Skip to contentSkip to conversation

unify the file-closing stuff in fs/file.c

RFC In review Al Viro 4 participants 10 messages May 12, 2022 – May 16, 2022

Subsystems (2)
lore
Share & tools
lore

Al Viro

author maintainer
#0 lore raw

RFC 1/1 unify the file-closing stuff in fs/file.c

Right now we have two places that do such removals - pick_file()
and {__,}close_fd_get_file().

They are almost identical - the only difference is in calling
conventions (well, and the fact that __... is called with descriptor table locked).

Calling conventions are... interesting.

1) pick_file() - returns file or ERR_PTR(-EBADF) or ERR_PTR(-EINVAL). The latter is for "descriptor is greater than size of descriptor table". One of the callers treats all ERR_PTR(...) as "return -EBADF"; another uses ERR_PTR(-EINVAL) as "end the loop now" indicator.

2) {__,}close_fd_get_file() returns 0 or -ENOENT (huh?), with file (or NULL) passed to caller by way of struct file ** argument. One of the callers (binder) ignores the return value completely and checks if the file is NULL. Another (io_uring) checks for return value being negative, then maps -ENOENT to -EBADF, not that any other value would be possible.

ERR_PTR(-EINVAL) magic in case of pick_file() is borderline defensible; {__,}close_fd_get_file() conventions are insane. The older caller (in binder) had never even looked at return value; the newer one patches the bogus -ENOENT to what it wants to report, with strange "defensive" BS logics just in case __close_fd_get_file() would somehow find a different error to report.

At the very least, {__,}close_fd_get_file() callers would've been happier if it just returned file or NULL. What's more, I'm seriously tempted to make pick_file() do the same thing. close_fd() won't care (checking for NULL is just as easy as for IS_ERR) and __range_close() could just as well cap the max_fd argument with last_fd(files_fdtable(current->files)).

Does anybody see problems with the following?

commit 8819510a641800a63ab10d6b5ab283cada1cbd50

Author: Al Viro <[email protected]>
Date:   Thu May 12 17:08:03 2022 -0400

Unify the primitives for file descriptor closing

Currently we have 3 primitives for removing an opened file from descriptor table - pick_file(), __close_fd_get_file() and close_fd_get_file(). Their calling conventions are rather odd and there's a code duplication for no good reason. They can be unified -

1) have __range_close() cap max_fd in the very beginning; that way we don't need separate way for pick_file() to report being past the end of descriptor table.

2) make {__,}close_fd_get_file() return file (or NULL) directly, rather than returning it via struct file ** argument. Don't bother with (bogus) return value - nobody wants that -ENOENT.

3) make pick_file() return NULL on unopened descriptor - the only caller that used to care about the distinction between descriptor past the end of descriptor table and finding NULL in descriptor table doesn't give a damn after (1).

4) lift ->files_lock out of pick_file()

That actually simplifies the callers, as well as the primitives themselves. Code duplication is also gone...

drivers/android/binder.c +1 -1
index 8351c5638880..27c9b004823a 100644
@@ -1855,7 +1855,7 @@ static void binder_deferred_fd_close(int fd)
 	if (!twcb)
 		return;
 	init_task_work(&twcb->twork, binder_do_fd_close);
-	close_fd_get_file(fd, &twcb->file);
+	twcb->file = close_fd_get_file(fd);
 	if (twcb->file) {
 		filp_close(twcb->file, current->files);
 		task_work_add(current, &twcb->twork, TWA_RESUME);
fs/file.c +28 -49

130 lines not on this page · raw message

fs/internal.h +1 -1
index 08503dc68d2b..4065e2679103 100644
@@ -125,7 +125,7 @@ extern struct file *do_file_open_root(const struct path *,
 		const char *, const struct open_flags *);
 extern struct open_how build_open_how(int flags, umode_t mode);
 extern int build_open_flags(const struct open_how *how, struct open_flags *op);
-extern int __close_fd_get_file(unsigned int fd, struct file **res);
+extern struct file *__close_fd_get_file(unsigned int fd);
 
 long do_sys_ftruncate(unsigned int fd, loff_t length, int small);
 int chmod_common(const struct path *path, umode_t mode);
fs/io_uring.c +2 -5
index dc580a30723d..7257b0870353 100644
@@ -5137,13 +5137,10 @@ static int io_close(struct io_kiocb *req, unsigned int issue_flags)
 		return -EAGAIN;
 	}
 
-	ret = __close_fd_get_file(close->fd, &file);
+	file = __close_fd_get_file(close->fd);
 	spin_unlock(&files->file_lock);
-	if (ret < 0) {
-		if (ret == -ENOENT)
-			ret = -EBADF;
+	if (!file)
 		goto err;
-	}
 
 	/* No ->flush() or already async, safely close from here */
 	ret = filp_close(file, current->files);
include/linux/fdtable.h +1 -1
index d0e78174874a..e066816f3519 100644
@@ -125,7 +125,7 @@ int iterate_fd(struct files_struct *, unsigned,
 
 extern int close_fd(unsigned int fd);
 extern int __close_range(unsigned int fd, unsigned int max_fd, unsigned int flags);
-extern int close_fd_get_file(unsigned int fd, struct file **res);
+extern struct file *close_fd_get_file(unsigned int fd);
 extern int unshare_fd(unsigned long unshare_flags, unsigned int max_fds,
 		      struct files_struct **new_fdp);
 
author maintainer
#1 lore raw

RFC PATCH 1/1 get rid of the remnants of 'batched' fget/fput stuff

Hadn't been used since 62906e89e63b "io_uring: remove file batch-get optimisation", should've been killed back then...

---

fs/file.c +13 -19
index 9780888fa2da..9fbc0c653930 100644
@@ -850,7 +850,7 @@ void do_close_on_exec(struct files_struct *files)
 }
 
 static inline struct file *__fget_files_rcu(struct files_struct *files,
-	unsigned int fd, fmode_t mask, unsigned int refs)
+	unsigned int fd, fmode_t mask)
 {
 	for (;;) {
 		struct file *file;
@@ -876,10 +876,10 @@ static inline struct file *__fget_files_rcu(struct files_struct *files,
 		 * Such a race can take two forms:
 		 *
 		 *  (a) the file ref already went down to zero,
-		 *      and get_file_rcu_many() fails. Just try
+		 *      and get_file_rcu() fails. Just try
 		 *      again:
 		 */
-		if (unlikely(!get_file_rcu_many(file, refs)))
+		if (unlikely(!get_file_rcu(file)))
 			continue;
 
 		/*
@@ -888,11 +888,11 @@ static inline struct file *__fget_files_rcu(struct files_struct *files,
 		 *       pointer having changed, because it always goes
 		 *       hand-in-hand with 'fdt'.
 		 *
-		 * If so, we need to put our refs and try again.
+		 * If so, we need to put our ref and try again.
 		 */
 		if (unlikely(rcu_dereference_raw(files->fdt) != fdt) ||
 		    unlikely(rcu_dereference_raw(*fdentry) != file)) {
-			fput_many(file, refs);
+			fput(file);
 			continue;
 		}
 
@@ -905,37 +905,31 @@ static inline struct file *__fget_files_rcu(struct files_struct *files,
 }
 
 static struct file *__fget_files(struct files_struct *files, unsigned int fd,
-				 fmode_t mask, unsigned int refs)
+				 fmode_t mask)
 {
 	struct file *file;
 
 	rcu_read_lock();
-	file = __fget_files_rcu(files, fd, mask, refs);
+	file = __fget_files_rcu(files, fd, mask);
 	rcu_read_unlock();
 
 	return file;
 }
 
-static inline struct file *__fget(unsigned int fd, fmode_t mask,
-				  unsigned int refs)
-{
-	return __fget_files(current->files, fd, mask, refs);
-}
-
-struct file *fget_many(unsigned int fd, unsigned int refs)
+static inline struct file *__fget(unsigned int fd, fmode_t mask)
 {
-	return __fget(fd, FMODE_PATH, refs);
+	return __fget_files(current->files, fd, mask);
 }
 
 struct file *fget(unsigned int fd)
 {
-	return __fget(fd, FMODE_PATH, 1);
+	return __fget(fd, FMODE_PATH);
 }
 EXPORT_SYMBOL(fget);
 
 struct file *fget_raw(unsigned int fd)
 {
-	return __fget(fd, 0, 1);
+	return __fget(fd, 0);
 }
 EXPORT_SYMBOL(fget_raw);
 
@@ -945,7 +939,7 @@ struct file *fget_task(struct task_struct *task, unsigned int fd)
 
 	task_lock(task);
 	if (task->files)
-		file = __fget_files(task->files, fd, 0, 1);
+		file = __fget_files(task->files, fd, 0);
 	task_unlock(task);
 
 	return file;
@@ -1014,7 +1008,7 @@ static unsigned long __fget_light(unsigned int fd, fmode_t mask)
 			return 0;
 		return (unsigned long)file;
 	} else {
-		file = __fget(fd, mask, 1);
+		file = __fget(fd, mask);
 		if (!file)
 			return 0;
 		return FDPUT_FPUT | (unsigned long)file;
fs/file_table.c +2 -7
index 7d2e692b66a9..1ffd74bbbed6 100644
@@ -368,9 +368,9 @@ EXPORT_SYMBOL_GPL(flush_delayed_fput);
 
 static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 
-void fput_many(struct file *file, unsigned int refs)
+void fput(struct file *file)
 {
-	if (atomic_long_sub_and_test(refs, &file->f_count)) {
+	if (atomic_long_dec_and_test(&file->f_count)) {
 		struct task_struct *task = current;
 
 		if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
@@ -389,11 +389,6 @@ void fput_many(struct file *file, unsigned int refs)
 	}
 }
 
-void fput(struct file *file)
-{
-	fput_many(file, 1);
-}
-
 /*
  * synchronous analog of fput(); for kernel threads that might be needed
  * in some umount() (and thus can't use flush_delayed_fput() without
include/linux/file.h +0 -2
index 51e830b4fe3a..39704eae83e2 100644
@@ -14,7 +14,6 @@
 struct file;
 
 extern void fput(struct file *);
-extern void fput_many(struct file *, unsigned int);
 
 struct file_operations;
 struct task_struct;
@@ -47,7 +46,6 @@ static inline void fdput(struct fd fd)
 }
 
 extern struct file *fget(unsigned int fd);
-extern struct file *fget_many(unsigned int fd, unsigned int refs);
 extern struct file *fget_raw(unsigned int fd);
 extern struct file *fget_task(struct task_struct *task, unsigned int fd);
 extern unsigned long __fdget(unsigned int fd);
include/linux/fs.h +2 -3
index bbde95387a23..0521f0b1356b 100644
@@ -981,9 +981,8 @@ static inline struct file *get_file(struct file *f)
 	atomic_long_inc(&f->f_count);
 	return f;
 }
-#define get_file_rcu_many(x, cnt)	\
-	atomic_long_add_unless(&(x)->f_count, (cnt), 0)
-#define get_file_rcu(x) get_file_rcu_many((x), 1)
+#define get_file_rcu(x)	\
+	atomic_long_add_unless(&(x)->f_count, 1, 0)
 #define file_count(x)	atomic_long_read(&(x)->f_count)
 
 #define	MAX_NON_LFS	((1UL<<31) - 1)
Acked-by on v1 1/1
#2 lore raw

Re: [RFC][PATCH] get rid of the remnants of 'batched' fget/fput stuff

Quoted 2 lines · Al Viro

Hadn't been used since 62906e89e63b "io_uring: remove file batch-get optimisation", should've been killed back then...

I'm pretty sure this has been sent out before, forget from whom. So it's not like it hasn't been suggested or posted... In case it matters:

Signature · 1 line
Jens Axboe
author maintainer
#3 lore raw

Re: [RFC][PATCH] get rid of the remnants of 'batched' fget/fput stuff

Quoted 5 lines · Jens Axboe
Quoted 2 lines · Al Viro

Hadn't been used since 62906e89e63b "io_uring: remove file batch-get optimisation", should've been killed back then...

I'm pretty sure this has been sent out before, forget from whom.

Right you are... From Gou Hao, had fallen through the cracks back in November ;-/ Rebased and replaced my variant with it now...

#5 lore raw
In reply to Al Viro
Quoted 27 lines · Al Viro

Right now we have two places that do such removals - pick_file()
and {__,}close_fd_get_file().

