xfs_db, xfs_repair: improve CRC error detection
cover xfs_db, xfs_repair: improve CRC error detection
Hi folks,
After a conversion with a user on #IRC this morning, it was clear that xfs_repair and xfs_db weren't handling metadata blocks with CRC errors in them particularly well. xfs_metadump was reporting blocks with errors, but xfs_db was reporting them as having a correct CRC, which wasn't actually the case - they were unchecked, and the code saw the absence of error flags as meaning they were good.
Repair had a similar problem - buffers that were prefetched never had the verifier run on them when they were read by the checking code as they were uptodate in the cache. Hence the prefetch code needed to mark the buffers as unchecked so that the code that checked the metadata ran the verifier and appropriately.
This then showed up the fact that there were many places where repair was not catching the CRC error and rewriting the buffer to correct the bad CRC.
This then showed up that we weren't actually handling remote attribute properly for the CRC enabled format.
And so I fixed all of them. I've verified the code by manually corrupting blocks with xfs_db by writing garabges into unused regions of the blocks so that CRC errors are triggered. In each case repair detected the CRC error and took appropriate action. The CRC error was not found on a second run of xfs_repair. This really needs to be turned into a xfstest, but I haven't had time to do that yet. Any volunteers?
Anyway, these fixes mean we'll definitely need a 3.2.0-rc2 release in the not too distant future. Comments, flames and testing all welcome....
-Dave.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 1/9 db: don't claim unchecked CRCs are correct
From: Dave Chinner <[email protected]>
Currently xfs_db will claim the CRC on a structure is correct if the buffer is not marked with an error. However, buffers may have been read without a verifier, and hence have not had their CRCs validated. in this case, we shoul dreport "unchecked" rather than "correct". For example:
xfs_db> fsb 0x6003fxfs_db> type dir3xfs_db> p
dhdr.hdr.magic = 0x58444433
dhdr.hdr.crc = 0x2d0f9c9d (unchecked)
....
--- db/fprint.c | 15 ++++++++++++++- db/io.c | 2 ++ db/io.h | 12 +++++++++--- 3 files changed, 25 insertions(+), 4 deletions(-)
db/fprint.c +14 -1
@@ -206,7 +206,20 @@ fp_crc(__int64_t val;char *ok; - ok =iocur_crc_valid () ?"correct" :"bad" ; +switch (iocur_crc_valid ()) { +case -1 : + ok ="unchecked" ; +break ; +case 0 : + ok ="bad" ; +break ; +case 1 : + ok ="correct" ; +break ; +default : + ok ="unknown state" ; +break ; + } for (i = 0, bitpos = bit; i < count && !seenint();
db/io.c +2 -0
@@ -531,6 +531,8 @@ set_cur(return ; iocur_top->buf = bp->b_addr; iocur_top->bp = bp; +if (!ops) + bp->b_flags |=LIBXFS_B_UNCHECKED ; iocur_top->bb = d; iocur_top->blen = c;
db/io.h +9 -3
@@ -63,10 +63,16 @@ extern void set_cur(const struct typ *t, __int64_t d, int c, int ring_add, bbmap_t *bbmap);extern void ring_add (void ); -static inline bool +/* +* returns -1 for unchecked, 0 for bad and 1 for good +*/ +static inline int iocur_crc_valid () { -return (iocur_top->bp && - iocur_top->bp->b_error !=EFSBADCRC && +if (!iocur_top->bp) +return -1 ; +if (iocur_top->bp->b_flags &LIBXFS_B_UNCHECKED ) +return -1 ; +return (iocur_top->bp->b_error !=EFSBADCRC && (!iocur_top->ino_buf || iocur_top->ino_crc_ok)); }
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 1/9] db: don't claim unchecked CRCs are correct
Quoted 4 lines
db/fprint.c +4
-0+switch (iocur_crc_valid ()) { +case -1 : + ok ="unchecked" ; +break ;
db/fprint.c fp_crc( · line 222
Quoted 10 lines
db/fprint.c +10
-0+case 0 : + ok ="bad" ; +break ; +case 1 : + ok ="correct" ; +break ; +default : + ok ="unknown state" ; +break ; + }
We should have symbolic constants for these return values. But then again iocur_crc_valid only has a single caller currently, is it even worth the effort, or should we simply inline it?
Otherwise looks good,
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 1/9] db: don't claim unchecked CRCs are correct
Quoted 10 lines · Christoph Hellwig
Quoted 10 lines
db/fprint.c +10
-0+switch (iocur_crc_valid ()) { +case -1 : + ok ="unchecked" ; +break ; +case 0 : + ok ="bad" ; +break ; +case 1 : + ok ="correct" ; +break ;
Quoted 8 lines
db/fprint.c fp_crc( · line 222
Quoted 4 lines
db/fprint.c +4
-0+default : + ok ="unknown state" ; +break ; + }We should have symbolic constants for these return values. But then again
iocur_crc_validonly has a single caller currently, is it even worth the effort, or should we simply inline it?
If we grow another user, I'll add an enum for it...
Quoted 3 lines
Otherwise looks good,
Reviewed-by Christoph Hellwig <[email protected]>
Thanks.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
PATCH 2/9 db: verify buffer on type change
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Currently when the type command is run, we simply change the type associated with the buffer, but don't verify it. This results in unchecked CRCs being displayed. Hence when changing the type, run the verifier associated with the type to determine if the buffer contents is valid or not.
--- db/io.c | 25 +++++++++++++++++++++++++ db/io.h | 1 + db/type.c | 9 +++++---- 3 files changed, 31 insertions(+), 4 deletions(-)
db/io.c +25 -0
@@ -552,6 +552,31 @@ set_cur(ring_add (); } +void +set_iocur_type ( +const typ_t *t) +{ +const struct xfs_buf_ops *ops = t ? t->bops :NULL ; +struct xfs_buf *bp = iocur_top->bp; + + iocur_top->typ = t; + +/* verify the buffer if the type has one. */ +if (!bp) +return ; +if (!ops) { + bp->b_ops =NULL ; + bp->b_flags |=LIBXFS_B_UNCHECKED ; +return ; + } +if (!(bp->b_flags &LIBXFS_B_UPTODATE )) +return ; + bp->b_error =0 ; + bp->b_ops = ops; + bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; +} +static void stack_help (void ) {
db/io.h +1 -0
@@ -62,6 +62,7 @@ extern void write_cur(void);extern void set_cur (const struct typ *t,__int64_t d,int c,int ring_add,bbmap_t *bbmap);extern void ring_add (void ); +extern void set_iocur_type (const struct typ *t);/* * returns -1 for unchecked, 0 for bad and 1 for good
db/type.c +5 -4
@@ -162,10 +162,11 @@ type_f(if (tt ==NULL ) {dbprintf (_ ("no such type %s\n" ), argv[1 ]); }else { -if (iocur_top->typ ==NULL ) { -dbprintf (_ ("no current object\n" )); - }else { - iocur_top->typ = cur_typ = tt; +if (iocur_top->typ ==NULL ) +dbprintf (_ ("no current object\n" )); +else { + cur_typ = tt; +set_iocur_type (tt); } } }
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 2/9] db: verify buffer on type change
Quoted 7 lines
db/io.c +7
-0+void +set_iocur_type ( +const typ_t *t) +{ +const struct xfs_buf_ops *ops = t ? t->bops :NULL ; +struct xfs_buf *bp = iocur_top->bp; +
db/io.c set_cur( · line 571
Quoted 10 lines
db/io.c +10
-0+ iocur_top->typ = t; + +/* verify the buffer if the type has one. */ +if (!bp) +return ; +if (!ops) { + bp->b_ops =NULL ; + bp->b_flags |=LIBXFS_B_UNCHECKED ; +return ; + }
The only caller currently makes sure we never pass a NULL t argument, and I think keeping it that way is sensible. If we want to allow clearing the type we should add a separate clear_iocur_type helper for it.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 2/9] db: verify buffer on type change
Quoted 14 lines · Christoph Hellwig
Quoted 14 lines
db/io.c +14
-0+void +set_iocur_type ( +const typ_t *t) +{ +const struct xfs_buf_ops *ops = t ? t->bops :NULL ; +struct xfs_buf *bp = iocur_top->bp; + + iocur_top->typ = t; + +/* verify the buffer if the type has one. */ +if (!bp) +return ; +if (!ops) { + bp->b_ops =NULL ;
Quoted 8 lines
db/io.c set_cur( · line 571
Quoted 3 lines
db/io.c +3
-0+ bp->b_flags |=LIBXFS_B_UNCHECKED ; +return ; + }The only caller currently makes sure we never pass a NULL t argument, and I think keeping it that way is sensible. If we want to allow clearing the type we should add a separate
clear_iocur_typehelper for it.
Ok, I'll clear the conditional initialisation of *ops.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
PATCH 3/9 repair: ensure prefetched buffers have CRCs validated
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Prefetch currently does not do CRC validation when the IO completes due to the optimisation it performs and the fact that it does not know what the type of metadata into the buffer is supposed to be. Hence, mark all prefetched buffers as "suspect" so that when the end user tries to read it with a supplied validation function the validation is run even though the buffer was already in the cache.
--- include/libxfs.h | 1 + libxfs/rdwr.c | 36 +++++++++++++++++++++++++++++++----- repair/prefetch.c | 3 +++ 3 files changed, 35 insertions(+), 5 deletions(-)
include/libxfs.h +1 -0
@@ -333,6 +333,7 @@ enum xfs_buf_flags_t { /* b_flags bits */LIBXFS_B_STALE =0x0004 ,/* buffer marked as invalid */ LIBXFS_B_UPTODATE =0x0008 ,/* buffer is sync'd to disk */ LIBXFS_B_DISCONTIG =0x0010 ,/* discontiguous buffer */ + LIBXFS_B_UNCHECKED = 0x0020, /* needs verification */ };#define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))
libxfs/rdwr.c +31 -5
@@ -718,12 +718,25 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, bp =libxfs_getbuf (btp, blkno, len);if (!bp)return NULL ; -if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) + +/* +* if the buffer was prefetched, it is likely that it was not +* validated. Hence if we are supplied an ops function and the +* buffer is marked as unchecked, we need to validate it now. +*/ +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_error =0 ; + bp->b_ops = ops; + bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + }return bp; + }/* -* only set the ops on a cache miss (i.e. first physical read) as the -* verifier may change the ops to match the typ eof buffer it contains. +* Set the ops on a cache miss (i.e. first physical read) as the +* verifier may change the ops to match the type of buffer it contains. * A cache hit might reset the verifier to the original type if we set * it again, but it won't get called again and set to match the buffer * contents. *cough* xfs_da_node_buf_ops *cough*.
@@ -733,8 +746,10 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, error =libxfs_readbufr (btp, blkno, bp, len, flags);if (error) bp->b_error = error; -else if (bp->b_ops) +else if (bp->b_ops) { bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + }return bp; }
@@ -786,6 +801,14 @@ libxfs_readbuf_map(struct xfs_buftarg *btp, struct xfs_buf_map *map, int nmaps,return NULL ; bp->b_error =0 ; +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_ops = ops; + bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + } +return bp; + } bp->b_ops = ops;if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY )))return bp;
@@ -793,8 +816,10 @@ libxfs_readbuf_map(struct xfs_buftarg *btp, struct xfs_buf_map *map, int nmaps, error =libxfs_readbufr_map (btp, bp, flags);if (!error) { bp->b_flags |=LIBXFS_B_UPTODATE ; -if (bp->b_ops) +if (bp->b_ops) { bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + } }#ifdef IO_DEBUG printf("%lx: %s: read %lu bytes, error %d, blkno=%llu(%llu), %p\n",
@@ -889,6 +914,7 @@ libxfs_writebufr(xfs_buf_t *bp)if (!error) { bp->b_flags |=LIBXFS_B_UPTODATE ; bp->b_flags &= ~(LIBXFS_B_DIRTY |LIBXFS_B_EXIT ); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; }return error; }
repair/prefetch.c +3 -0
@@ -389,6 +389,7 @@ pf_read_inode_dirs( bp->b_ops = &xfs_inode_buf_ops; bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ;if (bp->b_error)return ;
@@ -460,6 +461,7 @@ pf_read_discontig(pthread_mutex_unlock (&args->lock);libxfs_readbufr_map (mp->m_ddev_targp, bp,0 ); + bp->b_flags |=LIBXFS_B_UNCHECKED ;libxfs_putbuf (bp);pthread_mutex_lock (&args->lock); }
@@ -583,6 +585,7 @@ pf_batch_read(break ;memcpy (XFS_BUF_PTR (bplist[i]), pbuf, size); bplist[i]->b_flags |=LIBXFS_B_UPTODATE ; + bplist[i]->b_flags |=LIBXFS_B_UNCHECKED ; len -= size;if (B_IS_INODE (XFS_BUF_PRIORITY (bplist[i])))pf_read_inode_dirs (args, bplist[i]);
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 3/9] repair: ensure prefetched buffers have CRCs validated
Quoted 21 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Prefetch currently does not do CRC validation when the IO completes due to the optimisation it performs and the fact that it does not know what the type of metadata into the buffer is supposed to be. Hence, mark all prefetched buffers as "suspect" so that when the end user tries to read it with a supplied validation function the validation is run even though the buffer was already in the cache.
Signed-off-by Dave Chinner <[email protected]>--- include/libxfs.h | 1 + libxfs/rdwr.c | 36 +++++++++++++++++++++++++++++++----- repair/prefetch.c | 3 +++ 3 files changed, 35 insertions(+), 5 deletions(-)include/libxfs.h +0
-0
include/libxfs.h enum xfs_buf_flags_t { /* b_flags bits */ · line 336
Quoted 4 lines
include/libxfs.h +1
-0@@ -333,6 +333,7 @@ enum xfs_buf_flags_t { /* b_flags bits */LIBXFS_B_STALE =0x0004 ,/* buffer marked as invalid */ LIBXFS_B_UPTODATE =0x0008 ,/* buffer is sync'd to disk */ LIBXFS_B_DISCONTIG =0x0010 ,/* discontiguous buffer */ + LIBXFS_B_UNCHECKED = 0x0020, /* needs verification */
This is used in the first couple patches, so it should probably be defined earlier (or shuffle those patches appropriately).
Quoted 14 lines
include/libxfs.h +0
-0};#define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))libxfs/rdwr.c +2
-1@@ -718,12 +718,25 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, bp =libxfs_getbuf (btp, blkno, len);if (!bp)return NULL ; -if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) + +/*
libxfs/rdwr.c libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, · line 732
Quoted 10 lines
libxfs/rdwr.c +10
-0+* if the buffer was prefetched, it is likely that it was not +* validated. Hence if we are supplied an ops function and the +* buffer is marked as unchecked, we need to validate it now. +*/ +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_error =0 ; + bp->b_ops = ops; + bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ;
Should we always expect an unchecked buffer to be read with an ops vector before being written? Even if so, this might look cleaner if we didn't encode the possibility of running a read verifier on a dirty buffer. I presume that would always fail as the crc is updated in the write verifier.
Quoted 25 lines
libxfs/rdwr.c +7
-3+ }return bp; + }/* -* only set the ops on a cache miss (i.e. first physical read) as the -* verifier may change the ops to match the typ eof buffer it contains. +* Set the ops on a cache miss (i.e. first physical read) as the +* verifier may change the ops to match the type of buffer it contains. * A cache hit might reset the verifier to the original type if we set * it again, but it won't get called again and set to match the buffer * contents. *cough* xfs_da_node_buf_ops *cough*. @@ -733,8 +746,10 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, error =libxfs_readbufr (btp, blkno, bp, len, flags);if (error) bp->b_error = error; -else if (bp->b_ops) +else if (bp->b_ops) { bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + }return bp; }
libxfs/rdwr.c libxfs_readbuf_map(struct xfs_buftarg *btp, struct xfs_buf_map *map, int nmaps, · line 809
Quoted 9 lines
libxfs/rdwr.c +6
-0@@ -786,6 +801,14 @@ libxfs_readbuf_map(struct xfs_buftarg *btp, struct xfs_buf_map *map, int nmaps,return NULL ; bp->b_error =0 ; +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_ops = ops; + bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + }
Same comment here.
Brian
Quoted 60 lines
libxfs/rdwr.c +6
-1+return bp; + } bp->b_ops = ops;if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY )))return bp;@@ -793,8 +816,10 @@ libxfs_readbuf_map(struct xfs_buftarg *btp, struct xfs_buf_map *map, int nmaps, error =libxfs_readbufr_map (btp, bp, flags);if (!error) { bp->b_flags |=LIBXFS_B_UPTODATE ; -if (bp->b_ops) +if (bp->b_ops) { bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; + } }#ifdef IO_DEBUG printf("%lx: %s: read %lu bytes, error %d, blkno=%llu(%llu), %p\n",@@ -889,6 +914,7 @@ libxfs_writebufr(xfs_buf_t *bp)if (!error) { bp->b_flags |=LIBXFS_B_UPTODATE ; bp->b_flags &= ~(LIBXFS_B_DIRTY |LIBXFS_B_EXIT ); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ; }return error; }repair/prefetch.c +3
-0@@ -389,6 +389,7 @@ pf_read_inode_dirs( bp->b_ops = &xfs_inode_buf_ops; bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ;if (bp->b_error)return ;@@ -460,6 +461,7 @@ pf_read_discontig(pthread_mutex_unlock (&args->lock);libxfs_readbufr_map (mp->m_ddev_targp, bp,0 ); + bp->b_flags |=LIBXFS_B_UNCHECKED ;libxfs_putbuf (bp);pthread_mutex_lock (&args->lock); }@@ -583,6 +585,7 @@ pf_batch_read(break ;memcpy (XFS_BUF_PTR (bplist[i]), pbuf, size); bplist[i]->b_flags |=LIBXFS_B_UPTODATE ; + bplist[i]->b_flags |=LIBXFS_B_UNCHECKED ; len -= size;if (B_IS_INODE (XFS_BUF_PRIORITY (bplist[i])))pf_read_inode_dirs (args, bplist[i]);Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 3/9] repair: ensure prefetched buffers have CRCs validated
Quoted 17 lines · Brian Foster
Quoted 16 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Prefetch currently does not do CRC validation when the IO completes due to the optimisation it performs and the fact that it does not know what the type of metadata into the buffer is supposed to be. Hence, mark all prefetched buffers as "suspect" so that when the end user tries to read it with a supplied validation function the validation is run even though the buffer was already in the cache.
Signed-off-by Dave Chinner <[email protected]>--- include/libxfs.h | 1 + libxfs/rdwr.c | 36 +++++++++++++++++++++++++++++++----- repair/prefetch.c | 3 +++ 3 files changed, 35 insertions(+), 5 deletions(-)
Quoted 12 lines
include/libxfs.h enum xfs_buf_flags_t { /* b_flags bits */ · line 336
Quoted 9 lines
include/libxfs.h +1
-0@@ -333,6 +333,7 @@ enum xfs_buf_flags_t { /* b_flags bits */LIBXFS_B_STALE =0x0004 ,/* buffer marked as invalid */ LIBXFS_B_UPTODATE =0x0008 ,/* buffer is sync'd to disk */ LIBXFS_B_DISCONTIG =0x0010 ,/* discontiguous buffer */ + LIBXFS_B_UNCHECKED = 0x0020, /* needs verification */This is used in the first couple patches, so it should probably be defined earlier (or shuffle those patches appropriately).
Ah, I busted that on shuffling the patchset, and hadn't done a patch-by-patch compile. Well spotted!
Quoted 23 lines
Quoted 23 lines
include/libxfs.h +0
-0};#define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))libxfs/rdwr.c +10
-1@@ -718,12 +718,25 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, bp =libxfs_getbuf (btp, blkno, len);if (!bp)return NULL ; -if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) + +/* +* if the buffer was prefetched, it is likely that it was not +* validated. Hence if we are supplied an ops function and the +* buffer is marked as unchecked, we need to validate it now. +*/ +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_error =0 ; + bp->b_ops = ops;
Quoted 8 lines
libxfs/rdwr.c libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, · line 732
Quoted 2 lines
libxfs/rdwr.c +2
-0+ bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ;Should we always expect an unchecked buffer to be read with an ops vector before being written? Even if so, this might look cleaner if we didn't encode the possibility of running a read verifier on a dirty buffer. I presume that would always fail as the crc is updated in the write verifier.
It should fail, and that's a good thing because writing to an unchecked buffer would indicate that we didn't validate it properly in the first place. Hence I thought that doing it this way leaves a canary that traps other problem usage with unchecked buffers.
Realistically, we shouldn't be writing unchecked buffers - prefetch doesn't touch buffers, it just does IO, and so someone else has to read the buffers before they can be dirtied. If it's read without an ops structure then modified and read again with an ops structure, we'll catch it...
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 3/9] repair: ensure prefetched buffers have CRCs validated
Quoted 71 lines · Dave Chinner
Quoted 17 lines · Brian Foster
Quoted 16 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Prefetch currently does not do CRC validation when the IO completes due to the optimisation it performs and the fact that it does not know what the type of metadata into the buffer is supposed to be. Hence, mark all prefetched buffers as "suspect" so that when the end user tries to read it with a supplied validation function the validation is run even though the buffer was already in the cache.
Signed-off-by Dave Chinner <[email protected]>--- include/libxfs.h | 1 + libxfs/rdwr.c | 36 +++++++++++++++++++++++++++++++----- repair/prefetch.c | 3 +++ 3 files changed, 35 insertions(+), 5 deletions(-)Quoted 12 lines
include/libxfs.h enum xfs_buf_flags_t { /* b_flags bits */ · line 336
Quoted 9 lines
include/libxfs.h +1
-0@@ -333,6 +333,7 @@ enum xfs_buf_flags_t { /* b_flags bits */LIBXFS_B_STALE =0x0004 ,/* buffer marked as invalid */ LIBXFS_B_UPTODATE =0x0008 ,/* buffer is sync'd to disk */ LIBXFS_B_DISCONTIG =0x0010 ,/* discontiguous buffer */ + LIBXFS_B_UNCHECKED = 0x0020, /* needs verification */This is used in the first couple patches, so it should probably be defined earlier (or shuffle those patches appropriately).
Ah, I busted that on shuffling the patchset, and hadn't done a patch-by-patch compile. Well spotted!
Quoted 23 lines
Quoted 23 lines
include/libxfs.h +0
-0};#define XFS_BUF_DADDR_NULL ((xfs_daddr_t) (-1LL))libxfs/rdwr.c +10
-1@@ -718,12 +718,25 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, bp =libxfs_getbuf (btp, blkno, len);if (!bp)return NULL ; -if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) + +/* +* if the buffer was prefetched, it is likely that it was not +* validated. Hence if we are supplied an ops function and the +* buffer is marked as unchecked, we need to validate it now. +*/ +if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) { +if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { + bp->b_error =0 ; + bp->b_ops = ops;Quoted 8 lines
libxfs/rdwr.c libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags, · line 732
Quoted 2 lines
libxfs/rdwr.c +2
-0+ bp->b_ops->verify_read (bp); + bp->b_flags &= ~LIBXFS_B_UNCHECKED ;Should we always expect an unchecked buffer to be read with an ops vector before being written? Even if so, this might look cleaner if we didn't encode the possibility of running a read verifier on a dirty buffer. I presume that would always fail as the crc is updated in the write verifier.
It should fail, and that's a good thing because writing to an unchecked buffer would indicate that we didn't validate it properly in the first place. Hence I thought that doing it this way leaves a canary that traps other problem usage with unchecked buffers.
Quoted 6 lines
Realistically, we shouldn't be writing unchecked buffers - prefetch doesn't touch buffers, it just does IO, and so someone else has to read the buffers before they can be dirtied. If it's read without an ops structure then modified and read again with an ops structure, we'll catch it...
Ah, I see. That sounds good, but a small comment there with the reasoning to allow a read verifier to run on a dirty buffer would be nice. :)
Brian
Quoted 6 lines
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 3/9] repair: ensure prefetched buffers have CRCs validated
Quoted 20 lines · Brian Foster
Quoted 12 lines · Dave Chinner
Quoted 5 lines · Brian Foster
Should we always expect an unchecked buffer to be read with an ops vector before being written? Even if so, this might look cleaner if we didn't encode the possibility of running a read verifier on a dirty buffer. I presume that would always fail as the crc is updated in the write verifier.
It should fail, and that's a good thing because writing to an unchecked buffer would indicate that we didn't validate it properly in the first place. Hence I thought that doing it this way leaves a canary that traps other problem usage with unchecked buffers.
Quoted 6 lines
Realistically, we shouldn't be writing unchecked buffers - prefetch doesn't touch buffers, it just does IO, and so someone else has to read the buffers before they can be dirtied. If it's read without an ops structure then modified and read again with an ops structure, we'll catch it...
Quoted 3 lines
Ah, I see. That sounds good, but a small comment there with the reasoning to allow a read verifier to run on a dirty buffer would be nice. :)
Ok, I'll add one.
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
PATCH 4/9 repair: detect and correct CRC errors in directory blocks
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
repair doesn't currently verifier errors in directory blocks - they cause repair to ignore blocks and hence fail because it can't read critical blocks from the directory.
Fix this by having the directory buffer read code detect a verifier error and retry the read without the verifier if the verifier has detected an error. Then pass the verifer error with the successfully read buffer back to the caller, so the caller can handle the error appropriately. In most cases, this is simply marking the directory as needing a rebuild, so once the directory entries have been checked and repaired, it will rewrite all the directory buffers (including the clean ones) and in the process recalculate all the the CRC on the directory blocks.
Hence pure CRC errors in directory blocks are now handled correctly by xfs_repair.
--- libxfs/rdwr.c | 3 +- repair/phase6.c | 108 ++++++++++++++++++++++++++++++++++++++++++++++---------- 2 files changed, 90 insertions(+), 21 deletions(-)
libxfs/rdwr.c +1 -2
@@ -724,9 +724,9 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags,* validated. Hence if we are supplied an ops function and the * buffer is marked as unchecked, we need to validate it now. */ + bp->b_error =0 ;if ((bp->b_flags & (LIBXFS_B_UPTODATE |LIBXFS_B_DIRTY ))) {if (ops && (bp->b_flags &LIBXFS_B_UNCHECKED )) { - bp->b_error =0 ; bp->b_ops = ops; bp->b_ops->verify_read (bp); bp->b_flags &= ~LIBXFS_B_UNCHECKED ;
@@ -741,7 +741,6 @@ libxfs_readbuf(struct xfs_buftarg *btp, xfs_daddr_t blkno, int len, int flags,* it again, but it won't get called again and set to match the buffer * contents. *cough* xfs_da_node_buf_ops *cough*. */ - bp->b_error =0 ; bp->b_ops = ops; error =libxfs_readbufr (btp, blkno, bp, len, flags);if (error)
repair/phase6.c +89 -19
182 lines not on this page · raw message
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 4/9] repair: detect and correct CRC errors in directory blocks
Quoted 9 lines · Dave Chinner
+ * Need to handle CRC and validation errors specially here. If
+ * there is a validator error, re-read without the verifier so
+ * that we get a buffer we can check and repair. Re-attach the
+ * ops to the buffer after the read so that when it is rewritten
+ * the CRC is recalculated.
+ *
+ * Returns a positive error on failure, 0 for success, and negative
+ * error if a verifier error occurred and we reread the block without
+ * the verifier successfully.
This comment doesn't come clear to using up the 80 character allowance :)
I have to say the positive and negative errno convention is a bit confusing to me. I'd rather always return 0 if we could read it, and pass a bool *crc_error instead that indicates if it needs fixing.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 5/9 repair: detect CRC errors in AG headers
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
repair doesn't currently detect verifier errors in AG header blocks - apart from the primary superblock they are not detected. They are, fortunately, corrected in the important cases (AGF, AGI and AGFL) because these structures are rebuilt in phase 5, but if you run xfs_repair in checking mode it won't report them as bad.
