Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  * Copyright 1993 by Theodore Ts'o.
0004  */
0005 #include <linux/module.h>
0006 #include <linux/moduleparam.h>
0007 #include <linux/sched.h>
0008 #include <linux/fs.h>
0009 #include <linux/pagemap.h>
0010 #include <linux/file.h>
0011 #include <linux/stat.h>
0012 #include <linux/errno.h>
0013 #include <linux/major.h>
0014 #include <linux/wait.h>
0015 #include <linux/blkpg.h>
0016 #include <linux/init.h>
0017 #include <linux/swap.h>
0018 #include <linux/slab.h>
0019 #include <linux/compat.h>
0020 #include <linux/suspend.h>
0021 #include <linux/freezer.h>
0022 #include <linux/mutex.h>
0023 #include <linux/writeback.h>
0024 #include <linux/completion.h>
0025 #include <linux/highmem.h>
0026 #include <linux/splice.h>
0027 #include <linux/sysfs.h>
0028 #include <linux/miscdevice.h>
0029 #include <linux/falloc.h>
0030 #include <linux/uio.h>
0031 #include <linux/ioprio.h>
0032 #include <linux/blk-cgroup.h>
0033 #include <linux/sched/mm.h>
0034 #include <linux/statfs.h>
0035 #include <linux/uaccess.h>
0036 #include <linux/blk-mq.h>
0037 #include <linux/spinlock.h>
0038 #include <uapi/linux/loop.h>
0039 
0040 /* Possible states of device */
0041 enum {
0042     Lo_unbound,
0043     Lo_bound,
0044     Lo_rundown,
0045     Lo_deleting,
0046 };
0047 
0048 struct loop_func_table;
0049 
0050 struct loop_device {
0051     int     lo_number;
0052     loff_t      lo_offset;
0053     loff_t      lo_sizelimit;
0054     int     lo_flags;
0055     char        lo_file_name[LO_NAME_SIZE];
0056 
0057     struct file *   lo_backing_file;
0058     struct block_device *lo_device;
0059 
0060     gfp_t       old_gfp_mask;
0061 
0062     spinlock_t      lo_lock;
0063     int         lo_state;
0064     spinlock_t              lo_work_lock;
0065     struct workqueue_struct *workqueue;
0066     struct work_struct      rootcg_work;
0067     struct list_head        rootcg_cmd_list;
0068     struct list_head        idle_worker_list;
0069     struct rb_root          worker_tree;
0070     struct timer_list       timer;
0071     bool            use_dio;
0072     bool            sysfs_inited;
0073 
0074     struct request_queue    *lo_queue;
0075     struct blk_mq_tag_set   tag_set;
0076     struct gendisk      *lo_disk;
0077     struct mutex        lo_mutex;
0078     bool            idr_visible;
0079 };
0080 
0081 struct loop_cmd {
0082     struct list_head list_entry;
0083     bool use_aio; /* use AIO interface to handle I/O */
0084     atomic_t ref; /* only for aio */
0085     long ret;
0086     struct kiocb iocb;
0087     struct bio_vec *bvec;
0088     struct cgroup_subsys_state *blkcg_css;
0089     struct cgroup_subsys_state *memcg_css;
0090 };
0091 
0092 #define LOOP_IDLE_WORKER_TIMEOUT (60 * HZ)
0093 #define LOOP_DEFAULT_HW_Q_DEPTH (128)
0094 
0095 static DEFINE_IDR(loop_index_idr);
0096 static DEFINE_MUTEX(loop_ctl_mutex);
0097 static DEFINE_MUTEX(loop_validate_mutex);
0098 
0099 /**
0100  * loop_global_lock_killable() - take locks for safe loop_validate_file() test
0101  *
0102  * @lo: struct loop_device
0103  * @global: true if @lo is about to bind another "struct loop_device", false otherwise
0104  *
0105  * Returns 0 on success, -EINTR otherwise.
0106  *
0107  * Since loop_validate_file() traverses on other "struct loop_device" if
0108  * is_loop_device() is true, we need a global lock for serializing concurrent
0109  * loop_configure()/loop_change_fd()/__loop_clr_fd() calls.
0110  */
0111 static int loop_global_lock_killable(struct loop_device *lo, bool global)
0112 {
0113     int err;
0114 
0115     if (global) {
0116         err = mutex_lock_killable(&loop_validate_mutex);
0117         if (err)
0118             return err;
0119     }
0120     err = mutex_lock_killable(&lo->lo_mutex);
0121     if (err && global)
0122         mutex_unlock(&loop_validate_mutex);
0123     return err;
0124 }
0125 
0126 /**
0127  * loop_global_unlock() - release locks taken by loop_global_lock_killable()
0128  *
0129  * @lo: struct loop_device
0130  * @global: true if @lo was about to bind another "struct loop_device", false otherwise
0131  */
0132 static void loop_global_unlock(struct loop_device *lo, bool global)
0133 {
0134     mutex_unlock(&lo->lo_mutex);
0135     if (global)
0136         mutex_unlock(&loop_validate_mutex);
0137 }
0138 
0139 static int max_part;
0140 static int part_shift;
0141 
0142 static loff_t get_size(loff_t offset, loff_t sizelimit, struct file *file)
0143 {
0144     loff_t loopsize;
0145 
0146     /* Compute loopsize in bytes */
0147     loopsize = i_size_read(file->f_mapping->host);
0148     if (offset > 0)
0149         loopsize -= offset;
0150     /* offset is beyond i_size, weird but possible */
0151     if (loopsize < 0)
0152         return 0;
0153 
0154     if (sizelimit > 0 && sizelimit < loopsize)
0155         loopsize = sizelimit;
0156     /*
0157      * Unfortunately, if we want to do I/O on the device,
0158      * the number of 512-byte sectors has to fit into a sector_t.
0159      */
0160     return loopsize >> 9;
0161 }
0162 
0163 static loff_t get_loop_size(struct loop_device *lo, struct file *file)
0164 {
0165     return get_size(lo->lo_offset, lo->lo_sizelimit, file);
0166 }
0167 
0168 static void __loop_update_dio(struct loop_device *lo, bool dio)
0169 {
0170     struct file *file = lo->lo_backing_file;
0171     struct address_space *mapping = file->f_mapping;
0172     struct inode *inode = mapping->host;
0173     unsigned short sb_bsize = 0;
0174     unsigned dio_align = 0;
0175     bool use_dio;
0176 
0177     if (inode->i_sb->s_bdev) {
0178         sb_bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
0179         dio_align = sb_bsize - 1;
0180     }
0181 
0182     /*
0183      * We support direct I/O only if lo_offset is aligned with the
0184      * logical I/O size of backing device, and the logical block
0185      * size of loop is bigger than the backing device's.
0186      *
0187      * TODO: the above condition may be loosed in the future, and
0188      * direct I/O may be switched runtime at that time because most
0189      * of requests in sane applications should be PAGE_SIZE aligned
0190      */
0191     if (dio) {
0192         if (queue_logical_block_size(lo->lo_queue) >= sb_bsize &&
0193             !(lo->lo_offset & dio_align) &&
0194             (file->f_mode & FMODE_CAN_ODIRECT))
0195             use_dio = true;
0196         else
0197             use_dio = false;
0198     } else {
0199         use_dio = false;
0200     }
0201 
0202     if (lo->use_dio == use_dio)
0203         return;
0204 
0205     /* flush dirty pages before changing direct IO */
0206     vfs_fsync(file, 0);
0207 
0208     /*
0209      * The flag of LO_FLAGS_DIRECT_IO is handled similarly with
0210      * LO_FLAGS_READ_ONLY, both are set from kernel, and losetup
0211      * will get updated by ioctl(LOOP_GET_STATUS)
0212      */
0213     if (lo->lo_state == Lo_bound)
0214         blk_mq_freeze_queue(lo->lo_queue);
0215     lo->use_dio = use_dio;
0216     if (use_dio) {
0217         blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, lo->lo_queue);
0218         lo->lo_flags |= LO_FLAGS_DIRECT_IO;
0219     } else {
0220         blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
0221         lo->lo_flags &= ~LO_FLAGS_DIRECT_IO;
0222     }
0223     if (lo->lo_state == Lo_bound)
0224         blk_mq_unfreeze_queue(lo->lo_queue);
0225 }
0226 
0227 /**
0228  * loop_set_size() - sets device size and notifies userspace
0229  * @lo: struct loop_device to set the size for
0230  * @size: new size of the loop device
0231  *
0232  * Callers must validate that the size passed into this function fits into
0233  * a sector_t, eg using loop_validate_size()
0234  */
0235 static void loop_set_size(struct loop_device *lo, loff_t size)
0236 {
0237     if (!set_capacity_and_notify(lo->lo_disk, size))
0238         kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
0239 }
0240 
0241 static int lo_write_bvec(struct file *file, struct bio_vec *bvec, loff_t *ppos)
0242 {
0243     struct iov_iter i;
0244     ssize_t bw;
0245 
0246     iov_iter_bvec(&i, WRITE, bvec, 1, bvec->bv_len);
0247 
0248     file_start_write(file);
0249     bw = vfs_iter_write(file, &i, ppos, 0);
0250     file_end_write(file);
0251 
0252     if (likely(bw ==  bvec->bv_len))
0253         return 0;
0254 
0255     printk_ratelimited(KERN_ERR
0256         "loop: Write error at byte offset %llu, length %i.\n",
0257         (unsigned long long)*ppos, bvec->bv_len);
0258     if (bw >= 0)
0259         bw = -EIO;
0260     return bw;
0261 }
0262 
0263 static int lo_write_simple(struct loop_device *lo, struct request *rq,
0264         loff_t pos)
0265 {
0266     struct bio_vec bvec;
0267     struct req_iterator iter;
0268     int ret = 0;
0269 
0270     rq_for_each_segment(bvec, rq, iter) {
0271         ret = lo_write_bvec(lo->lo_backing_file, &bvec, &pos);
0272         if (ret < 0)
0273             break;
0274         cond_resched();
0275     }
0276 
0277     return ret;
0278 }
0279 
0280 static int lo_read_simple(struct loop_device *lo, struct request *rq,
0281         loff_t pos)
0282 {
0283     struct bio_vec bvec;
0284     struct req_iterator iter;
0285     struct iov_iter i;
0286     ssize_t len;
0287 
0288     rq_for_each_segment(bvec, rq, iter) {
0289         iov_iter_bvec(&i, READ, &bvec, 1, bvec.bv_len);
0290         len = vfs_iter_read(lo->lo_backing_file, &i, &pos, 0);
0291         if (len < 0)
0292             return len;
0293 
0294         flush_dcache_page(bvec.bv_page);
0295 
0296         if (len != bvec.bv_len) {
0297             struct bio *bio;
0298 
0299             __rq_for_each_bio(bio, rq)
0300                 zero_fill_bio(bio);
0301             break;
0302         }
0303         cond_resched();
0304     }
0305 
0306     return 0;
0307 }
0308 
0309 static int lo_fallocate(struct loop_device *lo, struct request *rq, loff_t pos,
0310             int mode)
0311 {
0312     /*
0313      * We use fallocate to manipulate the space mappings used by the image
0314      * a.k.a. discard/zerorange.
