Jesper Dangaard Brouer <brouer@xxxxxxxxxx> writes:
> This patch change BPF syscall to avoid returning file descriptor value zero.
>
> As mentioned in cover letter, it is very impractical when extending kABI
> that the file-descriptor value 'zero' is valid, as this requires new fields
> must be initialised as minus-1. First step is to change the kernel such that
> BPF-syscall simply doesn't return value zero as a FD number.
>
> This patch achieves this by similar code to anon_inode_getfd(), with the
> exception of getting unused FD starting from 1. The kernel already supports
> starting from a specific FD value, as this is used by f_dupfd(). It seems
> simpler to replicate part of anon_inode_getfd() code and use this start from
> offset feature, instead of using f_dupfd() handling afterwards.
>
> Signed-off-by: Jesper Dangaard Brouer <brouer@xxxxxxxxxx>
> ---
> fs/file.c | 2 +-
> include/linux/file.h | 1 +
> kernel/bpf/syscall.c | 38 ++++++++++++++++++++++++++++++++------
> 3 files changed, 34 insertions(+), 7 deletions(-)
>
> diff --git a/fs/file.c b/fs/file.c
> index abb8b7081d7a..122185cb7707 100644
> --- a/fs/file.c
> +++ b/fs/file.c
> @@ -535,7 +535,7 @@ int __alloc_fd(struct files_struct *files,
> return error;
> }
>
> -static int alloc_fd(unsigned start, unsigned flags)
> +int alloc_fd(unsigned start, unsigned flags)
Missing an EXPORT_SYMBOL() to go with this.
> {
> return __alloc_fd(current->files, start, rlimit(RLIMIT_NOFILE), flags);
> }
> diff --git a/include/linux/file.h b/include/linux/file.h
> index 122f80084a3e..927fb6c2582d 100644
> --- a/include/linux/file.h
> +++ b/include/linux/file.h
> @@ -85,6 +85,7 @@ extern int f_dupfd(unsigned int from, struct file *file, unsigned flags);
> extern int replace_fd(unsigned fd, struct file *file, unsigned flags);
> extern void set_close_on_exec(unsigned int fd, int flag);
> extern bool get_close_on_exec(unsigned int fd);
> +extern int alloc_fd(unsigned start, unsigned flags);
> extern int __get_unused_fd_flags(unsigned flags, unsigned long nofile);
> extern int get_unused_fd_flags(unsigned flags);
> extern void put_unused_fd(unsigned int fd);
> diff --git a/kernel/bpf/syscall.c b/kernel/bpf/syscall.c
> index 4d530b1d5683..6eba236aacd1 100644
> --- a/kernel/bpf/syscall.c
> +++ b/kernel/bpf/syscall.c
> @@ -688,6 +688,32 @@ const struct file_operations bpf_map_fops = {
> .poll = bpf_map_poll,
> };
>
> +/* Code is similar to anon_inode_getfd(), except starts at FD 1 */
> +int bpf_anon_inode_getfd(const char *name, const struct file_operations *fops,
> + void *priv, int flags)
> +{
I think it's a little sad that we have to essentially re-implement
anon_inode_getfd(). I guess an alternative could be to add an extra
parameter to the existing function, either with a different name and a
wrapper function, or just change all the callers (by my count there are
only 13 call sites outside of BPF). Not sure if that is better, though?
> + int error, fd;
> + struct file *file;
> +
> + error = alloc_fd(1, flags);
> + if (error < 0)
> + return error;
> + fd = error;
> +
> + file = anon_inode_getfile(name, fops, priv, flags);
> + if (IS_ERR(file)) {
> + error = PTR_ERR(file);
> + goto err_put_unused_fd;
> + }
> + fd_install(fd, file);
> +
> + return fd;
> +
> +err_put_unused_fd:
> + put_unused_fd(fd);
> + return error;
> +}
> +
> int bpf_map_new_fd(struct bpf_map *map, int flags)
> {
> int ret;
> @@ -696,8 +722,8 @@ int bpf_map_new_fd(struct bpf_map *map, int flags)
> if (ret < 0)
> return ret;
>
> - return anon_inode_getfd("bpf-map", &bpf_map_fops, map,
> - flags | O_CLOEXEC);
> + return bpf_anon_inode_getfd("bpf-map", &bpf_map_fops, map,
> + flags | O_CLOEXEC);
> }
>
> int bpf_get_file_flag(int flags)
> @@ -1840,8 +1866,8 @@ int bpf_prog_new_fd(struct bpf_prog *prog)
> if (ret < 0)
> return ret;
>
> - return anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
> - O_RDWR | O_CLOEXEC);
> + return bpf_anon_inode_getfd("bpf-prog", &bpf_prog_fops, prog,
> + O_RDWR | O_CLOEXEC);
> }
>
> static struct bpf_prog *____bpf_prog_get(struct fd f)
> @@ -2471,7 +2497,7 @@ int bpf_link_settle(struct bpf_link_primer *primer)
>
> int bpf_link_new_fd(struct bpf_link *link)
> {
> - return anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
> + return bpf_anon_inode_getfd("bpf-link", &bpf_link_fops, link, O_CLOEXEC);
> }
>
> struct bpf_link *bpf_link_get_from_fd(u32 ufd)
> @@ -4024,7 +4050,7 @@ static int bpf_enable_runtime_stats(void)
> return -EBUSY;
> }
>
> - fd = anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
> + fd = bpf_anon_inode_getfd("bpf-stats", &bpf_stats_fops, NULL, O_CLOEXEC);
> if (fd >= 0)
> static_key_slow_inc(&bpf_stats_enabled_key.key);
>
I guess you should also fix __btf_new_fd() in btf.c?
