]> git.ipfire.org Git - thirdparty/xfsprogs-dev.git/commitdiff
xfs: prepare xfs_btree_cur for dynamic cursor heights
authorDarrick J. Wong <djwong@kernel.org>
Thu, 28 Apr 2022 19:39:03 +0000 (15:39 -0400)
committerEric Sandeen <sandeen@sandeen.net>
Thu, 28 Apr 2022 19:39:03 +0000 (15:39 -0400)
Source kernel commit: 6ca444cfd663545e9e1c19ad2695836ffafad0a6

Split out the btree level information into a separate struct and put it
at the end of the cursor structure as a VLA.  Files with huge data forks
(and in the future, the realtime rmap btree) will require the ability to
support many more levels than a per-AG btree cursor, which means that
we're going to create per-btree type cursor caches to conserve memory
for the more common case.

Note that a subsequent patch actually introduces dynamic cursor heights.
This one merely rearranges the structure to prepare for that.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Chandan Babu R <chandan.babu@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@sandeen.net>
libxfs/init.c
libxfs/xfs_alloc.c
libxfs/xfs_bmap.c
libxfs/xfs_btree.c
libxfs/xfs_btree.h

index 5a3bf1697f641d0c902baf3e1a469b6d0f738b83..0cd4b929c92e1ec7a6ea2c7a1e4837a8cd8a5dc4 100644 (file)
@@ -242,7 +242,8 @@ init_zones(void)
        xfs_da_state_zone = kmem_zone_init(
                        sizeof(struct xfs_da_state), "xfs_da_state");
        xfs_btree_cur_zone = kmem_zone_init(
-                       sizeof(struct xfs_btree_cur), "xfs_btree_cur");
+                       xfs_btree_cur_sizeof(XFS_BTREE_MAXLEVELS),
+                       "xfs_btree_cur");
        xfs_bmap_free_item_zone = kmem_zone_init(
                        sizeof(struct xfs_extent_free_item),
                        "xfs_bmap_free_item");
index fe050d8e9b74db7e4bb653f3f5f61bb102d46781..b8857204d4d3cb64790cc87a210f5a84a4109345 100644 (file)
@@ -484,8 +484,8 @@ xfs_alloc_fixup_trees(
                struct xfs_btree_block  *bnoblock;
                struct xfs_btree_block  *cntblock;
 
-               bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_bufs[0]);
-               cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_bufs[0]);
+               bnoblock = XFS_BUF_TO_BLOCK(bno_cur->bc_levels[0].bp);
+               cntblock = XFS_BUF_TO_BLOCK(cnt_cur->bc_levels[0].bp);
 
                if (XFS_IS_CORRUPT(mp,
                                   bnoblock->bb_numrecs !=
@@ -1508,7 +1508,7 @@ xfs_alloc_ag_vextent_lastblock(
         * than minlen.
         */
        if (*len || args->alignment > 1) {
-               acur->cnt->bc_ptrs[0] = 1;
+               acur->cnt->bc_levels[0].ptr = 1;
                do {
                        error = xfs_alloc_get_rec(acur->cnt, bno, len, &i);
                        if (error)
index 371aedc2eda6d039dc718921a2d603cef84c7769..fa8d688027de2046429171aba9ff5d7cf1d47a8b 100644 (file)
@@ -233,10 +233,10 @@ xfs_bmap_get_bp(
                return NULL;
 
        for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
-               if (!cur->bc_bufs[i])
+               if (!cur->bc_levels[i].bp)
                        break;
-               if (xfs_buf_daddr(cur->bc_bufs[i]) == bno)
-                       return cur->bc_bufs[i];
+               if (xfs_buf_daddr(cur->bc_levels[i].bp) == bno)
+                       return cur->bc_levels[i].bp;
        }
 
        /* Chase down all the log items to see if the bp is there */
@@ -622,8 +622,8 @@ xfs_bmap_btree_to_extents(
        ip->i_nblocks--;
        xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
        xfs_trans_binval(tp, cbp);
-       if (cur->bc_bufs[0] == cbp)
-               cur->bc_bufs[0] = NULL;
+       if (cur->bc_levels[0].bp == cbp)
+               cur->bc_levels[0].bp = NULL;
        xfs_iroot_realloc(ip, -1, whichfork);
        ASSERT(ifp->if_broot == NULL);
        ifp->if_format = XFS_DINODE_FMT_EXTENTS;
index 426ab7f8684636fe4b8e95a714822f67aefec821..2e144dc272eea8a1f68a1d1d0481fd4ab88f612b 100644 (file)
@@ -364,8 +364,8 @@ xfs_btree_del_cursor(
         * way we won't have initialized all the entries down to 0.
         */
        for (i = 0; i < cur->bc_nlevels; i++) {
-               if (cur->bc_bufs[i])
-                       xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
+               if (cur->bc_levels[i].bp)
+                       xfs_trans_brelse(cur->bc_tp, cur->bc_levels[i].bp);
                else if (!error)
                        break;
        }
@@ -412,9 +412,9 @@ xfs_btree_dup_cursor(
         * For each level current, re-get the buffer and copy the ptr value.
         */
        for (i = 0; i < new->bc_nlevels; i++) {
-               new->bc_ptrs[i] = cur->bc_ptrs[i];
-               new->bc_ra[i] = cur->bc_ra[i];
-               bp = cur->bc_bufs[i];
+               new->bc_levels[i].ptr = cur->bc_levels[i].ptr;
+               new->bc_levels[i].ra = cur->bc_levels[i].ra;
+               bp = cur->bc_levels[i].bp;
                if (bp) {
                        error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
                                                   xfs_buf_daddr(bp), mp->m_bsize,
@@ -426,7 +426,7 @@ xfs_btree_dup_cursor(
                                return error;
                        }
                }
-               new->bc_bufs[i] = bp;
+               new->bc_levels[i].bp = bp;
        }
        *ncur = new;
        return 0;
@@ -678,7 +678,7 @@ xfs_btree_get_block(
                return xfs_btree_get_iroot(cur);
        }
 
-       *bpp = cur->bc_bufs[level];
+       *bpp = cur->bc_levels[level].bp;
        return XFS_BUF_TO_BLOCK(*bpp);
 }
 
@@ -708,7 +708,7 @@ xfs_btree_firstrec(
        /*
         * Set the ptr value to 1, that's the first record/key.
         */
-       cur->bc_ptrs[level] = 1;
+       cur->bc_levels[level].ptr = 1;
        return 1;
 }
 
@@ -738,7 +738,7 @@ xfs_btree_lastrec(
        /*
         * Set the ptr value to numrecs, that's the last record/key.
         */
-       cur->bc_ptrs[level] = be16_to_cpu(block->bb_numrecs);
+       cur->bc_levels[level].ptr = be16_to_cpu(block->bb_numrecs);
        return 1;
 }
 
@@ -919,11 +919,11 @@ xfs_btree_readahead(
            (lev == cur->bc_nlevels - 1))
                return 0;
 
-       if ((cur->bc_ra[lev] | lr) == cur->bc_ra[lev])
+       if ((cur->bc_levels[lev].ra | lr) == cur->bc_levels[lev].ra)
                return 0;
 
-       cur->bc_ra[lev] |= lr;
-       block = XFS_BUF_TO_BLOCK(cur->bc_bufs[lev]);
+       cur->bc_levels[lev].ra |= lr;
+       block = XFS_BUF_TO_BLOCK(cur->bc_levels[lev].bp);
 
        if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
                return xfs_btree_readahead_lblock(cur, lr, block);
@@ -988,22 +988,22 @@ xfs_btree_setbuf(
 {
        struct xfs_btree_block  *b;     /* btree block */
 
-       if (cur->bc_bufs[lev])
-               xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[lev]);
-       cur->bc_bufs[lev] = bp;
-       cur->bc_ra[lev] = 0;
+       if (cur->bc_levels[lev].bp)
+               xfs_trans_brelse(cur->bc_tp, cur->bc_levels[lev].bp);
+       cur->bc_levels[lev].bp = bp;
+       cur->bc_levels[lev].ra = 0;
 
        b = XFS_BUF_TO_BLOCK(bp);
        if (cur->bc_flags & XFS_BTREE_LONG_PTRS) {
                if (b->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK))
-                       cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
+                       cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
                if (b->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK))
-                       cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
+                       cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
        } else {
                if (b->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK))
-                       cur->bc_ra[lev] |= XFS_BTCUR_LEFTRA;
+                       cur->bc_levels[lev].ra |= XFS_BTCUR_LEFTRA;
                if (b->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK))
-                       cur->bc_ra[lev] |= XFS_BTCUR_RIGHTRA;
+                       cur->bc_levels[lev].ra |= XFS_BTCUR_RIGHTRA;
        }
 }
 
@@ -1545,7 +1545,7 @@ xfs_btree_increment(
 #endif
 
        /* We're done if we remain in the block after the increment. */
-       if (++cur->bc_ptrs[level] <= xfs_btree_get_numrecs(block))
+       if (++cur->bc_levels[level].ptr <= xfs_btree_get_numrecs(block))
                goto out1;
 
        /* Fail if we just went off the right edge of the tree. */
@@ -1568,7 +1568,7 @@ xfs_btree_increment(
                        goto error0;
 #endif
 
-               if (++cur->bc_ptrs[lev] <= xfs_btree_get_numrecs(block))
+               if (++cur->bc_levels[lev].ptr <= xfs_btree_get_numrecs(block))
                        break;
 
                /* Read-ahead the right block for the next loop. */
@@ -1595,14 +1595,14 @@ xfs_btree_increment(
        for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
                union xfs_btree_ptr     *ptrp;
 
-               ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
+               ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
                --lev;
                error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
                if (error)
                        goto error0;
 
                xfs_btree_setbuf(cur, lev, bp);
-               cur->bc_ptrs[lev] = 1;
+               cur->bc_levels[lev].ptr = 1;
        }
 out1:
        *stat = 1;
@@ -1638,7 +1638,7 @@ xfs_btree_decrement(
        xfs_btree_readahead(cur, level, XFS_BTCUR_LEFTRA);
 
        /* We're done if we remain in the block after the decrement. */
-       if (--cur->bc_ptrs[level] > 0)
+       if (--cur->bc_levels[level].ptr > 0)
                goto out1;
 
        /* Get a pointer to the btree block. */
@@ -1662,7 +1662,7 @@ xfs_btree_decrement(
         * Stop when we don't go off the left edge of a block.
         */
        for (lev = level + 1; lev < cur->bc_nlevels; lev++) {
-               if (--cur->bc_ptrs[lev] > 0)
+               if (--cur->bc_levels[lev].ptr > 0)
                        break;
                /* Read-ahead the left block for the next loop. */
                xfs_btree_readahead(cur, lev, XFS_BTCUR_LEFTRA);
@@ -1688,13 +1688,13 @@ xfs_btree_decrement(
        for (block = xfs_btree_get_block(cur, lev, &bp); lev > level; ) {
                union xfs_btree_ptr     *ptrp;
 
-               ptrp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[lev], block);
+               ptrp = xfs_btree_ptr_addr(cur, cur->bc_levels[lev].ptr, block);
                --lev;
                error = xfs_btree_read_buf_block(cur, ptrp, 0, &block, &bp);
                if (error)
                        goto error0;
                xfs_btree_setbuf(cur, lev, bp);
-               cur->bc_ptrs[lev] = xfs_btree_get_numrecs(block);
+               cur->bc_levels[lev].ptr = xfs_btree_get_numrecs(block);
        }
 out1:
        *stat = 1;
@@ -1732,7 +1732,7 @@ xfs_btree_lookup_get_block(
         *
         * Otherwise throw it away and get a new one.
         */
-       bp = cur->bc_bufs[level];
+       bp = cur->bc_levels[level].bp;
        error = xfs_btree_ptr_to_daddr(cur, pp, &daddr);
        if (error)
                return error;
@@ -1861,7 +1861,7 @@ xfs_btree_lookup(
                                        return -EFSCORRUPTED;
                                }
 
-                               cur->bc_ptrs[0] = dir != XFS_LOOKUP_LE;
+                               cur->bc_levels[0].ptr = dir != XFS_LOOKUP_LE;
                                *stat = 0;
                                return 0;
                        }
@@ -1913,7 +1913,7 @@ xfs_btree_lookup(
                        if (error)
                                goto error0;
 
-                       cur->bc_ptrs[level] = keyno;
+                       cur->bc_levels[level].ptr = keyno;
                }
        }
 
@@ -1930,7 +1930,7 @@ xfs_btree_lookup(
                    !xfs_btree_ptr_is_null(cur, &ptr)) {
                        int     i;
 
-                       cur->bc_ptrs[0] = keyno;
+                       cur->bc_levels[0].ptr = keyno;
                        error = xfs_btree_increment(cur, 0, &i);
                        if (error)
                                goto error0;
@@ -1941,7 +1941,7 @@ xfs_btree_lookup(
                }
        } else if (dir == XFS_LOOKUP_LE && diff > 0)
                keyno--;
-       cur->bc_ptrs[0] = keyno;
+       cur->bc_levels[0].ptr = keyno;
 
        /* Return if we succeeded or not. */
        if (keyno == 0 || keyno > xfs_btree_get_numrecs(block))
@@ -2101,7 +2101,7 @@ __xfs_btree_updkeys(
                if (error)
                        return error;
 #endif
-               ptr = cur->bc_ptrs[level];
+               ptr = cur->bc_levels[level].ptr;
                nlkey = xfs_btree_key_addr(cur, ptr, block);
                nhkey = xfs_btree_high_key_addr(cur, ptr, block);
                if (!force_all &&
@@ -2168,7 +2168,7 @@ xfs_btree_update_keys(
                if (error)
                        return error;
 #endif
-               ptr = cur->bc_ptrs[level];
+               ptr = cur->bc_levels[level].ptr;
                kp = xfs_btree_key_addr(cur, ptr, block);
                xfs_btree_copy_keys(cur, kp, &key, 1);
                xfs_btree_log_keys(cur, bp, ptr, ptr);
@@ -2202,7 +2202,7 @@ xfs_btree_update(
                goto error0;
 #endif
        /* Get the address of the rec to be updated. */
-       ptr = cur->bc_ptrs[0];
+       ptr = cur->bc_levels[0].ptr;
        rp = xfs_btree_rec_addr(cur, ptr, block);
 
        /* Fill in the new contents and log them. */
@@ -2277,7 +2277,7 @@ xfs_btree_lshift(
         * If the cursor entry is the one that would be moved, don't
         * do it... it's too complicated.
         */
-       if (cur->bc_ptrs[level] <= 1)
+       if (cur->bc_levels[level].ptr <= 1)
                goto out0;
 
        /* Set up the left neighbor as "left". */
@@ -2411,7 +2411,7 @@ xfs_btree_lshift(
                goto error0;
 
        /* Slide the cursor value left one. */
-       cur->bc_ptrs[level]--;
+       cur->bc_levels[level].ptr--;
 
        *stat = 1;
        return 0;
@@ -2473,7 +2473,7 @@ xfs_btree_rshift(
         * do it... it's too complicated.
         */
        lrecs = xfs_btree_get_numrecs(left);
-       if (cur->bc_ptrs[level] >= lrecs)
+       if (cur->bc_levels[level].ptr >= lrecs)
                goto out0;
 
        /* Set up the right neighbor as "right". */
@@ -2661,7 +2661,7 @@ __xfs_btree_split(
         */
        lrecs = xfs_btree_get_numrecs(left);
        rrecs = lrecs / 2;
-       if ((lrecs & 1) && cur->bc_ptrs[level] <= rrecs + 1)
+       if ((lrecs & 1) && cur->bc_levels[level].ptr <= rrecs + 1)
                rrecs++;
        src_index = (lrecs - rrecs + 1);
 
