Skip to contentSkip to conversation

erofs-utils: mkfs: Use mtime to seed ctimes

PATCH In review David Anderson via Linux-erofs 2 participants 7 messages Mar 1, 2022 – Mar 11, 2022 Library code

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 erofs-utils: mkfs: Use mtime to seed ctimes

Currently mkfs.erofs picks up whatever the system time happened to be when the input file structure was created. Since there's no (easy) way for userspace to control ctime, there's no way to control the per-file ctime that mkfs.erofs uses.

Switching to mtime allows this tuning, which is important when the timestamp of files is used to detect staleness.

Change-Id I9cab662398bedc43d6d68ae798912f33360814e3
---
 lib/inode.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/inode.c +3 -3
index 461c797..f0a71a8 100644
@@ -806,12 +806,12 @@ static int erofs_fill_inode(struct erofs_inode *inode,
 	inode->i_mode = st->st_mode;
 	inode->i_uid = cfg.c_uid == -1 ? st->st_uid : cfg.c_uid;
 	inode->i_gid = cfg.c_gid == -1 ? st->st_gid : cfg.c_gid;
-	inode->i_ctime = st->st_ctime;
-	inode->i_ctime_nsec = ST_CTIM_NSEC(st);
+	inode->i_ctime = st->st_mtime;
+	inode->i_ctime_nsec = ST_MTIM_NSEC(st);
 
 	switch (cfg.c_timeinherit) {
 	case TIMESTAMP_CLAMPING:
-		if (st->st_ctime < sbi.build_time)
+		if (st->st_mtime < sbi.build_time)
 			break;
 	case TIMESTAMP_FIXED:
 		inode->i_ctime = sbi.build_time;
Signature · 1 line
2.35.1.574.g5d30c73bfb-goog
#1 lore raw

Hi David,

Quoted 7 lines · David Anderson via Linux-erofs

Currently mkfs.erofs picks up whatever the system time happened to be when the input file structure was created. Since there's no (easy) way for userspace to control ctime, there's no way to control the per-file ctime that mkfs.erofs uses.

Switching to mtime allows this tuning, which is important when the timestamp of files is used to detect staleness.

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

Also should we introduce a new compat feature to indicate that new mkfs records mtime instead?

Thanks,
Gao Xiang

Quoted 29 lines
Change-Id I9cab662398bedc43d6d68ae798912f33360814e3
---
 lib/inode.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
lib/inode.c +3 -3
index 461c797..f0a71a8 100644
@@ -806,12 +806,12 @@ static int erofs_fill_inode(struct erofs_inode *inode,
 	inode->i_mode = st->st_mode;
 	inode->i_uid = cfg.c_uid == -1 ? st->st_uid : cfg.c_uid;
 	inode->i_gid = cfg.c_gid == -1 ? st->st_gid : cfg.c_gid;
-	inode->i_ctime = st->st_ctime;
-	inode->i_ctime_nsec = ST_CTIM_NSEC(st);
+	inode->i_ctime = st->st_mtime;
+	inode->i_ctime_nsec = ST_MTIM_NSEC(st);
 
 	switch (cfg.c_timeinherit) {
 	case TIMESTAMP_CLAMPING:
-		if (st->st_ctime < sbi.build_time)
+		if (st->st_mtime < sbi.build_time)
 			break;
 	case TIMESTAMP_FIXED:
 		inode->i_ctime = sbi.build_time;
Signature · 1 line
2.35.1.574.g5d30c73bfb-goog
#2 lore raw
Quoted 7 lines · Gao Xiang

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

That's a good idea, I'll repost with a patch to rename to mtime. Initially I figured it was ok, but the "ctime" name would cause problems if EROFS ever stores both timestamps.

Quoted 2 lines

Also should we introduce a new compat feature to indicate that new mkfs records mtime instead?

Will do. Is there anything in the kernel that would need to care about the new flag? (It looks like no but asking just in case.)

Best,

-David

#3 lore raw
Quoted 6 lines · David Anderson
Quoted 5 lines · Gao Xiang

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Quoted 6 lines
Quoted 2 lines

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

That's a good idea, I'll repost with a patch to rename to mtime. Initially I figured it was ok, but the "ctime" name would cause problems if EROFS ever stores both timestamps.

Yeah, I recommend that we could use mtime from now on, but in case that users are confused, we may need add a compatible feature to indicate that.

Quoted 6 lines
Quoted 3 lines

Also should we introduce a new compat feature to indicate that new mkfs records mtime instead?

Will do. Is there anything in the kernel that would need to care about the new flag? (It looks like no but asking just in case.)

No I think, old kernels are still compatible with a new compatible feature. It's just for marking.

Thanks,
Gao Xiang

Quoted 4 lines

Best,

-David

#4 lore raw

Hi David,

Quoted 10 lines · Gao Xiang
Quoted 9 lines · David Anderson
Quoted 7 lines · Gao Xiang

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

Quoted 6 lines
Quoted 3 lines

That's a good idea, I'll repost with a patch to rename to mtime. Initially I figured it was ok, but the "ctime" name would cause problems if EROFS ever stores both timestamps.

Yeah, I recommend that we could use mtime from now on, but in case that users are confused, we may need add a compatible feature to indicate that.

Would you mind sending a kernel patch to rename them recently if you have some free slot?

We're in 5.17-rc7 cycle.. I think we have to prepare patches for 5.18 now so we can use new timestamp behavior from 5.18..

Thanks,
Gao Xiang

#5 lore raw

Sorry for not replying sooner, things got very busy - I am working on this now and will hopefully have something tested + uploaded by EOD.

Best,

-David

Quoted 30 lines · Gao Xiang

Hi David,

Quoted 10 lines · Gao Xiang
Quoted 9 lines · David Anderson
Quoted 7 lines · Gao Xiang

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

Quoted 6 lines
Quoted 3 lines

That's a good idea, I'll repost with a patch to rename to mtime. Initially I figured it was ok, but the "ctime" name would cause problems if EROFS ever stores both timestamps.

Yeah, I recommend that we could use mtime from now on, but in case that users are confused, we may need add a compatible feature to indicate that.

Would you mind sending a kernel patch to rename them recently if you have some free slot?

We're in 5.17-rc7 cycle.. I think we have to prepare patches for 5.18 now so we can use new timestamp behavior from 5.18..

Thanks,
Gao Xiang

#6 lore raw
Quoted 2 lines · David Anderson

Sorry for not replying sooner, things got very busy - I am working on this now and will hopefully have something tested + uploaded by EOD.

Yeah, ok. I think it's not urgent to us on the kernel side anyway since it's always compatible with old kernels and mtime == ctime == atime as always, we could also clarify later if it doesn't catch the next merge window.. We could delay it to the next release..

Thanks,
Gao Xiang

Quoted 36 lines

Best,

-David

Quoted 30 lines · Gao Xiang

Hi David,

Quoted 10 lines · Gao Xiang
Quoted 9 lines · David Anderson
Quoted 7 lines · Gao Xiang

Yeah, I agree I should think more when I planned to store ctime' at the first time [my original thought was to keep metadata time (including uid, gid, etc..), so I selected ctime' instead of `mtime'].

Should we change what's described in 'Documentation/filesystems/erofs.rst' and even rename i_ctime to i_mtime?

Quoted 6 lines
Quoted 3 lines

That's a good idea, I'll repost with a patch to rename to mtime. Initially I figured it was ok, but the "ctime" name would cause problems if EROFS ever stores both timestamps.

Yeah, I recommend that we could use mtime from now on, but in case that users are confused, we may need add a compatible feature to indicate that.

Would you mind sending a kernel patch to rename them recently if you have some free slot?

We're in 5.17-rc7 cycle.. I think we have to prepare patches for 5.18 now so we can use new timestamp behavior from 5.18..

Thanks,
Gao Xiang