They are almost identical - the only difference is in calling
conventions (well, and the fact that __... is called with descriptor table locked).

Calling conventions are... interesting.

1) pick_file() - returns file or ERR_PTR(-EBADF) or ERR_PTR(-EINVAL). The latter is for "descriptor is greater than size of descriptor table". One of the callers treats all ERR_PTR(...) as "return -EBADF"; another uses ERR_PTR(-EINVAL) as "end the loop now" indicator.

2) {__,}close_fd_get_file() returns 0 or -ENOENT (huh?), with file (or NULL) passed to caller by way of struct file ** argument. One of the callers (binder) ignores the return value completely and checks if the file is NULL. Another (io_uring) checks for return value being negative, then maps -ENOENT to -EBADF, not that any other value would be possible.

ERR_PTR(-EINVAL) magic in case of pick_file() is borderline defensible; {__,}close_fd_get_file() conventions are insane. The older caller (in binder) had never even looked at return value; the newer one patches the bogus -ENOENT to what it wants to report, with strange "defensive" BS logics just in case __close_fd_get_file() would somehow find a different error to report.

Quoted 7 lines

At the very least, {__,}close_fd_get_file() callers would've been happier if it just returned file or NULL. What's more, I'm seriously tempted to make pick_file() do the same thing. close_fd() won't care (checking for NULL is just as easy as for IS_ERR) and __range_close() could just as well cap the max_fd argument with last_fd(files_fdtable(current->files)).

Does anybody see problems with the following?

Looks good to me, and much better than passing in the pointer to the file pointer imho.

Signature · 1 line
Jens Axboe
#6 lore raw
Quoted 35 lines · Jens Axboe
Quoted 33 lines · Al Viro

Right now we have two places that do such removals - pick_file()
and {__,}close_fd_get_file().

They are almost identical - the only difference is in calling
conventions (well, and the fact that __... is called with descriptor table locked).

Calling conventions are... interesting.

1) pick_file() - returns file or ERR_PTR(-EBADF) or ERR_PTR(-EINVAL). The latter is for "descriptor is greater than size of descriptor table". One of the callers treats all ERR_PTR(...) as "return -EBADF"; another uses ERR_PTR(-EINVAL) as "end the loop now" indicator.

2) {__,}close_fd_get_file() returns 0 or -ENOENT (huh?), with file (or NULL) passed to caller by way of struct file ** argument. One of the callers (binder) ignores the return value completely and checks if the file is NULL. Another (io_uring) checks for return value being negative, then maps -ENOENT to -EBADF, not that any other value would be possible.

ERR_PTR(-EINVAL) magic in case of pick_file() is borderline defensible; {__,}close_fd_get_file() conventions are insane. The older caller (in binder) had never even looked at return value; the newer one patches the bogus -ENOENT to what it wants to report, with strange "defensive" BS logics just in case __close_fd_get_file() would somehow find a different error to report.

At the very least, {__,}close_fd_get_file() callers would've been happier if it just returned file or NULL. What's more, I'm seriously tempted to make pick_file() do the same thing. close_fd() won't care (checking for NULL is just as easy as for IS_ERR) and __range_close() could just as well cap the max_fd argument with last_fd(files_fdtable(current->files)).

Quoted 4 lines
Quoted 1 line

Does anybody see problems with the following?

Looks good to me, and much better than passing in the pointer to the file pointer imho.

I agree. This looks good.

Quoted 4 lines

--
Jens Axboe

maintainer Reviewed-by on v1 1/1
#7 lore raw
In reply to Al Viro
Quoted 27 lines · Al Viro

Right now we have two places that do such removals - pick_file()
and {__,}close_fd_get_file().

They are almost identical - the only difference is in calling
conventions (well, and the fact that __... is called with descriptor table locked).

Calling conventions are... interesting.

1) pick_file() - returns file or ERR_PTR(-EBADF) or ERR_PTR(-EINVAL). The latter is for "descriptor is greater than size of descriptor table". One of the callers treats all ERR_PTR(...) as "return -EBADF"; another uses ERR_PTR(-EINVAL) as "end the loop now" indicator.

