- 01 Aug, 2017 1 commit
-
-
Trond Myklebust authored
The verifier is allocated on the stack, but the EXCHANGE_ID RPC call was changed to be asynchronous by commit 8d89bd70. If we interrrupt the call to rpc_wait_for_completion_task(), we can therefore end up transmitting random stack contents in lieu of the verifier. Fixes: 8d89bd70 ("NFS setup async exchange_id") Cc: stable@vger.kernel.org # v4.9+ Signed-off-by:
Trond Myklebust <trond.myklebust@primarydata.com> Signed-off-by:
Anna Schumaker <Anna.Schumaker@Netapp.com>
-
- 28 Jul, 2017 1 commit
-
-
Benjamin Coddington authored
nfs4_retry_setlk() sets the task's state to TASK_INTERRUPTIBLE within the same region protected by the wait_queue's lock after checking for a notification from CB_NOTIFY_LOCK callback. However, after releasing that lock, a wakeup for that task may race in before the call to freezable_schedule_timeout_interruptible() and set TASK_WAKING, then freezable_schedule_timeout_interruptible() will set the state back to TASK_INTERRUPTIBLE before the task will sleep. The result is that the task will sleep for the entire duration of the timeout. Since we've already set TASK_INTERRUPTIBLE in the locked section, just use freezable_schedule_timout() instead. Fixes: a1d617d8 ("nfs: allow blocking locks to be awoken by lock callbacks") Signed-off-by:
Benjamin Coddington <bcodding@redhat.com> Reviewed-by:
Jeff Layton <jlayton@redhat.com> Cc: stable@vger.kernel.org # v4.9+ Signed-off-by:
-
- 27 Jul, 2017 2 commits
-
-
NeilBrown authored
posix_fallocate() will allocate space in an NFS file by considering the last byte of every 4K block. If it is before EOF, it will read the byte and if it is zero, a zero is written out. If it is after EOF, the zero is unconditionally written. For the blocks beyond EOF, if NFS believes its cache is valid, it will expand these writes to write full pages, and then will merge the pages. This results if (typically) 1MB writes. If NFS believes its cache is not valid (particularly if NFS_INO_INVALID_DATA or NFS_INO_REVAL_PAGECACHE are set - see nfs_write_pageuptodate()), it will send the individual 1-byte writes. This results in (typically) 256 times as many RPC requests, and can be substantially slower. Currently nfs_revalidate_mapping() is only used when reading a file or mmapping a file, as these are times when the content needs to be up-to-date. Writes don't generally need the cache to be up-to-date, but writes beyond EOF can benefit, particularly in the posix_fallocate() case. So this patch calls nfs_revalidate_mapping() when writing beyond EOF - i.e. when there is a gap between the end of the file and the start of the write. If the cache is thought to be out of date (as happens after taking a file lock), this will cause a GETATTR, and the two flags mentioned above will be cleared. With this, posix_fallocate() on a newly locked file does not generate excessive tiny writes. Signed-off-by:
NeilBrown <neilb@suse.com> Signed-off-by:
-
NeilBrown authored
Prior to commit ca0daa27 ("NFS: Cache aggressively when file is open for writing"), NFS would revalidate, or invalidate, the file size when taking a lock. Since that commit it only invalidates the file content. If the file size is changed on the server while wait for the lock, the client will have an incorrect understanding of the file size and could corrupt data. This particularly happens when writing beyond the (supposed) end of file and can be easily be demonstrated with posix_fallocate(). If an application opens an empty file, waits for a write lock, and then calls posix_fallocate(), glibc will determine that the underlying filesystem doesn't support fallocate (assuming version 4.1 or earlier) and will write out a '0' byte at the end of each 4K page in the region being fallocated that is after the end of the file. NFS will (usually) detect that these writes are beyond EOF and will expand them to cover the whole page, and then will merge the pages. Consequently, NFS will write out large blocks of zeroes beyond where it thought EOF was. If EOF had moved, the pre-existing part of the file will be over-written. Locking should have protected against this, but it doesn't. This patch restores the use of nfs_zap_caches() which invalidated the cached attributes. When posix_fallocate() asks for the file size, the request will go to the server and get a correct answer. cc: stable@vger.kernel.org (v4.8+) Fixes: ca0daa27 ("NFS: Cache aggressively when file is open for writing") Signed-off-by:
-
- 26 Jul, 2017 1 commit
-
-
Anna Schumaker authored
Commit bd8b2441 ("NFS: Store the raw NFS access mask in the inode's access cache") changed how the access results are stored after an access() call. An NFS v4 OPEN might have access bits returned with the opendata, so we should use the NFS4_ACCESS values when determining the return value in nfs4_opendata_access(). Fixes: bd8b2441 ("NFS: Store the raw NFS access mask in the inode's access cache") Reported-by:
Eryu Guan <eguan@redhat.com> Signed-off-by:
Takashi Iwai <tiwai@suse.de>
-
- 21 Jul, 2017 7 commits
-
-
Trond Myklebust authored
We must set fl->dsaddr once, and once only, even if there are multiple processes calling filelayout_check_deviceid() for the same layout segment. Reported-by:
Olga Kornievskaia <kolga@netapp.com> Signed-off-by:
-
Trond Myklebust authored
When mapping a directory, we want the MAY_WRITE permissions to reflect whether or not we have permission to modify, add and delete the directory entries. MAY_EXEC must map to lookup permissions. On the other hand, for files, we want MAY_WRITE to reflect a permission to modify and extend the file. Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
NeilBrown authored
Commit 3d476263 ("tcp: remove poll() flakes when receiving RST") in v4.12 changed the order in which ->sk_state_change() and ->sk_error_report() are called when a socket is shut down - sk_state_change() is now called first. This causes xs_tcp_state_change() -> xs_sock_mark_closed() -> xprt_disconnect_done() to wake all pending tasked with -EAGAIN. When the ->sk_error_report() callback arrives, it is too late to pass the error on, and it is lost. As easy way to demonstrate the problem caused is to try to start rpc.nfsd while rcpbind isn't running. nfsd will attempt a tcp connection to rpcbind. A ECONNREFUSED error is returned, but sunrpc code loses the error and keeps retrying. If it saw the ECONNREFUSED, it would abort. To fix this, handle the sk->sk_err in the TCP_CLOSE branch of xs_tcp_state_change(). Fixes: 3d476263 ("tcp: remove poll() flakes when receiving RST") Cc: stable@vger.kernel.org (v4.12) Signed-off-by:
-
Eryu Guan authored
Array size of mnt3_counts should be the size of array mnt3_procedures, not mnt_procedures, though they're same in size right now. Found this by code inspection. Fixes: 1c5876dd ("sunrpc: move p_count out of struct rpc_procinfo") Cc: Christoph Hellwig <hch@lst.de> Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
- 19 Jul, 2017 5 commits
-
-
Trond Myklebust authored
Doing the test without taking any locks is racy, and so really it makes more sense to do it in the flexfiles code (which is the only case that cares). Signed-off-by:
-
Trond Myklebust authored
If the layout has expired due to a fencing event, then we should not attempt to commit to the DS. Signed-off-by:
-
Trond Myklebust authored
We must make sure that cinfo->ds->ncommitting is in sync with the commit list, since it is checked as part of pnfs_commit_list(). Signed-off-by:
-
Trond Myklebust authored
We must make sure that cinfo->ds->nwritten is in sync with the commit list, since it is checked as part of pnfs_scan_commit_lists(). Signed-off-by:
-
Steve Dickson authored
Doing this copy eliminates the "port=0" entry in the /proc/mounts entries Fixes: https://bugzilla.kernel.org/show_bug.cgi?id=69241Signed-off-by:
Steve Dickson <steved@redhat.com> Signed-off-by:
-
- 13 Jul, 2017 23 commits
-
-
Trond Myklebust authored
"perf lock" shows fairly heavy contention for the bit waitqueue locks when doing an I/O heavy workload. Use a bit to tell whether or not there has been contention for a lock so that we can optimise away the bit waitqueue options in those cases. Signed-off-by:
-
Peng Tao authored
This support for opening files on NFS by file handle, both through the open_by_handle syscall, and for re-exporting NFS (for example using a different version). The support is very basic for now, as each open by handle will have to do an NFSv4 open operation on the wire. In the future this will hopefully be mitigated by an open file cache, as well as various optimizations in NFS for this specific case. Signed-off-by:
Peng Tao <tao.peng@primarydata.com> [hch: incorporated various changes, resplit the patches, new changelog] Signed-off-by:
-
Jeff Layton authored
This will be needed in order to implement the get_parent export op for nfsd. Signed-off-by:
Jeff Layton <jeff.layton@primarydata.com> Signed-off-by:
-
Peng Tao authored
This helper will allow to find an existing NFS inode by the file handle and fattr. Signed-off-by:
-
Peng Tao authored
It's a trival change but follows knfsd export document that asks for d_splice_alias during lookup. Signed-off-by:
-
Jason A. Donenfeld authored
Otherwise, we enable a MAC forgery via timing attack. Signed-off-by:
Jason A. Donenfeld <Jason@zx2c4.com> Cc: "J. Bruce Fields" <bfields@fieldses.org> Cc: Jeff Layton <jlayton@poochiereds.net> Cc: Trond Myklebust <trond.myklebust@primarydata.com> Cc: Anna Schumaker <anna.schumaker@netapp.com> Cc: linux-nfs@vger.kernel.org Cc: stable@vger.kernel.org Signed-off-by:
-
Olga Kornievskaia authored
Return size of COPY is u64 but it was assigned to an "int" status. Signed-off-by:
-
Chuck Lever authored
Clean up. FASTREG and LOCAL_INV WRs are typically not signaled. localinv_wake is used for the last LOCAL_INV WR in a chain, which is always signaled. The documenting comments should reflect that. Signed-off-by:
Chuck Lever <chuck.lever@oracle.com> Signed-off-by:
-
Chuck Lever authored
Clean up. Reported by: Geliang Tang <geliangtang@gmail.com> Signed-off-by:
-
Chuck Lever authored
Clean up. Commit 38f1932e ("xprtrdma: Remove FMRs from the unmap list after unmapping") utilized list_del_init() to try to prevent some list corruption. The corruption was actually caused by the reply handler racing with a signal. Now that MR invalidation is properly serialized, list_del_init() can safely be replaced. Signed-off-by:
-
Chuck Lever authored
Some have complained about the log messages generated when xprtrdma opens or closes a connection to a server. When an NFS mount is mostly idle these can appear every few minutes as the client idles out the connection and reconnects. Connection and disconnection is a normal part of operation, and not exceptional, so change these to dprintk's for now. At some point all of these will be converted to tracepoints, but that's for another day. Signed-off-by:
-
Chuck Lever authored
Transparent State Migration copies a client's lease state from the server where a filesystem used to reside to the server where it now resides. When an NFSv4.1 client first contacts that destination server, it uses EXCHANGE_ID to detect trunking relationships. The lease that was copied there is returned to that client, but the destination server sets EXCHGID4_FLAG_CONFIRMED_R when replying to the client. This is because the lease was confirmed on the source server (before it was copied). When CONFIRMED_R is set, the client throws away the sequence ID returned by the server. During a Transparent State Migration, however there's no other way for the client to know what sequence ID to use with a lease that's been migrated. Therefore, the client must save and use the contrived slot sequence value returned by the destination server even when CONFIRMED_R is set. Note that some servers always return a seqid of 1 after a migration. Reported-by:
Xuan Qi <xuan.qi@oracle.com> Signed-off-by:
-
Chuck Lever authored
Transparent State Migration copies a client's lease state from the server where a filesystem used to reside to the server where it now resides. When an NFSv4.1 client first contacts that destination server, it uses EXCHANGE_ID to detect trunking relationships. The lease that was copied there is returned to that client, but the destination server sets EXCHGID4_FLAG_CONFIRMED_R when replying to the client. This is because the lease was confirmed on the source server (before it was copied). Normally, when CONFIRMED_R is set, a client purges the lease and creates a new one. However, that throws away the entire benefit of Transparent State Migration. Therefore, the client must not purge that lease when it is possible that Transparent State Migration has occurred. Reported-by:
-
Chuck Lever authored
Deferred MR recovery does a DMA-unmapping of the MW. However, ro_map invokes rpcrdma_defer_mr_recovery in some error cases where the MW has not even been DMA-mapped yet. Avoid a DMA-unmapping error replacing rpcrdma_defer_mr_recovery. Also note that if ib_dma_map_sg is asked to map 0 nents, it will return 0. So the extra "if (i == 0)" check is no longer needed. Fixes: 42fe28f6 ("xprtrdma: Do not leak an MW during a DMA ...") Fixes: 505bbe64 ("xprtrdma: Refactor MR recovery work queues") Signed-off-by:
