Cannot set mtime on a symbolic link #412
Milestone
Comments
|
Thanks for the bug an potential fix. What you propose may in fact be exactly the right fix but I need to give it some more thought and do some testing first. |
mmaybee
pushed a commit
to mmaybee/openzfs
that referenced
this issue
Aug 27, 2021
Ingest written blocks to the zettacache. Note that this is an on-disk format change due to some code cleanup. Several synergistic improvements as well: don't allow overwrites for sync-to-convergence; clean up now-unused code in Agent limit disk i/o queue depth. This prevents starvation of the 512 Tokio blocking threads. Otherwise we could use up all of these threads with the write_raw() blocking task that does the pwrite() syscall. separate disk i/o queue depth limits for reads vs writes. This prevents starvation of reads (for cache hits) when the writes (for inserts) are at the max depth limit buffering of cache insertions. When the buffer is full, the insertion is dropped (ignored). When the cache is slower than the network (which it is when ingesting writes, but not when inserting from cache misses), we don't want to wait for the cache, slowing down overall write performance; instead we buffer a small amount (about a second's worth) and once that limit is reached we don't insert the next block to the cache. Victim blocks are selected randomly (based on arrival order). In the future we might want to have the kernel tell us which writes contain data that's more likely to be read in the future (e.g. ZFS metadata). block allocator: continue looking from last allocation location. This vastly reduces CPU usage, once evictions have happened and the free space is fragmented. TARGET_CACHE_SIZE_PCT default to 80 more destructuring code cleanup in range_tree.rs zettacache: remove unnecessary chunk_summary fix DOSE-602 cross-PendingFreesLog object consolidation breaks pool
sdimitro
pushed a commit
to sdimitro/zfs
that referenced
this issue
May 23, 2022
Signed-off-by: Paul Dagnelie <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Hi,
If you try to do a utimensat(AT_SYMLINK_NOFOLLOW) on a symlink, the mtime of the symlink is set to the current time and not to the requested time.
To work around this I have made the change below - however I don't know if there are any nasty side-effects from this change.....
diff --git a/module/zfs/zpl_inode.c b/module/zfs/zpl_inode.c
index dbfe61a..17acf37 100644
--- a/module/zfs/zpl_inode.c
+++ b/module/zfs/zpl_inode.c
@@ -354,6 +354,8 @@ const struct inode_operations zpl_symlink_inode_operations = {
.readlink = generic_readlink,
.follow_link = zpl_follow_link,
.put_link = zpl_put_link,
};
const struct inode_operations zpl_special_inode_operations = {
regards,
James.
The text was updated successfully, but these errors were encountered: