xfs: add refcount btree operations
authorDarrick J. Wong <darrick.wong@oracle.com>
Mon, 3 Oct 2016 16:11:19 +0000 (09:11 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Mon, 3 Oct 2016 16:11:19 +0000 (09:11 -0700)
Implement the generic btree operations required to manipulate refcount
btree blocks.  The implementation is similar to the bmapbt, though it
will only allocate and free blocks from the AG.

Since the refcount root and level fields are separate from the
existing roots and levels array, they need a separate logging flag.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
[hch: fix logging of AGF refcount btree fields]
Signed-off-by: Christoph Hellwig <hch@lst.de>
fs/xfs/Makefile
fs/xfs/libxfs/xfs_alloc.c
fs/xfs/libxfs/xfs_format.h
fs/xfs/libxfs/xfs_refcount.c [new file with mode: 0644]
fs/xfs/libxfs/xfs_refcount.h [new file with mode: 0644]
fs/xfs/libxfs/xfs_refcount_btree.c

index 8d749f2df99e62385fc06cc9af9cc7a3108313d8..98b24279b75f563317467a4092142011c0d6a159 100644 (file)
@@ -55,6 +55,7 @@ xfs-y                         += $(addprefix libxfs/, \
                                   xfs_ag_resv.o \
                                   xfs_rmap.o \
                                   xfs_rmap_btree.o \
+                                  xfs_refcount.o \
                                   xfs_refcount_btree.o \
                                   xfs_sb.o \
                                   xfs_symlink_remote.o \
index aa0e1ca7b6a5a710a914d1c4071b5f8013db051a..be7e3fcefd73c989c7bb302bc77732af74f3f058 100644 (file)
@@ -2334,6 +2334,9 @@ xfs_alloc_log_agf(
                offsetof(xfs_agf_t, agf_btreeblks),
                offsetof(xfs_agf_t, agf_uuid),
                offsetof(xfs_agf_t, agf_rmap_blocks),
+               offsetof(xfs_agf_t, agf_refcount_blocks),
+               offsetof(xfs_agf_t, agf_refcount_root),
+               offsetof(xfs_agf_t, agf_refcount_level),
                /* needed so that we don't log the whole rest of the structure: */
                offsetof(xfs_agf_t, agf_spare64),
                sizeof(xfs_agf_t)
index 97c74f4a22b5964a67d64b0f6afc71dfb0a0ae0e..8b826102dfe343b05aab29c40148b66bd56a855b 100644 (file)
@@ -684,8 +684,11 @@ typedef struct xfs_agf {
 #define        XFS_AGF_BTREEBLKS       0x00000800
 #define        XFS_AGF_UUID            0x00001000
 #define        XFS_AGF_RMAP_BLOCKS     0x00002000
-#define        XFS_AGF_SPARE64         0x00004000
-#define        XFS_AGF_NUM_BITS        15
+#define        XFS_AGF_REFCOUNT_BLOCKS 0x00004000
+#define        XFS_AGF_REFCOUNT_ROOT   0x00008000
+#define        XFS_AGF_REFCOUNT_LEVEL  0x00010000
+#define        XFS_AGF_SPARE64         0x00020000
+#define        XFS_AGF_NUM_BITS        18
 #define        XFS_AGF_ALL_BITS        ((1 << XFS_AGF_NUM_BITS) - 1)
 
 #define XFS_AGF_FLAGS \
@@ -703,6 +706,9 @@ typedef struct xfs_agf {
        { XFS_AGF_BTREEBLKS,    "BTREEBLKS" }, \
        { XFS_AGF_UUID,         "UUID" }, \
        { XFS_AGF_RMAP_BLOCKS,  "RMAP_BLOCKS" }, \
+       { XFS_AGF_REFCOUNT_BLOCKS,      "REFCOUNT_BLOCKS" }, \
+       { XFS_AGF_REFCOUNT_ROOT,        "REFCOUNT_ROOT" }, \
+       { XFS_AGF_REFCOUNT_LEVEL,       "REFCOUNT_LEVEL" }, \
        { XFS_AGF_SPARE64,      "SPARE64" }
 
