Skip to contentSkip to conversation

nvme-cli: close fd before return

PATCH New Chaitanya Kulkarni 3 participants 8 messages May 19, 2019 – May 20, 2019 NVMe

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nvme-cli: close fd before return

Instead of returning on the erro use existing close_fd label before we return in create_ns() and format().

Signed-off-by Chaitanya Kulkarni <chaitanya.kulkarni at wdc.com>
---
 nvme.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
nvme.c +3 -2
index e0f7552..d6e5b21 100644
@@ -1216,7 +1216,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
 				fprintf(stderr, "identify failed\n");
 				show_nvme_status(err);
 			}
-			return err;
+			goto close_fd;
 		}
 		for (i = 0; i < 16; ++i) {
 			if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0) {
@@ -1245,6 +1245,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
 	else
 		perror("create namespace");
 
+ close_fd:
 	close(fd);
 
 	return err;
@@ -3198,7 +3199,7 @@ static int format(int argc, char **argv, struct command *cmd, struct plugin *plu
 				fprintf(stderr, "identify failed\n");
 				show_nvme_status(err);
 			}
-			return err;
+			goto close_fd;
 		}
 		prev_lbaf = ns.flbas & 0xf;
 
Signature · 1 line
2.17.0
#1 lore raw

PATCH 1/1 nvme-cli: close fd before return

This kind of patches are already in Github PR:
https://github.com/linux-nvme/nvme-cli/pull/490

Quoted 10 lines
nvme.c +1 -1
@@ -1216,7 +1216,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
 				fprintf(stderr, "identify failed\n");
 				show_nvme_status(err);
 			}
-			return err;
+			goto close_fd;
 		}
 		for (i = 0; i < 16; ++i) {
 			if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0) {

nvme.c static int create_ns(int argc, char **argv, struct command *cmd, struct plugin * · line 1251

Quoted 7 lines
nvme.c +1 -0
@@ -1245,6 +1245,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
 	else
 		perror("create namespace");
 
+ close_fd:
 	close(fd);
 
 	return err;

If this patch wants to free the leaked file descriptor in case of errors, it needs to cover more parts than it shows above. You can see the patch for the create_ns() from:
https://github.com/linux-nvme/nvme-cli/pull/490/commits/b0c8a309266c2daf6ebadf9ab14884c6954765a1

nvme.c static int format(int argc, char **argv, struct command *cmd, struct plugin *plu · line 3204

Quoted 8 lines
nvme.c +1 -1
@@ -3198,7 +3199,7 @@ static int format(int argc, char **argv, struct command *cmd, struct plugin *plu
 				fprintf(stderr, "identify failed\n");
 				show_nvme_status(err);
 			}
-			return err;
+			goto close_fd;
 		}
 		prev_lbaf = ns.flbas & 0xf;

It also needs to be coverted with more parts to close the leaked file descriptor.
You can see the patch for the format() from:
https://github.com/linux-nvme/nvme-cli/pull/490/commits/e3c487c6d9c145ba0b90d4ef227510b4faa33e98

#2 lore raw

PATCH 1/1 nvme-cli: close fd before return

This only covers the code path for nvme commands failure and not the errno codes ans thi ps mainly to make all the nvme error and return codes consistent.

Quoted 40 lines · Minwoo Im

This kind of patches are already in Github PR:

   https://github.com/linux-nvme/nvme-cli/pull/490
Quoted 10 lines
nvme.c +1 -1
@@ -1216,7 +1216,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
  				fprintf(stderr, "identify failed\n");
  				show_nvme_status(err);
  			}
-			return err;
+			goto close_fd;
  		}
  		for (i = 0; i < 16; ++i) {
  			if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0) {

nvme.c static int create_ns(int argc, char **argv, struct command *cmd, struct plugin * · line 1251

Quoted 7 lines
nvme.c +1 -0
@@ -1245,6 +1245,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
  	else
  		perror("create namespace");
 
+ close_fd:
  	close(fd);
 
  	return err;

If this patch wants to free the leaked file descriptor in case of errors, it needs to cover more parts than it shows above. You can see the patch for the create_ns() from:

   https://github.com/linux-nvme/nvme-cli/pull/490/commits/b0c8a309266c2daf6ebadf9ab14884c6954765a1

nvme.c static int format(int argc, char **argv, struct command *cmd, struct plugin *plu · line 3204

Quoted 8 lines
nvme.c +1 -1
@@ -3198,7 +3199,7 @@ static int format(int argc, char **argv, struct command *cmd, struct plugin *plu
  				fprintf(stderr, "identify failed\n");
  				show_nvme_status(err);
  			}
-			return err;
+			goto close_fd;
  		}
  		prev_lbaf = ns.flbas & 0xf;

It also needs to be coverted with more parts to close the leaked file descriptor.
You can see the patch for the format() from:

   https://github.com/linux-nvme/nvme-cli/pull/490/commits/e3c487c6d9c145ba0b90d4ef227510b4faa33e98
#3 lore raw

PATCH 1/1 nvme-cli: close fd before return

Quoted 3 lines

This only covers the code path for nvme commands failure and not the errno codes ans thi ps mainly to make all the nvme error and return codes consistent.

I think I don't agree with that. If you want to fix the leaked fd before returning directly, then it should make it go to close_fd for the EINVAL cases without adding -(negative) on it.

I think the PR can be merged regardless to errno update now, though.

#4 lore raw

PATCH 1/1 nvme-cli: close fd before return

Quoted 10 lines · Minwoo Im
Quoted 3 lines

This only covers the code path for nvme commands failure and not the errno codes ans thi ps mainly to make all the nvme error and return codes consistent.

I think I don't agree with that. If you want to fix the leaked fd before returning directly, then it should make it go to close_fd for the EINVAL cases without adding -(negative) on it.

I think the PR can be merged regardless to errno update now, though.

Before we add your patch-series we need to make all the returns consistent which is not present in the format() and create_ns() for nvme error status. Ofcourse eventually all the fd leaks needs to be fixed. This patch makes it sure the leaks for the only nvme error code are consistent and on the top of that your series fits well.

Without this we need to add two return calls in the create_ns ()
http://lists.infradead.org/pipermail/linux-nvme/2019-May/024307.html :-

static int attach_ns(int argc, char **argv, struct command *cmd, struct plugin *plugin)
@@ -1216,7 +1210,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
fprintf(stderr, "identify failed\n"); show_nvme_status(err);
}
- return err;
+ return nvme_status_to_errno(err, false);
}
for (i = 0; i < 16; ++i) {
if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0) {
@@ -1246,8 +1240,7 @@ static int create_ns(int argc, char **argv, struct command *cmd, struct plugin *
perror("create namespace");

close(fd);
-
- return err;
+ return nvme_status_to_errno(err, false);
}

It should be consistent with all the over functions with one return.

This what I'm trying to avoid (by ignoring other leaks for now), can you please explain why it is not a good idea to avoid this ?

#5 lore raw

PATCH 1/1 nvme-cli: close fd before return

Quoted 32 lines

Before we add your patch-series we need to make all the returns consistent which is not present in the format() and create_ns() for nvme error status. Ofcourse eventually all the fd leaks needs to be fixed. This patch makes it sure the leaks for the only nvme error code are consistent and on the top of that your series fits well.

Without this we need to add two return calls in the create_ns ()
http://lists.infradead.org/pipermail/linux-nvme/2019-May/024307.html :-

static int attach_ns(int argc, char **argv, struct command *cmd, struct plugin *plugin)

(quoted diff) +0 -0
@@ -1216,7 +1210,7 @@ static int create_ns(int argc, char **argv, struct

command *cmd, struct plugin *
fprintf(stderr, "identify failed\n"); show_nvme_status(err);
}

nvme.c +1 -1
-			return err;
+			return nvme_status_to_errno(err, false);
  		}
  		for (i = 0; i < 16; ++i) {
  			if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0)

{

(quoted diff) +0 -0
@@ -1246,8 +1240,7 @@ static int create_ns(int argc, char **argv, struct

command *cmd, struct plugin *
perror("create namespace");

  	close(fd);
nvme.c +1 -2
-
-	return err;
+	return nvme_status_to_errno(err, false);
  }
Quoted 4 lines

It should be consistent with all the over functions with one return.

This what I'm trying to avoid (by ignoring other leaks for now), can you please explain why it is not a good idea to avoid this ?

I never thought it was not a good idea. I just told you that it's just a duplicated patch with on-going PR in the Github.

If we are returning the nvme_status_to_errno() in that other patch that I have posted twice, is it that critical factor which blocks the patchsets to be merged?

I don't know what exactly the issues are between "first merge patch serias and code-cleanup there", and "first merge this fix and rebase the patch series". When I was preparing the patch series about errno, I was also hate that codes you have posted above, but I need to make it focused first to make people review this series first even there are two times calling like above.

If you really want to make it consistent, then why don't we just title it like "Do not return in case of errors in the middle"

If you really want to close fd to fix the leak issue, then I think it needs to cover more than that.

