]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/commitdiff
[XFS] implement generic xfs_btree_split
authorChristoph Hellwig <hch@infradead.org>
Thu, 30 Oct 2008 05:57:03 +0000 (16:57 +1100)
committerLachlan McIlroy <lachlan@sgi.com>
Thu, 30 Oct 2008 05:57:03 +0000 (16:57 +1100)
Make the btree split code generic. Based on a patch from David Chinner
with lots of changes to follow the original btree implementations more
closely. While this loses some of the generic helper routines for
inserting/moving/removing records it also solves some of the one off bugs
in the original code and makes it easier to verify.

SGI-PV: 985583

SGI-Modid: xfs-linux-melb:xfs-kern:32198a

Signed-off-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Lachlan McIlroy <lachlan@sgi.com>
Signed-off-by: Bill O'Donnell <billodo@sgi.com>
Signed-off-by: David Chinner <david@fromorbit.com>
fs/xfs/xfs_alloc_btree.c
fs/xfs/xfs_bmap_btree.c
fs/xfs/xfs_btree.c
fs/xfs/xfs_btree.h
fs/xfs/xfs_ialloc_btree.c

index 974a412ebc8a34d571f91e3f895cae0afccca12a..8a8d1aeec52abd4077719e5d35c89ecc1ca896f4 100644 (file)
@@ -35,6 +35,7 @@
 #include "xfs_dinode.h"
 #include "xfs_inode.h"
 #include "xfs_btree.h"
+#include "xfs_btree_trace.h"
 #include "xfs_ialloc.h"
 #include "xfs_alloc.h"
 #include "xfs_error.h"
@@ -48,8 +49,6 @@ STATIC void xfs_alloc_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_alloc_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_alloc_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC int xfs_alloc_newroot(xfs_btree_cur_t *, int *);
-STATIC int xfs_alloc_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
-               xfs_alloc_key_t *, xfs_btree_cur_t **, int *);
 
 /*
  * Internal functions.
@@ -695,15 +694,18 @@ xfs_alloc_insrec(
                        if (i)
                                optr = ptr = cur->bc_ptrs[level];
                        else {
+                               union xfs_btree_ptr bno = { .s = cpu_to_be32(nbno) };
                                /*
                                 * Next, try splitting the current block in
                                 * half. If this works we have to re-set our
                                 * variables because we could be in a
                                 * different block now.
                                 */
-                               if ((error = xfs_alloc_split(cur, level, &nbno,
-                                               &nkey, &ncur, &i)))
+                               if ((error = xfs_btree_split(cur, level, &bno,
+                                               (union xfs_btree_key *)&nkey,
+                                               &ncur, &i)))
                                        return error;
+                               nbno = be32_to_cpu(bno.s);
                                if (i) {
                                        bp = cur->bc_bufs[level];
                                        block = XFS_BUF_TO_ALLOC_BLOCK(bp);
@@ -1089,160 +1091,6 @@ xfs_alloc_newroot(
        return 0;
 }
 
-/*
- * Split cur/level block in half.
- * Return new block number and its first record (to be inserted into parent).
- */
-STATIC int                             /* error */
-xfs_alloc_split(
-       xfs_btree_cur_t         *cur,   /* btree cursor */
-       int                     level,  /* level to split */
-       xfs_agblock_t           *bnop,  /* output: block number allocated */
-       xfs_alloc_key_t         *keyp,  /* output: first key of new block */
-       xfs_btree_cur_t         **curp, /* output: new cursor */
-       int                     *stat)  /* success/failure */
-{
-       int                     error;  /* error return value */
-       int                     i;      /* loop index/record number */
-       xfs_agblock_t           lbno;   /* left (current) block number */
-       xfs_buf_t               *lbp;   /* buffer for left block */
-       xfs_alloc_block_t       *left;  /* left (current) btree block */
-       xfs_agblock_t           rbno;   /* right (new) block number */
-       xfs_buf_t               *rbp;   /* buffer for right block */
-       xfs_alloc_block_t       *right; /* right (new) btree block */
-
-       /*
-        * Allocate the new block from the freelist.
-        * If we can't do it, we're toast.  Give up.
-        */
-       error = xfs_alloc_get_freelist(cur->bc_tp,
-                                        cur->bc_private.a.agbp, &rbno, 1);
-       if (error)
-               return error;
-       if (rbno == NULLAGBLOCK) {
-               *stat = 0;
-               return 0;
-       }
-       xfs_trans_agbtree_delta(cur->bc_tp, 1);
-       rbp = xfs_btree_get_bufs(cur->bc_mp, cur->bc_tp, cur->bc_private.a.agno,
-               rbno, 0);
-       /*
-        * Set up the new block as "right".
-        */
-       right = XFS_BUF_TO_ALLOC_BLOCK(rbp);
-       /*
-        * "Left" is the current (according to the cursor) block.
-        */
-       lbp = cur->bc_bufs[level];
-       left = XFS_BUF_TO_ALLOC_BLOCK(lbp);
-#ifdef DEBUG
-       if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-               return error;
-#endif
-       /*
-        * Fill in the btree header for the new block.
-        */
-       right->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
-       right->bb_level = left->bb_level;
-       right->bb_numrecs = cpu_to_be16(be16_to_cpu(left->bb_numrecs) / 2);
-       /*
-        * Make sure that if there's an odd number of entries now, that
-        * each new block will have the same number of entries.
-        */
-       if ((be16_to_cpu(left->bb_numrecs) & 1) &&
-           cur->bc_ptrs[level] <= be16_to_cpu(right->bb_numrecs) + 1)
-               be16_add_cpu(&right->bb_numrecs, 1);
-       i = be16_to_cpu(left->bb_numrecs) - be16_to_cpu(right->bb_numrecs) + 1;
-       /*
-        * For non-leaf blocks, copy keys and addresses over to the new block.
-        */
-       if (level > 0) {
-               xfs_alloc_key_t *lkp;   /* left btree key pointer */
-               xfs_alloc_ptr_t *lpp;   /* left btree address pointer */
-               xfs_alloc_key_t *rkp;   /* right btree key pointer */
-               xfs_alloc_ptr_t *rpp;   /* right btree address pointer */
-
-               lkp = XFS_ALLOC_KEY_ADDR(left, i, cur);
-               lpp = XFS_ALLOC_PTR_ADDR(left, i, cur);
-               rkp = XFS_ALLOC_KEY_ADDR(right, 1, cur);
-               rpp = XFS_ALLOC_PTR_ADDR(right, 1, cur);
-#ifdef DEBUG
-               for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
-                       if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level)))
-                               return error;
-               }
-#endif
-               memcpy(rkp, lkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-               memcpy(rpp, lpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-               xfs_alloc_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               xfs_alloc_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               *keyp = *rkp;
-       }
-       /*
-        * For leaf blocks, copy records over to the new block.
-        */
-       else {
-               xfs_alloc_rec_t *lrp;   /* left btree record pointer */
-               xfs_alloc_rec_t *rrp;   /* right btree record pointer */
-
-               lrp = XFS_ALLOC_REC_ADDR(left, i, cur);
-               rrp = XFS_ALLOC_REC_ADDR(right, 1, cur);
-               memcpy(rrp, lrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-               xfs_alloc_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               keyp->ar_startblock = rrp->ar_startblock;
-               keyp->ar_blockcount = rrp->ar_blockcount;
-       }
-       /*
-        * Find the left block number by looking in the buffer.
-        * Adjust numrecs, sibling pointers.
-        */
-       lbno = XFS_DADDR_TO_AGBNO(cur->bc_mp, XFS_BUF_ADDR(lbp));
-       be16_add_cpu(&left->bb_numrecs, -(be16_to_cpu(right->bb_numrecs)));
-       right->bb_rightsib = left->bb_rightsib;
-       left->bb_rightsib = cpu_to_be32(rbno);
-       right->bb_leftsib = cpu_to_be32(lbno);
-       xfs_alloc_log_block(cur->bc_tp, rbp, XFS_BB_ALL_BITS);
-       xfs_alloc_log_block(cur->bc_tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
-       /*
-        * If there's a block to the new block's right, make that block
-        * point back to right instead of to left.
-        */
-       if (be32_to_cpu(right->bb_rightsib) != NULLAGBLOCK) {
-               xfs_alloc_block_t       *rrblock;       /* rr btree block */
-               xfs_buf_t               *rrbp;          /* buffer for rrblock */
-
-               if ((error = xfs_btree_read_bufs(cur->bc_mp, cur->bc_tp,
-                               cur->bc_private.a.agno, be32_to_cpu(right->bb_rightsib), 0,
-                               &rrbp, XFS_ALLOC_BTREE_REF)))
-                       return error;
-               rrblock = XFS_BUF_TO_ALLOC_BLOCK(rrbp);
-               if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
-                       return error;
-               rrblock->bb_leftsib = cpu_to_be32(rbno);
-               xfs_alloc_log_block(cur->bc_tp, rrbp, XFS_BB_LEFTSIB);
-       }
-       /*
-        * If the cursor is really in the right block, move it there.
-        * If it's just pointing past the last entry in left, then we'll
-        * insert there, so don't change anything in that case.
-        */
-       if (cur->bc_ptrs[level] > be16_to_cpu(left->bb_numrecs) + 1) {
-               xfs_btree_setbuf(cur, level, rbp);
-               cur->bc_ptrs[level] -= be16_to_cpu(left->bb_numrecs);
-       }
-       /*
-        * If there are more levels, we'll need another cursor which refers to
-        * the right block, no matter where this cursor was.
-        */
-       if (level + 1 < cur->bc_nlevels) {
-               if ((error = xfs_btree_dup_cursor(cur, curp)))
-                       return error;
-               (*curp)->bc_ptrs[level + 1]++;
-       }
-       *bnop = rbno;
-       *stat = 1;
-       return 0;
-}
 
 /*
  * Externally visible routines.
@@ -1396,6 +1244,41 @@ xfs_allocbt_dup_cursor(
                        cur->bc_btnum);
 }
 
+STATIC int
+xfs_allocbt_alloc_block(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *start,
+       union xfs_btree_ptr     *new,
+       int                     length,
+       int                     *stat)
+{
+       int                     error;
+       xfs_agblock_t           bno;
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+
+       /* Allocate the new block from the freelist. If we can't, give up.  */
+       error = xfs_alloc_get_freelist(cur->bc_tp, cur->bc_private.a.agbp,
+                                      &bno, 1);
+       if (error) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+               return error;
+       }
+
+       if (bno == NULLAGBLOCK) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+               *stat = 0;
+               return 0;
+       }
+
+       xfs_trans_agbtree_delta(cur->bc_tp, 1);
+       new->s = cpu_to_be32(bno);
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+       *stat = 1;
+       return 0;
+}
+
 /*
  * Update the longest extent in the AGF
  */
@@ -1557,6 +1440,7 @@ static const struct xfs_btree_ops xfs_allocbt_ops = {
        .key_len                = sizeof(xfs_alloc_key_t),
 
        .dup_cursor             = xfs_allocbt_dup_cursor,
+       .alloc_block            = xfs_allocbt_alloc_block,
        .update_lastrec         = xfs_allocbt_update_lastrec,
        .get_maxrecs            = xfs_allocbt_get_maxrecs,
        .init_key_from_rec      = xfs_allocbt_init_key_from_rec,
index 809bd6ee177ed1a148aef58b915ac9489a633df1..e7539263457fb429122a32d65398b5aa7184e431 100644 (file)
@@ -52,8 +52,6 @@
 STATIC int xfs_bmbt_killroot(xfs_btree_cur_t *);
 STATIC void xfs_bmbt_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_bmbt_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
-STATIC int xfs_bmbt_split(xfs_btree_cur_t *, int, xfs_fsblock_t *,
-               __uint64_t *, xfs_btree_cur_t **, int *);
 
 #undef EXIT
 
@@ -550,13 +548,17 @@ xfs_bmbt_insrec(
                                if (i) {
                                        optr = ptr = cur->bc_ptrs[level];
                                } else {
-                                       if ((error = xfs_bmbt_split(cur, level,
-                                                       &nbno, &startoff, &ncur,
+                                       union xfs_btree_ptr bno = { .l = cpu_to_be64(nbno) };
+                                       union xfs_btree_key skey;
+                                       if ((error = xfs_btree_split(cur, level,
+                                                       &bno, &skey, &ncur,
                                                        &i))) {
                                                XFS_BMBT_TRACE_CURSOR(cur,
                                                        ERROR);
                                                return error;
                                        }
+                                       nbno = be64_to_cpu(bno.l);
+                                       startoff = be64_to_cpu(skey.bmbt.br_startoff);
                                        if (i) {
                                                block = xfs_bmbt_get_block(
                                                            cur, level, &bp);
@@ -825,184 +827,6 @@ xfs_extent_state(
        return XFS_EXT_NORM;
 }
 