0315      */
0316     struct file *file = lo->lo_backing_file;
0317     int ret;
0318 
0319     mode |= FALLOC_FL_KEEP_SIZE;
0320 
0321     if (!bdev_max_discard_sectors(lo->lo_device))
0322         return -EOPNOTSUPP;
0323 
0324     ret = file->f_op->fallocate(file, mode, pos, blk_rq_bytes(rq));
0325     if (unlikely(ret && ret != -EINVAL && ret != -EOPNOTSUPP))
0326         return -EIO;
0327     return ret;
0328 }
0329 
0330 static int lo_req_flush(struct loop_device *lo, struct request *rq)
0331 {
0332     int ret = vfs_fsync(lo->lo_backing_file, 0);
0333     if (unlikely(ret && ret != -EINVAL))
0334         ret = -EIO;
0335 
0336     return ret;
0337 }
0338 
0339 static void lo_complete_rq(struct request *rq)
0340 {
0341     struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
0342     blk_status_t ret = BLK_STS_OK;
0343 
0344     if (!cmd->use_aio || cmd->ret < 0 || cmd->ret == blk_rq_bytes(rq) ||
0345         req_op(rq) != REQ_OP_READ) {
0346         if (cmd->ret < 0)
0347             ret = errno_to_blk_status(cmd->ret);
0348         goto end_io;
0349     }
0350 
0351     /*
0352      * Short READ - if we got some data, advance our request and
0353      * retry it. If we got no data, end the rest with EIO.
0354      */
0355     if (cmd->ret) {
0356         blk_update_request(rq, BLK_STS_OK, cmd->ret);
0357         cmd->ret = 0;
0358         blk_mq_requeue_request(rq, true);
0359     } else {
0360         if (cmd->use_aio) {
0361             struct bio *bio = rq->bio;
0362 
0363             while (bio) {
0364                 zero_fill_bio(bio);
0365                 bio = bio->bi_next;
0366             }
0367         }
0368         ret = BLK_STS_IOERR;
0369 end_io:
0370         blk_mq_end_request(rq, ret);
0371     }
0372 }
0373 
0374 static void lo_rw_aio_do_completion(struct loop_cmd *cmd)
0375 {
0376     struct request *rq = blk_mq_rq_from_pdu(cmd);
0377 
0378     if (!atomic_dec_and_test(&cmd->ref))
0379         return;
0380     kfree(cmd->bvec);
0381     cmd->bvec = NULL;
0382     if (likely(!blk_should_fake_timeout(rq->q)))
0383         blk_mq_complete_request(rq);
0384 }
0385 
0386 static void lo_rw_aio_complete(struct kiocb *iocb, long ret)
0387 {
0388     struct loop_cmd *cmd = container_of(iocb, struct loop_cmd, iocb);
0389 
0390     cmd->ret = ret;
0391     lo_rw_aio_do_completion(cmd);
0392 }
0393 
0394 static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
0395              loff_t pos, bool rw)
0396 {
0397     struct iov_iter iter;
0398     struct req_iterator rq_iter;
0399     struct bio_vec *bvec;
0400     struct request *rq = blk_mq_rq_from_pdu(cmd);
0401     struct bio *bio = rq->bio;
0402     struct file *file = lo->lo_backing_file;
0403     struct bio_vec tmp;
0404     unsigned int offset;
0405     int nr_bvec = 0;
0406     int ret;
0407 
0408     rq_for_each_bvec(tmp, rq, rq_iter)
0409         nr_bvec++;
0410 
0411     if (rq->bio != rq->biotail) {
0412 
0413         bvec = kmalloc_array(nr_bvec, sizeof(struct bio_vec),
0414                      GFP_NOIO);
0415         if (!bvec)
0416             return -EIO;
0417         cmd->bvec = bvec;
0418 
0419         /*
0420          * The bios of the request may be started from the middle of
0421          * the 'bvec' because of bio splitting, so we can't directly
0422          * copy bio->bi_iov_vec to new bvec. The rq_for_each_bvec
0423          * API will take care of all details for us.
0424          */
0425         rq_for_each_bvec(tmp, rq, rq_iter) {
0426             *bvec = tmp;
0427             bvec++;
0428         }
0429         bvec = cmd->bvec;
0430         offset = 0;
0431     } else {
0432         /*
0433          * Same here, this bio may be started from the middle of the
0434          * 'bvec' because of bio splitting, so offset from the bvec
0435          * must be passed to iov iterator
0436          */
0437         offset = bio->bi_iter.bi_bvec_done;
0438         bvec = __bvec_iter_bvec(bio->bi_io_vec, bio->bi_iter);
0439     }
0440     atomic_set(&cmd->ref, 2);
0441 
0442     iov_iter_bvec(&iter, rw, bvec, nr_bvec, blk_rq_bytes(rq));
0443     iter.iov_offset = offset;
0444 
0445     cmd->iocb.ki_pos = pos;
0446     cmd->iocb.ki_filp = file;
0447     cmd->iocb.ki_complete = lo_rw_aio_complete;
0448     cmd->iocb.ki_flags = IOCB_DIRECT;
0449     cmd->iocb.ki_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, 0);
0450 
0451     if (rw == WRITE)
0452         ret = call_write_iter(file, &cmd->iocb, &iter);
0453     else
0454         ret = call_read_iter(file, &cmd->iocb, &iter);
0455 
0456     lo_rw_aio_do_completion(cmd);
0457 
0458     if (ret != -EIOCBQUEUED)
0459         lo_rw_aio_complete(&cmd->iocb, ret);
0460     return 0;
0461 }
0462 
0463 static int do_req_filebacked(struct loop_device *lo, struct request *rq)
0464 {
0465     struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
0466     loff_t pos = ((loff_t) blk_rq_pos(rq) << 9) + lo->lo_offset;
0467 
0468     /*
0469      * lo_write_simple and lo_read_simple should have been covered
0470      * by io submit style function like lo_rw_aio(), one blocker
0471      * is that lo_read_simple() need to call flush_dcache_page after
0472      * the page is written from kernel, and it isn't easy to handle
0473      * this in io submit style function which submits all segments
0474      * of the req at one time. And direct read IO doesn't need to
0475      * run flush_dcache_page().
0476      */
0477     switch (req_op(rq)) {
0478     case REQ_OP_FLUSH:
0479         return lo_req_flush(lo, rq);
0480     case REQ_OP_WRITE_ZEROES:
0481         /*
0482          * If the caller doesn't want deallocation, call zeroout to
0483          * write zeroes the range.  Otherwise, punch them out.
0484          */
0485         return lo_fallocate(lo, rq, pos,
0486             (rq->cmd_flags & REQ_NOUNMAP) ?
0487                 FALLOC_FL_ZERO_RANGE :
0488                 FALLOC_FL_PUNCH_HOLE);
0489     case REQ_OP_DISCARD:
0490         return lo_fallocate(lo, rq, pos, FALLOC_FL_PUNCH_HOLE);
0491     case REQ_OP_WRITE:
0492         if (cmd->use_aio)
0493             return lo_rw_aio(lo, cmd, pos, WRITE);
0494         else
0495             return lo_write_simple(lo, rq, pos);
0496     case REQ_OP_READ:
0497         if (cmd->use_aio)
0498             return lo_rw_aio(lo, cmd, pos, READ);
0499         else
0500             return lo_read_simple(lo, rq, pos);
0501     default:
0502         WARN_ON_ONCE(1);
0503         return -EIO;
0504     }
0505 }
0506 
0507 static inline void loop_update_dio(struct loop_device *lo)
0508 {
0509     __loop_update_dio(lo, (lo->lo_backing_file->f_flags & O_DIRECT) |
0510                 lo->use_dio);
0511 }
0512 
0513 static void loop_reread_partitions(struct loop_device *lo)
0514 {
0515     int rc;
0516 
0517     mutex_lock(&lo->lo_disk->open_mutex);
0518     rc = bdev_disk_changed(lo->lo_disk, false);
0519     mutex_unlock(&lo->lo_disk->open_mutex);
0520     if (rc)
0521         pr_warn("%s: partition scan of loop%d (%s) failed (rc=%d)\n",
0522             __func__, lo->lo_number, lo->lo_file_name, rc);
0523 }
0524 
0525 static inline int is_loop_device(struct file *file)
0526 {
0527     struct inode *i = file->f_mapping->host;
0528 
0529     return i && S_ISBLK(i->i_mode) && imajor(i) == LOOP_MAJOR;
0530 }
0531 
0532 static int loop_validate_file(struct file *file, struct block_device *bdev)
0533 {
0534     struct inode    *inode = file->f_mapping->host;
0535     struct file *f = file;
0536 
0537     /* Avoid recursion */
0538     while (is_loop_device(f)) {
0539         struct loop_device *l;
0540 
0541         lockdep_assert_held(&loop_validate_mutex);
0542         if (f->f_mapping->host->i_rdev == bdev->bd_dev)
0543             return -EBADF;
0544 
0545         l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
0546         if (l->lo_state != Lo_bound)
0547             return -EINVAL;
0548         /* Order wrt setting lo->lo_backing_file in loop_configure(). */
0549         rmb();
0550         f = l->lo_backing_file;
0551     }
0552     if (!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))
0553         return -EINVAL;
0554     return 0;
0555 }
0556 
0557 /*
0558  * loop_change_fd switched the backing store of a loopback device to
0559  * a new file. This is useful for operating system installers to free up
0560  * the original file and in High Availability environments to switch to
0561  * an alternative location for the content in case of server meltdown.
0562  * This can only work if the loop device is used read-only, and if the
0563  * new backing store is the same size and type as the old backing store.
