Merge tag 'xfs-for-linus-3.15-rc2' of git://oss.sgi.com/xfs/xfs

Pull xfs bug fixes from Dave Chinner:
 "The fixes are for data corruption issues, memory corruption and
  regressions for changes merged in -rc1.

  Data corruption fixes:
   - fix a bunch of delayed allocation state mismatches
   - fix collapse/zero range bugs
   - fix a direct IO block mapping bug @ EOF

  Other fixes:
   - fix a use after free on metadata IO error
   - fix a use after free on IO error during unmount
   - fix an incorrect error sign on direct IO write errors
   - add missing O_TMPFILE inode security context initialisation"

* tag 'xfs-for-linus-3.15-rc2' of git://oss.sgi.com/xfs/xfs:
  xfs: fix tmpfile/selinux deadlock and initialize security
  xfs: fix buffer use after free on IO error
  xfs: wrong error sign conversion during failed DIO writes
  xfs: unmount does not wait for shutdown during unmount
  xfs: collapse range is delalloc challenged
  xfs: don't map ranges that span EOF for direct IO
  xfs: zeroing space needs to punch delalloc blocks
  xfs: xfs_vm_write_end truncates too much on failure
  xfs: write failure beyond EOF truncates too much data
  xfs: kill buffers over failed write ranges properly

fs/xfs/xfs_aops.c

@@ -1344,6 +1344,14 @@ __xfs_get_blocks(
 	/*
 	 * If this is O_DIRECT or the mpage code calling tell them how large
 	 * the mapping is, so that we can avoid repeated get_blocks calls.
+	 *
+	 * If the mapping spans EOF, then we have to break the mapping up as the
+	 * mapping for blocks beyond EOF must be marked new so that sub block
+	 * regions can be correctly zeroed. We can't do this for mappings within
+	 * EOF unless the mapping was just allocated or is unwritten, otherwise
+	 * the callers would overwrite existing data with zeros. Hence we have
+	 * to split the mapping into a range up to and including EOF, and a
+	 * second mapping for beyond EOF.
 	 */
 	if (direct || size > (1 << inode->i_blkbits)) {
 		xfs_off_t		mapping_size;
@@ -1354,6 +1362,12 @@ __xfs_get_blocks(
 		ASSERT(mapping_size > 0);
 		if (mapping_size > size)
 			mapping_size = size;
+		if (offset < i_size_read(inode) &&
+		    offset + mapping_size >= i_size_read(inode)) {
+			/* limit mapping to block that spans EOF */
+			mapping_size = roundup_64(i_size_read(inode) - offset,
+						  1 << inode->i_blkbits);
+		}
 		if (mapping_size > LONG_MAX)
 			mapping_size = LONG_MAX;
 
@@ -1566,6 +1580,16 @@ xfs_vm_write_failed(
 
 		xfs_vm_kill_delalloc_range(inode, block_offset,
 					   block_offset + bh->b_size);
+
+		/*
+		 * This buffer does not contain data anymore. make sure anyone
+		 * who finds it knows that for certain.
+		 */
+		clear_buffer_delay(bh);
+		clear_buffer_uptodate(bh);
+		clear_buffer_mapped(bh);
+		clear_buffer_new(bh);
+		clear_buffer_dirty(bh);
 	}
 
 }
@@ -1599,12 +1623,21 @@ xfs_vm_write_begin(
 	status = __block_write_begin(page, pos, len, xfs_get_blocks);
 	if (unlikely(status)) {
 		struct inode	*inode = mapping->host;
+		size_t		isize = i_size_read(inode);
 
 		xfs_vm_write_failed(inode, page, pos, len);
 		unlock_page(page);
 
-		if (pos + len > i_size_read(inode))
-			truncate_pagecache(inode, i_size_read(inode));
+		/*
+		 * If the write is beyond EOF, we only want to kill blocks
+		 * allocated in this write, not blocks that were previously
+		 * written successfully.
+		 */
+		if (pos + len > isize) {
+			ssize_t start = max_t(ssize_t, pos, isize);
+
+			truncate_pagecache_range(inode, start, pos + len);
+		}
 
 		page_cache_release(page);
 		page = NULL;
@@ -1615,9 +1648,12 @@ xfs_vm_write_begin(
 }
 
