From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: Gao Xiang <hsiangkao@redhat.com>
Cc: linux-xfs@vger.kernel.org, Dave Chinner <dchinner@redhat.com>,
Eric Sandeen <sandeen@sandeen.net>
Subject: Re: [RFC PATCH] xfs_repair: fix rebuilding btree node block less than minrecs
Date: Tue, 9 Jun 2020 15:12:39 -0700 [thread overview]
Message-ID: <20200609221239.GE11245@magnolia> (raw)
In-Reply-To: <20200609201423.GD11245@magnolia>
On Tue, Jun 09, 2020 at 01:14:23PM -0700, Darrick J. Wong wrote:
> On Tue, Jun 09, 2020 at 07:40:53PM +0800, Gao Xiang wrote:
> > In production, we found that sometimes xfs_repair phase 5
> > rebuilds freespace node block with pointers less than minrecs
> > and if we trigger xfs_repair again it would report such
> > the following message:
> >
> > bad btree nrecs (39, min=40, max=80) in btbno block 0/7882
> >
> > The background is that xfs_repair starts to rebuild AGFL
> > after the freespace btree is settled in phase 5 so we may
> > need to leave necessary room in advance for each btree
> > leaves in order to avoid freespace btree split and then
> > result in AGFL rebuild fails. The old mathematics uses
> > ceil(num_extents / maxrecs) to decide the number of node
> > blocks. That would be fine without leaving extra space
> > since minrecs = maxrecs / 2 but if some slack was decreased
> > from maxrecs, the result would be larger than what is
> > expected and cause num_recs_pb less than minrecs, i.e:
> >
> > num_extents = 79, adj_maxrecs = 80 - 2 (slack) = 78
> >
> > so we'd get
> >
> > num_blocks = ceil(79 / 78) = 2,
> > num_recs_pb = 79 / 2 = 39, which is less than
> > minrecs = 80 / 2 = 40
> >
> > OTOH, btree bulk loading code behaves in a different way.
> > As in xfs_btree_bload_level_geometry it wrote
> >
> > num_blocks = floor(num_extents / maxrecs)
> >
> > which will never go below minrecs. And when it goes
> > above maxrecs, just increment num_blocks and recalculate
> > so we can get the reasonable results.
> >
> > In the long term, btree bulk loader will replace the current
> > repair code as well as to resolve AGFL dependency issue.
> > But we may still want to look for a backportable solution
> > for stable versions. Hence, use the same logic to avoid the
> > freespace btree minrecs underflow for now.
> >
> > Cc: "Darrick J. Wong" <darrick.wong@oracle.com>
> > Cc: Dave Chinner <dchinner@redhat.com>
> > Cc: Eric Sandeen <sandeen@sandeen.net>
> > Fixes: 9851fd79bfb1 ("repair: AGFL rebuild fails if btree split required")
> > Signed-off-by: Gao Xiang <hsiangkao@redhat.com>
> > ---
> > not heavy tested yet..
> >
> > repair/phase5.c | 101 +++++++++++++++++++++---------------------------
> > 1 file changed, 45 insertions(+), 56 deletions(-)
> >
> > diff --git a/repair/phase5.c b/repair/phase5.c
> > index abae8a08..997804a5 100644
> > --- a/repair/phase5.c
> > +++ b/repair/phase5.c
> > @@ -348,11 +348,29 @@ finish_cursor(bt_status_t *curs)
> > * failure at runtime. Hence leave a couple of records slack space in
> > * each block to allow immediate modification of the tree without
> > * requiring splits to be done.
> > - *
> > - * XXX(hch): any reason we don't just look at mp->m_alloc_mxr?
> > */
> > -#define XR_ALLOC_BLOCK_MAXRECS(mp, level) \
> > - (libxfs_allocbt_maxrecs((mp), (mp)->m_sb.sb_blocksize, (level) == 0) - 2)
> > +static void
> > +compute_level_geometry(xfs_mount_t *mp, bt_stat_level_t *lptr,
> > + uint64_t nr_this_level, bool leaf)
>
> Please don't use structure typedefs here.
>
> Also, please indent the argument list like the rest of xfs code, e.g.
>
> static void
> compute_bnobt_geometry(
> struct xfs_mount *mp,
> struct bt_stat_level *lptr,
>
> (etc)
>
> > +{
> > + unsigned int maxrecs = mp->m_alloc_mxr[!leaf];
> > + int slack = leaf ? 2 : 0;
> > + unsigned int desired_npb;
> > +
> > + desired_npb = max(mp->m_alloc_mnr[!leaf], maxrecs - slack);
> > + lptr->num_recs_tot = nr_this_level;
> > + lptr->num_blocks = max(1ULL, nr_this_level / desired_npb);
> > +
> > + lptr->num_recs_pb = nr_this_level / lptr->num_blocks;
> > + lptr->modulo = nr_this_level % lptr->num_blocks;
> > + if (lptr->num_recs_pb > maxrecs || (lptr->num_recs_pb == maxrecs &&
> > + lptr->modulo)) {
>
> Indentation....
>
> if (lptr->num_recs_pb > maxrecs ||
> (lptr->num_recs_pb == maxrecs && lptr->modulo)) {
> lptr->num_blocks++;
> ...
> }
>
> > + lptr->num_blocks++;
> > +
> > + lptr->num_recs_pb = nr_this_level / lptr->num_blocks;
> > + lptr->modulo = nr_this_level % lptr->num_blocks;
Oops, I hit send too quickly; the computation looks ok to me.
Granted that's probably due to having written another btree geometry
calculation function. :D
Also, I think init_rmapbt_cursor does a similar trick and therefore
suffers from a similar bug. See the comment "Leave enough slack in the
rmapbt that we can insert..." around line 1412 or so?
--D
> > + }
> > +}
> >
> > /*
> > * this calculates a freespace cursor for an ag.
> > @@ -370,6 +388,7 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > int i;
> > int extents_used;
> > int extra_blocks;
> > + uint64_t old_blocks;
> > bt_stat_level_t *lptr;
> > bt_stat_level_t *p_lptr;
> > extent_tree_node_t *ext_ptr;
> > @@ -388,10 +407,7 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > * of the tree and set up the cursor for the leaf level
> > * (note that the same code is duplicated further down)
> > */
> > - lptr->num_blocks = howmany(num_extents, XR_ALLOC_BLOCK_MAXRECS(mp, 0));
> > - lptr->num_recs_pb = num_extents / lptr->num_blocks;
> > - lptr->modulo = num_extents % lptr->num_blocks;
> > - lptr->num_recs_tot = num_extents;
> > + compute_level_geometry(mp, lptr, num_extents, true);
> > level = 1;
> >
> > #ifdef XR_BLD_FREE_TRACE
> > @@ -406,18 +422,12 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > * per level is the # of blocks in the level below it
> > */
> > if (lptr->num_blocks > 1) {
> > - for (; btree_curs->level[level - 1].num_blocks > 1
> > - && level < XFS_BTREE_MAXLEVELS;
> > - level++) {
> > + do {
> > + p_lptr = lptr;
> > lptr = &btree_curs->level[level];
> > - p_lptr = &btree_curs->level[level - 1];
> > - lptr->num_blocks = howmany(p_lptr->num_blocks,
> > - XR_ALLOC_BLOCK_MAXRECS(mp, level));
> > - lptr->modulo = p_lptr->num_blocks
> > - % lptr->num_blocks;
> > - lptr->num_recs_pb = p_lptr->num_blocks
> > - / lptr->num_blocks;
> > - lptr->num_recs_tot = p_lptr->num_blocks;
> > +
> > + compute_level_geometry(mp, lptr,
> > + p_lptr->num_blocks, false);
> > #ifdef XR_BLD_FREE_TRACE
> > fprintf(stderr, "%s %d %d %d %d %d\n", __func__,
> > level,
> > @@ -426,7 +436,9 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > lptr->modulo,
> > lptr->num_recs_tot);
> > #endif
> > - }
> > + level++;
> > + } while (lptr->num_blocks > 1);
> > + ASSERT (level < XFS_BTREE_MAXLEVELS);
> > }
> >
> > ASSERT(lptr->num_blocks == 1);
> > @@ -496,8 +508,11 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > * see if the number of leaf blocks will change as a result
> > * of the number of extents changing
> > */
> > - if (howmany(num_extents, XR_ALLOC_BLOCK_MAXRECS(mp, 0))
> > - != btree_curs->level[0].num_blocks) {
> > + old_blocks = btree_curs->level[0].num_blocks;
> > + compute_level_geometry(mp, &btree_curs->level[0], num_extents, true);
> > + extra_blocks = 0;
> > +
> > + if (old_blocks != btree_curs->level[0].num_blocks) {
> > /*
> > * yes -- recalculate the cursor. If the number of
> > * excess (overallocated) blocks is < xfs_agfl_size/2, we're ok.
