[XFS] handle inode creating race

SGI-PV: 921072
SGI-Modid: xfs-linux:xfs-kern:181657a
Signed-off-by: Christoph Hellwig <hch@sgi.com>
Signed-off-by: Nathan Scott <nathans@sgi.com>

fs/xfs/linux-2.6/xfs_ioctl.c

@@ -270,7 +270,7 @@ xfs_vget_fsop_handlereq(
 	/*
 	 * Get the XFS inode, building a vnode to go with it.
 	 */
-	error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, 0);
+	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
 	if (error)
 		return error;
 	if (ip == NULL)

fs/xfs/linux-2.6/xfs_super.c

@@ -234,6 +234,10 @@ xfs_initialize_vnode(
 		vp->v_type = IFTOVT(ip->i_d.di_mode);
 		xfs_revalidate_inode(XFS_BHVTOM(bdp), vp, ip);
 		xfs_set_inodeops(inode);
+	
+		ip->i_flags &= ~XFS_INEW;
+		barrier();
+
 		unlock_new_inode(inode);
 	}
 }

fs/xfs/quota/xfs_qm.c

@@ -1782,7 +1782,7 @@ xfs_qm_dqusage_adjust(
 	 * the case in all other instances. It's OK that we do this because
 	 * quotacheck is done only at mount time.
 	 */
-	if ((error = xfs_iget(mp, NULL, ino, XFS_ILOCK_EXCL, &ip, bno))) {
+	if ((error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_EXCL, &ip, bno))) {
 		*res = BULKSTAT_RV_NOTHING;
 		return (error);
 	}
@@ -2010,14 +2010,14 @@ xfs_qm_init_quotainos(
 		    mp->m_sb.sb_uquotino != NULLFSINO) {
 			ASSERT(mp->m_sb.sb_uquotino > 0);
 			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
-					     0, &uip, 0)))
+					     0, 0, &uip, 0)))
 				return XFS_ERROR(error);
 		}
 		if (XFS_IS_GQUOTA_ON(mp) &&
 		    mp->m_sb.sb_gquotino != NULLFSINO) {
 			ASSERT(mp->m_sb.sb_gquotino > 0);
 			if ((error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
-					     0, &gip, 0))) {
+					     0, 0, &gip, 0))) {
 				if (uip)
 					VN_RELE(XFS_ITOV(uip));
 				return XFS_ERROR(error);

fs/xfs/quota/xfs_qm_syscalls.c

@@ -404,7 +404,7 @@ xfs_qm_scall_trunc_qfiles(
 	}
 
 	if ((flags & XFS_DQ_USER) && mp->m_sb.sb_uquotino != NULLFSINO) {
-		error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino, 0, &qip, 0);
+		error = xfs_iget(mp, NULL, mp->m_sb.sb_uquotino, 0, 0, &qip, 0);
 		if (! error) {
 			(void) xfs_truncate_file(mp, qip);
 			VN_RELE(XFS_ITOV(qip));
@@ -412,7 +412,7 @@ xfs_qm_scall_trunc_qfiles(
 	}
 
 	if ((flags & XFS_DQ_GROUP) && mp->m_sb.sb_gquotino != NULLFSINO) {
-		error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino, 0, &qip, 0);
+		error = xfs_iget(mp, NULL, mp->m_sb.sb_gquotino, 0, 0, &qip, 0);
 		if (! error) {
 			(void) xfs_truncate_file(mp, qip);
 			VN_RELE(XFS_ITOV(qip));
@@ -555,11 +555,13 @@ xfs_qm_scall_getqstat(
 		gip = mp->m_quotainfo->qi_gquotaip;
 	}
 	if (!uip && mp->m_sb.sb_uquotino != NULLFSINO) {
-		if (xfs_iget(mp, NULL, mp->m_sb.sb_uquotino, 0, &uip, 0) == 0)
+		if (xfs_iget(mp, NULL, mp->m_sb.sb_uquotino,
+					0, 0, &uip, 0) == 0)
 			tempuqip = B_TRUE;
 	}
 	if (!gip && mp->m_sb.sb_gquotino != NULLFSINO) {
-		if (xfs_iget(mp, NULL, mp->m_sb.sb_gquotino, 0, &gip, 0) == 0)
+		if (xfs_iget(mp, NULL, mp->m_sb.sb_gquotino,
+					0, 0, &gip, 0) == 0)
 			tempgqip = B_TRUE;
 	}
 	if (uip) {
@@ -1338,7 +1340,7 @@ xfs_qm_internalqcheck_adjust(
 	ipreleased = B_FALSE;
  again:
 	lock_flags = XFS_ILOCK_SHARED;
-	if ((error = xfs_iget(mp, NULL, ino, lock_flags, &ip, bno))) {
+	if ((error = xfs_iget(mp, NULL, ino, 0, lock_flags, &ip, bno))) {
 		*res = BULKSTAT_RV_NOTHING;
 		return (error);
 	}

