BLK_CGROUP 81087579e9bSDan Schatzberg static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 811e03a3d7aSMing Lei { 81287579e9bSDan Schatzberg return !css || css == blkcg_root_css; 81387579e9bSDan Schatzberg } 81487579e9bSDan Schatzberg #else 81587579e9bSDan Schatzberg static inline int queue_on_root_worker(struct cgroup_subsys_state *css) 81687579e9bSDan Schatzberg { 81787579e9bSDan Schatzberg return !css; 81887579e9bSDan Schatzberg } 81987579e9bSDan Schatzberg #endif 82087579e9bSDan Schatzberg 82187579e9bSDan Schatzberg static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd) 82287579e9bSDan Schatzberg { 823413ec805SColin Ian King struct rb_node **node, *parent = NULL; 82487579e9bSDan Schatzberg struct loop_worker *cur_worker, *worker = NULL; 82587579e9bSDan Schatzberg struct work_struct *work; 82687579e9bSDan Schatzberg struct list_head *cmd_list; 82787579e9bSDan Schatzberg 82887579e9bSDan Schatzberg spin_lock_irq(&lo->lo_work_lock); 82987579e9bSDan Schatzberg 830c74d40e8SDan Schatzberg if (queue_on_root_worker(cmd->blkcg_css)) 83187579e9bSDan Schatzberg goto queue_work; 83287579e9bSDan Schatzberg 83387579e9bSDan Schatzberg node = &lo->worker_tree.rb_node; 83487579e9bSDan Schatzberg 83587579e9bSDan Schatzberg while (*node) { 83687579e9bSDan Schatzberg parent = *node; 83787579e9bSDan Schatzberg cur_worker = container_of(*node, struct loop_worker, rb_node); 838c74d40e8SDan Schatzberg if (cur_worker->blkcg_css == cmd->blkcg_css) { 83987579e9bSDan Schatzberg worker = cur_worker; 84087579e9bSDan Schatzberg break; 841c74d40e8SDan Schatzberg } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) { 84287579e9bSDan Schatzberg node = &(*node)->rb_left; 84387579e9bSDan Schatzberg } else { 84487579e9bSDan Schatzberg node = &(*node)->rb_right; 84587579e9bSDan Schatzberg } 84687579e9bSDan Schatzberg } 84787579e9bSDan Schatzberg if (worker) 84887579e9bSDan Schatzberg goto queue_work; 84987579e9bSDan Schatzberg 85087579e9bSDan Schatzberg worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN); 85187579e9bSDan Schatzberg /* 85287579e9bSDan Schatzberg * In the event we cannot allocate a worker, just queue on the 853c74d40e8SDan Schatzberg * rootcg worker and issue the I/O as the rootcg 85487579e9bSDan Schatzberg */ 855c74d40e8SDan Schatzberg if (!worker) { 856c74d40e8SDan Schatzberg cmd->blkcg_css = NULL; 857c74d40e8SDan Schatzberg if (cmd->memcg_css) 858c74d40e8SDan Schatzberg css_put(cmd->memcg_css); 859c74d40e8SDan Schatzberg cmd->memcg_css = NULL; 86087579e9bSDan Schatzberg goto queue_work; 861c74d40e8SDan Schatzberg } 86287579e9bSDan Schatzberg 863c74d40e8SDan Schatzberg worker->blkcg_css = cmd->blkcg_css; 864c74d40e8SDan Schatzberg css_get(worker->blkcg_css); 86587579e9bSDan Schatzberg INIT_WORK(&worker->work, loop_workfn); 86687579e9bSDan Schatzberg INIT_LIST_HEAD(&worker->cmd_list); 86787579e9bSDan Schatzberg INIT_LIST_HEAD(&worker->idle_list); 86887579e9bSDan Schatzberg worker->lo = lo; 86987579e9bSDan Schatzberg rb_link_node(&worker->rb_node, parent, node); 87087579e9bSDan Schatzberg rb_insert_color(&worker->rb_node, &lo->worker_tree); 87187579e9bSDan Schatzberg queue_work: 87287579e9bSDan Schatzberg if (worker) { 87387579e9bSDan Schatzberg /* 87487579e9bSDan Schatzberg * We need to remove from the idle list here while 87587579e9bSDan Schatzberg * holding the lock so that the idle timer doesn't 87687579e9bSDan Schatzberg * free the worker 87787579e9bSDan Schatzberg */ 87887579e9bSDan Schatzberg if (!list_empty(&worker->idle_list)) 87987579e9bSDan Schatzberg list_del_init(&worker->idle_list); 88087579e9bSDan Schatzberg work = &worker->work; 88187579e9bSDan Schatzberg cmd_list = &worker->cmd_list; 88287579e9bSDan Schatzberg } else { 88387579e9bSDan Schatzberg work = &lo->rootcg_work; 88487579e9bSDan Schatzberg cmd_list = &lo->rootcg_cmd_list; 88587579e9bSDan Schatzberg } 88687579e9bSDan Schatzberg list_add_tail(&cmd->list_entry, cmd_list); 88787579e9bSDan Schatzberg queue_work(lo->workqueue, work); 88887579e9bSDan Schatzberg spin_unlock_irq(&lo->lo_work_lock); 889e03a3d7aSMing Lei } 890e03a3d7aSMing Lei 89156a85fd8SHolger Hoffstätte static void loop_update_rotational(struct loop_device *lo) 89256a85fd8SHolger Hoffstätte { 89356a85fd8SHolger Hoffstätte struct file *file = lo->lo_backing_file; 89456a85fd8SHolger Hoffstätte struct inode *file_inode = file->f_mapping->host; 89556a85fd8SHolger Hoffstätte struct block_device *file_bdev = file_inode->i_sb->s_bdev; 89656a85fd8SHolger Hoffstätte struct request_queue *q = lo->lo_queue; 89756a85fd8SHolger Hoffstätte bool nonrot = true; 89856a85fd8SHolger Hoffstätte 89956a85fd8SHolger Hoffstätte /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */ 90056a85fd8SHolger Hoffstätte if (file_bdev) 90110f0d2a5SChristoph Hellwig nonrot = bdev_nonrot(file_bdev); 90256a85fd8SHolger Hoffstätte 90356a85fd8SHolger Hoffstätte if (nonrot) 90456a85fd8SHolger Hoffstätte blk_queue_flag_set(QUEUE_FLAG_NONROT, q); 90556a85fd8SHolger Hoffstätte else 90656a85fd8SHolger Hoffstätte blk_queue_flag_clear(QUEUE_FLAG_NONROT, q); 90756a85fd8SHolger Hoffstätte } 90856a85fd8SHolger Hoffstätte 90962ab466cSMartijn Coenen /** 91062ab466cSMartijn Coenen * loop_set_status_from_info - configure device from loop_info 91162ab466cSMartijn Coenen * @lo: struct loop_device to configure 91262ab466cSMartijn Coenen * @info: struct loop_info64 to configure the device with 91362ab466cSMartijn Coenen * 91462ab466cSMartijn Coenen * Configures the loop device parameters according to the passed 91562ab466cSMartijn Coenen * in loop_info64 configuration. 91662ab466cSMartijn Coenen */ 91762ab466cSMartijn Coenen static int 91862ab466cSMartijn Coenen loop_set_status_from_info(struct loop_device *lo, 91962ab466cSMartijn Coenen const struct loop_info64 *info) 92062ab466cSMartijn Coenen { 92162ab466cSMartijn Coenen if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE) 92262ab466cSMartijn Coenen return -EINVAL; 92362ab466cSMartijn Coenen 92447e96246SChristoph Hellwig switch (info->lo_encrypt_type) { 92547e96246SChristoph Hellwig case LO_CRYPT_NONE: 92647e96246SChristoph Hellwig break; 92747e96246SChristoph Hellwig case LO_CRYPT_XOR: 92847e96246SChristoph Hellwig pr_warn("support for the xor transformation has been removed.\n"); 92962ab466cSMartijn Coenen return -EINVAL; 93047e96246SChristoph Hellwig case LO_CRYPT_CRYPTOAPI: 93147e96246SChristoph Hellwig pr_warn("support for cryptoloop has been removed. Use dm-crypt instead.\n"); 93262ab466cSMartijn Coenen return -EINVAL; 93347e96246SChristoph Hellwig default: 93447e96246SChristoph Hellwig return -EINVAL; 93547e96246SChristoph Hellwig } 93662ab466cSMartijn Coenen 93762ab466cSMartijn Coenen lo->lo_offset = info->lo_offset; 93862ab466cSMartijn Coenen lo->lo_sizelimit = info->lo_sizelimit; 93962ab466cSMartijn Coenen memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE); 94062ab466cSMartijn Coenen lo->lo_file_name[LO_NAME_SIZE-1] = 0; 941faf1d254SMartijn Coenen lo->lo_flags = info->lo_flags; 94262ab466cSMartijn Coenen return 0; 94362ab466cSMartijn Coenen } 94462ab466cSMartijn Coenen 9453448914eSMartijn Coenen static int loop_configure(struct loop_device *lo, fmode_t mode, 9463448914eSMartijn Coenen struct block_device *bdev, 9473448914eSMartijn Coenen const struct loop_config *config) 9481da177e4SLinus Torvalds { 9493ce6e1f6STetsuo Handa struct file *file = fget(config->fd); 9501da177e4SLinus Torvalds struct inode *inode; 9511da177e4SLinus Torvalds struct address_space *mapping; 9521da177e4SLinus Torvalds int error; 9531da177e4SLinus Torvalds loff_t size; 95485b0a54aSJan Kara bool partscan; 9553448914eSMartijn Coenen unsigned short bsize; 9563ce6e1f6STetsuo Handa bool is_loop; 9573ce6e1f6STetsuo Handa 9583ce6e1f6STetsuo Handa if (!file) 9593ce6e1f6STetsuo Handa return -EBADF; 9603ce6e1f6STetsuo Handa is_loop = is_loop_device(file); 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds /* This is safe, since we have a reference from open(). */ 9631da177e4SLinus Torvalds __module_get(THIS_MODULE); 9641da177e4SLinus Torvalds 96533ec3e53SJan Kara /* 96633ec3e53SJan Kara * If we don't hold exclusive handle for the device, upgrade to it 96733ec3e53SJan Kara * here to avoid changing device under exclusive owner. 96833ec3e53SJan Kara */ 96933ec3e53SJan Kara if (!(mode & FMODE_EXCL)) { 97037c3fc9aSChristoph Hellwig error = bd_prepare_to_claim(bdev, loop_configure); 971ecbe6bc0SChristoph Hellwig if (error) 972757ecf40SJan Kara goto out_putf; 97333ec3e53SJan Kara } 97433ec3e53SJan Kara 9753ce6e1f6STetsuo Handa error = loop_global_lock_killable(lo, is_loop); 97633ec3e53SJan Kara if (error) 97733ec3e53SJan Kara goto out_bdev; 978757ecf40SJan Kara 9791da177e4SLinus Torvalds error = -EBUSY; 9801da177e4SLinus Torvalds if (lo->lo_state != Lo_unbound) 981757ecf40SJan Kara goto out_unlock; 9821da177e4SLinus Torvalds 983d2ac838eSTheodore Ts'o error = loop_validate_file(file, bdev); 984d2ac838eSTheodore Ts'o if (error) 985757ecf40SJan Kara goto out_unlock; 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds mapping = file->f_mapping; 9881da177e4SLinus Torvalds inode = mapping->host; 9891da177e4SLinus Torvalds 9903448914eSMartijn Coenen if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) { 9913448914eSMartijn Coenen error = -EINVAL; 9923448914eSMartijn Coenen goto out_unlock; 9933448914eSMartijn Coenen } 9943448914eSMartijn Coenen 9953448914eSMartijn Coenen if (config->block_size) { 996af3c570fSXie Yongji error = blk_validate_block_size(config->block_size); 9973448914eSMartijn Coenen if (error) 9983448914eSMartijn Coenen goto out_unlock; 9993448914eSMartijn Coenen } 10003448914eSMartijn Coenen 10013448914eSMartijn Coenen error = loop_set_status_from_info(lo, &config->info); 10023448914eSMartijn Coenen if (error) 10033448914eSMartijn Coenen goto out_unlock; 10043448914eSMartijn Coenen 1005456be148SChristoph Hellwig if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) || 1006283e7e5dSAl Viro !file->f_op->write_iter) 10073448914eSMartijn Coenen lo->lo_flags |= LO_FLAGS_READ_ONLY; 1008083a6a50SMartijn Coenen 100987579e9bSDan Schatzberg lo->workqueue = alloc_workqueue("loop%d", 101087579e9bSDan Schatzberg WQ_UNBOUND | WQ_FREEZABLE, 101187579e9bSDan Schatzberg 0, 101287579e9bSDan Schatzberg lo->lo_number); 101387579e9bSDan Schatzberg if (!lo->workqueue) { 101487579e9bSDan Schatzberg error = -ENOMEM; 1015757ecf40SJan Kara goto out_unlock; 101687579e9bSDan Schatzberg } 10171da177e4SLinus Torvalds 10189f65c489SMatteo Croce disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 10197a2f0ce1SChristoph Hellwig set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0); 10201da177e4SLinus Torvalds 102187579e9bSDan Schatzberg INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn); 102287579e9bSDan Schatzberg INIT_LIST_HEAD(&lo->rootcg_cmd_list); 102387579e9bSDan Schatzberg INIT_LIST_HEAD(&lo->idle_worker_list); 102487579e9bSDan Schatzberg lo->worker_tree = RB_ROOT; 102587579e9bSDan Schatzberg timer_setup(&lo->timer, loop_free_idle_workers, 102687579e9bSDan Schatzberg TIMER_DEFERRABLE); 10273448914eSMartijn Coenen lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO; 10281da177e4SLinus Torvalds lo->lo_device = bdev; 10291da177e4SLinus Torvalds lo->lo_backing_file = file; 10301da177e4SLinus Torvalds lo->old_gfp_mask = mapping_gfp_mask(mapping); 10311da177e4SLinus Torvalds mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS)); 10321da177e4SLinus Torvalds 10333448914eSMartijn Coenen if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync) 103421d0727fSJens Axboe blk_queue_write_cache(lo->lo_queue, true, false); 103568db1961SNikanth Karthikesan 10363448914eSMartijn Coenen if (config->block_size) 10373448914eSMartijn Coenen bsize = config->block_size; 103896ed320dSLinus Torvalds else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev) 103985560117SMartijn Coenen /* In case of direct I/O, match underlying block size */ 10403448914eSMartijn Coenen bsize = bdev_logical_block_size(inode->i_sb->s_bdev); 10413448914eSMartijn Coenen else 10423448914eSMartijn Coenen bsize = 512; 104385560117SMartijn Coenen 104485560117SMartijn Coenen blk_queue_logical_block_size(lo->lo_queue, bsize); 104585560117SMartijn Coenen blk_queue_physical_block_size(lo->lo_queue, bsize); 104685560117SMartijn Coenen blk_queue_io_min(lo->lo_queue, bsize); 104785560117SMartijn Coenen 10482b9ac22bSKristian Klausen loop_config_discard(lo); 104956a85fd8SHolger Hoffstätte loop_update_rotational(lo); 10502e5ab5f3SMing Lei loop_update_dio(lo); 1051ee862730SMilan Broz loop_sysfs_init(lo); 105279e5dc59SMartijn Coenen 105379e5dc59SMartijn Coenen size = get_loop_size(lo, file); 10545795b6f5SMartijn Coenen loop_set_size(lo, size); 10551da177e4SLinus Torvalds 10563ce6e1f6STetsuo Handa /* Order wrt reading lo_state in loop_validate_file(). */ 10573ce6e1f6STetsuo Handa wmb(); 10583ce6e1f6STetsuo Handa 10596c997918SSerge E. Hallyn lo->lo_state = Lo_bound; 1060e03c8dd1SKay Sievers if (part_shift) 1061e03c8dd1SKay Sievers lo->lo_flags |= LO_FLAGS_PARTSCAN; 106285b0a54aSJan Kara partscan = lo->lo_flags & LO_FLAGS_PARTSCAN; 1063fe6a8fc5SLennart Poettering if (partscan) 106446e7eac6SChristoph Hellwig lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 1065c1681bf8SAnatol Pomozov 10663ce6e1f6STetsuo Handa loop_global_unlock(lo, is_loop); 106785b0a54aSJan Kara if (partscan) 10680384264eSChristoph Hellwig loop_reread_partitions(lo); 106937c3fc9aSChristoph Hellwig if (!(mode & FMODE_EXCL)) 107037c3fc9aSChristoph Hellwig bd_abort_claiming(bdev, loop_configure); 10711da177e4SLinus Torvalds return 0; 10721da177e4SLinus Torvalds 1073757ecf40SJan Kara out_unlock: 10743ce6e1f6STetsuo Handa loop_global_unlock(lo, is_loop); 107533ec3e53SJan Kara out_bdev: 107637c3fc9aSChristoph Hellwig if (!(mode & FMODE_EXCL)) 107737c3fc9aSChristoph Hellwig bd_abort_claiming(bdev, loop_configure); 10781da177e4SLinus Torvalds out_putf: 10791da177e4SLinus Torvalds fput(file); 10801da177e4SLinus Torvalds /* This is safe: open() is still holding a reference. */ 10811da177e4SLinus Torvalds module_put(THIS_MODULE); 10821da177e4SLinus Torvalds return error; 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 1085bf23747eSTetsuo Handa static void __loop_clr_fd(struct loop_device *lo, bool release) 10861da177e4SLinus Torvalds { 10876050fa4cSTetsuo Handa struct file *filp; 1088b4e3ca1aSAl Viro gfp_t gfp = lo->old_gfp_mask; 108987579e9bSDan Schatzberg struct loop_worker *pos, *worker; 10901da177e4SLinus Torvalds 10913ce6e1f6STetsuo Handa /* 10923ce6e1f6STetsuo Handa * Flush loop_configure() and loop_change_fd(). It is acceptable for 10933ce6e1f6STetsuo Handa * loop_validate_file() to succeed, for actual clear operation has not 10943ce6e1f6STetsuo Handa * started yet. 10953ce6e1f6STetsuo Handa */ 10963ce6e1f6STetsuo Handa mutex_lock(&loop_validate_mutex); 10973ce6e1f6STetsuo Handa mutex_unlock(&loop_validate_mutex); 10983ce6e1f6STetsuo Handa /* 10993ce6e1f6STetsuo Handa * loop_validate_file() now fails because l->lo_state != Lo_bound 11003ce6e1f6STetsuo Handa * became visible. 11013ce6e1f6STetsuo Handa */ 11023ce6e1f6STetsuo Handa 11036050fa4cSTetsuo Handa /* 11046050fa4cSTetsuo Handa * Since this function is called upon "ioctl(LOOP_CLR_FD)" xor "close() 11056050fa4cSTetsuo Handa * after ioctl(LOOP_CLR_FD)", it is a sign of something going wrong if 11066050fa4cSTetsuo Handa * lo->lo_state has changed while waiting for lo->lo_mutex. 11076050fa4cSTetsuo Handa */ 11086cc8e743SPavel Tatashin mutex_lock(&lo->lo_mutex); 11096050fa4cSTetsuo Handa BUG_ON(lo->lo_state != Lo_rundown); 11106050fa4cSTetsuo Handa mutex_unlock(&lo->lo_mutex); 11111da177e4SLinus Torvalds 11124ceddce5SMauricio Faria de Oliveira if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags)) 11134ceddce5SMauricio Faria de Oliveira blk_queue_write_cache(lo->lo_queue, false, false); 11144ceddce5SMauricio Faria de Oliveira 1115f8933667SMing Lei /* freeze request queue during the transition */ 1116f8933667SMing Lei blk_mq_freeze_queue(lo->lo_queue); 1117f8933667SMing Lei 111887579e9bSDan Schatzberg destroy_workqueue(lo->workqueue); 111987579e9bSDan Schatzberg spin_lock_irq(&lo->lo_work_lock); 112087579e9bSDan Schatzberg list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 112187579e9bSDan Schatzberg idle_list) { 112287579e9bSDan Schatzberg list_del(&worker->idle_list); 112387579e9bSDan Schatzberg rb_erase(&worker->rb_node, &lo->worker_tree); 1124c74d40e8SDan Schatzberg css_put(worker->blkcg_css); 112587579e9bSDan Schatzberg kfree(worker); 112687579e9bSDan Schatzberg } 112787579e9bSDan Schatzberg spin_unlock_irq(&lo->lo_work_lock); 112887579e9bSDan Schatzberg del_timer_sync(&lo->timer); 112987579e9bSDan Schatzberg 11301da177e4SLinus Torvalds spin_lock_irq(&lo->lo_lock); 11316050fa4cSTetsuo Handa filp = lo->lo_backing_file; 11321da177e4SLinus Torvalds lo->lo_backing_file = NULL; 113305eb0f25SKay Sievers spin_unlock_irq(&lo->lo_lock); 11341da177e4SLinus Torvalds 11351da177e4SLinus Torvalds lo->lo_device = NULL; 11361da177e4SLinus Torvalds lo->lo_offset = 0; 11371da177e4SLinus Torvalds lo->lo_sizelimit = 0; 11381da177e4SLinus Torvalds memset(lo->lo_file_name, 0, LO_NAME_SIZE); 113989e4fdecSOmar Sandoval blk_queue_logical_block_size(lo->lo_queue, 512); 1140bf093753SOmar Sandoval blk_queue_physical_block_size(lo->lo_queue, 512); 1141bf093753SOmar Sandoval blk_queue_io_min(lo->lo_queue, 512); 1142e515be8fSXie Yongji invalidate_disk(lo->lo_disk); 114351a0bb0cSMilan Broz loop_sysfs_exit(lo); 1144c3473c63SDavid Zeuthen /* let user-space know about this change */ 114519f553dbSXie Yongji kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE); 11461da177e4SLinus Torvalds mapping_set_gfp_mask(filp->f_mapping, gfp); 1147bf23747eSTetsuo Handa /* This is safe: open() is still holding a reference. */ 1148bf23747eSTetsuo Handa module_put(THIS_MODULE); 1149f8933667SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 1150f8933667SMing Lei 11519f65c489SMatteo Croce disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE); 11526050fa4cSTetsuo Handa 11536050fa4cSTetsuo Handa if (lo->lo_flags & LO_FLAGS_PARTSCAN) { 11546050fa4cSTetsuo Handa int err; 11556050fa4cSTetsuo Handa 1156bf23747eSTetsuo Handa /* 1157bf23747eSTetsuo Handa * open_mutex has been held already in release path, so don't 1158bf23747eSTetsuo Handa * acquire it if this function is called in such case. 1159bf23747eSTetsuo Handa * 1160bf23747eSTetsuo Handa * If the reread partition isn't from release path, lo_refcnt 1161bf23747eSTetsuo Handa * must be at least one and it can only become zero when the 1162bf23747eSTetsuo Handa * current holder is released. 