-
-/*
- * Split cur/level block in half.
- * Return new block number and its first record (to be inserted into parent).
- */
-STATIC int                                     /* error */
-xfs_bmbt_split(
-       xfs_btree_cur_t         *cur,
-       int                     level,
-       xfs_fsblock_t           *bnop,
-       __uint64_t              *startoff,
-       xfs_btree_cur_t         **curp,
-       int                     *stat)          /* success/failure */
-{
-       xfs_alloc_arg_t         args;           /* block allocation args */
-       int                     error;          /* error return value */
-       int                     i;              /* loop counter */
-       xfs_fsblock_t           lbno;           /* left sibling block number */
-       xfs_buf_t               *lbp;           /* left buffer pointer */
-       xfs_bmbt_block_t        *left;          /* left btree block */
-       xfs_bmbt_key_t          *lkp;           /* left btree key */
-       xfs_bmbt_ptr_t          *lpp;           /* left address pointer */
-       xfs_bmbt_rec_t          *lrp;           /* left record pointer */
-       xfs_buf_t               *rbp;           /* right buffer pointer */
-       xfs_bmbt_block_t        *right;         /* right btree block */
-       xfs_bmbt_key_t          *rkp;           /* right btree key */
-       xfs_bmbt_ptr_t          *rpp;           /* right address pointer */
-       xfs_bmbt_block_t        *rrblock;       /* right-right btree block */
-       xfs_buf_t               *rrbp;          /* right-right buffer pointer */
-       xfs_bmbt_rec_t          *rrp;           /* right record pointer */
-
-       XFS_BMBT_TRACE_CURSOR(cur, ENTRY);
-       // disable until merged into common code
-//     XFS_BMBT_TRACE_ARGIFK(cur, level, *bnop, *startoff);
-       args.tp = cur->bc_tp;
-       args.mp = cur->bc_mp;
-       lbp = cur->bc_bufs[level];
-       lbno = XFS_DADDR_TO_FSB(args.mp, XFS_BUF_ADDR(lbp));
-       left = XFS_BUF_TO_BMBT_BLOCK(lbp);
-       args.fsbno = cur->bc_private.b.firstblock;
-       args.firstblock = args.fsbno;
-       args.minleft = 0;
-       if (args.fsbno == NULLFSBLOCK) {
-               args.fsbno = lbno;
-               args.type = XFS_ALLOCTYPE_START_BNO;
-               /*
-                * Make sure there is sufficient room left in the AG to
-                * complete a full tree split for an extent insert.  If
-                * we are converting the middle part of an extent then
-                * we may need space for two tree splits.
-                *
-                * We are relying on the caller to make the correct block
-                * reservation for this operation to succeed.  If the
-                * reservation amount is insufficient then we may fail a
-                * block allocation here and corrupt the filesystem.
-                */
-               args.minleft = xfs_trans_get_block_res(args.tp);
-       } else if (cur->bc_private.b.flist->xbf_low)
-               args.type = XFS_ALLOCTYPE_START_BNO;
-       else
-               args.type = XFS_ALLOCTYPE_NEAR_BNO;
-       args.mod = args.alignment = args.total = args.isfl =
-               args.userdata = args.minalignslop = 0;
-       args.minlen = args.maxlen = args.prod = 1;
-       args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
-       if (!args.wasdel && xfs_trans_get_block_res(args.tp) == 0) {
-               XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-               return XFS_ERROR(ENOSPC);
-       }
-       if ((error = xfs_alloc_vextent(&args))) {
-               XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-               return error;
-       }
-       if (args.fsbno == NULLFSBLOCK && args.minleft) {
-               /*
-                * Could not find an AG with enough free space to satisfy
-                * a full btree split.  Try again without minleft and if
-                * successful activate the lowspace algorithm.
-                */
-               args.fsbno = 0;
-               args.type = XFS_ALLOCTYPE_FIRST_AG;
-               args.minleft = 0;
-               if ((error = xfs_alloc_vextent(&args))) {
-                       XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-                       return error;
-               }
-               cur->bc_private.b.flist->xbf_low = 1;
-       }
-       if (args.fsbno == NULLFSBLOCK) {
-               XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-               *stat = 0;
-               return 0;
-       }
-       ASSERT(args.len == 1);
-       cur->bc_private.b.firstblock = args.fsbno;
-       cur->bc_private.b.allocated++;
-       cur->bc_private.b.ip->i_d.di_nblocks++;
-       xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
-       XFS_TRANS_MOD_DQUOT_BYINO(args.mp, args.tp, cur->bc_private.b.ip,
-                       XFS_TRANS_DQ_BCOUNT, 1L);
-       rbp = xfs_btree_get_bufl(args.mp, args.tp, args.fsbno, 0);
-       right = XFS_BUF_TO_BMBT_BLOCK(rbp);
-#ifdef DEBUG
-       if ((error = xfs_btree_check_lblock(cur, left, level, rbp))) {
-               XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-               return error;
-       }
-#endif
-       right->bb_magic = cpu_to_be32(XFS_BMAP_MAGIC);
-       right->bb_level = left->bb_level;
-       right->bb_numrecs = cpu_to_be16(be16_to_cpu(left->bb_numrecs) / 2);
-       if ((be16_to_cpu(left->bb_numrecs) & 1) &&
-           cur->bc_ptrs[level] <= be16_to_cpu(right->bb_numrecs) + 1)
-               be16_add_cpu(&right->bb_numrecs, 1);
-       i = be16_to_cpu(left->bb_numrecs) - be16_to_cpu(right->bb_numrecs) + 1;
-       if (level > 0) {
-               lkp = XFS_BMAP_KEY_IADDR(left, i, cur);
-               lpp = XFS_BMAP_PTR_IADDR(left, i, cur);
-               rkp = XFS_BMAP_KEY_IADDR(right, 1, cur);
-               rpp = XFS_BMAP_PTR_IADDR(right, 1, cur);
-#ifdef DEBUG
-               for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
-                       if ((error = xfs_btree_check_lptr_disk(cur, lpp[i], level))) {
-                               XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-                               return error;
-                       }
-               }
-#endif
-               memcpy(rkp, lkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-               memcpy(rpp, lpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-               xfs_bmbt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               xfs_bmbt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               *startoff = be64_to_cpu(rkp->br_startoff);
-       } else {
-               lrp = XFS_BMAP_REC_IADDR(left, i, cur);
-               rrp = XFS_BMAP_REC_IADDR(right, 1, cur);
-               memcpy(rrp, lrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-               xfs_bmbt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               *startoff = xfs_bmbt_disk_get_startoff(rrp);
-       }
-       be16_add_cpu(&left->bb_numrecs, -(be16_to_cpu(right->bb_numrecs)));
-       right->bb_rightsib = left->bb_rightsib;
-       left->bb_rightsib = cpu_to_be64(args.fsbno);
-       right->bb_leftsib = cpu_to_be64(lbno);
-       xfs_bmbt_log_block(cur, rbp, XFS_BB_ALL_BITS);
-       xfs_bmbt_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
-       if (be64_to_cpu(right->bb_rightsib) != NULLDFSBNO) {
-               if ((error = xfs_btree_read_bufl(args.mp, args.tp,
-                               be64_to_cpu(right->bb_rightsib), 0, &rrbp,
-                               XFS_BMAP_BTREE_REF))) {
-                       XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-                       return error;
-               }
-               rrblock = XFS_BUF_TO_BMBT_BLOCK(rrbp);
-               if ((error = xfs_btree_check_lblock(cur, rrblock, level, rrbp))) {
-                       XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-                       return error;
-               }
-               rrblock->bb_leftsib = cpu_to_be64(args.fsbno);
-               xfs_bmbt_log_block(cur, rrbp, XFS_BB_LEFTSIB);
-       }
-       if (cur->bc_ptrs[level] > be16_to_cpu(left->bb_numrecs) + 1) {
-               xfs_btree_setbuf(cur, level, rbp);
-               cur->bc_ptrs[level] -= be16_to_cpu(left->bb_numrecs);
-       }
-       if (level + 1 < cur->bc_nlevels) {
-               if ((error = xfs_btree_dup_cursor(cur, curp))) {
-                       XFS_BMBT_TRACE_CURSOR(cur, ERROR);
-                       return error;
-               }
-               (*curp)->bc_ptrs[level + 1]++;
-       }
-       *bnop = args.fsbno;
-       XFS_BMBT_TRACE_CURSOR(cur, EXIT);
-       *stat = 1;
-       return 0;
-}
-
 /*
  * Convert on-disk form of btree root to in-memory form.
  */
@@ -1737,6 +1561,92 @@ xfs_bmbt_dup_cursor(
        return new;
 }
 