0564  */
0565 static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
0566               unsigned int arg)
0567 {
0568     struct file *file = fget(arg);
0569     struct file *old_file;
0570     int error;
0571     bool partscan;
0572     bool is_loop;
0573 
0574     if (!file)
0575         return -EBADF;
0576 
0577     /* suppress uevents while reconfiguring the device */
0578     dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
0579 
0580     is_loop = is_loop_device(file);
0581     error = loop_global_lock_killable(lo, is_loop);
0582     if (error)
0583         goto out_putf;
0584     error = -ENXIO;
0585     if (lo->lo_state != Lo_bound)
0586         goto out_err;
0587 
0588     /* the loop device has to be read-only */
0589     error = -EINVAL;
0590     if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
0591         goto out_err;
0592 
0593     error = loop_validate_file(file, bdev);
0594     if (error)
0595         goto out_err;
0596 
0597     old_file = lo->lo_backing_file;
0598 
0599     error = -EINVAL;
0600 
0601     /* size of the new backing store needs to be the same */
0602     if (get_loop_size(lo, file) != get_loop_size(lo, old_file))
0603         goto out_err;
0604 
0605     /* and ... switch */
0606     disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
0607     blk_mq_freeze_queue(lo->lo_queue);
0608     mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
0609     lo->lo_backing_file = file;
0610     lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
0611     mapping_set_gfp_mask(file->f_mapping,
0612                  lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
0613     loop_update_dio(lo);
0614     blk_mq_unfreeze_queue(lo->lo_queue);
0615     partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
0616     loop_global_unlock(lo, is_loop);
0617 
0618     /*
0619      * Flush loop_validate_file() before fput(), for l->lo_backing_file
0620      * might be pointing at old_file which might be the last reference.
0621      */
0622     if (!is_loop) {
0623         mutex_lock(&loop_validate_mutex);
0624         mutex_unlock(&loop_validate_mutex);
0625     }
0626     /*
0627      * We must drop file reference outside of lo_mutex as dropping
0628      * the file ref can take open_mutex which creates circular locking
0629      * dependency.
0630      */
0631     fput(old_file);
0632     if (partscan)
0633         loop_reread_partitions(lo);
0634 
0635     error = 0;
0636 done:
0637     /* enable and uncork uevent now that we are done */
0638     dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
0639     return error;
0640 
0641 out_err:
0642     loop_global_unlock(lo, is_loop);
0643 out_putf:
0644     fput(file);
0645     goto done;
0646 }
0647 
0648 /* loop sysfs attributes */
0649 
0650 static ssize_t loop_attr_show(struct device *dev, char *page,
0651                   ssize_t (*callback)(struct loop_device *, char *))
0652 {
0653     struct gendisk *disk = dev_to_disk(dev);
0654     struct loop_device *lo = disk->private_data;
0655 
0656     return callback(lo, page);
0657 }
0658 
0659 #define LOOP_ATTR_RO(_name)                     \
0660 static ssize_t loop_attr_##_name##_show(struct loop_device *, char *);  \
0661 static ssize_t loop_attr_do_show_##_name(struct device *d,      \
0662                 struct device_attribute *attr, char *b) \
0663 {                                   \
0664     return loop_attr_show(d, b, loop_attr_##_name##_show);      \
0665 }                                   \
0666 static struct device_attribute loop_attr_##_name =          \
0667     __ATTR(_name, 0444, loop_attr_do_show_##_name, NULL);
0668 
0669 static ssize_t loop_attr_backing_file_show(struct loop_device *lo, char *buf)
0670 {
0671     ssize_t ret;
0672     char *p = NULL;
0673 
0674     spin_lock_irq(&lo->lo_lock);
0675     if (lo->lo_backing_file)
0676         p = file_path(lo->lo_backing_file, buf, PAGE_SIZE - 1);
0677     spin_unlock_irq(&lo->lo_lock);
0678 
0679     if (IS_ERR_OR_NULL(p))
0680         ret = PTR_ERR(p);
0681     else {
0682         ret = strlen(p);
0683         memmove(buf, p, ret);
0684         buf[ret++] = '\n';
0685         buf[ret] = 0;
0686     }
0687 
0688     return ret;
0689 }
0690 
0691 static ssize_t loop_attr_offset_show(struct loop_device *lo, char *buf)
0692 {
0693     return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_offset);
0694 }
0695 
0696 static ssize_t loop_attr_sizelimit_show(struct loop_device *lo, char *buf)
0697 {
0698     return sysfs_emit(buf, "%llu\n", (unsigned long long)lo->lo_sizelimit);
0699 }
0700 
0701 static ssize_t loop_attr_autoclear_show(struct loop_device *lo, char *buf)
0702 {
0703     int autoclear = (lo->lo_flags & LO_FLAGS_AUTOCLEAR);
0704 
0705     return sysfs_emit(buf, "%s\n", autoclear ? "1" : "0");
0706 }
0707 
0708 static ssize_t loop_attr_partscan_show(struct loop_device *lo, char *buf)
0709 {
0710     int partscan = (lo->lo_flags & LO_FLAGS_PARTSCAN);
0711 
0712     return sysfs_emit(buf, "%s\n", partscan ? "1" : "0");
0713 }
0714 
0715 static ssize_t loop_attr_dio_show(struct loop_device *lo, char *buf)
0716 {
0717     int dio = (lo->lo_flags & LO_FLAGS_DIRECT_IO);
0718 
0719     return sysfs_emit(buf, "%s\n", dio ? "1" : "0");
0720 }
0721 
0722 LOOP_ATTR_RO(backing_file);
0723 LOOP_ATTR_RO(offset);
0724 LOOP_ATTR_RO(sizelimit);
0725 LOOP_ATTR_RO(autoclear);
0726 LOOP_ATTR_RO(partscan);
0727 LOOP_ATTR_RO(dio);
0728 
0729 static struct attribute *loop_attrs[] = {
0730     &loop_attr_backing_file.attr,
0731     &loop_attr_offset.attr,
0732     &loop_attr_sizelimit.attr,
0733     &loop_attr_autoclear.attr,
0734     &loop_attr_partscan.attr,
0735     &loop_attr_dio.attr,
0736     NULL,
0737 };
0738 
0739 static struct attribute_group loop_attribute_group = {
0740     .name = "loop",
0741     .attrs= loop_attrs,
0742 };
0743 
0744 static void loop_sysfs_init(struct loop_device *lo)
0745 {
0746     lo->sysfs_inited = !sysfs_create_group(&disk_to_dev(lo->lo_disk)->kobj,
0747                         &loop_attribute_group);
0748 }
0749 
0750 static void loop_sysfs_exit(struct loop_device *lo)
0751 {
0752     if (lo->sysfs_inited)
0753         sysfs_remove_group(&disk_to_dev(lo->lo_disk)->kobj,
0754                    &loop_attribute_group);
0755 }
0756 
0757 static void loop_config_discard(struct loop_device *lo)
0758 {
0759     struct file *file = lo->lo_backing_file;
0760     struct inode *inode = file->f_mapping->host;
0761     struct request_queue *q = lo->lo_queue;
0762     u32 granularity, max_discard_sectors;
0763 
0764     /*
0765      * If the backing device is a block device, mirror its zeroing
0766      * capability. Set the discard sectors to the block device's zeroing
0767      * capabilities because loop discards result in blkdev_issue_zeroout(),
0768      * not blkdev_issue_discard(). This maintains consistent behavior with
0769      * file-backed loop devices: discarded regions read back as zero.
0770      */
0771     if (S_ISBLK(inode->i_mode)) {
0772         struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
0773 
0774         max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
0775         granularity = bdev_discard_granularity(I_BDEV(inode)) ?:
0776             queue_physical_block_size(backingq);
0777 
0778     /*
0779      * We use punch hole to reclaim the free space used by the
0780      * image a.k.a. discard.
0781      */
0782     } else if (!file->f_op->fallocate) {
0783         max_discard_sectors = 0;
0784         granularity = 0;
0785 
0786     } else {
0787         struct kstatfs sbuf;
0788 
0789         max_discard_sectors = UINT_MAX >> 9;
0790         if (!vfs_statfs(&file->f_path, &sbuf))
0791             granularity = sbuf.f_bsize;
0792         else
0793             max_discard_sectors = 0;
0794     }
0795 
0796     if (max_discard_sectors) {
0797         q->limits.discard_granularity = granularity;
0798         blk_queue_max_discard_sectors(q, max_discard_sectors);
0799         blk_queue_max_write_zeroes_sectors(q, max_discard_sectors);
0800     } else {
0801         q->limits.discard_granularity = 0;
0802         blk_queue_max_discard_sectors(q, 0);
0803         blk_queue_max_write_zeroes_sectors(q, 0);
0804     }
0805 }
0806 
0807 struct loop_worker {
0808     struct rb_node rb_node;
0809     struct work_struct work;
0810     struct list_head cmd_list;
0811     struct list_head idle_list;
0812     struct loop_device *lo;
0813     struct cgroup_subsys_state *blkcg_css;
0814     unsigned long last_ran_at;
0815 };
0816 
0817 static void loop_workfn(struct work_struct *work);
0818 
0819 #ifdef CONFIG_BLK_CGROUP
0820 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
0821 {
0822     return !css || css == blkcg_root_css;
0823 }
0824 #else
0825 static inline int queue_on_root_worker(struct cgroup_subsys_state *css)
0826 {
0827     return !css;
0828 }
0829 #endif
0830 
0831 static void loop_queue_work(struct loop_device *lo, struct loop_cmd *cmd)
0832 {
0833     struct rb_node **node, *parent = NULL;
0834     struct loop_worker *cur_worker, *worker = NULL;
0835     struct work_struct *work;
0836     struct list_head *cmd_list;
0837 
0838     spin_lock_irq(&lo->lo_work_lock);
0839 
0840     if (queue_on_root_worker(cmd->blkcg_css))
0841         goto queue_work;
0842 
0843     node = &lo->worker_tree.rb_node;
0844 
0845     while (*node) {
0846         parent = *node;
0847         cur_worker = container_of(*node, struct loop_worker, rb_node);
0848         if (cur_worker->blkcg_css == cmd->blkcg_css) {
0849             worker = cur_worker;
0850             break;
0851         } else if ((long)cur_worker->blkcg_css < (long)cmd->blkcg_css) {
0852             node = &(*node)->rb_left;
0853         } else {
0854             node = &(*node)->rb_right;
0855         }
0856     }
0857     if (worker)
0858         goto queue_work;
0859 
0860     worker = kzalloc(sizeof(struct loop_worker), GFP_NOWAIT | __GFP_NOWARN);
0861     /*
0862      * In the event we cannot allocate a worker, just queue on the
0863      * rootcg worker and issue the I/O as the rootcg
0864      */
0865     if (!worker) {
0866         cmd->blkcg_css = NULL;
0867         if (cmd->memcg_css)
0868             css_put(cmd->memcg_css);
0869         cmd->memcg_css = NULL;
0870         goto queue_work;
0871     }
0872 
0873     worker->blkcg_css = cmd->blkcg_css;
0874     css_get(worker->blkcg_css);
0875     INIT_WORK(&worker->work, loop_workfn);
0876     INIT_LIST_HEAD(&worker->cmd_list);
0877     INIT_LIST_HEAD(&worker->idle_list);
0878     worker->lo = lo;
0879     rb_link_node(&worker->rb_node, parent, node);
0880     rb_insert_color(&worker->rb_node, &lo->worker_tree);
0881 queue_work:
0882     if (worker) {
0883         /*
0884          * We need to remove from the idle list here while
0885          * holding the lock so that the idle timer doesn't
0886          * free the worker
0887          */
0888         if (!list_empty(&worker->idle_list))
0889             list_del_init(&worker->idle_list);
0890         work = &worker->work;
0891         cmd_list = &worker->cmd_list;
0892     } else {
0893         work = &lo->rootcg_work;
0894         cmd_list = &lo->rootcg_cmd_list;
0895     }
0896     list_add_tail(&cmd->list_entry, cmd_list);
0897     queue_work(lo->workqueue, work);
0898     spin_unlock_irq(&lo->lo_work_lock);
0899 }
0900 
0901 static void loop_set_timer(struct loop_device *lo)
0902 {
0903     timer_reduce(&lo->timer, jiffies + LOOP_IDLE_WORKER_TIMEOUT);
0904 }
0905 
0906 static void loop_free_idle_workers(struct loop_device *lo, bool delete_all)
0907 {
0908     struct loop_worker *pos, *worker;
0909 
0910     spin_lock_irq(&lo->lo_work_lock);
0911     list_for_each_entry_safe(worker, pos, &lo->idle_worker_list,
0912                 idle_list) {
0913         if (!delete_all &&
0914             time_is_after_jiffies(worker->last_ran_at +
0915                       LOOP_IDLE_WORKER_TIMEOUT))
0916             break;
0917         list_del(&worker->idle_list);
0918         rb_erase(&worker->rb_node, &lo->worker_tree);
0919         css_put(worker->blkcg_css);
0920         kfree(worker);
0921     }
0922     if (!list_empty(&lo->idle_worker_list))
0923         loop_set_timer(lo);
0924     spin_unlock_irq(&lo->lo_work_lock);
0925 }
0926 
0927 static void loop_free_idle_workers_timer(struct timer_list *timer)
0928 {
0929     struct loop_device *lo = container_of(timer, struct loop_device, timer);
0930 
0931     return loop_free_idle_workers(lo, false);
0932 }
0933 
0934 static void loop_update_rotational(struct loop_device *lo)
0935 {
0936     struct file *file = lo->lo_backing_file;
0937     struct inode *file_inode = file->f_mapping->host;
0938     struct block_device *file_bdev = file_inode->i_sb->s_bdev;
0939     struct request_queue *q = lo->lo_queue;
0940     bool nonrot = true;
0941 
0942     /* not all filesystems (e.g. tmpfs) have a sb->s_bdev */
0943     if (file_bdev)
0944         nonrot = bdev_nonrot(file_bdev);
0945 
0946     if (nonrot)
0947         blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
0948     else
0949         blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
0950 }
0951 
0952 /**
0953  * loop_set_status_from_info - configure device from loop_info
0954  * @lo: struct loop_device to configure
0955  * @info: struct loop_info64 to configure the device with
0956  *
0957  * Configures the loop device parameters according to the passed
0958  * in loop_info64 configuration.
0959  */
0960 static int
0961 loop_set_status_from_info(struct loop_device *lo,
0962               const struct loop_info64 *info)
0963 {
0964     if ((unsigned int) info->lo_encrypt_key_size > LO_KEY_SIZE)
0965         return -EINVAL;
0966 
0967     switch (info->lo_encrypt_type) {
0968     case LO_CRYPT_NONE:
0969         break;
0970     case LO_CRYPT_XOR:
0971         pr_warn("support for the xor transformation has been removed.\n");
0972         return -EINVAL;
0973     case LO_CRYPT_CRYPTOAPI:
0974         pr_warn("support for cryptoloop has been removed.  Use dm-crypt instead.\n");
0975         return -EINVAL;
0976     default:
0977         return -EINVAL;
0978     }
0979 
0980     lo->lo_offset = info->lo_offset;
0981     lo->lo_sizelimit = info->lo_sizelimit;
0982 
0983     /* loff_t vars have been assigned __u64 */
0984     if (lo->lo_offset < 0 || lo->lo_sizelimit < 0)
0985         return -EOVERFLOW;
0986 
0987     memcpy(lo->lo_file_name, info->lo_file_name, LO_NAME_SIZE);
0988     lo->lo_file_name[LO_NAME_SIZE-1] = 0;
0989     lo->lo_flags = info->lo_flags;
0990     return 0;
0991 }
0992 
0993 static int loop_configure(struct loop_device *lo, fmode_t mode,
0994               struct block_device *bdev,
0995               const struct loop_config *config)
0996 {
0997     struct file *file = fget(config->fd);
0998     struct inode *inode;
0999     struct address_space *mapping;
1000     int error;
1001     loff_t size;
1002     bool partscan;
1003     unsigned short bsize;
1004     bool is_loop;
1005 
1006     if (!file)
1007         return -EBADF;
1008     is_loop = is_loop_device(file);
1009 
1010     /* This is safe, since we have a reference from open(). */
1011     __module_get(THIS_MODULE);
1012 
1013     /* suppress uevents while reconfiguring the device */
1014     dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 1);
1015 
1016     /*
1017      * If we don't hold exclusive handle for the device, upgrade to it
1018      * here to avoid changing device under exclusive owner.
1019      */
1020     if (!(mode & FMODE_EXCL)) {
1021         error = bd_prepare_to_claim(bdev, loop_configure);
1022         if (error)
1023             goto out_putf;
1024     }
1025 
1026     error = loop_global_lock_killable(lo, is_loop);
1027     if (error)
1028         goto out_bdev;
1029 
1030     error = -EBUSY;
1031     if (lo->lo_state != Lo_unbound)
1032         goto out_unlock;
1033 
1034     error = loop_validate_file(file, bdev);
1035     if (error)
1036         goto out_unlock;
1037 
1038     mapping = file->f_mapping;
1039     inode = mapping->host;
1040 
1041     if ((config->info.lo_flags & ~LOOP_CONFIGURE_SETTABLE_FLAGS) != 0) {
1042         error = -EINVAL;
1043         goto out_unlock;
1044     }
1045 
1046     if (config->block_size) {
1047         error = blk_validate_block_size(config->block_size);
1048         if (error)
1049             goto out_unlock;
1050     }
1051 
1052     error = loop_set_status_from_info(lo, &config->info);
1053     if (error)
1054         goto out_unlock;
1055 
1056     if (!(file->f_mode & FMODE_WRITE) || !(mode & FMODE_WRITE) ||
1057         !file->f_op->write_iter)
1058         lo->lo_flags |= LO_FLAGS_READ_ONLY;
1059 
1060     if (!lo->workqueue) {
1061         lo->workqueue = alloc_workqueue("loop%d",
1062                         WQ_UNBOUND | WQ_FREEZABLE,
1063                         0, lo->lo_number);
1064         if (!lo->workqueue) {
1065             error = -ENOMEM;
1066             goto out_unlock;
1067         }
1068     }
1069 
1070     disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1071     set_disk_ro(lo->lo_disk, (lo->lo_flags & LO_FLAGS_READ_ONLY) != 0);
1072 
1073     lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
1074     lo->lo_device = bdev;
1075     lo->lo_backing_file = file;
1076     lo->old_gfp_mask = mapping_gfp_mask(mapping);
1077     mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
1078 
1079     if (!(lo->lo_flags & LO_FLAGS_READ_ONLY) && file->f_op->fsync)
1080         blk_queue_write_cache(lo->lo_queue, true, false);
1081 
1082     if (config->block_size)
1083         bsize = config->block_size;
1084     else if ((lo->lo_backing_file->f_flags & O_DIRECT) && inode->i_sb->s_bdev)
1085         /* In case of direct I/O, match underlying block size */
1086         bsize = bdev_logical_block_size(inode->i_sb->s_bdev);
1087     else
1088         bsize = 512;
1089 
1090     blk_queue_logical_block_size(lo->lo_queue, bsize);
1091     blk_queue_physical_block_size(lo->lo_queue, bsize);
1092     blk_queue_io_min(lo->lo_queue, bsize);
1093 
1094     loop_config_discard(lo);
1095     loop_update_rotational(lo);
1096     loop_update_dio(lo);
1097     loop_sysfs_init(lo);
1098 
1099     size = get_loop_size(lo, file);
1100     loop_set_size(lo, size);
1101 
1102     /* Order wrt reading lo_state in loop_validate_file(). */
1103     wmb();
1104 
1105     lo->lo_state = Lo_bound;
1106     if (part_shift)
1107         lo->lo_flags |= LO_FLAGS_PARTSCAN;
1108     partscan = lo->lo_flags & LO_FLAGS_PARTSCAN;
1109     if (partscan)
1110         clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1111 
1112     loop_global_unlock(lo, is_loop);
1113     if (partscan)
1114         loop_reread_partitions(lo);
1115     if (!(mode & FMODE_EXCL))
1116         bd_abort_claiming(bdev, loop_configure);
1117 
1118     error = 0;
1119 done:
1120     /* enable and uncork uevent now that we are done */
1121     dev_set_uevent_suppress(disk_to_dev(lo->lo_disk), 0);
1122     return error;
1123 
1124 out_unlock:
1125     loop_global_unlock(lo, is_loop);
1126 out_bdev:
1127     if (!(mode & FMODE_EXCL))
1128         bd_abort_claiming(bdev, loop_configure);
1129 out_putf:
1130     fput(file);
1131     /* This is safe: open() is still holding a reference. */
1132     module_put(THIS_MODULE);
1133     goto done;
1134 }
1135 
1136 static void __loop_clr_fd(struct loop_device *lo, bool release)
1137 {
1138     struct file *filp;
1139     gfp_t gfp = lo->old_gfp_mask;
1140 
1141     if (test_bit(QUEUE_FLAG_WC, &lo->lo_queue->queue_flags))
1142         blk_queue_write_cache(lo->lo_queue, false, false);
1143 
1144     /*
1145      * Freeze the request queue when unbinding on a live file descriptor and
1146      * thus an open device.  When called from ->release we are guaranteed
1147      * that there is no I/O in progress already.
1148      */
1149     if (!release)
1150         blk_mq_freeze_queue(lo->lo_queue);
1151 
1152     spin_lock_irq(&lo->lo_lock);
1153     filp = lo->lo_backing_file;
1154     lo->lo_backing_file = NULL;
1155     spin_unlock_irq(&lo->lo_lock);
1156 
1157     lo->lo_device = NULL;
1158     lo->lo_offset = 0;
1159     lo->lo_sizelimit = 0;
1160     memset(lo->lo_file_name, 0, LO_NAME_SIZE);
1161     blk_queue_logical_block_size(lo->lo_queue, 512);
1162     blk_queue_physical_block_size(lo->lo_queue, 512);
1163     blk_queue_io_min(lo->lo_queue, 512);
1164     invalidate_disk(lo->lo_disk);
1165     loop_sysfs_exit(lo);
1166     /* let user-space know about this change */
1167     kobject_uevent(&disk_to_dev(lo->lo_disk)->kobj, KOBJ_CHANGE);
1168     mapping_set_gfp_mask(filp->f_mapping, gfp);
1169     /* This is safe: open() is still holding a reference. */
1170     module_put(THIS_MODULE);
1171     if (!release)
1172         blk_mq_unfreeze_queue(lo->lo_queue);
1173 
1174     disk_force_media_change(lo->lo_disk, DISK_EVENT_MEDIA_CHANGE);
1175 
1176     if (lo->lo_flags & LO_FLAGS_PARTSCAN) {
1177         int err;
1178 
1179         /*
1180          * open_mutex has been held already in release path, so don't
1181          * acquire it if this function is called in such case.
1182          *
1183          * If the reread partition isn't from release path, lo_refcnt
1184          * must be at least one and it can only become zero when the
1185          * current holder is released.
1186          */
1187         if (!release)
1188             mutex_lock(&lo->lo_disk->open_mutex);
1189         err = bdev_disk_changed(lo->lo_disk, false);
1190         if (!release)
1191             mutex_unlock(&lo->lo_disk->open_mutex);
1192         if (err)
1193             pr_warn("%s: partition scan of loop%d failed (rc=%d)\n",
1194                 __func__, lo->lo_number, err);
1195         /* Device is gone, no point in returning error */
1196     }
1197 
1198     /*
1199      * lo->lo_state is set to Lo_unbound here after above partscan has
1200      * finished. There cannot be anybody else entering __loop_clr_fd() as
1201      * Lo_rundown state protects us from all the other places trying to
1202      * change the 'lo' device.
1203      */
1204     lo->lo_flags = 0;
1205     if (!part_shift)
1206         set_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1207     mutex_lock(&lo->lo_mutex);
1208     lo->lo_state = Lo_unbound;
1209     mutex_unlock(&lo->lo_mutex);
1210 
1211     /*
1212      * Need not hold lo_mutex to fput backing file. Calling fput holding
1213      * lo_mutex triggers a circular lock dependency possibility warning as
1214      * fput can take open_mutex which is usually taken before lo_mutex.
1215      */
1216     fput(filp);
1217 }
1218 
1219 static int loop_clr_fd(struct loop_device *lo)
1220 {
1221     int err;
1222 
1223     /*
1224      * Since lo_ioctl() is called without locks held, it is possible that
1225      * loop_configure()/loop_change_fd() and loop_clr_fd() run in parallel.
1226      *
1227      * Therefore, use global lock when setting Lo_rundown state in order to
1228      * make sure that loop_validate_file() will fail if the "struct file"
1229      * which loop_configure()/loop_change_fd() found via fget() was this
1230      * loop device.
1231      */
1232     err = loop_global_lock_killable(lo, true);
1233     if (err)
1234         return err;
1235     if (lo->lo_state != Lo_bound) {
1236         loop_global_unlock(lo, true);
1237         return -ENXIO;
1238     }
1239     /*
1240      * If we've explicitly asked to tear down the loop device,
1241      * and it has an elevated reference count, set it for auto-teardown when
1242      * the last reference goes away. This stops $!~#$@ udev from
1243      * preventing teardown because it decided that it needs to run blkid on
1244      * the loopback device whenever they appear. xfstests is notorious for
1245      * failing tests because blkid via udev races with a losetup
1246      * <dev>/do something like mkfs/losetup -d <dev> causing the losetup -d
1247      * command to fail with EBUSY.
1248      */
1249     if (disk_openers(lo->lo_disk) > 1) {
1250         lo->lo_flags |= LO_FLAGS_AUTOCLEAR;
1251         loop_global_unlock(lo, true);
1252         return 0;
1253     }
1254     lo->lo_state = Lo_rundown;
1255     loop_global_unlock(lo, true);
1256 
1257     __loop_clr_fd(lo, false);
1258     return 0;
1259 }
1260 
1261 static int
1262 loop_set_status(struct loop_device *lo, const struct loop_info64 *info)
1263 {
1264     int err;
1265     int prev_lo_flags;
1266     bool partscan = false;
1267     bool size_changed = false;
1268 
1269     err = mutex_lock_killable(&lo->lo_mutex);
1270     if (err)
1271         return err;
1272     if (lo->lo_state != Lo_bound) {
1273         err = -ENXIO;
1274         goto out_unlock;
1275     }
1276 
1277     if (lo->lo_offset != info->lo_offset ||
1278         lo->lo_sizelimit != info->lo_sizelimit) {
1279         size_changed = true;
1280         sync_blockdev(lo->lo_device);
1281         invalidate_bdev(lo->lo_device);
1282     }
1283 
1284     /* I/O need to be drained during transfer transition */
1285     blk_mq_freeze_queue(lo->lo_queue);
1286 
1287     prev_lo_flags = lo->lo_flags;
1288 
1289     err = loop_set_status_from_info(lo, info);
1290     if (err)
1291         goto out_unfreeze;
1292 
1293     /* Mask out flags that can't be set using LOOP_SET_STATUS. */
1294     lo->lo_flags &= LOOP_SET_STATUS_SETTABLE_FLAGS;
1295     /* For those flags, use the previous values instead */
1296     lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_SETTABLE_FLAGS;
1297     /* For flags that can't be cleared, use previous values too */
1298     lo->lo_flags |= prev_lo_flags & ~LOOP_SET_STATUS_CLEARABLE_FLAGS;
1299 
1300     if (size_changed) {
1301         loff_t new_size = get_size(lo->lo_offset, lo->lo_sizelimit,
1302                        lo->lo_backing_file);
1303         loop_set_size(lo, new_size);
1304     }
1305 
1306     loop_config_discard(lo);
1307 
1308     /* update dio if lo_offset or transfer is changed */
1309     __loop_update_dio(lo, lo->use_dio);
1310 
1311 out_unfreeze:
1312     blk_mq_unfreeze_queue(lo->lo_queue);
1313 
1314     if (!err && (lo->lo_flags & LO_FLAGS_PARTSCAN) &&
1315          !(prev_lo_flags & LO_FLAGS_PARTSCAN)) {
1316         clear_bit(GD_SUPPRESS_PART_SCAN, &lo->lo_disk->state);
1317         partscan = true;
1318     }
1319 out_unlock:
1320     mutex_unlock(&lo->lo_mutex);
1321     if (partscan)
1322         loop_reread_partitions(lo);
1323 
1324     return err;
1325 }
1326 
1327 static int
1328 loop_get_status(struct loop_device *lo, struct loop_info64 *info)
1329 {
1330     struct path path;
1331     struct kstat stat;
1332     int ret;
1333 
1334     ret = mutex_lock_killable(&lo->lo_mutex);
1335     if (ret)
1336         return ret;
1337     if (lo->lo_state != Lo_bound) {
1338         mutex_unlock(&lo->lo_mutex);
1339         return -ENXIO;
1340     }
1341 
1342     memset(info, 0, sizeof(*info));
1343     info->lo_number = lo->lo_number;
1344     info->lo_offset = lo->lo_offset;
1345     info->lo_sizelimit = lo->lo_sizelimit;
1346     info->lo_flags = lo->lo_flags;
1347     memcpy(info->lo_file_name, lo->lo_file_name, LO_NAME_SIZE);
1348 
1349     /* Drop lo_mutex while we call into the filesystem. */
1350     path = lo->lo_backing_file->f_path;
1351     path_get(&path);
1352     mutex_unlock(&lo->lo_mutex);
1353     ret = vfs_getattr(&path, &stat, STATX_INO, AT_STATX_SYNC_AS_STAT);
1354     if (!ret) {
1355         info->lo_device = huge_encode_dev(stat.dev);
1356         info->lo_inode = stat.ino;
1357         info->lo_rdevice = huge_encode_dev(stat.rdev);
1358     }
1359     path_put(&path);
1360     return ret;
1361 }
1362 
1363 static void
1364 loop_info64_from_old(const struct loop_info *info, struct loop_info64 *info64)
1365 {
1366     memset(info64, 0, sizeof(*info64));
1367     info64->lo_number = info->lo_number;
1368     info64->lo_device = info->lo_device;
1369     info64->lo_inode = info->lo_inode;
1370     info64->lo_rdevice = info->lo_rdevice;
1371     info64->lo_offset = info->lo_offset;
1372     info64->lo_sizelimit = 0;
1373     info64->lo_flags = info->lo_flags;
1374     memcpy(info64->lo_file_name, info->lo_name, LO_NAME_SIZE);
1375 }
1376 
1377 static int
1378 loop_info64_to_old(const struct loop_info64 *info64, struct loop_info *info)
1379 {
1380     memset(info, 0, sizeof(*info));
1381     info->lo_number = info64->lo_number;
1382     info->lo_device = info64->lo_device;
1383     info->lo_inode = info64->lo_inode;
1384     info->lo_rdevice = info64->lo_rdevice;
1385     info->lo_offset = info64->lo_offset;
1386     info->lo_flags = info64->lo_flags;
1387     memcpy(info->lo_name, info64->lo_file_name, LO_NAME_SIZE);
1388 
1389     /* error in case values were truncated */
1390     if (info->lo_device != info64->lo_device ||
1391         info->lo_rdevice != info64->lo_rdevice ||
1392         info->lo_inode != info64->lo_inode ||
1393         info->lo_offset != info64->lo_offset)
1394         return -EOVERFLOW;
1395 
1396     return 0;
1397 }
1398 
1399 static int
1400 loop_set_status_old(struct loop_device *lo, const struct loop_info __user *arg)
1401 {
1402     struct loop_info info;
1403     struct loop_info64 info64;
1404 
1405     if (copy_from_user(&info, arg, sizeof (struct loop_info)))
1406         return -EFAULT;
1407     loop_info64_from_old(&info, &info64);
1408     return loop_set_status(lo, &info64);
1409 }
1410 
1411 static int
1412 loop_set_status64(struct loop_device *lo, const struct loop_info64 __user *arg)
1413 {
1414     struct loop_info64 info64;
1415 
1416     if (copy_from_user(&info64, arg, sizeof (struct loop_info64)))
1417         return -EFAULT;
1418     return loop_set_status(lo, &info64);
1419 }
1420 
1421 static int
1422 loop_get_status_old(struct loop_device *lo, struct loop_info __user *arg) {
1423     struct loop_info info;
1424     struct loop_info64 info64;
1425     int err;
1426 
