- 02 Dec, 2020 38 commits
-
-
Trond Myklebust authored
Use the existing xdr_stream_decode_string_dup() to safely decode into kmalloced strings. Signed-off-by:Trond Myklebust <trond.myklebust@hammerspace.com>
-
Trond Myklebust authored
Signed-off-by: -
Trond Myklebust authored
Ensure that we report the correct netid when using UDP or RDMA transports to the DSes. Signed-off-by: -
Trond Myklebust authored
We want to enable RDMA and UDP as valid transport methods if a GETDEVICEINFO call specifies it. Do so by adding a parser for the netid that translates it to an appropriate argument for the RPC transport layer. Signed-off-by: -
Trond Myklebust authored
Signed-off-by: -
Trond Myklebust authored
If the pNFS metadata server advertises multiple addresses for the same data server, we should try to connect to just one protocol family and transport type on the assumption that homogeneity will improve performance. Signed-off-by: -
Trond Myklebust authored
Signed-off-by: -
Trond Myklebust authored
Switch the mount code to use xprt_find_transport_ident() and to check the results before allowing the mount to proceed. Signed-off-by: -
Trond Myklebust authored
Signed-off-by: -
Trond Myklebust authored
After we've looked up the transport module, we need to ensure it can't go away until we've finished running the transport setup code. Signed-off-by: -
Trond Myklebust authored
According to RFC5666, the correct netid for an IPv6 addressed RDMA transport is "rdma6", which we've supported as a mount option since Linux-4.7. The problem is when we try to load the module "xprtrdma6", that will fail, since there is no modulealias of that name. Fixes: 181342c5 ("xprtrdma: Add rdma6 option to support NFS/RDMA IPv6") Signed-off-by:
-
Trond Myklebust authored
If the directory is changing, causing the page cache to get invalidated while we are listing the contents, then the NFS client is currently forced to read in the entire directory contents from scratch, because it needs to perform a linear search for the readdir cookie. While this is not an issue for small directories, it does not scale to directories with millions of entries. In order to be able to deal with large directories that are changing, add a heuristic to ensure that if the page cache is empty, and we are searching for a cookie that is not the zero cookie, we just default to performing uncached readdir. Signed-off-by:
Benjamin Coddington <bcodding@redhat.com> Tested-by:
Dave Wysochanski <dwysocha@redhat.com>
-
Trond Myklebust authored
If we're doing uncached readdir, allocate multiple pages in order to try to avoid duplicate RPC calls for the same getdents() call. Signed-off-by:
-
Trond Myklebust authored
If the server is handing out monotonically increasing readdir cookie values, then we can optimise away searches through pages that contain cookies that lie outside our search range. Signed-off-by:
-
Trond Myklebust authored
If the server insists on using the readdir verifiers in order to allow cookies to expire, then we should ensure that we cache the verifier with the cookie, so that we can return an error if the application tries to use the expired cookie. Signed-off-by:
-
Trond Myklebust authored
If the server returns NFS4ERR_NOT_SAME or tells us that the cookie is bad in response to a READDIR call, then we should empty the page cache so that we can fill it from scratch again. Signed-off-by:
-
Trond Myklebust authored
If we're ever going to allow support for servers that use the readdir verifier, then that use needs to be managed by the middle layers as those need to be able to reject cookies from other verifiers. Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Trond Myklebust authored
The descriptor and the struct nfs_entry are both large structures, so don't allocate them from the stack. Signed-off-by:
-
Trond Myklebust authored
Clean up nfs_do_filldir(). Signed-off-by:
-
Trond Myklebust authored
Remove the redundant caching of the credential in struct nfs_open_dir_context. Pass the buffer size as an argument to nfs_readdir_xdr_filler(). Signed-off-by:
-
Trond Myklebust authored
Support readdir buffers of up to 1MB in size so that we can read large directories using few RPC calls. Signed-off-by:
-
Trond Myklebust authored
We don't need to store a hash, so replace struct qstr with a simple const char pointer and length. Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Trond Myklebust authored
The kmapped pointer is only used once per loop to check if we need to exit. Signed-off-by:
-
Trond Myklebust authored
If a readdir call returns more data than we can fit into one page cache page, then allocate a new one for that data rather than discarding the data. Signed-off-by:
-
Trond Myklebust authored
Refactor to use pagecache_get_page() so that we can fill the page in multiple stages. Signed-off-by:
-
Trond Myklebust authored
Clean up handling of the case where there are no entries in the readdir reply. Signed-off-by:
-
Trond Myklebust authored
Since the 'eof_index' is only ever used as a flag, make it so. Also add a flag to detect if the page has been completely filled. Signed-off-by:
-
Trond Myklebust authored
Ensure that the contents of struct nfs_open_dir_context are consistent by setting them under the file->f_lock from a private copy (that is known to be consistent). Signed-off-by:
-
Olga Kornievskaia authored
Currently, the client will always ask for security_labels if the server returns that it supports that feature regardless of any LSM modules (such as Selinux) enforcing security policy. This adds performance penalty to the READDIR operation. Client adjusts superblock's support of the security_label based on the server's support but also current client's configuration of the LSM modules. Thus, prior to using the default bitmask in READDIR, this patch checks the server's capabilities and then instructs READDIR to remove FATTR4_WORD2_SECURITY_LABEL from the bitmask. v5: fixing silly mistakes of the rushed v4 v4: simplifying logic v3: changing label's initialization per Ondrej's comment v2: dropping selinux hook and using the sb cap. Suggested-by:
Ondrej Mosnacek <omosnace@redhat.com> Suggested-by:
Scott Mayhew <smayhew@redhat.com> Signed-off-by:
Olga Kornievskaia <kolga@netapp.com> Fixes: 2b0143b5 ("VFS: normal filesystems (and lustre): d_inode() annotations") Signed-off-by:
-
Trond Myklebust authored
Currently, we wake up the tasks by priority queue ordering, which means that we ignore the batching that is supposed to help with QoS issues. Fixes: c049f8ea ("SUNRPC: Remove the bh-safe lock requirement on the rpc_wait_queue->lock") Signed-off-by:
-
Trond Myklebust authored
We need to respect the NFS_MOUNT_SOFTREVAL flag in _nfs4_proc_lookupp, by timing out if the server is unavailable. Signed-off-by: -
Trond Myklebust authored
In order to use the open_by_filehandle() operations on NFSv3, we need to be able to emulate lookupp() so that nfs_get_parent() can be used to convert disconnected dentries into connected ones. Signed-off-by: -
Trond Myklebust authored
We want to reuse the lookup code in NFSv3 in order to emulate the NFSv4 lookupp operation. Signed-off-by: -
Dai Ngo authored
Since commit b4868b44 ("NFSv4: Wait for stateid updates after CLOSE/OPEN_DOWNGRADE"), every inter server copy operation suffers 5 seconds delay regardless of the size of the copy. The delay is from nfs_set_open_stateid_locked when the check by nfs_stateid_is_sequential fails because the seqid in both nfs4_state and nfs4_stateid are 0. Fix __nfs42_ssc_open to delay setting of NFS_OPEN_STATE in nfs4_state, until after the call to update_open_stateid, to indicate this is the 1st open. This fix is part of a 2 patches, the other patch is the fix in the source server to return the stateid for COPY_NOTIFY request with seqid 1 instead of 0. Fixes: ce0887ac ("NFSD add nfs4 inter ssc to nfsd4_copy") Signed-off-by:
Dai Ngo <dai.ngo@oracle.com> Signed-off-by:
Anna Schumaker <Anna.Schumaker@Netapp.com>
-
Chuck Lever authored
By switching to an XFS-backed export, I am able to reproduce the ibcomp worker crash on my client with xfstests generic/013. For the failing LISTXATTRS operation, xdr_inline_pages() is called with page_len=12 and buflen=128. - When ->send_request() is called, rpcrdma_marshal_req() does not set up a Reply chunk because buflen is smaller than the inline threshold. Thus rpcrdma_convert_iovs() does not get invoked at all and the transport's XDRBUF_SPARSE_PAGES logic is not invoked on the receive buffer. - During reply processing, rpcrdma_inline_fixup() tries to copy received data into rq_rcv_buf->pages because page_len is positive. But there are no receive pages because rpcrdma_marshal_req() never allocated them. The result is that the ibcomp worker faults and dies. Sometimes that causes a visible crash, and sometimes it results in a transport hang without other symptoms. RPC/RDMA's XDRBUF_SPARSE_PAGES support is not entirely correct, and should eventually be fixed or replaced. However, my preference is that upper-layer operations should explicitly allocate their receive buffers (using GFP_KERNEL) when possible, rather than relying on XDRBUF_SPARSE_PAGES. Reported-by:
Chuck Lever <chuck.lever@oracle.com> Reviewed-by:
Frank van der Linden <fllinden@amazon.com> Tested-by:
-
Chuck Lever authored
There's no need to defer allocation of pages for the receive buffer. - This upcall is quite infrequent - gssp_alloc_receive_pages() can allocate the pages with GFP_KERNEL, unlike the transport - gssp_alloc_receive_pages() knows exactly how many pages are needed Signed-off-by:
Olga Kornievskaia <aglo@umich.edu> Signed-off-by:
-
- 30 Nov, 2020 1 commit
-
-
Trond Myklebust authored
If the flexfiles mirroring is enabled, then the read code expects to be able to set pgio->pg_mirror_idx to point to the data server that is being used for this particular read. However it does not change the pg_mirror_count because we only need to send a single read. Signed-off-by:
-
- 29 Nov, 2020 1 commit
-
-
Linus Torvalds authored
-