--- repair/scan.c | 66 ++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 28 deletions(-)
repair/scan.c +38 -28
@@ -1207,28 +1207,31 @@ scan_ag( void *arg) {struct aghdr_cnts *agcnts = arg; -xfs_agf_t *agf; -xfs_buf_t *agfbuf; +struct xfs_agf *agf; +struct xfs_buf *agfbuf =NULL ;int agf_dirty =0 ; -xfs_agi_t *agi; -xfs_buf_t *agibuf; +struct xfs_agi *agi; +struct xfs_buf *agibuf =NULL ;int agi_dirty =0 ; -xfs_sb_t *sb; -xfs_buf_t *sbbuf; +struct xfs_sb *sb =NULL ; +struct xfs_buf *sbbuf =NULL ;int sb_dirty =0 ;int status; +char *objname =NULL ; sbbuf =libxfs_readbuf (mp->m_dev,XFS_AG_DADDR (mp, agno,XFS_SB_DADDR ),XFS_FSS_TO_BB (mp,1 ),0 , &xfs_sb_buf_ops);if (!sbbuf) { -do_error (_ ("can't get root superblock for ag %d\n" ), agno); -return ; + objname =_ ("root superblock" ); + goto out_free; } +if (sbbuf->b_error ==EFSBADCRC || sbbuf->b_error ==EFSCORRUPTED ) + sb_dirty =1 ; + sb = (xfs_sb_t *)calloc (BBSIZE ,1 );if (!sb) {do_error (_ ("can't allocate memory for superblock\n" )); -libxfs_putbuf (sbbuf); -return ; + goto out_free; }libxfs_sb_from_disk (sb,XFS_BUF_TO_SBP (sbbuf));
@@ -1236,23 +1239,22 @@ scan_ag(XFS_AG_DADDR (mp, agno,XFS_AGF_DADDR (mp)), XFS_FSS_TO_BB(mp, 1), 0, &xfs_agf_buf_ops);if (!agfbuf) { -do_error (_ ("can't read agf block for ag %d\n" ), agno); -libxfs_putbuf (sbbuf); -free (sb); -return ; + objname =_ ("agf block" ); + goto out_free; } +if (agfbuf->b_error ==EFSBADCRC || agfbuf->b_error ==EFSCORRUPTED ) + agf_dirty =1 ; agf =XFS_BUF_TO_AGF (agfbuf); agibuf =libxfs_readbuf (mp->m_dev,XFS_AG_DADDR (mp, agno,XFS_AGI_DADDR (mp)),XFS_FSS_TO_BB (mp,1 ),0 , &xfs_agi_buf_ops);if (!agibuf) { -do_error (_ ("can't read agi block for ag %d\n" ), agno); -libxfs_putbuf (agfbuf); -libxfs_putbuf (sbbuf); -free (sb); -return ; + objname =_ ("agi block" ); + goto out_free; } +if (agibuf->b_error ==EFSBADCRC || agibuf->b_error ==EFSCORRUPTED ) + agi_dirty =1 ; agi =XFS_BUF_TO_AGI (agibuf);/* fix up bad ag headers */
@@ -1277,7 +1279,7 @@ scan_ag(do_warn (_ ("would reset bad sb for ag %d\n" ), agno); } } -if (status &XR_AG_AGF ) { +if (agf_dirty || status &XR_AG_AGF ) {if (!no_modify) {do_warn (_ ("reset bad agf for ag %d\n" ), agno); agf_dirty =1 ;
@@ -1285,7 +1287,7 @@ scan_ag(do_warn (_ ("would reset bad agf for ag %d\n" ), agno); } } -if (status &XR_AG_AGI ) { +if (agi_dirty || status &XR_AG_AGI ) {if (!no_modify) {do_warn (_ ("reset bad agi for ag %d\n" ), agno); agi_dirty =1 ;
@@ -1295,15 +1297,9 @@ scan_ag( }if (status && no_modify) { -libxfs_putbuf (agibuf); -libxfs_putbuf (agfbuf); -libxfs_putbuf (sbbuf); -free (sb); -do_warn (_ ("bad uncorrected agheader %d, skipping ag...\n" ), agno); - -return ; + goto out_free; }scan_freelist (agf, agcnts);
@@ -1341,6 +1337,20 @@ scan_ag(print_inode_list (i);#endif return ; + +out_free: +if (sb) +free (sb); +if (agibuf) +libxfs_putbuf (agibuf); +if (agfbuf) +libxfs_putbuf (agfbuf); +if (sbbuf) +libxfs_putbuf (sbbuf); +if (objname) +do_error (_ ("can't get %s for ag %d\n" ), objname, agno); +return ; + }#define SCAN_THREADS 32
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 101 lines · Dave Chinner
From: Dave Chinner <[email protected]>
repair doesn't currently detect verifier errors in AG header blocks - apart from the primary superblock they are not detected. They are, fortunately, corrected in the important cases (AGF, AGI and AGFL) because these structures are rebuilt in phase 5, but if you run
xfs_repairin checking mode it won't report them as bad.Signed-off-by Dave Chinner <[email protected]>--- repair/scan.c | 66 ++++++++++++++++++++++++++++++++++------------------------- 1 file changed, 38 insertions(+), 28 deletions(-)repair/scan.c +22
-20@@ -1207,28 +1207,31 @@ scan_ag( void *arg) {struct aghdr_cnts *agcnts = arg; -xfs_agf_t *agf; -xfs_buf_t *agfbuf; +struct xfs_agf *agf; +struct xfs_buf *agfbuf =NULL ;int agf_dirty =0 ; -xfs_agi_t *agi; -xfs_buf_t *agibuf; +struct xfs_agi *agi; +struct xfs_buf *agibuf =NULL ;int agi_dirty =0 ; -xfs_sb_t *sb; -xfs_buf_t *sbbuf; +struct xfs_sb *sb =NULL ; +struct xfs_buf *sbbuf =NULL ;int sb_dirty =0 ;int status; +char *objname =NULL ; sbbuf =libxfs_readbuf (mp->m_dev,XFS_AG_DADDR (mp, agno,XFS_SB_DADDR ),XFS_FSS_TO_BB (mp,1 ),0 , &xfs_sb_buf_ops);if (!sbbuf) { -do_error (_ ("can't get root superblock for ag %d\n" ), agno); -return ; + objname =_ ("root superblock" ); + goto out_free; } +if (sbbuf->b_error ==EFSBADCRC || sbbuf->b_error ==EFSCORRUPTED ) + sb_dirty =1 ; + sb = (xfs_sb_t *)calloc (BBSIZE ,1 );if (!sb) {do_error (_ ("can't allocate memory for superblock\n" )); -libxfs_putbuf (sbbuf); -return ; + goto out_free; }libxfs_sb_from_disk (sb,XFS_BUF_TO_SBP (sbbuf));@@ -1236,23 +1239,22 @@ scan_ag(XFS_AG_DADDR (mp, agno,XFS_AGF_DADDR (mp)), XFS_FSS_TO_BB(mp, 1), 0, &xfs_agf_buf_ops);if (!agfbuf) { -do_error (_ ("can't read agf block for ag %d\n" ), agno); -libxfs_putbuf (sbbuf); -free (sb); -return ; + objname =_ ("agf block" ); + goto out_free; } +if (agfbuf->b_error ==EFSBADCRC || agfbuf->b_error ==EFSCORRUPTED ) + agf_dirty =1 ; agf =XFS_BUF_TO_AGF (agfbuf); agibuf =libxfs_readbuf (mp->m_dev,XFS_AG_DADDR (mp, agno,XFS_AGI_DADDR (mp)),XFS_FSS_TO_BB (mp,1 ),0 , &xfs_agi_buf_ops);if (!agibuf) { -do_error (_ ("can't read agi block for ag %d\n" ), agno); -libxfs_putbuf (agfbuf); -libxfs_putbuf (sbbuf); -free (sb); -return ; + objname =_ ("agi block" ); + goto out_free; } +if (agibuf->b_error ==EFSBADCRC || agibuf->b_error ==EFSCORRUPTED ) + agi_dirty =1 ; agi =XFS_BUF_TO_AGI (agibuf);/* fix up bad ag headers */ @@ -1277,7 +1279,7 @@ scan_ag(do_warn (_ ("would reset bad sb for ag %d\n" ), agno); } } -if (status &XR_AG_AGF ) { +if (agf_dirty || status &XR_AG_AGF ) {if (!no_modify) {do_warn (_ ("reset bad agf for ag %d\n" ), agno); agf_dirty =1 ;
repair/scan.c scan_ag( · line 1293
Quoted 8 lines
repair/scan.c +1
-1@@ -1285,7 +1287,7 @@ scan_ag(do_warn (_ ("would reset bad agf for ag %d\n" ), agno); } } -if (status &XR_AG_AGI ) { +if (agi_dirty || status &XR_AG_AGI ) {if (!no_modify) {do_warn (_ ("reset bad agi for ag %d\n" ), agno); agi_dirty =1 ;
There are a few asserts a bit further down this function that assume *_dirty is set only when in !no_modify mode. E.g.:
ASSERT(agi_dirty == 0 || (agi_dirty && !no_modify));
You'll probably want to remove those. Or...
Quoted 5 lines
repair/scan.c +0
-1@@ -1295,15 +1297,9 @@ scan_ag( }if (status && no_modify) { -libxfs_putbuf (agibuf);
repair/scan.c scan_ag( · line 1303
Quoted 10 lines
repair/scan.c +1
-6-libxfs_putbuf (agfbuf); -libxfs_putbuf (sbbuf); -free (sb); -do_warn (_ ("bad uncorrected agheader %d, skipping ag...\n" ), agno); - -return ; + goto out_free; }
Would we want to skip the ag, as such, on a CRC error in no_modify mode? If so, perhaps we could set the status variable on crc errors and bitwise or the value returned from verify_set_agheader().
Brian
Quoted 30 lines
repair/scan.c +14
-0scan_freelist (agf, agcnts);@@ -1341,6 +1337,20 @@ scan_ag(print_inode_list (i);#endif return ; + +out_free: +if (sb) +free (sb); +if (agibuf) +libxfs_putbuf (agibuf); +if (agfbuf) +libxfs_putbuf (agfbuf); +if (sbbuf) +libxfs_putbuf (sbbuf); +if (objname) +do_error (_ ("can't get %s for ag %d\n" ), objname, agno); +return ; + }#define SCAN_THREADS 32Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 10 lines · Brian Foster
Quoted 9 lines · Dave Chinner
From: Dave Chinner <[email protected]>
repair doesn't currently detect verifier errors in AG header blocks - apart from the primary superblock they are not detected. They are, fortunately, corrected in the important cases (AGF, AGI and AGFL) because these structures are rebuilt in phase 5, but if you run
xfs_repairin checking mode it won't report them as bad.Signed-off-by Dave Chinner <[email protected]>
....
Quoted 13 lines
repair/scan.c scan_ag( · line 1293
Quoted 9 lines
repair/scan.c +1
-1@@ -1285,7 +1287,7 @@ scan_ag(do_warn (_ ("would reset bad agf for ag %d\n" ), agno); } } -if (status &XR_AG_AGI ) { +if (agi_dirty || status &XR_AG_AGI ) {if (!no_modify) {do_warn (_ ("reset bad agi for ag %d\n" ), agno); agi_dirty =1 ;There are a few asserts a bit further down this function that assume
*_dirtyis set only when in !no_modify mode. E.g.:
Quoted 3 lines
ASSERT(agi_dirty == 0 || (agi_dirty && !no_modify));You'll probably want to remove those. Or...
I thought I caught all of those...
Quoted 4 lines
Quoted 4 lines
(quoted diff) +0
-0@@ -1295,15 +1297,9 @@ scan_ag( }
Quoted 16 lines
repair/scan.c scan_ag( · line 1303
Quoted 12 lines
repair/scan.c +1
-7if (status && no_modify) { -libxfs_putbuf (agibuf); -libxfs_putbuf (agfbuf); -libxfs_putbuf (sbbuf); -free (sb); -do_warn (_ ("bad uncorrected agheader %d, skipping ag...\n" ), agno); - -return ; + goto out_free; }Would we want to skip the ag, as such, on a CRC error in
no_modifymode? If so, perhaps we could set the status variable on crc errors and bitwise or the value returned fromverify_set_agheader().
I figured that for CRC errors we really should try to validate everything if we can. If nothing else fails the validation, then it might be a bit flip in an unused portion of the sector and in that case we can actually continue onwards. IOWs, a CRC error is not necessarily fatal....
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
In reply to Dave ChinnerQuoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
repair doesn't currently detect verifier errors in AG header blocks - apart from the primary superblock they are not detected. They are, fortunately, corrected in the important cases (AGF, AGI and AGFL) because these structures are rebuilt in phase 5, but if you run
xfs_repairin checking mode it won't report them as bad.
Shouldn't we apply the same scheme as for directories here, that is if it fails with a verifier error re-read without the verifier and then still do the full check as well?
Btw, it might make sense to have special read_buf variants in libxfs that always return a valid buffer even if the verifier fails, just telling us about it without having to re-read.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 9 lines · Christoph Hellwig
Quoted 7 lines · Dave Chinner
From: Dave Chinner <[email protected]>
repair doesn't currently detect verifier errors in AG header blocks - apart from the primary superblock they are not detected. They are, fortunately, corrected in the important cases (AGF, AGI and AGFL) because these structures are rebuilt in phase 5, but if you run
xfs_repairin checking mode it won't report them as bad.
Quoted 3 lines
Shouldn't we apply the same scheme as for directories here, that is if it fails with a verifier error re-read without the verifier and then still do the full check as well?
The directory code is the special case - it uses xfs_trans_read_buf* interfaces, which return either a good buffer with no error or an error with no buffer. Hence for the directory code, we have to re-read the buffer without the verifier to grab the unchecked buffer from the cache when the verifier detects an error.
Quoted 3 lines
Btw, it might make sense to have special
read_bufvariants in libxfs that always return a valid buffer even if the verifier fails, just telling us about it without having to re-read.
That's exactly what they do now - you get the xfs_buf, the data in bp->b_addr and the verifier error in bp->b_error all in one call.
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 9 lines · Dave Chinner
Quoted 3 lines
Shouldn't we apply the same scheme as for directories here, that is if it fails with a verifier error re-read without the verifier and then still do the full check as well?
The directory code is the special case - it uses xfs_trans_read_buf* interfaces, which return either a good buffer with no error or an error with no buffer. Hence for the directory code, we have to re-read the buffer without the verifier to grab the unchecked buffer from the cache when the verifier detects an error.
How about just having a variant of xfs_da_read_buf that doesn't use xfs_trans_read_buf *? xfs_da_read_buf is pretty simple, especially when removing the magic test that has been superceeded by the verifiers.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · Dave Chinner
Quoted 3 lines
Shouldn't we apply the same scheme as for directories here, that is if it fails with a verifier error re-read without the verifier and then still do the full check as well?
The directory code is the special case - it uses xfs_trans_read_buf* interfaces, which return either a good buffer with no error or an error with no buffer. Hence for the directory code, we have to re-read the buffer without the verifier to grab the unchecked buffer from the cache when the verifier detects an error.
Quoted 3 lines
How about just having a variant of
xfs_da_read_bufthat doesn't usexfs_trans_read_buf*?xfs_da_read_bufis pretty simple, especially when removing the magic test that has been superceeded by the verifiers.
Right now I ijust want to keep the change as small as possible and get a release out.
Yes, I agree there's much more cleanup that is needed in this code, but at this point is seems unnecessary and doesn't matter for the purpose of providing this functionality initially. We have to provide the same functionality for the kernel code for it to be able to handle CRC errors sanely, and so we're going to need to restructure xfs_trans_read_buf() to handle it. in doing this, we solve the libxfs problem, too, because what we do to the kernel code will flow on to userspace...
So really, I'd prefer to leave the userspace code doing this retry for this one piece of infrastructure and do all the major restructing of the directory buffer read code in the kernel code first. That way we can pick up all the changes for userspace when the libxfs code is updated.
Keep in mind that right now we've got a *lot* of updates from the kernel to the libxfs code pending that are stalled waiting for a release to be done...
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 5/9] repair: detect CRC errors in AG headers
Quoted 2 lines · Dave Chinner
Right now I ijust want to keep the change as small as possible and get a release out.
Oh well. I'll probably ACK it next time it comes around. I think the series needs enough changes for a resend anyway.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 6/9 repair: report AG btree verifier errors
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
When we scan the filesystem freespace and inode maps in phase 2, we don't report CRC errors that are found. We don't really care from a repair perspective, because the trees are completely rebuilt from the ground up in phase 5, but froma checking perspective we need to inform the user that we found inconsistencies.
--- repair/scan.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-)
repair/scan.c +24 -3
@@ -82,6 +82,12 @@ scan_sbtree(do_error (_ ("can't read btree block %d/%d\n" ), agno, root);return ; } +if (bp->b_error ==EFSBADCRC || bp->b_error ==EFSCORRUPTED ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), + agno, root, bp->b_error); + suspect =1 ; + } + (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , root, agno, suspect, isroot, magic, priv);libxfs_putbuf (bp);
@@ -123,6 +129,7 @@ scan_lbtree(xfs_buf_t *bp;int err;int dirty =0 ; +bool badcrc = false; bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, root),XFS_FSB_TO_BB (mp,1 ),0 , ops);
@@ -132,14 +139,25 @@ scan_lbtree( XFS_FSB_TO_AGBNO(mp, root));return (1 ); } + +/* +* only check for bad CRC here - caller will determine if there +* is a corruption or not and whether it got corrected and so needs +* writing back. CRC errors always imply we need to write the block. +*/ +if (bp->b_error ==EFSBADCRC ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), +XFS_FSB_TO_AGNO (mp, root), +XFS_FSB_TO_AGBNO (mp, root), bp->b_error); + badcrc = true; + } + err = (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , type, whichfork, root, ino, tot, nex, blkmapp, bm_cursor, isroot, check_dups, &dirty, magic); -ASSERT (dirty ==0 || (dirty && !no_modify)); - -if (dirty && !no_modify) +if ((dirty || badcrc) && !no_modify)libxfs_writebuf (bp,0 );else libxfs_putbuf (bp);
@@ -1066,6 +1084,9 @@ scan_freelist(do_abort (_ ("can't read agfl block for ag %d\n" ), agno);return ; } +if (agflbuf->b_error ==EFSBADCRC ) +do_warn (_ ("agfl has bad CRC for ag %d\n" ), agno); + freelist =XFS_BUF_TO_AGFL_BNO (mp, agflbuf); i =be32_to_cpu (agf->agf_flfirst);
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 6/9] repair: report AG btree verifier errors
Quoted 52 lines · Dave Chinner
From: Dave Chinner <[email protected]>
When we scan the filesystem freespace and inode maps in phase 2, we don't report CRC errors that are found. We don't really care from a repair perspective, because the trees are completely rebuilt from the ground up in phase 5, but froma checking perspective we need to inform the user that we found inconsistencies.
Signed-off-by Dave Chinner <[email protected]>--- repair/scan.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-)repair/scan.c +17
-0@@ -82,6 +82,12 @@ scan_sbtree(do_error (_ ("can't read btree block %d/%d\n" ), agno, root);return ; } +if (bp->b_error ==EFSBADCRC || bp->b_error ==EFSCORRUPTED ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), + agno, root, bp->b_error); + suspect =1 ; + } + (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , root, agno, suspect, isroot, magic, priv);libxfs_putbuf (bp);@@ -123,6 +129,7 @@ scan_lbtree(xfs_buf_t *bp;int err;int dirty =0 ; +bool badcrc = false; bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, root),XFS_FSB_TO_BB (mp,1 ),0 , ops);@@ -132,14 +139,25 @@ scan_lbtree( XFS_FSB_TO_AGBNO(mp, root));return (1 ); } + +/* +* only check for bad CRC here - caller will determine if there +* is a corruption or not and whether it got corrected and so needs +* writing back. CRC errors always imply we need to write the block. +*/ +if (bp->b_error ==EFSBADCRC ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), +XFS_FSB_TO_AGNO (mp, root), +XFS_FSB_TO_AGBNO (mp, root), bp->b_error);
repair/scan.c scan_lbtree( · line 141
Quoted 10 lines
repair/scan.c +3
-2+ badcrc = true; + } + err = (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , type, whichfork, root, ino, tot, nex, blkmapp, bm_cursor, isroot, check_dups, &dirty, magic); -ASSERT (dirty ==0 || (dirty && !no_modify)); -
Here we remove the dirty assert, but don't change the semantics of dirty (we use badcrc).
Brian
Quoted 22 lines
repair/scan.c +4
-1-if (dirty && !no_modify) +if ((dirty || badcrc) && !no_modify)libxfs_writebuf (bp,0 );else libxfs_putbuf (bp);@@ -1066,6 +1084,9 @@ scan_freelist(do_abort (_ ("can't read agfl block for ag %d\n" ), agno);return ; } +if (agflbuf->b_error ==EFSBADCRC ) +do_warn (_ ("agfl has bad CRC for ag %d\n" ), agno); + freelist =XFS_BUF_TO_AGFL_BNO (mp, agflbuf); i =be32_to_cpu (agf->agf_flfirst);Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 6/9] repair: report AG btree verifier errors
Quoted 61 lines · Brian Foster
Quoted 60 lines · Dave Chinner
From: Dave Chinner <[email protected]>
When we scan the filesystem freespace and inode maps in phase 2, we don't report CRC errors that are found. We don't really care from a repair perspective, because the trees are completely rebuilt from the ground up in phase 5, but froma checking perspective we need to inform the user that we found inconsistencies.
