Back to home page

LXR

 
 

    


0001 /*
0002  *  linux/fs/block_dev.c
0003  *
0004  *  Copyright (C) 1991, 1992  Linus Torvalds
0005  *  Copyright (C) 2001  Andrea Arcangeli <andrea@suse.de> SuSE
0006  */
0007 
0008 #include <linux/init.h>
0009 #include <linux/mm.h>
0010 #include <linux/fcntl.h>
0011 #include <linux/slab.h>
0012 #include <linux/kmod.h>
0013 #include <linux/major.h>
0014 #include <linux/device_cgroup.h>
0015 #include <linux/highmem.h>
0016 #include <linux/blkdev.h>
0017 #include <linux/backing-dev.h>
0018 #include <linux/module.h>
0019 #include <linux/blkpg.h>
0020 #include <linux/magic.h>
0021 #include <linux/buffer_head.h>
0022 #include <linux/swap.h>
0023 #include <linux/pagevec.h>
0024 #include <linux/writeback.h>
0025 #include <linux/mpage.h>
0026 #include <linux/mount.h>
0027 #include <linux/uio.h>
0028 #include <linux/namei.h>
0029 #include <linux/log2.h>
0030 #include <linux/cleancache.h>
0031 #include <linux/dax.h>
0032 #include <linux/badblocks.h>
0033 #include <linux/task_io_accounting_ops.h>
0034 #include <linux/falloc.h>
0035 #include <linux/uaccess.h>
0036 #include "internal.h"
0037 
0038 struct bdev_inode {
0039     struct block_device bdev;
0040     struct inode vfs_inode;
0041 };
0042 
0043 static const struct address_space_operations def_blk_aops;
0044 
0045 static inline struct bdev_inode *BDEV_I(struct inode *inode)
0046 {
0047     return container_of(inode, struct bdev_inode, vfs_inode);
0048 }
0049 
0050 struct block_device *I_BDEV(struct inode *inode)
0051 {
0052     return &BDEV_I(inode)->bdev;
0053 }
0054 EXPORT_SYMBOL(I_BDEV);
0055 
0056 void __vfs_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
0057 {
0058     struct va_format vaf;
0059     va_list args;
0060 
0061     va_start(args, fmt);
0062     vaf.fmt = fmt;
0063     vaf.va = &args;
0064     printk_ratelimited("%sVFS (%s): %pV\n", prefix, sb->s_id, &vaf);
0065     va_end(args);
0066 }
0067 
0068 static void bdev_write_inode(struct block_device *bdev)
0069 {
0070     struct inode *inode = bdev->bd_inode;
0071     int ret;
0072 
0073     spin_lock(&inode->i_lock);
0074     while (inode->i_state & I_DIRTY) {
0075         spin_unlock(&inode->i_lock);
0076         ret = write_inode_now(inode, true);
0077         if (ret) {
0078             char name[BDEVNAME_SIZE];
0079             pr_warn_ratelimited("VFS: Dirty inode writeback failed "
0080                         "for block device %s (err=%d).\n",
0081                         bdevname(bdev, name), ret);
0082         }
0083         spin_lock(&inode->i_lock);
0084     }
0085     spin_unlock(&inode->i_lock);
0086 }
0087 
0088 /* Kill _all_ buffers and pagecache , dirty or not.. */
0089 void kill_bdev(struct block_device *bdev)
0090 {
0091     struct address_space *mapping = bdev->bd_inode->i_mapping;
0092 
0093     if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
0094         return;
0095 
0096     invalidate_bh_lrus();
0097     truncate_inode_pages(mapping, 0);
0098 }   
0099 EXPORT_SYMBOL(kill_bdev);
0100 
0101 /* Invalidate clean unused buffers and pagecache. */
0102 void invalidate_bdev(struct block_device *bdev)
0103 {
0104     struct address_space *mapping = bdev->bd_inode->i_mapping;
0105 
0106     if (mapping->nrpages == 0)
0107         return;
0108 
0109     invalidate_bh_lrus();
0110     lru_add_drain_all();    /* make sure all lru add caches are flushed */
0111     invalidate_mapping_pages(mapping, 0, -1);
0112     /* 99% of the time, we don't need to flush the cleancache on the bdev.
0113      * But, for the strange corners, lets be cautious
0114      */
0115     cleancache_invalidate_inode(mapping);
0116 }
0117 EXPORT_SYMBOL(invalidate_bdev);
0118 
0119 int set_blocksize(struct block_device *bdev, int size)
0120 {
0121     /* Size must be a power of two, and between 512 and PAGE_SIZE */
0122     if (size > PAGE_SIZE || size < 512 || !is_power_of_2(size))
0123         return -EINVAL;
0124 
0125     /* Size cannot be smaller than the size supported by the device */
0126     if (size < bdev_logical_block_size(bdev))
0127         return -EINVAL;
0128 
0129     /* Don't change the size if it is same as current */
0130     if (bdev->bd_block_size != size) {
0131         sync_blockdev(bdev);
0132         bdev->bd_block_size = size;
0133         bdev->bd_inode->i_blkbits = blksize_bits(size);
0134         kill_bdev(bdev);
0135     }
0136     return 0;
0137 }
0138 
0139 EXPORT_SYMBOL(set_blocksize);
0140 
0141 int sb_set_blocksize(struct super_block *sb, int size)
0142 {
0143     if (set_blocksize(sb->s_bdev, size))
0144         return 0;
0145     /* If we get here, we know size is power of two
0146      * and it's value is between 512 and PAGE_SIZE */
0147     sb->s_blocksize = size;
0148     sb->s_blocksize_bits = blksize_bits(size);
0149     return sb->s_blocksize;
0150 }
0151 
0152 EXPORT_SYMBOL(sb_set_blocksize);
0153 
0154 int sb_min_blocksize(struct super_block *sb, int size)
0155 {
0156     int minsize = bdev_logical_block_size(sb->s_bdev);
0157     if (size < minsize)
0158         size = minsize;
0159     return sb_set_blocksize(sb, size);
0160 }
0161 
0162 EXPORT_SYMBOL(sb_min_blocksize);
0163 
0164 static int
0165 blkdev_get_block(struct inode *inode, sector_t iblock,
0166         struct buffer_head *bh, int create)
0167 {
0168     bh->b_bdev = I_BDEV(inode);
0169     bh->b_blocknr = iblock;
0170     set_buffer_mapped(bh);
0171     return 0;
0172 }
0173 
0174 static struct inode *bdev_file_inode(struct file *file)
0175 {
0176     return file->f_mapping->host;
0177 }
0178 
0179 static unsigned int dio_bio_write_op(struct kiocb *iocb)
0180 {
0181     unsigned int op = REQ_OP_WRITE | REQ_SYNC | REQ_IDLE;
0182 
0183     /* avoid the need for a I/O completion work item */
0184     if (iocb->ki_flags & IOCB_DSYNC)
0185         op |= REQ_FUA;
0186     return op;
0187 }
0188 
0189 #define DIO_INLINE_BIO_VECS 4
0190 
0191 static void blkdev_bio_end_io_simple(struct bio *bio)
0192 {
0193     struct task_struct *waiter = bio->bi_private;
0194 
0195     WRITE_ONCE(bio->bi_private, NULL);
0196     wake_up_process(waiter);
0197 }
0198 
0199 static ssize_t
0200 __blkdev_direct_IO_simple(struct kiocb *iocb, struct iov_iter *iter,
0201         int nr_pages)
0202 {
0203     struct file *file = iocb->ki_filp;
0204     struct block_device *bdev = I_BDEV(bdev_file_inode(file));
0205     struct bio_vec inline_vecs[DIO_INLINE_BIO_VECS], *vecs, *bvec;
0206     loff_t pos = iocb->ki_pos;
0207     bool should_dirty = false;
0208     struct bio bio;
0209     ssize_t ret;
0210     blk_qc_t qc;
0211     int i;
0212 
0213     if ((pos | iov_iter_alignment(iter)) &
0214         (bdev_logical_block_size(bdev) - 1))
0215         return -EINVAL;
0216 
0217     if (nr_pages <= DIO_INLINE_BIO_VECS)
0218         vecs = inline_vecs;
0219     else {
0220         vecs = kmalloc(nr_pages * sizeof(struct bio_vec), GFP_KERNEL);
0221         if (!vecs)
0222             return -ENOMEM;
0223     }
0224 
0225     bio_init(&bio, vecs, nr_pages);
0226     bio.bi_bdev = bdev;
0227     bio.bi_iter.bi_sector = pos >> 9;
0228     bio.bi_private = current;
0229     bio.bi_end_io = blkdev_bio_end_io_simple;
0230 
0231     ret = bio_iov_iter_get_pages(&bio, iter);
0232     if (unlikely(ret))
0233         return ret;
0234     ret = bio.bi_iter.bi_size;
0235 
0236     if (iov_iter_rw(iter) == READ) {
0237         bio.bi_opf = REQ_OP_READ;
0238         if (iter_is_iovec(iter))
0239             should_dirty = true;
0240     } else {
0241         bio.bi_opf = dio_bio_write_op(iocb);
0242         task_io_account_write(ret);
0243     }
0244 
0245     qc = submit_bio(&bio);
0246     for (;;) {
0247         set_current_state(TASK_UNINTERRUPTIBLE);
0248         if (!READ_ONCE(bio.bi_private))
0249             break;
0250         if (!(iocb->ki_flags & IOCB_HIPRI) ||
0251             !blk_mq_poll(bdev_get_queue(bdev), qc))
0252             io_schedule();
0253     }
0254     __set_current_state(TASK_RUNNING);
