Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / xfs / xfs_attr_inactive.c
diff --git a/kernel/fs/xfs/xfs_attr_inactive.c b/kernel/fs/xfs/xfs_attr_inactive.c
new file mode 100644 (file)
index 0000000..3fbf167
--- /dev/null
@@ -0,0 +1,465 @@
+/*
+ * Copyright (c) 2000-2005 Silicon Graphics, Inc.
+ * Copyright (c) 2013 Red Hat, Inc.
+ * All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_bit.h"
+#include "xfs_mount.h"
+#include "xfs_da_format.h"
+#include "xfs_da_btree.h"
+#include "xfs_inode.h"
+#include "xfs_alloc.h"
+#include "xfs_attr_remote.h"
+#include "xfs_trans.h"
+#include "xfs_inode_item.h"
+#include "xfs_bmap.h"
+#include "xfs_attr.h"
+#include "xfs_attr_leaf.h"
+#include "xfs_error.h"
+#include "xfs_quota.h"
+#include "xfs_trace.h"
+#include "xfs_dir2.h"
+
+/*
+ * Look at all the extents for this logical region,
+ * invalidate any buffers that are incore/in transactions.
+ */
+STATIC int
+xfs_attr3_leaf_freextent(
+       struct xfs_trans        **trans,
+       struct xfs_inode        *dp,
+       xfs_dablk_t             blkno,
+       int                     blkcnt)
+{
+       struct xfs_bmbt_irec    map;
+       struct xfs_buf          *bp;
+       xfs_dablk_t             tblkno;
+       xfs_daddr_t             dblkno;
+       int                     tblkcnt;
+       int                     dblkcnt;
+       int                     nmap;
+       int                     error;
+
+       /*
+        * Roll through the "value", invalidating the attribute value's
+        * blocks.
+        */
+       tblkno = blkno;
+       tblkcnt = blkcnt;
+       while (tblkcnt > 0) {
+               /*
+                * Try to remember where we decided to put the value.
+                */
+               nmap = 1;
+               error = xfs_bmapi_read(dp, (xfs_fileoff_t)tblkno, tblkcnt,
+                                      &map, &nmap, XFS_BMAPI_ATTRFORK);
+               if (error) {
+                       return error;
+               }
+               ASSERT(nmap == 1);
+               ASSERT(map.br_startblock != DELAYSTARTBLOCK);
+
+               /*
+                * If it's a hole, these are already unmapped
+                * so there's nothing to invalidate.
+                */
+               if (map.br_startblock != HOLESTARTBLOCK) {
+
+                       dblkno = XFS_FSB_TO_DADDR(dp->i_mount,
+                                                 map.br_startblock);
+                       dblkcnt = XFS_FSB_TO_BB(dp->i_mount,
+                                               map.br_blockcount);
+                       bp = xfs_trans_get_buf(*trans,
+                                       dp->i_mount->m_ddev_targp,
+                                       dblkno, dblkcnt, 0);
+                       if (!bp)
+                               return -ENOMEM;
+                       xfs_trans_binval(*trans, bp);
+                       /*
+                        * Roll to next transaction.
+                        */
+                       error = xfs_trans_roll(trans, dp);
+                       if (error)
+                               return error;
+               }
+
+               tblkno += map.br_blockcount;
+               tblkcnt -= map.br_blockcount;
+       }
+
+       return 0;
+}
+
+/*
+ * Invalidate all of the "remote" value regions pointed to by a particular
+ * leaf block.
+ * Note that we must release the lock on the buffer so that we are not
+ * caught holding something that the logging code wants to flush to disk.
