]> www.infradead.org Git - users/hch/xfs.git/commitdiff
xfs: wire up a new inode fork type for the realtime refcount
authorDarrick J. Wong <djwong@kernel.org>
Tue, 15 Oct 2024 19:40:20 +0000 (12:40 -0700)
committerDarrick J. Wong <djwong@kernel.org>
Fri, 1 Nov 2024 20:47:09 +0000 (13:47 -0700)
Plumb in the pieces we need to embed the root of the realtime refcount
btree in an inode's data fork, complete with new fork type and
on-disk interpretation functions.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
fs/xfs/libxfs/xfs_format.h
fs/xfs/libxfs/xfs_inode_fork.c
fs/xfs/libxfs/xfs_ondisk.h
fs/xfs/libxfs/xfs_rtrefcount_btree.c
fs/xfs/libxfs/xfs_rtrefcount_btree.h
fs/xfs/xfs_inode_item_recover.c

index ad02e4ad85d2c70744c7b674f8c0cb2a5c24185c..8e3536d7759e0f0dbf65b90b605bb53b162ee4d9 100644 (file)
@@ -1807,6 +1807,14 @@ typedef __be32 xfs_refcount_ptr_t;
  */
 #define        XFS_RTREFC_CRC_MAGIC    0x52434e54      /* 'RCNT' */
 
+/*
+ * rt refcount root header, on-disk form only.
+ */
+struct xfs_rtrefcount_root {
+       __be16          bb_level;       /* 0 is a leaf */
+       __be16          bb_numrecs;     /* current # of data records */
+};
+
 /* inode-rooted btree pointer type */
 typedef __be64 xfs_rtrefcount_ptr_t;
 
index 6eb918e8b1ca08e0a6ad818659cf3776c12efce5..de033e23a7f2caab63c30d2043447e661874a799 100644 (file)
@@ -28,6 +28,7 @@
 #include "xfs_health.h"
 #include "xfs_symlink_remote.h"
 #include "xfs_rtrmap_btree.h"
+#include "xfs_rtrefcount_btree.h"
 
 struct kmem_cache *xfs_ifork_cache;
 
@@ -284,8 +285,7 @@ xfs_iformat_data_fork(
                        /* same comment about growfs and rmap inodes applies */
                        if (!xfs_has_reflink(ip->i_mount))
                                return -EFSCORRUPTED;
-                       ASSERT(0); /* to be implemented later */
-                       return -EFSCORRUPTED;
+                       return xfs_iformat_rtrefcount(ip, dip);
                default:
                        xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__,
                                        dip, sizeof(*dip), __this_address);
@@ -612,7 +612,9 @@ xfs_iflush_fork(
                break;
 
        case XFS_DINODE_FMT_REFCOUNT:
-               ASSERT(0); /* to be implemented later */
+               ASSERT(whichfork == XFS_DATA_FORK);
+               if (iip->ili_fields & brootflag[whichfork])
+                       xfs_iflush_rtrefcount(ip, dip);
                break;
 
        default:
index efb035050c009cd9fc4325ff04e689d5ec2f92e2..a85ecddaa48eed37ecbe796fb055ff7c6f5332f8 100644 (file)
@@ -86,6 +86,7 @@ xfs_check_ondisk_structs(void)
        XFS_CHECK_STRUCT_SIZE(xfs_rtrmap_ptr_t,                 8);
        XFS_CHECK_STRUCT_SIZE(struct xfs_rtrmap_root,           4);
        XFS_CHECK_STRUCT_SIZE(xfs_rtrefcount_ptr_t,             8);
+       XFS_CHECK_STRUCT_SIZE(struct xfs_rtrefcount_root,       4);
 
        /*
         * m68k has problems with struct xfs_attr_leaf_name_remote, but we pad
index 4adabbd0185081f9e1229219409c117fa57d8cfe..cdbca2f21b91f61921ef73bc6d6e02ecc6595dd7 100644 (file)
@@ -77,6 +77,41 @@ xfs_rtrefcountbt_get_maxrecs(
        return cur->bc_mp->m_rtrefc_mxr[level != 0];
 }
 