0255 
0256     bio_for_each_segment_all(bvec, &bio, i) {
0257         if (should_dirty && !PageCompound(bvec->bv_page))
0258             set_page_dirty_lock(bvec->bv_page);
0259         put_page(bvec->bv_page);
0260     }
0261 
0262     if (vecs != inline_vecs)
0263         kfree(vecs);
0264 
0265     if (unlikely(bio.bi_error))
0266         return bio.bi_error;
0267     return ret;
0268 }
0269 
0270 struct blkdev_dio {
0271     union {
0272         struct kiocb        *iocb;
0273         struct task_struct  *waiter;
0274     };
0275     size_t          size;
0276     atomic_t        ref;
0277     bool            multi_bio : 1;
0278     bool            should_dirty : 1;
0279     bool            is_sync : 1;
0280     struct bio      bio;
0281 };
0282 
0283 static struct bio_set *blkdev_dio_pool __read_mostly;
0284 
0285 static void blkdev_bio_end_io(struct bio *bio)
0286 {
0287     struct blkdev_dio *dio = bio->bi_private;
0288     bool should_dirty = dio->should_dirty;
0289 
0290     if (dio->multi_bio && !atomic_dec_and_test(&dio->ref)) {
0291         if (bio->bi_error && !dio->bio.bi_error)
0292             dio->bio.bi_error = bio->bi_error;
0293     } else {
0294         if (!dio->is_sync) {
0295             struct kiocb *iocb = dio->iocb;
0296             ssize_t ret = dio->bio.bi_error;
0297 
0298             if (likely(!ret)) {
0299                 ret = dio->size;
0300                 iocb->ki_pos += ret;
0301             }
0302 
0303             dio->iocb->ki_complete(iocb, ret, 0);
0304             bio_put(&dio->bio);
0305         } else {
0306             struct task_struct *waiter = dio->waiter;
0307 
0308             WRITE_ONCE(dio->waiter, NULL);
0309             wake_up_process(waiter);
0310         }
0311     }
0312 
0313     if (should_dirty) {
0314         bio_check_pages_dirty(bio);
0315     } else {
0316         struct bio_vec *bvec;
0317         int i;
0318 
0319         bio_for_each_segment_all(bvec, bio, i)
0320             put_page(bvec->bv_page);
0321         bio_put(bio);
0322     }
0323 }
0324 
0325 static ssize_t
0326 __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, int nr_pages)
0327 {
0328     struct file *file = iocb->ki_filp;
0329     struct inode *inode = bdev_file_inode(file);
0330     struct block_device *bdev = I_BDEV(inode);
0331     struct blk_plug plug;
0332     struct blkdev_dio *dio;
0333     struct bio *bio;
0334     bool is_read = (iov_iter_rw(iter) == READ), is_sync;
0335     loff_t pos = iocb->ki_pos;
0336     blk_qc_t qc = BLK_QC_T_NONE;
0337     int ret;
0338 
0339     if ((pos | iov_iter_alignment(iter)) &
0340         (bdev_logical_block_size(bdev) - 1))
0341         return -EINVAL;
0342 
0343     bio = bio_alloc_bioset(GFP_KERNEL, nr_pages, blkdev_dio_pool);
0344     bio_get(bio); /* extra ref for the completion handler */
0345 
0346     dio = container_of(bio, struct blkdev_dio, bio);
0347     dio->is_sync = is_sync = is_sync_kiocb(iocb);
0348     if (dio->is_sync)
0349         dio->waiter = current;
0350     else
0351         dio->iocb = iocb;
0352 
0353     dio->size = 0;
0354     dio->multi_bio = false;
0355     dio->should_dirty = is_read && (iter->type == ITER_IOVEC);
0356 
0357     blk_start_plug(&plug);
0358     for (;;) {
0359         bio->bi_bdev = bdev;
0360         bio->bi_iter.bi_sector = pos >> 9;
0361         bio->bi_private = dio;
0362         bio->bi_end_io = blkdev_bio_end_io;
0363 
0364         ret = bio_iov_iter_get_pages(bio, iter);
0365         if (unlikely(ret)) {
0366             bio->bi_error = ret;
0367             bio_endio(bio);
0368             break;
0369         }
0370 
0371         if (is_read) {
0372             bio->bi_opf = REQ_OP_READ;
0373             if (dio->should_dirty)
0374                 bio_set_pages_dirty(bio);
0375         } else {
0376             bio->bi_opf = dio_bio_write_op(iocb);
0377             task_io_account_write(bio->bi_iter.bi_size);
0378         }
0379 
0380         dio->size += bio->bi_iter.bi_size;
0381         pos += bio->bi_iter.bi_size;
0382 
0383         nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES);
0384         if (!nr_pages) {
0385             qc = submit_bio(bio);
0386             break;
0387         }
0388 
0389         if (!dio->multi_bio) {
0390             dio->multi_bio = true;
0391             atomic_set(&dio->ref, 2);
0392         } else {
0393             atomic_inc(&dio->ref);
0394         }
0395 
0396         submit_bio(bio);
0397         bio = bio_alloc(GFP_KERNEL, nr_pages);
0398     }
0399     blk_finish_plug(&plug);
0400 
0401     if (!is_sync)
0402         return -EIOCBQUEUED;
0403 
0404     for (;;) {
0405         set_current_state(TASK_UNINTERRUPTIBLE);
0406         if (!READ_ONCE(dio->waiter))
0407             break;
0408 
0409         if (!(iocb->ki_flags & IOCB_HIPRI) ||
0410             !blk_mq_poll(bdev_get_queue(bdev), qc))
0411             io_schedule();
0412     }
0413     __set_current_state(TASK_RUNNING);
0414 
0415     ret = dio->bio.bi_error;
0416     if (likely(!ret))
0417         ret = dio->size;
0418 
0419     bio_put(&dio->bio);
0420     return ret;
0421 }
0422 
0423 static ssize_t
0424 blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
0425 {
0426     int nr_pages;
0427 
0428     nr_pages = iov_iter_npages(iter, BIO_MAX_PAGES + 1);
0429     if (!nr_pages)
0430         return 0;
0431     if (is_sync_kiocb(iocb) && nr_pages <= BIO_MAX_PAGES)
0432         return __blkdev_direct_IO_simple(iocb, iter, nr_pages);
0433 
0434     return __blkdev_direct_IO(iocb, iter, min(nr_pages, BIO_MAX_PAGES));
0435 }
0436 
0437 static __init int blkdev_init(void)
0438 {
0439     blkdev_dio_pool = bioset_create(4, offsetof(struct blkdev_dio, bio));
0440     if (!blkdev_dio_pool)
0441         return -ENOMEM;
0442     return 0;
0443 }
0444 module_init(blkdev_init);
0445 
0446 int __sync_blockdev(struct block_device *bdev, int wait)
0447 {
0448     if (!bdev)
0449         return 0;
0450     if (!wait)
0451         return filemap_flush(bdev->bd_inode->i_mapping);
0452     return filemap_write_and_wait(bdev->bd_inode->i_mapping);
0453 }
0454 
0455 /*
0456  * Write out and wait upon all the dirty data associated with a block
0457  * device via its mapping.  Does not take the superblock lock.
0458  */
0459 int sync_blockdev(struct block_device *bdev)
0460 {
0461     return __sync_blockdev(bdev, 1);
0462 }
0463 EXPORT_SYMBOL(sync_blockdev);
0464 
0465 /*
0466  * Write out and wait upon all dirty data associated with this
0467  * device.   Filesystem data as well as the underlying block
0468  * device.  Takes the superblock lock.
0469  */
0470 int fsync_bdev(struct block_device *bdev)
0471 {
0472     struct super_block *sb = get_super(bdev);
0473     if (sb) {
0474         int res = sync_filesystem(sb);
0475         drop_super(sb);
0476         return res;
0477     }
0478     return sync_blockdev(bdev);
0479 }
0480 EXPORT_SYMBOL(fsync_bdev);
0481 
0482 /**
0483  * freeze_bdev  --  lock a filesystem and force it into a consistent state
0484  * @bdev:   blockdevice to lock
0485  *
0486  * If a superblock is found on this device, we take the s_umount semaphore
0487  * on it to make sure nobody unmounts until the snapshot creation is done.
0488  * The reference counter (bd_fsfreeze_count) guarantees that only the last
0489  * unfreeze process can unfreeze the frozen filesystem actually when multiple
0490  * freeze requests arrive simultaneously. It counts up in freeze_bdev() and
0491  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
0492  * actually.
0493  */
0494 struct super_block *freeze_bdev(struct block_device *bdev)
0495 {
0496     struct super_block *sb;
0497     int error = 0;
0498 
0499     mutex_lock(&bdev->bd_fsfreeze_mutex);
0500     if (++bdev->bd_fsfreeze_count > 1) {
0501         /*
0502          * We don't even need to grab a reference - the first call
0503          * to freeze_bdev grab an active reference and only the last
0504          * thaw_bdev drops it.
0505          */
0506         sb = get_super(bdev);
0507         if (sb)
0508             drop_super(sb);
0509         mutex_unlock(&bdev->bd_fsfreeze_mutex);
0510         return sb;
0511     }
0512 
0513     sb = get_active_super(bdev);
0514     if (!sb)
0515         goto out;
0516     if (sb->s_op->freeze_super)
0517         error = sb->s_op->freeze_super(sb);
0518     else
0519         error = freeze_super(sb);
0520     if (error) {
0521         deactivate_super(sb);
0522         bdev->bd_fsfreeze_count--;
0523         mutex_unlock(&bdev->bd_fsfreeze_mutex);
0524         return ERR_PTR(error);
0525     }
0526     deactivate_super(sb);
0527  out:
0528     sync_blockdev(bdev);
0529     mutex_unlock(&bdev->bd_fsfreeze_mutex);
0530     return sb;  /* thaw_bdev releases s->s_umount */
0531 }
0532 EXPORT_SYMBOL(freeze_bdev);
0533 
0534 /**
0535  * thaw_bdev  -- unlock filesystem
0536  * @bdev:   blockdevice to unlock
0537  * @sb:     associated superblock
0538  *
0539  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
0540  */
0541 int thaw_bdev(struct block_device *bdev, struct super_block *sb)
0542 {
0543     int error = -EINVAL;
0544 
0545     mutex_lock(&bdev->bd_fsfreeze_mutex);
0546     if (!bdev->bd_fsfreeze_count)
0547         goto out;
0548 
0549     error = 0;
0550     if (--bdev->bd_fsfreeze_count > 0)
0551         goto out;
0552 
0553     if (!sb)
0554         goto out;
0555 
0556     if (sb->s_op->thaw_super)
0557         error = sb->s_op->thaw_super(sb);
0558     else
0559         error = thaw_super(sb);
0560     if (error)
0561         bdev->bd_fsfreeze_count++;
0562 out:
0563     mutex_unlock(&bdev->bd_fsfreeze_mutex);
0564     return error;
0565 }
0566 EXPORT_SYMBOL(thaw_bdev);
0567 
0568 static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
0569 {
0570     return block_write_full_page(page, blkdev_get_block, wbc);
0571 }
0572 
0573 static int blkdev_readpage(struct file * file, struct page * page)
0574 {
0575     return block_read_full_page(page, blkdev_get_block);
0576 }
0577 
0578 static int blkdev_readpages(struct file *file, struct address_space *mapping,
0579             struct list_head *pages, unsigned nr_pages)
0580 {
0581     return mpage_readpages(mapping, pages, nr_pages, blkdev_get_block);
0582 }
0583 
0584 static int blkdev_write_begin(struct file *file, struct address_space *mapping,
0585             loff_t pos, unsigned len, unsigned flags,
0586             struct page **pagep, void **fsdata)
0587 {
0588     return block_write_begin(mapping, pos, len, flags, pagep,
0589                  blkdev_get_block);
0590 }
0591 
0592 static int blkdev_write_end(struct file *file, struct address_space *mapping,
0593             loff_t pos, unsigned len, unsigned copied,
0594             struct page *page, void *fsdata)
0595 {
0596     int ret;
0597     ret = block_write_end(file, mapping, pos, len, copied, page, fsdata);
0598 
0599     unlock_page(page);
0600     put_page(page);
0601 
0602     return ret;
0603 }
0604 
0605 /*
0606  * private llseek:
0607  * for a block special file file_inode(file)->i_size is zero
0608  * so we compute the size by hand (just as in block_read/write above)
0609  */
0610 static loff_t block_llseek(struct file *file, loff_t offset, int whence)
0611 {
0612     struct inode *bd_inode = bdev_file_inode(file);
0613     loff_t retval;
0614 
0615     inode_lock(bd_inode);
0616     retval = fixed_size_llseek(file, offset, whence, i_size_read(bd_inode));
0617     inode_unlock(bd_inode);
0618     return retval;
0619 }
0620     
0621 int blkdev_fsync(struct file *filp, loff_t start, loff_t end, int datasync)
0622 {
0623     struct inode *bd_inode = bdev_file_inode(filp);
0624     struct block_device *bdev = I_BDEV(bd_inode);
0625     int error;
0626     
0627     error = filemap_write_and_wait_range(filp->f_mapping, start, end);
0628     if (error)
0629         return error;
0630 
0631     /*
0632      * There is no need to serialise calls to blkdev_issue_flush with
0633      * i_mutex and doing so causes performance issues with concurrent
0634      * O_SYNC writers to a block device.
0635      */
0636     error = blkdev_issue_flush(bdev, GFP_KERNEL, NULL);
0637     if (error == -EOPNOTSUPP)
0638         error = 0;
0639 
0640     return error;
0641 }
0642 EXPORT_SYMBOL(blkdev_fsync);
0643 
0644 /**
0645  * bdev_read_page() - Start reading a page from a block device
0646  * @bdev: The device to read the page from
0647  * @sector: The offset on the device to read the page to (need not be aligned)
0648  * @page: The page to read
0649  *
0650  * On entry, the page should be locked.  It will be unlocked when the page
0651  * has been read.  If the block driver implements rw_page synchronously,
0652  * that will be true on exit from this function, but it need not be.
0653  *
0654  * Errors returned by this function are usually "soft", eg out of memory, or
0655  * queue full; callers should try a different route to read this page rather
0656  * than propagate an error back up the stack.
0657  *
0658  * Return: negative errno if an error occurs, 0 if submission was successful.
0659  */
0660 int bdev_read_page(struct block_device *bdev, sector_t sector,
0661             struct page *page)
0662 {
0663     const struct block_device_operations *ops = bdev->bd_disk->fops;
0664     int result = -EOPNOTSUPP;
0665 
0666     if (!ops->rw_page || bdev_get_integrity(bdev))
0667         return result;
0668 
0669     result = blk_queue_enter(bdev->bd_queue, false);
0670     if (result)
0671         return result;
0672     result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, false);
0673     blk_queue_exit(bdev->bd_queue);
0674     return result;
0675 }
0676 EXPORT_SYMBOL_GPL(bdev_read_page);
0677 
0678 /**
0679  * bdev_write_page() - Start writing a page to a block device
0680  * @bdev: The device to write the page to
0681  * @sector: The offset on the device to write the page to (need not be aligned)
0682  * @page: The page to write
0683  * @wbc: The writeback_control for the write
0684  *
0685  * On entry, the page should be locked and not currently under writeback.
0686  * On exit, if the write started successfully, the page will be unlocked and
0687  * under writeback.  If the write failed already (eg the driver failed to
0688  * queue the page to the device), the page will still be locked.  If the
0689  * caller is a ->writepage implementation, it will need to unlock the page.
0690  *
0691  * Errors returned by this function are usually "soft", eg out of memory, or
0692  * queue full; callers should try a different route to write this page rather
0693  * than propagate an error back up the stack.
0694  *
0695  * Return: negative errno if an error occurs, 0 if submission was successful.
0696  */
0697 int bdev_write_page(struct block_device *bdev, sector_t sector,
0698             struct page *page, struct writeback_control *wbc)
0699 {
0700     int result;
0701     const struct block_device_operations *ops = bdev->bd_disk->fops;
0702 
0703     if (!ops->rw_page || bdev_get_integrity(bdev))
0704         return -EOPNOTSUPP;
0705     result = blk_queue_enter(bdev->bd_queue, false);
0706     if (result)
0707         return result;
0708 
0709     set_page_writeback(page);
0710     result = ops->rw_page(bdev, sector + get_start_sect(bdev), page, true);
0711     if (result)
0712         end_page_writeback(page);
0713     else
0714         unlock_page(page);
0715     blk_queue_exit(bdev->bd_queue);
0716     return result;
0717 }
0718 EXPORT_SYMBOL_GPL(bdev_write_page);
0719 
0720 /**
0721  * bdev_direct_access() - Get the address for directly-accessibly memory
0722  * @bdev: The device containing the memory
0723  * @dax: control and output parameters for ->direct_access
0724  *
0725  * If a block device is made up of directly addressable memory, this function
0726  * will tell the caller the PFN and the address of the memory.  The address
0727  * may be directly dereferenced within the kernel without the need to call
0728  * ioremap(), kmap() or similar.  The PFN is suitable for inserting into
0729  * page tables.
0730  *
0731  * Return: negative errno if an error occurs, otherwise the number of bytes
0732  * accessible at this address.
0733  */
0734 long bdev_direct_access(struct block_device *bdev, struct blk_dax_ctl *dax)
0735 {
0736     sector_t sector = dax->sector;
0737     long avail, size = dax->size;
0738     const struct block_device_operations *ops = bdev->bd_disk->fops;
0739 
0740     /*
0741      * The device driver is allowed to sleep, in order to make the
0742      * memory directly accessible.
0743      */
0744     might_sleep();
0745 
0746     if (size < 0)
0747         return size;
0748     if (!blk_queue_dax(bdev_get_queue(bdev)) || !ops->direct_access)
0749         return -EOPNOTSUPP;
0750     if ((sector + DIV_ROUND_UP(size, 512)) >
0751                     part_nr_sects_read(bdev->bd_part))
0752         return -ERANGE;
0753     sector += get_start_sect(bdev);
0754     if (sector % (PAGE_SIZE / 512))
0755         return -EINVAL;
0756     avail = ops->direct_access(bdev, sector, &dax->addr, &dax->pfn, size);
0757     if (!avail)
0758         return -ERANGE;
0759     if (avail > 0 && avail & ~PAGE_MASK)
0760         return -ENXIO;
0761     return min(avail, size);
0762 }
0763 EXPORT_SYMBOL_GPL(bdev_direct_access);
0764 
0765 /**
0766  * bdev_dax_supported() - Check if the device supports dax for filesystem
0767  * @sb: The superblock of the device
0768  * @blocksize: The block size of the device
0769  *
0770  * This is a library function for filesystems to check if the block device
0771  * can be mounted with dax option.
0772  *
0773  * Return: negative errno if unsupported, 0 if supported.