-
Chuck Lever authored
When ib_post_send() fails, all LOCAL_INV WRs past @bad_wr have to be examined, and the MRs reset by hand. I'm not sure how the existing code can work by comparing R_keys. Restructure the logic so that instead it walks the chain of WRs, starting from the first bad one. Make sure to wait for completion if at least one WR was actually posted. Otherwise, if the ib_post_send fails, we can end up DMA-unmapping the MR while LOCAL_INV operations are in flight. Commit 7a89f9c6 ("xprtrdma: Honor ->send_request API contract") added the rdma_disconnect() call site. The disconnect actually causes more problems than it solves, and SQ overruns happen only as a result of software bugs. So remove it. Fixes: d7a21c1b ("xprtrdma: Reset MRs in frwr_op_unmap_sync()") Signed-off-by:
-
Chuck Lever authored
After a signal, the RPC client aborts synchronous RPCs running on behalf of the signaled application. The server is still executing those RPCs, and will write the results back into the client's memory when it's done. By the time the server writes the results, that memory is likely being used for other purposes. Therefore xprtrdma has to immediately invalidate all memory regions used by those aborted RPCs to prevent the server's writes from clobbering that re-used memory. With FMR memory registration, invalidation takes a relatively long time. In fact, the invalidation is often still running when the server tries to write the results into the memory regions that are being invalidated. This sets up a race between two processes: 1. After the signal, xprt_rdma_free calls ro_unmap_safe. 2. While ro_unmap_safe is still running, the server replies and rpcrdma_reply_handler runs, calling ro_unmap_sync. Both processes invoke ib_unmap_fmr on the same FMR. The mlx4 driver allows two ib_unmap_fmr calls on the same FMR at the same time, but HCAs generally don't tolerate this. Sometimes this can result in a system crash. If the HCA happens to survive, rpcrdma_reply_handler continues. It removes the rpc_rqst from rq_list and releases the transport_lock. This enables xprt_rdma_free to run in another process, and the rpc_rqst is released while rpcrdma_reply_handler is still waiting for the ib_unmap_fmr call to finish. But further down in rpcrdma_reply_handler, the transport_lock is taken again, and "rqst" is dereferenced. If "rqst" has already been released, this triggers a general protection fault. Since bottom- halves are disabled, the system locks up. Address both issues by reversing the order of the xprt_lookup_rqst call and the ro_unmap_sync call. Introduce a separate lookup mechanism for rpcrdma_req's to enable calling ro_unmap_sync before xprt_lookup_rqst. Now the handler takes the transport_lock once and holds it for the XID lookup and RPC completion. BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=305 Fixes: 68791649 ('xprtrdma: Invalidate in the RPC reply ... ') Signed-off-by: -
Chuck Lever authored
Clean up: I'm about to use the rl_free field for purposes other than a free list. So use a more generic name. This is a refactoring change only. BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=305 Fixes: 68791649 ('xprtrdma: Invalidate in the RPC reply ... ') Signed-off-by:
-
Chuck Lever authored
There are rare cases where an rpcrdma_req can be re-used (via rpcrdma_buffer_put) while the RPC reply handler is still running. This is due to a signal firing at just the wrong instant. Since commit 9d6b0409 ("xprtrdma: Place registered MWs on a per-req list"), rpcrdma_mws are self-contained; ie., they fully describe an MR and scatterlist, and no part of that information is stored in struct rpcrdma_req. As part of closing the above race window, pass only the req's list of registered MRs to ro_unmap_sync, rather than the rpcrdma_req itself. Some extra transport header sanity checking is removed. Since the client depends on its own recollection of what memory had been registered, there doesn't seem to be a way to abuse this change. And, the check was not terribly effective. If the client had sent Read chunks, the "list_empty" test is negative in both of the removed cases, which are actually looking for Write or Reply chunks. BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=305 Fixes: 68791649 ('xprtrdma: Invalidate in the RPC reply ... ') Signed-off-by:
-
Chuck Lever authored
There are rare cases where an rpcrdma_req and its matched rpcrdma_rep can be re-used, via rpcrdma_buffer_put, while the RPC reply handler is still using that req. This is typically due to a signal firing at just the wrong instant. As part of closing this race window, avoid using the wrong rpcrdma_rep to detect remotely invalidated MRs. Mark MRs as invalidated while we are sure the rep is still OK to use. BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=305 Fixes: 68791649 ('xprtrdma: Invalidate in the RPC reply ... ') Signed-off-by:
-
Chuck Lever authored
Callers assume the ro_unmap_sync and ro_unmap_safe methods empty the list of registered MRs. Ensure that all paths through fmr_op_unmap_sync() remove MWs from that list. Fixes: 9d6b0409 ("xprtrdma: Place registered MWs on a ... ") Signed-off-by:
-
NeilBrown authored
If an NFS server returns a filehandle that we have previously seen, and reports a different type, then nfs_refresh_inode() will log a warning and return an error. nfs_fhget() does not check for this error and may return an inode with a different type than the one that the server reported. This is likely to cause confusion, and is one way that ->open_context() could return a directory inode as discussed in the previous patch. So if nfs_refresh_inode() returns and error, return that error from nfs_fhget() to avoid the confusion propagating. Signed-off-by:
-
NeilBrown authored
A confused server could return a filehandle for an NFSv4 OPEN request, which it previously returned for a directory. So the inode returned by ->open_context() in nfs_atomic_open() could conceivably be a directory inode. This has particular implications for the call to nfs_file_set_open_context() in nfs_finish_open(). If that is called on a directory inode, then the nfs_open_context that gets stored in the filp->private_data will be linked to nfs_inode->open_files. When the directory is closed, nfs_closedir() will (ultimately) free the ->private_data, but not unlink it from nfs_inode->open_files (because it doesn't expect an nfs_open_context there). Subsequently the memory could get used for something else and eventually if the ->open_files list is walked, the walker will fall off the end and crash. So: change nfs_finish_open() to only call nfs_file_set_open_context() for regular-file inodes. This failure mode has been seen in a production setting (unknown NFS server implementation). The kernel was v3.0 and the specific sequence seen would not affect more recent kernels, but I think a risk is still present, and caution is wise. Signed-off-by:
-
NeilBrown authored
Since commit bafc9b75 ("vfs: More precise tests in d_invalidate") in v3.18, a return of '0' from ->d_revalidate() will cause the dentry to be invalidated even if it has filesystems mounted on or it or on a descendant. The mounted filesystem is unmounted. This means we need to be careful not to return 0 unless the directory referred to truly is invalid. So -ESTALE or -ENOENT should invalidate the directory. Other errors such a -EPERM or -ERESTARTSYS should be returned from ->d_revalidate() so they are propagated to the caller. A particular problem can be demonstrated by: 1/ mount an NFS filesystem using NFSv3 on /mnt 2/ mount any other filesystem on /mnt/foo 3/ ls /mnt/foo 4/ turn off network, or otherwise make the server unable to respond 5/ ls /mnt/foo & 6/ cat /proc/$!/stack # note that nfs_lookup_revalidate is in the call stack 7/ kill -9 $! # this results in -ERESTARTSYS being returned 8/ observe that /mnt/foo has been unmounted. This patch changes nfs_lookup_revalidate() to only treat -ESTALE from nfs_lookup_verify_inode() and -ESTALE or -ENOENT from ->lookup() as indicating an invalid inode. Other errors are returned. Also nfs_check_inode_attributes() is changed to return -ESTALE rather than -EIO. This is consistent with the error returned in similar circumstances from nfs_update_inode(). As this bug allows any user to unmount a filesystem mounted on an NFS filesystem, this fix is suitable for stable kernels. Fixes: bafc9b75 ("vfs: More precise tests in d_invalidate") Cc: stable@vger.kernel.org (v3.18+) Signed-off-by:
-
