fuse: invalidate inode attr in writeback cache mode
Subsystems (2)
RFC PATCH 1/1 fuse: invalidate inode attr in writeback cache mode
Under writeback mode, inode->i_blocks is not updated, making utils like du read st.blocks as 0.
For example, when using virtiofs (cache=always & nondax mode) with writeback_cache enabled, writing a new file and check its disk usage with du, du reports 0 usage.
# uname -r 5.6.0-rc6+ # mount -t virtiofs virtiofs /mnt/virtiofs # rm -f /mnt/virtiofs/testfile
# create new file and do extend write # xfs_io -fc "pwrite 0 4k" /mnt/virtiofs/testfile wrote 4096/4096 bytes at offset 0 4 KiB, 1 ops; 0.0001 sec (28.103 MiB/sec and 7194.2446 ops/sec) # stat -c %s,%b /mnt/virtiofs/testfile 4096,0 <==== i_size is correct, but block usage is 0
Fix it by invalidating attr in fuse_write_end(), like other write paths.
--- fs/fuse/file.c | 1 + 1 file changed, 1 insertion(+)
fs/fuse/file.c +1 -0
@@ -2228,6 +2228,7 @@ static int fuse_write_end(struct file *file, struct address_space *mapping, }fuse_write_update_size (inode, pos + copied); +fuse_invalidate_attr (inode);set_page_dirty (page); unlock:
Signature · 1 line
1.8.3.1
Miklos Szeredi
maintainerQuoted 7 lines · Eryu Guan
Under writeback mode,
inode->i_blocksis not updated, making utils like du read st.blocks as 0.For example, when using virtiofs (cache=always & nondax mode) with
writeback_cacheenabled, writing a new file and check its disk usage with du, du reports 0 usage.
Hmm... invalidating the attribute might also yield the wrong result as the server may not have received the WRITE request that modifies the underlying file.
Invalidating attributes at the end of fuse_flush() definitely makes sense, though.
If we wanted 100% correct behavior, we'd need to flush WRITE requests before each GETATTR request. That might be a performance bottleneck, though.
So first I'd just try doing the invalidation from fuse_flush().
Thanks,
Miklos
Quoted 5 lines · Miklos Szeredi
Quoted 4 lines · Eryu Guan
Under writeback mode,
inode->i_blocksis not updated, making utils like du read st.blocks as 0.
Quoted 7 lines
Quoted 3 lines
For example, when using virtiofs (cache=always & nondax mode) with
writeback_cacheenabled, writing a new file and check its disk usage with du, du reports 0 usage.Hmm... invalidating the attribute might also yield the wrong result as the server may not have received the WRITE request that modifies the underlying file.
That's true, I open-write a file then sleep without closing it, and check file's st_blocks in another terminal, the usage is 0, and remains 0 even after the file is closed, because the wrong attr is cached.
Quoted 3 lines
Invalidating attributes at the end of
fuse_flush()definitely makes sense, though.
du also reports 0 if I didn't close the file, but I got correct st_blocks when it's closed.
Quoted 6 lines
If we wanted 100% correct behavior, we'd need to flush WRITE requests before each GETATTR request. That might be a performance bottleneck, though.
So first I'd just try doing the invalidation from
fuse_flush().
Sure, will send v2 soon.
Quoted 3 lines
Thanks,
Miklos
Thanks for the review!
Eryu
PATCH v2 1/1 fuse: invalidate inode attr in writeback cache mode
In reply to Eryu GuanUnder writeback mode, inode->i_blocks is not updated, making utils du read st.blocks as 0.
For example, when using virtiofs (cache=always & nondax mode) with writeback_cache enabled, writing a new file and check its disk usage with du, du reports 0 usage.
# uname -r 5.6.0-rc6+ # mount -t virtiofs virtiofs /mnt/virtiofs # rm -f /mnt/virtiofs/testfile
# create new file and do extend write # xfs_io -fc "pwrite 0 4k" /mnt/virtiofs/testfile wrote 4096/4096 bytes at offset 0 4 KiB, 1 ops; 0.0001 sec (28.103 MiB/sec and 7194.2446 ops/sec) # du -k /mnt/virtiofs/testfile 0 <==== disk usage is 0 # stat -c %s,%b /mnt/virtiofs/testfile 4096,0 <==== i_size is correct, but st_blocks is 0
Fix it by invalidating attr in fuse_flush(), so we get up-to-date attr from server on next getattr.
---
v2:
- do fuse_invalidate_attr() in fuse_flush() as Miklos suggested
fs/fuse/file.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
fs/fuse/file.c +10 -1
@@ -445,8 +445,9 @@ static int fuse_flush(struct file *file, fl_owner_t id)if (is_bad_inode (inode))return -EIO ; + err =0 ;if (fc->no_flush) -return 0 ; + goto inval_attr_out; err =write_inode_now (inode,1 );if (err)
@@ -475,6 +476,14 @@ static int fuse_flush(struct file *file, fl_owner_t id) fc->no_flush =1 ; err =0 ; } + +inval_attr_out: +/* +* In memory i_blocks is not maintained by fuse, if writeback cache is +* enabled, i_blocks from cached attr may not be accurate. +*/ +if (!err && fc->writeback_cache) +fuse_invalidate_attr (inode);return err; }
Signature · 1 line
2.26.1.107.gefe3874
Quoted 22 lines · Eryu Guan
Under writeback mode,
inode->i_blocksis not updated, making utils du read st.blocks as 0.For example, when using virtiofs (cache=always & nondax mode) with
writeback_cacheenabled, writing a new file and check its disk usage with du, du reports 0 usage.# uname -r 5.6.0-rc6+ # mount -t virtiofs virtiofs /mnt/virtiofs # rm -f /mnt/virtiofs/testfile# create new file and do extend write # xfs_io -fc "pwrite 0 4k" /mnt/virtiofs/testfile wrote 4096/4096 bytes at offset 0 4 KiB, 1 ops; 0.0001 sec (28.103 MiB/sec and 7194.2446 ops/sec) # du -k /mnt/virtiofs/testfile 0 <==== disk usage is 0 # stat -c %s,%b /mnt/virtiofs/testfile 4096,0 <==== i_size is correct, but st_blocks is 0
Quoted 2 lines
Fix it by invalidating attr in
fuse_flush(), so we get up-to-date attr from server on next getattr.
Thanks, applied.
I started thinking: why is fuse_flush() only writing out dirty pages
if fc->no_flush is false? It just doesn't make sense... But that's
an independent bug, and I'll do a separate patch for that.
Thanks,
Miklos