 /* disk block (xfs_daddr_t) in the AG */
diff --git a/fs/xfs/libxfs/xfs_refcount.c b/fs/xfs/libxfs/xfs_refcount.c
new file mode 100644 (file)
index 0000000..de13406
--- /dev/null
@@ -0,0 +1,177 @@
+/*
+ * 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_sb.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_btree.h"
+#include "xfs_bmap.h"
+#include "xfs_refcount_btree.h"
+#include "xfs_alloc.h"
+#include "xfs_error.h"
+#include "xfs_trace.h"
+#include "xfs_cksum.h"
+#include "xfs_trans.h"
+#include "xfs_bit.h"
+#include "xfs_refcount.h"
+
+/*
+ * Look up the first record less than or equal to [bno, len] in the btree
+ * given by cur.
+ */
+int
+xfs_refcount_lookup_le(
+       struct xfs_btree_cur    *cur,
+       xfs_agblock_t           bno,
+       int                     *stat)
+{
+       trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
+                       XFS_LOOKUP_LE);
+       cur->bc_rec.rc.rc_startblock = bno;
+       cur->bc_rec.rc.rc_blockcount = 0;
+       return xfs_btree_lookup(cur, XFS_LOOKUP_LE, stat);
+}
+
+/*
+ * Look up the first record greater than or equal to [bno, len] in the btree
+ * given by cur.
+ */
+int
+xfs_refcount_lookup_ge(
+       struct xfs_btree_cur    *cur,
+       xfs_agblock_t           bno,
+       int                     *stat)
+{
+       trace_xfs_refcount_lookup(cur->bc_mp, cur->bc_private.a.agno, bno,
+                       XFS_LOOKUP_GE);
+       cur->bc_rec.rc.rc_startblock = bno;
+       cur->bc_rec.rc.rc_blockcount = 0;
+       return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
+}
+
+/*
+ * Get the data from the pointed-to record.
+ */
+int
+xfs_refcount_get_rec(
+       struct xfs_btree_cur            *cur,
+       struct xfs_refcount_irec        *irec,
+       int                             *stat)
+{
+       union xfs_btree_rec     *rec;
+       int                     error;
+
+       error = xfs_btree_get_rec(cur, &rec, stat);
+       if (!error && *stat == 1) {
+               irec->rc_startblock = be32_to_cpu(rec->refc.rc_startblock);
+               irec->rc_blockcount = be32_to_cpu(rec->refc.rc_blockcount);
+               irec->rc_refcount = be32_to_cpu(rec->refc.rc_refcount);
+               trace_xfs_refcount_get(cur->bc_mp, cur->bc_private.a.agno,
+                               irec);
+       }
+       return error;
+}
+
+/*
+ * Update the record referred to by cur to the value given
+ * by [bno, len, refcount].
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_update(
+       struct xfs_btree_cur            *cur,
+       struct xfs_refcount_irec        *irec)
+{
+       union xfs_btree_rec     rec;
+       int                     error;
+
+       trace_xfs_refcount_update(cur->bc_mp, cur->bc_private.a.agno, irec);
+       rec.refc.rc_startblock = cpu_to_be32(irec->rc_startblock);
+       rec.refc.rc_blockcount = cpu_to_be32(irec->rc_blockcount);
+       rec.refc.rc_refcount = cpu_to_be32(irec->rc_refcount);
+       error = xfs_btree_update(cur, &rec);
+       if (error)
+               trace_xfs_refcount_update_error(cur->bc_mp,
+                               cur->bc_private.a.agno, error, _RET_IP_);
+       return error;
+}
+
+/*
+ * Insert the record referred to by cur to the value given
+ * by [bno, len, refcount].
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_insert(
+       struct xfs_btree_cur            *cur,
+       struct xfs_refcount_irec        *irec,
+       int                             *i)
+{
+       int                             error;
+
+       trace_xfs_refcount_insert(cur->bc_mp, cur->bc_private.a.agno, irec);
+       cur->bc_rec.rc.rc_startblock = irec->rc_startblock;
+       cur->bc_rec.rc.rc_blockcount = irec->rc_blockcount;
+       cur->bc_rec.rc.rc_refcount = irec->rc_refcount;
+       error = xfs_btree_insert(cur, i);
+       XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
+out_error:
+       if (error)
+               trace_xfs_refcount_insert_error(cur->bc_mp,
+                               cur->bc_private.a.agno, error, _RET_IP_);
+       return error;
+}
+
+/*
+ * Remove the record referred to by cur, then set the pointer to the spot
+ * where the record could be re-inserted, in case we want to increment or
+ * decrement the cursor.
+ * This either works (return 0) or gets an EFSCORRUPTED error.
+ */
+STATIC int
+xfs_refcount_delete(
+       struct xfs_btree_cur    *cur,
+       int                     *i)
+{
+       struct xfs_refcount_irec        irec;
+       int                     found_rec;
+       int                     error;
+
+       error = xfs_refcount_get_rec(cur, &irec, &found_rec);
+       if (error)
+               goto out_error;
+       XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, found_rec == 1, out_error);
+       trace_xfs_refcount_delete(cur->bc_mp, cur->bc_private.a.agno, &irec);
+       error = xfs_btree_delete(cur, i);
+       XFS_WANT_CORRUPTED_GOTO(cur->bc_mp, *i == 1, out_error);
+       if (error)
+               goto out_error;
+       error = xfs_refcount_lookup_ge(cur, irec.rc_startblock, &found_rec);
+out_error:
+       if (error)
+               trace_xfs_refcount_delete_error(cur->bc_mp,
+                               cur->bc_private.a.agno, error, _RET_IP_);
+       return error;
+}
diff --git a/fs/xfs/libxfs/xfs_refcount.h b/fs/xfs/libxfs/xfs_refcount.h
new file mode 100644 (file)
index 0000000..4dc335a
--- /dev/null
@@ -0,0 +1,30 @@
+/*
+ * 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.
+ */
+#ifndef __XFS_REFCOUNT_H__
+#define __XFS_REFCOUNT_H__
+
+extern int xfs_refcount_lookup_le(struct xfs_btree_cur *cur,
+               xfs_agblock_t bno, int *stat);
+extern int xfs_refcount_lookup_ge(struct xfs_btree_cur *cur,
+               xfs_agblock_t bno, int *stat);
+extern int xfs_refcount_get_rec(struct xfs_btree_cur *cur,
+               struct xfs_refcount_irec *irec, int *stat);
+
+#endif /* __XFS_REFCOUNT_H__ */
index 359cf0c3a6b434489d4b8067b9c98a9ca93d743d..81d58b09af718eccb6851f7ec923af1756ca4d69 100644 (file)
@@ -34,6 +34,7 @@
 #include "xfs_cksum.h"
 #include "xfs_trans.h"
 #include "xfs_bit.h"
+#include "xfs_rmap.h"
 
 static struct xfs_btree_cur *
 xfs_refcountbt_dup_cursor(
@@ -44,6 +45,178 @@ xfs_refcountbt_dup_cursor(
                        cur->bc_private.a.dfops);
 }
 
+STATIC void
+xfs_refcountbt_set_root(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *ptr,
+       int                     inc)
+{
+       struct xfs_buf          *agbp = cur->bc_private.a.agbp;
+       struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
+       xfs_agnumber_t          seqno = be32_to_cpu(agf->agf_seqno);
+       struct xfs_perag        *pag = xfs_perag_get(cur->bc_mp, seqno);
+
+       ASSERT(ptr->s != 0);
+
+       agf->agf_refcount_root = ptr->s;
+       be32_add_cpu(&agf->agf_refcount_level, inc);
+       pag->pagf_refcount_level += inc;
+       xfs_perag_put(pag);
+
+       xfs_alloc_log_agf(cur->bc_tp, agbp,
+                       XFS_AGF_REFCOUNT_ROOT | XFS_AGF_REFCOUNT_LEVEL);
+}
+
+STATIC int
+xfs_refcountbt_alloc_block(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *start,
+       union xfs_btree_ptr     *new,
+       int                     *stat)
+{
+       struct xfs_buf          *agbp = cur->bc_private.a.agbp;
+       struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
+       struct xfs_alloc_arg    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.type = XFS_ALLOCTYPE_NEAR_BNO;
+       args.fsbno = XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_private.a.agno,
+                       xfs_refc_block(args.mp));
+       args.firstblock = args.fsbno;
+       xfs_rmap_ag_owner(&args.oinfo, XFS_RMAP_OWN_REFC);
+       args.minlen = args.maxlen = args.prod = 1;
+
+       error = xfs_alloc_vextent(&args);
+       if (error)
+               goto out_error;
+       trace_xfs_refcountbt_alloc_block(cur->bc_mp, cur->bc_private.a.agno,
+                       args.agbno, 1);
+       if (args.fsbno == NULLFSBLOCK) {
+               XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+               *stat = 0;
+               return 0;
+       }
+       ASSERT(args.agno == cur->bc_private.a.agno);
+       ASSERT(args.len == 1);
+
+       new->s = cpu_to_be32(args.agbno);
+       be32_add_cpu(&agf->agf_refcount_blocks, 1);
+       xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
+
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_EXIT);
+       *stat = 1;
+       return 0;
+
+out_error:
+       XFS_BTREE_TRACE_CURSOR(cur, XBT_ERROR);
+       return error;
+}
+
+STATIC int
+xfs_refcountbt_free_block(
+       struct xfs_btree_cur    *cur,
+       struct xfs_buf          *bp)
+{
+       struct xfs_mount        *mp = cur->bc_mp;
+       struct xfs_buf          *agbp = cur->bc_private.a.agbp;
+       struct xfs_agf          *agf = XFS_BUF_TO_AGF(agbp);
+       xfs_fsblock_t           fsbno = XFS_DADDR_TO_FSB(mp, XFS_BUF_ADDR(bp));
+       struct xfs_owner_info   oinfo;
+
+       trace_xfs_refcountbt_free_block(cur->bc_mp, cur->bc_private.a.agno,
+                       XFS_FSB_TO_AGBNO(cur->bc_mp, fsbno), 1);
+       xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_REFC);
+       be32_add_cpu(&agf->agf_refcount_blocks, -1);
+       xfs_alloc_log_agf(cur->bc_tp, agbp, XFS_AGF_REFCOUNT_BLOCKS);
+       xfs_bmap_add_free(mp, cur->bc_private.a.dfops, fsbno, 1,
+                       &oinfo);
+
+       return 0;
+}
+
+STATIC int
+xfs_refcountbt_get_minrecs(
+       struct xfs_btree_cur    *cur,
+       int                     level)
+{
+       return cur->bc_mp->m_refc_mnr[level != 0];
+}
+
+STATIC int
+xfs_refcountbt_get_maxrecs(
+       struct xfs_btree_cur    *cur,
+       int                     level)
+{
+       return cur->bc_mp->m_refc_mxr[level != 0];
+}
+
+STATIC void
+xfs_refcountbt_init_key_from_rec(
+       union xfs_btree_key     *key,
+       union xfs_btree_rec     *rec)
+{
+       key->refc.rc_startblock = rec->refc.rc_startblock;
+}
+
+STATIC void
+xfs_refcountbt_init_high_key_from_rec(
+       union xfs_btree_key     *key,
+       union xfs_btree_rec     *rec)
+{
+       __u32                   x;
+
+       x = be32_to_cpu(rec->refc.rc_startblock);
+       x += be32_to_cpu(rec->refc.rc_blockcount) - 1;
+       key->refc.rc_startblock = cpu_to_be32(x);
+}
+
+STATIC void
+xfs_refcountbt_init_rec_from_cur(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_rec     *rec)
+{
+       rec->refc.rc_startblock = cpu_to_be32(cur->bc_rec.rc.rc_startblock);
+       rec->refc.rc_blockcount = cpu_to_be32(cur->bc_rec.rc.rc_blockcount);
+       rec->refc.rc_refcount = cpu_to_be32(cur->bc_rec.rc.rc_refcount);
+}
+
+STATIC void
+xfs_refcountbt_init_ptr_from_cur(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_ptr     *ptr)
+{
+       struct xfs_agf          *agf = XFS_BUF_TO_AGF(cur->bc_private.a.agbp);
+
+       ASSERT(cur->bc_private.a.agno == be32_to_cpu(agf->agf_seqno));
+       ASSERT(agf->agf_refcount_root != 0);
+
+       ptr->s = agf->agf_refcount_root;
+}
+
+STATIC __int64_t
+xfs_refcountbt_key_diff(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_key     *key)
+{
+       struct xfs_refcount_irec        *rec = &cur->bc_rec.rc;
+       struct xfs_refcount_key         *kp = &key->refc;
+
+       return (__int64_t)be32_to_cpu(kp->rc_startblock) - rec->rc_startblock;
+}
+
+STATIC __int64_t
+xfs_refcountbt_diff_two_keys(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_key     *k1,
+       union xfs_btree_key     *k2)
+{
+       return (__int64_t)be32_to_cpu(k1->refc.rc_startblock) -
+                         be32_to_cpu(k2->refc.rc_startblock);
+}
+
 STATIC bool
 xfs_refcountbt_verify(
        struct xfs_buf          *bp)
@@ -106,12 +279,50 @@ const struct xfs_buf_ops xfs_refcountbt_buf_ops = {
        .verify_write           = xfs_refcountbt_write_verify,
 };
 
+#if defined(DEBUG) || defined(XFS_WARN)
+STATIC int
+xfs_refcountbt_keys_inorder(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_key     *k1,
+       union xfs_btree_key     *k2)
+{
+       return be32_to_cpu(k1->refc.rc_startblock) <
+              be32_to_cpu(k2->refc.rc_startblock);
+}
+
+STATIC int
+xfs_refcountbt_recs_inorder(
+       struct xfs_btree_cur    *cur,
+       union xfs_btree_rec     *r1,
+       union xfs_btree_rec     *r2)
+{
+       return  be32_to_cpu(r1->refc.rc_startblock) +
+               be32_to_cpu(r1->refc.rc_blockcount) <=
+               be32_to_cpu(r2->refc.rc_startblock);
+}
+#endif
+
 static const struct xfs_btree_ops xfs_refcountbt_ops = {
        .rec_len                = sizeof(struct xfs_refcount_rec),
        .key_len                = sizeof(struct xfs_refcount_key),
 
        .dup_cursor             = xfs_refcountbt_dup_cursor,
+       .set_root               = xfs_refcountbt_set_root,
+       .alloc_block            = xfs_refcountbt_alloc_block,
+       .free_block             = xfs_refcountbt_free_block,
+       .get_minrecs            = xfs_refcountbt_get_minrecs,
+       .get_maxrecs            = xfs_refcountbt_get_maxrecs,
+       .init_key_from_rec      = xfs_refcountbt_init_key_from_rec,
+       .init_high_key_from_rec = xfs_refcountbt_init_high_key_from_rec,
+       .init_rec_from_cur      = xfs_refcountbt_init_rec_from_cur,
+       .init_ptr_from_cur      = xfs_refcountbt_init_ptr_from_cur,
+       .key_diff               = xfs_refcountbt_key_diff,
        .buf_ops                = &xfs_refcountbt_buf_ops,
+       .diff_two_keys          = xfs_refcountbt_diff_two_keys,
+#if defined(DEBUG) || defined(XFS_WARN)
+       .keys_inorder           = xfs_refcountbt_keys_inorder,
+       .recs_inorder           = xfs_refcountbt_recs_inorder,
+#endif
 };
 
 /*