0774  */
0775 int bdev_dax_supported(struct super_block *sb, int blocksize)
0776 {
0777     struct blk_dax_ctl dax = {
0778         .sector = 0,
0779         .size = PAGE_SIZE,
0780     };
0781     int err;
0782 
0783     if (blocksize != PAGE_SIZE) {
0784         vfs_msg(sb, KERN_ERR, "error: unsupported blocksize for dax");
0785         return -EINVAL;
0786     }
0787 
0788     err = bdev_direct_access(sb->s_bdev, &dax);
0789     if (err < 0) {
0790         switch (err) {
0791         case -EOPNOTSUPP:
0792             vfs_msg(sb, KERN_ERR,
0793                 "error: device does not support dax");
0794             break;
0795         case -EINVAL:
0796             vfs_msg(sb, KERN_ERR,
0797                 "error: unaligned partition for dax");
0798             break;
0799         default:
0800             vfs_msg(sb, KERN_ERR,
0801                 "error: dax access failed (%d)", err);
0802         }
0803         return err;
0804     }
0805 
0806     return 0;
0807 }
0808 EXPORT_SYMBOL_GPL(bdev_dax_supported);
0809 
0810 /**
0811  * bdev_dax_capable() - Return if the raw device is capable for dax
0812  * @bdev: The device for raw block device access
0813  */
0814 bool bdev_dax_capable(struct block_device *bdev)
0815 {
0816     struct blk_dax_ctl dax = {
0817         .size = PAGE_SIZE,
0818     };
0819 
0820     if (!IS_ENABLED(CONFIG_FS_DAX))
0821         return false;
0822 
0823     dax.sector = 0;
0824     if (bdev_direct_access(bdev, &dax) < 0)
0825         return false;
0826 
0827     dax.sector = bdev->bd_part->nr_sects - (PAGE_SIZE / 512);
0828     if (bdev_direct_access(bdev, &dax) < 0)
0829         return false;
0830 
0831     return true;
0832 }
0833 
0834 /*
0835  * pseudo-fs
0836  */
0837 
0838 static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
0839 static struct kmem_cache * bdev_cachep __read_mostly;
0840 
0841 static struct inode *bdev_alloc_inode(struct super_block *sb)
0842 {
0843     struct bdev_inode *ei = kmem_cache_alloc(bdev_cachep, GFP_KERNEL);
0844     if (!ei)
0845         return NULL;
0846     return &ei->vfs_inode;
0847 }
0848 
0849 static void bdev_i_callback(struct rcu_head *head)
0850 {
0851     struct inode *inode = container_of(head, struct inode, i_rcu);
0852     struct bdev_inode *bdi = BDEV_I(inode);
0853 
0854     kmem_cache_free(bdev_cachep, bdi);
0855 }
0856 
0857 static void bdev_destroy_inode(struct inode *inode)
0858 {
0859     call_rcu(&inode->i_rcu, bdev_i_callback);
0860 }
0861 
0862 static void init_once(void *foo)
0863 {
0864     struct bdev_inode *ei = (struct bdev_inode *) foo;
0865     struct block_device *bdev = &ei->bdev;
0866 
0867     memset(bdev, 0, sizeof(*bdev));
0868     mutex_init(&bdev->bd_mutex);
0869     INIT_LIST_HEAD(&bdev->bd_list);
0870 #ifdef CONFIG_SYSFS
0871     INIT_LIST_HEAD(&bdev->bd_holder_disks);
0872 #endif
0873     inode_init_once(&ei->vfs_inode);
0874     /* Initialize mutex for freeze. */
0875     mutex_init(&bdev->bd_fsfreeze_mutex);
0876 }
0877 
0878 static void bdev_evict_inode(struct inode *inode)
0879 {
0880     struct block_device *bdev = &BDEV_I(inode)->bdev;
0881     truncate_inode_pages_final(&inode->i_data);
0882     invalidate_inode_buffers(inode); /* is it needed here? */
0883     clear_inode(inode);
0884     spin_lock(&bdev_lock);
0885     list_del_init(&bdev->bd_list);
0886     spin_unlock(&bdev_lock);
0887 }
0888 
0889 static const struct super_operations bdev_sops = {
0890     .statfs = simple_statfs,
0891     .alloc_inode = bdev_alloc_inode,
0892     .destroy_inode = bdev_destroy_inode,
0893     .drop_inode = generic_delete_inode,
0894     .evict_inode = bdev_evict_inode,
0895 };
0896 
0897 static struct dentry *bd_mount(struct file_system_type *fs_type,
0898     int flags, const char *dev_name, void *data)
0899 {
0900     struct dentry *dent;
0901     dent = mount_pseudo(fs_type, "bdev:", &bdev_sops, NULL, BDEVFS_MAGIC);
0902     if (!IS_ERR(dent))
0903         dent->d_sb->s_iflags |= SB_I_CGROUPWB;
0904     return dent;
0905 }
0906 
0907 static struct file_system_type bd_type = {
0908     .name       = "bdev",
0909     .mount      = bd_mount,
0910     .kill_sb    = kill_anon_super,
0911 };
0912 
0913 struct super_block *blockdev_superblock __read_mostly;
0914 EXPORT_SYMBOL_GPL(blockdev_superblock);
0915 
0916 void __init bdev_cache_init(void)
0917 {
0918     int err;
0919     static struct vfsmount *bd_mnt;
0920 
0921     bdev_cachep = kmem_cache_create("bdev_cache", sizeof(struct bdev_inode),
0922             0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
0923                 SLAB_MEM_SPREAD|SLAB_ACCOUNT|SLAB_PANIC),
0924             init_once);
0925     err = register_filesystem(&bd_type);
0926     if (err)
0927         panic("Cannot register bdev pseudo-fs");
0928     bd_mnt = kern_mount(&bd_type);
0929     if (IS_ERR(bd_mnt))
0930         panic("Cannot create bdev pseudo-fs");
0931     blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
0932 }
0933 
0934 /*
0935  * Most likely _very_ bad one - but then it's hardly critical for small
0936  * /dev and can be fixed when somebody will need really large one.
0937  * Keep in mind that it will be fed through icache hash function too.
0938  */
0939 static inline unsigned long hash(dev_t dev)
0940 {
0941     return MAJOR(dev)+MINOR(dev);
0942 }
0943 
0944 static int bdev_test(struct inode *inode, void *data)
0945 {
0946     return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
0947 }
0948 
0949 static int bdev_set(struct inode *inode, void *data)
0950 {
0951     BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
0952     return 0;
0953 }
0954 
0955 static LIST_HEAD(all_bdevs);
0956 
0957 struct block_device *bdget(dev_t dev)
0958 {
0959     struct block_device *bdev;
0960     struct inode *inode;
0961 
0962     inode = iget5_locked(blockdev_superblock, hash(dev),
0963             bdev_test, bdev_set, &dev);
0964 
0965     if (!inode)
0966         return NULL;
0967 
0968     bdev = &BDEV_I(inode)->bdev;
0969 
0970     if (inode->i_state & I_NEW) {
0971         bdev->bd_contains = NULL;
0972         bdev->bd_super = NULL;
0973         bdev->bd_inode = inode;
0974         bdev->bd_block_size = (1 << inode->i_blkbits);
0975         bdev->bd_part_count = 0;
0976         bdev->bd_invalidated = 0;
0977         inode->i_mode = S_IFBLK;
0978         inode->i_rdev = dev;
0979         inode->i_bdev = bdev;
0980         inode->i_data.a_ops = &def_blk_aops;
0981         mapping_set_gfp_mask(&inode->i_data, GFP_USER);
0982         spin_lock(&bdev_lock);
0983         list_add(&bdev->bd_list, &all_bdevs);
0984         spin_unlock(&bdev_lock);
0985         unlock_new_inode(inode);
0986     }
0987     return bdev;
0988 }
0989 
0990 EXPORT_SYMBOL(bdget);
0991 
0992 /**
0993  * bdgrab -- Grab a reference to an already referenced block device
0994  * @bdev:   Block device to grab a reference to.
0995  */
0996 struct block_device *bdgrab(struct block_device *bdev)
0997 {
0998     ihold(bdev->bd_inode);
0999     return bdev;
1000 }
1001 EXPORT_SYMBOL(bdgrab);
1002 
1003 long nr_blockdev_pages(void)
1004 {
1005     struct block_device *bdev;
1006     long ret = 0;
1007     spin_lock(&bdev_lock);
1008     list_for_each_entry(bdev, &all_bdevs, bd_list) {
1009         ret += bdev->bd_inode->i_mapping->nrpages;
1010     }
1011     spin_unlock(&bdev_lock);
1012     return ret;
1013 }
1014 
1015 void bdput(struct block_device *bdev)
1016 {
1017     iput(bdev->bd_inode);
1018 }
1019 
1020 EXPORT_SYMBOL(bdput);
1021  
1022 static struct block_device *bd_acquire(struct inode *inode)
1023 {
1024     struct block_device *bdev;
1025 
1026     spin_lock(&bdev_lock);
1027     bdev = inode->i_bdev;
1028     if (bdev) {
1029         bdgrab(bdev);
1030         spin_unlock(&bdev_lock);
1031         return bdev;
1032     }
1033     spin_unlock(&bdev_lock);
1034 
1035     bdev = bdget(inode->i_rdev);
1036     if (bdev) {
1037         spin_lock(&bdev_lock);
1038         if (!inode->i_bdev) {
1039             /*
1040              * We take an additional reference to bd_inode,
1041              * and it's released in clear_inode() of inode.
1042              * So, we can access it via ->i_mapping always
1043              * without igrab().
1044              */
1045             bdgrab(bdev);
1046             inode->i_bdev = bdev;
1047             inode->i_mapping = bdev->bd_inode->i_mapping;
1048         }
1049         spin_unlock(&bdev_lock);
1050     }
1051     return bdev;
1052 }
1053 
1054 /* Call when you free inode */
1055 
1056 void bd_forget(struct inode *inode)
1057 {
1058     struct block_device *bdev = NULL;
1059 
1060     spin_lock(&bdev_lock);
1061     if (!sb_is_blkdev_sb(inode->i_sb))
1062         bdev = inode->i_bdev;
1063     inode->i_bdev = NULL;
1064     inode->i_mapping = &inode->i_data;
1065     spin_unlock(&bdev_lock);
1066 
1067     if (bdev)
1068         bdput(bdev);
1069 }
1070 
1071 /**
1072  * bd_may_claim - test whether a block device can be claimed
1073  * @bdev: block device of interest
1074  * @whole: whole block device containing @bdev, may equal @bdev
1075  * @holder: holder trying to claim @bdev
1076  *
1077  * Test whether @bdev can be claimed by @holder.
1078  *
1079  * CONTEXT:
1080  * spin_lock(&bdev_lock).
1081  *
1082  * RETURNS:
1083  * %true if @bdev can be claimed, %false otherwise.
1084  */
1085 static bool bd_may_claim(struct block_device *bdev, struct block_device *whole,
1086              void *holder)
1087 {
1088     if (bdev->bd_holder == holder)
1089         return true;     /* already a holder */
1090     else if (bdev->bd_holder != NULL)
1091         return false;    /* held by someone else */
1092     else if (whole == bdev)
1093         return true;     /* is a whole device which isn't held */
1094 
1095     else if (whole->bd_holder == bd_may_claim)
1096         return true;     /* is a partition of a device that is being partitioned */
1097     else if (whole->bd_holder != NULL)
1098         return false;    /* is a partition of a held device */
1099     else
1100         return true;     /* is a partition of an un-held device */
1101 }
1102 
1103 /**
1104  * bd_prepare_to_claim - prepare to claim a block device
1105  * @bdev: block device of interest
1106  * @whole: the whole device containing @bdev, may equal @bdev
1107  * @holder: holder trying to claim @bdev
1108  *
1109  * Prepare to claim @bdev.  This function fails if @bdev is already
1110  * claimed by another holder and waits if another claiming is in
1111  * progress.  This function doesn't actually claim.  On successful
1112  * return, the caller has ownership of bd_claiming and bd_holder[s].