1163bf23747eSTetsuo Handa */ 1164bf23747eSTetsuo Handa if (!release) 11650384264eSChristoph Hellwig mutex_lock(&lo->lo_disk->open_mutex); 11660384264eSChristoph Hellwig err = bdev_disk_changed(lo->lo_disk, false); 1167bf23747eSTetsuo Handa if (!release) 11680384264eSChristoph Hellwig mutex_unlock(&lo->lo_disk->open_mutex); 116940853d6fSDongli Zhang if (err) 1170d57f3374SJan Kara pr_warn("%s: partition scan of loop%d failed (rc=%d)\n", 11716050fa4cSTetsuo Handa __func__, lo->lo_number, err); 1172d57f3374SJan Kara /* Device is gone, no point in returning error */ 1173d57f3374SJan Kara } 1174758a58d0SDongli Zhang 1175bf23747eSTetsuo Handa /* 1176bf23747eSTetsuo Handa * lo->lo_state is set to Lo_unbound here after above partscan has 1177bf23747eSTetsuo Handa * finished. There cannot be anybody else entering __loop_clr_fd() as 1178bf23747eSTetsuo Handa * Lo_rundown state protects us from all the other places trying to 1179bf23747eSTetsuo Handa * change the 'lo' device. 1180bf23747eSTetsuo Handa */ 1181758a58d0SDongli Zhang lo->lo_flags = 0; 1182758a58d0SDongli Zhang if (!part_shift) 118346e7eac6SChristoph Hellwig lo->lo_disk->flags |= GENHD_FL_NO_PART; 11846050fa4cSTetsuo Handa mutex_lock(&lo->lo_mutex); 1185758a58d0SDongli Zhang lo->lo_state = Lo_unbound; 11866cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1187758a58d0SDongli Zhang 1188bf23747eSTetsuo Handa /* 1189bf23747eSTetsuo Handa * Need not hold lo_mutex to fput backing file. Calling fput holding 1190bf23747eSTetsuo Handa * lo_mutex triggers a circular lock dependency possibility warning as 1191bf23747eSTetsuo Handa * fput can take open_mutex which is usually taken before lo_mutex. 1192bf23747eSTetsuo Handa */ 1193bf23747eSTetsuo Handa fput(filp); 11941da177e4SLinus Torvalds } 11951da177e4SLinus Torvalds 1196a2505b79SJan Kara static int loop_clr_fd(struct loop_device *lo) 1197a2505b79SJan Kara { 11987ccd0791SJan Kara int err; 11997ccd0791SJan Kara 12006cc8e743SPavel Tatashin err = mutex_lock_killable(&lo->lo_mutex); 12017ccd0791SJan Kara if (err) 12027ccd0791SJan Kara return err; 12037ccd0791SJan Kara if (lo->lo_state != Lo_bound) { 12046cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1205a2505b79SJan Kara return -ENXIO; 12067ccd0791SJan Kara } 1207a2505b79SJan Kara /* 1208a2505b79SJan Kara * If we've explicitly asked to tear down the loop device, 1209a2505b79SJan Kara * and it has an elevated reference count, set it for auto-teardown when 1210a2505b79SJan Kara * the last reference goes away. This stops $!~#$@ udev from 1211a2505b79SJan Kara * preventing teardown because it decided that it needs to run blkid on 1212a2505b79SJan Kara * the loopback device whenever they appear. xfstests is notorious for 1213a2505b79SJan Kara * failing tests because blkid via udev races with a losetup 1214a2505b79SJan Kara * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d 1215a2505b79SJan Kara * command to fail with EBUSY. 1216a2505b79SJan Kara */ 1217a2505b79SJan Kara if (atomic_read(&lo->lo_refcnt) > 1) { 1218a2505b79SJan Kara lo->lo_flags |= LO_FLAGS_AUTOCLEAR; 12196cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1220a2505b79SJan Kara return 0; 1221a2505b79SJan Kara } 1222a2505b79SJan Kara lo->lo_state = Lo_rundown; 12236cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1224a2505b79SJan Kara 1225bf23747eSTetsuo Handa __loop_clr_fd(lo, false); 12266050fa4cSTetsuo Handa return 0; 1227a2505b79SJan Kara } 1228a2505b79SJan Kara 12291da177e4SLinus Torvalds static int 12301da177e4SLinus Torvalds loop_set_status(struct loop_device *lo, const struct loop_info64 *info) 12311da177e4SLinus Torvalds { 12321da177e4SLinus Torvalds int err; 1233faf1d254SMartijn Coenen int prev_lo_flags; 123485b0a54aSJan Kara bool partscan = false; 12350c3796c2SMartijn Coenen bool size_changed = false; 12361da177e4SLinus Torvalds 12376cc8e743SPavel Tatashin err = mutex_lock_killable(&lo->lo_mutex); 1238550df5fdSJan Kara if (err) 1239550df5fdSJan Kara return err; 1240550df5fdSJan Kara if (lo->lo_state != Lo_bound) { 1241550df5fdSJan Kara err = -ENXIO; 1242550df5fdSJan Kara goto out_unlock; 1243550df5fdSJan Kara } 12441da177e4SLinus Torvalds 12455db470e2SJaegeuk Kim if (lo->lo_offset != info->lo_offset || 12465db470e2SJaegeuk Kim lo->lo_sizelimit != info->lo_sizelimit) { 12470c3796c2SMartijn Coenen size_changed = true; 12485db470e2SJaegeuk Kim sync_blockdev(lo->lo_device); 1249f4bd34b1SZheng Bin invalidate_bdev(lo->lo_device); 12505db470e2SJaegeuk Kim } 12515db470e2SJaegeuk Kim 1252ecdd0959SMing Lei /* I/O need to be drained during transfer transition */ 1253ecdd0959SMing Lei blk_mq_freeze_queue(lo->lo_queue); 1254ecdd0959SMing Lei 12550c3796c2SMartijn Coenen if (size_changed && lo->lo_device->bd_inode->i_mapping->nrpages) { 1256f4bd34b1SZheng Bin /* If any pages were dirtied after invalidate_bdev(), try again */ 12575db470e2SJaegeuk Kim err = -EAGAIN; 1258d9a74051SColin Ian King pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 12595db470e2SJaegeuk Kim __func__, lo->lo_number, lo->lo_file_name, 12605db470e2SJaegeuk Kim lo->lo_device->bd_inode->i_mapping->nrpages); 12615db470e2SJaegeuk Kim goto out_unfreeze; 12625db470e2SJaegeuk Kim } 12630c3796c2SMartijn Coenen 1264faf1d254SMartijn Coenen prev_lo_flags = lo->lo_flags; 1265faf1d254SMartijn Coenen 12660c3796c2SMartijn Coenen err = loop_set_status_from_info(lo, info); 12670c3796c2SMartijn Coenen if (err) 1268550df5fdSJan Kara goto out_unfreeze; 12690c3796c2SMartijn Coenen 1270faf1d254SMartijn Coenen /* Mask out flags that can't be set using LOOP_SET_STATUS. */ 12716ac92fb5SMartijn Coenen lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS; 1272faf1d254SMartijn Coenen /* For those flags, use the previous values instead */ 1273faf1d254SMartijn Coenen lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS; 1274faf1d254SMartijn Coenen /* For flags that can't be cleared, use previous values too */ 1275faf1d254SMartijn Coenen lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS; 1276faf1d254SMartijn Coenen 1277b0bd158dSMartijn Coenen if (size_changed) { 1278b0bd158dSMartijn Coenen loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit, 1279b0bd158dSMartijn Coenen lo->lo_backing_file); 1280b0bd158dSMartijn Coenen loop_set_size(lo, new_size); 1281b040ad9cSArnd Bergmann } 1282541c742aSGuo Chao 1283dfaa2ef6SLukas Czerner loop_config_discard(lo); 12841da177e4SLinus Torvalds 12852e5ab5f3SMing Lei /* update dio if lo_offset or transfer is changed */ 12862e5ab5f3SMing Lei __loop_update_dio(lo, lo->use_dio); 12872e5ab5f3SMing Lei 1288550df5fdSJan Kara out_unfreeze: 1289ecdd0959SMing Lei blk_mq_unfreeze_queue(lo->lo_queue); 1290e02898b4SOmar Sandoval 1291faf1d254SMartijn Coenen if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) && 1292faf1d254SMartijn Coenen !(prev_lo_flags & LO_FLAGS_PARTSCAN)) { 129346e7eac6SChristoph Hellwig lo->lo_disk->flags &= ~GENHD_FL_NO_PART; 129485b0a54aSJan Kara partscan = true; 1295e02898b4SOmar Sandoval } 1296550df5fdSJan Kara out_unlock: 12976cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 129885b0a54aSJan Kara if (partscan) 12990384264eSChristoph Hellwig loop_reread_partitions(lo); 1300e02898b4SOmar Sandoval 1301ecdd0959SMing Lei return err; 13021da177e4SLinus Torvalds } 13031da177e4SLinus Torvalds 13041da177e4SLinus Torvalds static int 13051da177e4SLinus Torvalds loop_get_status(struct loop_device *lo, struct loop_info64 *info) 13061da177e4SLinus Torvalds { 1307b1ab5fa3STetsuo Handa struct path path; 13081da177e4SLinus Torvalds struct kstat stat; 13092d1d4c1eSOmar Sandoval int ret; 13101da177e4SLinus Torvalds 13116cc8e743SPavel Tatashin ret = mutex_lock_killable(&lo->lo_mutex); 13124a5ce9baSJan Kara if (ret) 13134a5ce9baSJan Kara return ret; 13142d1d4c1eSOmar Sandoval if (lo->lo_state != Lo_bound) { 13156cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 13161da177e4SLinus Torvalds return -ENXIO; 13172d1d4c1eSOmar Sandoval } 13182d1d4c1eSOmar Sandoval 13191da177e4SLinus Torvalds memset(info, 0, sizeof(*info)); 13201da177e4SLinus Torvalds info->lo_number = lo->lo_number; 13211da177e4SLinus Torvalds info->lo_offset = lo->lo_offset; 13221da177e4SLinus Torvalds info->lo_sizelimit = lo->lo_sizelimit; 13231da177e4SLinus Torvalds info->lo_flags = lo->lo_flags; 13241da177e4SLinus Torvalds memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE); 13252d1d4c1eSOmar Sandoval 13266cc8e743SPavel Tatashin /* Drop lo_mutex while we call into the filesystem. */ 1327b1ab5fa3STetsuo Handa path = lo->lo_backing_file->f_path; 1328b1ab5fa3STetsuo Handa path_get(&path); 13296cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1330b1ab5fa3STetsuo Handa ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT); 13312d1d4c1eSOmar Sandoval if (!ret) { 13322d1d4c1eSOmar Sandoval info->lo_device = huge_encode_dev(stat.dev); 13332d1d4c1eSOmar Sandoval info->lo_inode = stat.ino; 13342d1d4c1eSOmar Sandoval info->lo_rdevice = huge_encode_dev(stat.rdev); 13352d1d4c1eSOmar Sandoval } 1336b1ab5fa3STetsuo Handa path_put(&path); 13372d1d4c1eSOmar Sandoval return ret; 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds static void 13411da177e4SLinus Torvalds loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64) 13421da177e4SLinus Torvalds { 13431da177e4SLinus Torvalds memset(info64, 0, sizeof(*info64)); 13441da177e4SLinus Torvalds info64->lo_number = info->lo_number; 13451da177e4SLinus Torvalds info64->lo_device = info->lo_device; 13461da177e4SLinus Torvalds info64->lo_inode = info->lo_inode; 13471da177e4SLinus Torvalds info64->lo_rdevice = info->lo_rdevice; 13481da177e4SLinus Torvalds info64->lo_offset = info->lo_offset; 13491da177e4SLinus Torvalds info64->lo_sizelimit = 0; 13501da177e4SLinus Torvalds info64->lo_flags = info->lo_flags; 13511da177e4SLinus Torvalds memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE); 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 13541da177e4SLinus Torvalds static int 13551da177e4SLinus Torvalds loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info) 13561da177e4SLinus Torvalds { 13571da177e4SLinus Torvalds memset(info, 0, sizeof(*info)); 13581da177e4SLinus Torvalds info->lo_number = info64->lo_number; 13591da177e4SLinus Torvalds info->lo_device = info64->lo_device; 13601da177e4SLinus Torvalds info->lo_inode = info64->lo_inode; 13611da177e4SLinus Torvalds info->lo_rdevice = info64->lo_rdevice; 13621da177e4SLinus Torvalds info->lo_offset = info64->lo_offset; 13631da177e4SLinus Torvalds info->lo_flags = info64->lo_flags; 13641da177e4SLinus Torvalds memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE); 13651da177e4SLinus Torvalds 13661da177e4SLinus Torvalds /* error in case values were truncated */ 13671da177e4SLinus Torvalds if (info->lo_device != info64->lo_device || 13681da177e4SLinus Torvalds info->lo_rdevice != info64->lo_rdevice || 13691da177e4SLinus Torvalds info->lo_inode != info64->lo_inode || 13701da177e4SLinus Torvalds info->lo_offset != info64->lo_offset) 13711da177e4SLinus Torvalds return -EOVERFLOW; 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds return 0; 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds static int 13771da177e4SLinus Torvalds loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg) 13781da177e4SLinus Torvalds { 13791da177e4SLinus Torvalds struct loop_info info; 13801da177e4SLinus Torvalds struct loop_info64 info64; 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds if (copy_from_user(&info, arg, sizeof (struct loop_info))) 13831da177e4SLinus Torvalds return -EFAULT; 13841da177e4SLinus Torvalds loop_info64_from_old(&info, &info64); 13851da177e4SLinus Torvalds return loop_set_status(lo, &info64); 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 13881da177e4SLinus Torvalds static int 13891da177e4SLinus Torvalds loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg) 13901da177e4SLinus Torvalds { 13911da177e4SLinus Torvalds struct loop_info64 info64; 13921da177e4SLinus Torvalds 13931da177e4SLinus Torvalds if (copy_from_user(&info64, arg, sizeof (struct loop_info64))) 13941da177e4SLinus Torvalds return -EFAULT; 13951da177e4SLinus Torvalds return loop_set_status(lo, &info64); 13961da177e4SLinus Torvalds } 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds static int 13991da177e4SLinus Torvalds loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) { 14001da177e4SLinus Torvalds struct loop_info info; 14011da177e4SLinus Torvalds struct loop_info64 info64; 1402bdac616dSOmar Sandoval int err; 14031da177e4SLinus Torvalds 14044a5ce9baSJan Kara if (!arg) 1405bdac616dSOmar Sandoval return -EINVAL; 14061da177e4SLinus Torvalds err = loop_get_status(lo, &info64); 14071da177e4SLinus Torvalds if (!err) 14081da177e4SLinus Torvalds err = loop_info64_to_old(&info64, &info); 14091da177e4SLinus Torvalds if (!err && copy_to_user(arg, &info, sizeof(info))) 14101da177e4SLinus Torvalds err = -EFAULT; 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds return err; 14131da177e4SLinus Torvalds } 14141da177e4SLinus Torvalds 14151da177e4SLinus Torvalds static int 14161da177e4SLinus Torvalds loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) { 14171da177e4SLinus Torvalds struct loop_info64 info64; 1418bdac616dSOmar Sandoval int err; 14191da177e4SLinus Torvalds 14204a5ce9baSJan Kara if (!arg) 1421bdac616dSOmar Sandoval return -EINVAL; 14221da177e4SLinus Torvalds err = loop_get_status(lo, &info64); 14231da177e4SLinus Torvalds if (!err && copy_to_user(arg, &info64, sizeof(info64))) 14241da177e4SLinus Torvalds err = -EFAULT; 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds return err; 14271da177e4SLinus Torvalds } 14281da177e4SLinus Torvalds 142951001b7dSHannes Reinecke static int loop_set_capacity(struct loop_device *lo) 143053d66608SJ. R. Okajima { 14310a6ed1b5SMartijn Coenen loff_t size; 14320a6ed1b5SMartijn Coenen 143353d66608SJ. R. Okajima if (unlikely(lo->lo_state != Lo_bound)) 14347b0576a3SGuo Chao return -ENXIO; 143553d66608SJ. R. Okajima 14360a6ed1b5SMartijn Coenen size = get_loop_size(lo, lo->lo_backing_file); 14370a6ed1b5SMartijn Coenen loop_set_size(lo, size); 1438083a6a50SMartijn Coenen 1439083a6a50SMartijn Coenen return 0; 144053d66608SJ. R. Okajima } 144153d66608SJ. R. Okajima 1442ab1cb278SMing Lei static int loop_set_dio(struct loop_device *lo, unsigned long arg) 1443ab1cb278SMing Lei { 1444ab1cb278SMing Lei int error = -ENXIO; 1445ab1cb278SMing Lei if (lo->lo_state != Lo_bound) 1446ab1cb278SMing Lei goto out; 1447ab1cb278SMing Lei 1448ab1cb278SMing Lei __loop_update_dio(lo, !!arg); 1449ab1cb278SMing Lei if (lo->use_dio == !!arg) 1450ab1cb278SMing Lei return 0; 1451ab1cb278SMing Lei error = -EINVAL; 1452ab1cb278SMing Lei out: 1453ab1cb278SMing Lei return error; 1454ab1cb278SMing Lei } 1455ab1cb278SMing Lei 145689e4fdecSOmar Sandoval static int loop_set_block_size(struct loop_device *lo, unsigned long arg) 145789e4fdecSOmar Sandoval { 14585db470e2SJaegeuk Kim int err = 0; 14595db470e2SJaegeuk Kim 146089e4fdecSOmar Sandoval if (lo->lo_state != Lo_bound) 146189e4fdecSOmar Sandoval return -ENXIO; 146289e4fdecSOmar Sandoval 1463af3c570fSXie Yongji err = blk_validate_block_size(arg); 14643448914eSMartijn Coenen if (err) 14653448914eSMartijn Coenen return err; 146689e4fdecSOmar Sandoval 14677e81f99aSMartijn Coenen if (lo->lo_queue->limits.logical_block_size == arg) 14687e81f99aSMartijn Coenen return 0; 14697e81f99aSMartijn Coenen 14705db470e2SJaegeuk Kim sync_blockdev(lo->lo_device); 1471f4bd34b1SZheng Bin invalidate_bdev(lo->lo_device); 14725db470e2SJaegeuk Kim 147389e4fdecSOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 147489e4fdecSOmar Sandoval 1475f4bd34b1SZheng Bin /* invalidate_bdev should have truncated all the pages */ 14767e81f99aSMartijn Coenen if (lo->lo_device->bd_inode->i_mapping->nrpages) { 14775db470e2SJaegeuk Kim err = -EAGAIN; 1478d9a74051SColin Ian King pr_warn("%s: loop%d (%s) still has dirty pages (nrpages=%lu)\n", 14795db470e2SJaegeuk Kim __func__, lo->lo_number, lo->lo_file_name, 14805db470e2SJaegeuk Kim lo->lo_device->bd_inode->i_mapping->nrpages); 14815db470e2SJaegeuk Kim goto out_unfreeze; 14825db470e2SJaegeuk Kim } 14835db470e2SJaegeuk Kim 148489e4fdecSOmar Sandoval blk_queue_logical_block_size(lo->lo_queue, arg); 1485bf093753SOmar Sandoval blk_queue_physical_block_size(lo->lo_queue, arg); 1486bf093753SOmar Sandoval blk_queue_io_min(lo->lo_queue, arg); 148789e4fdecSOmar Sandoval loop_update_dio(lo); 14885db470e2SJaegeuk Kim out_unfreeze: 148989e4fdecSOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 149089e4fdecSOmar Sandoval 14915db470e2SJaegeuk Kim return err; 149289e4fdecSOmar Sandoval } 149389e4fdecSOmar Sandoval 1494a1316544SJan Kara static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd, 1495a1316544SJan Kara unsigned long arg) 14961da177e4SLinus Torvalds { 14971da177e4SLinus Torvalds int err; 14981da177e4SLinus Torvalds 14996cc8e743SPavel Tatashin err = mutex_lock_killable(&lo->lo_mutex); 15003148ffbdSOmar Sandoval if (err) 1501a1316544SJan Kara return err; 15021da177e4SLinus Torvalds switch (cmd) { 150353d66608SJ. R. Okajima case LOOP_SET_CAPACITY: 150451001b7dSHannes Reinecke err = loop_set_capacity(lo); 150553d66608SJ. R. Okajima break; 1506ab1cb278SMing Lei case LOOP_SET_DIRECT_IO: 1507ab1cb278SMing Lei err = loop_set_dio(lo, arg); 1508ab1cb278SMing Lei break; 150989e4fdecSOmar Sandoval case LOOP_SET_BLOCK_SIZE: 151089e4fdecSOmar Sandoval err = loop_set_block_size(lo, arg); 151189e4fdecSOmar Sandoval break; 15121da177e4SLinus Torvalds default: 151347e96246SChristoph Hellwig err = -EINVAL; 15141da177e4SLinus Torvalds } 15156cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1516a1316544SJan Kara return err; 1517a1316544SJan Kara } 1518f028f3b2SNikanth Karthikesan 1519a1316544SJan Kara static int lo_ioctl(struct block_device *bdev, fmode_t mode, 1520a1316544SJan Kara unsigned int cmd, unsigned long arg) 1521a1316544SJan Kara { 1522a1316544SJan Kara struct loop_device *lo = bdev->bd_disk->private_data; 1523571fae6eSMartijn Coenen void __user *argp = (void __user *) arg; 1524a1316544SJan Kara int err; 1525a1316544SJan Kara 1526a1316544SJan Kara switch (cmd) { 15273448914eSMartijn Coenen case LOOP_SET_FD: { 15283448914eSMartijn Coenen /* 15293448914eSMartijn Coenen * Legacy case - pass in a zeroed out struct loop_config with 15303448914eSMartijn Coenen * only the file descriptor set , which corresponds with the 15313448914eSMartijn Coenen * default parameters we'd have used otherwise. 15323448914eSMartijn Coenen */ 15333448914eSMartijn Coenen struct loop_config config; 15343448914eSMartijn Coenen 15353448914eSMartijn Coenen memset(&config, 0, sizeof(config)); 15363448914eSMartijn Coenen config.fd = arg; 15373448914eSMartijn Coenen 15383448914eSMartijn Coenen return loop_configure(lo, mode, bdev, &config); 15393448914eSMartijn Coenen } 15403448914eSMartijn Coenen case LOOP_CONFIGURE: { 15413448914eSMartijn Coenen struct loop_config config; 15423448914eSMartijn Coenen 15433448914eSMartijn Coenen if (copy_from_user(&config, argp, sizeof(config))) 15443448914eSMartijn Coenen return -EFAULT; 15453448914eSMartijn Coenen 15463448914eSMartijn Coenen return loop_configure(lo, mode, bdev, &config); 15473448914eSMartijn Coenen } 1548a1316544SJan Kara case LOOP_CHANGE_FD: 1549c3710770SJan Kara return loop_change_fd(lo, bdev, arg); 1550a1316544SJan Kara case LOOP_CLR_FD: 15517ccd0791SJan Kara return loop_clr_fd(lo); 1552a1316544SJan Kara case LOOP_SET_STATUS: 1553a1316544SJan Kara err = -EPERM; 1554a1316544SJan Kara if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1555571fae6eSMartijn Coenen err = loop_set_status_old(lo, argp); 1556a1316544SJan Kara } 1557a1316544SJan Kara break; 1558a1316544SJan Kara case LOOP_GET_STATUS: 1559571fae6eSMartijn Coenen return loop_get_status_old(lo, argp); 1560a1316544SJan Kara case LOOP_SET_STATUS64: 1561a1316544SJan Kara err = -EPERM; 1562a1316544SJan Kara if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) { 1563571fae6eSMartijn Coenen err = loop_set_status64(lo, argp); 1564a1316544SJan Kara } 1565a1316544SJan Kara break; 1566a1316544SJan Kara case LOOP_GET_STATUS64: 1567571fae6eSMartijn Coenen return loop_get_status64(lo, argp); 1568a1316544SJan Kara case LOOP_SET_CAPACITY: 1569a1316544SJan Kara case LOOP_SET_DIRECT_IO: 1570a1316544SJan Kara case LOOP_SET_BLOCK_SIZE: 1571a1316544SJan Kara if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN)) 1572a1316544SJan Kara return -EPERM; 1573df561f66SGustavo A. R. Silva fallthrough; 1574a1316544SJan Kara default: 1575a1316544SJan Kara err = lo_simple_ioctl(lo, cmd, arg); 1576a1316544SJan Kara break; 1577a1316544SJan Kara } 1578a1316544SJan Kara 15791da177e4SLinus Torvalds return err; 15801da177e4SLinus Torvalds } 15811da177e4SLinus Torvalds 1582863d5b82SDavid Howells #ifdef CONFIG_COMPAT 1583863d5b82SDavid Howells struct compat_loop_info { 1584863d5b82SDavid Howells compat_int_t lo_number; /* ioctl r/o */ 1585863d5b82SDavid Howells compat_dev_t lo_device; /* ioctl r/o */ 1586863d5b82SDavid Howells compat_ulong_t lo_inode; /* ioctl r/o */ 1587863d5b82SDavid Howells compat_dev_t lo_rdevice; /* ioctl r/o */ 1588863d5b82SDavid Howells compat_int_t lo_offset; 1589f941c51eSCarlos Llamas compat_int_t lo_encrypt_type; /* obsolete, ignored */ 1590863d5b82SDavid Howells compat_int_t lo_encrypt_key_size; /* ioctl w/o */ 1591863d5b82SDavid Howells compat_int_t lo_flags; /* ioctl r/o */ 1592863d5b82SDavid Howells char lo_name[LO_NAME_SIZE]; 1593863d5b82SDavid Howells unsigned char lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */ 1594863d5b82SDavid Howells compat_ulong_t lo_init[2]; 1595863d5b82SDavid Howells char reserved[4]; 1596863d5b82SDavid Howells }; 1597863d5b82SDavid Howells 1598863d5b82SDavid Howells /* 1599863d5b82SDavid Howells * Transfer 32-bit compatibility structure in userspace to 64-bit loop info 1600863d5b82SDavid Howells * - noinlined to reduce stack space usage in main part of driver 1601863d5b82SDavid Howells */ 1602863d5b82SDavid Howells static noinline int 1603ba674cfcSAl Viro loop_info64_from_compat(const struct compat_loop_info __user *arg, 1604863d5b82SDavid Howells struct loop_info64 *info64) 1605863d5b82SDavid Howells { 1606863d5b82SDavid Howells struct compat_loop_info info; 1607863d5b82SDavid Howells 1608863d5b82SDavid Howells if (copy_from_user(&info, arg, sizeof(info))) 1609863d5b82SDavid Howells return -EFAULT; 1610863d5b82SDavid Howells 1611863d5b82SDavid Howells memset(info64, 0, sizeof(*info64)); 1612863d5b82SDavid Howells info64->lo_number = info.lo_number; 1613863d5b82SDavid Howells info64->lo_device = info.lo_device; 1614863d5b82SDavid Howells info64->lo_inode = info.lo_inode; 1615863d5b82SDavid Howells info64->lo_rdevice = info.lo_rdevice; 1616863d5b82SDavid Howells info64->lo_offset = info.lo_offset; 1617863d5b82SDavid Howells info64->lo_sizelimit = 0; 1618863d5b82SDavid Howells info64->lo_flags = info.lo_flags; 1619863d5b82SDavid Howells memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE); 1620863d5b82SDavid Howells return 0; 1621863d5b82SDavid Howells } 1622863d5b82SDavid Howells 1623863d5b82SDavid Howells /* 1624863d5b82SDavid Howells * Transfer 64-bit loop info to 32-bit compatibility structure in userspace 1625863d5b82SDavid Howells * - noinlined to reduce stack space usage in main part of driver 1626863d5b82SDavid Howells */ 1627863d5b82SDavid Howells static noinline int 1628863d5b82SDavid Howells loop_info64_to_compat(const struct loop_info64 *info64, 1629863d5b82SDavid Howells struct compat_loop_info __user *arg) 1630863d5b82SDavid Howells { 1631863d5b82SDavid Howells struct compat_loop_info info; 1632863d5b82SDavid Howells 1633863d5b82SDavid Howells memset(&info, 0, sizeof(info)); 1634863d5b82SDavid Howells info.lo_number = info64->lo_number; 1635863d5b82SDavid Howells info.lo_device = info64->lo_device; 1636863d5b82SDavid Howells info.lo_inode = info64->lo_inode; 1637863d5b82SDavid Howells info.lo_rdevice = info64->lo_rdevice; 1638863d5b82SDavid Howells info.lo_offset = info64->lo_offset; 1639863d5b82SDavid Howells info.lo_flags = info64->lo_flags; 1640863d5b82SDavid Howells memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE); 1641863d5b82SDavid Howells 1642863d5b82SDavid Howells /* error in case values were truncated */ 1643863d5b82SDavid Howells if (info.lo_device != info64->lo_device || 1644863d5b82SDavid Howells info.lo_rdevice != info64->lo_rdevice || 1645863d5b82SDavid Howells info.lo_inode != info64->lo_inode || 164647e96246SChristoph Hellwig info.lo_offset != info64->lo_offset) 1647863d5b82SDavid Howells return -EOVERFLOW; 1648863d5b82SDavid Howells 1649863d5b82SDavid Howells if (copy_to_user(arg, &info, sizeof(info))) 1650863d5b82SDavid Howells return -EFAULT; 1651863d5b82SDavid Howells return 0; 1652863d5b82SDavid Howells } 1653863d5b82SDavid Howells 1654863d5b82SDavid Howells static int 1655863d5b82SDavid Howells loop_set_status_compat(struct loop_device *lo, 1656863d5b82SDavid Howells const struct compat_loop_info __user *arg) 1657863d5b82SDavid Howells { 1658863d5b82SDavid Howells struct loop_info64 info64; 1659863d5b82SDavid Howells int ret; 1660863d5b82SDavid Howells 1661863d5b82SDavid Howells ret = loop_info64_from_compat(arg, &info64); 1662863d5b82SDavid Howells if (ret < 0) 1663863d5b82SDavid Howells return ret; 1664863d5b82SDavid Howells return loop_set_status(lo, &info64); 1665863d5b82SDavid Howells } 1666863d5b82SDavid Howells 1667863d5b82SDavid Howells static int 1668863d5b82SDavid Howells loop_get_status_compat(struct loop_device *lo, 1669863d5b82SDavid Howells struct compat_loop_info __user *arg) 1670863d5b82SDavid Howells { 1671863d5b82SDavid Howells struct loop_info64 info64; 1672bdac616dSOmar Sandoval int err; 1673863d5b82SDavid Howells 16744a5ce9baSJan Kara if (!arg) 1675bdac616dSOmar Sandoval return -EINVAL; 1676863d5b82SDavid Howells err = loop_get_status(lo, &info64); 1677863d5b82SDavid Howells if (!err) 1678863d5b82SDavid Howells err = loop_info64_to_compat(&info64, arg); 1679863d5b82SDavid Howells return err; 1680863d5b82SDavid Howells } 1681863d5b82SDavid Howells 1682bb214884SAl Viro static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode, 1683bb214884SAl Viro unsigned int cmd, unsigned long arg) 1684863d5b82SDavid Howells { 1685bb214884SAl Viro struct loop_device *lo = bdev->bd_disk->private_data; 1686863d5b82SDavid Howells int err; 1687863d5b82SDavid Howells 1688863d5b82SDavid Howells switch(cmd) { 1689863d5b82SDavid Howells case LOOP_SET_STATUS: 16903148ffbdSOmar Sandoval err = loop_set_status_compat(lo, 16913148ffbdSOmar Sandoval (const struct compat_loop_info __user *)arg); 1692863d5b82SDavid Howells break; 1693863d5b82SDavid Howells case LOOP_GET_STATUS: 16943148ffbdSOmar Sandoval err = loop_get_status_compat(lo, 16953148ffbdSOmar Sandoval (struct compat_loop_info __user *)arg); 1696863d5b82SDavid Howells break; 169753d66608SJ. R. Okajima case LOOP_SET_CAPACITY: 1698863d5b82SDavid Howells case LOOP_CLR_FD: 1699863d5b82SDavid Howells case LOOP_GET_STATUS64: 1700863d5b82SDavid Howells case LOOP_SET_STATUS64: 17013448914eSMartijn Coenen case LOOP_CONFIGURE: 1702863d5b82SDavid Howells arg = (unsigned long) compat_ptr(arg); 1703df561f66SGustavo A. R. Silva fallthrough; 1704863d5b82SDavid Howells case LOOP_SET_FD: 1705863d5b82SDavid Howells case LOOP_CHANGE_FD: 17069fea4b39SEvan Green case LOOP_SET_BLOCK_SIZE: 1707fdbe4eeeSAlessio Balsini case LOOP_SET_DIRECT_IO: 1708bb214884SAl Viro err = lo_ioctl(bdev, mode, cmd, arg); 1709863d5b82SDavid Howells break; 1710863d5b82SDavid Howells default: 1711863d5b82SDavid Howells err = -ENOIOCTLCMD; 1712863d5b82SDavid Howells break; 1713863d5b82SDavid Howells } 1714863d5b82SDavid Howells return err; 1715863d5b82SDavid Howells } 1716863d5b82SDavid Howells #endif 1717863d5b82SDavid Howells 1718bb214884SAl Viro static int lo_open(struct block_device *bdev, fmode_t mode) 17191da177e4SLinus Torvalds { 1720990e7811SChristoph Hellwig struct loop_device *lo = bdev->bd_disk->private_data; 17210a42e99bSJan Kara int err; 1722770fe30aSKay Sievers 17236cc8e743SPavel Tatashin err = mutex_lock_killable(&lo->lo_mutex); 17246cc8e743SPavel Tatashin if (err) 1725770fe30aSKay Sievers return err; 1726990e7811SChristoph Hellwig if (lo->lo_state == Lo_deleting) 1727990e7811SChristoph Hellwig err = -ENXIO; 1728990e7811SChristoph Hellwig else 17296cc8e743SPavel Tatashin atomic_inc(&lo->lo_refcnt); 17306cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1731990e7811SChristoph Hellwig return err; 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds 1734967d1dc1SJan Kara static void lo_release(struct gendisk *disk, fmode_t mode) 17351da177e4SLinus Torvalds { 17366cc8e743SPavel Tatashin struct loop_device *lo = disk->private_data; 17371da177e4SLinus Torvalds 17386cc8e743SPavel Tatashin mutex_lock(&lo->lo_mutex); 1739f8933667SMing Lei if (atomic_dec_return(&lo->lo_refcnt)) 17400a42e99bSJan Kara goto out_unlock; 1741f8933667SMing Lei 174214f27939SMilan Broz if (lo->lo_flags & LO_FLAGS_AUTOCLEAR) { 1743a2505b79SJan Kara if (lo->lo_state != Lo_bound) 1744a2505b79SJan Kara goto out_unlock; 1745a2505b79SJan Kara lo->lo_state = Lo_rundown; 17466cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 174714f27939SMilan Broz /* 174814f27939SMilan Broz * In autoclear mode, stop the loop thread 174914f27939SMilan Broz * and remove configuration after last close. 