NFSD: Fix last write offset handling in layoutcommit
Subsystems (3)
cover NFSD: Fix last write offset handling in layoutcommit
These patches correct the behavior of the pNFS server when the client sends a layoutcommit without a new file size and with zero number of block/scsi extents.
Tested manually for the pNFS block layout setup.
---
Sergey Bashirov (2):
NFSD: Minor cleanup in layoutcommit processing
NFSD: Fix last write offset handling in layoutcommit
fs/nfsd/blocklayout.c | 2 +- fs/nfsd/nfs4proc.c | 20 +++++++------------- 2 files changed, 8 insertions(+), 14 deletions(-)
Signature · 1 line
2.43.0
PATCH 1/2 NFSD: Minor cleanup in layoutcommit processing
Remove dprintk in nfsd4_layoutcommit. These are not needed in day to day usage, and the information is also available in Wireshark when capturing NFS traffic.
--- fs/nfsd/nfs4proc.c | 12 +++--------- 1 file changed, 3 insertions(+), 9 deletions(-)
fs/nfsd/nfs4proc.c +3 -9
@@ -2499,18 +2499,12 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp, inode =d_inode (current_fh->fh_dentry); nfserr = nfserr_inval; -if (new_size <= seg->offset) { -dprintk ("pnfsd: last write before layout segment\n" ); +if (new_size <= seg->offset) goto out; - } -if (new_size > seg->offset + seg->length) { -dprintk ("pnfsd: last write beyond layout segment\n" ); +if (new_size > seg->offset + seg->length) goto out; - } -if (!lcp->lc_newoffset && new_size >i_size_read (inode)) { -dprintk ("pnfsd: layoutcommit beyond EOF\n" ); +if (!lcp->lc_newoffset && new_size >i_size_read (inode)) goto out; - } nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lcp->lc_sid, false, lcp->lc_layout_type,
Signature · 1 line
2.43.0
Re: [PATCH 1/2] NFSD: Minor cleanup in layoutcommit processing
Looks good:
PATCH 2/2 NFSD: Fix last write offset handling in layoutcommit
In reply to Sergey BashirovThe data type of loca_last_write_offset is newoffset4 and is switched on a boolean value, no_newoffset, that indicates if a previous write occurred or not. If no_newoffset is FALSE, an offset is not given. This means that client does not try to update the file size. Thus, server should not try to calculate new file size and check if it fits into the seg range.
--- fs/nfsd/blocklayout.c | 2 +- fs/nfsd/nfs4proc.c | 16 ++++++++-------- 2 files changed, 9 insertions(+), 9 deletions(-)
fs/nfsd/blocklayout.c +1 -1
@@ -118,7 +118,7 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, struct iomap *iomaps, int nr_iomaps) {struct timespec64 mtime =inode_get_mtime (inode); -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;struct iattr iattr = { .ia_valid =0 };int error;
fs/nfsd/nfs4proc.c +8 -8
@@ -2482,7 +2482,7 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,const struct nfsd4_layout_seg *seg = &lcp->lc_seg;struct svc_fh *current_fh = &cstate->current_fh;const struct nfsd4_layout_ops *ops; -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;struct inode *inode;struct nfs4_layout_stateid *ls;__be32 nfserr;
@@ -2498,13 +2498,13 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp, goto out; inode =d_inode (current_fh->fh_dentry); - nfserr = nfserr_inval; -if (new_size <= seg->offset) - goto out; -if (new_size > seg->offset + seg->length) - goto out; -if (!lcp->lc_newoffset && new_size >i_size_read (inode)) - goto out; +if (new_size) { + nfserr = nfserr_inval; +if (new_size <= seg->offset) + goto out; +if (new_size > seg->offset + seg->length) + goto out; + } nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lcp->lc_sid, false, lcp->lc_layout_type,
Signature · 1 line
2.43.0
Hi Sergey, Konstantin -
Quoted 6 lines · Sergey Bashirov
The data type of
loca_last_write_offsetis newoffset4 and is switched on a boolean value,no_newoffset, that indicates if a previous write occurred or not. Ifno_newoffsetis FALSE, an offset is not given. This means that client does not try to update the file size. Thus, server should not try to calculate new file size and check if it fits into the seg range.
The patch description should describe the impact of the current incorrect logic -- does it result in file corruption, failed tests, etc? That way support engineers at distributions can more easily find this patch if a customer runs across bad behavior.
Also, let's reference RFC 8881 Section 12.5.4.2, where the properly compliant behavior is specified.
Quoted 13 lines
Co-developed-by Konstantin Evtushenko <[email protected]>Signed-off-by Konstantin Evtushenko <[email protected]>Signed-off-by Sergey Bashirov <[email protected]>--- fs/nfsd/blocklayout.c | 2 +- fs/nfsd/nfs4proc.c | 16 ++++++++-------- 2 files changed, 9 insertions(+), 9 deletions(-)fs/nfsd/blocklayout.c +0
-0
fs/nfsd/blocklayout.c nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, · line 123
Quoted 7 lines
fs/nfsd/blocklayout.c +1
-1@@ -118,7 +118,7 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, struct iomap *iomaps, int nr_iomaps) {struct timespec64 mtime =inode_get_mtime (inode); -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;struct iattr iattr = { .ia_valid =0 };int error;
See below for an alternative.
