kern/drivers/dev/tmpfs.c

/* Copyright (c) 2018 Google Inc
 * Barret Rhoden <brho@cs.berkeley.edu>
 * See LICENSE for details.
 *
 * #tmpfs, in-memory ram filesystem instance.
 *
 * Compared to #kfs, there can be many #tmpfs instances, all of which are
 * independent and clean themeselves up when they are detached and the last chan
 * is closed.  A few notes:
 * - The FS root ("/") can't be removed.  When you unmount, it gets
 *   closed/decreffed.
 * - The tmpfs will exist for the life of any open chans, including the mount.
 *   You can e.g. cd into a subdir, unmount /, and the tmpfs will stay alive til
 *   you leave the directory (basically close the chan).  You can do the same
 *   thing with open_at().
 * - The root tree_file will not be deleted so long as you have an open chan.
 *   Any open chan on a subdir/subfile will hold refs on the root.  The mount
 *   point will also hold those refs.  We also hold an additional +1 on the root
 *   TF, which we drop once we have no users and we've purged the tree. */

#include <ns.h>
#include <kmalloc.h>
#include <string.h>
#include <stdio.h>
#include <assert.h>
#include <error.h>
#include <tree_file.h>
#include <pmap.h>
#include <cpio.h>

struct dev tmpfs_devtab;

/* This is a little wasteful - a u16_pool will work too, though I'd like to
 * exercise bits of the arena code. */
static struct arena *tmpfs_ids;

struct tmpfs {
	struct tree_filesystem		tfs;
	atomic_t			qid_file;
	uint16_t			id;
	struct kref			users;
};

static struct tmpfs *tree_file_to_tmpfs(struct tree_file *tf)
{
	return (struct tmpfs*)(tf->tfs);
}

#define TMPFS_MAX_QID_FILE ((1ULL << 48) - 1)
#define TMPFS_MAX_ID ((1 << 16) - 1)

/* Qids are made up of 16 bits for the ID, 48 for the file. */
static uint64_t tmpfs_get_qid_path(struct tmpfs *tmpfs)
{
	uint64_t ret = atomic_fetch_and_add(&tmpfs->qid_file, 1);

	/* Wrap-around of qid_file won't happen in our lifetime */
	if (ret > TMPFS_MAX_QID_FILE)
		error(ENOSPC, "out of tmpfs files");
	return ((uint64_t)tmpfs->id << 16) | ret;
}

static char *devname(void)
{
	return tmpfs_devtab.name;
}

static void tmpfs_tf_free(struct tree_file *tf)
{
	/* We have nothing special hanging off the TF */
}

static void tmpfs_tf_unlink(struct tree_file *parent, struct tree_file *child)
{
	/* This is the "+1 for existing" ref. */
	tf_kref_put(child);
}

static void __tmpfs_tf_init(struct tree_file *tf, int dir_type, int dir_dev,
                            struct username *user, int perm)
{
	struct dir *dir = &tf->file.dir;

	fs_file_init_dir(&tf->file, dir_type, dir_dev, user, perm);
	dir->qid.path = tmpfs_get_qid_path(tree_file_to_tmpfs(tf));
	dir->qid.vers = 0;
	/* This is the "+1 for existing" ref.  There is no backing store for the
	 * FS, such as a disk or 9p, so we can't get rid of a file until it is
	 * unlinked and decreffed.  Note that KFS doesn't use pruners or
	 * anything else. */
	__kref_get(&tf->kref, 1);
}