1427     if (!arg)
1428         return -EINVAL;
1429     err = loop_get_status(lo, &info64);
1430     if (!err)
1431         err = loop_info64_to_old(&info64, &info);
1432     if (!err && copy_to_user(arg, &info, sizeof(info)))
1433         err = -EFAULT;
1434 
1435     return err;
1436 }
1437 
1438 static int
1439 loop_get_status64(struct loop_device *lo, struct loop_info64 __user *arg) {
1440     struct loop_info64 info64;
1441     int err;
1442 
1443     if (!arg)
1444         return -EINVAL;
1445     err = loop_get_status(lo, &info64);
1446     if (!err && copy_to_user(arg, &info64, sizeof(info64)))
1447         err = -EFAULT;
1448 
1449     return err;
1450 }
1451 
1452 static int loop_set_capacity(struct loop_device *lo)
1453 {
1454     loff_t size;
1455 
1456     if (unlikely(lo->lo_state != Lo_bound))
1457         return -ENXIO;
1458 
1459     size = get_loop_size(lo, lo->lo_backing_file);
1460     loop_set_size(lo, size);
1461 
1462     return 0;
1463 }
1464 
1465 static int loop_set_dio(struct loop_device *lo, unsigned long arg)
1466 {
1467     int error = -ENXIO;
1468     if (lo->lo_state != Lo_bound)
1469         goto out;
1470 
1471     __loop_update_dio(lo, !!arg);
1472     if (lo->use_dio == !!arg)
1473         return 0;
1474     error = -EINVAL;
1475  out:
1476     return error;
1477 }
1478 
1479 static int loop_set_block_size(struct loop_device *lo, unsigned long arg)
1480 {
1481     int err = 0;
1482 
1483     if (lo->lo_state != Lo_bound)
1484         return -ENXIO;
1485 
1486     err = blk_validate_block_size(arg);
1487     if (err)
1488         return err;
1489 
1490     if (lo->lo_queue->limits.logical_block_size == arg)
1491         return 0;
1492 
1493     sync_blockdev(lo->lo_device);
1494     invalidate_bdev(lo->lo_device);
1495 
1496     blk_mq_freeze_queue(lo->lo_queue);
1497     blk_queue_logical_block_size(lo->lo_queue, arg);
1498     blk_queue_physical_block_size(lo->lo_queue, arg);
1499     blk_queue_io_min(lo->lo_queue, arg);
1500     loop_update_dio(lo);
1501     blk_mq_unfreeze_queue(lo->lo_queue);
1502 
1503     return err;
1504 }
1505 
1506 static int lo_simple_ioctl(struct loop_device *lo, unsigned int cmd,
1507                unsigned long arg)
1508 {
1509     int err;
1510 
1511     err = mutex_lock_killable(&lo->lo_mutex);
1512     if (err)
1513         return err;
1514     switch (cmd) {
1515     case LOOP_SET_CAPACITY:
1516         err = loop_set_capacity(lo);
1517         break;
1518     case LOOP_SET_DIRECT_IO:
1519         err = loop_set_dio(lo, arg);
1520         break;
1521     case LOOP_SET_BLOCK_SIZE:
1522         err = loop_set_block_size(lo, arg);
1523         break;
1524     default:
1525         err = -EINVAL;
1526     }
1527     mutex_unlock(&lo->lo_mutex);
1528     return err;
1529 }
1530 
1531 static int lo_ioctl(struct block_device *bdev, fmode_t mode,
1532     unsigned int cmd, unsigned long arg)
1533 {
1534     struct loop_device *lo = bdev->bd_disk->private_data;
1535     void __user *argp = (void __user *) arg;
1536     int err;
1537 
1538     switch (cmd) {
1539     case LOOP_SET_FD: {
1540         /*
1541          * Legacy case - pass in a zeroed out struct loop_config with
1542          * only the file descriptor set , which corresponds with the
1543          * default parameters we'd have used otherwise.
1544          */
1545         struct loop_config config;
1546 
1547         memset(&config, 0, sizeof(config));
1548         config.fd = arg;
1549 
1550         return loop_configure(lo, mode, bdev, &config);
1551     }
1552     case LOOP_CONFIGURE: {
1553         struct loop_config config;
1554 
1555         if (copy_from_user(&config, argp, sizeof(config)))
1556             return -EFAULT;
1557 
1558         return loop_configure(lo, mode, bdev, &config);
1559     }
1560     case LOOP_CHANGE_FD:
1561         return loop_change_fd(lo, bdev, arg);
1562     case LOOP_CLR_FD:
1563         return loop_clr_fd(lo);
1564     case LOOP_SET_STATUS:
1565         err = -EPERM;
1566         if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1567             err = loop_set_status_old(lo, argp);
1568         }
1569         break;
1570     case LOOP_GET_STATUS:
1571         return loop_get_status_old(lo, argp);
1572     case LOOP_SET_STATUS64:
1573         err = -EPERM;
1574         if ((mode & FMODE_WRITE) || capable(CAP_SYS_ADMIN)) {
1575             err = loop_set_status64(lo, argp);
1576         }
1577         break;
1578     case LOOP_GET_STATUS64:
1579         return loop_get_status64(lo, argp);
1580     case LOOP_SET_CAPACITY:
1581     case LOOP_SET_DIRECT_IO:
1582     case LOOP_SET_BLOCK_SIZE:
1583         if (!(mode & FMODE_WRITE) && !capable(CAP_SYS_ADMIN))
1584             return -EPERM;
1585         fallthrough;
1586     default:
1587         err = lo_simple_ioctl(lo, cmd, arg);
1588         break;
1589     }
1590 
1591     return err;
1592 }
1593 
1594 #ifdef CONFIG_COMPAT
1595 struct compat_loop_info {
1596     compat_int_t    lo_number;      /* ioctl r/o */
1597     compat_dev_t    lo_device;      /* ioctl r/o */
1598     compat_ulong_t  lo_inode;       /* ioctl r/o */
1599     compat_dev_t    lo_rdevice;     /* ioctl r/o */
1600     compat_int_t    lo_offset;
1601     compat_int_t    lo_encrypt_type;        /* obsolete, ignored */
1602     compat_int_t    lo_encrypt_key_size;    /* ioctl w/o */
1603     compat_int_t    lo_flags;       /* ioctl r/o */
1604     char        lo_name[LO_NAME_SIZE];
1605     unsigned char   lo_encrypt_key[LO_KEY_SIZE]; /* ioctl w/o */
1606     compat_ulong_t  lo_init[2];
1607     char        reserved[4];
1608 };
1609 
1610 /*
1611  * Transfer 32-bit compatibility structure in userspace to 64-bit loop info
1612  * - noinlined to reduce stack space usage in main part of driver
1613  */
1614 static noinline int
1615 loop_info64_from_compat(const struct compat_loop_info __user *arg,
1616             struct loop_info64 *info64)
1617 {
1618     struct compat_loop_info info;
1619 
1620     if (copy_from_user(&info, arg, sizeof(info)))
1621         return -EFAULT;
1622 
1623     memset(info64, 0, sizeof(*info64));
1624     info64->lo_number = info.lo_number;
1625     info64->lo_device = info.lo_device;
1626     info64->lo_inode = info.lo_inode;
1627     info64->lo_rdevice = info.lo_rdevice;
1628     info64->lo_offset = info.lo_offset;
1629     info64->lo_sizelimit = 0;
1630     info64->lo_flags = info.lo_flags;
1631     memcpy(info64->lo_file_name, info.lo_name, LO_NAME_SIZE);
1632     return 0;
1633 }
1634 
1635 /*
1636  * Transfer 64-bit loop info to 32-bit compatibility structure in userspace
1637  * - noinlined to reduce stack space usage in main part of driver
1638  */
1639 static noinline int
1640 loop_info64_to_compat(const struct loop_info64 *info64,
1641               struct compat_loop_info __user *arg)
1642 {
1643     struct compat_loop_info info;
1644 
1645     memset(&info, 0, sizeof(info));
1646     info.lo_number = info64->lo_number;
1647     info.lo_device = info64->lo_device;
1648     info.lo_inode = info64->lo_inode;
1649     info.lo_rdevice = info64->lo_rdevice;
1650     info.lo_offset = info64->lo_offset;
1651     info.lo_flags = info64->lo_flags;
1652     memcpy(info.lo_name, info64->lo_file_name, LO_NAME_SIZE);
1653 
1654     /* error in case values were truncated */
1655     if (info.lo_device != info64->lo_device ||
1656         info.lo_rdevice != info64->lo_rdevice ||
1657         info.lo_inode != info64->lo_inode ||
1658         info.lo_offset != info64->lo_offset)
1659         return -EOVERFLOW;
1660 
1661     if (copy_to_user(arg, &info, sizeof(info)))
1662         return -EFAULT;
1663     return 0;
1664 }
1665 
1666 static int
1667 loop_set_status_compat(struct loop_device *lo,
1668                const struct compat_loop_info __user *arg)
1669 {
1670     struct loop_info64 info64;
1671     int ret;
1672 
1673     ret = loop_info64_from_compat(arg, &info64);
1674     if (ret < 0)
1675         return ret;
1676     return loop_set_status(lo, &info64);
1677 }
1678 
1679 static int
1680 loop_get_status_compat(struct loop_device *lo,
1681                struct compat_loop_info __user *arg)
1682 {
1683     struct loop_info64 info64;
1684     int err;
1685 
1686     if (!arg)
1687         return -EINVAL;
1688     err = loop_get_status(lo, &info64);
1689     if (!err)
1690         err = loop_info64_to_compat(&info64, arg);
1691     return err;
1692 }
1693 
1694 static int lo_compat_ioctl(struct block_device *bdev, fmode_t mode,
1695                unsigned int cmd, unsigned long arg)
1696 {
1697     struct loop_device *lo = bdev->bd_disk->private_data;
1698     int err;
1699 
1700     switch(cmd) {
1701     case LOOP_SET_STATUS:
1702         err = loop_set_status_compat(lo,
1703                  (const struct compat_loop_info __user *)arg);
1704         break;
1705     case LOOP_GET_STATUS:
1706         err = loop_get_status_compat(lo,
1707                      (struct compat_loop_info __user *)arg);
1708         break;
1709     case LOOP_SET_CAPACITY:
1710     case LOOP_CLR_FD:
1711     case LOOP_GET_STATUS64:
1712     case LOOP_SET_STATUS64:
1713     case LOOP_CONFIGURE:
1714         arg = (unsigned long) compat_ptr(arg);
1715         fallthrough;
1716     case LOOP_SET_FD:
1717     case LOOP_CHANGE_FD:
1718     case LOOP_SET_BLOCK_SIZE:
1719     case LOOP_SET_DIRECT_IO:
1720         err = lo_ioctl(bdev, mode, cmd, arg);
1721         break;
1722     default:
1723         err = -ENOIOCTLCMD;
1724         break;
1725     }
1726     return err;
1727 }
1728 #endif
1729 
1730 static void lo_release(struct gendisk *disk, fmode_t mode)
1731 {
1732     struct loop_device *lo = disk->private_data;
1733 
1734     if (disk_openers(disk) > 0)
1735         return;
1736 
1737     mutex_lock(&lo->lo_mutex);
1738     if (lo->lo_state == Lo_bound && (lo->lo_flags & LO_FLAGS_AUTOCLEAR)) {
1739         lo->lo_state = Lo_rundown;
1740         mutex_unlock(&lo->lo_mutex);
1741         /*
1742          * In autoclear mode, stop the loop thread
1743          * and remove configuration after last close.
1744          */
1745         __loop_clr_fd(lo, true);
1746         return;
1747     }
1748     mutex_unlock(&lo->lo_mutex);
1749 }
1750 
1751 static void lo_free_disk(struct gendisk *disk)
1752 {
1753     struct loop_device *lo = disk->private_data;
1754 
1755     if (lo->workqueue)
1756         destroy_workqueue(lo->workqueue);
1757     loop_free_idle_workers(lo, true);
1758     del_timer_sync(&lo->timer);
1759     mutex_destroy(&lo->lo_mutex);
1760     kfree(lo);
1761 }
1762 
1763 static const struct block_device_operations lo_fops = {
1764     .owner =    THIS_MODULE,
1765     .release =  lo_release,
1766     .ioctl =    lo_ioctl,
1767 #ifdef CONFIG_COMPAT
1768     .compat_ioctl = lo_compat_ioctl,
1769 #endif
1770     .free_disk =    lo_free_disk,
1771 };
1772 
1773 /*
1774  * And now the modules code and kernel interface.
1775  */
1776 static int max_loop;
1777 module_param(max_loop, int, 0444);
1778 MODULE_PARM_DESC(max_loop, "Maximum number of loop devices");
1779 module_param(max_part, int, 0444);
1780 MODULE_PARM_DESC(max_part, "Maximum number of partitions per loop device");
1781 
1782 static int hw_queue_depth = LOOP_DEFAULT_HW_Q_DEPTH;
1783 
1784 static int loop_set_hw_queue_depth(const char *s, const struct kernel_param *p)
1785 {
1786     int ret = kstrtoint(s, 10, &hw_queue_depth);
1787 
1788     return (ret || (hw_queue_depth < 1)) ? -EINVAL : 0;
1789 }
1790 
1791 static const struct kernel_param_ops loop_hw_qdepth_param_ops = {
1792     .set    = loop_set_hw_queue_depth,
1793     .get    = param_get_int,
1794 };
1795 
1796 device_param_cb(hw_queue_depth, &loop_hw_qdepth_param_ops, &hw_queue_depth, 0444);
1797 MODULE_PARM_DESC(hw_queue_depth, "Queue depth for each hardware queue. Default: 128");
1798 
1799 MODULE_LICENSE("GPL");
1800 MODULE_ALIAS_BLOCKDEV_MAJOR(LOOP_MAJOR);
1801 
1802 static blk_status_t loop_queue_rq(struct blk_mq_hw_ctx *hctx,
1803         const struct blk_mq_queue_data *bd)
1804 {
1805     struct request *rq = bd->rq;
1806     struct loop_cmd *cmd = blk_mq_rq_to_pdu(rq);
1807     struct loop_device *lo = rq->q->queuedata;
1808 
1809     blk_mq_start_request(rq);
1810 
1811     if (lo->lo_state != Lo_bound)
1812         return BLK_STS_IOERR;
1813 
1814     switch (req_op(rq)) {
1815     case REQ_OP_FLUSH:
1816     case REQ_OP_DISCARD:
1817     case REQ_OP_WRITE_ZEROES:
1818         cmd->use_aio = false;
1819         break;
1820     default:
1821         cmd->use_aio = lo->use_dio;
1822         break;
1823     }
1824 
1825     /* always use the first bio's css */
1826     cmd->blkcg_css = NULL;
1827     cmd->memcg_css = NULL;
1828 #ifdef CONFIG_BLK_CGROUP
1829     if (rq->bio) {
1830         cmd->blkcg_css = bio_blkcg_css(rq->bio);
1831 #ifdef CONFIG_MEMCG
1832         if (cmd->blkcg_css) {
1833             cmd->memcg_css =
1834                 cgroup_get_e_css(cmd->blkcg_css->cgroup,
1835                         &memory_cgrp_subsys);
1836         }
1837 #endif
1838     }
1839 #endif
1840     loop_queue_work(lo, cmd);
1841 
1842     return BLK_STS_OK;
1843 }
1844 
1845 static void loop_handle_cmd(struct loop_cmd *cmd)
1846 {
1847     struct request *rq = blk_mq_rq_from_pdu(cmd);
1848     const bool write = op_is_write(req_op(rq));
1849     struct loop_device *lo = rq->q->queuedata;
1850     int ret = 0;
1851     struct mem_cgroup *old_memcg = NULL;
1852 
1853     if (write && (lo->lo_flags & LO_FLAGS_READ_ONLY)) {
1854         ret = -EIO;
1855         goto failed;
1856     }
1857 
1858     if (cmd->blkcg_css)
1859         kthread_associate_blkcg(cmd->blkcg_css);
1860     if (cmd->memcg_css)
1861         old_memcg = set_active_memcg(
1862             mem_cgroup_from_css(cmd->memcg_css));
1863 
1864     ret = do_req_filebacked(lo, rq);
1865 
1866     if (cmd->blkcg_css)
1867         kthread_associate_blkcg(NULL);
1868 
1869     if (cmd->memcg_css) {
1870         set_active_memcg(old_memcg);
1871         css_put(cmd->memcg_css);
1872     }
1873  failed:
1874     /* complete non-aio request */
1875     if (!cmd->use_aio || ret) {
1876         if (ret == -EOPNOTSUPP)
1877             cmd->ret = ret;
1878         else
1879             cmd->ret = ret ? -EIO : 0;
1880         if (likely(!blk_should_fake_timeout(rq->q)))
1881             blk_mq_complete_request(rq);
1882     }
1883 }
1884 
1885 static void loop_process_work(struct loop_worker *worker,
1886             struct list_head *cmd_list, struct loop_device *lo)
1887 {
1888     int orig_flags = current->flags;
1889     struct loop_cmd *cmd;
1890 
1891     current->flags |= PF_LOCAL_THROTTLE | PF_MEMALLOC_NOIO;
1892     spin_lock_irq(&lo->lo_work_lock);
1893     while (!list_empty(cmd_list)) {
1894         cmd = container_of(
1895             cmd_list->next, struct loop_cmd, list_entry);
1896         list_del(cmd_list->next);
1897         spin_unlock_irq(&lo->lo_work_lock);
1898 
1899         loop_handle_cmd(cmd);
1900         cond_resched();
1901 
1902         spin_lock_irq(&lo->lo_work_lock);
1903     }
1904 
1905     /*
1906      * We only add to the idle list if there are no pending cmds
1907      * *and* the worker will not run again which ensures that it
1908      * is safe to free any worker on the idle list
1909      */
1910     if (worker && !work_pending(&worker->work)) {
1911         worker->last_ran_at = jiffies;
1912         list_add_tail(&worker->idle_list, &lo->idle_worker_list);
1913         loop_set_timer(lo);
1914     }
1915     spin_unlock_irq(&lo->lo_work_lock);
1916     current->flags = orig_flags;
1917 }
1918 
1919 static void loop_workfn(struct work_struct *work)
1920 {
1921     struct loop_worker *worker =
1922         container_of(work, struct loop_worker, work);
1923     loop_process_work(worker, &worker->cmd_list, worker->lo);
1924 }
1925 
1926 static void loop_rootcg_workfn(struct work_struct *work)
1927 {
1928     struct loop_device *lo =
1929         container_of(work, struct loop_device, rootcg_work);
1930     loop_process_work(NULL, &lo->rootcg_cmd_list, lo);
1931 }
1932 
1933 static const struct blk_mq_ops loop_mq_ops = {
1934     .queue_rq       = loop_queue_rq,
1935     .complete   = lo_complete_rq,
1936 };
1937 
1938 static int loop_add(int i)
1939 {
1940     struct loop_device *lo;
1941     struct gendisk *disk;
1942     int err;
1943 
1944     err = -ENOMEM;
1945     lo = kzalloc(sizeof(*lo), GFP_KERNEL);
1946     if (!lo)
1947         goto out;
1948     lo->worker_tree = RB_ROOT;
1949     INIT_LIST_HEAD(&lo->idle_worker_list);
1950     timer_setup(&lo->timer, loop_free_idle_workers_timer, TIMER_DEFERRABLE);
1951     lo->lo_state = Lo_unbound;
1952 
1953     err = mutex_lock_killable(&loop_ctl_mutex);
1954     if (err)
1955         goto out_free_dev;
1956 
1957     /* allocate id, if @id >= 0, we're requesting that specific id */
1958     if (i >= 0) {
1959         err = idr_alloc(&loop_index_idr, lo, i, i + 1, GFP_KERNEL);
1960         if (err == -ENOSPC)
1961             err = -EEXIST;
1962     } else {
1963         err = idr_alloc(&loop_index_idr, lo, 0, 0, GFP_KERNEL);
1964     }
1965     mutex_unlock(&loop_ctl_mutex);
1966     if (err < 0)
1967         goto out_free_dev;
1968     i = err;
1969 
1970     lo->tag_set.ops = &loop_mq_ops;
1971     lo->tag_set.nr_hw_queues = 1;
1972     lo->tag_set.queue_depth = hw_queue_depth;
1973     lo->tag_set.numa_node = NUMA_NO_NODE;
1974     lo->tag_set.cmd_size = sizeof(struct loop_cmd);
1975     lo->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING |
1976         BLK_MQ_F_NO_SCHED_BY_DEFAULT;
1977     lo->tag_set.driver_data = lo;
1978 
1979     err = blk_mq_alloc_tag_set(&lo->tag_set);
1980     if (err)
1981         goto out_free_idr;
1982 
1983     disk = lo->lo_disk = blk_mq_alloc_disk(&lo->tag_set, lo);
1984     if (IS_ERR(disk)) {
1985         err = PTR_ERR(disk);
1986         goto out_cleanup_tags;
1987     }
1988     lo->lo_queue = lo->lo_disk->queue;
1989 
1990     blk_queue_max_hw_sectors(lo->lo_queue, BLK_DEF_MAX_SECTORS);
1991 
1992     /*
1993      * By default, we do buffer IO, so it doesn't make sense to enable
1994      * merge because the I/O submitted to backing file is handled page by
1995      * page. For directio mode, merge does help to dispatch bigger request
1996      * to underlayer disk. We will enable merge once directio is enabled.