fs/xfs/xfs_iget.c

@@ -169,6 +169,7 @@ xfs_iget_core(
 	xfs_mount_t	*mp,
 	xfs_trans_t	*tp,
 	xfs_ino_t	ino,
+	uint		flags,
 	uint		lock_flags,
 	xfs_inode_t	**ipp,
 	xfs_daddr_t	bno)
@@ -180,7 +181,6 @@ xfs_iget_core(
 	ulong		version;
 	int		error;
 	/* REFERENCED */
-	int		newnode;
 	xfs_chash_t	*ch;
 	xfs_chashlist_t	*chl, *chlnew;
 	SPLDECL(s);
@@ -193,11 +193,22 @@ xfs_iget_core(
 
 	for (ip = ih->ih_next; ip != NULL; ip = ip->i_next) {
 		if (ip->i_ino == ino) {
+			/*
+			 * If INEW is set this inode is being set up
+			 * we need to pause and try again.
+			 */
+			if (ip->i_flags & XFS_INEW) {
+				read_unlock(&ih->ih_lock);
+				delay(1);
+				XFS_STATS_INC(xs_ig_frecycle);
 
-			inode_vp = XFS_ITOV_NULL(ip);
+				goto again;
+			}
 
+			inode_vp = XFS_ITOV_NULL(ip);
 			if (inode_vp == NULL) {
-				/* If IRECLAIM is set this inode is
+				/*
+				 * If IRECLAIM is set this inode is
 				 * on its way out of the system,
 				 * we need to pause and try again.
 				 */
@@ -250,14 +261,15 @@ xfs_iget_core(
 			XFS_STATS_INC(xs_ig_found);
 
 finish_inode:
-			if (lock_flags != 0) {
-				xfs_ilock(ip, lock_flags);
-			}
-
-			newnode = (ip->i_d.di_mode == 0);
-			if (newnode) {
+			if (ip->i_d.di_mode == 0) {
+				if (!(flags & IGET_CREATE))
+					return ENOENT;
 				xfs_iocore_inode_reinit(ip);
 			}
+	
+			if (lock_flags != 0)
+				xfs_ilock(ip, lock_flags);
+
 			ip->i_flags &= ~XFS_ISTALE;
 
 			vn_trace_exit(vp, "xfs_iget.found",
@@ -293,6 +305,11 @@ xfs_iget_core(
 	if (lock_flags != 0) {
 		xfs_ilock(ip, lock_flags);
 	}
+		
+	if ((ip->i_d.di_mode == 0) && !(flags & IGET_CREATE)) {
+		xfs_idestroy(ip);
+		return ENOENT;
+	}
 
 	/*
 	 * Put ip on its hash chain, unless someone else hashed a duplicate
@@ -324,6 +341,7 @@ xfs_iget_core(
 	ih->ih_next = ip;
 	ip->i_udquot = ip->i_gdquot = NULL;
 	ih->ih_version++;
+	ip->i_flags |= XFS_INEW;
 
 	write_unlock(&ih->ih_lock);
 
@@ -404,8 +422,6 @@ xfs_iget_core(
 
 	XFS_MOUNT_IUNLOCK(mp);
 
