xfs: introduce interval queries on btrees
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 3 Aug 2016 01:10:21 +0000 (11:10 +1000)
committerDave Chinner <david@fromorbit.com>
Wed, 3 Aug 2016 01:10:21 +0000 (11:10 +1000)
Create a function to enable querying of btree records mapping to a
range of keys.  This will be used in subsequent patches to allow
querying the reverse mapping btree to find the extents mapped to a
range of physical blocks, though the generic code can be used for
any range query.

The overlapped query range function needs to use the btree get_block
helper because the root block could be an inode, in which case
bc_bufs[nlevels-1] will be NULL.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Dave Chinner <david@fromorbit.com>
fs/xfs/libxfs/xfs_btree.c
fs/xfs/libxfs/xfs_btree.h
fs/xfs/xfs_trace.h

index 5c1b35a23a3dfdd97952cd582924d4263d2ec093..d79dcb08ed1845385fa5bf02b4a3ebf425f8b569 100644 (file)
@@ -4521,3 +4521,267 @@ xfs_btree_compute_maxlevels(
                maxblocks = (maxblocks + limits[1] - 1) / limits[1];
        return level;
 }
+
+/*
+ * Query a regular btree for all records overlapping a given interval.
+ * Start with a LE lookup of the key of low_rec and return all records
+ * until we find a record with a key greater than the key of high_rec.
+ */
+STATIC int
+xfs_btree_simple_query_range(
+       struct xfs_btree_cur            *cur,
+       union xfs_btree_key             *low_key,
+       union xfs_btree_key             *high_key,
+       xfs_btree_query_range_fn        fn,
+       void                            *priv)
+{
+       union xfs_btree_rec             *recp;
+       union xfs_btree_key             rec_key;
+       __int64_t                       diff;
+       int                             stat;
+       bool                            firstrec = true;
+       int                             error;
+
+       ASSERT(cur->bc_ops->init_high_key_from_rec);
+       ASSERT(cur->bc_ops->diff_two_keys);
+
+       /*
+        * Find the leftmost record.  The btree cursor must be set
+        * to the low record used to generate low_key.
+        */
+       stat = 0;
+       error = xfs_btree_lookup(cur, XFS_LOOKUP_LE, &stat);
+       if (error)
+               goto out;
+
+       while (stat) {
+               /* Find the record. */
+               error = xfs_btree_get_rec(cur, &recp, &stat);
+               if (error || !stat)
+                       break;
+               cur->bc_ops->init_high_key_from_rec(&rec_key, recp);
+
+               /* Skip if high_key(rec) < low_key. */
+               if (firstrec) {
+                       firstrec = false;
+                       diff = cur->bc_ops->diff_two_keys(cur, low_key,
+                                       &rec_key);
+                       if (diff > 0)
+                               goto advloop;
+               }
+
+               /* Stop if high_key < low_key(rec). */
+               diff = cur->bc_ops->diff_two_keys(cur, &rec_key, high_key);
+               if (diff > 0)
+                       break;
+
+               /* Callback */
+               error = fn(cur, recp, priv);
+               if (error < 0 || error == XFS_BTREE_QUERY_RANGE_ABORT)
+                       break;
+
+advloop:
+               /* Move on to the next record. */
+               error = xfs_btree_increment(cur, 0, &stat);
+               if (error)
+                       break;
+       }
+
+out:
+       return error;
+}
+
+/*
+ * Query an overlapped interval btree for all records overlapping a given
+ * interval.  This function roughly follows the algorithm given in
+ * "Interval Trees" of _Introduction to Algorithms_, which is section
+ * 14.3 in the 2nd and 3rd editions.
+ *
+ * First, generate keys for the low and high records passed in.
+ *
+ * For any leaf node, generate the high and low keys for the record.
+ * If the record keys overlap with the query low/high keys, pass the
+ * record to the function iterator.
+ *
+ * For any internal node, compare the low and high keys of each
+ * pointer against the query low/high keys.  If there's an overlap,
+ * follow the pointer.
+ *
+ * As an optimization, we stop scanning a block when we find a low key
+ * that is greater than the query's high key.
+ */
+STATIC int
+xfs_btree_overlapped_query_range(
+       struct xfs_btree_cur            *cur,
+       union xfs_btree_key             *low_key,
+       union xfs_btree_key             *high_key,
+       xfs_btree_query_range_fn        fn,
+       void                            *priv)
+{
+       union xfs_btree_ptr             ptr;
+       union xfs_btree_ptr             *pp;
+       union xfs_btree_key             rec_key;
+       union xfs_btree_key             rec_hkey;
+       union xfs_btree_key             *lkp;
+       union xfs_btree_key             *hkp;
+       union xfs_btree_rec             *recp;
+       struct xfs_btree_block          *block;
+       __int64_t                       ldiff;
+       __int64_t                       hdiff;
+       int                             level;
+       struct xfs_buf                  *bp;
+       int                             i;
+       int                             error;
+
+       /* Load the root of the btree. */
+       level = cur->bc_nlevels - 1;
+       cur->bc_ops->init_ptr_from_cur(cur, &ptr);
+       error = xfs_btree_lookup_get_block(cur, level, &ptr, &block);
+       if (error)
+               return error;
+       xfs_btree_get_block(cur, level, &bp);
