xfs: log rmap intent items
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 3 Aug 2016 02:09:48 +0000 (12:09 +1000)
committerDave Chinner <david@fromorbit.com>
Wed, 3 Aug 2016 02:09:48 +0000 (12:09 +1000)
Provide a mechanism for higher levels to create RUI/RUD items, submit
them to the log, and a stub function to deal with recovered RUI items.
These parts will be connected to the rmapbt in a later patch.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>
fs/xfs/Makefile
fs/xfs/xfs_log_recover.c
fs/xfs/xfs_rmap_item.c
fs/xfs/xfs_rmap_item.h
fs/xfs/xfs_trans.h
fs/xfs/xfs_trans_rmap.c [new file with mode: 0644]

index 6c90393..3124210 100644 (file)
@@ -109,6 +109,7 @@ xfs-y                               += xfs_log.o \
                                   xfs_trans_buf.o \
                                   xfs_trans_extfree.o \
                                   xfs_trans_inode.o \
+                                  xfs_trans_rmap.o \
 
 # optional features
 xfs-$(CONFIG_XFS_QUOTA)                += xfs_dquot.o \
index f64d4d0..365aef6 100644 (file)
@@ -43,6 +43,7 @@
 #include "xfs_bmap_btree.h"
 #include "xfs_error.h"
 #include "xfs_dir2.h"
+#include "xfs_rmap_item.h"
 
 #define BLK_AVG(blk1, blk2)    ((blk1+blk2) >> 1)
 
@@ -1911,6 +1912,8 @@ xlog_recover_reorder_trans(
                case XFS_LI_QUOTAOFF:
                case XFS_LI_EFD:
                case XFS_LI_EFI:
+               case XFS_LI_RUI:
+               case XFS_LI_RUD:
                        trace_xfs_log_recover_item_reorder_tail(log,
                                                        trans, item, pass);
                        list_move_tail(&item->ri_list, &inode_list);
@@ -3415,6 +3418,101 @@ xlog_recover_efd_pass2(
 }
 