175014f27939SMilan Broz */ 1751bf23747eSTetsuo Handa __loop_clr_fd(lo, true); 17520a42e99bSJan Kara return; 175343cade80SOmar Sandoval } else if (lo->lo_state == Lo_bound) { 175414f27939SMilan Broz /* 175514f27939SMilan Broz * Otherwise keep thread (if running) and config, 175614f27939SMilan Broz * but flush possible ongoing bios in thread. 175714f27939SMilan Broz */ 175843cade80SOmar Sandoval blk_mq_freeze_queue(lo->lo_queue); 175943cade80SOmar Sandoval blk_mq_unfreeze_queue(lo->lo_queue); 176014f27939SMilan Broz } 176196c58655SDavid Woodhouse 17620a42e99bSJan Kara out_unlock: 17636cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 1764ae665016SLinus Torvalds } 1765ae665016SLinus Torvalds 176683d5cde4SAlexey Dobriyan static const struct block_device_operations lo_fops = { 17671da177e4SLinus Torvalds .owner = THIS_MODULE, 1768bb214884SAl Viro .open = lo_open, 1769bb214884SAl Viro .release = lo_release, 1770bb214884SAl Viro .ioctl = lo_ioctl, 1771863d5b82SDavid Howells #ifdef CONFIG_COMPAT 1772bb214884SAl Viro .compat_ioctl = lo_compat_ioctl, 1773863d5b82SDavid Howells #endif 17741da177e4SLinus Torvalds }; 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds /* 17771da177e4SLinus Torvalds * And now the modules code and kernel interface. 17781da177e4SLinus Torvalds */ 177973285082SKen Chen static int max_loop; 17805657a819SJoe Perches module_param(max_loop, int, 0444); 1781a47653fcSKen Chen MODULE_PARM_DESC(max_loop, "Maximum number of loop devices"); 17825657a819SJoe Perches module_param(max_part, int, 0444); 1783476a4813SLaurent Vivier MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device"); 1784ef44c508SChaitanya Kulkarni 1785ef44c508SChaitanya Kulkarni static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH; 1786ef44c508SChaitanya Kulkarni 1787ef44c508SChaitanya Kulkarni static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p) 1788ef44c508SChaitanya Kulkarni { 1789ef44c508SChaitanya Kulkarni int ret = kstrtoint(s, 10, &hw_queue_depth); 1790ef44c508SChaitanya Kulkarni 1791ef44c508SChaitanya Kulkarni return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0; 1792ef44c508SChaitanya Kulkarni } 1793ef44c508SChaitanya Kulkarni 1794ef44c508SChaitanya Kulkarni static const struct kernel_param_ops loop_hw_qdepth_param_ops = { 1795ef44c508SChaitanya Kulkarni .set = loop_set_hw_queue_depth, 1796ef44c508SChaitanya Kulkarni .get = param_get_int, 1797ef44c508SChaitanya Kulkarni }; 1798ef44c508SChaitanya Kulkarni 1799ef44c508SChaitanya Kulkarni device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444); 1800ef44c508SChaitanya Kulkarni MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128"); 1801ef44c508SChaitanya Kulkarni 18021da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 18031da177e4SLinus Torvalds MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR); 18041da177e4SLinus Torvalds 1805fc17b653SChristoph Hellwig static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx, 1806b5dd2f60SMing Lei const struct blk_mq_queue_data *bd) 1807b5dd2f60SMing Lei { 18081894e916SJens Axboe struct request *rq = bd->rq; 18091894e916SJens Axboe struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq); 18101894e916SJens Axboe struct loop_device *lo = rq->q->queuedata; 1811b5dd2f60SMing Lei 18121894e916SJens Axboe blk_mq_start_request(rq); 1813b5dd2f60SMing Lei 1814f4aa4c7bSMing Lei if (lo->lo_state != Lo_bound) 1815fc17b653SChristoph Hellwig return BLK_STS_IOERR; 1816f4aa4c7bSMing Lei 18171894e916SJens Axboe switch (req_op(rq)) { 1818f0225cacSChristoph Hellwig case REQ_OP_FLUSH: 1819f0225cacSChristoph Hellwig case REQ_OP_DISCARD: 182019372e27SChristoph Hellwig case REQ_OP_WRITE_ZEROES: 1821bc07c10aSMing Lei cmd->use_aio = false; 1822f0225cacSChristoph Hellwig break; 1823f0225cacSChristoph Hellwig default: 1824f0225cacSChristoph Hellwig cmd->use_aio = lo->use_dio; 1825f0225cacSChristoph Hellwig break; 1826f0225cacSChristoph Hellwig } 1827bc07c10aSMing Lei 1828d4478e92SShaohua Li /* always use the first bio's css */ 1829c74d40e8SDan Schatzberg cmd->blkcg_css = NULL; 1830c74d40e8SDan Schatzberg cmd->memcg_css = NULL; 18310b508bc9SShaohua Li #ifdef CONFIG_BLK_CGROUP 1832c74d40e8SDan Schatzberg if (rq->bio && rq->bio->bi_blkg) { 1833c74d40e8SDan Schatzberg cmd->blkcg_css = &bio_blkcg(rq->bio)->css; 1834c74d40e8SDan Schatzberg #ifdef CONFIG_MEMCG 1835c74d40e8SDan Schatzberg cmd->memcg_css = 1836c74d40e8SDan Schatzberg cgroup_get_e_css(cmd->blkcg_css->cgroup, 1837c74d40e8SDan Schatzberg &memory_cgrp_subsys); 1838d4478e92SShaohua Li #endif 1839c74d40e8SDan Schatzberg } 1840c74d40e8SDan Schatzberg #endif 184187579e9bSDan Schatzberg loop_queue_work(lo, cmd); 1842b5dd2f60SMing Lei 1843fc17b653SChristoph Hellwig return BLK_STS_OK; 1844b5dd2f60SMing Lei } 1845b5dd2f60SMing Lei 1846b5dd2f60SMing Lei static void loop_handle_cmd(struct loop_cmd *cmd) 1847b5dd2f60SMing Lei { 18481894e916SJens Axboe struct request *rq = blk_mq_rq_from_pdu(cmd); 18491894e916SJens Axboe const bool write = op_is_write(req_op(rq)); 18501894e916SJens Axboe struct loop_device *lo = rq->q->queuedata; 1851f4829a9bSChristoph Hellwig int ret = 0; 1852c74d40e8SDan Schatzberg struct mem_cgroup *old_memcg = NULL; 1853b5dd2f60SMing Lei 1854f4829a9bSChristoph Hellwig if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) { 1855f4829a9bSChristoph Hellwig ret = -EIO; 1856b5dd2f60SMing Lei goto failed; 1857f4829a9bSChristoph Hellwig } 1858b5dd2f60SMing Lei 1859c74d40e8SDan Schatzberg if (cmd->blkcg_css) 1860c74d40e8SDan Schatzberg kthread_associate_blkcg(cmd->blkcg_css); 1861c74d40e8SDan Schatzberg if (cmd->memcg_css) 1862c74d40e8SDan Schatzberg old_memcg = set_active_memcg( 1863c74d40e8SDan Schatzberg mem_cgroup_from_css(cmd->memcg_css)); 1864c74d40e8SDan Schatzberg 18651894e916SJens Axboe ret = do_req_filebacked(lo, rq); 1866c74d40e8SDan Schatzberg 1867c74d40e8SDan Schatzberg if (cmd->blkcg_css) 1868c74d40e8SDan Schatzberg kthread_associate_blkcg(NULL); 1869c74d40e8SDan Schatzberg 1870c74d40e8SDan Schatzberg if (cmd->memcg_css) { 1871c74d40e8SDan Schatzberg set_active_memcg(old_memcg); 1872c74d40e8SDan Schatzberg css_put(cmd->memcg_css); 1873c74d40e8SDan Schatzberg } 1874b5dd2f60SMing Lei failed: 1875bc07c10aSMing Lei /* complete non-aio request */ 1876fe2cb290SChristoph Hellwig if (!cmd->use_aio || ret) { 18778cd55087SEvan Green if (ret == -EOPNOTSUPP) 18788cd55087SEvan Green cmd->ret = ret; 18798cd55087SEvan Green else 1880fe2cb290SChristoph Hellwig cmd->ret = ret ? -EIO : 0; 188115f73f5bSChristoph Hellwig if (likely(!blk_should_fake_timeout(rq->q))) 18821894e916SJens Axboe blk_mq_complete_request(rq); 1883fe2cb290SChristoph Hellwig } 1884b5dd2f60SMing Lei } 1885b5dd2f60SMing Lei 188687579e9bSDan Schatzberg static void loop_set_timer(struct loop_device *lo) 1887b5dd2f60SMing Lei { 188887579e9bSDan Schatzberg timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT); 1889b5dd2f60SMing Lei } 1890b5dd2f60SMing Lei 189187579e9bSDan Schatzberg static void loop_process_work(struct loop_worker *worker, 189287579e9bSDan Schatzberg struct list_head *cmd_list, struct loop_device *lo) 1893b5dd2f60SMing Lei { 189487579e9bSDan Schatzberg int orig_flags = current->flags; 189587579e9bSDan Schatzberg struct loop_cmd *cmd; 1896b5dd2f60SMing Lei 189787579e9bSDan Schatzberg current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO; 189887579e9bSDan Schatzberg spin_lock_irq(&lo->lo_work_lock); 189987579e9bSDan Schatzberg while (!list_empty(cmd_list)) { 190087579e9bSDan Schatzberg cmd = container_of( 190187579e9bSDan Schatzberg cmd_list->next, struct loop_cmd, list_entry); 190287579e9bSDan Schatzberg list_del(cmd_list->next); 190387579e9bSDan Schatzberg spin_unlock_irq(&lo->lo_work_lock); 190487579e9bSDan Schatzberg 190587579e9bSDan Schatzberg loop_handle_cmd(cmd); 190687579e9bSDan Schatzberg cond_resched(); 190787579e9bSDan Schatzberg 190887579e9bSDan Schatzberg spin_lock_irq(&lo->lo_work_lock); 190987579e9bSDan Schatzberg } 191087579e9bSDan Schatzberg 191187579e9bSDan Schatzberg /* 191287579e9bSDan Schatzberg * We only add to the idle list if there are no pending cmds 191387579e9bSDan Schatzberg * *and* the worker will not run again which ensures that it 191487579e9bSDan Schatzberg * is safe to free any worker on the idle list 191587579e9bSDan Schatzberg */ 191687579e9bSDan Schatzberg if (worker && !work_pending(&worker->work)) { 191787579e9bSDan Schatzberg worker->last_ran_at = jiffies; 191887579e9bSDan Schatzberg list_add_tail(&worker->idle_list, &lo->idle_worker_list); 191987579e9bSDan Schatzberg loop_set_timer(lo); 192087579e9bSDan Schatzberg } 192187579e9bSDan Schatzberg spin_unlock_irq(&lo->lo_work_lock); 192287579e9bSDan Schatzberg current->flags = orig_flags; 192387579e9bSDan Schatzberg } 192487579e9bSDan Schatzberg 192587579e9bSDan Schatzberg static void loop_workfn(struct work_struct *work) 192687579e9bSDan Schatzberg { 192787579e9bSDan Schatzberg struct loop_worker *worker = 192887579e9bSDan Schatzberg container_of(work, struct loop_worker, work); 192987579e9bSDan Schatzberg