-	newnode = 1;
-
  return_ip:
 	ASSERT(ip->i_df.if_ext_max ==
 	       XFS_IFORK_DSIZE(ip) / sizeof(xfs_bmbt_rec_t));
@@ -434,6 +450,7 @@ xfs_iget(
 	xfs_mount_t	*mp,
 	xfs_trans_t	*tp,
 	xfs_ino_t	ino,
+	uint		flags,
 	uint		lock_flags,
 	xfs_inode_t	**ipp,
 	xfs_daddr_t	bno)
@@ -454,8 +471,8 @@ xfs_iget(
 		if (inode->i_state & I_NEW) {
 inode_allocate:
 			vn_initialize(inode);
-			error = xfs_iget_core(vp, mp, tp, ino,
-						lock_flags, ipp, bno);
+			error = xfs_iget_core(vp, mp, tp, ino, flags,
+					lock_flags, ipp, bno);
 			if (error) {
 				vn_mark_bad(vp);
 				if (inode->i_state & I_NEW)
@@ -576,6 +593,10 @@ xfs_iput_new(xfs_inode_t	*ip,
 
 	vn_trace_entry(vp, "xfs_iput_new", (inst_t *)__return_address);
 
+	if ((ip->i_d.di_mode == 0)) {
+		ASSERT(!(ip->i_flags & XFS_IRECLAIMABLE));
+		vn_mark_bad(vp);
+	}
 	if (inode->i_state & I_NEW)
 		unlock_new_inode(inode);
 	if (lock_flags)

fs/xfs/xfs_inode.c

@@ -1163,7 +1163,8 @@ xfs_ialloc(
 	 * This is because we're setting fields here we need
 	 * to prevent others from looking at until we're done.
 	 */
-	error = xfs_trans_iget(tp->t_mountp, tp, ino, XFS_ILOCK_EXCL, &ip);
+	error = xfs_trans_iget(tp->t_mountp, tp, ino,
+			IGET_CREATE, XFS_ILOCK_EXCL, &ip);
 	if (error != 0) {
 		return error;
 	}

fs/xfs/xfs_inode.h

@@ -381,6 +381,7 @@ void xfs_ifork_next_set(xfs_inode_t *ip, int w, int n);
 #define XFS_IRECLAIM    0x0008  /* we have started reclaiming this inode    */
 #define XFS_ISTALE	0x0010	/* inode has been staled */
 #define XFS_IRECLAIMABLE 0x0020 /* inode can be reclaimed */
+#define XFS_INEW	0x0040
 
 /*
  * Flags for inode locking.
@@ -465,6 +466,9 @@ xfs_inode_t *xfs_bhvtoi(struct bhv_desc *bhvp);
 /*
  * xfs_iget.c prototypes.
  */
+
+#define IGET_CREATE	1
+
 void		xfs_ihash_init(struct xfs_mount *);
 void		xfs_ihash_free(struct xfs_mount *);
 void		xfs_chash_init(struct xfs_mount *);
@@ -473,7 +477,7 @@ xfs_inode_t	*xfs_inode_incore(struct xfs_mount *, xfs_ino_t,
 				  struct xfs_trans *);
 void            xfs_inode_lock_init(xfs_inode_t *, struct vnode *);
 int		xfs_iget(struct xfs_mount *, struct xfs_trans *, xfs_ino_t,
-			 uint, xfs_inode_t **, xfs_daddr_t);
+			 uint, uint, xfs_inode_t **, xfs_daddr_t);
 void		xfs_iput(xfs_inode_t *, uint);
 void		xfs_iput_new(xfs_inode_t *, uint);
 void		xfs_ilock(xfs_inode_t *, uint);

fs/xfs/xfs_itable.c

@@ -102,7 +102,7 @@ xfs_bulkstat_one(
 		/* We're not being passed a pointer to a dinode.  This happens
 		 * if BULKSTAT_FG_IGET is selected.  Do the iget.
 		 */
-		error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, bno);
+		error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
 		if (error) {
 			*stat = BULKSTAT_RV_NOTHING;
 			return error;