+/*
+ * Calculate number of records in a realtime refcount btree inode root.
+ */
+unsigned int
+xfs_rtrefcountbt_droot_maxrecs(
+       unsigned int            blocklen,
+       bool                    leaf)
+{
+       blocklen -= sizeof(struct xfs_rtrefcount_root);
+
+       if (leaf)
+               return blocklen / sizeof(struct xfs_refcount_rec);
+       return blocklen / (2 * sizeof(struct xfs_refcount_key) +
+                       sizeof(xfs_rtrefcount_ptr_t));
+}
+
+/*
+ * Get the maximum records we could store in the on-disk format.
+ *
+ * For non-root nodes this is equivalent to xfs_rtrefcountbt_get_maxrecs, but
+ * for the root node this checks the available space in the dinode fork so that
+ * we can resize the in-memory buffer to match it.  After a resize to the
+ * maximum size this function returns the same value as
+ * xfs_rtrefcountbt_get_maxrecs for the root node, too.
+ */
+STATIC int
+xfs_rtrefcountbt_get_dmaxrecs(
+       struct xfs_btree_cur    *cur,
+       int                     level)
+{
+       if (level != cur->bc_nlevels - 1)
+               return cur->bc_mp->m_rtrefc_mxr[level != 0];
+       return xfs_rtrefcountbt_droot_maxrecs(cur->bc_ino.forksize, level == 0);
+}
+
 STATIC void
 xfs_rtrefcountbt_init_key_from_rec(
        union xfs_btree_key             *key,
@@ -247,6 +282,89 @@ xfs_rtrefcountbt_keys_contiguous(
                                 be32_to_cpu(key2->refc.rc_startblock));
 }
 
+static inline void
+xfs_rtrefcountbt_move_ptrs(
+       struct xfs_mount        *mp,
+       struct xfs_btree_block  *broot,
+       short                   old_size,
+       size_t                  new_size,
+       unsigned int            numrecs)
+{
+       void                    *dptr;
+       void                    *sptr;
+
+       sptr = xfs_rtrefcount_broot_ptr_addr(mp, broot, 1, old_size);
+       dptr = xfs_rtrefcount_broot_ptr_addr(mp, broot, 1, new_size);
+       memmove(dptr, sptr, numrecs * sizeof(xfs_rtrefcount_ptr_t));
+}
+
+static struct xfs_btree_block *
+xfs_rtrefcountbt_broot_realloc(
+       struct xfs_btree_cur    *cur,
+       unsigned int            new_numrecs)
+{
+       struct xfs_mount        *mp = cur->bc_mp;
+       struct xfs_ifork        *ifp = xfs_btree_ifork_ptr(cur);
+       struct xfs_btree_block  *broot;
+       unsigned int            new_size;
+       unsigned int            old_size = ifp->if_broot_bytes;
+       const unsigned int      level = cur->bc_nlevels - 1;
+
+       new_size = xfs_rtrefcount_broot_space_calc(mp, level, new_numrecs);
+
+       /* Handle the nop case quietly. */
+       if (new_size == old_size)
+               return ifp->if_broot;
+
+       if (new_size > old_size) {
+               unsigned int    old_numrecs;
+
+               /*
+                * If there wasn't any memory allocated before, just allocate
+                * it now and get out.
+                */
+               if (old_size == 0)
+                       return xfs_broot_realloc(ifp, new_size);
+
+               /*
+                * If there is already an existing if_broot, then we need to
+                * realloc it and possibly move the node block pointers because
+                * those are not butted up against the btree block header.
+                */
+               old_numrecs = xfs_rtrefcountbt_maxrecs(mp, old_size, level);
+               broot = xfs_broot_realloc(ifp, new_size);
+               if (level > 0)
+                       xfs_rtrefcountbt_move_ptrs(mp, broot, old_size,
+                                       new_size, old_numrecs);
+               goto out_broot;
+       }
+
+       /*
+        * We're reducing numrecs.  If we're going all the way to zero, just
+        * free the block.
+        */
+       ASSERT(ifp->if_broot != NULL && old_size > 0);
+       if (new_size == 0) {
+               xfs_broot_free(ifp);
+               return NULL;
+       }
+
+       /*
+        * Shrink the btree root by possibly moving the rtrmapbt pointers,
+        * since they are not butted up against the btree block header.  Then
+        * reallocate broot.
+        */
+       if (level > 0)
+               xfs_rtrefcountbt_move_ptrs(mp, ifp->if_broot, old_size,
+                               new_size, new_numrecs);
+       broot = xfs_broot_realloc(ifp, new_size);
+
+out_broot:
+       ASSERT(xfs_rtrefcount_droot_space(broot) <=
+              xfs_inode_fork_size(cur->bc_ino.ip, cur->bc_ino.whichfork));
+       return broot;
+}
+
 const struct xfs_btree_ops xfs_rtrefcountbt_ops = {
        .name                   = "rtrefcount",
        .type                   = XFS_BTREE_TYPE_INODE,
@@ -264,6 +382,7 @@ const struct xfs_btree_ops xfs_rtrefcountbt_ops = {
        .free_block             = xfs_btree_free_metafile_block,
        .get_minrecs            = xfs_rtrefcountbt_get_minrecs,
        .get_maxrecs            = xfs_rtrefcountbt_get_maxrecs,
+       .get_dmaxrecs           = xfs_rtrefcountbt_get_dmaxrecs,
        .init_key_from_rec      = xfs_rtrefcountbt_init_key_from_rec,
        .init_high_key_from_rec = xfs_rtrefcountbt_init_high_key_from_rec,
        .init_rec_from_cur      = xfs_rtrefcountbt_init_rec_from_cur,
@@ -274,6 +393,7 @@ const struct xfs_btree_ops xfs_rtrefcountbt_ops = {
        .keys_inorder           = xfs_rtrefcountbt_keys_inorder,
        .recs_inorder           = xfs_rtrefcountbt_recs_inorder,
        .keys_contiguous        = xfs_rtrefcountbt_keys_contiguous,
+       .broot_realloc          = xfs_rtrefcountbt_broot_realloc,
 };
 
 /* Allocate a new rt refcount btree cursor. */
@@ -457,3 +577,141 @@ xfs_rtrefcountbt_calc_reserves(
 
        return xfs_rtrefcountbt_max_size(mp, mp->m_sb.sb_rgextents);
 }
+
+/*
+ * Convert on-disk form of btree root to in-memory form.
+ */
+STATIC void
+xfs_rtrefcountbt_from_disk(
+       struct xfs_inode                *ip,
+       struct xfs_rtrefcount_root      *dblock,
+       int                             dblocklen,
+       struct xfs_btree_block          *rblock)
+{
+       struct xfs_mount                *mp = ip->i_mount;
+       struct xfs_refcount_key *fkp;
+       __be64                          *fpp;
+       struct xfs_refcount_key *tkp;
+       __be64                          *tpp;
+       struct xfs_refcount_rec *frp;
+       struct xfs_refcount_rec *trp;
+       unsigned int                    numrecs;
+       unsigned int                    maxrecs;
+       unsigned int                    rblocklen;
+
+       rblocklen = xfs_rtrefcount_broot_space(mp, dblock);
+
+       xfs_btree_init_block(mp, rblock, &xfs_rtrefcountbt_ops, 0, 0,
+                       ip->i_ino);
+
+       rblock->bb_level = dblock->bb_level;
+       rblock->bb_numrecs = dblock->bb_numrecs;
+
+       if (be16_to_cpu(rblock->bb_level) > 0) {
+               maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
+               fkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
+               tkp = xfs_rtrefcount_key_addr(rblock, 1);
+               fpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
+               tpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
+               numrecs = be16_to_cpu(dblock->bb_numrecs);
+               memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
+               memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
+       } else {
+               frp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
+               trp = xfs_rtrefcount_rec_addr(rblock, 1);
+               numrecs = be16_to_cpu(dblock->bb_numrecs);
+               memcpy(trp, frp, sizeof(*frp) * numrecs);
+       }
+}
+
+/* Load a realtime reference count btree root in from disk. */
+int
+xfs_iformat_rtrefcount(
+       struct xfs_inode        *ip,
+       struct xfs_dinode       *dip)
+{
+       struct xfs_mount        *mp = ip->i_mount;
+       struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
+       struct xfs_btree_block  *broot;
+       unsigned int            numrecs;
+       unsigned int            level;
+       int                     dsize;
+
+       dsize = XFS_DFORK_SIZE(dip, mp, XFS_DATA_FORK);
+       numrecs = be16_to_cpu(dfp->bb_numrecs);
+       level = be16_to_cpu(dfp->bb_level);
+
+       if (level > mp->m_rtrefc_maxlevels ||
+           xfs_rtrefcount_droot_space_calc(level, numrecs) > dsize)
+               return -EFSCORRUPTED;
+
+       broot = xfs_broot_realloc(xfs_ifork_ptr(ip, XFS_DATA_FORK),
+                       xfs_rtrefcount_broot_space_calc(mp, level, numrecs));
+       if (broot)
+               xfs_rtrefcountbt_from_disk(ip, dfp, dsize, broot);
+       return 0;
+}
+
+/*
+ * Convert in-memory form of btree root to on-disk form.
+ */
+void
+xfs_rtrefcountbt_to_disk(
+       struct xfs_mount                *mp,
+       struct xfs_btree_block          *rblock,
