Merge tag 'xfs-4.19-merge-6' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux

Pull xfs updates from Darrick Wong:
 "This is the second part of the XFS changes for 4.19.

  The biggest changes are the removal of buffer heads frm XFS, a massive
  reworking of the deferred transaction operations handling code, the
  removal of the long defunct barrier/nobarrier mount options, and the
  addition of a few more online repair functions.

  Summary:

   - Use extent maps to track pagecache page status instead of
     bufferhead state.

   - Refactor pagecache read and write paths to use the new iomap
     library functions, which enable us to drop the old bufferhead code
     for pagesize == blocksize filesystems.

   - Set up parallel per-block-per-page metadata to track subpage
     information that was tracked by buffer heads, which enables us to
     drop the old bufferhead code for pagesize > blocksize filesystems.

   - Tie a deferred ops control structure to a transaction so that we
     can take advantage of an upper-level dfops without having to plumb
     pointer passing through the code.

   - Refactor the deferred ops code to track deferred ops as part of the
     transaction structure (instead of as a separate data structure) so
     that we can simplify the scoping rules around defer_ops.

   - Refactor twisty delwri buffer submission code to avoid deadlocks.

   - Shorten and fix indenting problems in the scrub code.

   - Detect obviously bad summary counts at mount and fix them.

   - Directly associate deferred ops control structure with a
     transaction so that callers no longer have to manage it themselves.

   - Remove a couple of IRIX-era inode macros.

   - Remove the long-deprecated 'barrier' and 'nobarrier' mount options.

   - Clean up the inode fork structure a bit.

   - Check for bad fs summary counter values in the superblock.

   - Reduce COW fork lookups during writeback.

   - Refactor the deferred ops control structures into the transaction
     structure, thereby eliminating the need for transaction users to
     handle the deferred ops as a separate data structure.

   - Add the ability to repair AG headers online.

   - Fix a crash due to insufficient return value checking.

   - Various fixes and cleanups"

* tag 'xfs-4.19-merge-6' of git://git.kernel.org/pub/scm/fs/xfs/xfs-linux: (155 commits)
  xfs: fix a null pointer dereference in xfs_bmap_extents_to_btree
  xfs: remove b_last_holder & associated macros
  iomap: Switch to offset_in_page for clarity
  xfs: Close race between direct IO and xfs_break_layouts()
  xfs: repair the AGI
  xfs: repair the AGFL
  xfs: repair the AGF
  xfs: remove dead error handling code in xfs_dquot_disk_alloc()
  xfs: use WRITE_ONCE to update if_seq
  xfs: fix a comment in xfs_log_reserve
  xfs: only validate summary counts on primary superblock
  xfs: substitute spaces with tabs
  xfs: fold dfops into the transaction
  xfs: always defer agfl block frees
  xfs: pass transaction to xfs_defer_add()
  xfs: replace xfs_defer_ops ->dop_pending with on-stack list
  xfs: cancel dfops on xfs_defer_finish() error
  xfs: clean out superfluous dfops dop params/vars
  xfs: drop dop param from xfs_defer_op_type ->finish_item() callback
  xfs: automatic dfops inode relogging
  ...

1 files changed, 72 insertions, 28 deletions

diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index a3378252baa1..99db27d6ac8a 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -207,6 +207,9 @@ xfs_initialize_perag(
 		if (xfs_buf_hash_init(pag))
 			goto out_free_pag;
 		init_waitqueue_head(&pag->pagb_wait);
+		spin_lock_init(&pag->pagb_lock);
+		pag->pagb_count = 0;
+		pag->pagb_tree = RB_ROOT;
 
 		if (radix_tree_preload(GFP_NOFS))
 			goto out_hash_destroy;
@@ -606,6 +609,56 @@ xfs_default_resblks(xfs_mount_t *mp)
 	return resblks;
 }
 
+/* Ensure the summary counts are correct. */
+STATIC int
+xfs_check_summary_counts(
+	struct xfs_mount	*mp)
+{
+	/*
+	 * The AG0 superblock verifier rejects in-progress filesystems,
+	 * so we should never see the flag set this far into mounting.
+	 */
+	if (mp->m_sb.sb_inprogress) {
+		xfs_err(mp, "sb_inprogress set after log recovery??");
+		WARN_ON(1);
+		return -EFSCORRUPTED;
+	}
+
+	/*
+	 * Now the log is mounted, we know if it was an unclean shutdown or
+	 * not. If it was, with the first phase of recovery has completed, we
+	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
+	 * but they are recovered transactionally in the second recovery phase
+	 * later.
+	 *
+	 * If the log was clean when we mounted, we can check the summary
+	 * counters.  If any of them are obviously incorrect, we can recompute
+	 * them from the AGF headers in the next step.
+	 */
+	if (XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
+	    (mp->m_sb.sb_fdblocks > mp->m_sb.sb_dblocks ||
+	     mp->m_sb.sb_ifree > mp->m_sb.sb_icount))
+		mp->m_flags |= XFS_MOUNT_BAD_SUMMARY;
+
+	/*
+	 * We can safely re-initialise incore superblock counters from the
+	 * per-ag data. These may not be correct if the filesystem was not
+	 * cleanly unmounted, so we waited for recovery to finish before doing
+	 * this.
+	 *
+	 * If the filesystem was cleanly unmounted or the previous check did
+	 * not flag anything weird, then we can trust the values in the
+	 * superblock to be correct and we don't need to do anything here.
+	 * Otherwise, recalculate the summary counters.
+	 */
+	if ((!xfs_sb_version_haslazysbcount(&mp->m_sb) ||
+	     XFS_LAST_UNMOUNT_WAS_CLEAN(mp)) &&
+	    !(mp->m_flags & XFS_MOUNT_BAD_SUMMARY))
+		return 0;
+
+	return xfs_initialize_perag_data(mp, mp->m_sb.sb_agcount);
+}
+
 /*
  * This function does the following on an initial mount of a file system:
  *	- reads the superblock from disk and init the mount struct
@@ -831,32 +884,10 @@ xfs_mountfs(
 		goto out_fail_wait;
 	}
 
-	/*
-	 * Now the log is mounted, we know if it was an unclean shutdown or
-	 * not. If it was, with the first phase of recovery has completed, we
-	 * have consistent AG blocks on disk. We have not recovered EFIs yet,
-	 * but they are recovered transactionally in the second recovery phase
-	 * later.
-	 *
-	 * Hence we can safely re-initialise incore superblock counters from
-	 * the per-ag data. These may not be correct if the filesystem was not
-	 * cleanly unmounted, so we need to wait for recovery to finish before
-	 * doing this.
-	 *
-	 * If the filesystem was cleanly unmounted, then we can trust the
-	 * values in the superblock to be correct and we don't need to do
-	 * anything here.
-	 *
-	 * If we are currently making the filesystem, the initialisation will
-	 * fail as the perag data is in an undefined state.
-	 */
-	if (xfs_sb_version_haslazysbcount(&mp->m_sb) &&
-	    !XFS_LAST_UNMOUNT_WAS_CLEAN(mp) &&
-	     !mp->m_sb.sb_inprogress) {
-		error = xfs_initialize_perag_data(mp, sbp->sb_agcount);
-		if (error)
-			goto out_log_dealloc;
-	}
+	/* Make sure the summary counts are ok. */
+	error = xfs_check_summary_counts(mp);
+	if (error)
+		goto out_log_dealloc;
 
 	/*
 	 * Get and sanity-check the root inode.
@@ -1011,7 +1042,7 @@ xfs_mountfs(
  out_rtunmount:
 	xfs_rtunmount_inodes(mp);
  out_rele_rip:
-	IRELE(rip);
+	xfs_irele(rip);
 	/* Clean out dquots that might be in memory after quotacheck. */
 	xfs_qm_unmount(mp);
 	/*
@@ -1067,7 +1098,7 @@ xfs_unmountfs(
 	xfs_fs_unreserve_ag_blocks(mp);
 	xfs_qm_unmount_quotas(mp);
 	xfs_rtunmount_inodes(mp);
-	IRELE(mp->m_rootip);
+	xfs_irele(mp->m_rootip);
 
 	/*
 	 * We can potentially deadlock here if we have an inode cluster
@@ -1395,3 +1426,16 @@ xfs_dev_is_read_only(
 	}
 	return 0;
 }
+
+/* Force the summary counters to be recalculated at next mount. */
+void
+xfs_force_summary_recalc(
+	struct xfs_mount	*mp)
+{
+	if (!xfs_sb_version_haslazysbcount(&mp->m_sb))
+		return;
+
+	spin_lock(&mp->m_sb_lock);
+	mp->m_flags |= XFS_MOUNT_BAD_SUMMARY;
+	spin_unlock(&mp->m_sb_lock);
+}