Merge tag 'xfs-5.13-merge-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux

Pull xfs updates from Darrick Wong:
 "The notable user-visible addition this cycle is ability to remove
  space from the last AG in a filesystem. This is the first of many
  changes needed for full-fledged support for shrinking a filesystem.
  Still needed are (a) the ability to reorganize files and metadata away
  from the end of the fs; (b) the ability to remove entire allocation
  groups; (c) shrink support for realtime volumes; and (d) thorough
  testing of (a-c).

  There are a number of performance improvements in this code drop: Dave
  streamlined various parts of the buffer logging code and reduced the
  cost of various debugging checks, and added the ability to pre-create
  the xattr structures while creating files. Brian eliminated
  transaction reservations that were being held across writeback (thus
  reducing livelock potential.

  Other random pieces: Pavel fixed the repetitve warnings about
  deprecated mount options, I fixed online fsck to behave itself when a
  readonly remount comes in during scrub, and refactored various other
  parts of that code, Christoph contributed a lot of refactoring this
  cycle. The xfs_icdinode structure has been absorbed into the (incore)
  xfs_inode structure, and the format and flags handling around
  xfs_inode_fork structures has been simplified. Chandan provided a
  number of fixes for extent count overflow related problems that have
  been shaken out by debugging knobs added during 5.12.

  Summary:

   - Various minor fixes in online scrub.

   - Prevent metadata files from being automatically inactivated.

   - Validate btree heights by the computed per-btree limits.

   - Don't warn about remounting with deprecated mount options.

   - Initialize attr forks at create time if we suspect we're going to
     need to store them.

   - Reduce memory reallocation workouts in the logging code.

   - Fix some theoretical math calculation errors in logged buffers that
     span multiple discontig memory ranges but contiguous ondisk
     regions.

   - Speedups in dirty buffer bitmap handling.

   - Make type verifier functions more inline-happy to reduce overhead.

   - Reduce debug overhead in directory checking code.

   - Many many typo fixes.

   - Begin to handle the permanent loss of the very end of a filesystem.

   - Fold struct xfs_icdinode into xfs_inode.

   - Deprecate the long defunct BMV_IF_NO_DMAPI_READ from the bmapx
     ioctl.

   - Remove a broken directory block format check from online scrub.

   - Fix a bug where we could produce an unnecessarily tall data fork
     btree when creating an attr fork.

   - Fix scrub and readonly remounts racing.

   - Fix a writeback ioend log deadlock problem by dropping the behavior
     where we could preallocate a setfilesize transaction.

   - Fix some bugs in the new extent count checking code.

   - Fix some bugs in the attr fork preallocation code.

   - Refactor if_flags out of the incore inode fork data structure"

* tag 'xfs-5.13-merge-3' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: (77 commits)
  xfs: remove xfs_quiesce_attr declaration
  xfs: remove XFS_IFEXTENTS
  xfs: remove XFS_IFINLINE
  xfs: remove XFS_IFBROOT
  xfs: only look at the fork format in xfs_idestroy_fork
  xfs: simplify xfs_attr_remove_args
  xfs: rename and simplify xfs_bmap_one_block
  xfs: move the XFS_IFEXTENTS check into xfs_iread_extents
  xfs: drop unnecessary setfilesize helper
  xfs: drop unused ioend private merge and setfilesize code
  xfs: open code ioend needs workqueue helper
  xfs: drop submit side trans alloc for append ioends
  xfs: fix return of uninitialized value in variable error
  xfs: get rid of the ip parameter to xchk_setup_*
  xfs: fix scrub and remount-ro protection when running scrub
  xfs: move the check for post-EOF mappings into xfs_can_free_eofblocks
  xfs: move the xfs_can_free_eofblocks call under the IOLOCK
  xfs: precalculate default inode attribute offset
  xfs: default attr fork size does not handle device inodes
  xfs: inode fork allocation depends on XFS_IFEXTENT flag
  ...
This commit is contained in:
Linus Torvalds
2021-04-29 10:43:51 -07:00
92 changed files with 1482 additions and 1326 deletions

View File

@ -33,7 +33,7 @@ xfs_readlink_bmap_ilocked(
struct xfs_buf *bp;
xfs_daddr_t d;
char *cur_chunk;
int pathlen = ip->i_d.di_size;
int pathlen = ip->i_disk_size;
int nmaps = XFS_SYMLINK_MAPS;
int byte_cnt;
int n;
@ -86,7 +86,7 @@ xfs_readlink_bmap_ilocked(
}
ASSERT(pathlen == 0);
link[ip->i_d.di_size] = '\0';
link[ip->i_disk_size] = '\0';
error = 0;
out:
@ -104,14 +104,14 @@ xfs_readlink(
trace_xfs_readlink(ip);
ASSERT(!(ip->i_df.if_flags & XFS_IFINLINE));
ASSERT(ip->i_df.if_format != XFS_DINODE_FMT_LOCAL);
if (XFS_FORCED_SHUTDOWN(mp))
return -EIO;
xfs_ilock(ip, XFS_ILOCK_SHARED);
pathlen = ip->i_d.di_size;
pathlen = ip->i_disk_size;
if (!pathlen)
goto out;
@ -210,7 +210,7 @@ xfs_symlink(
/*
* Check whether the directory allows new symlinks or not.
*/
if (dp->i_d.di_flags & XFS_DIFLAG_NOSYMLINKS) {
if (dp->i_diflags & XFS_DIFLAG_NOSYMLINKS) {
error = -EPERM;
goto out_trans_cancel;
}
@ -224,7 +224,7 @@ xfs_symlink(
* Allocate an inode for the symlink.
*/
error = xfs_dir_ialloc(mnt_userns, &tp, dp, S_IFLNK | (mode & ~S_IFMT),
1, 0, prid, &ip);
1, 0, prid, false, &ip);
if (error)
goto out_trans_cancel;
@ -250,7 +250,7 @@ xfs_symlink(
if (pathlen <= XFS_IFORK_DSIZE(ip)) {
xfs_init_local_fork(ip, XFS_DATA_FORK, target_path, pathlen);
ip->i_d.di_size = pathlen;
ip->i_disk_size = pathlen;
ip->i_df.if_format = XFS_DINODE_FMT_LOCAL;
xfs_trans_log_inode(tp, ip, XFS_ILOG_DDATA | XFS_ILOG_CORE);
} else {
@ -265,7 +265,7 @@ xfs_symlink(
goto out_trans_cancel;
resblks -= fs_blocks;
ip->i_d.di_size = pathlen;
ip->i_disk_size = pathlen;
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
cur_chunk = target_path;
@ -300,7 +300,7 @@ xfs_symlink(
}
ASSERT(pathlen == 0);
}
i_size_write(VFS_I(ip), ip->i_d.di_size);
i_size_write(VFS_I(ip), ip->i_disk_size);
/*
* Create the directory entry for the symlink.
@ -377,7 +377,7 @@ xfs_inactive_symlink_rmt(
xfs_trans_t *tp;
mp = ip->i_mount;
ASSERT(ip->i_df.if_flags & XFS_IFEXTENTS);
ASSERT(!xfs_need_iread_extents(&ip->i_df));
/*
* We're freeing a symlink that has some
* blocks allocated to it. Free the
@ -400,8 +400,8 @@ xfs_inactive_symlink_rmt(
* locked for the second transaction. In the error paths we need it
* held so the cancel won't rele it, see below.
*/
size = (int)ip->i_d.di_size;
ip->i_d.di_size = 0;
size = (int)ip->i_disk_size;
ip->i_disk_size = 0;
VFS_I(ip)->i_mode = (VFS_I(ip)->i_mode & ~S_IFMT) | S_IFREG;
xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
/*
@ -477,7 +477,7 @@ xfs_inactive_symlink(
return -EIO;
xfs_ilock(ip, XFS_ILOCK_EXCL);
pathlen = (int)ip->i_d.di_size;
pathlen = (int)ip->i_disk_size;
ASSERT(pathlen);
if (pathlen <= 0 || pathlen > XFS_SYMLINK_MAXLEN) {
@ -492,7 +492,7 @@ xfs_inactive_symlink(
* Inline fork state gets removed by xfs_difree() so we have nothing to
* do here in that case.
*/
if (ip->i_df.if_flags & XFS_IFINLINE) {
if (ip->i_df.if_format == XFS_DINODE_FMT_LOCAL) {
xfs_iunlock(ip, XFS_ILOCK_EXCL);
return 0;
}