- 25 May, 2016 2 commits
-
-
Weston Andros Adamson authored
Since the patch "NFS: Allow multiple commit requests in flight per file" we can run multiple simultaneous commits on the same inode. This introduced a race over collecting pages to commit that made it possible to call nfs_init_commit() with an empty list - which causes crashes like the one below. The fix is to catch this race and avoid calling nfs_init_commit and initiate_commit when there is no work to do. Here is the crash: [600522.076832] BUG: unable to handle kernel NULL pointer dereference at 0000000000000040 [600522.078475] IP: [<ffffffffa0479e72>] nfs_init_commit+0x22/0x130 [nfs] [600522.078745] PGD 4272b1067 PUD 4272cb067 PMD 0 [600522.078972] Oops: 0000 [#1] SMP [600522.079204] Modules linked in: nfsv3 nfs_layout_flexfiles rpcsec_gss_krb5 nfsv4 dns_resolver nfs fscache dcdbas ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ebtable_nat ebtable_broute bridge stp llc ebtable_filter ebtables ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_security ip6table_raw ip6table_filter ip6_tables iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack iptable_mangle iptable_security iptable_raw vmw_vsock_vmci_transport vsock bonding ipmi_devintf ipmi_msghandler coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel ppdev vmw_balloon parport_pc parport acpi_cpufreq vmw_vmci i2c_piix4 shpchp nfsd auth_rpcgss nfs_acl lockd grace sunrpc xfs libcrc32c vmwgfx drm_kms_helper ttm drm crc32c_intel serio_raw vmxnet3 [600522.081380] vmw_pvscsi ata_generic pata_acpi [600522.081809] CPU: 3 PID: 15667 Comm: /usr/bin/python Not tainted 4.1.9-100.pd.88.el7.x86_64 #1 [600522.082281] Hardware name: VMware, Inc. VMware Virtual Platform/440BX Desktop Reference Platform, BIOS 6.00 09/30/2014 [600522.082814] task: ffff8800bbbfa780 ti: ffff88042ae84000 task.ti: ffff88042ae84000 [600522.083378] RIP: 0010:[<ffffffffa0479e72>] [<ffffffffa0479e72>] nfs_init_commit+0x22/0x130 [nfs] [600522.083973] RSP: 0018:ffff88042ae87438 EFLAGS: 00010246 [600522.084571] RAX: 0000000000000000 RBX: ffff880003485e40 RCX: ffff88042ae87588 [600522.085188] RDX: 0000000000000000 RSI: ffff88042ae874b0 RDI: ffff880003485e40 [600522.085756] RBP: ffff88042ae87448 R08: ffff880003486010 R09: ffff88042ae874b0 [600522.086332] R10: 0000000000000000 R11: 0000000000000005 R12: ffff88042ae872d0 [600522.086905] R13: ffff88042ae874b0 R14: ffff880003485e40 R15: ffff88042704c840 [600522.087484] FS: 00007f4728ff2740(0000) GS:ffff88043fd80000(0000) knlGS:0000000000000000 [600522.088070] CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b [600522.088663] CR2: 0000000000000040 CR3: 000000042b6aa000 CR4: 00000000001406e0 [600522.089327] Stack: [600522.089926] 0000000000000001 ffff88042ae87588 ffff88042ae874f8 ffffffffa04f09fa [600522.090549] 0000000000017840 0000000000017840 ffff88042ae87588 ffff8803258d9930 [600522.091169] ffff88042ae87578 ffffffffa0563d80 0000000000000000 ffff88042704c840 [600522.091789] Call Trace: [600522.092420] [<ffffffffa04f09fa>] pnfs_generic_commit_pagelist+0x1da/0x320 [nfsv4] [600522.093052] [<ffffffffa0563d80>] ? ff_layout_commit_prepare_v3+0x30/0x30 [nfs_layout_flexfiles] [600522.093696] [<ffffffffa0562645>] ff_layout_commit_pagelist+0x15/0x20 [nfs_layout_flexfiles] [600522.094359] [<ffffffffa047bc78>] nfs_generic_commit_list+0xe8/0x120 [nfs] [600522.095032] [<ffffffffa047bd6a>] nfs_commit_inode+0xba/0x110 [nfs] [600522.095719] [<ffffffffa046ac54>] nfs_release_page+0x44/0xd0 [nfs] [600522.096410] [<ffffffff811a8122>] try_to_release_page+0x32/0x50 [600522.097109] [<ffffffff811bd4f1>] shrink_page_list+0x961/0xb30 [600522.097812] [<ffffffff811bdced>] shrink_inactive_list+0x1cd/0x550 [600522.098530] [<ffffffff811bea65>] shrink_lruvec+0x635/0x840 [600522.099250] [<ffffffff811bed60>] shrink_zone+0xf0/0x2f0 [600522.099974] [<ffffffff811bf312>] do_try_to_free_pages+0x192/0x470 [600522.100709] [<ffffffff811bf6ca>] try_to_free_pages+0xda/0x170 [600522.101464] [<ffffffff811b2198>] __alloc_pages_nodemask+0x588/0x970 [600522.102235] [<ffffffff811fbbd5>] alloc_pages_vma+0xb5/0x230 [600522.103000] [<ffffffff813a1589>] ? cpumask_any_but+0x39/0x50 [600522.103774] [<ffffffff811d6115>] wp_page_copy.isra.55+0x95/0x490 [600522.104558] [<ffffffff810e3438>] ? __wake_up+0x48/0x60 [600522.105357] [<ffffffff811d7d3b>] do_wp_page+0xab/0x4f0 [600522.106137] [<ffffffff810a1bbb>] ? release_task+0x36b/0x470 [600522.106902] [<ffffffff8126dbd7>] ? eventfd_ctx_read+0x67/0x1c0 [600522.107659] [<ffffffff811da2a8>] handle_mm_fault+0xc78/0x1900 [600522.108431] [<ffffffff81067ef1>] __do_page_fault+0x181/0x420 [600522.109173] [<ffffffff811446a6>] ? __audit_syscall_exit+0x1e6/0x280 [600522.109893] [<ffffffff810681c0>] do_page_fault+0x30/0x80 [600522.110594] [<ffffffff81024f36>] ? syscall_trace_leave+0xc6/0x120 [600522.111288] [<ffffffff81790a58>] page_fault+0x28/0x30 [600522.111947] Code: 5d c3 0f 1f 80 00 00 00 00 0f 1f 44 00 00 55 4c 8d 87 d0 01 00 00 48 89 e5 53 48 89 fb 48 83 ec 08 4c 8b 0e 49 8b 41 18 4c 39 ce <48> 8b 40 40 4c 8b 50 30 74 24 48 8b 87 d0 01 00 00 48 8b 7e 08 [600522.113343] RIP [<ffffffffa0479e72>] nfs_init_commit+0x22/0x130 [nfs] [600522.114003] RSP <ffff88042ae87438> [600522.114636] CR2: 0000000000000040 Fixes: af7cf057 (NFS: Allow multiple commit requests in flight per file) CC: stable@vger.kernel.org Signed-off-by:
Weston Andros Adamson <dros@primarydata.com> Signed-off-by:
Anna Schumaker <Anna.Schumaker@Netapp.com>
-
Dan Carpenter authored
nfs_create_request() doesn't return NULL, it returns error pointers. Fixes: 67911c8f ('NFS: Add nfs_commit_file()') Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
- 17 May, 2016 38 commits
-
-
Jeff Layton authored
It can return NULL if layoutgets are blocked currently. Fix it to return -EAGAIN in that case, so we can properly handle it in pnfs_update_layout. Also, clean up and simplify the error handling -- eliminate "status" and just use "lseg". Signed-off-by:
Jeff Layton <jeff.layton@primarydata.com> Signed-off-by:
-
Jeff Layton authored
There are several problems in the way a stateid is selected for a LAYOUTGET operation: We pick a stateid to use in the RPC prepare op, but that makes it difficult to serialize LAYOUTGETs that use the open stateid. That serialization is done in pnfs_update_layout, which occurs well before the rpc_prepare operation. Between those two events, the i_lock is dropped and reacquired. pnfs_update_layout can find that the list has lsegs in it and not do any serialization, but then later pnfs_choose_layoutget_stateid ends up choosing the open stateid. This patch changes the client to select the stateid to use in the LAYOUTGET earlier, when we're searching for a usable layout segment. This way we can do it all while holding the i_lock the first time, and ensure that we serialize any LAYOUTGET call that uses a non-layout stateid. This also means a rework of how LAYOUTGET replies are handled, as we must now get the latest stateid if we want to retransmit in response to a retryable error. Most of those errors boil down to the fact that the layout state has changed in some fashion. Thus, what we really want to do is to re-search for a layout when it fails with a retryable error, so that we can avoid reissuing the RPC at all if possible. While the LAYOUTGET RPC is async, the initiating thread always waits for it to complete, so it's effectively synchronous anyway. Currently, when we need to retry a LAYOUTGET because of an error, we drive that retry via the rpc state machine. This means that once the call has been submitted, it runs until it completes. So, we must move the error handling for this RPC out of the rpc_call_done operation and into the caller. In order to handle errors like NFS4ERR_DELAY properly, we must also pass a pointer to the sliding timeout, which is now moved to the stack in pnfs_update_layout. The complicating errors are -NFS4ERR_RECALLCONFLICT and -NFS4ERR_LAYOUTTRYLATER, as those involve a timeout after which we give up and return NULL back to the caller. So, there is some special handling for those errors to ensure that the layers driving the retries can handle that appropriately. Signed-off-by:
-
Jeff Layton authored
If we get back something like NFS4ERR_OLD_STATEID, that will be translated into -EAGAIN, and the do/while loop in send_layoutget will drive the call again. This is not quite what we want, I think. An error like that is a sign that something has changed. That something could have been a concurrent LAYOUTGET that would give us a usable lseg. Lift the retry logic into pnfs_update_layout instead. That allows us to redo the layout search, and may spare us from having to issue an RPC. Signed-off-by:
-
Jeff Layton authored
Currently, the code will clear the fail bit if we get back a fatal error. I don't think that's correct -- we want to clear that bit if we do not get a fatal error. Fixes: 0bcbf039 (nfs: handle request add failure properly) Signed-off-by:
-
Jeff Layton authored
Signed-off-by:
-
Jeff Layton authored
Setting just the NFS_LAYOUT_RETURN_REQUESTED flag doesn't do anything, unless there are lsegs that are also being marked for return. At the point where that happens this flag is also set, so these set_bit calls don't do anything useful. Signed-off-by:
-
Jeff Layton authored
LAYOUTRETURN is "special" in that servers and clients are expected to work with old stateids. When the client sends a LAYOUTRETURN with an old stateid in it then the server is expected to only tear down layout segments that were present when that seqid was current. Ensure that the client handles its accounting accordingly. Signed-off-by:
-
Jeff Layton authored
When we want to selectively do a LAYOUTRETURN, we need to specify a stateid that represents most recent layout acquisition that is to be returned. When we mark a layout stateid to be returned, we update the return sequence number in the layout header with that value, if it's newer than the existing one. Then, when we go to do a LAYOUTRETURN on layout header put, we overwrite the seqid in the stateid with the saved one, and then zero it out. Signed-off-by:
-
Jeff Layton authored
In later patches, we're going to teach the client to be more selective about how it returns layouts. This means keeping a record of what the stateid's seqid was at the time that the server handed out a layout segment. Signed-off-by:
-
Jeff Layton authored
Otherwise, we'll end up returning layouts that we've just received if the client issues a new LAYOUTGET prior to the LAYOUTRETURN. Signed-off-by:
-
Tom Haynes authored
If we are initializing reads or writes and can not connect to a DS, then check whether or not IO is allowed through the MDS. If it is allowed, reset to the MDS. Else, fail the layout segment and force a retry of a new layout segment. Signed-off-by:
Tom Haynes <loghyr@primarydata.com> Signed-off-by:
-
Tom Haynes authored
Whenever we check to see if we have the needed number of DSes for the action, we may also have to check to see whether IO is allowed to go to the MDS or not. [jlayton: fix merge conflict due to lack of localio patches here] Signed-off-by:
-
Trond Myklebust authored
This patch fixes a problem whereby the pNFS client falls back to doing reads and writes through the metadata server even when the layout flag FF_FLAGS_NO_IO_THRU_MDS is set. Signed-off-by:
Trond Myklebust <trond.myklebust@primarydata.com> Signed-off-by:
-
Trond Myklebust authored
No need to make them a priority any more, or to make them succeed. Signed-off-by:
-
Trond Myklebust authored
When we're using a delegation to represent our open state, we should ensure that we use the stateid that was used to create that delegation. Signed-off-by:
-
Trond Myklebust authored
In order to more easily distinguish what kind of stateid we are dealing with, introduce a type that can be used to label the stateid structure. The label will be useful both for debugging, but also when dealing with operations like SETATTR, READ and WRITE that can take several different types of stateid as arguments. Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Trond Myklebust authored
Signed-off-by:
-
Chuck Lever authored
Clean up. After "xprtrdma: Remove ro_unmap() from all registration modes", there are no longer any sites that take rpcrdma_ia::qplock for read. The one site that takes it for write is always single-threaded. It is safe to remove it. Signed-off-by:
Chuck Lever <chuck.lever@oracle.com> Signed-off-by:
-
Chuck Lever authored
In a cluster failover scenario, it is desirable for the client to attempt to reconnect quickly, as an alternate NFS server is already waiting to take over for the down server. The client can't see that a server IP address has moved to a new server until the existing connection is gone. For fabrics and devices where it is meaningful, set a definite upper bound on the amount of time before it is determined that a connection is no longer valid. This allows the RPC client to detect connection loss in a timely matter, then perform a fresh resolution of the server GUID in case it has changed (cluster failover). Signed-off-by:
Steve Wise <swise@opengridcomputing.com> Signed-off-by:
-
Chuck Lever authored
Clean up: The ro_unmap method is no longer used. Signed-off-by:
Sagi Grimberg <sagi@grimberg.me> Signed-off-by:
-
Chuck Lever authored
There needs to be a safe method of releasing registered memory resources when an RPC terminates. Safe can mean a number of things: + Doesn't have to sleep + Doesn't rely on having a QP in RTS ro_unmap_safe will be that safe method. It can be used in cases where synchronous memory invalidation can deadlock, or needs to have an active QP. The important case is fencing an RPC's memory regions after it is signaled (^C) and before it exits. If this is not done, there is a window where the server can write an RPC reply into memory that the client has released and re-used for some other purpose. Note that this is a full solution for FRWR, but FMR and physical still have some gaps where a particularly bad server can wreak some havoc on the client. These gaps are not made worse by this patch and are expected to be exceptionally rare and timing-based. They are noted in documenting comments. Signed-off-by:
-
Chuck Lever authored
Separate the DMA unmap operation from freeing the MW. In a subsequent patch they will not always be done at the same time, and they are not related operations (except by order; freeing the MW must be the last step during invalidation). Signed-off-by:
-
Chuck Lever authored
In a subsequent patch, the fr_xprt and fr_worker fields will be needed by another memory registration mode. Move them into the generic rpcrdma_mw structure that wraps struct rpcrdma_frmr. Signed-off-by:
-
Chuck Lever authored
Maintain the order of invalidation and DMA unmapping when doing a background MR reset. Signed-off-by:
-
Chuck Lever authored
frwr_op_unmap_sync() is now invoked in a workqueue context, the same as __frwr_queue_recovery(). There's no need to defer MR reset if posting LOCAL_INV MRs fails. This means that even when ib_post_send() fails (which should occur very rarely) the invalidation and DMA unmapping steps are still done in the correct order. Signed-off-by:
-
Chuck Lever authored
Move the the I/O direction field from rpcrdma_mr_seg into the rpcrdma_frmr. This makes it possible to DMA-unmap the frwr long after an RPC has exited and its rpcrdma_mr_seg array has been released and re-used. This might occur if an RPC times out while waiting for a new connection to be established. Signed-off-by:
-
Chuck Lever authored
Clean up: Follow same naming convention as other fields in struct rpcrdma_frwr. Signed-off-by:
-
Chuck Lever authored
Clean up: Replace rpcrdma_flush_cqs() and rpcrdma_clean_cqs() with the new ib_drain_qp() API. Signed-off-by:
Leon Romanovsky <leonro@mellanox.com> Tested-by:
-
Chuck Lever authored
rpcrdma_create_chunks() has been replaced, and can be removed. Signed-off-by:
-
Chuck Lever authored
rpcrdma_marshal_req() makes a simplifying assumption: that NFS operations with large Call messages have small Reply messages, and vice versa. Therefore with RPC-over-RDMA, only one chunk type is ever needed for each Call/Reply pair, because one direction needs chunks, the other direction will always fit inline. In fact, this assumption is asserted in the code: if (rtype != rpcrdma_noch && wtype != rpcrdma_noch) { dprintk("RPC: %s: cannot marshal multiple chunk lists\n", __func__); return -EIO; } But RPCGSS_SEC breaks this assumption. Because krb5i and krb5p perform data transformation on RPC messages before they are transmitted, direct data placement techniques cannot be used, thus RPC messages must be sent via a Long call in both directions. All such calls are sent with a Position Zero Read chunk, and all such replies are handled with a Reply chunk. Thus the client must provide every Call/Reply pair with both a Read list and a Reply chunk. Without any special security in effect, NFSv4 WRITEs may now also use the Read list and provide a Reply chunk. The marshal_req logic was preventing that, meaning an NFSv4 WRITE with a large payload that included a GETATTR result larger than the inline threshold would fail. The code that encodes each chunk list is now completely contained in its own function. There is some code duplication, but the trade-off is that the overall logic should be more clear. Note that all three chunk lists now share the rl_segments array. Some additional per-req accounting is necessary to track this usage. For the same reasons that the above simplifying assumption has held true for so long, I don't expect more array elements are needed at this time. Signed-off-by: -
Chuck Lever authored
Update documenting comments to reflect code changes over the past year. Signed-off-by:
-
Chuck Lever authored
Avoid the latency and interrupt overhead of registering a Write chunk when handling NFS READ requests of a few hundred bytes or less. This change does not interoperate with Linux NFS/RDMA servers that do not have commit 9d11b51c ('svcrdma: Fix send_reply() scatter/gather set-up'). Commit 9d11b51c was introduced in v4.3, and is included in 4.2.y, 4.1.y, and 3.18.y. Oracle bug 22925946 has been filed to request that the above fix be included in the Oracle Linux UEK4 NFS/RDMA server. Red Hat bugzillas 1327280 and 1327554 have been filed to request that RHEL NFS/RDMA server backports include the above fix. Workaround: Replace the "proto=rdma,port=20049" mount options with "proto=tcp" until commit 9d11b51c is applied to your NFS server. Signed-off-by:
-
Chuck Lever authored
When deciding whether to send a Call inline, rpcrdma_marshal_req doesn't take into account header bytes consumed by chunk lists. This results in Call messages on the wire that are sometimes larger than the inline threshold. Likewise, when a Write list or Reply chunk is in play, the server's reply has to emit an RDMA Send that includes a larger-than-minimal RPC-over-RDMA header. The actual size of a Call message cannot be estimated until after the chunk lists have been registered. Thus the size of each RPC-over-RDMA header can be estimated only after chunks are registered; but the decision to register chunks is based on the size of that header. Chicken, meet egg. The best a client can do is estimate header size based on the largest header that might occur, and then ensure that inline content is always smaller than that. Signed-off-by:
-
Chuck Lever authored
Send buffer space is shared between the RPC-over-RDMA header and an RPC message. A large RPC-over-RDMA header means less space is available for the associated RPC message, which then has to be moved via an RDMA Read or Write. As more segments are added to the chunk lists, the header increases in size. Typical modern hardware needs only a few segments to convey the maximum payload size, but some devices and registration modes may need a lot of segments to convey data payload. Sometimes so many are needed that the remaining space in the Send buffer is not enough for the RPC message. Sending such a message usually fails. To ensure a transport can always make forward progress, cap the number of RDMA segments that are allowed in chunk lists. This prevents less-capable devices and memory registrations from consuming a large portion of the Send buffer by reducing the maximum data payload that can be conveyed with such devices. For now I choose an arbitrary maximum of 8 RDMA segments. This allows a maximum size RPC-over-RDMA header to fit nicely in the current 1024 byte inline threshold with over 700 bytes remaining for an inline RPC message. The current maximum data payload of NFS READ or WRITE requests is one megabyte. To convey that payload on a client with 4KB pages, each chunk segment would need to handle 32 or more data pages. This is well within the capabilities of FMR. For physical registration, the maximum payload size on platforms with 4KB pages is reduced to 32KB. For FRWR, a device's maximum page list depth would need to be at least 34 to support the maximum 1MB payload. A device with a smaller maximum page list depth means the maximum data payload is reduced when using that device. Signed-off-by:
-
Chuck Lever authored
Currently the sysctls that allow setting the inline threshold allow any value to be set. Small values only make the transport run slower. The default 1KB setting is as low as is reasonable. And the logic that decides how to divide a Send buffer between RPC-over-RDMA header and RPC message assumes (but does not check) that the lower bound is not crazy (say, 57 bytes). Send and receive buffers share a page with some control information. Values larger than about 3KB can't be supported, currently. Signed-off-by:
-
Chuck Lever authored
RPC-over-RDMA transports have a limit on how large a backward direction (backchannel) RPC message can be. Ensure that the NFSv4.x CREATE_SESSION operation advertises this limit to servers. Signed-off-by:
-
Chuck Lever authored
Commit 176e21ee ("SUNRPC: Support for RPC over AF_LOCAL transports") added a 5-character netid, but did not bump RPCBIND_MAXNETIDLEN from 4 to 5. Fixes: 176e21ee ("SUNRPC: Support for RPC over AF_LOCAL ...") Signed-off-by:
-