1113  *
1114  * CONTEXT:
1115  * spin_lock(&bdev_lock).  Might release bdev_lock, sleep and regrab
1116  * it multiple times.
1117  *
1118  * RETURNS:
1119  * 0 if @bdev can be claimed, -EBUSY otherwise.
1120  */
1121 static int bd_prepare_to_claim(struct block_device *bdev,
1122                    struct block_device *whole, void *holder)
1123 {
1124 retry:
1125     /* if someone else claimed, fail */
1126     if (!bd_may_claim(bdev, whole, holder))
1127         return -EBUSY;
1128 
1129     /* if claiming is already in progress, wait for it to finish */
1130     if (whole->bd_claiming) {
1131         wait_queue_head_t *wq = bit_waitqueue(&whole->bd_claiming, 0);
1132         DEFINE_WAIT(wait);
1133 
1134         prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
1135         spin_unlock(&bdev_lock);
1136         schedule();
1137         finish_wait(wq, &wait);
1138         spin_lock(&bdev_lock);
1139         goto retry;
1140     }
1141 
1142     /* yay, all mine */
1143     return 0;
1144 }
1145 
1146 /**
1147  * bd_start_claiming - start claiming a block device
1148  * @bdev: block device of interest
1149  * @holder: holder trying to claim @bdev
1150  *
1151  * @bdev is about to be opened exclusively.  Check @bdev can be opened
1152  * exclusively and mark that an exclusive open is in progress.  Each
1153  * successful call to this function must be matched with a call to
1154  * either bd_finish_claiming() or bd_abort_claiming() (which do not
1155  * fail).
1156  *
1157  * This function is used to gain exclusive access to the block device
1158  * without actually causing other exclusive open attempts to fail. It
1159  * should be used when the open sequence itself requires exclusive
1160  * access but may subsequently fail.
1161  *
1162  * CONTEXT:
1163  * Might sleep.
1164  *
1165  * RETURNS:
1166  * Pointer to the block device containing @bdev on success, ERR_PTR()
1167  * value on failure.
1168  */
1169 static struct block_device *bd_start_claiming(struct block_device *bdev,
1170                           void *holder)
1171 {
1172     struct gendisk *disk;
1173     struct block_device *whole;
1174     int partno, err;
1175 
1176     might_sleep();
1177 
1178     /*
1179      * @bdev might not have been initialized properly yet, look up
1180      * and grab the outer block device the hard way.
1181      */
1182     disk = get_gendisk(bdev->bd_dev, &partno);
1183     if (!disk)
1184         return ERR_PTR(-ENXIO);
1185 
1186     /*
1187      * Normally, @bdev should equal what's returned from bdget_disk()
1188      * if partno is 0; however, some drivers (floppy) use multiple
1189      * bdev's for the same physical device and @bdev may be one of the
1190      * aliases.  Keep @bdev if partno is 0.  This means claimer
1191      * tracking is broken for those devices but it has always been that
1192      * way.
1193      */
1194     if (partno)
1195         whole = bdget_disk(disk, 0);
1196     else
1197         whole = bdgrab(bdev);
1198 
1199     module_put(disk->fops->owner);
1200     put_disk(disk);
1201     if (!whole)
1202         return ERR_PTR(-ENOMEM);
1203 
1204     /* prepare to claim, if successful, mark claiming in progress */
1205     spin_lock(&bdev_lock);
1206 
1207     err = bd_prepare_to_claim(bdev, whole, holder);
1208     if (err == 0) {
1209         whole->bd_claiming = holder;
1210         spin_unlock(&bdev_lock);
1211         return whole;
1212     } else {
1213         spin_unlock(&bdev_lock);
1214         bdput(whole);
1215         return ERR_PTR(err);
1216     }
1217 }
1218 
1219 #ifdef CONFIG_SYSFS
1220 struct bd_holder_disk {
1221     struct list_head    list;
1222     struct gendisk      *disk;
1223     int         refcnt;
1224 };
1225 
1226 static struct bd_holder_disk *bd_find_holder_disk(struct block_device *bdev,
1227                           struct gendisk *disk)
1228 {
1229     struct bd_holder_disk *holder;
1230 
1231     list_for_each_entry(holder, &bdev->bd_holder_disks, list)
1232         if (holder->disk == disk)
1233             return holder;
1234     return NULL;
1235 }
1236 
1237 static int add_symlink(struct kobject *from, struct kobject *to)
1238 {
1239     return sysfs_create_link(from, to, kobject_name(to));
1240 }
1241 
1242 static void del_symlink(struct kobject *from, struct kobject *to)
1243 {
1244     sysfs_remove_link(from, kobject_name(to));
1245 }
1246 
1247 /**
1248  * bd_link_disk_holder - create symlinks between holding disk and slave bdev
1249  * @bdev: the claimed slave bdev
1250  * @disk: the holding disk
1251  *
1252  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1253  *
1254  * This functions creates the following sysfs symlinks.
1255  *
1256  * - from "slaves" directory of the holder @disk to the claimed @bdev
1257  * - from "holders" directory of the @bdev to the holder @disk
1258  *
1259  * For example, if /dev/dm-0 maps to /dev/sda and disk for dm-0 is
1260  * passed to bd_link_disk_holder(), then:
1261  *
1262  *   /sys/block/dm-0/slaves/sda --> /sys/block/sda
1263  *   /sys/block/sda/holders/dm-0 --> /sys/block/dm-0
1264  *
1265  * The caller must have claimed @bdev before calling this function and
1266  * ensure that both @bdev and @disk are valid during the creation and
1267  * lifetime of these symlinks.
1268  *
1269  * CONTEXT:
1270  * Might sleep.
1271  *
1272  * RETURNS:
1273  * 0 on success, -errno on failure.
1274  */
1275 int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
1276 {
1277     struct bd_holder_disk *holder;
1278     int ret = 0;
1279 
1280     mutex_lock(&bdev->bd_mutex);
1281 
1282     WARN_ON_ONCE(!bdev->bd_holder);
1283 
1284     /* FIXME: remove the following once add_disk() handles errors */
1285     if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
1286         goto out_unlock;
1287 
1288     holder = bd_find_holder_disk(bdev, disk);
1289     if (holder) {
1290         holder->refcnt++;
1291         goto out_unlock;
1292     }
1293 
1294     holder = kzalloc(sizeof(*holder), GFP_KERNEL);
1295     if (!holder) {
1296         ret = -ENOMEM;
1297         goto out_unlock;
1298     }
1299 
1300     INIT_LIST_HEAD(&holder->list);
1301     holder->disk = disk;
1302     holder->refcnt = 1;
1303 
1304     ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1305     if (ret)
1306         goto out_free;
1307 
1308     ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
1309     if (ret)
1310         goto out_del;
1311     /*
1312      * bdev could be deleted beneath us which would implicitly destroy
1313      * the holder directory.  Hold on to it.
1314      */
1315     kobject_get(bdev->bd_part->holder_dir);
1316 
1317     list_add(&holder->list, &bdev->bd_holder_disks);
1318     goto out_unlock;
1319 
1320 out_del:
1321     del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1322 out_free:
1323     kfree(holder);
1324 out_unlock:
1325     mutex_unlock(&bdev->bd_mutex);
1326     return ret;
1327 }
1328 EXPORT_SYMBOL_GPL(bd_link_disk_holder);
1329 
1330 /**
1331  * bd_unlink_disk_holder - destroy symlinks created by bd_link_disk_holder()
1332  * @bdev: the calimed slave bdev
1333  * @disk: the holding disk
1334  *
1335  * DON'T USE THIS UNLESS YOU'RE ALREADY USING IT.
1336  *
1337  * CONTEXT:
1338  * Might sleep.
1339  */
1340 void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
1341 {
1342     struct bd_holder_disk *holder;
1343 
1344     mutex_lock(&bdev->bd_mutex);
1345 
1346     holder = bd_find_holder_disk(bdev, disk);
1347 
1348     if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
1349         del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
1350         del_symlink(bdev->bd_part->holder_dir,
1351                 &disk_to_dev(disk)->kobj);
1352         kobject_put(bdev->bd_part->holder_dir);
1353         list_del_init(&holder->list);
1354         kfree(holder);
1355     }
1356 
1357     mutex_unlock(&bdev->bd_mutex);
1358 }
1359 EXPORT_SYMBOL_GPL(bd_unlink_disk_holder);
1360 #endif
1361 
1362 /**
1363  * flush_disk - invalidates all buffer-cache entries on a disk
1364  *
1365  * @bdev:      struct block device to be flushed
1366  * @kill_dirty: flag to guide handling of dirty inodes
1367  *
1368  * Invalidates all buffer-cache entries on a disk. It should be called
1369  * when a disk has been changed -- either by a media change or online
1370  * resize.
1371  */
1372 static void flush_disk(struct block_device *bdev, bool kill_dirty)
1373 {
1374     if (__invalidate_device(bdev, kill_dirty)) {
1375         printk(KERN_WARNING "VFS: busy inodes on changed media or "
1376                "resized disk %s\n",
1377                bdev->bd_disk ? bdev->bd_disk->disk_name : "");
1378     }
1379 
1380     if (!bdev->bd_disk)
1381         return;
1382     if (disk_part_scan_enabled(bdev->bd_disk))
1383         bdev->bd_invalidated = 1;
1384 }
1385 
1386 /**
1387  * check_disk_size_change - checks for disk size change and adjusts bdev size.
1388  * @disk: struct gendisk to check
1389  * @bdev: struct bdev to adjust.
1390  *
1391  * This routine checks to see if the bdev size does not match the disk size
1392  * and adjusts it if it differs.
1393  */
1394 void check_disk_size_change(struct gendisk *disk, struct block_device *bdev)
1395 {
1396     loff_t disk_size, bdev_size;
1397 
1398     disk_size = (loff_t)get_capacity(disk) << 9;
1399     bdev_size = i_size_read(bdev->bd_inode);
1400     if (disk_size != bdev_size) {
1401         printk(KERN_INFO
1402                "%s: detected capacity change from %lld to %lld\n",
1403                disk->disk_name, bdev_size, disk_size);
1404         i_size_write(bdev->bd_inode, disk_size);
1405         flush_disk(bdev, false);
1406     }
1407 }
1408 EXPORT_SYMBOL(check_disk_size_change);
1409 
1410 /**
1411  * revalidate_disk - wrapper for lower-level driver's revalidate_disk call-back
1412  * @disk: struct gendisk to be revalidated
1413  *
1414  * This routine is a wrapper for lower-level driver's revalidate_disk
1415  * call-backs.  It is used to do common pre and post operations needed
1416  * for all revalidate_disk operations.
1417  */
1418 int revalidate_disk(struct gendisk *disk)
1419 {
1420     struct block_device *bdev;
1421     int ret = 0;
1422 
1423     if (disk->fops->revalidate_disk)
1424         ret = disk->fops->revalidate_disk(disk);
1425     blk_integrity_revalidate(disk);
1426     bdev = bdget_disk(disk, 0);
1427     if (!bdev)
1428         return ret;
1429 
1430     mutex_lock(&bdev->bd_mutex);
1431     check_disk_size_change(disk, bdev);
1432     bdev->bd_invalidated = 0;
1433     mutex_unlock(&bdev->bd_mutex);
1434     bdput(bdev);
1435     return ret;
1436 }
1437 EXPORT_SYMBOL(revalidate_disk);
1438 
1439 /*
1440  * This routine checks whether a removable media has been changed,
1441  * and invalidates all buffer-cache-entries in that case. This
1442  * is a relatively slow routine, so we have to try to minimize using
1443  * it. Thus it is called only upon a 'mount' or 'open'. This
1444  * is the best way of combining speed and utility, I think.
1445  * People changing diskettes in the middle of an operation deserve
1446  * to lose :-)
1447  */
1448 int check_disk_change(struct block_device *bdev)
1449 {
1450     struct gendisk *disk = bdev->bd_disk;
1451     const struct block_device_operations *bdops = disk->fops;
1452     unsigned int events;
1453 
1454     events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
1455                    DISK_EVENT_EJECT_REQUEST);
1456     if (!(events & DISK_EVENT_MEDIA_CHANGE))
1457         return 0;
1458 
1459     flush_disk(bdev, true);
1460     if (bdops->revalidate_disk)
1461         bdops->revalidate_disk(bdev->bd_disk);
1462     return 1;
1463 }
1464 
1465 EXPORT_SYMBOL(check_disk_change);
1466 
1467 void bd_set_size(struct block_device *bdev, loff_t size)
1468 {
1469     unsigned bsize = bdev_logical_block_size(bdev);
1470 
1471     inode_lock(bdev->bd_inode);
1472     i_size_write(bdev->bd_inode, size);
1473     inode_unlock(bdev->bd_inode);
1474     while (bsize < PAGE_SIZE) {
1475         if (size & bsize)
1476             break;
1477         bsize <<= 1;
1478     }
1479     bdev->bd_block_size = bsize;
1480     bdev->bd_inode->i_blkbits = blksize_bits(bsize);
1481 }
1482 EXPORT_SYMBOL(bd_set_size);
1483 
1484 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
1485 
1486 /*
1487  * bd_mutex locking:
1488  *
1489  *  mutex_lock(part->bd_mutex)
1490  *    mutex_lock_nested(whole->bd_mutex, 1)
1491  */
1492 
1493 static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
1494 {
1495     struct gendisk *disk;
1496     struct module *owner;
1497     int ret;
1498     int partno;
1499     int perm = 0;
1500 
1501     if (mode & FMODE_READ)
1502         perm |= MAY_READ;
1503     if (mode & FMODE_WRITE)
1504         perm |= MAY_WRITE;
1505     /*
1506      * hooks: /n/, see "layering violations".
1507      */
1508     if (!for_part) {
1509         ret = devcgroup_inode_permission(bdev->bd_inode, perm);
1510         if (ret != 0) {
1511             bdput(bdev);
1512             return ret;
1513         }
1514     }
1515 
1516  restart:
1517 
1518     ret = -ENXIO;
1519     disk = get_gendisk(bdev->bd_dev, &partno);
1520     if (!disk)
1521         goto out;
1522     owner = disk->fops->owner;
1523 
1524     disk_block_events(disk);
1525     mutex_lock_nested(&bdev->bd_mutex, for_part);
1526     if (!bdev->bd_openers) {
1527         bdev->bd_disk = disk;
1528         bdev->bd_queue = disk->queue;
1529         bdev->bd_contains = bdev;
1530 
1531         if (!partno) {
1532             ret = -ENXIO;
1533             bdev->bd_part = disk_get_part(disk, partno);
1534             if (!bdev->bd_part)
1535                 goto out_clear;
1536 
1537             ret = 0;
1538             if (disk->fops->open) {
1539                 ret = disk->fops->open(bdev, mode);
1540                 if (ret == -ERESTARTSYS) {
1541                     /* Lost a race with 'disk' being
1542                      * deleted, try again.
1543                      * See md.c
1544                      */
1545                     disk_put_part(bdev->bd_part);
1546                     bdev->bd_part = NULL;
1547                     bdev->bd_disk = NULL;
1548                     bdev->bd_queue = NULL;
1549                     mutex_unlock(&bdev->bd_mutex);
1550                     disk_unblock_events(disk);
1551                     put_disk(disk);
1552                     module_put(owner);
1553                     goto restart;
1554                 }
1555             }
1556 
1557             if (!ret)
1558                 bd_set_size(bdev,(loff_t)get_capacity(disk)<<9);
1559 
1560             /*
1561              * If the device is invalidated, rescan partition
1562              * if open succeeded or failed with -ENOMEDIUM.
1563              * The latter is necessary to prevent ghost
1564              * partitions on a removed medium.
1565              */
1566             if (bdev->bd_invalidated) {
1567                 if (!ret)
1568                     rescan_partitions(disk, bdev);
1569                 else if (ret == -ENOMEDIUM)
1570                     invalidate_partitions(disk, bdev);
1571             }
1572 
1573             if (ret)
1574                 goto out_clear;
1575         } else {
1576             struct block_device *whole;
1577             whole = bdget_disk(disk, 0);
1578             ret = -ENOMEM;
1579             if (!whole)
1580                 goto out_clear;
1581             BUG_ON(for_part);
1582             ret = __blkdev_get(whole, mode, 1);
1583             if (ret)
1584                 goto out_clear;
1585             bdev->bd_contains = whole;
1586             bdev->bd_part = disk_get_part(disk, partno);
1587             if (!(disk->flags & GENHD_FL_UP) ||
1588                 !bdev->bd_part || !bdev->bd_part->nr_sects) {
1589                 ret = -ENXIO;
1590                 goto out_clear;
1591             }
1592             bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
1593         }
1594     } else {
1595         if (bdev->bd_contains == bdev) {
1596             ret = 0;
1597             if (bdev->bd_disk->fops->open)
1598                 ret = bdev->bd_disk->fops->open(bdev, mode);
1599             /* the same as first opener case, read comment there */
1600             if (bdev->bd_invalidated) {
1601                 if (!ret)
1602                     rescan_partitions(bdev->bd_disk, bdev);
1603                 else if (ret == -ENOMEDIUM)
1604                     invalidate_partitions(bdev->bd_disk, bdev);
1605             }
1606             if (ret)
1607                 goto out_unlock_bdev;
1608         }
1609         /* only one opener holds refs to the module and disk */
1610         put_disk(disk);
1611         module_put(owner);
1612     }
1613     bdev->bd_openers++;
1614     if (for_part)
1615         bdev->bd_part_count++;
1616     mutex_unlock(&bdev->bd_mutex);
1617     disk_unblock_events(disk);
1618     return 0;
1619 
1620  out_clear:
1621     disk_put_part(bdev->bd_part);
1622     bdev->bd_disk = NULL;
1623     bdev->bd_part = NULL;
1624     bdev->bd_queue = NULL;
1625     if (bdev != bdev->bd_contains)
1626         __blkdev_put(bdev->bd_contains, mode, 1);
1627     bdev->bd_contains = NULL;
1628  out_unlock_bdev:
1629     mutex_unlock(&bdev->bd_mutex);
1630     disk_unblock_events(disk);
1631     put_disk(disk);
1632     module_put(owner);
1633  out:
1634     bdput(bdev);
1635 
1636     return ret;
1637 }
1638 
1639 /**
1640  * blkdev_get - open a block device
1641  * @bdev: block_device to open
1642  * @mode: FMODE_* mask
1643  * @holder: exclusive holder identifier
1644  *
1645  * Open @bdev with @mode.  If @mode includes %FMODE_EXCL, @bdev is
1646  * open with exclusive access.  Specifying %FMODE_EXCL with %NULL
1647  * @holder is invalid.  Exclusive opens may nest for the same @holder.
1648  *
1649  * On success, the reference count of @bdev is unchanged.  On failure,
1650  * @bdev is put.
1651  *
1652  * CONTEXT:
1653  * Might sleep.
1654  *
1655  * RETURNS:
1656  * 0 on success, -errno on failure.