/* Note: If your TFS doesn't support symlinks, you need to error out */
static void tmpfs_tf_create(struct tree_file *parent, struct tree_file *child,
                            int perm)
{
	__tmpfs_tf_init(child, parent->file.dir.type, parent->file.dir.dev,
			&eve, perm);
}

static void tmpfs_tf_rename(struct tree_file *tf, struct tree_file *old_parent,
                            struct tree_file *new_parent, const char *name,
                            int flags)
{
	/* We don't have a backend, so we don't need to do anything additional
	 * for rename. */
}

static bool tmpfs_tf_has_children(struct tree_file *parent)
{
	/* The tree_file parent list is complete and not merely a cache for a
	 * real backend. */
	return !list_empty(&parent->children);
}

struct tree_file_ops tmpfs_tf_ops = {
	.free = tmpfs_tf_free,
	.unlink = tmpfs_tf_unlink,
	.lookup = NULL,
	.create = tmpfs_tf_create,
	.rename = tmpfs_tf_rename,
	.has_children = tmpfs_tf_has_children,
};

/* Fills page with its contents from its backing store file.  For KFS, that
 * means we're creating or extending a file, and the contents are 0.  Note the
 * page/offset might be beyond the current file length, based on the current
 * pagemap code. */
static int tmpfs_pm_readpage(struct page_map *pm, struct page *pg)
{
	memset(page2kva(pg), 0, PGSIZE);
	atomic_or(&pg->pg_flags, PG_UPTODATE);
	/* Pretend that we blocked while filing this page.  This catches a lot
	 * of bugs.  It does slightly slow down the kernel, but it's only when
	 * filling the page cache, and considering we are using a RAMFS, you
	 * shouldn't measure things that actually rely on KFS's performance. */
	kthread_usleep(1);
	return 0;
}

/* Meant to take the page from PM and flush to backing store.  There is no
 * backing store. */
static int tmpfs_pm_writepage(struct page_map *pm, struct page *pg)
{
	return 0;
}

static void tmpfs_fs_punch_hole(struct fs_file *f, off64_t begin, off64_t end)
{
}

static bool tmpfs_fs_can_grow_to(struct fs_file *f, size_t len)
{
	/* TODO: implement some sort of memory limit */
	return true;
}

struct fs_file_ops tmpfs_fs_ops = {
	.readpage = tmpfs_pm_readpage,
	.writepage = tmpfs_pm_writepage,
	.punch_hole = tmpfs_fs_punch_hole,
	.can_grow_to = tmpfs_fs_can_grow_to,
};

static void purge_cb(struct tree_file *tf)
{
	/* this is the +1 for existing */
	tf_kref_put(tf);
}

static void tmpfs_release(struct kref *kref)
{
	struct tmpfs *tmpfs = container_of(kref, struct tmpfs, users);

	tfs_frontend_purge(&tmpfs->tfs, purge_cb);
	/* this is the ref from attach */
	assert(kref_refcnt(&tmpfs->tfs.root->kref) == 1);
	tf_kref_put(tmpfs->tfs.root);
	/* ensures __tf_free() happens before tfs_destroy */
	rcu_barrier();
	tfs_destroy(&tmpfs->tfs);
	arena_free(tmpfs_ids, (void*)(uintptr_t)tmpfs->id, 1);
	kfree(tmpfs);
}

static struct tmpfs *chan_to_tmpfs(struct chan *c)
{
	return tree_file_to_tmpfs(chan_to_tree_file(c));
}

static void incref_tmpfs_chan(struct chan *c)
{
	kref_get(&chan_to_tmpfs(c)->users, 1);
}

static void decref_tmpfs_chan(struct chan *c)
{
	kref_put(&chan_to_tmpfs(c)->users);
}

static void tmpfs_reset(void)
{
	/* Note you can't make an arena that hands out '0'.  That's baked into
	 * the interface.  I'm OK with it for now.  Because of that, the *size*
	 * of the span is 2^16 - 1; can't count '0'. */
	tmpfs_ids = arena_create("tmpfs_ids", (void*)1, TMPFS_MAX_ID, 1,
				 NULL, NULL, NULL, 0, ARENA_NEXTFIT | MEM_WAIT);
}

