io_uring: let IORING_OP_FILES_UPDATE support to choose fixed file slots
PATCH v2 1/1 io_uring: let IORING_OP_FILES_UPDATE support to choose fixed file slots
One big issue with file registration feature is that it needs user space apps to maintain free slot info about io_uring's fixed file table, which really is a burden for development. Now since io_uring starts to choose free file slot for user space apps by using IORING_FILE_INDEX_ALLOC flag in accept or open operations, but they need app to uses direct accept or direct open, which as far as I know, some apps are not prepared to use direct accept or open yet.
To support apps, who still need real fds, use registration feature easier, let IORING_OP_FILES_UPDATE support to choose fixed file slots, which will store picked fixed files slots in fd array and let cqe return the number of slots allocated.
---
V2:
Add illegal flags check in io_close_prep().
--- fs/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++------ include/uapi/linux/io_uring.h | 1 + 2 files changed, 66 insertions(+), 10 deletions(-)
fs/io_uring.c +65 -10
@@ -574,6 +574,7 @@ struct io_close {struct file *file;int fd;u32 file_slot; +u32 flags; };struct io_timeout_data {
@@ -1366,7 +1367,9 @@ static int io_req_prep_async(struct io_kiocb *req);static int io_install_fixed_file (struct io_kiocb *req,struct file *file,unsigned int issue_flags,u32 slot_index); -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, +unsigned int offset); +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags);static enum hrtimer_restart io_link_timeout_fn (struct hrtimer *timer);static void io_eventfd_signal (struct io_ring_ctx *ctx);
@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +static int io_close_prep (struct io_kiocb *req,const struct io_uring_sqe *sqe) { -if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) +if (sqe->off || sqe->addr || sqe->len || sqe->buf_index)return -EINVAL ;if (req->flags &REQ_F_FIXED_FILE )return -EBADF ; req->close.fd =READ_ONCE (sqe->fd); req->close.file_slot =READ_ONCE (sqe->file_index); -if (req->close.file_slot && req->close.fd) + req->close.flags =READ_ONCE (sqe->close_flags); +if (req->close.flags & ~IORING_CLOSE_FD_AND_FILE_SLOT ) +return -EINVAL ; +if (!(req->close.flags &IORING_CLOSE_FD_AND_FILE_SLOT ) && + req->close.file_slot && req->close.fd)return -EINVAL ;return 0 ;
@@ -5970,7 +5979,8 @@ static int io_close(struct io_kiocb *req, unsigned int issue_flags)if (req->close.file_slot) { ret =io_close_fixed (req, issue_flags); - goto err; +if (ret || !(req->close.flags &IORING_CLOSE_FD_AND_FILE_SLOT )) + goto err; }spin_lock (&files->file_lock);
@@ -8003,6 +8013,42 @@ static int io_files_update_prep(struct io_kiocb *req,return 0 ; } +static int io_files_update_with_index_alloc (struct io_kiocb *req, +unsigned int issue_flags) +{ +__s32 __user *fds =u64_to_user_ptr (req->rsrc_update.arg); +struct file *file; +unsigned int done, nr_fds = req->rsrc_update.nr_args; +int ret, fd; + +for (done =0 ; done < nr_fds; done++) { +if (copy_from_user (&fd, &fds[done],sizeof (fd))) { + ret = -EFAULT ; +break ; + } + + file =fget (fd); +if (!file) { + ret = -EBADF ; + goto out; + } + ret =io_fixed_fd_install (req, issue_flags, file, +IORING_FILE_INDEX_ALLOC ); +if (ret <0 ) + goto out; +if (copy_to_user (&fds[done], &ret,sizeof (ret))) { + ret = -EFAULT ; +__io_close_fixed (req, issue_flags, ret); +break ; + } + } + +out: +if (done) +return done; +return ret; +} +static int io_files_update (struct io_kiocb *req,unsigned int issue_flags) {struct io_ring_ctx *ctx = req->ctx;
@@ -8016,10 +8062,14 @@ static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) up.resv =0 ; up.resv2 =0 ; -io_ring_submit_lock (ctx, issue_flags); - ret =__io_register_rsrc_update (ctx,IORING_RSRC_FILE , - &up, req->rsrc_update.nr_args); -io_ring_submit_unlock (ctx, issue_flags); +if (req->rsrc_update.offset ==IORING_FILE_INDEX_ALLOC ) { + ret =io_files_update_with_index_alloc (req, issue_flags); + }else { +io_ring_submit_lock (ctx, issue_flags); + ret =__io_register_rsrc_update (ctx,IORING_RSRC_FILE , + &up, req->rsrc_update.nr_args); +io_ring_submit_unlock (ctx, issue_flags); + }if (ret <0 )req_set_fail (req);
@@ -10183,9 +10233,9 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file,return ret; } -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags) +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, +unsigned int offset) { -unsigned int offset = req->close.file_slot -1 ;struct io_ring_ctx *ctx = req->ctx;struct io_fixed_file *file_slot;struct file *file;
@@ -10222,6 +10272,11 @@ static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags)return ret; } +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags) +{ +return __io_close_fixed (req, issue_flags, req->close.file_slot -1 ); +} +static int __io_sqe_files_update (struct io_ring_ctx *ctx,struct io_uring_rsrc_update2 *up,unsigned nr_args)
include/uapi/linux/io_uring.h +1 -0
@@ -47,6 +47,7 @@ struct io_uring_sqe {__u32 unlink_flags;__u32 hardlink_flags;__u32 xattr_flags; +__u32 close_flags; };__u64 user_data;/* data to be passed back at completion time */ /* pack this to avoid bogus arm OABI complaints */
Signature · 1 line
2.14.4.44.g2045bb6
Jens Axboe
maintainerfs/io_uring.c static int io_statx(struct io_kiocb *req, unsigned int issue_flags) · line 5952
Quoted 6 lines · Xiaoguang Wang
fs/io_uring.c +2
-0@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +
This should go into uapi/linux/io_uring.h - I'll just move it, no need for a v3 for that. Test case should add it too.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerQuoted 10 lines · Jens Axboe
fs/io_uring.c static int io_statx(struct io_kiocb *req, unsigned int issue_flags) · line 5952
Quoted 6 lines · Xiaoguang Wang
fs/io_uring.c +2
-0@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +This should go into
uapi/linux/io_uring.h- I'll just move it, no need for a v3 for that. Test case should add it too.
Here's what I merged so far:
Changes:
- I re-wrote the commit message slightly
- Move flag to header where it belongs
- Get rid of 'goto' in io_files_update_with_index_alloc()
- Drop unneeded variable in io_files_update_with_index_alloc()
Signature · 1 line
Jens Axboe
hi,
Quoted 12 lines · Jens Axboe
Quoted 9 lines · Xiaoguang Wang
fs/io_uring.c static int io_statx(struct io_kiocb *req, unsigned int issue_flags) · line 5952
Quoted 6 lines
fs/io_uring.c +2
-0@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +This should go into
uapi/linux/io_uring.h- I'll just move it, no need for a v3 for that. Test case should add it too.Here's what I merged so far:
Quoted 8 lines
Changes:
- I re-wrote the commit message slightly
- Move flag to header where it belongs
- Get rid of 'goto' inio_files_update_with_index_alloc()
- Drop unneeded variable inio_files_update_with_index_alloc()
I think file registration feature is much easier to use now, thanks!
Regards,
Xiaoguang Wang
Quoted 1 line
Jens Axboe
maintainerQuoted 16 lines · Xiaoguang Wang
hi,
Quoted 14 lines · Jens Axboe
Quoted 9 lines · Xiaoguang Wang
fs/io_uring.c static int io_statx(struct io_kiocb *req, unsigned int issue_flags) · line 5952
Quoted 6 lines
fs/io_uring.c +2
-0@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +This should go into
uapi/linux/io_uring.h- I'll just move it, no need for a v3 for that. Test case should add it too.Here's what I merged so far:
Quoted 7 lines
Quoted 6 lines
Changes:
- I re-wrote the commit message slightly
- Move flag to header where it belongs
- Get rid of 'goto' inio_files_update_with_index_alloc()
- Drop unneeded variable inio_files_update_with_index_alloc()I think file registration feature is much easier to use now, thanks!
Thanks for making the change! Will you send a v2 of the liburing test? Then I'll get that queued up too.
