xref: /openbmc/linux/Documentation/filesystems/fuse-io.rst (revision 4b4193256c8d3bc3a5397b5cd9494c2ad386317d)
1*ba302d2aSMauro Carvalho Chehab.. SPDX-License-Identifier: GPL-2.0
2*ba302d2aSMauro Carvalho Chehab
3*ba302d2aSMauro Carvalho Chehab==============
4*ba302d2aSMauro Carvalho ChehabFuse I/O Modes
5*ba302d2aSMauro Carvalho Chehab==============
6*ba302d2aSMauro Carvalho Chehab
7*ba302d2aSMauro Carvalho ChehabFuse supports the following I/O modes:
8*ba302d2aSMauro Carvalho Chehab
9*ba302d2aSMauro Carvalho Chehab- direct-io
10*ba302d2aSMauro Carvalho Chehab- cached
11*ba302d2aSMauro Carvalho Chehab  + write-through
12*ba302d2aSMauro Carvalho Chehab  + writeback-cache
13*ba302d2aSMauro Carvalho Chehab
14*ba302d2aSMauro Carvalho ChehabThe direct-io mode can be selected with the FOPEN_DIRECT_IO flag in the
15*ba302d2aSMauro Carvalho ChehabFUSE_OPEN reply.
16*ba302d2aSMauro Carvalho Chehab
17*ba302d2aSMauro Carvalho ChehabIn direct-io mode the page cache is completely bypassed for reads and writes.
18*ba302d2aSMauro Carvalho ChehabNo read-ahead takes place. Shared mmap is disabled.
19*ba302d2aSMauro Carvalho Chehab
20*ba302d2aSMauro Carvalho ChehabIn cached mode reads may be satisfied from the page cache, and data may be
21*ba302d2aSMauro Carvalho Chehabread-ahead by the kernel to fill the cache.  The cache is always kept consistent
22*ba302d2aSMauro Carvalho Chehabafter any writes to the file.  All mmap modes are supported.
23*ba302d2aSMauro Carvalho Chehab
24*ba302d2aSMauro Carvalho ChehabThe cached mode has two sub modes controlling how writes are handled.  The
25*ba302d2aSMauro Carvalho Chehabwrite-through mode is the default and is supported on all kernels.  The
26*ba302d2aSMauro Carvalho Chehabwriteback-cache mode may be selected by the FUSE_WRITEBACK_CACHE flag in the
27*ba302d2aSMauro Carvalho ChehabFUSE_INIT reply.
28*ba302d2aSMauro Carvalho Chehab
29*ba302d2aSMauro Carvalho ChehabIn write-through mode each write is immediately sent to userspace as one or more
30*ba302d2aSMauro Carvalho ChehabWRITE requests, as well as updating any cached pages (and caching previously
31*ba302d2aSMauro Carvalho Chehabuncached, but fully written pages).  No READ requests are ever sent for writes,
32*ba302d2aSMauro Carvalho Chehabso when an uncached page is partially written, the page is discarded.
33*ba302d2aSMauro Carvalho Chehab
34*ba302d2aSMauro Carvalho ChehabIn writeback-cache mode (enabled by the FUSE_WRITEBACK_CACHE flag) writes go to
35*ba302d2aSMauro Carvalho Chehabthe cache only, which means that the write(2) syscall can often complete very
36*ba302d2aSMauro Carvalho Chehabfast.  Dirty pages are written back implicitly (background writeback or page
37*ba302d2aSMauro Carvalho Chehabreclaim on memory pressure) or explicitly (invoked by close(2), fsync(2) and
38*ba302d2aSMauro Carvalho Chehabwhen the last ref to the file is being released on munmap(2)).  This mode
39*ba302d2aSMauro Carvalho Chehabassumes that all changes to the filesystem go through the FUSE kernel module
40*ba302d2aSMauro Carvalho Chehab(size and atime/ctime/mtime attributes are kept up-to-date by the kernel), so
41*ba302d2aSMauro Carvalho Chehabit's generally not suitable for network filesystems.  If a partial page is
42*ba302d2aSMauro Carvalho Chehabwritten, then the page needs to be first read from userspace.  This means, that
43*ba302d2aSMauro Carvalho Chehabeven for files opened for O_WRONLY it is possible that READ requests will be
44*ba302d2aSMauro Carvalho Chehabgenerated by the kernel.
45