 /*
- * On failure, we only need to kill delalloc blocks beyond EOF because they
- * will never be written. For blocks within EOF, generic_write_end() zeros them
- * so they are safe to leave alone and be written with all the other valid data.
+ * On failure, we only need to kill delalloc blocks beyond EOF in the range of
+ * this specific write because they will never be written. Previous writes
+ * beyond EOF where block allocation succeeded do not need to be trashed, so
+ * only new blocks from this write should be trashed. For blocks within
+ * EOF, generic_write_end() zeros them so they are safe to leave alone and be
+ * written with all the other valid data.
  */
 STATIC int
 xfs_vm_write_end(
@@ -1640,8 +1676,11 @@ xfs_vm_write_end(
 		loff_t		to = pos + len;
 
 		if (to > isize) {
-			truncate_pagecache(inode, isize);
+			/* only kill blocks in this write beyond EOF */
+			if (pos > isize)
+				isize = pos;
 			xfs_vm_kill_delalloc_range(inode, isize, to);
+			truncate_pagecache_range(inode, isize, to);
 		}
 	}
 	return ret;

fs/xfs/xfs_bmap.c

@@ -5413,6 +5413,7 @@ xfs_bmap_shift_extents(
 	int				whichfork = XFS_DATA_FORK;
 	int				logflags;
 	xfs_filblks_t			blockcount = 0;
+	int				total_extents;
 
 	if (unlikely(XFS_TEST_ERROR(
 	    (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
@@ -5429,7 +5430,6 @@ xfs_bmap_shift_extents(
 	ASSERT(current_ext != NULL);
 
 	ifp = XFS_IFORK_PTR(ip, whichfork);
-
 	if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 		/* Read in all the extents */
 		error = xfs_iread_extents(tp, ip, whichfork);
@@ -5456,7 +5456,6 @@ xfs_bmap_shift_extents(
 
 	/* We are going to change core inode */
 	logflags = XFS_ILOG_CORE;
-
 	if (ifp->if_flags & XFS_IFBROOT) {
 		cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
 		cur->bc_private.b.firstblock = *firstblock;
@@ -5467,8 +5466,14 @@ xfs_bmap_shift_extents(
 		logflags |= XFS_ILOG_DEXT;
 	}
 
-	while (nexts++ < num_exts &&
-	       *current_ext <  XFS_IFORK_NEXTENTS(ip, whichfork)) {
+	/*
+	 * There may be delalloc extents in the data fork before the range we
+	 * are collapsing out, so we cannot
+	 * use the count of real extents here. Instead we have to calculate it
+	 * from the incore fork.
+	 */
+	total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
+	while (nexts++ < num_exts && *current_ext < total_extents) {
 
 		gotp = xfs_iext_get_ext(ifp, *current_ext);
 		xfs_bmbt_get_all(gotp, &got);
@@ -5556,10 +5561,11 @@ xfs_bmap_shift_extents(
 		}
 
 		(*current_ext)++;
+		total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
 	}
 
 	/* Check if we are done */
-	if (*current_ext ==  XFS_IFORK_NEXTENTS(ip, whichfork))
+	if (*current_ext == total_extents)
 		*done = 1;
 
 del_cursor:
@@ -5568,6 +5574,5 @@ xfs_bmap_shift_extents(
 			error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 
 	xfs_trans_log_inode(tp, ip, logflags);
-
 	return error;
 }

fs/xfs/xfs_bmap_util.c

@@ -1418,6 +1418,8 @@ xfs_zero_file_space(
 	xfs_off_t		end_boundary;
 	int			error;
 
+	trace_xfs_zero_file_space(ip);
+
 	granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
 
 	/*
@@ -1432,9 +1434,18 @@ xfs_zero_file_space(
 	ASSERT(end_boundary <= offset + len);
 
 	if (start_boundary < end_boundary - 1) {
-		/* punch out the page cache over the conversion range */
+		/*
+		 * punch out delayed allocation blocks and the page cache over
+		 * the conversion range
+		 */
+		xfs_ilock(ip, XFS_ILOCK_EXCL);
+		error = xfs_bmap_punch_delalloc_range(ip,
+				XFS_B_TO_FSBT(mp, start_boundary),
+				XFS_B_TO_FSB(mp, end_boundary - start_boundary));
+		xfs_iunlock(ip, XFS_ILOCK_EXCL);
 		truncate_pagecache_range(VFS_I(ip), start_boundary,
 					 end_boundary - 1);
+
 		/* convert the blocks */
 		error = xfs_alloc_file_space(ip, start_boundary,
 					end_boundary - start_boundary - 1,

fs/xfs/xfs_buf.c

@@ -1372,21 +1372,29 @@ xfs_buf_iorequest(
 		xfs_buf_wait_unpin(bp);
 	xfs_buf_hold(bp);
 
-	/* Set the count to 1 initially, this will stop an I/O
+	/*
+	 * Set the count to 1 initially, this will stop an I/O
 	 * completion callout which happens before we have started
 	 * all the I/O from calling xfs_buf_ioend too early.
 	 */
 	atomic_set(&bp->b_io_remaining, 1);
 	_xfs_buf_ioapply(bp);
-	_xfs_buf_ioend(bp, 1);
+	/*
+	 * If _xfs_buf_ioapply failed, we'll get back here with
+	 * only the reference we took above.  _xfs_buf_ioend will
+	 * drop it to zero, so we'd better not queue it for later,
+	 * or we'll free it before it's done.
+	 */
+	_xfs_buf_ioend(bp, bp->b_error ? 0 : 1);
 
 	xfs_buf_rele(bp);
 }
 
 /*
  * Waits for I/O to complete on the buffer supplied.  It returns immediately if
- * no I/O is pending or there is already a pending error on the buffer.  It
- * returns the I/O error code, if any, or 0 if there was no error.
+ * no I/O is pending or there is already a pending error on the buffer, in which
+ * case nothing will ever complete.  It returns the I/O error code, if any, or
+ * 0 if there was no error.
  */
 int
 xfs_buf_iowait(

fs/xfs/xfs_file.c

@@ -679,7 +679,7 @@ xfs_file_dio_aio_write(
 		goto out;
 
 	if (mapping->nrpages) {
-		ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
+		ret = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
 						    pos, -1);
 		if (ret)
 			goto out;

fs/xfs/xfs_inode.c

@@ -1334,7 +1334,8 @@ int
 xfs_create_tmpfile(
 	struct xfs_inode	*dp,
 	struct dentry		*dentry,
-	umode_t			mode)
+	umode_t			mode,
+	struct xfs_inode	**ipp)
 {
 	struct xfs_mount	*mp = dp->i_mount;
 	struct xfs_inode	*ip = NULL;
@@ -1402,7 +1403,6 @@ xfs_create_tmpfile(
 	xfs_qm_vop_create_dqattach(tp, ip, udqp, gdqp, pdqp);
 
 	ip->i_d.di_nlink--;
-	d_tmpfile(dentry, VFS_I(ip));
 	error = xfs_iunlink(tp, ip);
 	if (error)
 		goto out_trans_abort;
@@ -1415,6 +1415,7 @@ xfs_create_tmpfile(
 	xfs_qm_dqrele(gdqp);
 	xfs_qm_dqrele(pdqp);
 
+	*ipp = ip;
 	return 0;
 
  out_trans_abort:

fs/xfs/xfs_inode.h

@@ -334,7 +334,7 @@ int		xfs_lookup(struct xfs_inode *dp, struct xfs_name *name,
 int		xfs_create(struct xfs_inode *dp, struct xfs_name *name,
 			   umode_t mode, xfs_dev_t rdev, struct xfs_inode **ipp);
 int		xfs_create_tmpfile(struct xfs_inode *dp, struct dentry *dentry,
-			   umode_t mode);
+			   umode_t mode, struct xfs_inode **ipp);
 int		xfs_remove(struct xfs_inode *dp, struct xfs_name *name,
 			   struct xfs_inode *ip);
 int		xfs_link(struct xfs_inode *tdp, struct xfs_inode *sip,

fs/xfs/xfs_iops.c

@@ -1054,10 +1054,24 @@ xfs_vn_tmpfile(
 	umode_t		mode)
 {
 	int			error;
+	struct xfs_inode	*ip;
+	struct inode		*inode;
 
-	error = xfs_create_tmpfile(XFS_I(dir), dentry, mode);
+	error = xfs_create_tmpfile(XFS_I(dir), dentry, mode, &ip);
+	if (unlikely(error))
+		return -error;
 
+	inode = VFS_I(ip);
+
+	error = xfs_init_security(inode, dir, &dentry->d_name);
+	if (unlikely(error)) {
+		iput(inode);
 		return -error;
+	}
+
+	d_tmpfile(dentry, inode);
+
+	return 0;
 }
 
 static const struct inode_operations xfs_inode_operations = {

fs/xfs/xfs_log.c

@@ -1181,11 +1181,14 @@ xlog_iodone(xfs_buf_t *bp)
 	/* log I/O is always issued ASYNC */
 	ASSERT(XFS_BUF_ISASYNC(bp));
 	xlog_state_done_syncing(iclog, aborted);
+
 	/*
-	 * do not reference the buffer (bp) here as we could race
-	 * with it being freed after writing the unmount record to the
-	 * log.
+	 * drop the buffer lock now that we are done. Nothing references
+	 * the buffer after this, so an unmount waiting on this lock can now
+	 * tear it down safely. As such, it is unsafe to reference the buffer
+	 * (bp) after the unlock as we could race with it being freed.
 	 */
+	xfs_buf_unlock(bp);
 }
 
 /*
@@ -1368,8 +1371,16 @@ xlog_alloc_log(
 	bp = xfs_buf_alloc(mp->m_logdev_targp, 0, BTOBB(log->l_iclog_size), 0);
 	if (!bp)
 		goto out_free_log;
-	bp->b_iodone = xlog_iodone;
+
+	/*
+	 * The iclogbuf buffer locks are held over IO but we are not going to do
+	 * IO yet.  Hence unlock the buffer so that the log IO path can grab it
+	 * when appropriately.
+	 */
 	ASSERT(xfs_buf_islocked(bp));
+	xfs_buf_unlock(bp);
+
+	bp->b_iodone = xlog_iodone;
 	log->l_xbuf = bp;
 
 	spin_lock_init(&log->l_icloglock);
@@ -1398,6 +1409,9 @@ xlog_alloc_log(
 		if (!bp)
 			goto out_free_iclog;
 
+		ASSERT(xfs_buf_islocked(bp));
+		xfs_buf_unlock(bp);
+
 		bp->b_iodone = xlog_iodone;
 		iclog->ic_bp = bp;
 		iclog->ic_data = bp->b_addr;
@@ -1422,7 +1436,6 @@ xlog_alloc_log(
 		iclog->ic_callback_tail = &(iclog->ic_callback);
 		iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
 
-		ASSERT(xfs_buf_islocked(iclog->ic_bp));
 		init_waitqueue_head(&iclog->ic_force_wait);
 		init_waitqueue_head(&iclog->ic_write_wait);
 
@@ -1631,6 +1644,12 @@ xlog_cksum(
  * we transition the iclogs to IOERROR state *after* flushing all existing
  * iclogs to disk. This is because we don't want anymore new transactions to be
  * started or completed afterwards.
+ *
+ * We lock the iclogbufs here so that we can serialise against IO completion
+ * during unmount. We might be processing a shutdown triggered during unmount,
+ * and that can occur asynchronously to the unmount thread, and hence we need to
+ * ensure that completes before tearing down the iclogbufs. Hence we need to
+ * hold the buffer lock across the log IO to acheive that.
  */
 STATIC int
 xlog_bdstrat(
@@ -1638,6 +1657,7 @@ xlog_bdstrat(
 {
 	struct xlog_in_core	*iclog = bp->b_fspriv;
 
+	xfs_buf_lock(bp);
 	if (iclog->ic_state & XLOG_STATE_IOERROR) {
 		xfs_buf_ioerror(bp, EIO);
 		xfs_buf_stale(bp);
@@ -1645,7 +1665,8 @@ xlog_bdstrat(
 		/*
 		 * It would seem logical to return EIO here, but we rely on
 		 * the log state machine to propagate I/O errors instead of
-		 * doing it here.
+		 * doing it here. Similarly, IO completion will unlock the
+		 * buffer, so we don't do it here.
 		 */
 		return 0;
 	}
@@ -1847,14 +1868,28 @@ xlog_dealloc_log(
 	xlog_cil_destroy(log);
 
 	/*
-	 * always need to ensure that the extra buffer does not point to memory
-	 * owned by another log buffer before we free it.
+	 * Cycle all the iclogbuf locks to make sure all log IO completion
+	 * is done before we tear down these buffers.
 	 */
+	iclog = log->l_iclog;
+	for (i = 0; i < log->l_iclog_bufs; i++) {
+		xfs_buf_lock(iclog->ic_bp);
+		xfs_buf_unlock(iclog->ic_bp);
+		iclog = iclog->ic_next;
+	}
+
+	/*
+	 * Always need to ensure that the extra buffer does not point to memory
+	 * owned by another log buffer before we free it. Also, cycle the lock
+	 * first to ensure we've completed IO on it.
+	 */
+	xfs_buf_lock(log->l_xbuf);
+	xfs_buf_unlock(log->l_xbuf);
 	xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
 	xfs_buf_free(log->l_xbuf);
 
 	iclog = log->l_iclog;
-	for (i=0; i<log->l_iclog_bufs; i++) {
+	for (i = 0; i < log->l_iclog_bufs; i++) {
 		xfs_buf_free(iclog->ic_bp);
 		next_iclog = iclog->ic_next;
 		kmem_free(iclog);

fs/xfs/xfs_trace.h

@@ -603,6 +603,7 @@ DEFINE_INODE_EVENT(xfs_readlink);
 DEFINE_INODE_EVENT(xfs_inactive_symlink);
 DEFINE_INODE_EVENT(xfs_alloc_file_space);
 DEFINE_INODE_EVENT(xfs_free_file_space);
+DEFINE_INODE_EVENT(xfs_zero_file_space);
 DEFINE_INODE_EVENT(xfs_collapse_file_space);
 DEFINE_INODE_EVENT(xfs_readdir);
 #ifdef CONFIG_XFS_POSIX_ACL