+ */
+STATIC int
+xfs_attr3_leaf_inactive(
+       struct xfs_trans        **trans,
+       struct xfs_inode        *dp,
+       struct xfs_buf          *bp)
+{
+       struct xfs_attr_leafblock *leaf;
+       struct xfs_attr3_icleaf_hdr ichdr;
+       struct xfs_attr_leaf_entry *entry;
+       struct xfs_attr_leaf_name_remote *name_rmt;
+       struct xfs_attr_inactive_list *list;
+       struct xfs_attr_inactive_list *lp;
+       int                     error;
+       int                     count;
+       int                     size;
+       int                     tmp;
+       int                     i;
+       struct xfs_mount        *mp = bp->b_target->bt_mount;
+
+       leaf = bp->b_addr;
+       xfs_attr3_leaf_hdr_from_disk(mp->m_attr_geo, &ichdr, leaf);
+
+       /*
+        * Count the number of "remote" value extents.
+        */
+       count = 0;
+       entry = xfs_attr3_leaf_entryp(leaf);
+       for (i = 0; i < ichdr.count; entry++, i++) {
+               if (be16_to_cpu(entry->nameidx) &&
+                   ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
+                       name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
+                       if (name_rmt->valueblk)
+                               count++;
+               }
+       }
+
+       /*
+        * If there are no "remote" values, we're done.
+        */
+       if (count == 0) {
+               xfs_trans_brelse(*trans, bp);
+               return 0;
+       }
+
+       /*
+        * Allocate storage for a list of all the "remote" value extents.
+        */
+       size = count * sizeof(xfs_attr_inactive_list_t);
+       list = kmem_alloc(size, KM_SLEEP);
+
+       /*
+        * Identify each of the "remote" value extents.
+        */
+       lp = list;
+       entry = xfs_attr3_leaf_entryp(leaf);
+       for (i = 0; i < ichdr.count; entry++, i++) {
+               if (be16_to_cpu(entry->nameidx) &&
+                   ((entry->flags & XFS_ATTR_LOCAL) == 0)) {
+                       name_rmt = xfs_attr3_leaf_name_remote(leaf, i);
+                       if (name_rmt->valueblk) {
+                               lp->valueblk = be32_to_cpu(name_rmt->valueblk);
+                               lp->valuelen = xfs_attr3_rmt_blocks(dp->i_mount,
+                                                   be32_to_cpu(name_rmt->valuelen));
+                               lp++;
+                       }
+               }
+       }
+       xfs_trans_brelse(*trans, bp);   /* unlock for trans. in freextent() */
+
+       /*
+        * Invalidate each of the "remote" value extents.
+        */
+       error = 0;
+       for (lp = list, i = 0; i < count; i++, lp++) {
+               tmp = xfs_attr3_leaf_freextent(trans, dp,
+                               lp->valueblk, lp->valuelen);
+
+               if (error == 0)
+                       error = tmp;    /* save only the 1st errno */
+       }
+
+       kmem_free(list);
+       return error;
+}
+
+/*
+ * Recurse (gasp!) through the attribute nodes until we find leaves.
+ * We're doing a depth-first traversal in order to invalidate everything.
+ */
+STATIC int
+xfs_attr3_node_inactive(
+       struct xfs_trans **trans,
+       struct xfs_inode *dp,
+       struct xfs_buf  *bp,
+       int             level)
+{
+       xfs_da_blkinfo_t *info;
+       xfs_da_intnode_t *node;
+       xfs_dablk_t child_fsb;
+       xfs_daddr_t parent_blkno, child_blkno;
+       int error, i;
+       struct xfs_buf *child_bp;
+       struct xfs_da_node_entry *btree;
+       struct xfs_da3_icnode_hdr ichdr;
+
+       /*
+        * Since this code is recursive (gasp!) we must protect ourselves.
+        */
+       if (level > XFS_DA_NODE_MAXDEPTH) {
+               xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
+               return -EIO;
+       }
+
+       node = bp->b_addr;
+       dp->d_ops->node_hdr_from_disk(&ichdr, node);
+       parent_blkno = bp->b_bn;
+       if (!ichdr.count) {
+               xfs_trans_brelse(*trans, bp);
+               return 0;
+       }
+       btree = dp->d_ops->node_tree_p(node);
+       child_fsb = be32_to_cpu(btree[0].before);
+       xfs_trans_brelse(*trans, bp);   /* no locks for later trans */
+
+       /*
+        * If this is the node level just above the leaves, simply loop
+        * over the leaves removing all of them.  If this is higher up
+        * in the tree, recurse downward.