+       trace_xfs_btree_overlapped_query_range(cur, level, bp);
+#ifdef DEBUG
+       error = xfs_btree_check_block(cur, block, level, bp);
+       if (error)
+               goto out;
+#endif
+       cur->bc_ptrs[level] = 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)) {
+pop_up:
+                       if (level < cur->bc_nlevels - 1)
+                               cur->bc_ptrs[level + 1]++;
+                       level++;
+                       continue;
+               }
+
+               if (level == 0) {
+                       /* Handle a leaf node. */
+                       recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
+
+                       cur->bc_ops->init_high_key_from_rec(&rec_hkey, recp);
+                       ldiff = cur->bc_ops->diff_two_keys(cur, &rec_hkey,
+                                       low_key);
+
+                       cur->bc_ops->init_key_from_rec(&rec_key, recp);
+                       hdiff = cur->bc_ops->diff_two_keys(cur, high_key,
+                                       &rec_key);
+
+                       /*
+                        * If (record's high key >= query's low key) and
+                        *    (query's high key >= record's low key), then
+                        * this record overlaps the query range; callback.
+                        */
+                       if (ldiff >= 0 && hdiff >= 0) {
+                               error = fn(cur, recp, priv);
+                               if (error < 0 ||
+                                   error == XFS_BTREE_QUERY_RANGE_ABORT)
+                                       break;
+                       } else if (hdiff < 0) {
+                               /* Record is larger than high key; pop. */
+                               goto pop_up;
+                       }
+                       cur->bc_ptrs[level]++;
+                       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);
+
+               ldiff = cur->bc_ops->diff_two_keys(cur, hkp, low_key);
+               hdiff = cur->bc_ops->diff_two_keys(cur, high_key, lkp);
+
+               /*
+                * If (pointer's high key >= query's low key) and
+                *    (query's high key >= pointer's low key), then
+                * this record overlaps the query range; follow pointer.
+                */
+               if (ldiff >= 0 && hdiff >= 0) {
+                       level--;
+                       error = xfs_btree_lookup_get_block(cur, level, pp,
+                                       &block);
+                       if (error)
+                               goto out;
+                       xfs_btree_get_block(cur, level, &bp);
+                       trace_xfs_btree_overlapped_query_range(cur, level, bp);
+#ifdef DEBUG
+                       error = xfs_btree_check_block(cur, block, level, bp);
+                       if (error)
+                               goto out;
+#endif
+                       cur->bc_ptrs[level] = 1;
+                       continue;
+               } else if (hdiff < 0) {
+                       /* The low key is larger than the upper range; pop. */
+                       goto pop_up;
+               }
+               cur->bc_ptrs[level]++;
+       }
+
+out:
+       /*
+        * If we don't end this function with the cursor pointing at a record
+        * block, a subsequent non-error cursor deletion will not release
+        * node-level buffers, causing a buffer leak.  This is quite possible
+        * with a zero-results range query, so release the buffers if we
+        * failed to return any results.
+        */
+       if (cur->bc_bufs[0] == 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;
+                       }
+               }
+       }
+
+       return error;
+}
+
+/*
+ * Query a btree for all records overlapping a given interval of keys.  The
+ * supplied function will be called with each record found; return one of the
+ * XFS_BTREE_QUERY_RANGE_{CONTINUE,ABORT} values or the usual negative error
+ * code.  This function returns XFS_BTREE_QUERY_RANGE_ABORT, zero, or a
+ * negative error code.
+ */
+int
+xfs_btree_query_range(
+       struct xfs_btree_cur            *cur,
+       union xfs_btree_irec            *low_rec,
+       union xfs_btree_irec            *high_rec,
+       xfs_btree_query_range_fn        fn,
+       void                            *priv)
+{
+       union xfs_btree_rec             rec;
+       union xfs_btree_key             low_key;
+       union xfs_btree_key             high_key;
+
+       /* Find the keys of both ends of the interval. */
+       cur->bc_rec = *high_rec;
+       cur->bc_ops->init_rec_from_cur(cur, &rec);
+       cur->bc_ops->init_key_from_rec(&high_key, &rec);
+
+       cur->bc_rec = *low_rec;
+       cur->bc_ops->init_rec_from_cur(cur, &rec);
+       cur->bc_ops->init_key_from_rec(&low_key, &rec);
+
+       /* Enforce low key < high key. */
+       if (cur->bc_ops->diff_two_keys(cur, &low_key, &high_key) > 0)
+               return -EINVAL;
+
+       if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
+               return xfs_btree_simple_query_range(cur, &low_key,
+                               &high_key, fn, priv);
+       return xfs_btree_overlapped_query_range(cur, &low_key, &high_key,
+                       fn, priv);
+}
index bce6daacb1f73e6087c8f49cf4230e933fc9abd7..34c13616d255b87ca8a687e0f4bda1dbd35a18d6 100644 (file)
@@ -227,6 +227,12 @@ struct xfs_btree_ops {
 #define LASTREC_DELREC 2
 
 
+union xfs_btree_irec {
+       struct xfs_alloc_rec_incore     a;
+       struct xfs_bmbt_irec            b;
+       struct xfs_inobt_rec_incore     i;
+};
+
 /*
  * Btree cursor structure.
  * This collects all information needed by the btree code in one place.
@@ -237,11 +243,7 @@ typedef struct xfs_btree_cur
        struct xfs_mount        *bc_mp; /* file system mount struct */
        const struct xfs_btree_ops *bc_ops;
        uint                    bc_flags; /* btree features - below */
-       union {
-               xfs_alloc_rec_incore_t  a;
-               xfs_bmbt_irec_t         b;
-               xfs_inobt_rec_incore_t  i;
-       }               bc_rec;         /* current insert/search record value */
+       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 */
@@ -524,4 +526,14 @@ void xfs_btree_get_node_keys_overlapped(struct xfs_btree_cur *cur,
                struct xfs_btree_block *block, union xfs_btree_key *key);
 int xfs_btree_update_keys_overlapped(struct xfs_btree_cur *cur, int level);
 
+/* return codes */
+#define XFS_BTREE_QUERY_RANGE_CONTINUE 0       /* keep iterating */
+#define XFS_BTREE_QUERY_RANGE_ABORT    1       /* stop iterating */
+typedef int (*xfs_btree_query_range_fn)(struct xfs_btree_cur *cur,
+               union xfs_btree_rec *rec, void *priv);
+
+int xfs_btree_query_range(struct xfs_btree_cur *cur,
+               union xfs_btree_irec *low_rec, union xfs_btree_irec *high_rec,
+               xfs_btree_query_range_fn fn, void *priv);
+
 #endif /* __XFS_BTREE_H__ */
index 8fb59e644f807e968e40dfa2983f71e719384fdc..a5862680db319e24f45c390e886971667bbda0dd 100644 (file)
@@ -2220,6 +2220,7 @@ DEFINE_EVENT(xfs_btree_cur_class, name, \
        TP_PROTO(struct xfs_btree_cur *cur, int level, struct xfs_buf *bp), \
        TP_ARGS(cur, level, bp))
 DEFINE_BTREE_CUR_EVENT(xfs_btree_updkeys);
+DEFINE_BTREE_CUR_EVENT(xfs_btree_overlapped_query_range);
 
 #endif /* _TRACE_XFS_H */