@@ -2757,9 +2757,9 @@ __xfs_btree_split(
         * 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) {
+       if (cur->bc_levels[level].ptr > lrecs + 1) {
                xfs_btree_setbuf(cur, level, rbp);
-               cur->bc_ptrs[level] -= lrecs;
+               cur->bc_levels[level].ptr -= lrecs;
        }
        /*
         * If there are more levels, we'll need another cursor which refers
@@ -2769,7 +2769,7 @@ __xfs_btree_split(
                error = xfs_btree_dup_cursor(cur, curp);
                if (error)
                        goto error0;
-               (*curp)->bc_ptrs[level + 1]++;
+               (*curp)->bc_levels[level + 1].ptr++;
        }
        *ptrp = rptr;
        *stat = 1;
@@ -2935,7 +2935,7 @@ xfs_btree_new_iroot(
        xfs_btree_set_numrecs(block, 1);
        cur->bc_nlevels++;
        ASSERT(cur->bc_nlevels <= XFS_BTREE_MAXLEVELS);
-       cur->bc_ptrs[level + 1] = 1;
+       cur->bc_levels[level + 1].ptr = 1;
 
        kp = xfs_btree_key_addr(cur, 1, block);
        ckp = xfs_btree_key_addr(cur, 1, cblock);
@@ -3096,7 +3096,7 @@ xfs_btree_new_root(
 
        /* Fix up the cursor. */
        xfs_btree_setbuf(cur, cur->bc_nlevels, nbp);
-       cur->bc_ptrs[cur->bc_nlevels] = nptr;
+       cur->bc_levels[cur->bc_nlevels].ptr = nptr;
        cur->bc_nlevels++;
        ASSERT(cur->bc_nlevels <= XFS_BTREE_MAXLEVELS);
        *stat = 1;
@@ -3155,7 +3155,7 @@ xfs_btree_make_block_unfull(
                return error;
 
        if (*stat) {
-               *oindex = *index = cur->bc_ptrs[level];
+               *oindex = *index = cur->bc_levels[level].ptr;
                return 0;
        }
 
@@ -3170,7 +3170,7 @@ xfs_btree_make_block_unfull(
                return error;
 
 
-       *index = cur->bc_ptrs[level];
+       *index = cur->bc_levels[level].ptr;
        return 0;
 }
 
@@ -3217,7 +3217,7 @@ xfs_btree_insrec(
        }
 
        /* If we're off the left edge, return failure. */
-       ptr = cur->bc_ptrs[level];
+       ptr = cur->bc_levels[level].ptr;
        if (ptr == 0) {
                *stat = 0;
                return 0;
@@ -3560,7 +3560,7 @@ xfs_btree_kill_iroot(
        if (error)
                return error;
 
-       cur->bc_bufs[level - 1] = NULL;
+       cur->bc_levels[level - 1].bp = NULL;
        be16_add_cpu(&block->bb_level, -1);
        xfs_trans_log_inode(cur->bc_tp, ip,
                XFS_ILOG_CORE | xfs_ilog_fbroot(cur->bc_ino.whichfork));
@@ -3593,8 +3593,8 @@ xfs_btree_kill_root(
        if (error)
                return error;
 
-       cur->bc_bufs[level] = NULL;
-       cur->bc_ra[level] = 0;
+       cur->bc_levels[level].bp = NULL;
+       cur->bc_levels[level].ra = 0;
        cur->bc_nlevels--;
 
        return 0;
@@ -3653,7 +3653,7 @@ xfs_btree_delrec(
        tcur = NULL;
 
        /* Get the index of the entry being deleted, check for nothing there. */
-       ptr = cur->bc_ptrs[level];
+       ptr = cur->bc_levels[level].ptr;
        if (ptr == 0) {
                *stat = 0;
                return 0;
@@ -3963,7 +3963,7 @@ xfs_btree_delrec(
                                xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
                                tcur = NULL;
                                if (level == 0)
-                                       cur->bc_ptrs[0]++;
+                                       cur->bc_levels[0].ptr++;
 
                                *stat = 1;
                                return 0;
@@ -4100,9 +4100,9 @@ xfs_btree_delrec(
         * cursor to the left block, and fix up the index.
         */
        if (bp != lbp) {
-               cur->bc_bufs[level] = lbp;
-               cur->bc_ptrs[level] += lrecs;
-               cur->bc_ra[level] = 0;
+               cur->bc_levels[level].bp = lbp;
+               cur->bc_levels[level].ptr += lrecs;
+               cur->bc_levels[level].ra = 0;
        }
        /*
         * If we joined with the right neighbor and there's a level above
@@ -4122,16 +4122,16 @@ xfs_btree_delrec(
         * We can't use decrement because it would change the next level up.
         */
        if (level > 0)
-               cur->bc_ptrs[level]--;
+               cur->bc_levels[level].ptr--;
 
