Re: [PATCH v2 1/3] quota: add quota in-memory format support

[Christian Brauner <brauner@kernel.org>]

On Mon, Nov 21, 2022 at 03:28:52PM +0100, Lukas Czerner wrote:
> In memory quota format relies on quota infrastructure to store dquot
> information for us. While conventional quota formats for file systems
> with persistent storage can load quota information into dquot from the
> storage on-demand and hence quota dquot shrinker can free any dquot that
> is not currently being used, it must be avoided here. Otherwise we can
> lose valuable information, user provided limits, because there is no
> persistent storage to load the information from afterwards.
> 
> One information that in-memory quota format needs to keep track of is a
> sorted list of ids for each quota type. This is done by utilizing an rb
> tree which root is stored in mem_dqinfo->dqi_priv for each quota type.
> 
> This format can be used to support quota on file system without persistent
> storage such as tmpfs.
> 
> Signed-off-by: Lukas Czerner <lczerner@redhat.com>
> ---
>  fs/quota/Kconfig           |   8 ++
>  fs/quota/Makefile          |   1 +
>  fs/quota/dquot.c           |   3 +
>  fs/quota/quota_mem.c       | 260 +++++++++++++++++++++++++++++++++++++
>  include/linux/quota.h      |   7 +-
>  include/uapi/linux/quota.h |   1 +
>  6 files changed, 279 insertions(+), 1 deletion(-)
>  create mode 100644 fs/quota/quota_mem.c
> 
> diff --git a/fs/quota/Kconfig b/fs/quota/Kconfig
> index b59cd172b5f9..8ea9656ca37b 100644
> --- a/fs/quota/Kconfig
> +++ b/fs/quota/Kconfig
> @@ -67,6 +67,14 @@ config QFMT_V2
>  	  also supports 64-bit inode and block quota limits. If you need this
>  	  functionality say Y here.
>  
> +config QFMT_MEM
> +	tristate "Quota in-memory format support "
> +	depends on QUOTA
> +	help
> +	  This config option enables kernel support for in-memory quota
> +	  format support. Useful to support quota on file system without
> +	  permanent storage. If you need this functionality say Y here.
> +
>  config QUOTACTL
>  	bool
>  	default n
> diff --git a/fs/quota/Makefile b/fs/quota/Makefile
> index 9160639daffa..935be3f7b731 100644
> --- a/fs/quota/Makefile
> +++ b/fs/quota/Makefile
> @@ -5,3 +5,4 @@ obj-$(CONFIG_QFMT_V2)		+= quota_v2.o
>  obj-$(CONFIG_QUOTA_TREE)	+= quota_tree.o
>  obj-$(CONFIG_QUOTACTL)		+= quota.o kqid.o
>  obj-$(CONFIG_QUOTA_NETLINK_INTERFACE)	+= netlink.o
> +obj-$(CONFIG_QFMT_MEM)		+= quota_mem.o
> diff --git a/fs/quota/dquot.c b/fs/quota/dquot.c
> index 0427b44bfee5..f1a7a03632a2 100644
> --- a/fs/quota/dquot.c
> +++ b/fs/quota/dquot.c
> @@ -736,6 +736,9 @@ dqcache_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>  	spin_lock(&dq_list_lock);
>  	while (!list_empty(&free_dquots) && sc->nr_to_scan) {
>  		dquot = list_first_entry(&free_dquots, struct dquot, dq_free);
> +		if (test_bit(DQ_NO_SHRINK_B, &dquot->dq_flags) &&
> +		    !test_bit(DQ_FAKE_B, &dquot->dq_flags))
> +			continue;
>  		remove_dquot_hash(dquot);
>  		remove_free_dquot(dquot);
>  		remove_inuse(dquot);
> diff --git a/fs/quota/quota_mem.c b/fs/quota/quota_mem.c
> new file mode 100644
> index 000000000000..7d5e82122143
> --- /dev/null
> +++ b/fs/quota/quota_mem.c
> @@ -0,0 +1,260 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * In memory quota format relies on quota infrastructure to store dquot
> + * information for us. While conventional quota formats for file systems
> + * with persistent storage can load quota information into dquot from the
> + * storage on-demand and hence quota dquot shrinker can free any dquot
> + * that is not currently being used, it must be avoided here. Otherwise we
> + * can lose valuable information, user provided limits, because there is
> + * no persistent storage to load the information from afterwards.
> + *
> + * One information that in-memory quota format needs to keep track of is
> + * a sorted list of ids for each quota type. This is done by utilizing
> + * an rb tree which root is stored in mem_dqinfo->dqi_priv for each quota
> + * type.
> + *
> + * This format can be used to support quota on file system without persistent
> + * storage such as tmpfs.
> + */
> +#include <linux/errno.h>
> +#include <linux/fs.h>
> +#include <linux/mount.h>
> +#include <linux/kernel.h>
> +#include <linux/init.h>
> +#include <linux/module.h>
> +#include <linux/slab.h>
> +#include <linux/rbtree.h>
> +
> +#include <linux/quotaops.h>
> +#include <linux/quota.h>
> +
> +MODULE_AUTHOR("Lukas Czerner");
> +MODULE_DESCRIPTION("Quota in-memory format support");
> +MODULE_LICENSE("GPL");
> +
> +/*
> + * The following constants define the amount of time given a user
> + * before the soft limits are treated as hard limits (usually resulting
> + * in an allocation failure). The timer is started when the user crosses
> + * their soft limit, it is reset when they go below their soft limit.
> + */
> +#define MAX_IQ_TIME  604800	/* (7*24*60*60) 1 week */
> +#define MAX_DQ_TIME  604800	/* (7*24*60*60) 1 week */
> +
> +struct quota_id {
> +	struct rb_node	node;
> +	qid_t		id;
> +};
> +
> +static int mem_check_quota_file(struct super_block *sb, int type)
> +{
> +	/* There is no real quota file, nothing to do */
> +	return 1;
> +}
> +
> +/*
> + * There is no real quota file. Just allocate rb_root for quota ids and
> + * set limits
> + */
> +static int mem_read_file_info(struct super_block *sb, int type)
> +{
> +	struct quota_info *dqopt = sb_dqopt(sb);
> +	struct mem_dqinfo *info = &dqopt->info[type];
> +	int ret = 0;
> +
> +	down_read(&dqopt->dqio_sem);
> +	if (info->dqi_fmt_id != QFMT_MEM_ONLY) {
> +		ret = -EINVAL;
> +		goto out_unlock;
> +	}
> +
> +	info->dqi_priv = kzalloc(sizeof(struct rb_root), GFP_NOFS);
> +	if (!info->dqi_priv) {
> +		ret = -ENOMEM;
> +		goto out_unlock;
> +	}
> +
> +	/*
> +	 * Used space is stored as unsigned 64-bit value in bytes but
> +	 * quota core supports only signed 64-bit values so use that
> +	 * as a limit
> +	 */
> +	info->dqi_max_spc_limit = 0x7fffffffffffffffLL; /* 2^63-1 */
> +	info->dqi_max_ino_limit = 0x7fffffffffffffffLL;
> +
> +	info->dqi_bgrace = MAX_DQ_TIME;
> +	info->dqi_igrace = MAX_IQ_TIME;
> +	info->dqi_flags = 0;
> +
> +out_unlock:
> +	up_read(&dqopt->dqio_sem);
> +	return ret;
> +}
> +
> +static int mem_write_file_info(struct super_block *sb, int type)
> +{
> +	/* There is no real quota file, nothing to do */
> +	return 0;
> +}
> +
> +/*
> + * Free all the quota_id entries in the rb tree and rb_root.
> + */
> +static int mem_free_file_info(struct super_block *sb, int type)
> +{
> +	struct mem_dqinfo *info = &sb_dqopt(sb)->info[type];
> +	struct rb_root *root = info->dqi_priv;
> +	struct quota_id *entry;
> +	struct rb_node *node;
> +
> +	info->dqi_priv = NULL;
> +	node = rb_first(root);
> +	while (node) {
> +		entry = rb_entry(node, struct quota_id, node);
> +		node = rb_next(&entry->node);
> +
> +		rb_erase(&entry->node, root);
> +		kfree(entry);
> +	}
> +
> +	kfree(root);
> +	return 0;
> +}
> +
> +/*
> + * There is no real quota file, nothing to read. Just insert the id in
> + * the rb tree.
> + */
> +static int mem_read_dquot(struct dquot *dquot)
> +{
> +	struct mem_dqinfo *info = sb_dqinfo(dquot->dq_sb, dquot->dq_id.type);
> +	struct rb_node **n = &((struct rb_root *)info->dqi_priv)->rb_node;
> +	struct rb_node *parent = NULL, *new_node = NULL;
> +	struct quota_id *new_entry, *entry;
> +	qid_t id = from_kqid(&init_user_ns, dquot->dq_id);

Hey Lukas,

tmpfs instances can be mounted inside of mount namespaces owned by user
namespaces as is the case in unprivileged containers. An easy example is:

unshare --mount --user --map-root
mount -t tmpfs tmpfs /mnt

This tmpfs instances will be mounted with sb->s_user_ns set to the
userns just created during the unshare call and not to init_user_ns. So
this means that the filesystem idmapping isn't a 1:1 mapping. This needs
to be taken into account:

qid_t id = from_kqid(sb->s_user_ns, dquot->dq_id);

similar below.

But dquot_load_quota_sb() which you use in a later patch is restricted
to the init_user_ns which means that your patch as it stands is only
useable for tmpfs instances mounted in the init_user_ns.

If that's intentional then the code above is probably fine but if it's
not then you need preliminary patches to support quotas from filesystems
mountable in non-initial user namespaces.

Enabling this shouldn't be a big deal as it mostly involves updating
callsites to account for sb->s_user_ns when reading and writing quotas.
I've looked at that a while ago but there was no filesystem with quota
support that was also mountable in a user namespaces. Idmapped mounts
are already taken care of.