From: "Darrick J. Wong" <djwong@kernel.org>
To: Brian Foster <bfoster@redhat.com>
Cc: sandeen@sandeen.net, linux-xfs@vger.kernel.org
Subject: Re: [PATCH 07/10] xfs_repair: set NEEDSREPAIR when we deliberately corrupt directories
Date: Tue, 9 Feb 2021 11:43:17 -0800 [thread overview]
Message-ID: <20210209194317.GY7193@magnolia> (raw)
In-Reply-To: <20210209191422.GL14273@bfoster>
On Tue, Feb 09, 2021 at 02:14:22PM -0500, Brian Foster wrote:
> On Tue, Feb 09, 2021 at 10:35:42AM -0800, Darrick J. Wong wrote:
> > On Tue, Feb 09, 2021 at 12:20:59PM -0500, Brian Foster wrote:
> > > On Mon, Feb 08, 2021 at 08:10:44PM -0800, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <djwong@kernel.org>
> > > >
> > > > There are a few places in xfs_repair's directory checking code where we
> > > > deliberately corrupt a directory entry as a sentinel to trigger a
> > > > correction in later repair phase. In the mean time, the filesystem is
> > > > inconsistent, so set the needsrepair flag to force a re-run of repair if
> > > > the system goes down.
> > > >
> > > > Signed-off-by: Darrick J. Wong <djwong@kernel.org>
> > > > ---
> > >
> > > Hmm.. this seems orthogonal to the rest of the series. I'm sure we can
> > > come up with various additional uses for the bit, but it seems a little
> > > odd to me that repair might set it in some cases after a crash but not
> > > others (if the filesystem happens to already be corrupt, for example).
> >
> > <nod> Another option I thought of is to add a hook to the buffer cache
> > so that the first time anyone tries to bwrite a buffer (either directly
> > or via a delwri list or normal buffer cache writeback) we'll also set
> > needsrepair on the ondisk primary super. That would protect us against
> > other scenarios like crashing after writing a new AGF but before writing
> > the new AGI, where the fs is left in an indeterminate state.
> >
>
> Yeah, that _seems_ more appropriate to me. It's not immediately clear to
> me what the implementation should look like, but in general behavior
> that sets needsrepair on first modification and clears it as a final
> step sounds more practical. Then the behavior can be easily explained as
> "once repair starts on an fs, it must be completed before it is allowed
> to mount." I do think this should be lifted off for a followon series so
> we can make progress on the feature upgrade bits without growing more
> requirements and complexity..
Oh, definitely. I'll withdraw this patch for now in the interests of
getting everything else going for Eric. :)
--D
> Brian
>
> > Hmm, maybe I should pursue /that/ instead.
> >
> > --D
> >
> > > Brian
> > >
> > > > repair/agheader.h | 2 ++
> > > > repair/dir2.c | 3 +++
> > > > repair/phase6.c | 7 +++++++
> > > > repair/xfs_repair.c | 37 +++++++++++++++++++++++++++++++++++++
> > > > 4 files changed, 49 insertions(+)
> > > >
> > > >
> > > > diff --git a/repair/agheader.h b/repair/agheader.h
> > > > index a63827c8..fa6fe596 100644
> > > > --- a/repair/agheader.h
> > > > +++ b/repair/agheader.h
> > > > @@ -82,3 +82,5 @@ typedef struct fs_geo_list {
> > > > #define XR_AG_AGF 0x2
> > > > #define XR_AG_AGI 0x4
> > > > #define XR_AG_SB_SEC 0x8
> > > > +
> > > > +void force_needsrepair(struct xfs_mount *mp);
> > > > diff --git a/repair/dir2.c b/repair/dir2.c
> > > > index eabdb4f2..922b8a3e 100644