loop_process_work(worker, &worker->cmd_list, worker->lo); 193087579e9bSDan Schatzberg } 193187579e9bSDan Schatzberg 193287579e9bSDan Schatzberg static void loop_rootcg_workfn(struct work_struct *work) 193387579e9bSDan Schatzberg { 193487579e9bSDan Schatzberg struct loop_device *lo = 193587579e9bSDan Schatzberg container_of(work, struct loop_device, rootcg_work); 193687579e9bSDan Schatzberg loop_process_work(NULL, &lo->rootcg_cmd_list, lo); 193787579e9bSDan Schatzberg } 193887579e9bSDan Schatzberg 193987579e9bSDan Schatzberg static void loop_free_idle_workers(struct timer_list *timer) 194087579e9bSDan Schatzberg { 194187579e9bSDan Schatzberg struct loop_device *lo = container_of(timer, struct loop_device, timer); 194287579e9bSDan Schatzberg struct loop_worker *pos, *worker; 194387579e9bSDan Schatzberg 194487579e9bSDan Schatzberg spin_lock_irq(&lo->lo_work_lock); 194587579e9bSDan Schatzberg list_for_each_entry_safe(worker, pos, &lo->idle_worker_list, 194687579e9bSDan Schatzberg idle_list) { 194787579e9bSDan Schatzberg if (time_is_after_jiffies(worker->last_ran_at + 194887579e9bSDan Schatzberg LOOP_IDLE_WORKER_TIMEOUT)) 194987579e9bSDan Schatzberg break; 195087579e9bSDan Schatzberg list_del(&worker->idle_list); 195187579e9bSDan Schatzberg rb_erase(&worker->rb_node, &lo->worker_tree); 1952c74d40e8SDan Schatzberg css_put(worker->blkcg_css); 195387579e9bSDan Schatzberg kfree(worker); 195487579e9bSDan Schatzberg } 195587579e9bSDan Schatzberg if (!list_empty(&lo->idle_worker_list)) 195687579e9bSDan Schatzberg loop_set_timer(lo); 195787579e9bSDan Schatzberg spin_unlock_irq(&lo->lo_work_lock); 1958b5dd2f60SMing Lei } 1959b5dd2f60SMing Lei 1960f363b089SEric Biggers static const struct blk_mq_ops loop_mq_ops = { 1961b5dd2f60SMing Lei .queue_rq = loop_queue_rq, 1962fe2cb290SChristoph Hellwig .complete = lo_complete_rq, 1963b5dd2f60SMing Lei }; 1964b5dd2f60SMing Lei 1965d6da83d0SChristoph Hellwig static int loop_add(int i) 19661da177e4SLinus Torvalds { 196773285082SKen Chen struct loop_device *lo; 196873285082SKen Chen struct gendisk *disk; 196934dd82afSKay Sievers int err; 19701da177e4SLinus Torvalds 197134dd82afSKay Sievers err = -ENOMEM; 197268d740d7SSilva Paulo lo = kzalloc(sizeof(*lo), GFP_KERNEL); 197368d740d7SSilva Paulo if (!lo) 197473285082SKen Chen goto out; 1975ef7e7c82SMikulas Patocka lo->lo_state = Lo_unbound; 1976ef7e7c82SMikulas Patocka 197718d1f200SChristoph Hellwig err = mutex_lock_killable(&loop_ctl_mutex); 197818d1f200SChristoph Hellwig if (err) 197918d1f200SChristoph Hellwig goto out_free_dev; 198018d1f200SChristoph Hellwig 1981c718aa65STejun Heo /* allocate id, if @id >= 0, we're requesting that specific id */ 198234dd82afSKay Sievers if (i >= 0) { 1983c718aa65STejun Heo err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL); 1984c718aa65STejun Heo if (err == -ENOSPC) 198534dd82afSKay Sievers err = -EEXIST; 198634dd82afSKay Sievers } else { 1987c718aa65STejun Heo err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL); 198834dd82afSKay Sievers } 19891c500ad7STetsuo Handa mutex_unlock(&loop_ctl_mutex); 199034dd82afSKay Sievers if (err < 0) 19911c500ad7STetsuo Handa goto out_free_dev; 1992c718aa65STejun Heo i = err; 19931da177e4SLinus Torvalds 1994b5dd2f60SMing Lei lo->tag_set.ops = &loop_mq_ops; 1995b5dd2f60SMing Lei lo->tag_set.nr_hw_queues = 1; 1996ef44c508SChaitanya Kulkarni lo->tag_set.queue_depth = hw_queue_depth; 1997b5dd2f60SMing Lei lo->tag_set.numa_node = NUMA_NO_NODE; 1998b5dd2f60SMing Lei lo->tag_set.cmd_size = sizeof(struct loop_cmd); 19992112f5c1SBart Van Assche lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING | 20002112f5c1SBart Van Assche BLK_MQ_F_NO_SCHED_BY_DEFAULT; 2001b5dd2f60SMing Lei lo->tag_set.driver_data = lo; 2002b5dd2f60SMing Lei 2003b5dd2f60SMing Lei err = blk_mq_alloc_tag_set(&lo->tag_set); 2004b5dd2f60SMing Lei if (err) 20053ec981e3SMikulas Patocka goto out_free_idr; 200673285082SKen Chen 20071c99502fSChristoph Hellwig disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo); 20081c99502fSChristoph Hellwig if (IS_ERR(disk)) { 20091c99502fSChristoph Hellwig err = PTR_ERR(disk); 2010b5dd2f60SMing Lei goto out_cleanup_tags; 2011b5dd2f60SMing Lei } 20121c99502fSChristoph Hellwig lo->lo_queue = lo->lo_disk->queue; 2013ef7e7c82SMikulas Patocka 201454bb0adeSShaohua Li blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS); 201540326d8aSShaohua Li 20165b5e20f4SMing Lei /* 201740326d8aSShaohua Li * By default, we do buffer IO, so it doesn't make sense to enable 201840326d8aSShaohua Li * merge because the I/O submitted to backing file is handled page by 201940326d8aSShaohua Li * page. For directio mode, merge does help to dispatch bigger request 202040326d8aSShaohua Li * to underlayer disk. We will enable merge once directio is enabled. 20215b5e20f4SMing Lei */ 20228b904b5bSBart Van Assche blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue); 20235b5e20f4SMing Lei 2024e03c8dd1SKay Sievers /* 2025e03c8dd1SKay Sievers * Disable partition scanning by default. The in-kernel partition 2026e03c8dd1SKay Sievers * scanning can be requested individually per-device during its 2027e03c8dd1SKay Sievers * setup. Userspace can always add and remove partitions from all 2028e03c8dd1SKay Sievers * devices. The needed partition minors are allocated from the 2029e03c8dd1SKay Sievers * extended minor space, the main loop device numbers will continue 2030e03c8dd1SKay Sievers * to match the loop minors, regardless of the number of partitions 2031e03c8dd1SKay Sievers * used. 2032e03c8dd1SKay Sievers * 2033e03c8dd1SKay Sievers * If max_part is given, partition scanning is globally enabled for 2034e03c8dd1SKay Sievers * all loop devices. The minors for the main loop devices will be 2035e03c8dd1SKay Sievers * multiples of max_part. 2036e03c8dd1SKay Sievers * 2037e03c8dd1SKay Sievers * Note: Global-for-all-devices, set-only-at-init, read-only module 2038e03c8dd1SKay Sievers * parameteters like 'max_loop' and 'max_part' make things needlessly 2039e03c8dd1SKay Sievers * complicated, are too static, inflexible and may surprise 2040e03c8dd1SKay Sievers * userspace tools. Parameters like this in general should be avoided. 2041e03c8dd1SKay Sievers */ 2042e03c8dd1SKay Sievers if (!part_shift) 204346e7eac6SChristoph Hellwig disk->flags |= GENHD_FL_NO_PART; 2044f8933667SMing Lei atomic_set(&lo->lo_refcnt, 0); 20456cc8e743SPavel Tatashin mutex_init(&lo->lo_mutex); 20461da177e4SLinus Torvalds lo->lo_number = i; 20471da177e4SLinus Torvalds spin_lock_init(&lo->lo_lock); 204887579e9bSDan Schatzberg spin_lock_init(&lo->lo_work_lock); 20491da177e4SLinus Torvalds disk->major = LOOP_MAJOR; 2050476a4813SLaurent Vivier disk->first_minor = i << part_shift; 20511c99502fSChristoph Hellwig disk->minors = 1 << part_shift; 20521da177e4SLinus Torvalds disk->fops = &lo_fops; 20531da177e4SLinus Torvalds disk->private_data = lo; 20541da177e4SLinus Torvalds disk->queue = lo->lo_queue; 20559f65c489SMatteo Croce disk->events = DISK_EVENT_MEDIA_CHANGE; 20569f65c489SMatteo Croce disk->event_flags = DISK_EVENT_FLAG_UEVENT; 205773285082SKen Chen sprintf(disk->disk_name, "loop%d", i); 20581c500ad7STetsuo Handa /* Make this loop device reachable from pathname. */ 2059905705f0SLuis Chamberlain err = add_disk(disk); 2060905705f0SLuis Chamberlain if (err) 2061905705f0SLuis Chamberlain goto out_cleanup_disk; 2062905705f0SLuis Chamberlain 20631c500ad7STetsuo Handa /* Show this loop device. */ 20641c500ad7STetsuo Handa mutex_lock(&loop_ctl_mutex); 20651c500ad7STetsuo Handa lo->idr_visible = true; 206618d1f200SChristoph Hellwig mutex_unlock(&loop_ctl_mutex); 2067905705f0SLuis Chamberlain 206818d1f200SChristoph Hellwig return i; 206973285082SKen Chen 2070905705f0SLuis Chamberlain out_cleanup_disk: 2071905705f0SLuis Chamberlain blk_cleanup_disk(disk); 2072b5dd2f60SMing Lei out_cleanup_tags: 2073b5dd2f60SMing Lei blk_mq_free_tag_set(&lo->tag_set); 20743ec981e3SMikulas Patocka out_free_idr: 20751c500ad7STetsuo Handa mutex_lock(&loop_ctl_mutex); 20763ec981e3SMikulas Patocka idr_remove(&loop_index_idr, i); 207718d1f200SChristoph Hellwig mutex_unlock(&loop_ctl_mutex); 207873285082SKen Chen out_free_dev: 207973285082SKen Chen kfree(lo); 208073285082SKen Chen out: 208134dd82afSKay Sievers return err; 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds 208434dd82afSKay Sievers static void loop_remove(struct loop_device *lo) 208573285082SKen Chen { 20861c500ad7STetsuo Handa /* Make this loop device unreachable from pathname. */ 20876cd18e71SNeilBrown del_gendisk(lo->lo_disk); 20881c99502fSChristoph Hellwig blk_cleanup_disk(lo->lo_disk); 2089b5dd2f60SMing Lei blk_mq_free_tag_set(&lo->tag_set); 2090ce8d7861SChristoph Hellwig 20911c500ad7STetsuo Handa mutex_lock(&loop_ctl_mutex); 20921c500ad7STetsuo Handa idr_remove(&loop_index_idr, lo->lo_number); 20931c500ad7STetsuo Handa mutex_unlock(&loop_ctl_mutex); 20941c500ad7STetsuo Handa /* There is no route which can find this loop device. */ 20956cc8e743SPavel Tatashin mutex_destroy(&lo->lo_mutex); 209673285082SKen Chen kfree(lo); 209773285082SKen Chen } 209873285082SKen Chen 20998410d38cSChristoph Hellwig static void loop_probe(dev_t dev) 210073285082SKen Chen { 21018410d38cSChristoph Hellwig int idx = MINOR(dev) >> part_shift; 21028410d38cSChristoph Hellwig 21038410d38cSChristoph Hellwig if (max_loop && idx >= max_loop) 21048410d38cSChristoph Hellwig return; 2105d6da83d0SChristoph Hellwig loop_add(idx); 2106f9d10764SChristoph Hellwig } 2107f9d10764SChristoph Hellwig 2108f9d10764SChristoph Hellwig static int loop_control_remove(int idx) 2109770fe30aSKay Sievers { 2110770fe30aSKay Sievers struct loop_device *lo; 21110a42e99bSJan Kara int ret; 2112e5d66a10SChristoph Hellwig 2113e5d66a10SChristoph Hellwig if (idx < 0) { 2114e3f9387aSTetsuo Handa pr_warn_once("deleting an unspecified loop device is not supported.\n"); 2115e5d66a10SChristoph Hellwig return -EINVAL; 2116e5d66a10SChristoph Hellwig } 2117770fe30aSKay Sievers 21181c500ad7STetsuo Handa /* Hide this loop device for serialization. */ 21190a42e99bSJan Kara ret = mutex_lock_killable(&loop_ctl_mutex); 21200a42e99bSJan Kara if (ret) 21210a42e99bSJan Kara return ret; 2122b9848081SChristoph Hellwig lo = idr_find(&loop_index_idr, idx); 21231c500ad7STetsuo Handa if (!lo || !lo->idr_visible) 2124b9848081SChristoph Hellwig ret = -ENODEV; 21251c500ad7STetsuo Handa else 21261c500ad7STetsuo Handa lo->idr_visible = false; 21271c500ad7STetsuo Handa mutex_unlock(&loop_ctl_mutex); 21281c500ad7STetsuo Handa if (ret) 21291c500ad7STetsuo Handa return ret; 2130f9d10764SChristoph Hellwig 21311c500ad7STetsuo Handa /* Check whether this loop device can be removed. */ 21326cc8e743SPavel Tatashin ret = mutex_lock_killable(&lo->lo_mutex); 21336cc8e743SPavel Tatashin if (ret) 21341c500ad7STetsuo Handa goto mark_visible; 2135f9d10764SChristoph Hellwig if (lo->lo_state != Lo_unbound || 2136f9d10764SChristoph Hellwig atomic_read(&lo->lo_refcnt) > 0) { 21376cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 2138770fe30aSKay Sievers ret = -EBUSY; 21391c500ad7STetsuo Handa goto mark_visible; 2140770fe30aSKay Sievers } 21411c500ad7STetsuo Handa /* Mark this loop device no longer open()-able. */ 2142990e7811SChristoph Hellwig lo->lo_state = Lo_deleting; 21436cc8e743SPavel Tatashin mutex_unlock(&lo->lo_mutex); 2144f9d10764SChristoph Hellwig 2145770fe30aSKay Sievers loop_remove(lo); 21461c500ad7STetsuo Handa return 0; 21471c500ad7STetsuo Handa 21481c500ad7STetsuo Handa mark_visible: 21491c500ad7STetsuo Handa /* Show this loop device again. */ 21501c500ad7STetsuo Handa mutex_lock(&loop_ctl_mutex); 21511c500ad7STetsuo Handa lo->idr_visible = true; 2152f9d10764SChristoph Hellwig mutex_unlock(&loop_ctl_mutex); 2153f9d10764SChristoph Hellwig return ret; 2154770fe30aSKay Sievers } 2155f9d10764SChristoph Hellwig 2156f9d10764SChristoph Hellwig static int loop_control_get_free(int idx) 2157f9d10764SChristoph Hellwig { 2158f9d10764SChristoph Hellwig struct loop_device *lo; 2159b9848081SChristoph Hellwig int id, ret; 2160f9d10764SChristoph Hellwig 2161f9d10764SChristoph Hellwig ret = mutex_lock_killable(&loop_ctl_mutex); 2162f9d10764SChristoph Hellwig if (ret) 2163f9d10764SChristoph Hellwig return ret; 2164b9848081SChristoph Hellwig idr_for_each_entry(&loop_index_idr, lo, id) { 21651c500ad7STetsuo Handa /* Hitting a race results in creating a new loop device which is harmless. */ 21661c500ad7STetsuo Handa if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound) 2167b9848081SChristoph Hellwig goto found; 2168b9848081SChristoph Hellwig } 21690a42e99bSJan Kara mutex_unlock(&loop_ctl_mutex); 217018d1f200SChristoph Hellwig return loop_add(-1); 2171b9848081SChristoph Hellwig found: 2172b9848081SChristoph Hellwig mutex_unlock(&loop_ctl_mutex); 2173b9848081SChristoph Hellwig return id; 2174770fe30aSKay Sievers } 2175770fe30aSKay Sievers 2176f9d10764SChristoph Hellwig static long loop_control_ioctl(struct file *file, unsigned int cmd, 2177f9d10764SChristoph Hellwig unsigned long parm) 2178f9d10764SChristoph Hellwig { 2179f9d10764SChristoph Hellwig switch (cmd) { 2180f9d10764SChristoph Hellwig case LOOP_CTL_ADD: 218118d1f200SChristoph Hellwig return loop_add(parm); 2182f9d10764SChristoph Hellwig case LOOP_CTL_REMOVE: 2183f9d10764SChristoph Hellwig return loop_control_remove(parm); 2184f9d10764SChristoph Hellwig case LOOP_CTL_GET_FREE: 2185f9d10764SChristoph Hellwig return loop_control_get_free(parm); 2186f9d10764SChristoph Hellwig default: 2187f9d10764SChristoph Hellwig return -ENOSYS; 2188f9d10764SChristoph Hellwig } 2189f9d10764SChristoph Hellwig } 2190f9d10764SChristoph Hellwig 2191770fe30aSKay Sievers static const struct file_operations loop_ctl_fops = { 2192770fe30aSKay Sievers .open = nonseekable_open, 2193770fe30aSKay Sievers .unlocked_ioctl = loop_control_ioctl, 2194770fe30aSKay Sievers .compat_ioctl = loop_control_ioctl, 2195770fe30aSKay Sievers .owner = THIS_MODULE, 2196770fe30aSKay Sievers .llseek = noop_llseek, 2197770fe30aSKay Sievers }; 2198770fe30aSKay Sievers 2199770fe30aSKay Sievers static struct miscdevice loop_misc = { 2200770fe30aSKay Sievers .minor = LOOP_CTRL_MINOR, 2201770fe30aSKay Sievers .name = "loop-control", 2202770fe30aSKay Sievers .fops = &loop_ctl_fops, 2203770fe30aSKay Sievers }; 2204770fe30aSKay Sievers 2205770fe30aSKay Sievers MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR); 2206770fe30aSKay Sievers MODULE_ALIAS("devname:loop-control"); 2207770fe30aSKay Sievers 220873285082SKen Chen static int __init loop_init(void) 220973285082SKen Chen { 2210a47653fcSKen Chen int i, nr; 2211770fe30aSKay Sievers int err; 2212a47653fcSKen Chen 2213476a4813SLaurent Vivier part_shift = 0; 2214ac04fee0SNamhyung Kim if (max_part > 0) { 2215476a4813SLaurent Vivier part_shift = fls(max_part); 2216476a4813SLaurent Vivier 2217ac04fee0SNamhyung Kim /* 2218ac04fee0SNamhyung Kim * Adjust max_part according to part_shift as it is exported 2219ac04fee0SNamhyung Kim * to user space so that user can decide correct minor number 2220ac04fee0SNamhyung Kim * if [s]he want to create more devices. 2221ac04fee0SNamhyung Kim * 2222ac04fee0SNamhyung Kim * Note that -1 is required because partition 0 is reserved 2223ac04fee0SNamhyung Kim * for the whole disk. 2224ac04fee0SNamhyung Kim */ 2225ac04fee0SNamhyung Kim max_part = (1UL << part_shift) - 1; 2226ac04fee0SNamhyung Kim } 2227ac04fee0SNamhyung Kim 2228b1a66504SGuo Chao if ((1UL << part_shift) > DISK_MAX_PARTS) { 2229b1a66504SGuo Chao err = -EINVAL; 2230a8c1d064SAnton Volkov goto err_out; 2231b1a66504SGuo Chao } 223278f4bb36SNamhyung Kim 2233b1a66504SGuo Chao if (max_loop > 1UL << (MINORBITS - part_shift)) { 2234b1a66504SGuo Chao err = -EINVAL; 2235a8c1d064SAnton Volkov goto err_out; 2236b1a66504SGuo Chao } 223773285082SKen Chen 2238d134b00bSKay Sievers /* 2239d134b00bSKay Sievers * If max_loop is specified, create that many devices upfront. 2240d134b00bSKay Sievers * This also becomes a hard limit. If max_loop is not specified, 2241d134b00bSKay Sievers * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module 2242d134b00bSKay Sievers * init time. Loop devices can be requested on-demand with the 2243d134b00bSKay Sievers * /dev/loop-control interface, or be instantiated by accessing 2244d134b00bSKay Sievers * a 'dead' device node. 2245d134b00bSKay Sievers */ 2246aeb2b0b1SLukas Bulwahn if (max_loop) 2247a47653fcSKen Chen nr = max_loop; 2248aeb2b0b1SLukas Bulwahn else 2249d134b00bSKay Sievers nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT; 2250a47653fcSKen Chen 2251a8c1d064SAnton Volkov err = misc_register(&loop_misc); 2252a8c1d064SAnton Volkov if (err < 0) 2253a8c1d064SAnton Volkov goto err_out; 2254a8c1d064SAnton Volkov 2255a8c1d064SAnton Volkov 22568410d38cSChristoph Hellwig if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) { 2257b1a66504SGuo Chao err = -EIO; 2258b1a66504SGuo Chao goto misc_out; 2259b1a66504SGuo Chao } 2260a47653fcSKen Chen 2261d134b00bSKay Sievers /* pre-create number of devices given by config or max_loop */ 226234dd82afSKay Sievers for (i = 0; i < nr; i++) 2263d6da83d0SChristoph Hellwig loop_add(i); 226434dd82afSKay Sievers 226573285082SKen Chen printk(KERN_INFO "loop: module loaded\n"); 22661da177e4SLinus Torvalds return 0; 2267b1a66504SGuo Chao 2268b1a66504SGuo Chao misc_out: 2269b1a66504SGuo Chao misc_deregister(&loop_misc); 2270a8c1d064SAnton Volkov err_out: 2271b1a66504SGuo Chao return err; 227234dd82afSKay Sievers } 2273a47653fcSKen Chen 227473285082SKen Chen static void __exit loop_exit(void) 22751da177e4SLinus Torvalds { 22768e60947dSChristoph Hellwig struct loop_device *lo; 22778e60947dSChristoph Hellwig int id; 22788e60947dSChristoph Hellwig 22798b52d8beSChristoph Hellwig unregister_blkdev(LOOP_MAJOR, "loop"); 22808b52d8beSChristoph Hellwig misc_deregister(&loop_misc); 2281200f9337SLuis Chamberlain 22821c500ad7STetsuo Handa /* 22831c500ad7STetsuo Handa * There is no need to use loop_ctl_mutex here, for nobody else can 22841c500ad7STetsuo Handa * access loop_index_idr when this module is unloading (unless forced 22851c500ad7STetsuo Handa * module unloading is requested). If this is not a clean unloading, 22861c500ad7STetsuo Handa * we have no means to avoid kernel crash. 22871c500ad7STetsuo Handa */ 22888e60947dSChristoph Hellwig idr_for_each_entry(&loop_index_idr, lo, id) 22898e60947dSChristoph Hellwig loop_remove(lo); 2290bd5c39edSChristoph Hellwig 2291bd5c39edSChristoph Hellwig idr_destroy(&loop_index_idr); 22921da177e4SLinus Torvalds } 22931da177e4SLinus Torvalds 22941da177e4SLinus Torvalds module_init(loop_init); 22951da177e4SLinus Torvalds module_exit(loop_exit); 22961da177e4SLinus Torvalds 22971da177e4SLinus Torvalds #ifndef MODULE 22981da177e4SLinus Torvalds static int __init max_loop_setup(char *str) 22991da177e4SLinus Torvalds { 23001da177e4SLinus Torvalds max_loop = simple_strtol(str, NULL, 0); 23011da177e4SLinus Torvalds return 1; 23021da177e4SLinus Torvalds } 23031da177e4SLinus Torvalds 23041da177e4SLinus Torvalds __setup("max_loop=", max_loop_setup); 23051da177e4SLinus Torvalds #endif 2306