+STATIC int
+xfs_bmbt_alloc_block(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *start,
+       union xfs_btree_ptr     *new,
+       int                     length,
+       int                     *stat)
+{
+       xfs_alloc_arg_t         args;           /* block allocation args */
+       int                     error;          /* error return value */
+
+       memset(&args, 0, sizeof(args));
+       args.tp = cur->bc_tp;
+       args.mp = cur->bc_mp;
+       args.fsbno = cur->bc_private.b.firstblock;
+       args.firstblock = args.fsbno;
+
+       if (args.fsbno == NULLFSBLOCK) {
+               args.fsbno = be64_to_cpu(start->l);
+               args.type = XFS_ALLOCTYPE_START_BNO;
+               /*
+                * Make sure there is sufficient room left in the AG to
+                * complete a full tree split for an extent insert.  If
+                * we are converting the middle part of an extent then
+                * we may need space for two tree splits.
+                *
+                * We are relying on the caller to make the correct block
+                * reservation for this operation to succeed.  If the
+                * reservation amount is insufficient then we may fail a
+                * block allocation here and corrupt the filesystem.
+                */
+               args.minleft = xfs_trans_get_block_res(args.tp);
+       } else if (cur->bc_private.b.flist->xbf_low) {
+               args.type = XFS_ALLOCTYPE_START_BNO;
+       } else {
+               args.type = XFS_ALLOCTYPE_NEAR_BNO;
+       }
+
+       args.minlen = args.maxlen = args.prod = 1;
+       args.wasdel = cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL;
+       if (!args.wasdel && xfs_trans_get_block_res(args.tp) == 0) {
+               error = XFS_ERROR(ENOSPC);
+               goto error0;
+       }
+       error = xfs_alloc_vextent(&args);
+       if (error)
+               goto error0;
+
+       if (args.fsbno == NULLFSBLOCK && args.minleft) {
+               /*
+                * Could not find an AG with enough free space to satisfy
+                * a full btree split.  Try again without minleft and if
+                * successful activate the lowspace algorithm.
+                */
+               args.fsbno = 0;
+               args.type = XFS_ALLOCTYPE_FIRST_AG;
+               args.minleft = 0;
+               error = xfs_alloc_vextent(&args);
+               if (error)
+                       goto error0;
+               cur->bc_private.b.flist->xbf_low = 1;
+       }
+       if (args.fsbno == NULLFSBLOCK) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+               *stat = 0;
+               return 0;
+       }
+       ASSERT(args.len == 1);
+       cur->bc_private.b.firstblock = args.fsbno;
+       cur->bc_private.b.allocated++;
+       cur->bc_private.b.ip->i_d.di_nblocks++;
+       xfs_trans_log_inode(args.tp, cur->bc_private.b.ip, XFS_ILOG_CORE);
+       XFS_TRANS_MOD_DQUOT_BYINO(args.mp, args.tp, cur->bc_private.b.ip,
+                       XFS_TRANS_DQ_BCOUNT, 1L);
+
+       new->l = cpu_to_be64(args.fsbno);
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+       *stat = 1;
+       return 0;
+
+ error0:
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+       return error;
+}
+
 STATIC int
 xfs_bmbt_get_maxrecs(
        struct xfs_btree_cur    *cur,
@@ -1861,6 +1771,7 @@ static const struct xfs_btree_ops xfs_bmbt_ops = {
        .key_len                = sizeof(xfs_bmbt_key_t),
 
        .dup_cursor             = xfs_bmbt_dup_cursor,
+       .alloc_block            = xfs_bmbt_alloc_block,
        .get_maxrecs            = xfs_bmbt_get_maxrecs,
        .init_key_from_rec      = xfs_bmbt_init_key_from_rec,
        .init_ptr_from_cur      = xfs_bmbt_init_ptr_from_cur,
index 2b0d1422c4c6f621bb8d9bd0ddaf77391781199d..80576695fbe5bc817d0edef8babb0f740f5a19b8 100644 (file)
@@ -988,6 +988,48 @@ xfs_btree_get_sibling(
        }
 }
 
+STATIC void
+xfs_btree_set_sibling(
+       struct xfs_btree_cur    *cur,
+       struct xfs_btree_block  *block,
+       union xfs_btree_ptr     *ptr,
+       int                     lr)
+{
+       ASSERT(lr == XFS_BB_LEFTSIB || lr == XFS_BB_RIGHTSIB);
+
+       if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
+               if (lr == XFS_BB_RIGHTSIB)
+                       block->bb_u.l.bb_rightsib = ptr->l;
+               else
+                       block->bb_u.l.bb_leftsib = ptr->l;
+       } else {
+               if (lr == XFS_BB_RIGHTSIB)
+                       block->bb_u.s.bb_rightsib = ptr->s;
+               else
+                       block->bb_u.s.bb_leftsib = ptr->s;
+       }
+}
+
+STATIC void
+xfs_btree_init_block(
+       struct xfs_btree_cur    *cur,
+       int                     level,
+       int                     numrecs,
+       struct xfs_btree_block  *new)   /* new block */
+{
+       new->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
+       new->bb_level = cpu_to_be16(level);
+       new->bb_numrecs = cpu_to_be16(numrecs);
+
+       if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
+               new->bb_u.l.bb_leftsib = cpu_to_be64(NULLFSBLOCK);
+               new->bb_u.l.bb_rightsib = cpu_to_be64(NULLFSBLOCK);
+       } else {
+               new->bb_u.s.bb_leftsib = cpu_to_be32(NULLAGBLOCK);
+               new->bb_u.s.bb_rightsib = cpu_to_be32(NULLAGBLOCK);
+       }
+}
+
 /*
  * Return true if ptr is the last record in the btree and
  * we need to track updateÑ• to this record.  The decision
@@ -1012,6 +1054,21 @@ xfs_btree_is_lastrec(
        return 1;
 }
 
+STATIC void
+xfs_btree_buf_to_ptr(
+       struct xfs_btree_cur    *cur,
+       struct xfs_buf          *bp,
+       union xfs_btree_ptr     *ptr)
+{
+       if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
+               ptr->l = cpu_to_be64(XFS_DADDR_TO_FSB(cur->bc_mp,
+                                       XFS_BUF_ADDR(bp)));
+       else {
+               ptr->s = cpu_to_be32(XFS_DADDR_TO_AGBNO(cur->bc_mp,
+                                       XFS_BUF_ADDR(bp)));
+       }
+}
+
 STATIC xfs_daddr_t
 xfs_btree_ptr_to_daddr(
        struct xfs_btree_cur    *cur,
@@ -1051,6 +1108,31 @@ xfs_btree_set_refs(
        }
 }
 
+STATIC int
+xfs_btree_get_buf_block(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *ptr,
+       int                     flags,
+       struct xfs_btree_block  **block,
+       struct xfs_buf          **bpp)
+{
+       struct xfs_mount        *mp = cur->bc_mp;
+       xfs_daddr_t             d;
+
+       /* need to sort out how callers deal with failures first */
+       ASSERT(!(flags & XFS_BUF_TRYLOCK));
+
+       d = xfs_btree_ptr_to_daddr(cur, ptr);
+       *bpp = xfs_trans_get_buf(cur->bc_tp, mp->m_ddev_targp, d,
+                                mp->m_bsize, flags);
+
+       ASSERT(*bpp);
+       ASSERT(!XFS_BUF_GETERROR(*bpp));
+
+       *block = XFS_BUF_TO_BLOCK(*bpp);
+       return 0;
+}
+
 /*
  * Read in the buffer at the given ptr and return the buffer and
  * the block pointer within the buffer.
@@ -2199,3 +2281,189 @@ error1:
        xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
        return error;
 }
+
+/*
+ * Split cur/level block in half.
+ * Return new block number and the key to its first
+ * record (to be inserted into parent).
+ */
+int                                            /* error */
+xfs_btree_split(
+       struct xfs_btree_cur    *cur,
+       int                     level,
+       union xfs_btree_ptr     *ptrp,
+       union xfs_btree_key     *key,
+       struct xfs_btree_cur    **curp,
+       int                     *stat)          /* success/failure */
+{
+       union xfs_btree_ptr     lptr;           /* left sibling block ptr */
+       struct xfs_buf          *lbp;           /* left buffer pointer */
+       struct xfs_btree_block  *left;          /* left btree block */
+       union xfs_btree_ptr     rptr;           /* right sibling block ptr */
+       struct xfs_buf          *rbp;           /* right buffer pointer */
+       struct xfs_btree_block  *right;         /* right btree block */
+       union xfs_btree_ptr     rrptr;          /* right-right sibling ptr */
+       struct xfs_buf          *rrbp;          /* right-right buffer pointer */
+       struct xfs_btree_block  *rrblock;       /* right-right btree block */
+       int                     lrecs;
+       int                     rrecs;
+       int                     src_index;
+       int                     error;          /* error return value */
+#ifdef DEBUG
+       int                     i;
+#endif
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+       XFS_BTREE_TRACE_ARGIPK(cur, level, *ptrp, key);
+
+       XFS_BTREE_STATS_INC(cur, split);
+
+       /* Set up left block (current one). */
+       left = xfs_btree_get_block(cur, level, &lbp);
+
+#ifdef DEBUG
+       error = xfs_btree_check_block(cur, left, level, lbp);
+       if (error)
+               goto error0;
+#endif
+
+       xfs_btree_buf_to_ptr(cur, lbp, &lptr);
+
+       /* Allocate the new block. If we can't do it, we're toast. Give up. */
+       error = cur->bc_ops->alloc_block(cur, &lptr, &rptr, 1, stat);
+       if (error)
+               goto error0;
+       if (*stat == 0)
+               goto out0;
+       XFS_BTREE_STATS_INC(cur, alloc);
+
+       /* Set up the new block as "right". */
+       error = xfs_btree_get_buf_block(cur, &rptr, 0, &right, &rbp);
+       if (error)
+               goto error0;
+
+       /* Fill in the btree header for the new right block. */
+       xfs_btree_init_block(cur, xfs_btree_get_level(left), 0, right);
+
+       /*
+        * Split the entries between the old and the new block evenly.
+        * Make sure that if there's an odd number of entries now, that
+        * each new block will have the same number of entries.
+        */
+       lrecs = xfs_btree_get_numrecs(left);
+       rrecs = lrecs / 2;
+       if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
+               rrecs++;
+       src_index = (lrecs - rrecs + 1);
+
+       XFS_BTREE_STATS_ADD(cur, moves, rrecs);
+
+       /*
+        * Copy btree block entries from the left block over to the
+        * new block, the right. Update the right block and log the
+        * changes.
+        */
+       if (level > 0) {
+               /* It's a non-leaf.  Move keys and pointers. */
+               union xfs_btree_key     *lkp;   /* left btree key */
+               union xfs_btree_ptr     *lpp;   /* left address pointer */
+               union xfs_btree_key     *rkp;   /* right btree key */
+               union xfs_btree_ptr     *rpp;   /* right address pointer */
+
+               lkp = xfs_btree_key_addr(cur, src_index, left);
+               lpp = xfs_btree_ptr_addr(cur, src_index, left);
+               rkp = xfs_btree_key_addr(cur, 1, right);
+               rpp = xfs_btree_ptr_addr(cur, 1, right);
+
+#ifdef DEBUG
+               for (i = src_index; i < rrecs; i++) {
+                       error = xfs_btree_check_ptr(cur, lpp, i, level);
+                       if (error)
+                               goto error0;
+               }
+#endif
+
+               xfs_btree_copy_keys(cur, rkp, lkp, rrecs);
+               xfs_btree_copy_ptrs(cur, rpp, lpp, rrecs);
+
+               xfs_btree_log_keys(cur, rbp, 1, rrecs);
+               xfs_btree_log_ptrs(cur, rbp, 1, rrecs);
+
+               /* Grab the keys to the entries moved to the right block */
+               xfs_btree_copy_keys(cur, key, rkp, 1);
+       } else {
+               /* It's a leaf.  Move records.  */
+               union xfs_btree_rec     *lrp;   /* left record pointer */
+               union xfs_btree_rec     *rrp;   /* right record pointer */
+
+               lrp = xfs_btree_rec_addr(cur, src_index, left);
+               rrp = xfs_btree_rec_addr(cur, 1, right);
+
+               xfs_btree_copy_recs(cur, rrp, lrp, rrecs);
+               xfs_btree_log_recs(cur, rbp, 1, rrecs);
+
+               cur->bc_ops->init_key_from_rec(key,
+                       xfs_btree_rec_addr(cur, 1, right));
+       }
+
+
+       /*
+        * Find the left block number by looking in the buffer.
+        * Adjust numrecs, sibling pointers.
+        */
+       xfs_btree_get_sibling(cur, left, &rrptr, XFS_BB_RIGHTSIB);
+       xfs_btree_set_sibling(cur, right, &rrptr, XFS_BB_RIGHTSIB);
+       xfs_btree_set_sibling(cur, right, &lptr, XFS_BB_LEFTSIB);
+       xfs_btree_set_sibling(cur, left, &rptr, XFS_BB_RIGHTSIB);
+
+       lrecs -= rrecs;
+       xfs_btree_set_numrecs(left, lrecs);
+       xfs_btree_set_numrecs(right, xfs_btree_get_numrecs(right) + rrecs);
+
+       xfs_btree_log_block(cur, rbp, XFS_BB_ALL_BITS);
+       xfs_btree_log_block(cur, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
+
+       /*
+        * If there's a block to the new block's right, make that block
+        * point back to right instead of to left.
+        */
+       if (!xfs_btree_ptr_is_null(cur, &rrptr)) {
+               error = xfs_btree_read_buf_block(cur, &rrptr, level,
+                                                       0, &rrblock, &rrbp);
+               if (error)
+                       goto error0;
+               xfs_btree_set_sibling(cur, rrblock, &rptr, XFS_BB_LEFTSIB);
+               xfs_btree_log_block(cur, rrbp, XFS_BB_LEFTSIB);
+       }
+       /*
+        * If the cursor is really in the right block, move it there.
+        * If it's just pointing past the last entry in left, then we'll
+        * insert there, so don't change anything in that case.
+        */
+       if (cur->bc_ptrs[level] > lrecs + 1) {
+               xfs_btree_setbuf(cur, level, rbp);
+               cur->bc_ptrs[level] -= lrecs;
+       }
+       /*
+        * If there are more levels, we'll need another cursor which refers
+        * the right block, no matter where this cursor was.
+        */
+       if (level + 1 < cur->bc_nlevels) {
+               error = xfs_btree_dup_cursor(cur, curp);
+               if (error)
+                       goto error0;
+               (*curp)->bc_ptrs[level + 1]++;
+       }
+       *ptrp = rptr;
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+       *stat = 1;
+       return 0;
+out0:
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+       *stat = 0;
+       return 0;
+
+error0:
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+       return error;
+}
index 7cde287b5c9c009205f6ce4178675c3e801b361a..354a6656fad5d7cdc766c3f18a58c465f84a26ec 100644 (file)
@@ -187,6 +187,12 @@ struct xfs_btree_ops {
        /* cursor operations */
        struct xfs_btree_cur *(*dup_cursor)(struct xfs_btree_cur *);
 
+       /* block allocation / freeing */
+       int     (*alloc_block)(struct xfs_btree_cur *cur,
+                              union xfs_btree_ptr *start_bno,
+                              union xfs_btree_ptr *new_bno,
+                              int length, int *stat);
+
        /* update last record information */
        void    (*update_lastrec)(struct xfs_btree_cur *cur,
                                  struct xfs_btree_block *block,
@@ -535,6 +541,8 @@ int xfs_btree_updkey(struct xfs_btree_cur *, union xfs_btree_key *, int);
 int xfs_btree_update(struct xfs_btree_cur *, union xfs_btree_rec *);
 int xfs_btree_lshift(struct xfs_btree_cur *, int, int *);
 int xfs_btree_rshift(struct xfs_btree_cur *, int, int *);
+int xfs_btree_split(struct xfs_btree_cur *, int, union xfs_btree_ptr *,
+               union xfs_btree_key *, struct xfs_btree_cur **, int *);
 
 /*
  * Helpers.
index 60f5db5d6dfa1d32877fd17453dbef4efca95498..c76190a83e4e1bd854a14055682671bb26db8ac5 100644 (file)
@@ -35,6 +35,7 @@
 #include "xfs_dinode.h"
 #include "xfs_inode.h"
 #include "xfs_btree.h"
+#include "xfs_btree_trace.h"
 #include "xfs_ialloc.h"
 #include "xfs_alloc.h"
 #include "xfs_error.h"
@@ -44,8 +45,6 @@ STATIC void xfs_inobt_log_keys(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_inobt_log_ptrs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC void xfs_inobt_log_recs(xfs_btree_cur_t *, xfs_buf_t *, int, int);
 STATIC int xfs_inobt_newroot(xfs_btree_cur_t *, int *);
-STATIC int xfs_inobt_split(xfs_btree_cur_t *, int, xfs_agblock_t *,
-               xfs_inobt_key_t *, xfs_btree_cur_t **, int *);
 
 /*
  * Single level of the xfs_inobt_delete record deletion routine.
@@ -620,15 +619,18 @@ xfs_inobt_insrec(
                        if (i) {
                                optr = ptr = cur->bc_ptrs[level];
                        } else {
+                               union xfs_btree_ptr bno = { .s = cpu_to_be32(nbno) };
                                /*
                                 * Next, try splitting the current block
                                 * in half. If this works we have to
                                 * re-set our variables because
                                 * we could be in a different block now.
                                 */
-                               if ((error = xfs_inobt_split(cur, level, &nbno,
-                                               &nkey, &ncur, &i)))
+                               if ((error = xfs_btree_split(cur, level, &bno,
+                                               (union xfs_btree_key *)&nkey,
+                                               &ncur, &i)))
                                        return error;
+                               nbno = be32_to_cpu(bno.s);
                                if (i) {
                                        bp = cur->bc_bufs[level];
                                        block = XFS_BUF_TO_INOBT_BLOCK(bp);
@@ -972,165 +974,6 @@ xfs_inobt_newroot(
        return 0;
 }
 
-/*
- * Split cur/level block in half.
- * Return new block number and its first record (to be inserted into parent).
- */
-STATIC int                             /* error */
-xfs_inobt_split(
-       xfs_btree_cur_t         *cur,   /* btree cursor */
-       int                     level,  /* level to split */
-       xfs_agblock_t           *bnop,  /* output: block number allocated */
-       xfs_inobt_key_t         *keyp,  /* output: first key of new block */
-       xfs_btree_cur_t         **curp, /* output: new cursor */
-       int                     *stat)  /* success/failure */
-{
-       xfs_alloc_arg_t         args;   /* allocation argument structure */
-       int                     error;  /* error return value */
-       int                     i;      /* loop index/record number */
-       xfs_agblock_t           lbno;   /* left (current) block number */
-       xfs_buf_t               *lbp;   /* buffer for left block */
-       xfs_inobt_block_t       *left;  /* left (current) btree block */
-       xfs_inobt_key_t         *lkp;   /* left btree key pointer */
-       xfs_inobt_ptr_t         *lpp;   /* left btree address pointer */
-       xfs_inobt_rec_t         *lrp;   /* left btree record pointer */
-       xfs_buf_t               *rbp;   /* buffer for right block */
-       xfs_inobt_block_t       *right; /* right (new) btree block */
-       xfs_inobt_key_t         *rkp;   /* right btree key pointer */
-       xfs_inobt_ptr_t         *rpp;   /* right btree address pointer */
-       xfs_inobt_rec_t         *rrp;   /* right btree record pointer */
-
-       /*
-        * Set up left block (current one).
-        */
-       lbp = cur->bc_bufs[level];
-       args.tp = cur->bc_tp;
-       args.mp = cur->bc_mp;
-       lbno = XFS_DADDR_TO_AGBNO(args.mp, XFS_BUF_ADDR(lbp));
-       /*
-        * Allocate the new block.
-        * If we can't do it, we're toast.  Give up.
-        */
-       args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_private.a.agno, lbno);
-       args.mod = args.minleft = args.alignment = args.total = args.wasdel =
-               args.isfl = args.userdata = args.minalignslop = 0;
-       args.minlen = args.maxlen = args.prod = 1;
-       args.type = XFS_ALLOCTYPE_NEAR_BNO;
-       if ((error = xfs_alloc_vextent(&args)))
-               return error;
-       if (args.fsbno == NULLFSBLOCK) {
-               *stat = 0;
-               return 0;
-       }
-       ASSERT(args.len == 1);
-       rbp = xfs_btree_get_bufs(args.mp, args.tp, args.agno, args.agbno, 0);
-       /*
-        * Set up the new block as "right".
-        */
-       right = XFS_BUF_TO_INOBT_BLOCK(rbp);
-       /*
-        * "Left" is the current (according to the cursor) block.
-        */
-       left = XFS_BUF_TO_INOBT_BLOCK(lbp);
-#ifdef DEBUG
-       if ((error = xfs_btree_check_sblock(cur, left, level, lbp)))
-               return error;
-#endif
-       /*
-        * Fill in the btree header for the new block.
-        */
-       right->bb_magic = cpu_to_be32(xfs_magics[cur->bc_btnum]);
-       right->bb_level = left->bb_level;
-       right->bb_numrecs = cpu_to_be16(be16_to_cpu(left->bb_numrecs) / 2);
-       /*
-        * Make sure that if there's an odd number of entries now, that
-        * each new block will have the same number of entries.
-        */
-       if ((be16_to_cpu(left->bb_numrecs) & 1) &&
-           cur->bc_ptrs[level] <= be16_to_cpu(right->bb_numrecs) + 1)
-               be16_add_cpu(&right->bb_numrecs, 1);
-       i = be16_to_cpu(left->bb_numrecs) - be16_to_cpu(right->bb_numrecs) + 1;
-       /*
-        * For non-leaf blocks, copy keys and addresses over to the new block.
-        */
-       if (level > 0) {
-               lkp = XFS_INOBT_KEY_ADDR(left, i, cur);
-               lpp = XFS_INOBT_PTR_ADDR(left, i, cur);
-               rkp = XFS_INOBT_KEY_ADDR(right, 1, cur);
-               rpp = XFS_INOBT_PTR_ADDR(right, 1, cur);
-#ifdef DEBUG
-               for (i = 0; i < be16_to_cpu(right->bb_numrecs); i++) {
-                       if ((error = xfs_btree_check_sptr(cur, be32_to_cpu(lpp[i]), level)))
-                               return error;
-               }
-#endif
-               memcpy(rkp, lkp, be16_to_cpu(right->bb_numrecs) * sizeof(*rkp));
-               memcpy(rpp, lpp, be16_to_cpu(right->bb_numrecs) * sizeof(*rpp));
-               xfs_inobt_log_keys(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               xfs_inobt_log_ptrs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               *keyp = *rkp;
-       }
-       /*
-        * For leaf blocks, copy records over to the new block.
-        */
-       else {
-               lrp = XFS_INOBT_REC_ADDR(left, i, cur);
-               rrp = XFS_INOBT_REC_ADDR(right, 1, cur);
-               memcpy(rrp, lrp, be16_to_cpu(right->bb_numrecs) * sizeof(*rrp));
-               xfs_inobt_log_recs(cur, rbp, 1, be16_to_cpu(right->bb_numrecs));
-               keyp->ir_startino = rrp->ir_startino;
-       }
-       /*
-        * Find the left block number by looking in the buffer.
-        * Adjust numrecs, sibling pointers.
-        */
-       be16_add_cpu(&left->bb_numrecs, -(be16_to_cpu(right->bb_numrecs)));
-       right->bb_rightsib = left->bb_rightsib;
-       left->bb_rightsib = cpu_to_be32(args.agbno);
-       right->bb_leftsib = cpu_to_be32(lbno);
-       xfs_inobt_log_block(args.tp, rbp, XFS_BB_ALL_BITS);
-       xfs_inobt_log_block(args.tp, lbp, XFS_BB_NUMRECS | XFS_BB_RIGHTSIB);
-       /*
-        * If there's a block to the new block's right, make that block
-        * point back to right instead of to left.
-        */
-       if (be32_to_cpu(right->bb_rightsib) != NULLAGBLOCK) {
-               xfs_inobt_block_t       *rrblock;       /* rr btree block */
-               xfs_buf_t               *rrbp;          /* buffer for rrblock */
-
-               if ((error = xfs_btree_read_bufs(args.mp, args.tp, args.agno,
-                               be32_to_cpu(right->bb_rightsib), 0, &rrbp,
-                               XFS_INO_BTREE_REF)))
-                       return error;
-               rrblock = XFS_BUF_TO_INOBT_BLOCK(rrbp);
-               if ((error = xfs_btree_check_sblock(cur, rrblock, level, rrbp)))
-                       return error;
-               rrblock->bb_leftsib = cpu_to_be32(args.agbno);
-               xfs_inobt_log_block(args.tp, rrbp, XFS_BB_LEFTSIB);
-       }
-       /*
-        * If the cursor is really in the right block, move it there.
-        * If it's just pointing past the last entry in left, then we'll
-        * insert there, so don't change anything in that case.
-        */
-       if (cur->bc_ptrs[level] > be16_to_cpu(left->bb_numrecs) + 1) {
-               xfs_btree_setbuf(cur, level, rbp);
-               cur->bc_ptrs[level] -= be16_to_cpu(left->bb_numrecs);
-       }
-       /*
-        * If there are more levels, we'll need another cursor which refers
-        * the right block, no matter where this cursor was.
-        */
-       if (level + 1 < cur->bc_nlevels) {
-               if ((error = xfs_btree_dup_cursor(cur, curp)))
-                       return error;
-               (*curp)->bc_ptrs[level + 1]++;
-       }
-       *bnop = args.agbno;
-       *stat = 1;
-       return 0;
-}
-
 /*
  * Externally visible routines.
  */
@@ -1285,6 +1128,48 @@ xfs_inobt_dup_cursor(
                        cur->bc_private.a.agbp, cur->bc_private.a.agno);
 }
 
+STATIC int
+xfs_inobt_alloc_block(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *start,
+       union xfs_btree_ptr     *new,
+       int                     length,
+       int                     *stat)
+{
+       xfs_alloc_arg_t         args;           /* block allocation args */
+       int                     error;          /* error return value */
+       xfs_agblock_t           sbno = be32_to_cpu(start->s);
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ENTRY);
+
+       memset(&args, 0, sizeof(args));
+       args.tp = cur->bc_tp;
+       args.mp = cur->bc_mp;
+       args.fsbno = XFS_AGB_TO_FSB(args.mp, cur->bc_private.a.agno, sbno);
+       args.minlen = 1;
+       args.maxlen = 1;
+       args.prod = 1;
+       args.type = XFS_ALLOCTYPE_NEAR_BNO;
+
+       error = xfs_alloc_vextent(&args);
+       if (error) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+               return error;
+       }
+       if (args.fsbno == NULLFSBLOCK) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+               *stat = 0;
+               return 0;
+       }
+       ASSERT(args.len == 1);
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+
+       new->s = cpu_to_be32(XFS_FSB_TO_AGBNO(args.mp, args.fsbno));
+       *stat = 1;
+       return 0;
+}
+
+
 STATIC int
 xfs_inobt_get_maxrecs(
        struct xfs_btree_cur    *cur,
@@ -1396,6 +1281,7 @@ static const struct xfs_btree_ops xfs_inobt_ops = {
        .key_len                = sizeof(xfs_inobt_key_t),
 
        .dup_cursor             = xfs_inobt_dup_cursor,
+       .alloc_block            = xfs_inobt_alloc_block,
        .get_maxrecs            = xfs_inobt_get_maxrecs,
        .init_key_from_rec      = xfs_inobt_init_key_from_rec,
        .init_ptr_from_cur      = xfs_inobt_init_ptr_from_cur,