io_uring: Add IO_U_F_DEVICE_ERROR to identify error types
PATCH 1/1 io_uring: Add IO_U_F_DEVICE_ERROR to identify error types
__io_u_log_error expects a positive value of io_u->error parsing it with strerror() expecting it to be an errno. io_uring_cmd (cmd_type=nvme), for example, has returned errno value as a positive value and device-specific CQE status type and code as well.
Commit 78831c6b35c5 ("io_uring: Fix the flip to negative of CQE status") has put the abs() to the cqe->res, and it caused confusions between the actual CQE stauts and the system error value (errno). Now we have Commit 2a13699a89dc ("io_uring: Add .errdetails to parse CQ status"), meaning that io_uring_cmd ioengines will parse the actual value of io_u->error value as CQE status value, so we should know if the value is for CQE status or errno.
This patch added a flag IO_U_F_DEVICE_ERROR to io_u to represent if io_u->error has device-specific error value, otherwise it's errno.
--- engines/io_uring.c | 22 ++++++++++++++++------ io_u.c | 9 ++++++--- io_u.h | 1 + 3 files changed, 23 insertions(+), 9 deletions(-)
engines/io_uring.c +16 -6
@@ -528,12 +528,9 @@ static struct io_u *fio_ioring_cmd_event(struct thread_data *td, int event) cqe = &ld->cq_ring.cqes[index]; io_u = (struct io_u *) (uintptr_t ) cqe->user_data; -if (cqe->res !=0 ) { - io_u->error =abs (cqe->res); -return io_u; - }else { - io_u->error =0 ; - } + io_u->error = cqe->res; +if (io_u->error !=0 ) + goto ret;if (o->cmd_type ==FIO_URING_CMD_NVME ) { data =FILE_ENG_DATA (io_u->file);
@@ -544,6 +541,16 @@ static struct io_u *fio_ioring_cmd_event(struct thread_data *td, int event) } } +ret: +/* +* If IO_U_F_DEVICE_ERROR is not set, io_u->error will be parsed as an +* errno, otherwise device-specific error value (status value in CQE). +*/ +if ((int )io_u->error >0 ) +io_u_set (td, io_u,IO_U_F_DEVICE_ERROR ); +else +io_u_clear (td, io_u,IO_U_F_DEVICE_ERROR ); + io_u->error =abs (io_u->error);return io_u; }
@@ -557,6 +564,9 @@ static char *fio_ioring_cmd_errdetails(struct thread_data *td,#define MAXMSGCHUNK 128char *msg, msgchunk[MAXMSGCHUNK ]; +if (!(io_u->flags &IO_U_F_DEVICE_ERROR )) +return NULL ; + msg =calloc (1 ,MAXERRDETAIL );strcpy (msg,"io_uring_cmd: " );
io_u.c +6 -3
@@ -1956,7 +1956,8 @@ static void __io_u_log_error(struct thread_data *td, struct io_u *io_u)log_err ("fio: io_u error%s%s: %s: %s offset=%llu, buflen=%llu\n" , io_u->file ?" on file " :"" , io_u->file ? io_u->file->file_name :"" , -strerror (io_u->error), + (io_u->flags &IO_U_F_DEVICE_ERROR ) ? +"Device-specific error" :strerror (io_u->error),io_ddir_name (io_u->ddir), io_u->offset, io_u->xfer_buflen);
@@ -1965,8 +1966,10 @@ static void __io_u_log_error(struct thread_data *td, struct io_u *io_u)if (td->io_ops->errdetails) {char *err = td->io_ops->errdetails (td, io_u); -log_err ("fio: %s\n" , err); -free (err); +if (err) { +log_err ("fio: %s\n" , err); +free (err); + } }if (!td->error)
io_u.h +1 -0
@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/*
Signature · 1 line
2.34.1
Jens Axboe
AppliedQuoted 13 lines · Minwoo Im
__io_u_log_errorexpects a positive value ofio_u->errorparsing it withstrerror()expecting it to be an errno.io_uring_cmd(cmd_type=nvme), for example, has returned errno value as a positive value and device-specific CQE status type and code as well.Commit
78831c6b35c5("io_uring: Fix the flip to negative of CQE status") has put theabs()to thecqe->res, and it caused confusions between the actual CQE stauts and the system error value (errno). Now we have Commit2a13699a89dc("io_uring: Add .errdetails to parse CQ status"), meaning thatio_uring_cmdioengines will parse the actual value ofio_u->errorvalue as CQE status value, so we should know if the value is for CQE status or errno.