 /*
+ * This routine is called to create an in-core extent rmap update
+ * item from the rui format structure which was logged on disk.
+ * It allocates an in-core rui, copies the extents from the format
+ * structure into it, and adds the rui to the AIL with the given
+ * LSN.
+ */
+STATIC int
+xlog_recover_rui_pass2(
+       struct xlog                     *log,
+       struct xlog_recover_item        *item,
+       xfs_lsn_t                       lsn)
+{
+       int                             error;
+       struct xfs_mount                *mp = log->l_mp;
+       struct xfs_rui_log_item         *ruip;
+       struct xfs_rui_log_format       *rui_formatp;
+
+       rui_formatp = item->ri_buf[0].i_addr;
+
+       ruip = xfs_rui_init(mp, rui_formatp->rui_nextents);
+       error = xfs_rui_copy_format(&item->ri_buf[0], &ruip->rui_format);
+       if (error) {
+               xfs_rui_item_free(ruip);
+               return error;
+       }
+       atomic_set(&ruip->rui_next_extent, rui_formatp->rui_nextents);
+
+       spin_lock(&log->l_ailp->xa_lock);
+       /*
+        * The RUI has two references. One for the RUD and one for RUI to ensure
+        * it makes it into the AIL. Insert the RUI into the AIL directly and
+        * drop the RUI reference. Note that xfs_trans_ail_update() drops the
+        * AIL lock.
+        */
+       xfs_trans_ail_update(log->l_ailp, &ruip->rui_item, lsn);
+       xfs_rui_release(ruip);
+       return 0;
+}
+
+
+/*
+ * This routine is called when an RUD format structure is found in a committed
+ * transaction in the log. Its purpose is to cancel the corresponding RUI if it
+ * was still in the log. To do this it searches the AIL for the RUI with an id
+ * equal to that in the RUD format structure. If we find it we drop the RUD
+ * reference, which removes the RUI from the AIL and frees it.
+ */
+STATIC int
+xlog_recover_rud_pass2(
+       struct xlog                     *log,
+       struct xlog_recover_item        *item)
+{
+       struct xfs_rud_log_format       *rud_formatp;
+       struct xfs_rui_log_item         *ruip = NULL;
+       struct xfs_log_item             *lip;
+       __uint64_t                      rui_id;
+       struct xfs_ail_cursor           cur;
+       struct xfs_ail                  *ailp = log->l_ailp;
+
+       rud_formatp = item->ri_buf[0].i_addr;
+       ASSERT(item->ri_buf[0].i_len == (sizeof(struct xfs_rud_log_format) +
+                       ((rud_formatp->rud_nextents - 1) *
+                       sizeof(struct xfs_map_extent))));
+       rui_id = rud_formatp->rud_rui_id;
+
+       /*
+        * Search for the RUI with the id in the RUD format structure in the
+        * AIL.
+        */
+       spin_lock(&ailp->xa_lock);
+       lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
+       while (lip != NULL) {
+               if (lip->li_type == XFS_LI_RUI) {
+                       ruip = (struct xfs_rui_log_item *)lip;
+                       if (ruip->rui_format.rui_id == rui_id) {
+                               /*
+                                * Drop the RUD reference to the RUI. This
+                                * removes the RUI from the AIL and frees it.
+                                */
+                               spin_unlock(&ailp->xa_lock);
+                               xfs_rui_release(ruip);
+                               spin_lock(&ailp->xa_lock);
+                               break;
+                       }
+               }
+               lip = xfs_trans_ail_cursor_next(ailp, &cur);
+       }
+
+       xfs_trans_ail_cursor_done(&cur);
+       spin_unlock(&ailp->xa_lock);
+
+       return 0;
+}
+
+/*
  * This routine is called when an inode create format structure is found in a
  * committed transaction in the log.  It's purpose is to initialise the inodes
  * being allocated on disk. This requires us to get inode cluster buffers that
@@ -3639,6 +3737,8 @@ xlog_recover_ra_pass2(
        case XFS_LI_EFI:
        case XFS_LI_EFD:
        case XFS_LI_QUOTAOFF:
+       case XFS_LI_RUI:
+       case XFS_LI_RUD:
        default:
                break;
        }
@@ -3662,6 +3762,8 @@ xlog_recover_commit_pass1(
        case XFS_LI_EFD:
        case XFS_LI_DQUOT:
        case XFS_LI_ICREATE:
+       case XFS_LI_RUI:
+       case XFS_LI_RUD:
                /* nothing to do in pass 1 */
                return 0;
        default:
@@ -3692,6 +3794,10 @@ xlog_recover_commit_pass2(
                return xlog_recover_efi_pass2(log, item, trans->r_lsn);
        case XFS_LI_EFD:
                return xlog_recover_efd_pass2(log, item);
+       case XFS_LI_RUI:
+               return xlog_recover_rui_pass2(log, item, trans->r_lsn);
+       case XFS_LI_RUD:
+               return xlog_recover_rud_pass2(log, item);
        case XFS_LI_DQUOT:
                return xlog_recover_dquot_pass2(log, buffer_list, item,
                                                trans->r_lsn);
@@ -4204,11 +4310,52 @@ xlog_recover_cancel_efi(
        spin_lock(&ailp->xa_lock);
 }
 
+/* Recover the RUI if necessary. */
+STATIC int
+xlog_recover_process_rui(
+       struct xfs_mount                *mp,
+       struct xfs_ail                  *ailp,
+       struct xfs_log_item             *lip)
+{
+       struct xfs_rui_log_item         *ruip;
+       int                             error;
+
+       /*
+        * Skip RUIs that we've already processed.
+        */
+       ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
+       if (test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags))
+               return 0;
+
+       spin_unlock(&ailp->xa_lock);
+       error = xfs_rui_recover(mp, ruip);
+       spin_lock(&ailp->xa_lock);
+
+       return error;
+}
+
+/* Release the RUI since we're cancelling everything. */
+STATIC void
+xlog_recover_cancel_rui(
+       struct xfs_mount                *mp,
+       struct xfs_ail                  *ailp,
+       struct xfs_log_item             *lip)
+{
+       struct xfs_rui_log_item         *ruip;
+
+       ruip = container_of(lip, struct xfs_rui_log_item, rui_item);
+
+       spin_unlock(&ailp->xa_lock);
+       xfs_rui_release(ruip);
+       spin_lock(&ailp->xa_lock);
+}
+
 /* Is this log item a deferred action intent? */
 static inline bool xlog_item_is_intent(struct xfs_log_item *lip)
 {
        switch (lip->li_type) {
        case XFS_LI_EFI:
+       case XFS_LI_RUI:
                return true;
        default:
                return false;
@@ -4269,6 +4416,9 @@ xlog_recover_process_intents(
                case XFS_LI_EFI:
                        error = xlog_recover_process_efi(log->l_mp, ailp, lip);
                        break;
+               case XFS_LI_RUI:
+                       error = xlog_recover_process_rui(log->l_mp, ailp, lip);
+                       break;
                }
                if (error)
                        goto out;
@@ -4313,6 +4463,9 @@ xlog_recover_cancel_intents(
                case XFS_LI_EFI:
                        xlog_recover_cancel_efi(log->l_mp, ailp, lip);
                        break;
+               case XFS_LI_RUI:
+                       xlog_recover_cancel_rui(log->l_mp, ailp, lip);
+                       break;
                }
 
                lip = xfs_trans_ail_cursor_next(ailp, &cur);
@@ -5130,6 +5283,7 @@ xlog_recover_finish(
                        xfs_alert(log->l_mp, "Failed to recover intents");
                        return error;
                }
+
                /*
                 * Sync the log to get all the intents out of the AIL.
                 * This isn't absolutely necessary, but it helps in
index 5398b84..fecd1e4 100644 (file)
@@ -22,6 +22,7 @@
 #include "xfs_format.h"
 #include "xfs_log_format.h"
 #include "xfs_trans_resv.h"
+#include "xfs_bit.h"
 #include "xfs_mount.h"
 #include "xfs_trans.h"
 #include "xfs_trans_priv.h"
@@ -457,3 +458,62 @@ xfs_rud_init(
 
        return rudp;
 }
+
+/*
+ * Process an rmap update intent item that was recovered from the log.
+ * We need to update the rmapbt.
+ */
+int
+xfs_rui_recover(
+       struct xfs_mount                *mp,
+       struct xfs_rui_log_item         *ruip)
+{
+       int                             i;
+       int                             error = 0;
+       struct xfs_map_extent           *rmap;
+       xfs_fsblock_t                   startblock_fsb;
+       bool                            op_ok;
+
+       ASSERT(!test_bit(XFS_RUI_RECOVERED, &ruip->rui_flags));
+
+       /*
+        * First check the validity of the extents described by the
+        * RUI.  If any are bad, then assume that all are bad and
+        * just toss the RUI.
+        */
+       for (i = 0; i < ruip->rui_format.rui_nextents; i++) {
+               rmap = &(ruip->rui_format.rui_extents[i]);
+               startblock_fsb = XFS_BB_TO_FSB(mp,
+                                  XFS_FSB_TO_DADDR(mp, rmap->me_startblock));
+               switch (rmap->me_flags & XFS_RMAP_EXTENT_TYPE_MASK) {
+               case XFS_RMAP_EXTENT_MAP:
+               case XFS_RMAP_EXTENT_UNMAP:
+               case XFS_RMAP_EXTENT_CONVERT:
+               case XFS_RMAP_EXTENT_ALLOC:
+               case XFS_RMAP_EXTENT_FREE:
+                       op_ok = true;
+                       break;
+               default:
+                       op_ok = false;
+                       break;
+               }
+               if (!op_ok || (startblock_fsb == 0) ||
+                   (rmap->me_len == 0) ||
+                   (startblock_fsb >= mp->m_sb.sb_dblocks) ||
+                   (rmap->me_len >= mp->m_sb.sb_agblocks) ||
+                   (rmap->me_flags & ~XFS_RMAP_EXTENT_FLAGS)) {
+                       /*
+                        * This will pull the RUI from the AIL and
+                        * free the memory associated with it.
+                        */
+                       set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags);
+                       xfs_rui_release(ruip);
+                       return -EIO;
+               }
+       }
+
+       /* XXX: do nothing for now */
+       set_bit(XFS_RUI_RECOVERED, &ruip->rui_flags);
+       xfs_rui_release(ruip);
+       return error;
+}
index bd36ab5..59ef3ec 100644 (file)
@@ -96,5 +96,6 @@ int xfs_rui_copy_format(struct xfs_log_iovec *buf,
                struct xfs_rui_log_format *dst_rui_fmt);
 void xfs_rui_item_free(struct xfs_rui_log_item *);
 void xfs_rui_release(struct xfs_rui_log_item *);
+int xfs_rui_recover(struct xfs_mount *mp, struct xfs_rui_log_item *ruip);
 
 #endif /* __XFS_RMAP_ITEM_H__ */
index 9bc368b..0226581 100644 (file)
@@ -233,4 +233,21 @@ void               xfs_trans_buf_copy_type(struct xfs_buf *dst_bp,
 extern kmem_zone_t     *xfs_trans_zone;
 extern kmem_zone_t     *xfs_log_item_desc_zone;
 
+enum xfs_rmap_intent_type;
+
+struct xfs_rui_log_item *xfs_trans_get_rui(struct xfs_trans *tp, uint nextents);
+void xfs_trans_log_start_rmap_update(struct xfs_trans *tp,
+               struct xfs_rui_log_item *ruip, enum xfs_rmap_intent_type type,
+               __uint64_t owner, int whichfork, xfs_fileoff_t startoff,
+               xfs_fsblock_t startblock, xfs_filblks_t blockcount,
+               xfs_exntst_t state);
+
+struct xfs_rud_log_item *xfs_trans_get_rud(struct xfs_trans *tp,
+               struct xfs_rui_log_item *ruip, uint nextents);
+int xfs_trans_log_finish_rmap_update(struct xfs_trans *tp,
+               struct xfs_rud_log_item *rudp, enum xfs_rmap_intent_type type,
+               __uint64_t owner, int whichfork, xfs_fileoff_t startoff,
+               xfs_fsblock_t startblock, xfs_filblks_t blockcount,
+               xfs_exntst_t state);
+
 #endif /* __XFS_TRANS_H__ */
diff --git a/fs/xfs/xfs_trans_rmap.c b/fs/xfs/xfs_trans_rmap.c
new file mode 100644 (file)
index 0000000..e3a5172
--- /dev/null
@@ -0,0 +1,205 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * 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; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * 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_mount.h"
+#include "xfs_defer.h"
+#include "xfs_trans.h"
+#include "xfs_trans_priv.h"
+#include "xfs_rmap_item.h"
+#include "xfs_alloc.h"
+#include "xfs_rmap.h"
+
+/*
+ * This routine is called to allocate an "rmap update intent"
+ * log item that will hold nextents worth of extents.  The
+ * caller must use all nextents extents, because we are not
+ * flexible about this at all.
+ */
+struct xfs_rui_log_item *
+xfs_trans_get_rui(
+       struct xfs_trans                *tp,
+       uint                            nextents)
+{
+       struct xfs_rui_log_item         *ruip;
+
+       ASSERT(tp != NULL);
+       ASSERT(nextents > 0);
+
+       ruip = xfs_rui_init(tp->t_mountp, nextents);
+       ASSERT(ruip != NULL);
+
+       /*
+        * Get a log_item_desc to point at the new item.
+        */
+       xfs_trans_add_item(tp, &ruip->rui_item);
+       return ruip;
+}
+
+/* Set the map extent flags for this reverse mapping. */
+static void
+xfs_trans_set_rmap_flags(
+       struct xfs_map_extent           *rmap,
+       enum xfs_rmap_intent_type       type,
+       int                             whichfork,
+       xfs_exntst_t                    state)
+{
+       rmap->me_flags = 0;
+       if (state == XFS_EXT_UNWRITTEN)
+               rmap->me_flags |= XFS_RMAP_EXTENT_UNWRITTEN;
+       if (whichfork == XFS_ATTR_FORK)
+               rmap->me_flags |= XFS_RMAP_EXTENT_ATTR_FORK;
+       switch (type) {
+       case XFS_RMAP_MAP:
+               rmap->me_flags |= XFS_RMAP_EXTENT_MAP;
+               break;
+       case XFS_RMAP_UNMAP:
+               rmap->me_flags |= XFS_RMAP_EXTENT_UNMAP;
+               break;
+       case XFS_RMAP_CONVERT:
+               rmap->me_flags |= XFS_RMAP_EXTENT_CONVERT;
+               break;
+       case XFS_RMAP_ALLOC:
+               rmap->me_flags |= XFS_RMAP_EXTENT_ALLOC;
+               break;
+       case XFS_RMAP_FREE:
+               rmap->me_flags |= XFS_RMAP_EXTENT_FREE;
+               break;
+       default:
+               ASSERT(0);
+       }
+}
+
+/*
+ * This routine is called to indicate that the described reverse
+ * mapping is to be logged as needing to be updated.  It should be
+ * called once for each mapping.
+ */
+void
+xfs_trans_log_start_rmap_update(
+       struct xfs_trans                *tp,
+       struct xfs_rui_log_item         *ruip,
+       enum xfs_rmap_intent_type       type,
+       __uint64_t                      owner,
+       int                             whichfork,
+       xfs_fileoff_t                   startoff,
+       xfs_fsblock_t                   startblock,
+       xfs_filblks_t                   blockcount,
+       xfs_exntst_t                    state)
+{
+       uint                            next_extent;
+       struct xfs_map_extent           *rmap;
+
+       tp->t_flags |= XFS_TRANS_DIRTY;
+       ruip->rui_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+       /*
+        * atomic_inc_return gives us the value after the increment;
+        * we want to use it as an array index so we need to subtract 1 from
+        * it.
+        */
+       next_extent = atomic_inc_return(&ruip->rui_next_extent) - 1;
+       ASSERT(next_extent < ruip->rui_format.rui_nextents);
+       rmap = &(ruip->rui_format.rui_extents[next_extent]);
+       rmap->me_owner = owner;
+       rmap->me_startblock = startblock;
+       rmap->me_startoff = startoff;
+       rmap->me_len = blockcount;
+       xfs_trans_set_rmap_flags(rmap, type, whichfork, state);
+}
+
+
+/*
+ * This routine is called to allocate an "rmap update done"
+ * log item that will hold nextents worth of extents.  The
+ * caller must use all nextents extents, because we are not
+ * flexible about this at all.
+ */
+struct xfs_rud_log_item *
+xfs_trans_get_rud(
+       struct xfs_trans                *tp,
+       struct xfs_rui_log_item         *ruip,
+       uint                            nextents)
+{
+       struct xfs_rud_log_item         *rudp;
+
+       ASSERT(tp != NULL);
+       ASSERT(nextents > 0);
+
+       rudp = xfs_rud_init(tp->t_mountp, ruip, nextents);
+       ASSERT(rudp != NULL);
+
+       /*
+        * Get a log_item_desc to point at the new item.
+        */
+       xfs_trans_add_item(tp, &rudp->rud_item);
+       return rudp;
+}
+
+/*
+ * Finish an rmap update and log it to the RUD. Note that the transaction is
+ * marked dirty regardless of whether the rmap update succeeds or fails to
+ * support the RUI/RUD lifecycle rules.
+ */
+int
+xfs_trans_log_finish_rmap_update(
+       struct xfs_trans                *tp,
+       struct xfs_rud_log_item         *rudp,
+       enum xfs_rmap_intent_type       type,
+       __uint64_t                      owner,
+       int                             whichfork,
+       xfs_fileoff_t                   startoff,
+       xfs_fsblock_t                   startblock,
+       xfs_filblks_t                   blockcount,
+       xfs_exntst_t                    state)
+{
+       uint                            next_extent;
+       struct xfs_map_extent           *rmap;
+       int                             error;
+
+       /* XXX: actually finish the rmap update here */
+       error = -EFSCORRUPTED;
+
+       /*
+        * Mark the transaction dirty, even on error. This ensures the
+        * transaction is aborted, which:
+        *
+        * 1.) releases the RUI and frees the RUD
+        * 2.) shuts down the filesystem
+        */
+       tp->t_flags |= XFS_TRANS_DIRTY;
+       rudp->rud_item.li_desc->lid_flags |= XFS_LID_DIRTY;
+
+       next_extent = rudp->rud_next_extent;
+       ASSERT(next_extent < rudp->rud_format.rud_nextents);
+       rmap = &(rudp->rud_format.rud_extents[next_extent]);
+       rmap->me_owner = owner;
+       rmap->me_startblock = startblock;
+       rmap->me_startoff = startoff;
+       rmap->me_len = blockcount;
+       xfs_trans_set_rmap_flags(rmap, type, whichfork, state);
+       rudp->rud_next_extent++;
+
+       return error;
+}