        /*
         * We combined blocks, so we have to update the parent keys if the
-        * btree supports overlapped intervals.  However, bc_ptrs[level + 1]
-        * points to the old block so that the caller knows which record to
-        * delete.  Therefore, the caller must be savvy enough to call updkeys
-        * for us if we return stat == 2.  The other exit points from this
-        * function don't require deletions further up the tree, so they can
-        * call updkeys directly.
+        * btree supports overlapped intervals.  However,
+        * bc_levels[level + 1].ptr points to the old block so that the caller
+        * knows which record to delete.  Therefore, the caller must be savvy
+        * enough to call updkeys for us if we return stat == 2.  The other
+        * exit points from this function don't require deletions further up
+        * the tree, so they can call updkeys directly.
         */
 
        /* Return value means the next level up has something to do. */
@@ -4185,7 +4185,7 @@ xfs_btree_delete(
 
        if (i == 0) {
                for (level = 1; level < cur->bc_nlevels; level++) {
-                       if (cur->bc_ptrs[level] == 0) {
+                       if (cur->bc_levels[level].ptr == 0) {
                                error = xfs_btree_decrement(cur, level, &i);
                                if (error)
                                        goto error0;
@@ -4216,7 +4216,7 @@ xfs_btree_get_rec(
        int                     error;  /* error return value */
 #endif
 
-       ptr = cur->bc_ptrs[0];
+       ptr = cur->bc_levels[0].ptr;
        block = xfs_btree_get_block(cur, 0, &bp);
 
 #ifdef DEBUG
@@ -4664,23 +4664,25 @@ xfs_btree_overlapped_query_range(
        if (error)
                goto out;
 #endif
-       cur->bc_ptrs[level] = 1;
+       cur->bc_levels[level].ptr = 1;
 
        while (level < cur->bc_nlevels) {
                block = xfs_btree_get_block(cur, level, &bp);
 
                /* End of node, pop back towards the root. */
-               if (cur->bc_ptrs[level] > be16_to_cpu(block->bb_numrecs)) {
+               if (cur->bc_levels[level].ptr >
+                                       be16_to_cpu(block->bb_numrecs)) {
 pop_up:
                        if (level < cur->bc_nlevels - 1)
-                               cur->bc_ptrs[level + 1]++;
+                               cur->bc_levels[level + 1].ptr++;
                        level++;
                        continue;
                }
 
                if (level == 0) {
                        /* Handle a leaf node. */
-                       recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
+                       recp = xfs_btree_rec_addr(cur, cur->bc_levels[0].ptr,
+                                       block);
 
                        cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
                        ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
@@ -4703,14 +4705,15 @@ pop_up:
                                /* Record is larger than high key; pop. */
                                goto pop_up;
                        }
-                       cur->bc_ptrs[level]++;
+                       cur->bc_levels[level].ptr++;
                        continue;
                }
 
                /* Handle an internal node. */
-               lkp = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
-               hkp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
-               pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
+               lkp = xfs_btree_key_addr(cur, cur->bc_levels[level].ptr, block);
+               hkp = xfs_btree_high_key_addr(cur, cur->bc_levels[level].ptr,
+                               block);
+               pp = xfs_btree_ptr_addr(cur, cur->bc_levels[level].ptr, block);
 
                ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
                hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
@@ -4733,13 +4736,13 @@ pop_up:
                        if (error)
                                goto out;
 #endif
-                       cur->bc_ptrs[level] = 1;
+                       cur->bc_levels[level].ptr = 1;
                        continue;
                } else if (hdiff < 0) {
                        /* The low key is larger than the upper range; pop. */
                        goto pop_up;
                }
-               cur->bc_ptrs[level]++;
+               cur->bc_levels[level].ptr++;
        }
 