fs/xfs/xfs_log_recover.c

@@ -3262,7 +3262,7 @@ xlog_recover_process_iunlinks(
 				xfs_buf_relse(agibp);
 
 				ino = XFS_AGINO_TO_INO(mp, agno, agino);
-				error = xfs_iget(mp, NULL, ino, 0, &ip, 0);
+				error = xfs_iget(mp, NULL, ino, 0, 0, &ip, 0);
 				ASSERT(error || (ip != NULL));
 
 				if (!error) {

fs/xfs/xfs_mount.c

@@ -975,7 +975,7 @@ xfs_mountfs(
 	 * Get and sanity-check the root inode.
 	 * Save the pointer to it in the mount structure.
 	 */
-	error = xfs_iget(mp, NULL, sbp->sb_rootino, XFS_ILOCK_EXCL, &rip, 0);
+	error = xfs_iget(mp, NULL, sbp->sb_rootino, 0, XFS_ILOCK_EXCL, &rip, 0);
 	if (error) {
 		cmn_err(CE_WARN, "XFS: failed to read root inode");
 		goto error3;

fs/xfs/xfs_rtalloc.c

@@ -149,7 +149,7 @@ xfs_growfs_rt_alloc(
 		/*
 		 * Lock the inode.
 		 */
-		if ((error = xfs_trans_iget(mp, tp, ino, XFS_ILOCK_EXCL, &ip)))
+		if ((error = xfs_trans_iget(mp, tp, ino, 0, XFS_ILOCK_EXCL, &ip)))
 			goto error_exit;
 		XFS_BMAP_INIT(&flist, &firstblock);
 		/*
@@ -189,7 +189,7 @@ xfs_growfs_rt_alloc(
 			/*
 			 * Lock the bitmap inode.
 			 */
-			if ((error = xfs_trans_iget(mp, tp, ino, XFS_ILOCK_EXCL,
+			if ((error = xfs_trans_iget(mp, tp, ino, 0, XFS_ILOCK_EXCL,
 					&ip)))
 				goto error_exit;
 			/*
@@ -2042,7 +2042,7 @@ xfs_growfs_rt(
 		/*
 		 * Lock out other callers by grabbing the bitmap inode lock.
 		 */
-		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino,
+		if ((error = xfs_trans_iget(mp, tp, 0, mp->m_sb.sb_rbmino,
 				XFS_ILOCK_EXCL, &ip)))
 			goto error_exit;
 		ASSERT(ip == mp->m_rbmip);
@@ -2057,7 +2057,7 @@ xfs_growfs_rt(
 		 * Get the summary inode into the transaction.
 		 */
 		if ((error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rsumino,
-				XFS_ILOCK_EXCL, &ip)))
+				0, XFS_ILOCK_EXCL, &ip)))
 			goto error_exit;
 		ASSERT(ip == mp->m_rsumip);
 		/*
@@ -2177,7 +2177,7 @@ xfs_rtallocate_extent(
 	/*
 	 * Lock out other callers by grabbing the bitmap inode lock.
 	 */
-	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, XFS_ILOCK_EXCL, &ip);
+	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0, XFS_ILOCK_EXCL, &ip);
 	if (error) {
 		return error;
 	}
@@ -2240,7 +2240,7 @@ xfs_rtfree_extent(
 	/*
 	 * Synchronize by locking the bitmap inode.
 	 */
-	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, XFS_ILOCK_EXCL, &ip);
+	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0, XFS_ILOCK_EXCL, &ip);
 	if (error) {
 		return error;
 	}
@@ -2348,12 +2348,12 @@ xfs_rtmount_inodes(
 	sbp = &mp->m_sb;
 	if (sbp->sb_rbmino == NULLFSINO)
 		return 0;
-	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, &mp->m_rbmip, 0);
+	error = xfs_iget(mp, NULL, sbp->sb_rbmino, 0, 0, &mp->m_rbmip, 0);
 	if (error)
 		return error;
 	ASSERT(mp->m_rbmip != NULL);
 	ASSERT(sbp->sb_rsumino != NULLFSINO);
-	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, &mp->m_rsumip, 0);
+	error = xfs_iget(mp, NULL, sbp->sb_rsumino, 0, 0, &mp->m_rsumip, 0);
 	if (error) {
 		VN_RELE(XFS_ITOV(mp->m_rbmip));
 		return error;
@@ -2384,7 +2384,7 @@ xfs_rtpick_extent(
 	__uint64_t	seq;		/* sequence number of file creation */
 	__uint64_t	*seqp;		/* pointer to seqno in inode */
 