Quoted 14 lines
fs/nfsd/nfs4proc.c +1
-1@@ -2482,7 +2482,7 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,const struct nfsd4_layout_seg *seg = &lcp->lc_seg;struct svc_fh *current_fh = &cstate->current_fh;const struct nfsd4_layout_ops *ops; -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;struct inode *inode;struct nfs4_layout_stateid *ls;__be32 nfserr;
fs/nfsd/nfs4proc.c nfsd4_layoutcommit(struct svc_rqst *rqstp, · line 2500
Quoted 3 lines
fs/nfsd/nfs4proc.c +0
-0@@ -2498,13 +2498,13 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp, goto out; inode =d_inode (current_fh->fh_dentry);
How about instead, drop the new_size initializer above, and do this:
lcp->lc_size_chg = false;
if (lcp->lc_newoffset) {
loff_t new_size = lcp->lc_last_wr + 1;
nfserr = nfserr_inval;
if (new_size <= seg->offset)
goto out;
if (new_size > seg->offset + seg->length)
goto out;
if (new_size > i_size_read(inode)) {
lcp->lc_size_chg = true;
lcp->lc_newsize = new_size;
}
}Quoted 7 lines
fs/nfsd/nfs4proc.c +0
-7- nfserr = nfserr_inval; -if (new_size <= seg->offset) - goto out; -if (new_size > seg->offset + seg->length) - goto out; -if (!lcp->lc_newoffset && new_size >i_size_read (inode)) - goto out;
fs/nfsd/nfs4proc.c nfsd4_layoutcommit(struct svc_rqst *rqstp, · line 2510
Quoted 10 lines
fs/nfsd/nfs4proc.c +7
-0+if (new_size) { + nfserr = nfserr_inval; +if (new_size <= seg->offset) + goto out; +if (new_size > seg->offset + seg->length) + goto out; + } nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lcp->lc_sid, false, lcp->lc_layout_type,
And lastly:
- if (new_size > i_size_read(inode)) {
- lcp->lc_size_chg = true;
- lcp->lc_newsize = new_size;
- } else {
- lcp->lc_size_chg = false;
- }
Also, I notice that nfsd4_decode_layoutcommit() has:
if (xdr_stream_decode_bool(argp->xdr, &lcp->lc_reclaim) < 0)
return nfserr_bad_xdr;but:
if (xdr_stream_decode_u32(argp->xdr, &lcp->lc_newoffset) < 0)
return nfserr_bad_xdr;The no_newoffset field should be decoded with xdr_stream_decode_bool too (though the end result is the same). For just this nit, please make a separate patch. Thanks!
Signature · 1 line
Chuck Lever
Hi Chuck,
Thanks for your kind guidance, it really helps to understand the process of code contribution better! I'm on vacation next week, will rework and resubmit updated patches when I get back.
--
Sergey Bashirov
Quoted 5 lines · Sergey Bashirov
Hi Chuck,
Thanks for your kind guidance, it really helps to understand the process of code contribution better! I'm on vacation next week, will rework and resubmit updated patches when I get back.
Dai, Jeff, and I appreciate your help with NFSD's pNFS implementation!
Signature · 1 line
Chuck Lever
Quoted 20 lines · Sergey Bashirov
The data type of
loca_last_write_offsetis newoffset4 and is switched on a boolean value,no_newoffset, that indicates if a previous write occurred or not. Ifno_newoffsetis FALSE, an offset is not given. This means that client does not try to update the file size. Thus, server should not try to calculate new file size and check if it fits into the seg range.Co-developed-by Konstantin Evtushenko <[email protected]>Signed-off-by Konstantin Evtushenko <[email protected]>Signed-off-by Sergey Bashirov <[email protected]>--- fs/nfsd/blocklayout.c | 2 +- fs/nfsd/nfs4proc.c | 16 ++++++++-------- 2 files changed, 9 insertions(+), 9 deletions(-)fs/nfsd/blocklayout.c +0
-0
fs/nfsd/blocklayout.c nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, · line 123
Quoted 7 lines
fs/nfsd/blocklayout.c +1
-1@@ -118,7 +118,7 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp, struct iomap *iomaps, int nr_iomaps) {struct timespec64 mtime =inode_get_mtime (inode); -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;struct iattr iattr = { .ia_valid =0 };int error;
Please guard the entire new_size check below instead, i.e.
if (lcp->lc_newoffset) {
loff_t new_size = lcp->lc_last_wr + 1;
if (new_size > i_size_read(inode)) {
iattr.ia_valid |= ATTR_SIZE;
iattr.ia_size = new_size;
}
}fs/nfsd/nfs4proc.c nfsd4_layoutcommit(struct svc_rqst *rqstp, · line 2485
Quoted 10 lines
fs/nfsd/nfs4proc.c +1
-1@@ -2482,7 +2482,7 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,const struct nfsd4_layout_seg *seg = &lcp->lc_seg;struct svc_fh *current_fh = &cstate->current_fh;const struct nfsd4_layout_ops *ops; -loff_t new_size = lcp->lc_last_wr +1 ; +loff_t new_size = (lcp->lc_newoffset) ? lcp->lc_last_wr +1 :0 ;
Same here.