#6 lore raw

PATCH 1/1 nvme-cli: close fd before return

Quoted 32 lines · Minwoo Im
Quoted 32 lines

Before we add your patch-series we need to make all the returns consistent which is not present in the format() and create_ns() for nvme error status. Ofcourse eventually all the fd leaks needs to be fixed. This patch makes it sure the leaks for the only nvme error code are consistent and on the top of that your series fits well.

Without this we need to add two return calls in the create_ns ()
http://lists.infradead.org/pipermail/linux-nvme/2019-May/024307.html :-

static int attach_ns(int argc, char **argv, struct command *cmd, struct plugin *plugin)

(quoted diff) +0 -0
@@ -1216,7 +1210,7 @@ static int create_ns(int argc, char **argv, struct

command *cmd, struct plugin *
fprintf(stderr, "identify failed\n"); show_nvme_status(err);
}

nvme.c +1 -1
-			return err;
+			return nvme_status_to_errno(err, false);
  		}
  		for (i = 0; i < 16; ++i) {
  			if ((1 << ns.lbaf[i].ds) == cfg.bs && ns.lbaf[i].ms == 0)

{

(quoted diff) +0 -0
@@ -1246,8 +1240,7 @@ static int create_ns(int argc, char **argv, struct

command *cmd, struct plugin *
perror("create namespace");

  	close(fd);
nvme.c +1 -2
-
-	return err;
+	return nvme_status_to_errno(err, false);
  }
Quoted 6 lines
Quoted 4 lines

It should be consistent with all the over functions with one return.

This what I'm trying to avoid (by ignoring other leaks for now), can you please explain why it is not a good idea to avoid this ?

I never thought it was not a good idea. I just told you that it's just a duplicated patch with on-going PR in the Github.

I've not said anything against PR. I was just explaining the idea behind sending

this patch. We can leave it to PR now.

But this is a good example of having things not be posted on the mailing list can

a create confusion, will be monitoring PR going forward.

Quoted 4 lines

If we are returning the nvme_status_to_errno() in that other patch that I have posted twice, is it that critical factor which blocks the patchsets to be merged?

No it is not but it just makes your series smooth.

Quoted 5 lines

I don't know what exactly the issues are between "first merge patch serias and code-cleanup there", and "first merge this fix and rebase the patch series". When I was preparing the patch series about errno, I was also hate that codes you have posted above,

This is exactly my intent behind this patch.

Quoted 2 lines

but I need to make it focused first to make
people review this series first even there are two times calling like above.

So far I've not encountered situation where PRs are being discussed on github and

patch-series is not posted on the mailing list, but it clear to me it can create a confusion

as some people are not following github as closely as mailing list and we don't have

the clarity over which has the priority when it comes to sending patches whether it is github or

the NVMe mailing list.

Quoted 1 line

From my side, I'll make sure to monitor the PR on github before posting

any further patches.

Quoted 2 lines

If you really want to make it consistent, then why don't we just title it like "Do not return in case of errors in the middle"

I don't mind that in fact I can resend this patch refer to your patch-set and mark it

as a preparation patch with changing title, if PR wasn't in process.

Quoted 3 lines

If you really want to close fd to fix the leak issue, then I think it needs to cover more than that.

I've never opposed to fix the leak, but this was not the intent.

As far as this patch goes, since you are already working on PR with github, will retract this

change and we should go with that change.

#7 lore raw

PATCH 1/1 nvme-cli: close fd before return

Quoted 9 lines

I've not said anything against PR. I was just explaining the idea behind sending

this patch. We can leave it to PR now.

But this is a good example of having things not be posted on the mailing list can

a create confusion, will be monitoring PR going forward.

Agree with your point. Actually, recently I'm trying to send patches of nvme-cli to linux-nvme mailing list, but I'm not sure whether the PR will not be allowed or not because I'm a just one who contributes this project :)

Quoted 9 lines

So far I've not encountered situation where PRs are being discussed on github and

patch-series is not posted on the mailing list, but it clear to me it can create a confusion

as some people are not following github as closely as mailing list and we don't have

Quoted 7 lines

the clarity over which has the priority when it comes to sending patches whether it is github or

the NVMe mailing list.

From my side, I'll make sure to monitor the PR on github before posting any further patches.

Thanks for your support, But I know it's hard to monitor Github PRs all the time.

Quoted 4 lines

I don't mind that in fact I can resend this patch refer to your patch-set and mark it

as a preparation patch with changing title, if PR wasn't in process.

Actually, I'm okay with your patch by making it focused onto something clear. Anyway, this patch looks good to me in code level, though.

Quoted 6 lines

I've never opposed to fix the leak, but this was not the intent.

As far as this patch goes, since you are already working on PR with github, will retract this

change and we should go with that change.

Let's see how we can go through it together :)