1997      */
1998     blk_queue_flag_set(QUEUE_FLAG_NOMERGES, lo->lo_queue);
1999 
2000     /*
2001      * Disable partition scanning by default. The in-kernel partition
2002      * scanning can be requested individually per-device during its
2003      * setup. Userspace can always add and remove partitions from all
2004      * devices. The needed partition minors are allocated from the
2005      * extended minor space, the main loop device numbers will continue
2006      * to match the loop minors, regardless of the number of partitions
2007      * used.
2008      *
2009      * If max_part is given, partition scanning is globally enabled for
2010      * all loop devices. The minors for the main loop devices will be
2011      * multiples of max_part.
2012      *
2013      * Note: Global-for-all-devices, set-only-at-init, read-only module
2014      * parameteters like 'max_loop' and 'max_part' make things needlessly
2015      * complicated, are too static, inflexible and may surprise
2016      * userspace tools. Parameters like this in general should be avoided.
2017      */
2018     if (!part_shift)
2019         set_bit(GD_SUPPRESS_PART_SCAN, &disk->state);
2020     mutex_init(&lo->lo_mutex);
2021     lo->lo_number       = i;
2022     spin_lock_init(&lo->lo_lock);
2023     spin_lock_init(&lo->lo_work_lock);
2024     INIT_WORK(&lo->rootcg_work, loop_rootcg_workfn);
2025     INIT_LIST_HEAD(&lo->rootcg_cmd_list);
2026     disk->major     = LOOP_MAJOR;
2027     disk->first_minor   = i << part_shift;
2028     disk->minors        = 1 << part_shift;
2029     disk->fops      = &lo_fops;
2030     disk->private_data  = lo;
2031     disk->queue     = lo->lo_queue;
2032     disk->events        = DISK_EVENT_MEDIA_CHANGE;
2033     disk->event_flags   = DISK_EVENT_FLAG_UEVENT;
2034     sprintf(disk->disk_name, "loop%d", i);
2035     /* Make this loop device reachable from pathname. */
2036     err = add_disk(disk);
2037     if (err)
2038         goto out_cleanup_disk;
2039 
2040     /* Show this loop device. */
2041     mutex_lock(&loop_ctl_mutex);
2042     lo->idr_visible = true;
2043     mutex_unlock(&loop_ctl_mutex);
2044 
2045     return i;
2046 
2047 out_cleanup_disk:
2048     put_disk(disk);
2049 out_cleanup_tags:
2050     blk_mq_free_tag_set(&lo->tag_set);
2051 out_free_idr:
2052     mutex_lock(&loop_ctl_mutex);
2053     idr_remove(&loop_index_idr, i);
2054     mutex_unlock(&loop_ctl_mutex);
2055 out_free_dev:
2056     kfree(lo);
2057 out:
2058     return err;
2059 }
2060 
2061 static void loop_remove(struct loop_device *lo)
2062 {
2063     /* Make this loop device unreachable from pathname. */
2064     del_gendisk(lo->lo_disk);
2065     blk_mq_free_tag_set(&lo->tag_set);
2066 
2067     mutex_lock(&loop_ctl_mutex);
2068     idr_remove(&loop_index_idr, lo->lo_number);
2069     mutex_unlock(&loop_ctl_mutex);
2070 
2071     put_disk(lo->lo_disk);
2072 }
2073 
2074 static void loop_probe(dev_t dev)
2075 {
2076     int idx = MINOR(dev) >> part_shift;
2077 
2078     if (max_loop && idx >= max_loop)
2079         return;
2080     loop_add(idx);
2081 }
2082 
2083 static int loop_control_remove(int idx)
2084 {
2085     struct loop_device *lo;
2086     int ret;
2087 
2088     if (idx < 0) {
2089         pr_warn_once("deleting an unspecified loop device is not supported.\n");
2090         return -EINVAL;
2091     }
2092         
2093     /* Hide this loop device for serialization. */
2094     ret = mutex_lock_killable(&loop_ctl_mutex);
2095     if (ret)
2096         return ret;
2097     lo = idr_find(&loop_index_idr, idx);
2098     if (!lo || !lo->idr_visible)
2099         ret = -ENODEV;
2100     else
2101         lo->idr_visible = false;
2102     mutex_unlock(&loop_ctl_mutex);
2103     if (ret)
2104         return ret;
2105 
2106     /* Check whether this loop device can be removed. */
2107     ret = mutex_lock_killable(&lo->lo_mutex);
2108     if (ret)
2109         goto mark_visible;
2110     if (lo->lo_state != Lo_unbound || disk_openers(lo->lo_disk) > 0) {
2111         mutex_unlock(&lo->lo_mutex);
2112         ret = -EBUSY;
2113         goto mark_visible;
2114     }
2115     /* Mark this loop device as no more bound, but not quite unbound yet */
2116     lo->lo_state = Lo_deleting;
2117     mutex_unlock(&lo->lo_mutex);
2118 
2119     loop_remove(lo);
2120     return 0;
2121 
2122 mark_visible:
2123     /* Show this loop device again. */
2124     mutex_lock(&loop_ctl_mutex);
2125     lo->idr_visible = true;
2126     mutex_unlock(&loop_ctl_mutex);
2127     return ret;
2128 }
2129 
2130 static int loop_control_get_free(int idx)
2131 {
2132     struct loop_device *lo;
2133     int id, ret;
2134 
2135     ret = mutex_lock_killable(&loop_ctl_mutex);
2136     if (ret)
2137         return ret;
2138     idr_for_each_entry(&loop_index_idr, lo, id) {
2139         /* Hitting a race results in creating a new loop device which is harmless. */
2140         if (lo->idr_visible && data_race(lo->lo_state) == Lo_unbound)
2141             goto found;
2142     }
2143     mutex_unlock(&loop_ctl_mutex);
2144     return loop_add(-1);
2145 found:
2146     mutex_unlock(&loop_ctl_mutex);
2147     return id;
2148 }
2149 
2150 static long loop_control_ioctl(struct file *file, unsigned int cmd,
2151                    unsigned long parm)
2152 {
2153     switch (cmd) {
2154     case LOOP_CTL_ADD:
2155         return loop_add(parm);
2156     case LOOP_CTL_REMOVE:
2157         return loop_control_remove(parm);
2158     case LOOP_CTL_GET_FREE:
2159         return loop_control_get_free(parm);
2160     default:
2161         return -ENOSYS;
2162     }
2163 }
2164 
2165 static const struct file_operations loop_ctl_fops = {
2166     .open       = nonseekable_open,
2167     .unlocked_ioctl = loop_control_ioctl,
2168     .compat_ioctl   = loop_control_ioctl,
2169     .owner      = THIS_MODULE,
2170     .llseek     = noop_llseek,
2171 };
2172 
2173 static struct miscdevice loop_misc = {
2174     .minor      = LOOP_CTRL_MINOR,
2175     .name       = "loop-control",
2176     .fops       = &loop_ctl_fops,
2177 };
2178 
2179 MODULE_ALIAS_MISCDEV(LOOP_CTRL_MINOR);
2180 MODULE_ALIAS("devname:loop-control");
2181 
2182 static int __init loop_init(void)
2183 {
2184     int i, nr;
2185     int err;
2186 
2187     part_shift = 0;
2188     if (max_part > 0) {
2189         part_shift = fls(max_part);
2190 
2191         /*
2192          * Adjust max_part according to part_shift as it is exported
2193          * to user space so that user can decide correct minor number
2194          * if [s]he want to create more devices.
2195          *
2196          * Note that -1 is required because partition 0 is reserved
2197          * for the whole disk.
2198          */
2199         max_part = (1UL << part_shift) - 1;
2200     }
2201 
2202     if ((1UL << part_shift) > DISK_MAX_PARTS) {
2203         err = -EINVAL;
2204         goto err_out;
2205     }
2206 
2207     if (max_loop > 1UL << (MINORBITS - part_shift)) {
2208         err = -EINVAL;
2209         goto err_out;
2210     }
2211 
2212     /*
2213      * If max_loop is specified, create that many devices upfront.
2214      * This also becomes a hard limit. If max_loop is not specified,
2215      * create CONFIG_BLK_DEV_LOOP_MIN_COUNT loop devices at module
2216      * init time. Loop devices can be requested on-demand with the
2217      * /dev/loop-control interface, or be instantiated by accessing
2218      * a 'dead' device node.
2219      */
2220     if (max_loop)
2221         nr = max_loop;
2222     else
2223         nr = CONFIG_BLK_DEV_LOOP_MIN_COUNT;
2224 
2225     err = misc_register(&loop_misc);
2226     if (err < 0)
2227         goto err_out;
2228 
2229 
2230     if (__register_blkdev(LOOP_MAJOR, "loop", loop_probe)) {
2231         err = -EIO;
2232         goto misc_out;
2233     }
2234 
2235     /* pre-create number of devices given by config or max_loop */
2236     for (i = 0; i < nr; i++)
2237         loop_add(i);
2238 
2239     printk(KERN_INFO "loop: module loaded\n");
2240     return 0;
2241 
2242 misc_out:
2243     misc_deregister(&loop_misc);
2244 err_out:
2245     return err;
2246 }
2247 
2248 static void __exit loop_exit(void)
2249 {
2250     struct loop_device *lo;
2251     int id;
2252 
2253     unregister_blkdev(LOOP_MAJOR, "loop");
2254     misc_deregister(&loop_misc);
2255 
2256     /*
2257      * There is no need to use loop_ctl_mutex here, for nobody else can
2258      * access loop_index_idr when this module is unloading (unless forced
2259      * module unloading is requested). If this is not a clean unloading,
2260      * we have no means to avoid kernel crash.
2261      */
2262     idr_for_each_entry(&loop_index_idr, lo, id)
2263         loop_remove(lo);
2264 
2265     idr_destroy(&loop_index_idr);
2266 }
2267 
2268 module_init(loop_init);
2269 module_exit(loop_exit);
2270 
2271 #ifndef MODULE
2272 static int __init max_loop_setup(char *str)
2273 {
2274     max_loop = simple_strtol(str, NULL, 0);
2275     return 1;
2276 }
2277 
2278 __setup("max_loop=", max_loop_setup);
2279 #endif