Signature · 1 line
Jens Axboe
Quoted 18 lines · Xiaoguang Wang
Quoted 16 lines
hi,
Quoted 14 lines · Jens Axboe
Quoted 9 lines · Xiaoguang Wang
fs/io_uring.c static int io_statx(struct io_kiocb *req, unsigned int issue_flags) · line 5952
Quoted 6 lines
fs/io_uring.c +2
-0@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +This should go into
uapi/linux/io_uring.h- I'll just move it, no need for a v3 for that. Test case should add it too.Here's what I merged so far:
Quoted 8 lines
Quoted 7 lines
Quoted 6 lines
Changes:
- I re-wrote the commit message slightly
- Move flag to header where it belongs
- Get rid of 'goto' inio_files_update_with_index_alloc()
- Drop unneeded variable inio_files_update_with_index_alloc()I think file registration feature is much easier to use now, thanks!
Thanks for making the change! Will you send a v2 of the liburing test?
Yeah, I'll do it, will send it soon.
Regards,
Xiaoguang Wang
Quoted 2 lines
Then I'll get that queued up too.
Quoted 8 lines · Xiaoguang Wang
One big issue with file registration feature is that it needs user space apps to maintain free slot info about io_uring's fixed file table, which really is a burden for development. Now since
io_uringstarts to choose free file slot for user space apps by usingIORING_FILE_INDEX_ALLOCflag in accept or open operations, but they need app to uses direct accept or direct open, which as far as I know, some apps are not prepared to use direct accept or open yet.
Quoted 4 lines
To support apps, who still need real fds, use registration feature easier, let
IORING_OP_FILES_UPDATEsupport to choose fixed file slots, which will store picked fixed files slots in fd array and let cqe return the number of slots allocated.
Why close bits are piggybacked in this patch without any mention in the commit message? What is error semantics of IORING_CLOSE_FD_AND_FILE_SLOT. Fail if any errored or both? How can it be reliably used? Why we do two separate things in one request with not clear semantics?
There is already one fix pending. Another problem on the surface is that it may call io_close() twice and the second will fail, e.g.
-> io_close()
---> close_fixed()
---> if (file->flush) return -EAGAIN
-> io_close()
---> close_fixed() // fails
I do think we need to revert the close change, and then remake properly and only if someone can describe sane semantics for it.
Quoted 166 lines
Suggested-by Hao Xu <[email protected]>Signed-off-by Xiaoguang Wang <[email protected]>---
V2:
Add illegal flags check inio_close_prep().--- fs/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++------ include/uapi/linux/io_uring.h | 1 + 2 files changed, 66 insertions(+), 10 deletions(-)fs/io_uring.c +65
-10@@ -574,6 +574,7 @@ struct io_close {struct file *file;int fd;u32 file_slot; +u32 flags; };struct io_timeout_data {@@ -1366,7 +1367,9 @@ static int io_req_prep_async(struct io_kiocb *req);static int io_install_fixed_file (struct io_kiocb *req,struct file *file,unsigned int issue_flags,u32 slot_index); -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, +unsigned int offset); +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags);static enum hrtimer_restart io_link_timeout_fn (struct hrtimer *timer);static void io_eventfd_signal (struct io_ring_ctx *ctx);@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags)return 0 ; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 +static int io_close_prep (struct io_kiocb *req,const struct io_uring_sqe *sqe) { -if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) +if (sqe->off || sqe->addr || sqe->len || sqe->buf_index)return -EINVAL ;if (req->flags &REQ_F_FIXED_FILE )return -EBADF ; req->close.fd =READ_ONCE (sqe->fd); req->close.file_slot =READ_ONCE (sqe->file_index); -if (req->close.file_slot && req->close.fd) + req->close.flags =READ_ONCE (sqe->close_flags); +if (req->close.flags & ~IORING_CLOSE_FD_AND_FILE_SLOT ) +return -EINVAL ; +if (!(req->close.flags &IORING_CLOSE_FD_AND_FILE_SLOT ) && + req->close.file_slot && req->close.fd)return -EINVAL ;return 0 ;@@ -5970,7 +5979,8 @@ static int io_close(struct io_kiocb *req, unsigned int issue_flags)if (req->close.file_slot) { ret =io_close_fixed (req, issue_flags); - goto err; +if (ret || !(req->close.flags &IORING_CLOSE_FD_AND_FILE_SLOT )) + goto err; }spin_lock (&files->file_lock);@@ -8003,6 +8013,42 @@ static int io_files_update_prep(struct io_kiocb *req,return 0 ; } +static int io_files_update_with_index_alloc (struct io_kiocb *req, +unsigned int issue_flags) +{ +__s32 __user *fds =u64_to_user_ptr (req->rsrc_update.arg); +struct file *file; +unsigned int done, nr_fds = req->rsrc_update.nr_args; +int ret, fd; + +for (done =0 ; done < nr_fds; done++) { +if (copy_from_user (&fd, &fds[done],sizeof (fd))) { + ret = -EFAULT ; +break ; + } + + file =fget (fd); +if (!file) { + ret = -EBADF ; + goto out; + } + ret =io_fixed_fd_install (req, issue_flags, file, +IORING_FILE_INDEX_ALLOC ); +if (ret <0 ) + goto out; +if (copy_to_user (&fds[done], &ret,sizeof (ret))) { + ret = -EFAULT ; +__io_close_fixed (req, issue_flags, ret); +break ; + } + } + +out: +if (done) +return done; +return ret; +} +static int io_files_update (struct io_kiocb *req,unsigned int issue_flags) {struct io_ring_ctx *ctx = req->ctx;@@ -8016,10 +8062,14 @@ static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) up.resv =0 ; up.resv2 =0 ; -io_ring_submit_lock (ctx, issue_flags); - ret =__io_register_rsrc_update (ctx,IORING_RSRC_FILE , - &up, req->rsrc_update.nr_args); -io_ring_submit_unlock (ctx, issue_flags); +if (req->rsrc_update.offset ==IORING_FILE_INDEX_ALLOC ) { + ret =io_files_update_with_index_alloc (req, issue_flags); + }else { +io_ring_submit_lock (ctx, issue_flags); + ret =__io_register_rsrc_update (ctx,IORING_RSRC_FILE , + &up, req->rsrc_update.nr_args); +io_ring_submit_unlock (ctx, issue_flags); + }if (ret <0 )req_set_fail (req);@@ -10183,9 +10233,9 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file,return ret; } -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags) +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, +unsigned int offset) { -unsigned int offset = req->close.file_slot -1 ;struct io_ring_ctx *ctx = req->ctx;struct io_fixed_file *file_slot;struct file *file;@@ -10222,6 +10272,11 @@ static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags)return ret; } +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags) +{ +return __io_close_fixed (req, issue_flags, req->close.file_slot -1 ); +} +static int __io_sqe_files_update (struct io_ring_ctx *ctx,struct io_uring_rsrc_update2 *up,unsigned nr_args)include/uapi/linux/io_uring.h +1
-0@@ -47,6 +47,7 @@ struct io_uring_sqe {__u32 unlink_flags;__u32 hardlink_flags;__u32 xattr_flags; +__u32 close_flags; };__u64 user_data;/* data to be passed back at completion time */ /* pack this to avoid bogus arm OABI complaints */
Signature · 1 line
Pavel Begunkov
hello Pavel,
Quoted 9 lines · Xiaoguang Wang
Quoted 8 lines
One big issue with file registration feature is that it needs user space apps to maintain free slot info about io_uring's fixed file table, which really is a burden for development. Now since
io_uringstarts to choose free file slot for user space apps by usingIORING_FILE_INDEX_ALLOCflag in accept or open operations, but they need app to uses direct accept or direct open, which as far as I know, some apps are not prepared to use direct accept or open yet.
Quoted 7 lines
Quoted 4 lines
To support apps, who still need real fds, use registration feature easier, let
IORING_OP_FILES_UPDATEsupport to choose fixed file slots, which will store picked fixed files slots in fd array and let cqe return the number of slots allocated.Why close bits are piggybacked in this patch without any mention in the commit message?
Sorry, it actually should be in one separate patch.
Quoted 4 lines
What is error semantics of
IORING_CLOSE_FD_AND_FILE_SLOT. Fail if any errored or both? How can it be reliably used? Why we do two separate things in one request with not clear semantics?
Yeah, indeed I know this issue when writing this patch, either normal fd close or file slot close may fail independently. I put them together just for efficiency, one CLOSE op can do two jobs.