+       int                             rblocklen,
+       struct xfs_rtrefcount_root      *dblock,
+       int                             dblocklen)
+{
+       struct xfs_refcount_key *fkp;
+       __be64                          *fpp;
+       struct xfs_refcount_key *tkp;
+       __be64                          *tpp;
+       struct xfs_refcount_rec *frp;
+       struct xfs_refcount_rec *trp;
+       unsigned int                    maxrecs;
+       unsigned int                    numrecs;
+
+       ASSERT(rblock->bb_magic == cpu_to_be32(XFS_RTREFC_CRC_MAGIC));
+       ASSERT(uuid_equal(&rblock->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid));
+       ASSERT(rblock->bb_u.l.bb_blkno == cpu_to_be64(XFS_BUF_DADDR_NULL));
+       ASSERT(rblock->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK));
+       ASSERT(rblock->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK));
+
+       dblock->bb_level = rblock->bb_level;
+       dblock->bb_numrecs = rblock->bb_numrecs;
+
+       if (be16_to_cpu(rblock->bb_level) > 0) {
+               maxrecs = xfs_rtrefcountbt_droot_maxrecs(dblocklen, false);
+               fkp = xfs_rtrefcount_key_addr(rblock, 1);
+               tkp = xfs_rtrefcount_droot_key_addr(dblock, 1);
+               fpp = xfs_rtrefcount_broot_ptr_addr(mp, rblock, 1, rblocklen);
+               tpp = xfs_rtrefcount_droot_ptr_addr(dblock, 1, maxrecs);
+               numrecs = be16_to_cpu(rblock->bb_numrecs);
+               memcpy(tkp, fkp, 2 * sizeof(*fkp) * numrecs);
+               memcpy(tpp, fpp, sizeof(*fpp) * numrecs);
+       } else {
+               frp = xfs_rtrefcount_rec_addr(rblock, 1);
+               trp = xfs_rtrefcount_droot_rec_addr(dblock, 1);
+               numrecs = be16_to_cpu(rblock->bb_numrecs);
+               memcpy(trp, frp, sizeof(*frp) * numrecs);
+       }
+}
+
+/* Flush a realtime reference count btree root out to disk. */
+void
+xfs_iflush_rtrefcount(
+       struct xfs_inode        *ip,
+       struct xfs_dinode       *dip)
+{
+       struct xfs_ifork        *ifp = xfs_ifork_ptr(ip, XFS_DATA_FORK);
+       struct xfs_rtrefcount_root *dfp = XFS_DFORK_PTR(dip, XFS_DATA_FORK);
+
+       ASSERT(ifp->if_broot != NULL);
+       ASSERT(ifp->if_broot_bytes > 0);
+       ASSERT(xfs_rtrefcount_droot_space(ifp->if_broot) <=
+                       xfs_inode_fork_size(ip, XFS_DATA_FORK));
+       xfs_rtrefcountbt_to_disk(ip->i_mount, ifp->if_broot,
+                       ifp->if_broot_bytes, dfp,
+                       XFS_DFORK_SIZE(dip, ip->i_mount, XFS_DATA_FORK));
+}
index 3cd44590c9304c599acba0d7cf216d16120ae225..e558a10c4744ad63adc9b1a1c956170c9721d00b 100644 (file)
@@ -25,6 +25,7 @@ void xfs_rtrefcountbt_commit_staged_btree(struct xfs_btree_cur *cur,
 unsigned int xfs_rtrefcountbt_maxrecs(struct xfs_mount *mp,
                unsigned int blocklen, bool leaf);
 void xfs_rtrefcountbt_compute_maxlevels(struct xfs_mount *mp);
+unsigned int xfs_rtrefcountbt_droot_maxrecs(unsigned int blocklen, bool leaf);
 
 /*
  * Addresses of records, keys, and pointers within an incore rtrefcountbt block.
@@ -71,4 +72,115 @@ xfs_filblks_t xfs_rtrefcountbt_calc_reserves(struct xfs_mount *mp);
 unsigned long long xfs_rtrefcountbt_calc_size(struct xfs_mount *mp,
                unsigned long long len);
 
