]>
Commit | Line | Data |
---|---|---|
6e4ffbcf SL |
1 | From 7b0e7b5d6d603aee5b85dfacdb3c3b24903e3810 Mon Sep 17 00:00:00 2001 |
2 | From: Kirill Smelkov <kirr@nexedi.com> | |
3 | Date: Wed, 27 Mar 2019 10:15:19 +0000 | |
4 | Subject: fuse: retrieve: cap requested size to negotiated max_write | |
5 | ||
6 | [ Upstream commit 7640682e67b33cab8628729afec8ca92b851394f ] | |
7 | ||
8 | FUSE filesystem server and kernel client negotiate during initialization | |
9 | phase, what should be the maximum write size the client will ever issue. | |
10 | Correspondingly the filesystem server then queues sys_read calls to read | |
11 | requests with buffer capacity large enough to carry request header + that | |
12 | max_write bytes. A filesystem server is free to set its max_write in | |
13 | anywhere in the range between [1*page, fc->max_pages*page]. In particular | |
14 | go-fuse[2] sets max_write by default as 64K, wheres default fc->max_pages | |
15 | corresponds to 128K. Libfuse also allows users to configure max_write, but | |
16 | by default presets it to possible maximum. | |
17 | ||
18 | If max_write is < fc->max_pages*page, and in NOTIFY_RETRIEVE handler we | |
19 | allow to retrieve more than max_write bytes, corresponding prepared | |
20 | NOTIFY_REPLY will be thrown away by fuse_dev_do_read, because the | |
21 | filesystem server, in full correspondence with server/client contract, will | |
22 | be only queuing sys_read with ~max_write buffer capacity, and | |
23 | fuse_dev_do_read throws away requests that cannot fit into server request | |
24 | buffer. In turn the filesystem server could get stuck waiting indefinitely | |
25 | for NOTIFY_REPLY since NOTIFY_RETRIEVE handler returned OK which is | |
26 | understood by clients as that NOTIFY_REPLY was queued and will be sent | |
27 | back. | |
28 | ||
29 | Cap requested size to negotiate max_write to avoid the problem. This | |
30 | aligns with the way NOTIFY_RETRIEVE handler works, which already | |
31 | unconditionally caps requested retrieve size to fuse_conn->max_pages. This | |
32 | way it should not hurt NOTIFY_RETRIEVE semantic if we return less data than | |
33 | was originally requested. | |
34 | ||
35 | Please see [1] for context where the problem of stuck filesystem was hit | |
36 | for real, how the situation was traced and for more involving patch that | |
37 | did not make it into the tree. | |
38 | ||
39 | [1] https://marc.info/?l=linux-fsdevel&m=155057023600853&w=2 | |
40 | [2] https://github.com/hanwen/go-fuse | |
41 | ||
42 | Signed-off-by: Kirill Smelkov <kirr@nexedi.com> | |
43 | Cc: Han-Wen Nienhuys <hanwen@google.com> | |
44 | Cc: Jakob Unterwurzacher <jakobunt@gmail.com> | |
45 | Signed-off-by: Miklos Szeredi <mszeredi@redhat.com> | |
46 | Signed-off-by: Sasha Levin <sashal@kernel.org> | |
47 | --- | |
48 | fs/fuse/dev.c | 2 +- | |
49 | 1 file changed, 1 insertion(+), 1 deletion(-) | |
50 | ||
51 | diff --git a/fs/fuse/dev.c b/fs/fuse/dev.c | |
52 | index 4c78b90bc83e..6dbc915520da 100644 | |
53 | --- a/fs/fuse/dev.c | |
54 | +++ b/fs/fuse/dev.c | |
55 | @@ -1688,7 +1688,7 @@ static int fuse_retrieve(struct fuse_conn *fc, struct inode *inode, | |
56 | offset = outarg->offset & ~PAGE_MASK; | |
57 | file_size = i_size_read(inode); | |
58 | ||
59 | - num = outarg->size; | |
60 | + num = min(outarg->size, fc->max_write); | |
61 | if (outarg->offset > file_size) | |
62 | num = 0; | |
63 | else if (outarg->offset + num > file_size) | |
64 | -- | |
65 | 2.20.1 | |
66 |