Signed-off-by Dave Chinner <[email protected]>--- repair/scan.c | 27 ++++++++++++++++++++++++--- 1 file changed, 24 insertions(+), 3 deletions(-)repair/scan.c +20
-0@@ -82,6 +82,12 @@ scan_sbtree(do_error (_ ("can't read btree block %d/%d\n" ), agno, root);return ; } +if (bp->b_error ==EFSBADCRC || bp->b_error ==EFSCORRUPTED ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), + agno, root, bp->b_error); + suspect =1 ; + } + (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , root, agno, suspect, isroot, magic, priv);libxfs_putbuf (bp);@@ -123,6 +129,7 @@ scan_lbtree(xfs_buf_t *bp;int err;int dirty =0 ; +bool badcrc = false; bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, root),XFS_FSB_TO_BB (mp,1 ),0 , ops);@@ -132,14 +139,25 @@ scan_lbtree( XFS_FSB_TO_AGBNO(mp, root));return (1 ); } + +/* +* only check for bad CRC here - caller will determine if there +* is a corruption or not and whether it got corrected and so needs +* writing back. CRC errors always imply we need to write the block. +*/ +if (bp->b_error ==EFSBADCRC ) { +do_warn (_ ("btree block %d/%d is suspect, error %d\n" ), +XFS_FSB_TO_AGNO (mp, root), +XFS_FSB_TO_AGBNO (mp, root), bp->b_error); + badcrc = true; + } + err = (*func)(XFS_BUF_TO_BLOCK (bp), nlevels -1 , type, whichfork, root, ino, tot, nex, blkmapp, bm_cursor, isroot, check_dups, &dirty, magic);
Quoted 5 lines
repair/scan.c scan_lbtree( · line 141
Quoted 2 lines
repair/scan.c +0
-2-ASSERT (dirty ==0 || (dirty && !no_modify)); -Here we remove the dirty assert, but don't change the semantics of dirty (we use badcrc).
Yeah, I went back and changed the mechanism for correcting pure CRC errors not to use the dirty flag. I'll re-instate the assert...
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
PATCH 7/9 repair: remove more dirv1 leftovers
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
get_bmapi() and it's children were only called by dirv1 code. There are no current callers, so remove them.
--- repair/dinode.c | 239 -------------------------------------------------------- repair/dinode.h | 6 -- 2 files changed, 245 deletions(-)
repair/dinode.c +0 -239
@@ -870,245 +870,6 @@ get_agino_buf(xfs_mount_t *mp, }/* -* these next routines return the filesystem blockno of the -* block containing the block "bno" in the file whose bmap -* tree (or extent list) is rooted by "rootblock". -* -* the next routines are utility routines for the third -* routine, get_bmapi(). -* -* NOTE: getfunc_extlist only used by dirv1 checking code -*/ -static xfs_dfsbno_t -getfunc_extlist (xfs_mount_t *mp, -xfs_ino_t ino, -xfs_dinode_t *dip, -xfs_dfiloff_t bno, -int whichfork) -{ -xfs_bmbt_irec_t irec; -xfs_dfsbno_t final_fsbno =NULLDFSBNO ; -xfs_bmbt_rec_t *rootblock = (xfs_bmbt_rec_t *) -XFS_DFORK_PTR (dip, whichfork); -xfs_extnum_t nextents =XFS_DFORK_NEXTENTS (dip, whichfork); -int i; - -for (i =0 ; i < nextents; i++) { -libxfs_bmbt_disk_get_all (rootblock + i, &irec); -if (irec.br_startoff <= bno && - bno < irec.br_startoff + irec.br_blockcount) { - final_fsbno = bno - irec.br_startoff + irec.br_startblock; -break ; - } - } - -return (final_fsbno); -} - -/* -* NOTE: getfunc_btree only used by dirv1 checking code... -*/ -static xfs_dfsbno_t -getfunc_btree (xfs_mount_t *mp, -xfs_ino_t ino, -xfs_dinode_t *dip, -xfs_dfiloff_t bno, -int whichfork) -{ -int i; -#ifdef DEBUG -int prev_level; -#endif -int found; -int numrecs; -xfs_bmbt_rec_t *rec; -xfs_bmbt_irec_t irec; -xfs_bmbt_ptr_t *pp; -xfs_bmbt_key_t *key; -xfs_bmdr_key_t *rkey; -xfs_bmdr_ptr_t *rp; -xfs_dfsbno_t fsbno; -xfs_buf_t *bp; -xfs_dfsbno_t final_fsbno =NULLDFSBNO ; -struct xfs_btree_block *block; -xfs_bmdr_block_t *rootblock = (xfs_bmdr_block_t *) -XFS_DFORK_PTR (dip, whichfork); - -ASSERT (rootblock->bb_level !=0 ); -/* -* deal with root block, it's got a slightly different -* header structure than interior nodes. We know that -* a btree should have at least 2 levels otherwise it -* would be an extent list. -*/ - rkey =XFS_BMDR_KEY_ADDR (rootblock,1 ); - rp =XFS_BMDR_PTR_ADDR (rootblock,1 , -xfs_bmdr_maxrecs (mp,XFS_DFORK_SIZE (dip, mp, whichfork),1 )); - found =-1 ; -for (i =0 ; i <be16_to_cpu (rootblock->bb_numrecs) -1 ; i++) { -if (be64_to_cpu (rkey[i].br_startoff) <= bno && - bno <be64_to_cpu (rkey[i +1 ].br_startoff)) { - found = i; -break ; - } - } -if (i ==be16_to_cpu (rootblock->bb_numrecs) -1 && - bno >=be64_to_cpu (rkey[i].br_startoff)) - found = i; - -ASSERT (found !=-1 ); - - fsbno =be64_to_cpu (rp[found]); - -ASSERT (verify_dfsbno (mp, fsbno)); - - bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, fsbno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL ); -if (!bp) { -do_error (_ ("cannot read bmap block %" PRIu64"\n" ), fsbno); -return (NULLDFSBNO ); - } - block =XFS_BUF_TO_BLOCK (bp); - numrecs =be16_to_cpu (block->bb_numrecs); - -/* -* ok, now traverse any interior btree nodes -*/ -#ifdef DEBUG - prev_level =be16_to_cpu (block->bb_level); -#endif - -while (be16_to_cpu (block->bb_level) >0 ) { -#ifdef DEBUG -ASSERT (be16_to_cpu (block->bb_level) < prev_level); - - prev_level =be16_to_cpu (block->bb_level); -#endif -if (numrecs > mp->m_bmap_dmxr[1 ]) { -do_warn ( -_ ("# of bmap records in inode %" PRIu64" exceeds max (%u, max - %u)\n" ), - ino, numrecs, - mp->m_bmap_dmxr[1 ]); -libxfs_putbuf (bp); -return (NULLDFSBNO ); - } -if (verbose && numrecs < mp->m_bmap_dmnr[1 ]) { -do_warn ( -_ ("- # of bmap records in inode %" PRIu64" less than minimum (%u, min - %u), proceeding ...\n" ), - ino, numrecs, mp->m_bmap_dmnr[1 ]); - } - key =XFS_BMBT_KEY_ADDR (mp, block,1 ); - pp =XFS_BMBT_PTR_ADDR (mp, block,1 , mp->m_bmap_dmxr[1 ]); -for (found =-1 , i =0 ; i < numrecs -1 ; i++) { -if (be64_to_cpu (key[i].br_startoff) <= bno && bno < -be64_to_cpu (key[i +1 ].br_startoff)) { - found = i; -break ; - } - } -if (i == numrecs -1 && bno >=be64_to_cpu (key[i].br_startoff)) - found = i; - -ASSERT (found !=-1 ); - fsbno =be64_to_cpu (pp[found]); - -ASSERT (verify_dfsbno (mp, fsbno)); - -/* -* release current btree block and read in the -* next btree block to be traversed -*/ -libxfs_putbuf (bp); - bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, fsbno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL ); -if (!bp) { -do_error (_ ("cannot read bmap block %" PRIu64"\n" ), - fsbno); -return (NULLDFSBNO ); - } - block =XFS_BUF_TO_BLOCK (bp); - numrecs =be16_to_cpu (block->bb_numrecs); - } - -/* -* current block must be a leaf block -*/ -ASSERT (be16_to_cpu (block->bb_level) ==0 ); -if (numrecs > mp->m_bmap_dmxr[0 ]) { -do_warn ( -_ ("# of bmap records in inode %" PRIu64" greater than maximum (%u, max - %u)\n" ), - ino, numrecs, mp->m_bmap_dmxr[0 ]); -libxfs_putbuf (bp); -return (NULLDFSBNO ); - } -if (verbose && numrecs < mp->m_bmap_dmnr[0 ]) -do_warn ( -_ ("- # of bmap records in inode %" PRIu64" less than minimum (%u, min - %u), continuing...\n" ), - ino, numrecs, mp->m_bmap_dmnr[0 ]); - - rec =XFS_BMBT_REC_ADDR (mp, block,1 ); -for (i =0 ; i < numrecs; i++) { -libxfs_bmbt_disk_get_all (rec + i, &irec); -if (irec.br_startoff <= bno && - bno < irec.br_startoff + irec.br_blockcount) { - final_fsbno = bno - irec.br_startoff + - irec.br_startblock; -break ; - } - } -libxfs_putbuf (bp); - -if (final_fsbno ==NULLDFSBNO ) -do_warn (_ ("could not map block %" PRIu64"\n" ), bno); - -return (final_fsbno); -} - -/* -* this could be smarter. maybe we should have an open inode -* routine that would get the inode buffer and return back -* an inode handle. I'm betting for the moment that this -* is used only by the directory and attribute checking code -* and that the avl tree find and buffer cache search are -* relatively cheap. If they're too expensive, we'll just -* have to fix this and add an inode handle to the da btree -* cursor. -* -* caller is responsible for checking doubly referenced blocks -* and references to holes -* -* NOTE: get_bmapi only used by dirv1 checking code -*/ -xfs_dfsbno_t -get_bmapi (xfs_mount_t *mp,xfs_dinode_t *dino_p, -xfs_ino_t ino_num,xfs_dfiloff_t bno,int whichfork) -{ -xfs_dfsbno_t fsbno; - -switch (XFS_DFORK_FORMAT (dino_p, whichfork)) { -case XFS_DINODE_FMT_EXTENTS : - fsbno =getfunc_extlist (mp, ino_num, dino_p, bno, whichfork); -break ; -case XFS_DINODE_FMT_BTREE : - fsbno =getfunc_btree (mp, ino_num, dino_p, bno, whichfork); -break ; -case XFS_DINODE_FMT_LOCAL : -do_error (_ ("get_bmapi() called for local inode %" PRIu64"\n" ), - ino_num); - fsbno =NULLDFSBNO ; -break ; -default : -/* -* shouldn't happen -*/ -do_error (_ ("bad inode format for inode %" PRIu64"\n" ), ino_num); - fsbno =NULLDFSBNO ; - } - -return (fsbno); -} - -/* * higher level inode processing stuff starts here: * first, one utility routine for each type of inode */
repair/dinode.h +0 -6
@@ -119,12 +119,6 @@ get_agino_buf(xfs_mount_t *mp,xfs_agino_t agino, xfs_dinode_t **dipp); -xfs_dfsbno_t -get_bmapi (xfs_mount_t *mp, -xfs_dinode_t *dip, -xfs_ino_t ino_num, -xfs_dfiloff_t bno, -int whichfork );void dinode_bmbt_translation_init (void );char *get_forkname (int whichfork);
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 7/9] repair: remove more dirv1 leftovers
Quoted 15 lines · Dave Chinner
From: Dave Chinner <[email protected]>
get_bmapi()and it's children were only called by dirv1 code. There are no current callers, so remove them.Signed-off-by Dave Chinner <[email protected]>--- repair/dinode.c | 239 -------------------------------------------------------- repair/dinode.h | 6 -- 2 files changed, 245 deletions(-)repair/dinode.c +0
-0
...
Quoted 6 lines
repair/dinode.c +0
-6-xfs_dfsbno_t -get_bmapi (xfs_mount_t *mp,xfs_dinode_t *dino_p, -xfs_ino_t ino_num,xfs_dfiloff_t bno,int whichfork) -{ -xfs_dfsbno_t fsbno; -
repair/dinode.c get_agino_buf(xfs_mount_t *mp, · line 1097
Quoted 10 lines
repair/dinode.c +0
-10-switch (XFS_DFORK_FORMAT (dino_p, whichfork)) { -case XFS_DINODE_FMT_EXTENTS : - fsbno =getfunc_extlist (mp, ino_num, dino_p, bno, whichfork); -break ; -case XFS_DINODE_FMT_BTREE : - fsbno =getfunc_btree (mp, ino_num, dino_p, bno, whichfork); -break ; -case XFS_DINODE_FMT_LOCAL : -do_error (_ ("get_bmapi() called for local inode %" PRIu64"\n" ), - ino_num);
I'm not quite familiar with how we manage the .po translation files, but FYI this string appears in a couple of them. Not sure whether there is value in keeping the translation around. The string does include the specific function name.
Brian
Quoted 41 lines
repair/dinode.c +0
-14- fsbno =NULLDFSBNO ; -break ; -default : -/* -* shouldn't happen -*/ -do_error (_ ("bad inode format for inode %" PRIu64"\n" ), ino_num); - fsbno =NULLDFSBNO ; - } - -return (fsbno); -} - -/* * higher level inode processing stuff starts here: * first, one utility routine for each type of inode */ repair/dinode.h +0
-6@@ -119,12 +119,6 @@ get_agino_buf(xfs_mount_t *mp,xfs_agino_t agino, xfs_dinode_t **dipp); -xfs_dfsbno_t -get_bmapi (xfs_mount_t *mp, -xfs_dinode_t *dip, -xfs_ino_t ino_num, -xfs_dfiloff_t bno, -int whichfork );void dinode_bmbt_translation_init (void );char *get_forkname (int whichfork);Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 7/9] repair: remove more dirv1 leftovers
Quoted 4 lines · Brian Foster
I'm not quite familiar with how we manage the .po translation files, but FYI this string appears in a couple of them. Not sure whether there is value in keeping the translation around. The string does include the specific function name.
They are dead but harmless after the original ones are removed. The next time the translators generate the translatable strings they won't be in there, and they should be removed with the translation update.
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 7/9] repair: remove more dirv1 leftovers
In reply to Dave ChinnerLooks good,
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 8/9 repair: handle remote sylmlink CRC errors
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
We can't really repair broken symlink buffer contents, but we can at least warn about it and correct the CRC error so the symlink is again readable.
--- repair/dinode.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)
repair/dinode.c +11 -1
@@ -1254,6 +1254,7 @@ process_symlink_remote(while (pathlen >0 ) {int blk_cnt =1 ;int byte_cnt; +int dirty =0 ; fsbno =blkmap_get (blkmap, i);if (fsbno ==NULLDFSBNO ) {
@@ -1284,6 +1285,12 @@ _("cannot read inode %" PRIu64 ", file block %d, disk block %" PRIu64 "\n"), lino, i, fsbno);return 1 ; } +if (bp->b_error ==EFSBADCRC ) { +do_warn ( +_ ("Bad symlink buffer CRC block %" PRIu64", inode %" PRIu64".\n" +"Correcting CRC, but symlink may be bad.\n" ), fsbno, lino); + dirty =1 ; + } byte_cnt =XFS_SYMLINK_BUF_SPACE (mp, byte_cnt); byte_cnt =MIN (pathlen, byte_cnt);
@@ -1307,7 +1314,10 @@ _("bad symlink header ino %" PRIu64 ", file block %d, disk block %" PRIu64 "\n") offset += byte_cnt; i++; -libxfs_putbuf (bp); +if (dirty) +libxfs_writebuf (bp,0 ); +else +libxfs_putbuf (bp); }return 0 ; }
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 8/9] repair: handle remote sylmlink CRC errors
Quoted 2 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Just a couple nits... otherwise it looks good to me. There's a typo in the patch subject...
Quoted 22 lines
We can't really repair broken symlink buffer contents, but we can at least warn about it and correct the CRC error so the symlink is again readable.
Signed-off-by Dave Chinner <[email protected]>--- repair/dinode.c | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-)repair/dinode.c +1
-0@@ -1254,6 +1254,7 @@ process_symlink_remote(while (pathlen >0 ) {int blk_cnt =1 ;int byte_cnt; +int dirty =0 ; fsbno =blkmap_get (blkmap, i);if (fsbno ==NULLDFSBNO ) {
repair/dinode.c _("cannot read inode %" PRIu64 ", file block %d, disk block %" PRIu64 "\n"), · line 1290
Quoted 6 lines
repair/dinode.c +3
-0@@ -1284,6 +1285,12 @@ _("cannot read inode %" PRIu64 ", file block %d, disk block %" PRIu64 "\n"), lino, i, fsbno);return 1 ; } +if (bp->b_error ==EFSBADCRC ) { + do_warn( +_("Bad symlink buffer CRC block %" PRIu64 ", inode %" PRIu64 ".\n"
"Bad symlink buffer CRC, block ..."
... and sticking a comma or some separator in that message makes it a little easier to read, IMO.
Brian
Quoted 25 lines
repair/dinode.c +7
-1+ "Correcting CRC, but symlink may be bad.\n"), fsbno, lino); + dirty =1 ; + } byte_cnt =XFS_SYMLINK_BUF_SPACE (mp, byte_cnt); byte_cnt =MIN (pathlen, byte_cnt);@@ -1307,7 +1314,10 @@ _("bad symlink header ino %" PRIu64 ", file block %d, disk block %" PRIu64 "\n") offset += byte_cnt; i++; -libxfs_putbuf (bp); +if (dirty) +libxfs_writebuf (bp,0 ); +else +libxfs_putbuf (bp); }return 0 ; }Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
PATCH 9/9 repair: detect and handle attribute tree CRC errors
In reply to Dave ChinnerFrom: Dave Chinner <[email protected]>
Currently the attribute code will not detect and correct errors in the attribute tree. It also fails to validate the CRCs and headers on remote attribute blocks. Ensure that all the attribute blocks are CRC checked and that the processing functions understand the correct block formats for decoding.
--- repair/attr_repair.c | 35 ++++++++++++++++++++++++++++------- 1 file changed, 28 insertions(+), 7 deletions(-)
repair/attr_repair.c +28 -7
@@ -611,6 +611,8 @@ verify_da_path(xfs_mount_t *mp,ASSERT (cursor->level[this_level].dirty ==0 || (cursor->level[this_level].dirty && !no_modify)); +if (bp->b_error ==EFSBADCRC ) + cursor->level[this_level].dirty++;if (cursor->level[this_level].dirty && !no_modify)libxfs_writebuf (cursor->level[this_level].bp,0 );else
@@ -974,6 +976,10 @@ rmtval_get(xfs_mount_t *mp, xfs_ino_t ino, blkmap_t *blkmap,xfs_dfsbno_t bno;xfs_buf_t *bp;int clearit =0 , i =0 , length =0 , amountdone =0 ; +int hdrsize =0 ; + +if (xfs_sb_version_hascrc (&mp->m_sb)) + hdrsize =sizeof (struct xfs_attr3_rmt_hdr );/* ASSUMPTION: valuelen is a valid number, so use it for looping */ /* Note that valuelen is not a multiple of blocksize */
@@ -986,16 +992,26 @@ rmtval_get(xfs_mount_t *mp, xfs_ino_t ino, blkmap_t *blkmap,break ; } bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, bno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL ); +XFS_FSB_TO_BB (mp,1 ),0 , + &xfs_attr3_rmt_buf_ops);if (!bp) {do_warn (_ ("can't read remote block for attributes of inode %" PRIu64"\n" ), ino); clearit =1 ;break ; } + +if (bp->b_error ==EFSBADCRC || bp->b_error ==EFSCORRUPTED ) { +do_warn ( +_ ("Corrupt remote block for attributes of inode %" PRIu64"\n" ), ino); + clearit =1 ; +break ; + } +ASSERT (mp->m_sb.sb_blocksize ==XFS_BUF_COUNT (bp)); - length =MIN (XFS_BUF_COUNT (bp), valuelen - amountdone); -memmove (value,XFS_BUF_PTR (bp), length); + + length =MIN (XFS_BUF_COUNT (bp) - hdrsize, valuelen - amountdone); +memmove (value,XFS_BUF_PTR (bp) + hdrsize, length); amountdone += length; value += length; i++;
@@ -1320,13 +1336,16 @@ process_leaf_attr_level(xfs_mount_t *mp, } bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, dev_bno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL ); +XFS_FSB_TO_BB (mp,1 ),0 , + &xfs_attr3_leaf_buf_ops);if (!bp) {do_warn (_ ("can't read file block %u (fsbno %" PRIu64") for attribute fork of inode %" PRIu64"\n" ), da_bno, dev_bno, ino); goto error_out; } +if (bp->b_error ==EFSBADCRC ) + repair++; leaf = bp->b_addr;xfs_attr3_leaf_hdr_from_disk (&leafhdr, leaf);
@@ -1382,9 +1401,9 @@ process_leaf_attr_level(xfs_mount_t *mp, current_hashval = greatest_hashval; -if (repair && !no_modify) +if (repair && !no_modify)libxfs_writebuf (bp,0 ); -else +else libxfs_putbuf (bp); }while (da_bno !=0 );
@@ -1512,6 +1531,8 @@ process_longform_attr( ino);return (1 ); } +if (bp->b_error ==EFSBADCRC ) + (*repair)++;/* verify leaf block */ leaf = (xfs_attr_leafblock_t *)XFS_BUF_PTR (bp);
@@ -1555,7 +1576,7 @@ process_longform_attr(case XFS_DA_NODE_MAGIC :/* btree-form attribute */ case XFS_DA3_NODE_MAGIC :/* must do this now, to release block 0 before the traversal */ -if (repairlinks) { +if (*repair || repairlinks) { *repair =1 ;libxfs_writebuf (bp,0 ); }else
Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 9/9] repair: detect and handle attribute tree CRC errors
Quoted 18 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Currently the attribute code will not detect and correct errors in the attribute tree. It also fails to validate the CRCs and headers on remote attribute blocks. Ensure that all the attribute blocks are CRC checked and that the processing functions understand the correct block formats for decoding.