> > > > --- a/repair/dir2.c
> > > > +++ b/repair/dir2.c
> > > > @@ -15,6 +15,7 @@
> > > > #include "da_util.h"
> > > > #include "prefetch.h"
> > > > #include "progress.h"
> > > > +#include "agheader.h"
> > > >
> > > > /*
> > > > * Known bad inode list. These are seen when the leaf and node
> > > > @@ -774,6 +775,7 @@ _("entry at block %u offset %" PRIdPTR " in directory inode %" PRIu64
> > > > do_warn(
> > > > _("\tclearing inode number in entry at offset %" PRIdPTR "...\n"),
> > > > (intptr_t)ptr - (intptr_t)d);
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > *dirty = 1;
> > > > } else {
> > > > @@ -914,6 +916,7 @@ _("entry \"%*.*s\" in directory inode %" PRIu64 " points to self: "),
> > > > */
> > > > if (junkit) {
> > > > if (!no_modify) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > *dirty = 1;
> > > > do_warn(_("clearing entry\n"));
> > > > diff --git a/repair/phase6.c b/repair/phase6.c
> > > > index 14464bef..5ecbe9b2 100644
> > > > --- a/repair/phase6.c
> > > > +++ b/repair/phase6.c
> > > > @@ -1649,6 +1649,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("entry \"%s\" in directory inode %" PRIu64 " points to non-existent inode %" PRIu64 ""),
> > > > fname, ip->i_ino, inum)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1666,6 +1667,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("entry \"%s\" in directory inode %" PRIu64 " points to free inode %" PRIu64),
> > > > fname, ip->i_ino, inum)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1684,6 +1686,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("%s (ino %" PRIu64 ") in root (%" PRIu64 ") is not a directory"),
> > > > ORPHANAGE, inum, ip->i_ino)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1706,6 +1709,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("entry \"%s\" (ino %" PRIu64 ") in dir %" PRIu64 " is a duplicate name"),
> > > > fname, inum, ip->i_ino)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1737,6 +1741,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("entry \"%s\" (ino %" PRIu64 ") in dir %" PRIu64 " is not in the the first block"), fname,
> > > > inum, ip->i_ino)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1764,6 +1769,7 @@ longform_dir2_entry_check_data(
> > > > if (entry_junked(
> > > > _("entry \"%s\" in dir %" PRIu64 " is not the first entry"),
> > > > fname, inum, ip->i_ino)) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > }
> > > > @@ -1852,6 +1858,7 @@ _("entry \"%s\" in dir inode %" PRIu64 " inconsistent with .. value (%" PRIu64 "
> > > > orphanage_ino = 0;
> > > > nbad++;
> > > > if (!no_modify) {
> > > > + force_needsrepair(mp);
> > > > dep->name[0] = '/';
> > > > libxfs_dir2_data_log_entry(&da, bp, dep);
> > > > if (verbose)
> > > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > > > index f607afcb..9dc73854 100644
> > > > --- a/repair/xfs_repair.c
> > > > +++ b/repair/xfs_repair.c
> > > > @@ -754,6 +754,43 @@ clear_needsrepair(
> > > > libxfs_buf_relse(bp);
> > > > }
> > > >
> > > > +/*
> > > > + * Mark the filesystem as needing repair. This should only be called by code
> > > > + * that deliberately sets invalid sentinel values in the on-disk metadata to
> > > > + * trigger a later reconstruction, and only after we've settled the primary
> > > > + * super contents (i.e. after phase 1).
