Merge tag 'fiemap_for_v5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
authorLinus Torvalds <torvalds@linux-foundation.org>
Mon, 30 Aug 2021 17:13:02 +0000 (10:13 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Mon, 30 Aug 2021 17:13:02 +0000 (10:13 -0700)
Pull FIEMAP cleanups from Jan Kara:
 "FIEMAP cleanups from Christoph transitioning all remaining filesystems
  supporting FIEMAP (ext2, hpfs) to iomap API and removing the old
  helper"

* tag 'fiemap_for_v5.15-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/jack/linux-fs:
  fs: remove generic_block_fiemap
  hpfs: use iomap_fiemap to implement ->fiemap
  ext2: use iomap_fiemap to implement ->fiemap
  ext2: make ext2_iomap_ops available unconditionally

fs/ext2/Kconfig
fs/ext2/inode.c
fs/hpfs/Kconfig
fs/hpfs/file.c
fs/ioctl.c
include/linux/fiemap.h

index 54eec91..1248ff4 100644 (file)
@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: GPL-2.0-only
 config EXT2_FS
        tristate "Second extended fs support"
+       select FS_IOMAP
        help
          Ext2 is a standard Linux file system for hard disks.
 
index dadb121..04f0def 100644 (file)
@@ -799,7 +799,6 @@ int ext2_get_block(struct inode *inode, sector_t iblock,
 
 }
 
-#ifdef CONFIG_FS_DAX
 static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
                unsigned flags, struct iomap *iomap, struct iomap *srcmap)
 {
@@ -852,16 +851,18 @@ const struct iomap_ops ext2_iomap_ops = {
        .iomap_begin            = ext2_iomap_begin,
        .iomap_end              = ext2_iomap_end,
 };
-#else
-/* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
-const struct iomap_ops ext2_iomap_ops;
-#endif /* CONFIG_FS_DAX */
 
 int ext2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
                u64 start, u64 len)
 {
-       return generic_block_fiemap(inode, fieinfo, start, len,
-                                   ext2_get_block);
+       int ret;
+
+       inode_lock(inode);
+       len = min_t(u64, len, i_size_read(inode));
+       ret = iomap_fiemap(inode, fieinfo, start, len, &ext2_iomap_ops);
+       inode_unlock(inode);
+
+       return ret;
 }
 
 static int ext2_writepage(struct page *page, struct writeback_control *wbc)
index 2b36dc6..ec975f4 100644 (file)
@@ -2,6 +2,7 @@
 config HPFS_FS
        tristate "OS/2 HPFS file system support"
        depends on BLOCK
+       select FS_IOMAP
        help
          OS/2 is IBM's operating system for PC's, the same as Warp, and HPFS
          is the file system used for organizing files on OS/2 hard disk
index c3a49aa..fb37f57 100644 (file)
@@ -9,6 +9,7 @@
 
 #include "hpfs_fn.h"
 #include <linux/mpage.h>
+#include <linux/iomap.h>
 #include <linux/fiemap.h>
 
 #define BLOCKS(size) (((size) + 511) >> 9)
@@ -116,6 +117,47 @@ static int hpfs_get_block(struct inode *inode, sector_t iblock, struct buffer_he
        return r;
 }
 
+static int hpfs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
+               unsigned flags, struct iomap *iomap, struct iomap *srcmap)
+{
+       struct super_block *sb = inode->i_sb;
+       unsigned int blkbits = inode->i_blkbits;
+       unsigned int n_secs;
+       secno s;
+
+       if (WARN_ON_ONCE(flags & (IOMAP_WRITE | IOMAP_ZERO)))
+               return -EINVAL;
+
+       iomap->bdev = inode->i_sb->s_bdev;
+       iomap->offset = offset;
+
+       hpfs_lock(sb);
+       s = hpfs_bmap(inode, offset >> blkbits, &n_secs);
+       if (s) {
+               n_secs = hpfs_search_hotfix_map_for_range(sb, s,
+                               min_t(loff_t, n_secs, length));
+               if (unlikely(!n_secs)) {
+                       s = hpfs_search_hotfix_map(sb, s);
+                       n_secs = 1;
+               }
+               iomap->type = IOMAP_MAPPED;
+               iomap->flags = IOMAP_F_MERGED;
+               iomap->addr = (u64)s << blkbits;
+               iomap->length = (u64)n_secs << blkbits;
+       } else {
+               iomap->type = IOMAP_HOLE;
+               iomap->addr = IOMAP_NULL_ADDR;
+               iomap->length = 1 << blkbits;
+       }
+
+       hpfs_unlock(sb);
+       return 0;
+}
+
+static const struct iomap_ops hpfs_iomap_ops = {
+       .iomap_begin            = hpfs_iomap_begin,
+};
+
 static int hpfs_readpage(struct file *file, struct page *page)
 {
        return mpage_readpage(page, hpfs_get_block);
@@ -192,7 +234,14 @@ static sector_t _hpfs_bmap(struct address_space *mapping, sector_t block)
 
 static int hpfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo, u64 start, u64 len)
 {
-       return generic_block_fiemap(inode, fieinfo, start, len, hpfs_get_block);
+       int ret;
+
+       inode_lock(inode);
+       len = min_t(u64, len, i_size_read(inode));
+       ret = iomap_fiemap(inode, fieinfo, start, len, &hpfs_iomap_ops);
+       inode_unlock(inode);
+
+       return ret;
 }
 
 const struct address_space_operations hpfs_aops = {
index 1e2204f..eea8267 100644 (file)
@@ -263,209 +263,6 @@ static long ioctl_file_clone_range(struct file *file,
                                args.src_length, args.dest_offset);
 }
 
-#ifdef CONFIG_BLOCK
-
-static inline sector_t logical_to_blk(struct inode *inode, loff_t offset)
-{
-       return (offset >> inode->i_blkbits);
-}
-
-static inline loff_t blk_to_logical(struct inode *inode, sector_t blk)
-{
-       return (blk << inode->i_blkbits);
-}
-
-/**
- * __generic_block_fiemap - FIEMAP for block based inodes (no locking)
- * @inode: the inode to map
- * @fieinfo: the fiemap info struct that will be passed back to userspace
- * @start: where to start mapping in the inode
- * @len: how much space to map
- * @get_block: the fs's get_block function
- *
- * This does FIEMAP for block based inodes.  Basically it will just loop
- * through get_block until we hit the number of extents we want to map, or we
- * go past the end of the file and hit a hole.
- *
- * If it is possible to have data blocks beyond a hole past @inode->i_size, then
- * please do not use this function, it will stop at the first unmapped block
- * beyond i_size.
- *
- * If you use this function directly, you need to do your own locking. Use
- * generic_block_fiemap if you want the locking done for you.
- */
-static int __generic_block_fiemap(struct inode *inode,
-                          struct fiemap_extent_info *fieinfo, loff_t start,
-                          loff_t len, get_block_t *get_block)
-{
-       struct buffer_head map_bh;
-       sector_t start_blk, last_blk;
-       loff_t isize = i_size_read(inode);
-       u64 logical = 0, phys = 0, size = 0;
-       u32 flags = FIEMAP_EXTENT_MERGED;
-       bool past_eof = false, whole_file = false;
-       int ret = 0;
-
-       ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_SYNC);
-       if (ret)
-               return ret;
-
-       /*
-        * Either the i_mutex or other appropriate locking needs to be held
-        * since we expect isize to not change at all through the duration of
-        * this call.
-        */
-       if (len >= isize) {
-               whole_file = true;
-               len = isize;
-       }
-
-       /*
-        * Some filesystems can't deal with being asked to map less than
-        * blocksize, so make sure our len is at least block length.
-        */
-       if (logical_to_blk(inode, len) == 0)
-               len = blk_to_logical(inode, 1);
-
-       start_blk = logical_to_blk(inode, start);
-       last_blk = logical_to_blk(inode, start + len - 1);
-
-       do {
-               /*
-                * we set b_size to the total size we want so it will map as
-                * many contiguous blocks as possible at once
-                */
-               memset(&map_bh, 0, sizeof(struct buffer_head));
-               map_bh.b_size = len;
-
-               ret = get_block(inode, start_blk, &map_bh, 0);
-               if (ret)
-                       break;
-
-               /* HOLE */
-               if (!buffer_mapped(&map_bh)) {
-                       start_blk++;
-
-                       /*
-                        * We want to handle the case where there is an
-                        * allocated block at the front of the file, and then
-                        * nothing but holes up to the end of the file properly,
-                        * to make sure that extent at the front gets properly
-                        * marked with FIEMAP_EXTENT_LAST
-                        */
-                       if (!past_eof &&
-                           blk_to_logical(inode, start_blk) >= isize)
-                               past_eof = 1;
-
-                       /*
-                        * First hole after going past the EOF, this is our
-                        * last extent
-                        */
-                       if (past_eof && size) {
-                               flags = FIEMAP_EXTENT_MERGED|FIEMAP_EXTENT_LAST;
-                               ret = fiemap_fill_next_extent(fieinfo, logical,
-                                                             phys, size,
-                                                             flags);
-                       } else if (size) {
-                               ret = fiemap_fill_next_extent(fieinfo, logical,
-                                                             phys, size, flags);
-                               size = 0;
-                       }
-
-                       /* if we have holes up to/past EOF then we're done */
-                       if (start_blk > last_blk || past_eof || ret)
-                               break;
-               } else {
-                       /*
-                        * We have gone over the length of what we wanted to
-                        * map, and it wasn't the entire file, so add the extent
-                        * we got last time and exit.
-                        *
-                        * This is for the case where say we want to map all the
-                        * way up to the second to the last block in a file, but
-                        * the last block is a hole, making the second to last
-                        * block FIEMAP_EXTENT_LAST.  In this case we want to
-                        * see if there is a hole after the second to last block
-                        * so we can mark it properly.  If we found data after
-                        * we exceeded the length we were requesting, then we
-                        * are good to go, just add the extent to the fieinfo
-                        * and break
-                        */
-                       if (start_blk > last_blk && !whole_file) {
-                               ret = fiemap_fill_next_extent(fieinfo, logical,
-                                                             phys, size,
-                                                             flags);
-                               break;
-                       }
-
-                       /*
-                        * if size != 0 then we know we already have an extent
-                        * to add, so add it.
-                        */
-                       if (size) {
-                               ret = fiemap_fill_next_extent(fieinfo, logical,
-                                                             phys, size,
-                                                             flags);
-                               if (ret)
-                                       break;
-                       }
-
-                       logical = blk_to_logical(inode, start_blk);
-                       phys = blk_to_logical(inode, map_bh.b_blocknr);
-                       size = map_bh.b_size;
-                       flags = FIEMAP_EXTENT_MERGED;
-
-                       start_blk += logical_to_blk(inode, size);
-
-                       /*
-                        * If we are past the EOF, then we need to make sure as
-                        * soon as we find a hole that the last extent we found
-                        * is marked with FIEMAP_EXTENT_LAST
-                        */
-                       if (!past_eof && logical + size >= isize)
-                               past_eof = true;
-               }
-               cond_resched();
-               if (fatal_signal_pending(current)) {
-                       ret = -EINTR;
-                       break;
-               }
-
-       } while (1);
-
-       /* If ret is 1 then we just hit the end of the extent array */
-       if (ret == 1)
-               ret = 0;
-
-       return ret;
-}
-
-/**
- * generic_block_fiemap - FIEMAP for block based inodes
- * @inode: The inode to map
- * @fieinfo: The mapping information
- * @start: The initial block to map
- * @len: The length of the extect to attempt to map
- * @get_block: The block mapping function for the fs
- *
- * Calls __generic_block_fiemap to map the inode, after taking
- * the inode's mutex lock.
- */
-
-int generic_block_fiemap(struct inode *inode,
-                        struct fiemap_extent_info *fieinfo, u64 start,
-                        u64 len, get_block_t *get_block)
-{
-       int ret;
-       inode_lock(inode);
-       ret = __generic_block_fiemap(inode, fieinfo, start, len, get_block);
-       inode_unlock(inode);
-       return ret;
-}
-EXPORT_SYMBOL(generic_block_fiemap);
-
-#endif  /*  CONFIG_BLOCK  */
-
 /*
  * This provides compatibility with legacy XFS pre-allocation ioctls
  * which predate the fallocate syscall.
index 4e624c4..c50882f 100644 (file)
@@ -18,8 +18,4 @@ int fiemap_prep(struct inode *inode, struct fiemap_extent_info *fieinfo,
 int fiemap_fill_next_extent(struct fiemap_extent_info *info, u64 logical,
                            u64 phys, u64 len, u32 flags);
 
-int generic_block_fiemap(struct inode *inode,
-               struct fiemap_extent_info *fieinfo, u64 start, u64 len,
-               get_block_t *get_block);
-
 #endif /* _LINUX_FIEMAP_H 1 */