Signed-off-by Dave Chinner <[email protected]>--- repair/attr_repair.c | 35 ++++++++++++++++++++++++++++------- 1 file changed, 28 insertions(+), 7 deletions(-)repair/attr_repair.c +0
-0
repair/attr_repair.c verify_da_path(xfs_mount_t *mp, · line 615
Quoted 5 lines
repair/attr_repair.c +2
-0@@ -611,6 +611,8 @@ verify_da_path(xfs_mount_t *mp,ASSERT (cursor->level[this_level].dirty ==0 || (cursor->level[this_level].dirty && !no_modify)); +if (bp->b_error ==EFSBADCRC ) + cursor->level[this_level].dirty++;
I was wondering why this wasn't checked closer to the readbuf call, then I noticed the assert. Any reason not to be consistent with the other changes, move this up closer to the call and nuke the assert?
Quoted 49 lines
repair/attr_repair.c +17
-4if (cursor->level[this_level].dirty && !no_modify)libxfs_writebuf (cursor->level[this_level].bp,0 );else @@ -974,6 +976,10 @@ rmtval_get(xfs_mount_t *mp, xfs_ino_t ino, blkmap_t *blkmap,xfs_dfsbno_t bno;xfs_buf_t *bp;int clearit =0 , i =0 , length =0 , amountdone =0 ; +int hdrsize =0 ; + +if (xfs_sb_version_hascrc (&mp->m_sb)) + hdrsize =sizeof (struct xfs_attr3_rmt_hdr );/* ASSUMPTION: valuelen is a valid number, so use it for looping */ /* Note that valuelen is not a multiple of blocksize */ @@ -986,16 +992,26 @@ rmtval_get(xfs_mount_t *mp, xfs_ino_t ino, blkmap_t *blkmap,break ; } bp =libxfs_readbuf (mp->m_dev,XFS_FSB_TO_DADDR (mp, bno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL ); +XFS_FSB_TO_BB (mp,1 ),0 , + &xfs_attr3_rmt_buf_ops);if (!bp) {do_warn (_ ("can't read remote block for attributes of inode %" PRIu64"\n" ), ino); clearit =1 ;break ; } + +if (bp->b_error ==EFSBADCRC || bp->b_error ==EFSCORRUPTED ) { +do_warn ( +_ ("Corrupt remote block for attributes of inode %" PRIu64"\n" ), ino); + clearit =1 ; +break ; + } +ASSERT (mp->m_sb.sb_blocksize ==XFS_BUF_COUNT (bp)); - length =MIN (XFS_BUF_COUNT (bp), valuelen - amountdone); -memmove (value,XFS_BUF_PTR (bp), length); + + length =MIN (XFS_BUF_COUNT (bp) - hdrsize, valuelen - amountdone); +memmove (value,XFS_BUF_PTR (bp) + hdrsize, length); amountdone += length; value += length; i++;@@ -1320,13 +1336,16 @@ process_leaf_attr_level(xfs_mount_t *mp, } bp = libxfs_readbuf(mp->m_dev, XFS_FSB_TO_DADDR(mp, dev_bno), -XFS_FSB_TO_BB (mp,1 ),0 ,NULL );
repair/attr_repair.c process_leaf_attr_level(xfs_mount_t *mp, · line 1348
Quoted 10 lines
repair/attr_repair.c +4
-0+XFS_FSB_TO_BB (mp,1 ),0 , + &xfs_attr3_leaf_buf_ops);if (!bp) {do_warn (_ ("can't read file block %u (fsbno %" PRIu64") for attribute fork of inode %" PRIu64"\n" ), da_bno, dev_bno, ino); goto error_out; } +if (bp->b_error ==EFSBADCRC ) + repair++;
Could you remind me why we only check EFSBADCRC in some places and EFSCORRUPTED as well in others?
Quoted 16 lines
repair/attr_repair.c +2
-2leaf = bp->b_addr;xfs_attr3_leaf_hdr_from_disk (&leafhdr, leaf);@@ -1382,9 +1401,9 @@ process_leaf_attr_level(xfs_mount_t *mp, current_hashval = greatest_hashval; -if (repair && !no_modify) +if (repair && !no_modify)libxfs_writebuf (bp,0 ); -else +else libxfs_putbuf (bp); }while (da_bno !=0 );
repair/attr_repair.c process_longform_attr( · line 1535
Quoted 5 lines
repair/attr_repair.c +2
-0@@ -1512,6 +1531,8 @@ process_longform_attr( ino);return (1 ); } +if (bp->b_error ==EFSBADCRC ) + (*repair)++;
Note that repair is unconditionally reset to 0 at the beginning of process_leaf_attr_block() (in the XFS_ATTR_LEAF_MAGIC case further down this function).
Brian
Quoted 19 lines
repair/attr_repair.c +1
-1/* verify leaf block */ leaf = (xfs_attr_leafblock_t *)XFS_BUF_PTR (bp);@@ -1555,7 +1576,7 @@ process_longform_attr(case XFS_DA_NODE_MAGIC :/* btree-form attribute */ case XFS_DA3_NODE_MAGIC :/* must do this now, to release block 0 before the traversal */ -if (repairlinks) { +if (*repair || repairlinks) { *repair =1 ;libxfs_writebuf (bp,0 ); }else Signature · 6 lines
1.9.0 _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs
_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Re: [PATCH 9/9] repair: detect and handle attribute tree CRC errors
Quoted 22 lines · Brian Foster
Quoted 21 lines · Dave Chinner
From: Dave Chinner <[email protected]>
Currently the attribute code will not detect and correct errors in the attribute tree. It also fails to validate the CRCs and headers on remote attribute blocks. Ensure that all the attribute blocks are CRC checked and that the processing functions understand the correct block formats for decoding.
Signed-off-by Dave Chinner <[email protected]>--- repair/attr_repair.c | 35 ++++++++++++++++++++++++++++------- 1 file changed, 28 insertions(+), 7 deletions(-)repair/attr_repair.c +0
-0@@ -611,6 +611,8 @@ verify_da_path(xfs_mount_t *mp,ASSERT (cursor->level[this_level].dirty ==0 || (cursor->level[this_level].dirty && !no_modify));
Quoted 6 lines
repair/attr_repair.c verify_da_path(xfs_mount_t *mp, · line 615
Quoted 2 lines
repair/attr_repair.c +2
-0+if (bp->b_error ==EFSBADCRC ) + cursor->level[this_level].dirty++;I was wondering why this wasn't checked closer to the readbuf call, then I noticed the assert. Any reason not to be consistent with the other changes, move this up closer to the call and nuke the assert?
Because I didn't want to modify the cursor state if the buffer failed the basic checks (i.e. if (bad) {... return} branch was taken) as if it is bad we aren't even going to try to repair it. Hence marking it dirty for writeout when it might not even be an attr block is probably a bad thing...
Quoted 11 lines
repair/attr_repair.c process_leaf_attr_level(xfs_mount_t *mp, · line 1348
Quoted 8 lines
repair/attr_repair.c +0
-0if (!bp) { do_warn( _("can't read file block %u (fsbno %" PRIu64 ") for attribute fork of inode %" PRIu64 "\n"),repair/attr_repair.c +2
-0da_bno, dev_bno, ino); goto error_out; } +if (bp->b_error ==EFSBADCRC ) + repair++;Could you remind me why we only check EFSBADCRC in some places and EFSCORRUPTED as well in others?
Here we are checking and repairing the buffer, so EFSCORRUPTED detection doesn't provide any value - if there is a corruption, repair will already handle it and fix it. Hence this is here to catch a buffer with a bad CRC but is otherwise good (e.g. bit error in usused/unreferenced space in the metadata block).
Quoted 16 lines
Quoted 16 lines
repair/attr_repair.c +2
-2leaf = bp->b_addr;xfs_attr3_leaf_hdr_from_disk (&leafhdr, leaf);@@ -1382,9 +1401,9 @@ process_leaf_attr_level(xfs_mount_t *mp, current_hashval = greatest_hashval; -if (repair && !no_modify) +if (repair && !no_modify)libxfs_writebuf (bp,0 ); -else +else libxfs_putbuf (bp); }while (da_bno !=0 );
Quoted 10 lines
repair/attr_repair.c process_longform_attr( · line 1535
Quoted 6 lines
repair/attr_repair.c +2
-0@@ -1512,6 +1531,8 @@ process_longform_attr( ino);return (1 ); } +if (bp->b_error ==EFSBADCRC ) + (*repair)++;Note that repair is unconditionally reset to 0 at the beginning of
process_leaf_attr_block()(in theXFS_ATTR_LEAF_MAGICcase further down this function).
Oh, so it is. I missed that. Good catch! Will fix.
Cheers,
Dave.
Signature · 7 lines
Dave Chinner [email protected] _______________________________________________ xfs mailing list [email protected] http://oss.sgi.com/mailman/listinfo/xfs