/* This is pretty hokey - we never call shutdown, and I don't know if 9ns has
 * any protections about device use before calling shutdown. */
static void tmpfs_shutdown(void)
{
	arena_destroy(tmpfs_ids);
	tmpfs_ids = NULL;
}

static struct chan *tmpfs_attach(char *spec)
{
	struct tree_filesystem *tfs = kzmalloc(sizeof(struct tmpfs), MEM_WAIT);
	struct tmpfs *tmpfs = (struct tmpfs*)tfs;

	/* All distinct chans get a ref on the filesystem, so that we can
	 * destroy it when the last user disconnects/closes. */
	kref_init(&tmpfs->users, tmpfs_release, 1);
	tmpfs->id = (uint16_t)(uintptr_t)arena_alloc(tmpfs_ids, 1, MEM_WAIT);

	/* This gives us one ref on root, dropped during tmpfs_release(). */
	tfs_init(tfs);
	tfs->tf_ops = tmpfs_tf_ops;
	tfs->fs_ops = tmpfs_fs_ops;

	/* This gives us an extra refcnt on tfs->root.  This is "+1 for
	 * existing." It is decreffed during the purge CB. */
	__tmpfs_tf_init(tfs->root, &tmpfs_devtab - devtab, 0, &eve, DMDIR |
			0777);
	/* This also increfs, copying tfs->root's ref for the chan it returns.*/
	return tree_file_alloc_chan(tfs->root, &tmpfs_devtab, "#tmpfs");
}

static struct walkqid *tmpfs_walk(struct chan *c, struct chan *nc, char **name,
                                  unsigned int nname)
{
	struct walkqid *wq = tree_chan_walk(c, nc, name, nname);

	if (wq && wq->clone && (wq->clone != c))
		incref_tmpfs_chan(wq->clone);
	return wq;
}

static void tmpfs_close(struct chan *c)
{
	tree_chan_close(c);
	decref_tmpfs_chan(c);
}

static void tmpfs_remove(struct chan *c)
{
	ERRSTACK(1);
	struct tmpfs *tmpfs = chan_to_tmpfs(c);

	/* This is a bit of a pain - when remove fails, we won't get a chance to
	 * close the chan.  See notes in tree_chan_remove() and sysremove(). */
	if (waserror()) {
		kref_put(&tmpfs->users);
		nexterror();
	}
	tree_chan_remove(c);
	kref_put(&tmpfs->users);
	poperror();
}

void tmpfs_rename(struct chan *c, struct chan *new_p_c, const char *name,
                  int flags)
{
	struct tmpfs *tmpfs_old = chan_to_tmpfs(c);
	struct tmpfs *tmpfs_new = chan_to_tmpfs(new_p_c);

	/* namec checked that both were from the same device.  It is our
	 * responsibility to make sure they are the same version. */
	if (tmpfs_old != tmpfs_new)
		error(EXDEV, "can't rename across tmpfs instances");
	tree_chan_rename(c, new_p_c, name, flags);
}

static unsigned long tmpfs_chan_ctl(struct chan *c, int op, unsigned long a1,
                                    unsigned long a2, unsigned long a3,
                                    unsigned long a4)
{
	switch (op) {
	case CCTL_SYNC:
		return 0;
	default:
		return tree_chan_ctl(c, op, a1, a2, a3, a4);
	}
}

struct dev tmpfs_devtab __devtab = {
	.name = "tmpfs",
	.reset = tmpfs_reset,
	.init = devinit,
	.shutdown = tmpfs_shutdown,
	.attach = tmpfs_attach,
	.walk = tmpfs_walk,
	.stat = tree_chan_stat,
	.open = tree_chan_open,
	.create = tree_chan_create,
	.close = tmpfs_close,
	.read = tree_chan_read,
	.bread = devbread,
	.write = tree_chan_write,
	.bwrite = devbwrite,
	.remove = tmpfs_remove,
	.rename = tmpfs_rename,
	.wstat = tree_chan_wstat,
	.power = devpower,
	.chaninfo = devchaninfo,
	.mmap = tree_chan_mmap,
	.chan_ctl = tmpfs_chan_ctl,
};