+        */
+       for (i = 0; i < ichdr.count; i++) {
+               /*
+                * Read the subsidiary block to see what we have to work with.
+                * Don't do this in a transaction.  This is a depth-first
+                * traversal of the tree so we may deal with many blocks
+                * before we come back to this one.
+                */
+               error = xfs_da3_node_read(*trans, dp, child_fsb, -2, &child_bp,
+                                               XFS_ATTR_FORK);
+               if (error)
+                       return error;
+               if (child_bp) {
+                                               /* save for re-read later */
+                       child_blkno = XFS_BUF_ADDR(child_bp);
+
+                       /*
+                        * Invalidate the subtree, however we have to.
+                        */
+                       info = child_bp->b_addr;
+                       switch (info->magic) {
+                       case cpu_to_be16(XFS_DA_NODE_MAGIC):
+                       case cpu_to_be16(XFS_DA3_NODE_MAGIC):
+                               error = xfs_attr3_node_inactive(trans, dp,
+                                                       child_bp, level + 1);
+                               break;
+                       case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
+                       case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
+                               error = xfs_attr3_leaf_inactive(trans, dp,
+                                                       child_bp);
+                               break;
+                       default:
+                               error = -EIO;
+                               xfs_trans_brelse(*trans, child_bp);
+                               break;
+                       }
+                       if (error)
+                               return error;
+
+                       /*
+                        * Remove the subsidiary block from the cache
+                        * and from the log.
+                        */
+                       error = xfs_da_get_buf(*trans, dp, 0, child_blkno,
+                               &child_bp, XFS_ATTR_FORK);
+                       if (error)
+                               return error;
+                       xfs_trans_binval(*trans, child_bp);
+               }
+
+               /*
+                * If we're not done, re-read the parent to get the next
+                * child block number.
+                */
+               if (i + 1 < ichdr.count) {
+                       error = xfs_da3_node_read(*trans, dp, 0, parent_blkno,
+                                                &bp, XFS_ATTR_FORK);
+                       if (error)
+                               return error;
+                       child_fsb = be32_to_cpu(btree[i + 1].before);
+                       xfs_trans_brelse(*trans, bp);
+               }
+               /*
+                * Atomically commit the whole invalidate stuff.
+                */
+               error = xfs_trans_roll(trans, dp);
+               if (error)
+                       return  error;
+       }
+
+       return 0;
+}
+
+/*
+ * Indiscriminately delete the entire attribute fork
+ *
+ * Recurse (gasp!) through the attribute nodes until we find leaves.
+ * We're doing a depth-first traversal in order to invalidate everything.
+ */
+int
+xfs_attr3_root_inactive(
+       struct xfs_trans        **trans,
+       struct xfs_inode        *dp)
+{
+       struct xfs_da_blkinfo   *info;
+       struct xfs_buf          *bp;
+       xfs_daddr_t             blkno;
+       int                     error;
+
+       /*
+        * Read block 0 to see what we have to work with.
+        * We only get here if we have extents, since we remove
+        * the extents in reverse order the extent containing
+        * block 0 must still be there.
+        */
+       error = xfs_da3_node_read(*trans, dp, 0, -1, &bp, XFS_ATTR_FORK);
+       if (error)
+               return error;
+       blkno = bp->b_bn;
+
+       /*
+        * Invalidate the tree, even if the "tree" is only a single leaf block.
+        * This is a depth-first traversal!