1657  */
1658 int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
1659 {
1660     struct block_device *whole = NULL;
1661     int res;
1662 
1663     WARN_ON_ONCE((mode & FMODE_EXCL) && !holder);
1664 
1665     if ((mode & FMODE_EXCL) && holder) {
1666         whole = bd_start_claiming(bdev, holder);
1667         if (IS_ERR(whole)) {
1668             bdput(bdev);
1669             return PTR_ERR(whole);
1670         }
1671     }
1672 
1673     res = __blkdev_get(bdev, mode, 0);
1674 
1675     if (whole) {
1676         struct gendisk *disk = whole->bd_disk;
1677 
1678         /* finish claiming */
1679         mutex_lock(&bdev->bd_mutex);
1680         spin_lock(&bdev_lock);
1681 
1682         if (!res) {
1683             BUG_ON(!bd_may_claim(bdev, whole, holder));
1684             /*
1685              * Note that for a whole device bd_holders
1686              * will be incremented twice, and bd_holder
1687              * will be set to bd_may_claim before being
1688              * set to holder
1689              */
1690             whole->bd_holders++;
1691             whole->bd_holder = bd_may_claim;
1692             bdev->bd_holders++;
1693             bdev->bd_holder = holder;
1694         }
1695 
1696         /* tell others that we're done */
1697         BUG_ON(whole->bd_claiming != holder);
1698         whole->bd_claiming = NULL;
1699         wake_up_bit(&whole->bd_claiming, 0);
1700 
1701         spin_unlock(&bdev_lock);
1702 
1703         /*
1704          * Block event polling for write claims if requested.  Any
1705          * write holder makes the write_holder state stick until
1706          * all are released.  This is good enough and tracking
1707          * individual writeable reference is too fragile given the
1708          * way @mode is used in blkdev_get/put().
1709          */
1710         if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
1711             (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
1712             bdev->bd_write_holder = true;
1713             disk_block_events(disk);
1714         }
1715 
1716         mutex_unlock(&bdev->bd_mutex);
1717         bdput(whole);
1718     }
1719 
1720     return res;
1721 }
1722 EXPORT_SYMBOL(blkdev_get);
1723 
1724 /**
1725  * blkdev_get_by_path - open a block device by name
1726  * @path: path to the block device to open
1727  * @mode: FMODE_* mask
1728  * @holder: exclusive holder identifier
1729  *
1730  * Open the blockdevice described by the device file at @path.  @mode
1731  * and @holder are identical to blkdev_get().
1732  *
1733  * On success, the returned block_device has reference count of one.
1734  *
1735  * CONTEXT:
1736  * Might sleep.
1737  *
1738  * RETURNS:
1739  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1740  */
1741 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
1742                     void *holder)
1743 {
1744     struct block_device *bdev;
1745     int err;
1746 
1747     bdev = lookup_bdev(path);
1748     if (IS_ERR(bdev))
1749         return bdev;
1750 
1751     err = blkdev_get(bdev, mode, holder);
1752     if (err)
1753         return ERR_PTR(err);
1754 
1755     if ((mode & FMODE_WRITE) && bdev_read_only(bdev)) {
1756         blkdev_put(bdev, mode);
1757         return ERR_PTR(-EACCES);
1758     }
1759 
1760     return bdev;
1761 }
1762 EXPORT_SYMBOL(blkdev_get_by_path);
1763 
1764 /**
1765  * blkdev_get_by_dev - open a block device by device number
1766  * @dev: device number of block device to open
1767  * @mode: FMODE_* mask
1768  * @holder: exclusive holder identifier
1769  *
1770  * Open the blockdevice described by device number @dev.  @mode and
1771  * @holder are identical to blkdev_get().
1772  *
1773  * Use it ONLY if you really do not have anything better - i.e. when
1774  * you are behind a truly sucky interface and all you are given is a
1775  * device number.  _Never_ to be used for internal purposes.  If you
1776  * ever need it - reconsider your API.
1777  *
1778  * On success, the returned block_device has reference count of one.
1779  *
1780  * CONTEXT:
1781  * Might sleep.
1782  *
1783  * RETURNS:
1784  * Pointer to block_device on success, ERR_PTR(-errno) on failure.
1785  */
1786 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder)
1787 {
1788     struct block_device *bdev;
1789     int err;
1790 
1791     bdev = bdget(dev);
1792     if (!bdev)
1793         return ERR_PTR(-ENOMEM);
1794 
1795     err = blkdev_get(bdev, mode, holder);
1796     if (err)
1797         return ERR_PTR(err);
1798 
1799     return bdev;
1800 }
1801 EXPORT_SYMBOL(blkdev_get_by_dev);
1802 
1803 static int blkdev_open(struct inode * inode, struct file * filp)
1804 {
1805     struct block_device *bdev;
1806 
1807     /*
1808      * Preserve backwards compatibility and allow large file access
1809      * even if userspace doesn't ask for it explicitly. Some mkfs
1810      * binary needs it. We might want to drop this workaround
1811      * during an unstable branch.
1812      */
1813     filp->f_flags |= O_LARGEFILE;
1814 
1815     if (filp->f_flags & O_NDELAY)
1816         filp->f_mode |= FMODE_NDELAY;
1817     if (filp->f_flags & O_EXCL)
1818         filp->f_mode |= FMODE_EXCL;
1819     if ((filp->f_flags & O_ACCMODE) == 3)
1820         filp->f_mode |= FMODE_WRITE_IOCTL;
1821 
1822     bdev = bd_acquire(inode);
1823     if (bdev == NULL)
1824         return -ENOMEM;
1825 
1826     filp->f_mapping = bdev->bd_inode->i_mapping;
1827 
1828     return blkdev_get(bdev, filp->f_mode, filp);
1829 }
1830 
1831 static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
1832 {
1833     struct gendisk *disk = bdev->bd_disk;
1834     struct block_device *victim = NULL;
1835 
1836     mutex_lock_nested(&bdev->bd_mutex, for_part);
1837     if (for_part)
1838         bdev->bd_part_count--;
1839 
1840     if (!--bdev->bd_openers) {
1841         WARN_ON_ONCE(bdev->bd_holders);
1842         sync_blockdev(bdev);
1843         kill_bdev(bdev);
1844 
1845         bdev_write_inode(bdev);
1846         /*
1847          * Detaching bdev inode from its wb in __destroy_inode()
1848          * is too late: the queue which embeds its bdi (along with
1849          * root wb) can be gone as soon as we put_disk() below.
1850          */
1851         inode_detach_wb(bdev->bd_inode);
1852     }
1853     if (bdev->bd_contains == bdev) {
1854         if (disk->fops->release)
1855             disk->fops->release(disk, mode);
1856     }
1857     if (!bdev->bd_openers) {
1858         struct module *owner = disk->fops->owner;
1859 
1860         disk_put_part(bdev->bd_part);
1861         bdev->bd_part = NULL;
1862         bdev->bd_disk = NULL;
1863         if (bdev != bdev->bd_contains)
1864             victim = bdev->bd_contains;
1865         bdev->bd_contains = NULL;
1866 
1867         put_disk(disk);
1868         module_put(owner);
1869     }
1870     mutex_unlock(&bdev->bd_mutex);
1871     bdput(bdev);
1872     if (victim)
1873         __blkdev_put(victim, mode, 1);
1874 }
1875 
1876 void blkdev_put(struct block_device *bdev, fmode_t mode)
1877 {
1878     mutex_lock(&bdev->bd_mutex);
1879 
1880     if (mode & FMODE_EXCL) {
1881         bool bdev_free;
1882 
1883         /*
1884          * Release a claim on the device.  The holder fields
1885          * are protected with bdev_lock.  bd_mutex is to
1886          * synchronize disk_holder unlinking.
1887          */
1888         spin_lock(&bdev_lock);
1889 
1890         WARN_ON_ONCE(--bdev->bd_holders < 0);
1891         WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
1892 
1893         /* bd_contains might point to self, check in a separate step */
1894         if ((bdev_free = !bdev->bd_holders))
1895             bdev->bd_holder = NULL;
1896         if (!bdev->bd_contains->bd_holders)
1897             bdev->bd_contains->bd_holder = NULL;
1898 
1899         spin_unlock(&bdev_lock);
1900 
1901         /*
1902          * If this was the last claim, remove holder link and
1903          * unblock evpoll if it was a write holder.
1904          */
1905         if (bdev_free && bdev->bd_write_holder) {
1906             disk_unblock_events(bdev->bd_disk);
1907             bdev->bd_write_holder = false;
1908         }
1909     }
1910 
1911     /*
1912      * Trigger event checking and tell drivers to flush MEDIA_CHANGE
1913      * event.  This is to ensure detection of media removal commanded
1914      * from userland - e.g. eject(1).
1915      */
1916     disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
1917 
1918     mutex_unlock(&bdev->bd_mutex);
1919 
1920     __blkdev_put(bdev, mode, 0);
1921 }
1922 EXPORT_SYMBOL(blkdev_put);
1923 
1924 static int blkdev_close(struct inode * inode, struct file * filp)
1925 {
1926     struct block_device *bdev = I_BDEV(bdev_file_inode(filp));
1927     blkdev_put(bdev, filp->f_mode);
1928     return 0;
1929 }
1930 
1931 static long block_ioctl(struct file *file, unsigned cmd, unsigned long arg)
1932 {
1933     struct block_device *bdev = I_BDEV(bdev_file_inode(file));
1934     fmode_t mode = file->f_mode;
1935 
1936     /*
1937      * O_NDELAY can be altered using fcntl(.., F_SETFL, ..), so we have
1938      * to updated it before every ioctl.
1939      */
1940     if (file->f_flags & O_NDELAY)
1941         mode |= FMODE_NDELAY;
1942     else
1943         mode &= ~FMODE_NDELAY;
1944 
1945     return blkdev_ioctl(bdev, mode, cmd, arg);
1946 }
1947 
1948 /*
1949  * Write data to the block device.  Only intended for the block device itself
1950  * and the raw driver which basically is a fake block device.
1951  *
1952  * Does not take i_mutex for the write and thus is not for general purpose
1953  * use.
1954  */
1955 ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
1956 {
1957     struct file *file = iocb->ki_filp;
1958     struct inode *bd_inode = bdev_file_inode(file);
1959     loff_t size = i_size_read(bd_inode);
1960     struct blk_plug plug;
1961     ssize_t ret;
1962 
1963     if (bdev_read_only(I_BDEV(bd_inode)))
1964         return -EPERM;
1965 
1966     if (!iov_iter_count(from))
1967         return 0;
1968 
1969     if (iocb->ki_pos >= size)
1970         return -ENOSPC;
1971 
1972     iov_iter_truncate(from, size - iocb->ki_pos);
1973 
1974     blk_start_plug(&plug);
1975     ret = __generic_file_write_iter(iocb, from);
1976     if (ret > 0)
1977         ret = generic_write_sync(iocb, ret);
1978     blk_finish_plug(&plug);
1979     return ret;
1980 }
1981 EXPORT_SYMBOL_GPL(blkdev_write_iter);
1982 
1983 ssize_t blkdev_read_iter(struct kiocb *iocb, struct iov_iter *to)
1984 {
1985     struct file *file = iocb->ki_filp;
1986     struct inode *bd_inode = bdev_file_inode(file);
1987     loff_t size = i_size_read(bd_inode);
1988     loff_t pos = iocb->ki_pos;
1989 
1990     if (pos >= size)
1991         return 0;
1992 
1993     size -= pos;
1994     iov_iter_truncate(to, size);
1995     return generic_file_read_iter(iocb, to);
1996 }
1997 EXPORT_SYMBOL_GPL(blkdev_read_iter);
1998 
1999 /*
2000  * Try to release a page associated with block device when the system
2001  * is under memory pressure.
2002  */
2003 static int blkdev_releasepage(struct page *page, gfp_t wait)
2004 {
2005     struct super_block *super = BDEV_I(page->mapping->host)->bdev.bd_super;
2006 
2007     if (super && super->s_op->bdev_try_to_free_page)
2008         return super->s_op->bdev_try_to_free_page(super, page, wait);
2009 
2010     return try_to_free_buffers(page);
2011 }
2012 
2013 static int blkdev_writepages(struct address_space *mapping,
2014                  struct writeback_control *wbc)
2015 {
2016     if (dax_mapping(mapping)) {
2017         struct block_device *bdev = I_BDEV(mapping->host);
2018 
2019         return dax_writeback_mapping_range(mapping, bdev, wbc);
2020     }
2021     return generic_writepages(mapping, wbc);
2022 }
2023 
2024 static const struct address_space_operations def_blk_aops = {
2025     .readpage   = blkdev_readpage,
2026     .readpages  = blkdev_readpages,
2027     .writepage  = blkdev_writepage,
2028     .write_begin    = blkdev_write_begin,
2029     .write_end  = blkdev_write_end,
2030     .writepages = blkdev_writepages,
2031     .releasepage    = blkdev_releasepage,
2032     .direct_IO  = blkdev_direct_IO,
2033     .is_dirty_writeback = buffer_check_dirty_writeback,
2034 };
2035 
2036 #define BLKDEV_FALLOC_FL_SUPPORTED                  \
2037         (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE |       \
2038          FALLOC_FL_ZERO_RANGE | FALLOC_FL_NO_HIDE_STALE)
2039 
2040 static long blkdev_fallocate(struct file *file, int mode, loff_t start,
2041                  loff_t len)
2042 {
2043     struct block_device *bdev = I_BDEV(bdev_file_inode(file));
2044     struct request_queue *q = bdev_get_queue(bdev);
2045     struct address_space *mapping;
2046     loff_t end = start + len - 1;
2047     loff_t isize;
2048     int error;
2049 
2050     /* Fail if we don't recognize the flags. */
2051     if (mode & ~BLKDEV_FALLOC_FL_SUPPORTED)
2052         return -EOPNOTSUPP;
2053 
2054     /* Don't go off the end of the device. */
2055     isize = i_size_read(bdev->bd_inode);
2056     if (start >= isize)
2057         return -EINVAL;
2058     if (end >= isize) {
2059         if (mode & FALLOC_FL_KEEP_SIZE) {
2060             len = isize - start;
2061             end = start + len - 1;
2062         } else
2063             return -EINVAL;
2064     }
2065 
2066     /*
2067      * Don't allow IO that isn't aligned to logical block size.
2068      */
2069     if ((start | len) & (bdev_logical_block_size(bdev) - 1))
2070         return -EINVAL;
2071 
2072     /* Invalidate the page cache, including dirty pages. */
2073     mapping = bdev->bd_inode->i_mapping;
2074     truncate_inode_pages_range(mapping, start, end);
2075 
2076     switch (mode) {
2077     case FALLOC_FL_ZERO_RANGE:
2078     case FALLOC_FL_ZERO_RANGE | FALLOC_FL_KEEP_SIZE:
2079         error = blkdev_issue_zeroout(bdev, start >> 9, len >> 9,
2080                         GFP_KERNEL, false);
2081         break;
2082     case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE:
2083         /* Only punch if the device can do zeroing discard. */
2084         if (!blk_queue_discard(q) || !q->limits.discard_zeroes_data)
2085             return -EOPNOTSUPP;
2086         error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2087                          GFP_KERNEL, 0);
2088         break;
2089     case FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE | FALLOC_FL_NO_HIDE_STALE:
2090         if (!blk_queue_discard(q))
2091             return -EOPNOTSUPP;
2092         error = blkdev_issue_discard(bdev, start >> 9, len >> 9,
2093                          GFP_KERNEL, 0);
2094         break;
2095     default:
2096         return -EOPNOTSUPP;
2097     }
2098     if (error)
2099         return error;
2100 
2101     /*
2102      * Invalidate again; if someone wandered in and dirtied a page,
2103      * the caller will be given -EBUSY.  The third argument is
2104      * inclusive, so the rounding here is safe.
2105      */
2106     return invalidate_inode_pages2_range(mapping,
2107                          start >> PAGE_SHIFT,
2108                          end >> PAGE_SHIFT);
2109 }
2110 
2111 const struct file_operations def_blk_fops = {
2112     .open       = blkdev_open,
2113     .release    = blkdev_close,
2114     .llseek     = block_llseek,
2115     .read_iter  = blkdev_read_iter,
2116     .write_iter = blkdev_write_iter,
2117     .mmap       = generic_file_mmap,
2118     .fsync      = blkdev_fsync,
2119     .unlocked_ioctl = block_ioctl,
2120 #ifdef CONFIG_COMPAT
2121     .compat_ioctl   = compat_blkdev_ioctl,
2122 #endif
2123     .splice_read    = generic_file_splice_read,
2124     .splice_write   = iter_file_splice_write,
2125     .fallocate  = blkdev_fallocate,
2126 };
2127 
2128 int ioctl_by_bdev(struct block_device *bdev, unsigned cmd, unsigned long arg)
2129 {
2130     int res;
2131     mm_segment_t old_fs = get_fs();
2132     set_fs(KERNEL_DS);
2133     res = blkdev_ioctl(bdev, 0, cmd, arg);
2134     set_fs(old_fs);
2135     return res;
2136 }
2137 
2138 EXPORT_SYMBOL(ioctl_by_bdev);
2139 
2140 /**
2141  * lookup_bdev  - lookup a struct block_device by name
2142  * @pathname:   special file representing the block device
2143  *
2144  * Get a reference to the blockdevice at @pathname in the current
2145  * namespace if possible and return it.  Return ERR_PTR(error)
2146  * otherwise.
2147  */
2148 struct block_device *lookup_bdev(const char *pathname)
2149 {
2150     struct block_device *bdev;
2151     struct inode *inode;
2152     struct path path;
2153     int error;
2154 
2155     if (!pathname || !*pathname)
2156         return ERR_PTR(-EINVAL);
2157 
2158     error = kern_path(pathname, LOOKUP_FOLLOW, &path);
2159     if (error)
2160         return ERR_PTR(error);
2161 
2162     inode = d_backing_inode(path.dentry);
2163     error = -ENOTBLK;
2164     if (!S_ISBLK(inode->i_mode))
2165         goto fail;
2166     error = -EACCES;
2167     if (!may_open_dev(&path))
2168         goto fail;
2169     error = -ENOMEM;
2170     bdev = bd_acquire(inode);
2171     if (!bdev)
2172         goto fail;
2173 out:
2174     path_put(&path);
2175     return bdev;
2176 fail:
2177     bdev = ERR_PTR(error);
2178     goto out;
2179 }
2180 EXPORT_SYMBOL(lookup_bdev);
2181 
2182 int __invalidate_device(struct block_device *bdev, bool kill_dirty)
2183 {
2184     struct super_block *sb = get_super(bdev);
2185     int res = 0;
2186 
2187     if (sb) {
2188         /*
2189          * no need to lock the super, get_super holds the
2190          * read mutex so the filesystem cannot go away
2191          * under us (->put_super runs with the write lock
2192          * hold).
2193          */
2194         shrink_dcache_sb(sb);
2195         res = invalidate_inodes(sb, kill_dirty);
2196         drop_super(sb);
2197     }
2198     invalidate_bdev(bdev);
2199     return res;
2200 }
2201 EXPORT_SYMBOL(__invalidate_device);
2202 
2203 void iterate_bdevs(void (*func)(struct block_device *, void *), void *arg)
2204 {
2205     struct inode *inode, *old_inode = NULL;
2206 
2207     spin_lock(&blockdev_superblock->s_inode_list_lock);
2208     list_for_each_entry(inode, &blockdev_superblock->s_inodes, i_sb_list) {
2209         struct address_space *mapping = inode->i_mapping;
2210         struct block_device *bdev;
2211 
2212         spin_lock(&inode->i_lock);
2213         if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW) ||
2214             mapping->nrpages == 0) {
2215             spin_unlock(&inode->i_lock);
2216             continue;
2217         }
2218         __iget(inode);
2219         spin_unlock(&inode->i_lock);
2220         spin_unlock(&blockdev_superblock->s_inode_list_lock);
2221         /*
2222          * We hold a reference to 'inode' so it couldn't have been
2223          * removed from s_inodes list while we dropped the
2224          * s_inode_list_lock  We cannot iput the inode now as we can
2225          * be holding the last reference and we cannot iput it under
2226          * s_inode_list_lock. So we keep the reference and iput it
2227          * later.
2228          */
2229         iput(old_inode);
2230         old_inode = inode;
2231         bdev = I_BDEV(inode);
2232 
2233         mutex_lock(&bdev->bd_mutex);
2234         if (bdev->bd_openers)
2235             func(bdev, arg);
2236         mutex_unlock(&bdev->bd_mutex);
2237 
2238         spin_lock(&blockdev_superblock->s_inode_list_lock);
2239     }
2240     spin_unlock(&blockdev_superblock->s_inode_list_lock);
2241     iput(old_inode);
2242 }