> > > > + */
> > > > +void
> > > > +force_needsrepair(
> > > > + struct xfs_mount *mp)
> > > > +{
> > > > + struct xfs_buf *bp;
> > > > + int error;
> > > > +
> > > > + if (!xfs_sb_version_hascrc(&mp->m_sb) ||
> > > > + xfs_sb_version_needsrepair(&mp->m_sb))
> > > > + return;
> > > > +
> > > > + bp = libxfs_getsb(mp);
> > > > + if (!bp || bp->b_error) {
> > > > + do_log(
> > > > + _("couldn't get superblock to set needsrepair, err=%d\n"),
> > > > + bp ? bp->b_error : ENOMEM);
> > > > + return;
> > > > + } else {
> > > > + mp->m_sb.sb_features_incompat |=
> > > > + XFS_SB_FEAT_INCOMPAT_NEEDSREPAIR;
> > > > + libxfs_sb_to_disk(bp->b_addr, &mp->m_sb);
> > > > +
> > > > + /* Force the primary super to disk immediately. */
> > > > + error = -libxfs_bwrite(bp);
> > > > + if (error)
> > > > + do_log(_("couldn't force needsrepair, err=%d\n"), error);
> > > > + }
> > > > + if (bp)
> > > > + libxfs_buf_relse(bp);
> > > > +}
> > > > +
> > > > int
> > > > main(int argc, char **argv)
> > > > {
> > > >
> > >
> >
>
next prev parent reply other threads:[~2021-02-09 19:47 UTC|newest]
Thread overview: 49+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-02-09 4:10 [PATCHSET v4 00/10] xfs: add the ability to flag a fs for repair Darrick J. Wong
2021-02-09 4:10 ` [PATCH 01/10] xfs_admin: clean up string quoting Darrick J. Wong
2021-02-09 9:07 ` Christoph Hellwig
2021-02-09 4:10 ` [PATCH 02/10] xfs_admin: support filesystems with realtime devices Darrick J. Wong
2021-02-09 9:08 ` Christoph Hellwig
2021-02-09 17:19 ` Brian Foster
2021-02-09 4:10 ` [PATCH 03/10] xfs_db: support the needsrepair feature flag in the version command Darrick J. Wong
2021-02-09 9:09 ` Christoph Hellwig
2021-02-09 17:15 ` Darrick J. Wong
2021-02-09 17:19 ` Brian Foster
2021-02-09 4:10 ` [PATCH 04/10] xfs_repair: fix unmount error message to have a newline Darrick J. Wong
2021-02-09 9:09 ` Christoph Hellwig
2021-02-09 4:10 ` [PATCH 05/10] xfs_repair: clear quota CHKD flags on the incore superblock too Darrick J. Wong
2021-02-09 9:10 ` Christoph Hellwig
2021-02-09 17:20 ` Brian Foster
2021-02-09 17:46 ` Darrick J. Wong
2021-02-09 4:10 ` [PATCH 06/10] xfs_repair: clear the needsrepair flag Darrick J. Wong
2021-02-09 9:12 ` Christoph Hellwig
2021-02-09 17:20 ` Brian Foster
2021-02-09 18:01 ` Darrick J. Wong
2021-02-09 4:10 ` [PATCH 07/10] xfs_repair: set NEEDSREPAIR when we deliberately corrupt directories Darrick J. Wong
2021-02-09 9:13 ` Christoph Hellwig
2021-02-09 18:45 ` Darrick J. Wong
2021-02-09 17:20 ` Brian Foster
2021-02-09 18:35 ` Darrick J. Wong
2021-02-09 19:14 ` Brian Foster
2021-02-09 19:43 ` Darrick J. Wong [this message]
2021-02-10 20:19 ` Eric Sandeen
2021-02-09 4:10 ` [PATCH 08/10] xfs_repair: allow setting the needsrepair flag Darrick J. Wong
2021-02-09 9:15 ` Christoph Hellwig
2021-02-09 14:41 ` Eric Sandeen
2021-02-09 16:47 ` Darrick J. Wong
2021-02-10 20:44 ` Eric Sandeen
2021-02-09 17:21 ` Brian Foster
2021-02-09 18:10 ` Darrick J. Wong
2021-02-10 20:26 ` Eric Sandeen
2021-02-09 4:10 ` [PATCH 09/10] xfs_repair: add a testing hook for NEEDSREPAIR Darrick J. Wong
2021-02-09 9:16 ` Christoph Hellwig
2021-02-09 17:21 ` Brian Foster
2021-02-09 18:17 ` Darrick J. Wong
2021-02-09 18:59 ` Brian Foster
2021-02-09 19:59 ` Darrick J. Wong
2021-02-09 20:32 ` Brian Foster
2021-02-10 21:41 ` Eric Sandeen
2021-02-11 1:30 ` Darrick J. Wong
2021-02-09 4:11 ` [PATCH 10/10] xfs_admin: support adding features to V5 filesystems Darrick J. Wong
2021-02-09 9:18 ` Christoph Hellwig
2021-02-09 17:22 ` Brian Foster
2021-02-09 18:22 ` Darrick J. Wong
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=20210209194317.GY7193@magnolia \
--to=djwong@kernel.org \
--cc=bfoster@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;
as well as URLs for NNTP newsgroup(s).