+        */
+       info = bp->b_addr;
+       switch (info->magic) {
+       case cpu_to_be16(XFS_DA_NODE_MAGIC):
+       case cpu_to_be16(XFS_DA3_NODE_MAGIC):
+               error = xfs_attr3_node_inactive(trans, dp, bp, 1);
+               break;
+       case cpu_to_be16(XFS_ATTR_LEAF_MAGIC):
+       case cpu_to_be16(XFS_ATTR3_LEAF_MAGIC):
+               error = xfs_attr3_leaf_inactive(trans, dp, bp);
+               break;
+       default:
+               error = -EIO;
+               xfs_trans_brelse(*trans, bp);
+               break;
+       }
+       if (error)
+               return error;
+
+       /*
+        * Invalidate the incore copy of the root block.
+        */
+       error = xfs_da_get_buf(*trans, dp, 0, blkno, &bp, XFS_ATTR_FORK);
+       if (error)
+               return error;
+       xfs_trans_binval(*trans, bp);   /* remove from cache */
+       /*
+        * Commit the invalidate and start the next transaction.
+        */
+       error = xfs_trans_roll(trans, dp);
+
+       return error;
+}
+
+/*
+ * xfs_attr_inactive kills all traces of an attribute fork on an inode. It
+ * removes both the on-disk and in-memory inode fork. Note that this also has to
+ * handle the condition of inodes without attributes but with an attribute fork
+ * configured, so we can't use xfs_inode_hasattr() here.
+ *
+ * The in-memory attribute fork is removed even on error.
+ */
+int
+xfs_attr_inactive(
+       struct xfs_inode        *dp)
+{
+       struct xfs_trans        *trans;
+       struct xfs_mount        *mp;
+       int                     cancel_flags = 0;
+       int                     lock_mode = XFS_ILOCK_SHARED;
+       int                     error = 0;
+
+       mp = dp->i_mount;
+       ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
+
+       xfs_ilock(dp, lock_mode);
+       if (!XFS_IFORK_Q(dp))
+               goto out_destroy_fork;
+       xfs_iunlock(dp, lock_mode);
+
+       /*
+        * Start our first transaction of the day.
+        *
+        * All future transactions during this code must be "chained" off
+        * this one via the trans_dup() call.  All transactions will contain
+        * the inode, and the inode will always be marked with trans_ihold().
+        * Since the inode will be locked in all transactions, we must log
+        * the inode in every transaction to let it float upward through
+        * the log.
+        */
+       lock_mode = 0;
+       trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
+       error = xfs_trans_reserve(trans, &M_RES(mp)->tr_attrinval, 0, 0);
+       if (error)
+               goto out_cancel;
+
+       lock_mode = XFS_ILOCK_EXCL;
+       cancel_flags = XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT;
+       xfs_ilock(dp, lock_mode);
+
+       if (!XFS_IFORK_Q(dp))
+               goto out_cancel;
+
+       /*
+        * No need to make quota reservations here. We expect to release some
+        * blocks, not allocate, in the common case.
+        */
+       xfs_trans_ijoin(trans, dp, 0);
+
+       /* invalidate and truncate the attribute fork extents */
+       if (dp->i_d.di_aformat != XFS_DINODE_FMT_LOCAL) {
+               error = xfs_attr3_root_inactive(&trans, dp);
+               if (error)
+                       goto out_cancel;
+
+               error = xfs_itruncate_extents(&trans, dp, XFS_ATTR_FORK, 0);
+               if (error)
+                       goto out_cancel;
+       }
+
+       /* Reset the attribute fork - this also destroys the in-core fork */
+       xfs_attr_fork_remove(dp, trans);
+
+       error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
+       xfs_iunlock(dp, lock_mode);
+       return error;
+
+out_cancel:
+       xfs_trans_cancel(trans, cancel_flags);
+out_destroy_fork:
+       /* kill the in-core attr fork before we drop the inode lock */
+       if (dp->i_afp)
+               xfs_idestroy_fork(dp, XFS_ATTR_FORK);
+       if (lock_mode)
+               xfs_iunlock(dp, lock_mode);
+       return error;
+}