 out:
@@ -4750,13 +4753,14 @@ out:
         * with a zero-results range query, so release the buffers if we
         * failed to return any results.
         */
-       if (cur->bc_bufs[0] == NULL) {
+       if (cur->bc_levels[0].bp == NULL) {
                for (i = 0; i < cur->bc_nlevels; i++) {
-                       if (cur->bc_bufs[i]) {
-                               xfs_trans_brelse(cur->bc_tp, cur->bc_bufs[i]);
-                               cur->bc_bufs[i] = NULL;
-                               cur->bc_ptrs[i] = 0;
-                               cur->bc_ra[i] = 0;
+                       if (cur->bc_levels[i].bp) {
+                               xfs_trans_brelse(cur->bc_tp,
+                                               cur->bc_levels[i].bp);
+                               cur->bc_levels[i].bp = NULL;
+                               cur->bc_levels[i].ptr = 0;
+                               cur->bc_levels[i].ra = 0;
                        }
                }
        }
@@ -4918,7 +4922,7 @@ xfs_btree_has_more_records(
        block = xfs_btree_get_block(cur, 0, &bp);
 
        /* There are still records in this block. */
-       if (cur->bc_ptrs[0] < xfs_btree_get_numrecs(block))
+       if (cur->bc_levels[0].ptr < xfs_btree_get_numrecs(block))
                return true;
 
        /* There are more record blocks. */
index 1018bcc43d660a1ddb3a2fdac67417cbd62d43ed..0181fc98bc129fa62504f7213dffa788f312a4c0 100644 (file)
@@ -212,6 +212,19 @@ struct xfs_btree_cur_ino {
 #define        XFS_BTCUR_BMBT_INVALID_OWNER    (1 << 1)
 };
 
+struct xfs_btree_level {
+       /* buffer pointer */
+       struct xfs_buf          *bp;
+
+       /* key/record number */
+       uint16_t                ptr;
+
+       /* readahead info */
+#define XFS_BTCUR_LEFTRA       (1 << 0) /* left sibling has been read-ahead */
+#define XFS_BTCUR_RIGHTRA      (1 << 1) /* right sibling has been read-ahead */
+       uint16_t                ra;
+};
+
 /*
  * Btree cursor structure.
  * This collects all information needed by the btree code in one place.
@@ -223,11 +236,6 @@ struct xfs_btree_cur
        const struct xfs_btree_ops *bc_ops;
        uint                    bc_flags; /* btree features - below */
        union xfs_btree_irec    bc_rec; /* current insert/search record value */
-       struct xfs_buf  *bc_bufs[XFS_BTREE_MAXLEVELS];  /* buf ptr per level */
-       int             bc_ptrs[XFS_BTREE_MAXLEVELS];   /* key/record # */
-       uint8_t         bc_ra[XFS_BTREE_MAXLEVELS];     /* readahead bits */
-#define        XFS_BTCUR_LEFTRA        1       /* left sibling has been read-ahead */
-#define        XFS_BTCUR_RIGHTRA       2       /* right sibling has been read-ahead */
        uint8_t         bc_nlevels;     /* number of levels in the tree */
        xfs_btnum_t     bc_btnum;       /* identifies which btree type */
        int             bc_statoff;     /* offset of btre stats array */
@@ -242,8 +250,22 @@ struct xfs_btree_cur
                struct xfs_btree_cur_ag bc_ag;
                struct xfs_btree_cur_ino bc_ino;
        };
+
+       /* Must be at the end of the struct! */
+       struct xfs_btree_level  bc_levels[];
 };
 
+/*
+ * Compute the size of a btree cursor that can handle a btree of a given
+ * height.  The bc_levels array handles node and leaf blocks, so its size
+ * is exactly nlevels.
+ */
+static inline size_t
+xfs_btree_cur_sizeof(unsigned int nlevels)
+{
+       return struct_size((struct xfs_btree_cur *)NULL, bc_levels, nlevels);
+}
+
 /* cursor flags */
 #define XFS_BTREE_LONG_PTRS            (1<<0)  /* pointers are 64bits long */
 #define XFS_BTREE_ROOT_IN_INODE                (1<<1)  /* root may be variable size */
@@ -257,7 +279,6 @@ struct xfs_btree_cur
  */
 #define XFS_BTREE_STAGING              (1<<5)
 
-
 #define        XFS_BTREE_NOERROR       0
 #define        XFS_BTREE_ERROR         1