kernel.git/fs/nfs/direct.c, branch linux-6.7.y

nfs: fix UAF in direct writes

2024-04-03T13:11:21Z

[ Upstream commit 17f46b803d4f23c66cacce81db35fef3adb8f2af ] In production we have been hitting the following warning consistently ------------[ cut here ]------------ refcount_t: underflow; use-after-free. WARNING: CPU: 17 PID: 1800359 at lib/refcount.c:28 refcount_warn_saturate+0x9c/0xe0 Workqueue: nfsiod nfs_direct_write_schedule_work [nfs] RIP: 0010:refcount_warn_saturate+0x9c/0xe0 PKRU: 55555554 Call Trace: ? __warn+0x9f/0x130 ? refcount_warn_saturate+0x9c/0xe0 ? report_bug+0xcc/0x150 ? handle_bug+0x3d/0x70 ? exc_invalid_op+0x16/0x40 ? asm_exc_invalid_op+0x16/0x20 ? refcount_warn_saturate+0x9c/0xe0 nfs_direct_write_schedule_work+0x237/0x250 [nfs] process_one_work+0x12f/0x4a0 worker_thread+0x14e/0x3b0 ? ZSTD_getCParams_internal+0x220/0x220 kthread+0xdc/0x120 ? __btf_name_valid+0xa0/0xa0 ret_from_fork+0x1f/0x30 This is because we're completing the nfs_direct_request twice in a row. The source of this is when we have our commit requests to submit, we process them and send them off, and then in the completion path for the commit requests we have if (nfs_commit_end(cinfo.mds)) nfs_direct_write_complete(dreq); However since we're submitting asynchronous requests we sometimes have one that completes before we submit the next one, so we end up calling complete on the nfs_direct_request twice. The only other place we use nfs_generic_commit_list() is in __nfs_commit_inode, which wraps this call in a nfs_commit_begin(); nfs_commit_end(); Which is a common pattern for this style of completion handling, one that is also repeated in the direct code with get_dreq()/put_dreq() calls around where we process events as well as in the completion paths. Fix this by using the same pattern for the commit requests. Before with my 200 node rocksdb stress running this warning would pop every 10ish minutes. With my patch the stress test has been running for several hours without popping. Signed-off-by: Josef Bacik Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust Signed-off-by: Sasha Levin

pNFS: Fix the pnfs block driver's calculation of layoutget size

2024-01-25T23:44:44Z

[ Upstream commit 8a6291bf3b0eae1bf26621e6419a91682f2d6227 ] Instead of relying on the value of the 'bytes_left' field, we should calculate the layout size based on the offset of the request that is being written out. Reported-by: Benjamin Coddington Signed-off-by: Trond Myklebust Fixes: 954998b60caa ("NFS: Fix error handling for O_DIRECT write scheduling") Reviewed-by: Benjamin Coddington Tested-by: Benjamin Coddington Reviewed-by: Christoph Hellwig Signed-off-by: Anna Schumaker Signed-off-by: Sasha Levin

NFS: More fixes for nfs_direct_write_reschedule_io()

2023-09-13T15:51:11Z

Ensure that all requests are put back onto the commit list so that they can be rescheduled. Fixes: 4daaeba93822 ("NFS: Fix nfs_direct_write_reschedule_io()") Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: Use the correct commit info in nfs_join_page_group()

2023-09-13T15:51:11Z

Ensure that nfs_clear_request_commit() updates the correct counters when it removes them from the commit list. Fixes: ed5d588fe47f ("NFS: Try to join page groups before an O_DIRECT retransmission") Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: More O_DIRECT accounting fixes for error paths

2023-09-13T15:51:10Z

If we hit a fatal error when retransmitting, we do need to record the removal of the request from the count of written bytes. Fixes: 031d73ed768a ("NFS: Fix O_DIRECT accounting of number of bytes read/written") Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: Fix O_DIRECT locking issues

2023-09-13T15:50:57Z

The dreq fields are protected by the dreq->lock. Fixes: 954998b60caa ("NFS: Fix error handling for O_DIRECT write scheduling") Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: Fix error handling for O_DIRECT write scheduling

2023-09-11T15:29:38Z

If we fail to schedule a request for transmission, there are 2 possibilities: 1) Either we hit a fatal error, and we just want to drop the remaining requests on the floor. 2) We were asked to try again, in which case we should allow the outstanding RPC calls to complete, so that we can recoalesce requests and try again. Fixes: d600ad1f2bdb ("NFS41: pop some layoutget errors to application") Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: Fix a potential data corruption

2023-08-24T17:24:15Z

We must ensure that the subrequests are joined back into the head before we can retransmit a request. If the head was not on the commit lists, because the server wrote it synchronously, we still need to add it back to the retransmission list. Add a call that mirrors the effect of nfs_cancel_remove_inode() for O_DIRECT. Fixes: ed5d588fe47f ("NFS: Try to join page groups before an O_DIRECT retransmission") Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker

NFS: Fix a use after free in nfs_direct_join_group()

2023-08-17T13:30:38Z

Be more careful when tearing down the subrequests of an O_DIRECT write as part of a retransmission. Reported-by: Chris Mason Fixes: ed5d588fe47f ("NFS: Try to join page groups before an O_DIRECT retransmission") Cc: stable@vger.kernel.org Signed-off-by: Trond Myklebust

NFS: Clean up O_DIRECT request allocation

2023-02-14T19:22:33Z

Rather than adjusting the index+offset after the call to nfs_create_request(), add a function nfs_page_create_from_page() that takes an offset. Signed-off-by: Trond Myklebust Signed-off-by: Anna Schumaker