2) {__,}close_fd_get_file() returns 0 or -ENOENT (huh?), with file (or NULL) passed to caller by way of struct file ** argument. One of the callers (binder) ignores the return value completely and checks if the file is NULL. Another (io_uring) checks for return value being negative, then maps -ENOENT to -EBADF, not that any other value would be possible.

ERR_PTR(-EINVAL) magic in case of pick_file() is borderline defensible; {__,}close_fd_get_file() conventions are insane. The older caller (in binder) had never even looked at return value; the newer one patches the bogus -ENOENT to what it wants to report, with strange "defensive" BS logics just in case __close_fd_get_file() would somehow find a different error to report.

Quoted 5 lines

At the very least, {__,}close_fd_get_file() callers would've been happier if it just returned file or NULL. What's more, I'm seriously tempted to make pick_file() do the same thing. close_fd() won't care (checking for NULL is just as easy as for IS_ERR) and __range_close() could just as well cap the max_fd argument with last_fd(files_fdtable(current->files)).

Originally, __close_range() did that last_fd() thing for both the cloexec and the non-cloexec case. But that proved buggy for the cloexec part (dumb oversight). So the cloexec part retrieves the last_fd() and marks cloexec under the spinlock to make sure it's stable.

We could've done the same that this patch does now and retrieved last_fd() in __range_close() too. But without having looked at {__,}close_fd_get_file() it was more obvious imho to let pick_file() tell the caller when to terminate the loop as it avoids fiddling last_fd() out from under the rcu lock given that we below we had to take the spinlock anyway.

With the motivation of reducing __close_fd_get_file() to pick_file() the other way is more sensible.

Quoted 27 lines

Does anybody see problems with the following?

commit 8819510a641800a63ab10d6b5ab283cada1cbd50

Author: Al Viro <[email protected]>
Date:   Thu May 12 17:08:03 2022 -0400

Unify the primitives for file descriptor closing

Currently we have 3 primitives for removing an opened file from descriptor table - pick_file(), __close_fd_get_file() and close_fd_get_file(). Their calling conventions are rather odd and there's a code duplication for no good reason. They can be unified -

1) have __range_close() cap max_fd in the very beginning; that way we don't need separate way for pick_file() to report being past the end of descriptor table.

2) make {__,}close_fd_get_file() return file (or NULL) directly, rather than returning it via struct file ** argument. Don't bother with (bogus) return value - nobody wants that -ENOENT.

3) make pick_file() return NULL on unopened descriptor - the only caller that used to care about the distinction between descriptor past the end of descriptor table and finding NULL in descriptor table doesn't give a damn after (1).

Quoted 6 lines

4) lift ->files_lock out of pick_file()

That actually simplifies the callers, as well as the primitives themselves. Code duplication is also gone...

The change in io_close() looked a bit subtle because ret was overriden in there by __close_fd_get_file() prior to changing it to return struct file but ret is set to -EBADF at the top of io_close() so it looks fine.

Since you change pick_file() to require the caller to hold the lock it'd be good to add a:

Context: Caller must hold files_lock.

to the kernel-doc I added; similar to what I did for last_fd().

Also, there's a bunch of regression tests I added in:

tools/testing/selftests/core/close_range_test.c

including various tests for issues reported by syzbot. Might be worth running to verify we didn't regress anything.

Thanks!

author maintainer
#8 lore raw
Quoted 1 line · Christian Brauner

Context: Caller must hold files_lock.

Done and force-pushed to #work.fd

Quoted 6 lines

Also, there's a bunch of regression tests I added in:

tools/testing/selftests/core/close_range_test.c

including various tests for issues reported by syzbot. Might be worth running to verify we didn't regress anything.

# PASSED: 7 / 7 tests passed.
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0

#9 lore raw
Quoted 8 lines · Al Viro
Quoted 1 line · Christian Brauner

Context: Caller must hold files_lock.

Done and force-pushed to #work.fd

Quoted 2 lines

Also, there's a bunch of regression tests I added in:

Quoted 7 lines
Quoted 4 lines

tools/testing/selftests/core/close_range_test.c

including various tests for issues reported by syzbot. Might be worth running to verify we didn't regress anything.

# PASSED: 7 / 7 tests passed.
# Totals: pass:7 fail:0 xfail:0 xpass:0 skip:0 error:0

Thank you, appreciate it!

Christian