Quoted 9 lines
There is already one fix pending. Another problem on the surface is that it may call
io_close()twice and the second will fail, e.g.->
io_close()
--->close_fixed()
---> if (file->flush) return-EAGAIN
->io_close()
--->close_fixed()// fails
Thanks for pointing this issue.
Quoted 4 lines
I do think we need to revert the close change, and then remake properly and only if someone can describe sane semantics for it.
Revert this close change seems reasonable, I thought for a while, didn't figure out better solutions yet.
Sorry for the trouble again.
Regards,
Xiaoguang Wang
Quoted 157 lines
Quoted 156 lines
Suggested-by Hao Xu <[email protected]>Signed-off-by Xiaoguang Wang <[email protected]>---
V2:
Add illegal flags check inio_close_prep().
---
fs/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++------include/uapi/linux/io_uring.h| 1 +2 files changed, 66 insertions(+), 10 deletions(-)fs/io_uring.c +4
-1@@ -574,6 +574,7 @@ struct io_close { struct file *file; int fd; u32 file_slot; + u32 flags; }; struct io_timeout_data {@@ -1366,7 +1367,9 @@ static int io_req_prep_async(struct io_kiocb *req); static int io_install_fixed_file(struct io_kiocb *req, struct file *file, unsigned int issue_flags, u32 slot_index); -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, + unsigned int offset); +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); static void io_eventfd_signal(struct io_ring_ctx *ctx);@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags) return 0; }+#define
IORING_CLOSE_FD_AND_FILE_SLOT1
+
static intio_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe){- if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) + if (sqe->off || sqe->addr || sqe->len || sqe->buf_index)return
-EINVAL;
if (req->flags&REQ_F_FIXED_FILE)
return-EBADF;
req->close.fd=READ_ONCE(sqe->fd);req->close.file_slot=READ_ONCE(sqe->file_index); - if (req->close.file_slot&&req->close.fd)
+req->close.flags=READ_ONCE(sqe->close_flags); + if (req->close.flags& ~IORING_CLOSE_FD_AND_FILE_SLOT) + return-EINVAL;
+ if (!(req->close.flags &IORING_CLOSE_FD_AND_FILE_SLOT) && +req->close.file_slot&&req->close.fd) return-EINVAL;
return 0;fs/io_uring.c +51
-5@@ -5970,7 +5979,8 @@ static int io_close(struct io_kiocb *req, unsigned int issue_flags) if (req->close.file_slot) { ret = io_close_fixed(req, issue_flags); - goto err; + if (ret || !(req->close.flags & IORING_CLOSE_FD_AND_FILE_SLOT)) + goto err; } spin_lock(&files->file_lock);@@ -8003,6 +8013,42 @@ static int io_files_update_prep(struct io_kiocb *req, return 0; } +static int io_files_update_with_index_alloc(struct io_kiocb *req, + unsigned int issue_flags) +{ + __s32 __user *fds = u64_to_user_ptr(req->rsrc_update.arg); + struct file *file; + unsigned int done, nr_fds = req->rsrc_update.nr_args; + int ret, fd; + + for (done = 0; done < nr_fds; done++) { + if (copy_from_user(&fd, &fds[done], sizeof(fd))) { + ret = -EFAULT; + break; + } + + file = fget(fd); + if (!file) { + ret = -EBADF; + goto out; + } + ret = io_fixed_fd_install(req, issue_flags, file, + IORING_FILE_INDEX_ALLOC); + if (ret < 0) + goto out; + if (copy_to_user(&fds[done], &ret, sizeof(ret))) { + ret = -EFAULT; + __io_close_fixed(req, issue_flags, ret); + break; + } + } + +out: + if (done) + return done; + return ret; +} + static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) { struct io_ring_ctx *ctx = req->ctx;@@ -8016,10 +8062,14 @@ static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) up.resv = 0; up.resv2 = 0; - io_ring_submit_lock(ctx, issue_flags); - ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, - &up, req->rsrc_update.nr_args); - io_ring_submit_unlock(ctx, issue_flags); + if (req->rsrc_update.offset == IORING_FILE_INDEX_ALLOC) { + ret = io_files_update_with_index_alloc(req, issue_flags); + } else { + io_ring_submit_lock(ctx, issue_flags); + ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, + &up, req->rsrc_update.nr_args); + io_ring_submit_unlock(ctx, issue_flags); + } if (ret < 0) req_set_fail(req);@@ -10183,9 +10233,9 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file, return ret; } -static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) +static int __io_close_fixed(struct io_kiocb *req, unsigned int issue_flags, + unsigned int offset) { - unsigned int offset = req->close.file_slot - 1; struct io_ring_ctx *ctx = req->ctx; struct io_fixed_file *file_slot; struct file *file;@@ -10222,6 +10272,11 @@ static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) return ret; } +static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) +{ + return __io_close_fixed(req, issue_flags, req->close.file_slot - 1); +} + static int __io_sqe_files_update(struct io_ring_ctx *ctx, struct io_uring_rsrc_update2 *up, unsigned nr_args)include/uapi/linux/io_uring.h +1
-0@@ -47,6 +47,7 @@ struct io_uring_sqe { __u32 unlink_flags; __u32 hardlink_flags; __u32 xattr_flags; + __u32 close_flags; };__u64 user_data; /* data to be passed back at completion time */ /* pack this to avoid bogus arm OABI complaints */
Quoted 41 lines · Xiaoguang Wang
hello Pavel,
Quoted 9 lines · Xiaoguang Wang
Quoted 8 lines
One big issue with file registration feature is that it needs user space apps to maintain free slot info about io_uring's fixed file table, which really is a burden for development. Now since
io_uringstarts to choose free file slot for user space apps by usingIORING_FILE_INDEX_ALLOCflag in accept or open operations, but they need app to uses direct accept or direct open, which as far as I know, some apps are not prepared to use direct accept or open yet.Quoted 7 lines
Quoted 4 lines
To support apps, who still need real fds, use registration feature easier, let
IORING_OP_FILES_UPDATEsupport to choose fixed file slots, which will store picked fixed files slots in fd array and let cqe return the number of slots allocated.Why close bits are piggybacked in this patch without any mention in the commit message?
Sorry, it actually should be in one separate patch.
Quoted 4 lines
What is error semantics of
IORING_CLOSE_FD_AND_FILE_SLOT. Fail if any errored or both? How can it be reliably used? Why we do two separate things in one request with not clear semantics?Yeah, indeed I know this issue when writing this patch, either normal fd close or file slot close may fail independently. I put them together just for efficiency, one CLOSE op can do two jobs.
Quoted 9 lines
There is already one fix pending. Another problem on the surface is that it may call
io_close()twice and the second will fail, e.g.->
io_close()
--->close_fixed()
---> if (file->flush) return-EAGAIN
->io_close()
--->close_fixed()// failsThanks for pointing this issue.
Quoted 2 lines
Quoted 5 lines
Quoted 2 lines
I do think we need to revert the close change, and then remake properly and only if someone can describe sane semantics for it.
Revert this close change seems reasonable, I thought for a while, didn't figure out better solutions yet.
Sorry for the trouble again.
No worries, let's kill it off then and return it back if and when someone comes up with a good semantaics for error handling.
Quoted 161 lines
Regards,
Xiaoguang WangQuoted 157 lines
Quoted 156 lines
Suggested-by Hao Xu <[email protected]>Signed-off-by Xiaoguang Wang <[email protected]>---
V2:
Add illegal flags check inio_close_prep().--- fs/io_uring.c | 75 +++++++++++++++++++++++++++++++++++++------ include/uapi/linux/io_uring.h | 1 + 2 files changed, 66 insertions(+), 10 deletions(-)fs/io_uring.c +62