> > @@ -553,30 +568,20 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> > }
> >
> > lptr = &btree_curs->level[0];
> > - lptr->num_blocks = howmany(num_extents,
> > - XR_ALLOC_BLOCK_MAXRECS(mp, 0));
> > - lptr->num_recs_pb = num_extents / lptr->num_blocks;
> > - lptr->modulo = num_extents % lptr->num_blocks;
> > - lptr->num_recs_tot = num_extents;
> > level = 1;
> >
> > /*
> > * if we need more levels, set them up
> > */
> > if (lptr->num_blocks > 1) {
> > - for (level = 1; btree_curs->level[level-1].num_blocks
> > - > 1 && level < XFS_BTREE_MAXLEVELS;
> > - level++) {
> > - lptr = &btree_curs->level[level];
> > - p_lptr = &btree_curs->level[level-1];
> > - lptr->num_blocks = howmany(p_lptr->num_blocks,
> > - XR_ALLOC_BLOCK_MAXRECS(mp, level));
> > - lptr->modulo = p_lptr->num_blocks
> > - % lptr->num_blocks;
> > - lptr->num_recs_pb = p_lptr->num_blocks
> > - / lptr->num_blocks;
> > - lptr->num_recs_tot = p_lptr->num_blocks;
> > - }
> > + do {
> > + p_lptr = lptr;
> > + lptr = &btree_curs->level[level++];
> > +
> > + compute_level_geometry(mp, lptr,
> > + p_lptr->num_blocks, false);
> > + } while (lptr->num_blocks > 1);
>
> /me wonders why it's necessary to wrap the do...while in an if test,
> as opposed to using a while loop with the if test up front?
>
> --D
>
> > + ASSERT (level < XFS_BTREE_MAXLEVELS);
> > }
> > ASSERT(lptr->num_blocks == 1);
> > btree_curs->num_levels = level;
> > @@ -591,22 +596,6 @@ calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
> >
> > ASSERT(blocks_allocated_total >= blocks_needed);
> > extra_blocks = blocks_allocated_total - blocks_needed;
> > - } else {
> > - if (extents_used > 0) {
> > - /*
> > - * reset the leaf level geometry to account
> > - * for consumed extents. we can leave the
> > - * rest of the cursor alone since the number
> > - * of leaf blocks hasn't changed.
> > - */
> > - lptr = &btree_curs->level[0];
> > -
> > - lptr->num_recs_pb = num_extents / lptr->num_blocks;
> > - lptr->modulo = num_extents % lptr->num_blocks;
> > - lptr->num_recs_tot = num_extents;
> > - }
> > -
> > - extra_blocks = 0;
> > }
> >
> > btree_curs->num_tot_blocks = blocks_allocated_pt;
> > --
> > 2.18.1
> >
next prev parent reply other threads:[~2020-06-09 22:12 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-06-09 11:40 [RFC PATCH] xfs_repair: fix rebuilding btree node block less than minrecs Gao Xiang
2020-06-09 14:35 ` Gao Xiang
2020-06-09 20:14 ` Darrick J. Wong
2020-06-09 22:12 ` Darrick J. Wong [this message]
2020-06-10 1:04 ` Gao Xiang
2020-06-10 3:58 ` [RFC PATCH v2] xfs_repair: fix rebuilding btree " Gao Xiang
2020-06-10 5:26 ` [RFC PATCH v3] " Gao Xiang
2020-06-16 16:11 ` Darrick J. Wong
2020-06-16 17:26 ` Gao Xiang
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20200609221239.GE11245@magnolia \
--to=darrick.wong@oracle.com \
--cc=dchinner@redhat.com \
--cc=hsiangkao@redhat.com \
--cc=linux-xfs@vger.kernel.org \
--cc=sandeen@sandeen.net \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox