hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/ICA1GK -------------------------------- When a container starts, it makes use of both mount namespace and chroot to achieve path isolation. So to get the /dev in the container, what we actually need is fs->root of container's init task (i.e. the reaper), not its mount namespace. Acquire reaper's fs->root, look up the /dev under it, and filter the partitions output based on the /dev content. Signed-off-by: GONG Ruiqi <gongruiqi1@huawei.com> --- block/genhd.c | 124 ++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 121 insertions(+), 3 deletions(-) diff --git a/block/genhd.c b/block/genhd.c index f19a86b11dea..b186530af3bd 100644 --- a/block/genhd.c +++ b/block/genhd.c @@ -29,6 +29,8 @@ #ifdef CONFIG_BPF_RVI #include <linux/bpf.h> #include <linux/pid_namespace.h> +#include <linux/namei.h> +#include <linux/fs_struct.h> #endif #include "blk-throttle.h" @@ -1457,13 +1459,125 @@ BTF_ID_LIST(btf_diststats_ids) BTF_ID(struct, block_device) BTF_ID(struct, disk_stats) +struct traverse_ctx { + struct dir_context ctx; + struct dentry *parent_dentry; + struct xarray *dev_list; + unsigned int index; +}; + +static bool filldir_callback(struct dir_context *ctx, const char *name, + int namelen, loff_t offset, u64 ino, + unsigned int d_type) +{ + struct traverse_ctx *tctx = container_of(ctx, struct traverse_ctx, ctx); + struct dentry *child_dentry; + struct inode *inode; + struct bdev_handle *handle; + void *rc; + + if (d_type != DT_BLK) + return true; + + child_dentry = lookup_one_len(name, tctx->parent_dentry, namelen); + if (IS_ERR(child_dentry)) { + pr_warn("Lookup failed for %s: %ld\n", name, PTR_ERR(child_dentry)); + return true; + } + + // double check if it's block dev + inode = d_inode(child_dentry); + if (!S_ISBLK(inode->i_mode)) + goto err_put; + + handle = bdev_open_by_dev(inode->i_rdev, BLK_OPEN_READ, NULL, NULL); + if (IS_ERR(handle)) { + pr_err("Failed to open block device %s (err=%ld)\n", + name, PTR_ERR(handle)); + goto err_put; + } + + rc = xa_store(tctx->dev_list, tctx->index++, handle->bdev, GFP_KERNEL); + if (xa_is_err(rc)) + pr_warn("xa_store() on %d failed\n", tctx->index - 1); + + bdev_release(handle); +err_put: + dput(child_dentry); + return true; +} + +static unsigned int get_targeted_dev(struct xarray *dev_list) +{ + struct task_struct *reaper; + struct path root_path, dev_path; + struct file *dir; + int ret; + struct traverse_ctx buf = { + .ctx.actor = filldir_callback, + .dev_list = dev_list, + .index = 0, + }; + + xa_init(dev_list); + reaper = get_current_level1_reaper(); + if (!reaper) + return 0; + + /* Reference: get_task_root() */ + task_lock(reaper); + if (!reaper->fs) { + task_unlock(reaper); + goto out_put_reaper; + } + get_fs_root(reaper->fs, &root_path); + task_unlock(reaper); + + /* + * For vfs_path_lookup(), @name being "dev" or "/dev" makes no + * difference, since struct nameidata.root is preset. + */ + ret = vfs_path_lookup(root_path.dentry, root_path.mnt, "dev", + LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &dev_path); + if (ret) + goto out_put_root; + + dir = dentry_open(&dev_path, O_RDONLY, current_cred()); + if (IS_ERR(dir)) { + ret = PTR_ERR(dir); + goto out_put_devpath; + } + buf.parent_dentry = dev_path.dentry; + + iterate_dir(dir, &buf.ctx); + + filp_close(dir, NULL); +out_put_devpath: + path_put(&dev_path); +out_put_root: + path_put(&root_path); +out_put_reaper: + put_task_struct(reaper); + return buf.index; +} + +struct partitions_seq_priv { + struct class_dev_iter iter; // must be the first + struct xarray dev_list; + unsigned int dev_list_size; +}; + static void *bpf_show_partition_start(struct seq_file *seqf, loff_t *pos) { + struct partitions_seq_priv *priv = seqf->private; void *p; p = bpf_disk_seqf_start(seqf, pos); if (!IS_ERR_OR_NULL(p) && !*pos) seq_puts(seqf, "major minor #blocks name\n\n"); + + priv->dev_list_size = get_targeted_dev(&priv->dev_list); + return p; } @@ -1503,6 +1617,7 @@ static void __show_partition(struct seq_file *seqf, struct block_device *part) /* Inconvenient to operate Xarray in bpf progs. */ static int bpf_show_partition(struct seq_file *seqf, void *v) { + struct partitions_seq_priv *priv = seqf->private; struct gendisk *sgp = v; struct block_device *part; unsigned long idx; @@ -1511,8 +1626,11 @@ static int bpf_show_partition(struct seq_file *seqf, void *v) return 0; rcu_read_lock(); - xa_for_each(&sgp->part_tbl, idx, part) - __show_partition(seqf, part); + xa_for_each(&sgp->part_tbl, idx, part) { + for (int i = 0; i < priv->dev_list_size; ++i) + if (part == xa_load(&priv->dev_list, i)) + __show_partition(seqf, part); + } rcu_read_unlock(); return 0; } @@ -1528,7 +1646,7 @@ static const struct bpf_iter_seq_info partitions_seq_info = { .seq_ops = &bpf_partitions_op, .init_seq_private = NULL, .fini_seq_private = NULL, - .seq_priv_size = sizeof(struct class_dev_iter), + .seq_priv_size = sizeof(struct partitions_seq_priv), }; static struct bpf_iter_reg partitions_reg_info = { -- 2.25.1