vfs: added better file aio_read aio_write operations presence check
PATCH 1/1 vfs: added better file aio_read aio_write operations presence check
From: Marco Stornelli <[email protected]>
If a filesystem in the file operations specifies for read and write operations only do_sync_read and do_sync_write without init aio_read and aio_write, there will be a kernel oops, because the vfs code check the presence of (to read for example) read OR aio_read method, then it calls read if it's pointer is not null. It's not sufficient because if the read function is actually a do_sync_read, it calls aio_read but without checking the presence. I think a BUG_ON check can be more useful.
---
fs/read_write.c renamed from fs/read_write.c.orig +2 -0
@@ -240,6 +240,7 @@ ssize_t do_sync_read(struct file *filp, kiocb.ki_pos = *ppos; kiocb.ki_left = len; +BUG_ON (!filp->f_op->aio_read);for (;;) { ret = filp->f_op->aio_read (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )
@@ -295,6 +296,7 @@ ssize_t do_sync_write(struct file *filp, kiocb.ki_pos = *ppos; kiocb.ki_left = len; +BUG_ON (!filp->f_op->aio_write);for (;;) { ret = filp->f_op->aio_write (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )
Quoted 6 lines · Marco Stornelli
From: Marco Stornelli <[email protected]>
If a filesystem in the file operations specifies for read and write operations only
do_sync_readanddo_sync_writewithout initaio_readandaio_write, there will be a kernel oops, because the vfs code check the presence of (to read for example) read ORaio_readmethod, then it calls read if it's pointer is not null. It's not sufficient because if the read function is actually ado_sync_read, it callsaio_readbut without checking the presence. I think aBUG_ONcheck can be more useful.
Instead of doing a BUG_ON() why can't we simply fall back to the generic_aio functions since most of the fs tend to do so as below.
--- fs/read_write.c | 10 ++++++++-- 1 files changed, 8 insertions(+), 2 deletions(-)
fs/read_write.c +0 -0
@@ -225,7 +225,11 @@ ssize_t do_sync_read(struct file *filp, char
__user *buf, size_t len, loff_t *pp
kiocb.ki_left = len;
for (;;) {
- ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
+ if (filp->f_op->aio_read)
+ ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
+ else
+ ret = generic_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
if (ret != -EIOCBRETRY)
break;
wait_on_retry_sync_kiocb(&kiocb);
@@ -280,7 +284,10 @@ ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, lof
kiocb.ki_left = len;
for (;;) {
- ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
+ if (filp->f_op->aio_write)
+ ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
+ else
+ ret = generic_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
if (ret != -EIOCBRETRY)
break;
wait_on_retry_sync_kiocb(&kiocb);
Signature · 5 lines
1.5.4.3 Thanks - Manish
Quoted 27 lines
Signed-off-by Marco Stornelli <[email protected]>---
fs/read_write.c 2008-09-16 11:01:13.000000000 +0200 renamed from fs/read_write.c.orig 2008-08-20 20:11:37.000000000 +0200 +2
-1@@ -240,6 +240,7 @@ ssize_t do_sync_read(struct file *filp, kiocb.ki_pos = *ppos; kiocb.ki_left = len; + BUG_ON(!filp->f_op->aio_read); for (;;) { ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos); if (ret != -EIOCBRETRY)@@ -295,6 +296,7 @@ ssize_t do_sync_write(struct file *filp, kiocb.ki_pos = *ppos; kiocb.ki_left = len; + BUG_ON(!filp->f_op->aio_write); for (;;) { ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos); if (ret != -EIOCBRETRY) --To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
BUG_ON it was a way to say: "hey you've used the do_sync_read/write as read/write operation but you don't specified an aio_read/write", but your solutions it's good too.
Manish Katiyar ha scritto:
Quoted 49 lines
Quoted 6 lines · Marco Stornelli
From: Marco Stornelli <[email protected]>
If a filesystem in the file operations specifies for read and write operations only
do_sync_readanddo_sync_writewithout initaio_readandaio_write, there will be a kernel oops, because the vfs code check the presence of (to read for example) read ORaio_readmethod, then it calls read if it's pointer is not null. It's not sufficient because if the read function is actually ado_sync_read, it callsaio_readbut without checking the presence. I think aBUG_ONcheck can be more useful.Instead of doing a
BUG_ON()why can't we simply fall back to thegeneric_aiofunctions since most of the fs tend to do so as below.Signed-off-by Manish Katiyar <[email protected]>--- fs/read_write.c | 10 ++++++++-- 1 files changed, 8 insertions(+), 2 deletions(-)fs/read_write.c +0
-0@@ -225,7 +225,11 @@ ssize_t do_sync_read(struct file *filp, char
__user*buf,size_tlen,loff_t*ppfs/read_write.c +4
-1kiocb.ki_left = len;for (;;) { - ret = filp->f_op->aio_read (&kiocb, &iov,1 , kiocb.ki_pos); +if (filp->f_op->aio_read) + ret = filp->f_op->aio_read (&kiocb, &iov,1 , kiocb.ki_pos); +else + ret =generic_file_aio_read (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )break ;wait_on_retry_sync_kiocb (&kiocb);@@ -280,7 +284,10 @@ ssize_t do_sync_write(struct file *filp, constchar
__user*buf,size_tlen, loffs/read_write.c +4
-1kiocb.ki_left = len;for (;;) { - ret = filp->f_op->aio_write (&kiocb, &iov,1 , kiocb.ki_pos); +if (filp->f_op->aio_write) + ret = filp->f_op->aio_write (&kiocb, &iov,1 , kiocb.ki_pos); +else + ret =generic_file_aio_write (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )break ;wait_on_retry_sync_kiocb (&kiocb);
Quoted 3 lines · Marco Stornelli
BUG_ONit was a way to say: "hey you've used the do_sync_read/write as read/write operation but you don't specified an aio_read/write", but your solutions it's good too.
Looks like I made some copy paste error while sending the patch. Below is the updated one.
--- fs/read_write.c | 10 ++++++++-- 1 files changed, 8 insertions(+), 2 deletions(-)
fs/read_write.c +0 -0
@@ -225,7 +225,10 @@ ssize_t do_sync_read(struct file *filp, char
__user *buf, size_t len, loff_t *pp
kiocb.ki_left = len;
for (;;) {
- ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
+ if (filp->f_op->aio_read)
+ ret = filp->f_op->aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
+ else
+ ret = generic_file_aio_read(&kiocb, &iov, 1, kiocb.ki_pos);
if (ret != -EIOCBRETRY)
break;
wait_on_retry_sync_kiocb(&kiocb);
@@ -280,7 +283,10 @@ ssize_t do_sync_write(struct file *filp, const char __user *buf, size_t len, lof
kiocb.ki_left = len;
for (;;) {
- ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
+ if (filp->f_op->aio_write)
+ ret = filp->f_op->aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
+ else
+ ret = generic_file_aio_write(&kiocb, &iov, 1, kiocb.ki_pos);
if (ret != -EIOCBRETRY)
break;
wait_on_retry_sync_kiocb(&kiocb);
Signature · 4 lines
1.5.4.3 Thanks - Manish
Quoted 53 lines
Manish Katiyar ha scritto:
Quoted 49 lines
Quoted 6 lines · Marco Stornelli
From: Marco Stornelli <[email protected]>
If a filesystem in the file operations specifies for read and write operations only
do_sync_readanddo_sync_writewithout initaio_readandaio_write, there will be a kernel oops, because the vfs code check the presence of (to read for example) read ORaio_readmethod, then it calls read if it's pointer is not null. It's not sufficient because if the read function is actually ado_sync_read, it callsaio_readbut without checking the presence. I think aBUG_ONcheck can be more useful.Instead of doing a
BUG_ON()why can't we simply fall back to thegeneric_aiofunctions since most of the fs tend to do so as below.Signed-off-by Manish Katiyar <[email protected]>--- fs/read_write.c | 10 ++++++++-- 1 files changed, 8 insertions(+), 2 deletions(-)fs/read_write.c +0
-0@@ -225,7 +225,11 @@ ssize_t do_sync_read(struct file *filp, char
__user*buf,size_tlen,loff_t*ppkiocb.ki_left = len; for (;;) {fs/read_write.c +4
-1- ret = filp->f_op->aio_read (&kiocb, &iov,1 , kiocb.ki_pos); +if (filp->f_op->aio_read) + ret = filp->f_op->aio_read (&kiocb, &iov,1 , kiocb.ki_pos); +else + ret =generic_file_aio_read (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )break ;wait_on_retry_sync_kiocb (&kiocb);@@ -280,7 +284,10 @@ ssize_t do_sync_write(struct file *filp, constchar
__user*buf,size_tlen, lofkiocb.ki_left = len; for (;;) {fs/read_write.c +4
-1- ret = filp->f_op->aio_write (&kiocb, &iov,1 , kiocb.ki_pos); +if (filp->f_op->aio_write) + ret = filp->f_op->aio_write (&kiocb, &iov,1 , kiocb.ki_pos); +else + ret =generic_file_aio_write (&kiocb, &iov,1 , kiocb.ki_pos);if (ret != -EIOCBRETRY )break ;wait_on_retry_sync_kiocb (&kiocb);
Quoted 6 lines · Marco Stornelli
From: Marco Stornelli <[email protected]>
If a filesystem in the file operations specifies for read and write operations only
do_sync_readanddo_sync_writewithout initaio_readandaio_write, there will be a kernel oops, because the vfs code check the presence of (to read for example) read ORaio_readmethod, then it calls read if it's pointer is not null. It's not sufficient because if the read function is actually ado_sync_read, it callsaio_readbut without checking the presence. I think aBUG_ONcheck can be more useful.
A NULL pointer derference is just as clear as the bug..