+/* Addresses of key, pointers, and records within an ondisk rtrefcount block. */
+
+static inline struct xfs_refcount_rec *
+xfs_rtrefcount_droot_rec_addr(
+       struct xfs_rtrefcount_root      *block,
+       unsigned int                    index)
+{
+       return (struct xfs_refcount_rec *)
+               ((char *)(block + 1) +
+                (index - 1) * sizeof(struct xfs_refcount_rec));
+}
+
+static inline struct xfs_refcount_key *
+xfs_rtrefcount_droot_key_addr(
+       struct xfs_rtrefcount_root      *block,
+       unsigned int                    index)
+{
+       return (struct xfs_refcount_key *)
+               ((char *)(block + 1) +
+                (index - 1) * sizeof(struct xfs_refcount_key));
+}
+
+static inline xfs_rtrefcount_ptr_t *
+xfs_rtrefcount_droot_ptr_addr(
+       struct xfs_rtrefcount_root      *block,
+       unsigned int                    index,
+       unsigned int                    maxrecs)
+{
+       return (xfs_rtrefcount_ptr_t *)
+               ((char *)(block + 1) +
+                maxrecs * sizeof(struct xfs_refcount_key) +
+                (index - 1) * sizeof(xfs_rtrefcount_ptr_t));
+}
+
+/*
+ * Address of pointers within the incore btree root.
+ *
+ * These are to be used when we know the size of the block and
+ * we don't have a cursor.
+ */
+static inline xfs_rtrefcount_ptr_t *
+xfs_rtrefcount_broot_ptr_addr(
+       struct xfs_mount        *mp,
+       struct xfs_btree_block  *bb,
+       unsigned int            index,
+       unsigned int            block_size)
+{
+       return xfs_rtrefcount_ptr_addr(bb, index,
+                       xfs_rtrefcountbt_maxrecs(mp, block_size, false));
+}
+
+/*
+ * Compute the space required for the incore btree root containing the given
+ * number of records.
+ */
+static inline size_t
+xfs_rtrefcount_broot_space_calc(
+       struct xfs_mount        *mp,
+       unsigned int            level,
+       unsigned int            nrecs)
+{
+       size_t                  sz = XFS_RTREFCOUNT_BLOCK_LEN;
+
+       if (level > 0)
+               return sz + nrecs * (sizeof(struct xfs_refcount_key) +
+                                    sizeof(xfs_rtrefcount_ptr_t));
+       return sz + nrecs * sizeof(struct xfs_refcount_rec);
+}
+
+/*
+ * Compute the space required for the incore btree root given the ondisk
+ * btree root block.
+ */
+static inline size_t
+xfs_rtrefcount_broot_space(struct xfs_mount *mp, struct xfs_rtrefcount_root *bb)
+{
+       return xfs_rtrefcount_broot_space_calc(mp, be16_to_cpu(bb->bb_level),
+                       be16_to_cpu(bb->bb_numrecs));
+}
+
+/* Compute the space required for the ondisk root block. */
+static inline size_t
+xfs_rtrefcount_droot_space_calc(
+       unsigned int            level,
+       unsigned int            nrecs)
+{
+       size_t                  sz = sizeof(struct xfs_rtrefcount_root);
+
+       if (level > 0)
+               return sz + nrecs * (sizeof(struct xfs_refcount_key) +
+                                    sizeof(xfs_rtrefcount_ptr_t));
+       return sz + nrecs * sizeof(struct xfs_refcount_rec);
+}
+
+/*
+ * Compute the space required for the ondisk root block given an incore root
+ * block.
+ */
+static inline size_t
+xfs_rtrefcount_droot_space(struct xfs_btree_block *bb)
+{
+       return xfs_rtrefcount_droot_space_calc(be16_to_cpu(bb->bb_level),
+                       be16_to_cpu(bb->bb_numrecs));
+}
+
+int xfs_iformat_rtrefcount(struct xfs_inode *ip, struct xfs_dinode *dip);
+void xfs_rtrefcountbt_to_disk(struct xfs_mount *mp,
+               struct xfs_btree_block *rblock, int rblocklen,
+               struct xfs_rtrefcount_root *dblock, int dblocklen);
+void xfs_iflush_rtrefcount(struct xfs_inode *ip, struct xfs_dinode *dip);
+
 #endif /* __XFS_RTREFCOUNT_BTREE_H__ */
index 6fb52cfdf6c89ad0b24e42cf4a7fe33488423a76..ee8ea8cfa3fef7abe7f260973cdc8f0f7416da5f 100644 (file)
@@ -23,6 +23,7 @@
 #include "xfs_icache.h"
 #include "xfs_bmap_btree.h"
 #include "xfs_rtrmap_btree.h"
+#include "xfs_rtrefcount_btree.h"
 
 STATIC void
 xlog_recover_inode_ra_pass2(
@@ -284,6 +285,9 @@ xlog_recover_inode_dbroot(
        case XFS_DINODE_FMT_RMAP:
                xfs_rtrmapbt_to_disk(mp, src, len, dfork, dsize);
                break;
+       case XFS_DINODE_FMT_REFCOUNT:
+               xfs_rtrefcountbt_to_disk(mp, src, len, dfork, dsize);
+               break;
        default:
                ASSERT(0);
                return -EFSCORRUPTED;