-8@@ -574,6 +574,7 @@ struct io_close { struct file *file; int fd; u32 file_slot; + u32 flags; }; struct io_timeout_data {@@ -1366,7 +1367,9 @@ static int io_req_prep_async(struct io_kiocb *req); static int io_install_fixed_file(struct io_kiocb *req, struct file *file, unsigned int issue_flags, u32 slot_index); -static int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); +static int __io_close_fixed (struct io_kiocb *req,unsigned int issue_flags, + unsigned int offset); +static inline int io_close_fixed (struct io_kiocb *req,unsigned int issue_flags); static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer); static void io_eventfd_signal(struct io_ring_ctx *ctx);@@ -5945,16 +5948,22 @@ static int io_statx(struct io_kiocb *req, unsigned int issue_flags) return 0; } +#define IORING_CLOSE_FD_AND_FILE_SLOT 1 + static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) { - if (sqe->off || sqe->addr || sqe->len || sqe->rw_flags || sqe->buf_index) + if (sqe->off || sqe->addr || sqe->len || sqe->buf_index) return -EINVAL; if (req->flags & REQ_F_FIXED_FILE) return -EBADF; req->close.fd = READ_ONCE(sqe->fd); req->close.file_slot = READ_ONCE(sqe->file_index); - if (req->close.file_slot && req->close.fd) + req->close.flags = READ_ONCE(sqe->close_flags); + if (req->close.flags & ~IORING_CLOSE_FD_AND_FILE_SLOT) + return -EINVAL; + if (!(req->close.flags & IORING_CLOSE_FD_AND_FILE_SLOT) && + req->close.file_slot && req->close.fd) return -EINVAL; return 0;@@ -5970,7 +5979,8 @@ static int io_close(struct io_kiocb *req, unsigned int issue_flags) if (req->close.file_slot) { ret = io_close_fixed(req, issue_flags); - goto err; + if (ret || !(req->close.flags & IORING_CLOSE_FD_AND_FILE_SLOT)) + goto err; } spin_lock(&files->file_lock);@@ -8003,6 +8013,42 @@ static int io_files_update_prep(struct io_kiocb *req, return 0; } +static int io_files_update_with_index_alloc(struct io_kiocb *req, + unsigned int issue_flags) +{ + __s32 __user *fds = u64_to_user_ptr(req->rsrc_update.arg); + struct file *file; + unsigned int done, nr_fds = req->rsrc_update.nr_args; + int ret, fd; + + for (done = 0; done < nr_fds; done++) { + if (copy_from_user(&fd, &fds[done], sizeof(fd))) { + ret = -EFAULT; + break; + } + + file = fget(fd); + if (!file) { + ret = -EBADF; + goto out; + } + ret = io_fixed_fd_install(req, issue_flags, file, + IORING_FILE_INDEX_ALLOC); + if (ret < 0) + goto out; + if (copy_to_user(&fds[done], &ret, sizeof(ret))) { + ret = -EFAULT; + __io_close_fixed(req, issue_flags, ret); + break; + } + } + +out: + if (done) + return done; + return ret; +} + static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) { struct io_ring_ctx *ctx = req->ctx;@@ -8016,10 +8062,14 @@ static int io_files_update(struct io_kiocb *req, unsigned int issue_flags) up.resv = 0; up.resv2 = 0; - io_ring_submit_lock(ctx, issue_flags); - ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, - &up, req->rsrc_update.nr_args); - io_ring_submit_unlock(ctx, issue_flags); + if (req->rsrc_update.offset == IORING_FILE_INDEX_ALLOC) { + ret = io_files_update_with_index_alloc(req, issue_flags); + } else { + io_ring_submit_lock(ctx, issue_flags); + ret = __io_register_rsrc_update(ctx, IORING_RSRC_FILE, + &up, req->rsrc_update.nr_args); + io_ring_submit_unlock(ctx, issue_flags); + } if (ret < 0) req_set_fail(req);@@ -10183,9 +10233,9 @@ static int io_install_fixed_file(struct io_kiocb *req, struct file *file, return ret; } -static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) +static int __io_close_fixed(struct io_kiocb *req, unsigned int issue_flags, + unsigned int offset) { - unsigned int offset = req->close.file_slot - 1; struct io_ring_ctx *ctx = req->ctx; struct io_fixed_file *file_slot; struct file *file;@@ -10222,6 +10272,11 @@ static int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) return ret; } +static inline int io_close_fixed(struct io_kiocb *req, unsigned int issue_flags) +{ + return __io_close_fixed(req, issue_flags, req->close.file_slot - 1); +} + static int __io_sqe_files_update(struct io_ring_ctx *ctx, struct io_uring_rsrc_update2 *up, unsigned nr_args)include/uapi/linux/io_uring.h +1
-0@@ -47,6 +47,7 @@ struct io_uring_sqe { __u32 unlink_flags; __u32 hardlink_flags; __u32 xattr_flags; + __u32 close_flags; }; __u64 user_data; /* data to be passed back at completion time */ /* pack this to avoid bogus arm OABI complaints */
Signature · 1 line
Pavel Begunkov