Quoted 1 line
[...]
Applied, thanks!
[1/1] io_uring: Add IO_U_F_DEVICE_ERROR to identify error types
commit: ebe67b667f25694ead4caa0198598318f891b345
Best regards,
Signature · 1 line
Jens Axboe
io_u.h enum { · line 28
Quoted 12 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/*
The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Signature · 1 line
Jens Axboe
Quoted 18 lines · Jens Axboe
io_u.h enum { · line 28
Quoted 12 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
Quoted 8 lines
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting abs(io_u->error) since error is unsigned. Please fix that up as well.
I do have a bot that automatically runs mailing list patches through our CI, but it does not report the results to the list and I am not always quick enough to manually report CI failures:
https://github.com/fiotestbot/fio/actions
Vincent
Quoted 22 lines · Vincent Fu
Quoted 21 lines · Jens Axboe
io_u.h enum { · line 27
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
Quoted 8 lines
Quoted 4 lines
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting
abs(io_u->error)since error is unsigned. Please fix that up as well.
Doh yes, I fixed that up now.
Quoted 5 lines
I do have a bot that automatically runs mailing list patches through our CI, but it does not report the results to the list and I am not always quick enough to manually report CI failures:
Send them to the list! It's not like it's a high traffic list, and that's super useful. Mostly because it gets the same coverage as a github pr then, but also because it'll inform the submitter that there's an issue without either you or me letting them now. Hence it saves time and cycles, which is a big win in my book.
Signature · 1 line
Jens Axboe
Quoted 28 lines · Jens Axboe
Quoted 27 lines · Vincent Fu
Quoted 25 lines · Jens Axboe
io_u.h enum { · line 27
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Quoted 5 lines
Quoted 3 lines
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting
abs(io_u->error)since error is unsigned. Please fix that up as well.Doh yes, I fixed that up now.
I appreciate for your fix!
Quoted 6 lines
Quoted 6 lines
I do have a bot that automatically runs mailing list patches through our CI, but it does not report the results to the list and I am not always quick enough to manually report CI failures:
Vincent,
I will have a look into the CI when I posted patches to the mailing list. It's super cool to have it.
Quoted 9 lines
Send them to the list! It's not like it's a high traffic list, and that's super useful. Mostly because it gets the same coverage as a github pr then, but also because it'll inform the submitter that there's an issue without either you or me letting them now. Hence it saves time and cycles, which is a big win in my book.
Signature · 1 line
Jens Axboe
Quoted 28 lines · Jens Axboe
Quoted 27 lines · Vincent Fu
Quoted 25 lines · Jens Axboe
io_u.h enum { · line 27
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Quoted 5 lines
Quoted 3 lines
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting
abs(io_u->error)since error is unsigned. Please fix that up as well.Doh yes, I fixed that up now.
Jens,
How about the following one for the fix? (checked with clang)
engines/io_uring.c +1 -1
@@ -589,7 +589,7 @@ ret:io_u_set (td, io_u,IO_U_F_DEVICE_ERROR ); elseio_u_clear (td ,io_u ,IO_U_F_DEVICE_ERROR ); - io_u->error = io_u->error; + io_u->error =abs ((int )io_u->error);return io_u; }
io_u->error is unsigned, but it can have negative value as an errno. io_u->error should be converted to a positive one especially when it's an errno. If we don't, it will be like Unknown error:
fio: io_u error on file /dev/ng0n1: Unknown error -22: write offset=429916160, buflen=1048576
Quoted 15 lines
Quoted 6 lines
I do have a bot that automatically runs mailing list patches through our CI, but it does not report the results to the list and I am not always quick enough to manually report CI failures:
Send them to the list! It's not like it's a high traffic list, and that's super useful. Mostly because it gets the same coverage as a github pr then, but also because it'll inform the submitter that there's an issue without either you or me letting them now. Hence it saves time and cycles, which is a big win in my book.
Signature · 1 line
Jens Axboe
Quoted 56 lines · Minwoo Im
Quoted 28 lines · Jens Axboe
Quoted 27 lines · Vincent Fu
Quoted 25 lines · Jens Axboe
io_u.h enum { · line 27
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Quoted 5 lines
Quoted 3 lines
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting
abs(io_u->error)since error is unsigned. Please fix that up as well.Doh yes, I fixed that up now.
Jens,
How about the following one for the fix? (checked with clang)
engines/io_uring.c +1
-1@@ -589,7 +589,7 @@ ret:io_u_set (td, io_u,IO_U_F_DEVICE_ERROR ); elseio_u_clear (td ,io_u ,IO_U_F_DEVICE_ERROR ); - io_u->error = io_u->error; + io_u->error =abs ((int )io_u->error);return io_u; }
io_u->erroris unsigned, but it can have negative value as an errno.io_u->errorshould be converted to a positive one especially when it's an errno. If we don't, it will be like Unknown error:
Quoted 1 line
fio:
io_uerror on file/dev/ng0n1: Unknown error -22: write offset=429916160, buflen=1048576
Yeah I think that looks good. Can you send this as a patch against the current tree? I did do a fixup, but way too jet lagged and it's obviously garbage.
Signature · 1 line
Jens Axboe
Quoted 57 lines · Jens Axboe
Quoted 56 lines · Minwoo Im
Quoted 28 lines · Jens Axboe
Quoted 27 lines · Vincent Fu
Quoted 25 lines · Jens Axboe
io_u.h enum { · line 27
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Quoted 5 lines
Quoted 3 lines
Also, Minwoo, we are seeing some build failures with your patches with compilers rejecting
abs(io_u->error)since error is unsigned. Please fix that up as well.Doh yes, I fixed that up now.
Jens,
How about the following one for the fix? (checked with clang)
engines/io_uring.c +1
-1@@ -589,7 +589,7 @@ ret:io_u_set (td, io_u,IO_U_F_DEVICE_ERROR ); elseio_u_clear (td ,io_u ,IO_U_F_DEVICE_ERROR ); - io_u->error = io_u->error; + io_u->error =abs ((int )io_u->error);return io_u; }
io_u->erroris unsigned, but it can have negative value as an errno.io_u->errorshould be converted to a positive one especially when it's an errno. If we don't, it will be like Unknown error:
Quoted 5 lines
Quoted 1 line
fio:
io_uerror on file/dev/ng0n1: Unknown error -22: write offset=429916160, buflen=1048576Yeah I think that looks good. Can you send this as a patch against the current tree? I did do a fixup, but way too jet lagged and it's obviously garbage.
Sure, I will, Thanks!
Quoted 12 lines · Jens Axboe
Quoted 11 lines · Minwoo Im
io_u.h +1
-0@@ -22,6 +22,7 @@ enum {IO_U_F_BARRIER =1 <<6 ,IO_U_F_VER_LIST =1 <<7 ,IO_U_F_PATTERN_DONE =1 <<8 , + IO_U_F_DEVICE_ERROR = 1 << 9, };
Quoted 5 lines
io_u.h enum { · line 28
Quoted 1 line
io_u.h +0
-0/* The patches you sent should've been a series, how are they supposed to both apply when you add an item here for each of them as if the other one doesn't exist?
My bad. Sorry for making mess here. Based on the feature, I think it should not be a series, but, yes, adding new flags with the same value is totally a mess.
Quoted 4 lines
I'll fix it up, but for the future, if patches depend on each other, it should be a series. Please check if everything works when it's pushed out, which should be shortly.
Okay.
Quoted 4 lines
And since I'm on a plane and this doesn't appear to want to send, when you do see it, please also add HOWTO additions similar to the fio.1 additions you made.
Will do it. Thanks!
Quoted 3 lines
Signature · 2 lines
-- Jens Axboe