-	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, XFS_ILOCK_EXCL, &ip);
+	error = xfs_trans_iget(mp, tp, mp->m_sb.sb_rbmino, 0, XFS_ILOCK_EXCL, &ip);
 	if (error)
 		return error;
 	ASSERT(ip == mp->m_rbmip);

fs/xfs/xfs_trans.h

@@ -1007,7 +1007,7 @@ void		xfs_trans_stale_inode_buf(xfs_trans_t *, struct xfs_buf *);
 void		xfs_trans_dquot_buf(xfs_trans_t *, struct xfs_buf *, uint);
 void		xfs_trans_inode_alloc_buf(xfs_trans_t *, struct xfs_buf *);
 int		xfs_trans_iget(struct xfs_mount *, xfs_trans_t *,
-			       xfs_ino_t , uint, struct xfs_inode **);
+			       xfs_ino_t , uint, uint, struct xfs_inode **);
 void		xfs_trans_ijoin(xfs_trans_t *, struct xfs_inode *, uint);
 void		xfs_trans_ihold(xfs_trans_t *, struct xfs_inode *);
 void		xfs_trans_ihold_release(xfs_trans_t *, struct xfs_inode *);

fs/xfs/xfs_trans_inode.c

@@ -95,6 +95,7 @@ xfs_trans_iget(
 	xfs_mount_t	*mp,
 	xfs_trans_t	*tp,
 	xfs_ino_t	ino,
+	uint		flags,
 	uint		lock_flags,
 	xfs_inode_t	**ipp)
 {
@@ -106,9 +107,8 @@ xfs_trans_iget(
 	 * If the transaction pointer is NULL, just call the normal
 	 * xfs_iget().
 	 */
-	if (tp == NULL) {
-		return (xfs_iget(mp, NULL, ino, lock_flags, ipp, 0));
-	}
+	if (tp == NULL)
+		return xfs_iget(mp, NULL, ino, flags, lock_flags, ipp, 0);
 
 	/*
 	 * If we find the inode in core with this transaction
@@ -148,7 +148,7 @@ xfs_trans_iget(
 	}
 
 	ASSERT(lock_flags & XFS_ILOCK_EXCL);
-	error = xfs_iget(tp->t_mountp, tp, ino, lock_flags, &ip, 0);
+	error = xfs_iget(tp->t_mountp, tp, ino, flags, lock_flags, &ip, 0);
 	if (error) {
 		return error;
 	}
@@ -186,7 +186,6 @@ xfs_trans_iget(
 	return 0;
 }
 
-
 /*
  * Add the locked inode to the transaction.
  * The inode must be locked, and it cannot be associated with any

fs/xfs/xfs_utils.c

@@ -110,7 +110,7 @@ xfs_dir_lookup_int(
 		 * reservation in the inactive routine.
 		 */
 		xfs_iunlock(dp, lock_mode);
-		error = xfs_iget(dp->i_mount, NULL, *inum, 0, ipp, 0);
+		error = xfs_iget(dp->i_mount, NULL, *inum, 0, 0, ipp, 0);
 		xfs_ilock(dp, lock_mode);
 
 		if (error) {

fs/xfs/xfs_vfsops.c

@@ -1610,7 +1610,7 @@ xfs_vget(
 	if (ino == 0)
 		return XFS_ERROR(ESTALE);
 
-	error = xfs_iget(mp, NULL, ino, XFS_ILOCK_SHARED, &ip, 0);
+	error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, 0);
 	if (error) {
 		*vpp = NULL;
 		return error;