From: Jens Axboe <[email protected]>
To: Stefan <[email protected]>, [email protected]
Subject: Re: madvise/fadvise 32-bit length
Date: Sat, 1 Jun 2024 09:05:01 -0600 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
On 6/1/24 8:19 AM, Jens Axboe wrote:
> On 6/1/24 3:43 AM, Stefan wrote:
>> io_uring uses the __u32 len field in order to pass the length to
>> madvise and fadvise, but these calls use an off_t, which is 64bit on
>> 64bit platforms.
>>
>> When using liburing, the length is silently truncated to 32bits (so
>> 8GB length would become zero, which has a different meaning of "until
>> the end of the file" for fadvise).
>>
>> If my understanding is correct, we could fix this by introducing new
>> operations MADVISE64 and FADVISE64, which use the addr3 field instead
>> of the length field for length.
>
> We probably just want to introduce a flag and ensure that older stable
> kernels check it, and then use a 64-bit field for it when the flag is
> set.
I think this should do it on the kernel side, as we already check these
fields and return -EINVAL as needed. Should also be trivial to backport.
Totally untested... Might want a FEAT flag for this, or something where
it's detectable, to make the liburing change straight forward.
diff --git a/io_uring/advise.c b/io_uring/advise.c
index 7085804c513c..cb7b881665e5 100644
--- a/io_uring/advise.c
+++ b/io_uring/advise.c
@@ -17,14 +17,14 @@
struct io_fadvise {
struct file *file;
u64 offset;
- u32 len;
+ u64 len;
u32 advice;
};
struct io_madvise {
struct file *file;
u64 addr;
- u32 len;
+ u64 len;
u32 advice;
};
@@ -33,11 +33,13 @@ int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
struct io_madvise *ma = io_kiocb_to_cmd(req, struct io_madvise);
- if (sqe->buf_index || sqe->off || sqe->splice_fd_in)
+ if (sqe->buf_index || sqe->splice_fd_in)
return -EINVAL;
ma->addr = READ_ONCE(sqe->addr);
- ma->len = READ_ONCE(sqe->len);
+ ma->len = READ_ONCE(sqe->off);
+ if (!ma->len)
+ ma->len = READ_ONCE(sqe->len);
ma->advice = READ_ONCE(sqe->fadvise_advice);
req->flags |= REQ_F_FORCE_ASYNC;
return 0;
@@ -78,11 +80,13 @@ int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
{
struct io_fadvise *fa = io_kiocb_to_cmd(req, struct io_fadvise);
- if (sqe->buf_index || sqe->addr || sqe->splice_fd_in)
+ if (sqe->buf_index || sqe->splice_fd_in)
return -EINVAL;
fa->offset = READ_ONCE(sqe->off);
- fa->len = READ_ONCE(sqe->len);
+ fa->len = READ_ONCE(sqe->addr);
+ if (!fa->len)
+ fa->len = READ_ONCE(sqe->len);
fa->advice = READ_ONCE(sqe->fadvise_advice);
if (io_fadvise_force_async(fa))
req->flags |= REQ_F_FORCE_ASYNC;
--
Jens Axboe
next prev parent reply other threads:[~2024-06-01 15:05 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-06-01 9:43 madvise/fadvise 32-bit length Stefan
2024-06-01 14:19 ` Jens Axboe
2024-06-01 15:05 ` Jens Axboe [this message]
2024-06-01 15:22 ` Stefan
2024-06-01 15:35 ` Jens Axboe
2024-06-01 15:51 ` Stefan
2024-06-01 18:33 ` Jens Axboe
2024-06-02 8:58 ` Stefan
2024-06-02 14:49 ` Jens Axboe
2024-06-05 5:25 ` Stefan
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox