Home | History | Annotate | Line # | Download | only in puffs
puffs_vfsops.c revision 1.32
      1  1.32    pooka /*	$NetBSD: puffs_vfsops.c,v 1.32 2007/03/29 16:30:07 pooka Exp $	*/
      2   1.1    pooka 
      3   1.1    pooka /*
      4   1.1    pooka  * Copyright (c) 2005, 2006  Antti Kantee.  All Rights Reserved.
      5   1.1    pooka  *
      6   1.1    pooka  * Development of this software was supported by the
      7   1.1    pooka  * Google Summer of Code program and the Ulla Tuominen Foundation.
      8   1.1    pooka  * The Google SoC project was mentored by Bill Studenmund.
      9   1.1    pooka  *
     10   1.1    pooka  * Redistribution and use in source and binary forms, with or without
     11   1.1    pooka  * modification, are permitted provided that the following conditions
     12   1.1    pooka  * are met:
     13   1.1    pooka  * 1. Redistributions of source code must retain the above copyright
     14   1.1    pooka  *    notice, this list of conditions and the following disclaimer.
     15   1.1    pooka  * 2. Redistributions in binary form must reproduce the above copyright
     16   1.1    pooka  *    notice, this list of conditions and the following disclaimer in the
     17   1.1    pooka  *    documentation and/or other materials provided with the distribution.
     18   1.1    pooka  * 3. The name of the company nor the name of the author may be used to
     19   1.1    pooka  *    endorse or promote products derived from this software without specific
     20   1.1    pooka  *    prior written permission.
     21   1.1    pooka  *
     22   1.1    pooka  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     23   1.1    pooka  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     24   1.1    pooka  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     25   1.1    pooka  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     26   1.1    pooka  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     27   1.1    pooka  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     28   1.1    pooka  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     29   1.1    pooka  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     30   1.1    pooka  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     31   1.1    pooka  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     32   1.1    pooka  * SUCH DAMAGE.
     33   1.1    pooka  */
     34   1.1    pooka 
     35   1.1    pooka #include <sys/cdefs.h>
     36  1.32    pooka __KERNEL_RCSID(0, "$NetBSD: puffs_vfsops.c,v 1.32 2007/03/29 16:30:07 pooka Exp $");
     37   1.1    pooka 
     38   1.1    pooka #include <sys/param.h>
     39   1.1    pooka #include <sys/mount.h>
     40   1.1    pooka #include <sys/malloc.h>
     41   1.1    pooka #include <sys/extattr.h>
     42   1.1    pooka #include <sys/queue.h>
     43   1.1    pooka #include <sys/vnode.h>
     44   1.1    pooka #include <sys/dirent.h>
     45   1.1    pooka #include <sys/kauth.h>
     46  1.26    pooka #include <sys/fstrans.h>
     47   1.1    pooka 
     48   1.1    pooka #include <lib/libkern/libkern.h>
     49   1.1    pooka 
     50   1.1    pooka #include <fs/puffs/puffs_msgif.h>
     51   1.1    pooka #include <fs/puffs/puffs_sys.h>
     52   1.1    pooka 
     53   1.1    pooka VFS_PROTOS(puffs);
     54   1.1    pooka 
     55  1.32    pooka MALLOC_JUSTDEFINE(M_PUFFS, "puffs", "Pass-to-Userspace Framework File System");
     56   1.1    pooka 
     57  1.22    pooka #ifndef PUFFS_PNODEBUCKETS
     58  1.22    pooka #define PUFFS_PNODEBUCKETS 256
     59  1.22    pooka #endif
     60  1.22    pooka #ifndef PUFFS_MAXPNODEBUCKETS
     61  1.22    pooka #define PUFFS_MAXPNODEBUCKETS 65536
     62  1.22    pooka #endif
     63  1.22    pooka int puffs_pnodebuckets = PUFFS_PNODEBUCKETS;
     64  1.22    pooka 
     65   1.1    pooka int
     66   1.1    pooka puffs_mount(struct mount *mp, const char *path, void *data,
     67   1.1    pooka 	    struct nameidata *ndp, struct lwp *l)
     68   1.1    pooka {
     69  1.22    pooka 	struct puffs_mount *pmp = NULL;
     70  1.17    pooka 	struct puffs_args *args;
     71  1.17    pooka 	char namebuf[PUFFSNAMESIZE+sizeof(PUFFS_NAMEPREFIX)+1]; /* spooky */
     72  1.22    pooka 	int error = 0, i;
     73   1.1    pooka 
     74   1.1    pooka 	if (mp->mnt_flag & MNT_GETARGS) {
     75   1.1    pooka 		pmp = MPTOPUFFSMP(mp);
     76   1.1    pooka 		return copyout(&pmp->pmp_args, data, sizeof(struct puffs_args));
     77   1.1    pooka 	}
     78   1.1    pooka 
     79   1.1    pooka 	/* update is not supported currently */
     80   1.1    pooka 	if (mp->mnt_flag & MNT_UPDATE)
     81   1.1    pooka 		return EOPNOTSUPP;
     82   1.1    pooka 
     83   1.1    pooka 	/*
     84   1.1    pooka 	 * We need the file system name
     85   1.1    pooka 	 */
     86   1.1    pooka 	if (!data)
     87   1.1    pooka 		return EINVAL;
     88   1.1    pooka 
     89  1.17    pooka 	MALLOC(args, struct puffs_args *, sizeof(struct puffs_args),
     90  1.17    pooka 	    M_PUFFS, M_WAITOK);
     91  1.17    pooka 
     92  1.17    pooka 	error = copyin(data, args, sizeof(struct puffs_args));
     93   1.1    pooka 	if (error)
     94  1.17    pooka 		goto out;
     95  1.17    pooka 
     96  1.17    pooka 	/* devel phase */
     97  1.17    pooka 	if (args->pa_vers != (PUFFSVERSION | PUFFSDEVELVERS)) {
     98  1.17    pooka 		printf("puffs_mount: development version mismatch\n");
     99  1.17    pooka 		error = EINVAL;
    100  1.17    pooka 		goto out;
    101  1.17    pooka 	}
    102   1.1    pooka 
    103   1.8    pooka 	/* nuke spy bits */
    104  1.17    pooka 	args->pa_flags &= PUFFS_KFLAG_MASK;
    105   1.8    pooka 
    106   1.1    pooka 	/* build real name */
    107  1.17    pooka 	(void)strlcpy(namebuf, PUFFS_NAMEPREFIX, sizeof(namebuf));
    108  1.17    pooka 	(void)strlcat(namebuf, args->pa_name, sizeof(namebuf));
    109   1.1    pooka 
    110   1.1    pooka 	/* inform user server if it got the max request size it wanted */
    111  1.17    pooka 	if (args->pa_maxreqlen == 0 || args->pa_maxreqlen > PUFFS_REQ_MAXSIZE)
    112  1.17    pooka 		args->pa_maxreqlen = PUFFS_REQ_MAXSIZE;
    113  1.17    pooka 	else if (args->pa_maxreqlen < PUFFS_REQSTRUCT_MAX)
    114  1.17    pooka 		args->pa_maxreqlen = PUFFS_REQSTRUCT_MAX;
    115  1.17    pooka 	(void)strlcpy(args->pa_name, namebuf, sizeof(args->pa_name));
    116   1.1    pooka 
    117  1.17    pooka 	error = copyout(args, data, sizeof(struct puffs_args));
    118   1.1    pooka 	if (error)
    119  1.17    pooka 		goto out;
    120   1.1    pooka 
    121   1.1    pooka 	error = set_statvfs_info(path, UIO_USERSPACE, namebuf,
    122   1.1    pooka 	    UIO_SYSSPACE, mp, l);
    123   1.1    pooka 	if (error)
    124  1.17    pooka 		goto out;
    125  1.10    pooka 	mp->mnt_stat.f_iosize = DEV_BSIZE;
    126   1.1    pooka 
    127   1.1    pooka 	MALLOC(pmp, struct puffs_mount *, sizeof(struct puffs_mount),
    128   1.1    pooka 	    M_PUFFS, M_WAITOK | M_ZERO);
    129   1.1    pooka 
    130   1.6    pooka 	mp->mnt_fs_bshift = DEV_BSHIFT;
    131   1.6    pooka 	mp->mnt_dev_bshift = DEV_BSHIFT;
    132   1.6    pooka 	mp->mnt_flag &= ~MNT_LOCAL; /* we don't really know, so ... */
    133   1.1    pooka 	mp->mnt_data = pmp;
    134  1.26    pooka 	mp->mnt_iflag |= IMNT_HAS_TRANS;
    135   1.6    pooka 
    136   1.1    pooka 	pmp->pmp_status = PUFFSTAT_MOUNTING;
    137   1.1    pooka 	pmp->pmp_nextreq = 0;
    138   1.1    pooka 	pmp->pmp_mp = mp;
    139  1.17    pooka 	pmp->pmp_req_maxsize = args->pa_maxreqlen;
    140  1.17    pooka 	pmp->pmp_args = *args;
    141   1.1    pooka 
    142  1.22    pooka 	/* puffs_node hash buckets */
    143  1.22    pooka 	pmp->pmp_npnodehash = puffs_pnodebuckets;
    144  1.22    pooka 	if (pmp->pmp_npnodehash < 1)
    145  1.22    pooka 		pmp->pmp_npnodehash = 1;
    146  1.22    pooka 	if (pmp->pmp_npnodehash > PUFFS_MAXPNODEBUCKETS)
    147  1.22    pooka 		pmp->pmp_npnodehash = PUFFS_MAXPNODEBUCKETS;
    148  1.22    pooka 	pmp->pmp_pnodehash = malloc
    149  1.22    pooka 	    (sizeof(struct puffs_pnode_hashlist *) * pmp->pmp_npnodehash,
    150  1.22    pooka 	    M_PUFFS, M_WAITOK);
    151  1.22    pooka 	for (i = 0; i < pmp->pmp_npnodehash; i++)
    152  1.22    pooka 		LIST_INIT(&pmp->pmp_pnodehash[i]);
    153  1.22    pooka 
    154   1.1    pooka 	/*
    155   1.1    pooka 	 * Inform the fileops processing code that we have a mountpoint.
    156   1.1    pooka 	 * If it doesn't know about anyone with our pid/fd having the
    157   1.1    pooka 	 * device open, punt
    158   1.1    pooka 	 */
    159  1.17    pooka 	if (puffs_setpmp(l->l_proc->p_pid, args->pa_fd, pmp)) {
    160  1.17    pooka 		error = ENOENT;
    161  1.17    pooka 		goto out;
    162   1.1    pooka 	}
    163   1.1    pooka 
    164  1.31    pooka 	mutex_init(&pmp->pmp_lock, MUTEX_DEFAULT, IPL_NONE);
    165  1.31    pooka 	cv_init(&pmp->pmp_req_waiter_cv, "puffsget");
    166  1.31    pooka 	cv_init(&pmp->pmp_req_waitersink_cv, "puffsink");
    167  1.31    pooka 	cv_init(&pmp->pmp_unmounting_cv, "puffsum");
    168  1.31    pooka 	cv_init(&pmp->pmp_suspend_cv, "pufsusum");
    169   1.1    pooka 	TAILQ_INIT(&pmp->pmp_req_touser);
    170   1.1    pooka 	TAILQ_INIT(&pmp->pmp_req_replywait);
    171   1.1    pooka 	TAILQ_INIT(&pmp->pmp_req_sizepark);
    172   1.1    pooka 
    173   1.1    pooka 	DPRINTF(("puffs_mount: mount point at %p, puffs specific at %p\n",
    174   1.1    pooka 	    mp, MPTOPUFFSMP(mp)));
    175   1.1    pooka 
    176   1.1    pooka 	vfs_getnewfsid(mp);
    177   1.1    pooka 
    178  1.17    pooka  out:
    179  1.22    pooka 	if (error && pmp && pmp->pmp_pnodehash)
    180  1.22    pooka 		free(pmp->pmp_pnodehash, M_PUFFS);
    181  1.22    pooka 	if (error && pmp)
    182  1.22    pooka 		FREE(pmp, M_PUFFS);
    183  1.17    pooka 	FREE(args, M_PUFFS);
    184  1.17    pooka 	return error;
    185   1.1    pooka }
    186   1.1    pooka 
    187   1.1    pooka /*
    188   1.1    pooka  * This is called from the first "Hello, I'm alive" ioctl
    189   1.1    pooka  * from userspace.
    190   1.1    pooka  */
    191   1.1    pooka int
    192   1.7    pooka puffs_start2(struct puffs_mount *pmp, struct puffs_startreq *sreq)
    193   1.1    pooka {
    194   1.1    pooka 	struct puffs_node *pn;
    195   1.1    pooka 	struct mount *mp;
    196   1.1    pooka 
    197   1.1    pooka 	mp = PMPTOMP(pmp);
    198   1.1    pooka 
    199  1.31    pooka 	mutex_enter(&pmp->pmp_lock);
    200   1.1    pooka 
    201   1.1    pooka 	/*
    202   1.1    pooka 	 * if someone has issued a VFS_ROOT() already, fill in the
    203   1.1    pooka 	 * vnode cookie.
    204   1.1    pooka 	 */
    205   1.2    pooka 	pn = NULL;
    206   1.1    pooka 	if (pmp->pmp_root) {
    207   1.1    pooka 		pn = VPTOPP(pmp->pmp_root);
    208   1.1    pooka 		pn->pn_cookie = sreq->psr_cookie;
    209   1.1    pooka 	}
    210   1.1    pooka 
    211   1.1    pooka 	/* We're good to fly */
    212   1.1    pooka 	pmp->pmp_rootcookie = sreq->psr_cookie;
    213   1.1    pooka 	pmp->pmp_status = PUFFSTAT_RUNNING;
    214  1.31    pooka 	mutex_exit(&pmp->pmp_lock);
    215   1.1    pooka 
    216   1.9    pooka 	/* do the VFS_STATVFS() we missed out on in sys_mount() */
    217   1.9    pooka 	copy_statvfs_info(&sreq->psr_sb, mp);
    218   1.9    pooka 	(void)memcpy(&mp->mnt_stat, &sreq->psr_sb, sizeof(mp->mnt_stat));
    219  1.11    pooka 	mp->mnt_stat.f_iosize = DEV_BSIZE;
    220   1.1    pooka 
    221   1.2    pooka 	DPRINTF(("puffs_start2: root vp %p, cur root pnode %p, cookie %p\n",
    222   1.1    pooka 	    pmp->pmp_root, pn, sreq->psr_cookie));
    223   1.1    pooka 
    224   1.1    pooka 	return 0;
    225   1.1    pooka }
    226   1.1    pooka 
    227   1.1    pooka int
    228   1.1    pooka puffs_start(struct mount *mp, int flags, struct lwp *l)
    229   1.1    pooka {
    230   1.1    pooka 
    231   1.1    pooka 	/*
    232   1.1    pooka 	 * This cannot travel to userspace, as this is called from
    233   1.1    pooka 	 * the kernel context of the process doing mount(2).  But
    234   1.1    pooka 	 * it's probably a safe bet that the process doing mount(2)
    235   1.1    pooka 	 * realizes it needs to start the filesystem also...
    236   1.1    pooka 	 */
    237   1.1    pooka 	return 0;
    238   1.1    pooka }
    239   1.1    pooka 
    240   1.1    pooka int
    241   1.1    pooka puffs_unmount(struct mount *mp, int mntflags, struct lwp *l)
    242   1.1    pooka {
    243   1.1    pooka 	struct puffs_mount *pmp;
    244   1.1    pooka 	int error, force;
    245   1.1    pooka 
    246   1.1    pooka 	PUFFS_VFSREQ(unmount);
    247   1.1    pooka 
    248   1.1    pooka 	error = 0;
    249   1.1    pooka 	force = mntflags & MNT_FORCE;
    250   1.1    pooka 	pmp = MPTOPUFFSMP(mp);
    251   1.1    pooka 
    252   1.1    pooka 	DPRINTF(("puffs_unmount: detach filesystem from vfs, current "
    253   1.1    pooka 	    "status 0x%x\n", pmp->pmp_status));
    254   1.1    pooka 
    255   1.1    pooka 	/*
    256   1.1    pooka 	 * flush all the vnodes.  VOP_RECLAIM() takes care that the
    257   1.1    pooka 	 * root vnode does not get flushed until unmount.  The
    258   1.1    pooka 	 * userspace root node cookie is stored in the mount
    259   1.1    pooka 	 * structure, so we can always re-instantiate a root vnode,
    260   1.1    pooka 	 * should userspace unmount decide it doesn't want to
    261   1.1    pooka 	 * cooperate.
    262   1.1    pooka 	 */
    263   1.1    pooka 	error = vflush(mp, NULLVP, force ? FORCECLOSE : 0);
    264   1.1    pooka 	if (error)
    265   1.1    pooka 		goto out;
    266   1.1    pooka 
    267   1.1    pooka 	/*
    268   1.1    pooka 	 * If we are not DYING, we should ask userspace's opinion
    269   1.1    pooka 	 * about the situation
    270   1.1    pooka 	 */
    271  1.31    pooka 	mutex_enter(&pmp->pmp_lock);
    272   1.1    pooka 	if (pmp->pmp_status != PUFFSTAT_DYING) {
    273  1.16    pooka 		pmp->pmp_unmounting = 1;
    274  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    275  1.16    pooka 
    276   1.1    pooka 		unmount_arg.pvfsr_flags = mntflags;
    277   1.1    pooka 		unmount_arg.pvfsr_pid = puffs_lwp2pid(l);
    278  1.16    pooka 
    279   1.1    pooka 		error = puffs_vfstouser(pmp, PUFFS_VFS_UNMOUNT,
    280   1.1    pooka 		     &unmount_arg, sizeof(unmount_arg));
    281  1.16    pooka 		DPRINTF(("puffs_unmount: error %d force %d\n", error, force));
    282  1.16    pooka 
    283  1.31    pooka 		mutex_enter(&pmp->pmp_lock);
    284  1.16    pooka 		pmp->pmp_unmounting = 0;
    285  1.31    pooka 		cv_broadcast(&pmp->pmp_unmounting_cv);
    286   1.1    pooka 	}
    287   1.1    pooka 
    288   1.1    pooka 	/*
    289   1.1    pooka 	 * if userspace cooperated or we really need to die,
    290   1.1    pooka 	 * screw what userland thinks and just die.
    291   1.1    pooka 	 */
    292   1.1    pooka 	if (error == 0 || force) {
    293  1.26    pooka 		/* tell waiters & other resources to go unwait themselves */
    294  1.26    pooka 		puffs_userdead(pmp);
    295   1.1    pooka 		puffs_nukebypmp(pmp);
    296  1.26    pooka 
    297  1.26    pooka 		/*
    298  1.26    pooka 		 * Sink waiters.  This is still not perfect, since the
    299  1.26    pooka 		 * draining is done after userret, not when they really
    300  1.26    pooka 		 * exit the file system.  It will probably work as almost
    301  1.26    pooka 		 * no call will block and therefore cause a context switch
    302  1.26    pooka 		 * and therefore will protected by the biglock after
    303  1.26    pooka 		 * exiting userspace.  But ... it's an imperfect world.
    304  1.26    pooka 		 */
    305  1.31    pooka 		while (pmp->pmp_req_waiters != 0)
    306  1.31    pooka 			cv_wait(&pmp->pmp_req_waitersink_cv, &pmp->pmp_lock);
    307  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    308  1.26    pooka 
    309  1.26    pooka 		/* free resources now that we hopefully have no waiters left */
    310  1.31    pooka 		cv_destroy(&pmp->pmp_req_waiter_cv);
    311  1.31    pooka 		cv_destroy(&pmp->pmp_req_waitersink_cv);
    312  1.31    pooka 		cv_destroy(&pmp->pmp_unmounting_cv);
    313  1.31    pooka 		cv_destroy(&pmp->pmp_suspend_cv);
    314  1.31    pooka 		mutex_destroy(&pmp->pmp_lock);
    315  1.31    pooka 
    316  1.22    pooka 		free(pmp->pmp_pnodehash, M_PUFFS);
    317   1.1    pooka 		FREE(pmp, M_PUFFS);
    318   1.1    pooka 		error = 0;
    319  1.16    pooka 	} else {
    320  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    321   1.1    pooka 	}
    322   1.1    pooka 
    323   1.1    pooka  out:
    324   1.2    pooka 	DPRINTF(("puffs_unmount: return %d\n", error));
    325   1.1    pooka 	return error;
    326   1.1    pooka }
    327   1.1    pooka 
    328   1.1    pooka /*
    329   1.1    pooka  * This doesn't need to travel to userspace
    330   1.1    pooka  */
    331   1.1    pooka int
    332   1.1    pooka puffs_root(struct mount *mp, struct vnode **vpp)
    333   1.1    pooka {
    334   1.1    pooka 	struct puffs_mount *pmp;
    335   1.1    pooka 	struct puffs_node *pn;
    336   1.1    pooka 	struct vnode *vp;
    337   1.1    pooka 
    338   1.1    pooka 	pmp = MPTOPUFFSMP(mp);
    339   1.1    pooka 
    340   1.1    pooka 	/*
    341   1.1    pooka 	 * pmp_lock must be held if vref()'ing or vrele()'ing the
    342  1.20    pooka 	 * root vnode.  the latter is controlled by puffs_inactive().
    343   1.1    pooka 	 */
    344  1.31    pooka 	mutex_enter(&pmp->pmp_lock);
    345   1.1    pooka 	vp = pmp->pmp_root;
    346   1.1    pooka 	if (vp) {
    347  1.25    pooka 		simple_lock(&vp->v_interlock);
    348  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    349   1.1    pooka 		pn = VPTOPP(vp);
    350  1.25    pooka 		if (vget(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK))
    351  1.20    pooka 			goto grabnew;
    352   1.1    pooka 		*vpp = vp;
    353   1.1    pooka 		return 0;
    354  1.25    pooka 	} else
    355  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    356  1.25    pooka 
    357  1.25    pooka 	/* XXX: this is wrong, so FIXME */
    358   1.1    pooka  grabnew:
    359   1.1    pooka 
    360   1.1    pooka 	/*
    361   1.1    pooka 	 * So, didn't have the magic root vnode available.
    362   1.1    pooka 	 * No matter, grab another an stuff it with the cookie.
    363   1.1    pooka 	 */
    364   1.6    pooka 	if (puffs_getvnode(mp, pmp->pmp_rootcookie, VDIR, 0, 0, &vp))
    365   1.1    pooka 		panic("sloppy programming");
    366   1.1    pooka 
    367  1.31    pooka 	mutex_enter(&pmp->pmp_lock);
    368   1.1    pooka 	/*
    369   1.1    pooka 	 * check if by mysterious force someone else created a root
    370   1.1    pooka 	 * vnode while we were executing.
    371   1.1    pooka 	 */
    372   1.1    pooka 	if (pmp->pmp_root) {
    373   1.1    pooka 		vref(pmp->pmp_root);
    374  1.31    pooka 		mutex_exit(&pmp->pmp_lock);
    375   1.1    pooka 		puffs_putvnode(vp);
    376   1.1    pooka 		vn_lock(pmp->pmp_root, LK_EXCLUSIVE | LK_RETRY);
    377   1.1    pooka 		*vpp = pmp->pmp_root;
    378   1.1    pooka 		return 0;
    379   1.1    pooka 	}
    380   1.1    pooka 
    381   1.1    pooka 	/* store cache */
    382   1.1    pooka 	vp->v_flag = VROOT;
    383   1.1    pooka 	pmp->pmp_root = vp;
    384  1.31    pooka 	mutex_exit(&pmp->pmp_lock);
    385   1.1    pooka 
    386   1.1    pooka 	vn_lock(pmp->pmp_root, LK_EXCLUSIVE | LK_RETRY);
    387   1.1    pooka 
    388   1.1    pooka 	*vpp = vp;
    389   1.1    pooka 	return 0;
    390   1.1    pooka }
    391   1.1    pooka 
    392   1.1    pooka int
    393   1.1    pooka puffs_quotactl(struct mount *mp, int cmd, uid_t uid, void *arg, struct lwp *l)
    394   1.1    pooka {
    395   1.1    pooka 
    396   1.1    pooka 	return EOPNOTSUPP;
    397   1.1    pooka }
    398   1.1    pooka 
    399   1.1    pooka int
    400   1.1    pooka puffs_statvfs(struct mount *mp, struct statvfs *sbp, struct lwp *l)
    401   1.1    pooka {
    402   1.1    pooka 	struct puffs_vfsreq_statvfs *statvfs_arg; /* too big for stack */
    403   1.1    pooka 	struct puffs_mount *pmp;
    404   1.1    pooka 	int error = 0;
    405   1.1    pooka 
    406   1.1    pooka 	pmp = MPTOPUFFSMP(mp);
    407   1.1    pooka 
    408   1.1    pooka 	/*
    409   1.1    pooka 	 * If we are mounting, it means that the userspace counterpart
    410   1.1    pooka 	 * is calling mount(2), but mount(2) also calls statvfs.  So
    411   1.1    pooka 	 * requesting statvfs from userspace would mean a deadlock.
    412   1.1    pooka 	 * Compensate.
    413   1.1    pooka 	 */
    414   1.1    pooka 	if (pmp->pmp_status == PUFFSTAT_MOUNTING)
    415   1.1    pooka 		return EINPROGRESS;
    416   1.1    pooka 
    417   1.1    pooka 	/* too big for stack */
    418   1.1    pooka 	MALLOC(statvfs_arg, struct puffs_vfsreq_statvfs *,
    419   1.1    pooka 	    sizeof(struct puffs_vfsreq_statvfs), M_PUFFS, M_WAITOK | M_ZERO);
    420   1.1    pooka 	statvfs_arg->pvfsr_pid = puffs_lwp2pid(l);
    421   1.1    pooka 
    422   1.1    pooka 	error = puffs_vfstouser(pmp, PUFFS_VFS_STATVFS,
    423   1.1    pooka 	    statvfs_arg, sizeof(*statvfs_arg));
    424  1.11    pooka 	statvfs_arg->pvfsr_sb.f_iosize = DEV_BSIZE;
    425  1.10    pooka 
    426   1.1    pooka 	/*
    427   1.1    pooka 	 * Try to produce a sensible result even in the event
    428   1.1    pooka 	 * of userspace error.
    429   1.1    pooka 	 *
    430   1.1    pooka 	 * XXX: cache the copy in non-error case
    431   1.1    pooka 	 */
    432   1.1    pooka 	if (!error) {
    433   1.1    pooka 		copy_statvfs_info(&statvfs_arg->pvfsr_sb, mp);
    434   1.1    pooka 		(void)memcpy(sbp, &statvfs_arg->pvfsr_sb,
    435   1.1    pooka 		    sizeof(struct statvfs));
    436   1.1    pooka 	} else {
    437   1.1    pooka 		copy_statvfs_info(sbp, mp);
    438   1.1    pooka 	}
    439   1.1    pooka 
    440   1.1    pooka 	FREE(statvfs_arg, M_PUFFS);
    441  1.14    pooka 	return error;
    442   1.1    pooka }
    443   1.1    pooka 
    444  1.26    pooka static int
    445  1.30    pooka pageflush(struct mount *mp, kauth_cred_t cred,
    446  1.30    pooka 	int waitfor, int suspending, struct lwp *l)
    447   1.1    pooka {
    448  1.26    pooka 	struct puffs_node *pn;
    449  1.18    pooka 	struct vnode *vp, *nvp;
    450  1.30    pooka 	int error, rv;
    451   1.1    pooka 
    452  1.26    pooka 	KASSERT(((waitfor == MNT_WAIT) && suspending) == 0);
    453  1.26    pooka 	KASSERT((suspending == 0)
    454  1.26    pooka 	    || (fstrans_is_owner(mp)
    455  1.27  hannken 	      && fstrans_getstate(mp) == FSTRANS_SUSPENDING));
    456   1.1    pooka 
    457  1.18    pooka 	error = 0;
    458  1.18    pooka 
    459  1.18    pooka 	/*
    460  1.24    pooka 	 * Sync all cached data from regular vnodes (which are not
    461  1.24    pooka 	 * currently locked, see below).  After this we call VFS_SYNC
    462  1.24    pooka 	 * for the fs server, which should handle data and metadata for
    463  1.24    pooka 	 * all the nodes it knows to exist.
    464  1.18    pooka 	 */
    465  1.18    pooka 	simple_lock(&mntvnode_slock);
    466  1.18    pooka  loop:
    467  1.18    pooka 	for (vp = TAILQ_FIRST(&mp->mnt_vnodelist); vp; vp = nvp) {
    468  1.18    pooka 		/* check if we're on the right list */
    469  1.18    pooka 		if (vp->v_mount != mp)
    470  1.18    pooka 			goto loop;
    471  1.18    pooka 
    472  1.18    pooka 		simple_lock(&vp->v_interlock);
    473  1.26    pooka 		pn = VPTOPP(vp);
    474  1.18    pooka 		nvp = TAILQ_NEXT(vp, v_mntvnodes);
    475  1.18    pooka 
    476  1.19    pooka 		if (vp->v_type != VREG || UVM_OBJ_IS_CLEAN(&vp->v_uobj)) {
    477  1.18    pooka 			simple_unlock(&vp->v_interlock);
    478  1.18    pooka 			continue;
    479  1.18    pooka 		}
    480  1.18    pooka 
    481  1.18    pooka 		simple_unlock(&mntvnode_slock);
    482  1.21    pooka 
    483  1.21    pooka 		/*
    484  1.21    pooka 		 * Here we try to get a reference to the vnode and to
    485  1.21    pooka 		 * lock it.  This is mostly cargo-culted, but I will
    486  1.21    pooka 		 * offer an explanation to why I believe this might
    487  1.21    pooka 		 * actually do the right thing.
    488  1.21    pooka 		 *
    489  1.21    pooka 		 * If the vnode is a goner, we quite obviously don't need
    490  1.21    pooka 		 * to sync it.
    491  1.21    pooka 		 *
    492  1.21    pooka 		 * If the vnode was busy, we don't need to sync it because
    493  1.21    pooka 		 * this is never called with MNT_WAIT except from
    494  1.21    pooka 		 * dounmount(), when we are wait-flushing all the dirty
    495  1.21    pooka 		 * vnodes through other routes in any case.  So there,
    496  1.21    pooka 		 * sync() doesn't actually sync.  Happy now?
    497  1.26    pooka 		 *
    498  1.26    pooka 		 * NOTE: if we're suspending, vget() does NOT lock.
    499  1.26    pooka 		 * See puffs_lock() for details.
    500  1.21    pooka 		 */
    501  1.18    pooka 		rv = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK);
    502  1.18    pooka 		if (rv) {
    503  1.18    pooka 			simple_lock(&mntvnode_slock);
    504  1.18    pooka 			if (rv == ENOENT)
    505  1.18    pooka 				goto loop;
    506  1.18    pooka 			continue;
    507  1.18    pooka 		}
    508  1.18    pooka 
    509  1.26    pooka 		/*
    510  1.26    pooka 		 * Thread information to puffs_strategy() through the
    511  1.26    pooka 		 * pnode flags: we want to issue the putpages operations
    512  1.26    pooka 		 * as FAF if we're suspending, since it's very probable
    513  1.26    pooka 		 * that our execution context is that of the userspace
    514  1.26    pooka 		 * daemon.  We can do this because:
    515  1.26    pooka 		 *   + we send the "going to suspend" prior to this part
    516  1.26    pooka 		 *   + if any of the writes fails in userspace, it's the
    517  1.26    pooka 		 *     file system server's problem to decide if this was a
    518  1.26    pooka 		 *     failed snapshot when it gets the "snapshot complete"
    519  1.26    pooka 		 *     notification.
    520  1.26    pooka 		 *   + if any of the writes fail in the kernel already, we
    521  1.26    pooka 		 *     immediately fail *and* notify the user server of
    522  1.26    pooka 		 *     failure.
    523  1.26    pooka 		 *
    524  1.26    pooka 		 * We also do FAFs if we're called from the syncer.  This
    525  1.26    pooka 		 * is just general optimization for trickle sync: no need
    526  1.26    pooka 		 * to really guarantee that the stuff ended on backing
    527  1.26    pooka 		 * storage.
    528  1.26    pooka 		 * TODO: Maybe also hint the user server of this twist?
    529  1.26    pooka 		 */
    530  1.30    pooka 		if (suspending || waitfor == MNT_LAZY) {
    531  1.30    pooka 			simple_lock(&vp->v_interlock);
    532  1.26    pooka 			pn->pn_stat |= PNODE_SUSPEND;
    533  1.30    pooka 			simple_unlock(&vp->v_interlock);
    534  1.30    pooka 		}
    535  1.30    pooka 		rv = VOP_FSYNC(vp, cred, waitfor, 0, 0, l);
    536  1.26    pooka 		if (suspending || waitfor == MNT_LAZY) {
    537  1.26    pooka 			simple_lock(&vp->v_interlock);
    538  1.26    pooka 			pn->pn_stat &= ~PNODE_SUSPEND;
    539  1.26    pooka 			simple_unlock(&vp->v_interlock);
    540  1.26    pooka 		}
    541  1.21    pooka 		if (rv)
    542  1.18    pooka 			error = rv;
    543  1.18    pooka 		vput(vp);
    544  1.18    pooka 		simple_lock(&mntvnode_slock);
    545  1.18    pooka 	}
    546  1.18    pooka 	simple_unlock(&mntvnode_slock);
    547  1.18    pooka 
    548  1.26    pooka 	return error;
    549  1.26    pooka }
    550  1.26    pooka 
    551  1.26    pooka int
    552  1.26    pooka puffs_sync(struct mount *mp, int waitfor, struct kauth_cred *cred,
    553  1.26    pooka 	struct lwp *l)
    554  1.26    pooka {
    555  1.26    pooka 	int error, rv;
    556  1.26    pooka 
    557  1.26    pooka 	PUFFS_VFSREQ(sync);
    558  1.26    pooka 
    559  1.30    pooka 	error = pageflush(mp, cred, waitfor, 0, l);
    560  1.26    pooka 
    561  1.18    pooka 	/* sync fs */
    562   1.1    pooka 	sync_arg.pvfsr_waitfor = waitfor;
    563   1.1    pooka 	puffs_credcvt(&sync_arg.pvfsr_cred, cred);
    564   1.1    pooka 	sync_arg.pvfsr_pid = puffs_lwp2pid(l);
    565   1.1    pooka 
    566  1.18    pooka 	rv = puffs_vfstouser(MPTOPUFFSMP(mp), PUFFS_VFS_SYNC,
    567   1.1    pooka 	    &sync_arg, sizeof(sync_arg));
    568  1.18    pooka 	if (rv)
    569  1.18    pooka 		error = rv;
    570   1.1    pooka 
    571   1.1    pooka 	return error;
    572   1.1    pooka }
    573   1.1    pooka 
    574   1.1    pooka int
    575   1.1    pooka puffs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
    576   1.1    pooka {
    577   1.1    pooka 
    578   1.1    pooka 	return EOPNOTSUPP;
    579   1.1    pooka }
    580   1.1    pooka 
    581   1.1    pooka #if 0
    582   1.1    pooka /*ARGSUSED*/
    583   1.1    pooka int
    584   1.1    pooka puffs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
    585   1.1    pooka {
    586   1.1    pooka 
    587   1.1    pooka 	return EOPNOTSUPP;
    588   1.1    pooka }
    589   1.1    pooka 
    590   1.1    pooka /*ARGSUSED*/
    591   1.1    pooka int
    592   1.1    pooka puffs_vptofh(struct vnode *vp, struct fid *fhp)
    593   1.1    pooka {
    594   1.1    pooka 
    595   1.1    pooka 	return EOPNOTSUPP;
    596   1.1    pooka }
    597   1.1    pooka #endif
    598   1.1    pooka 
    599   1.1    pooka void
    600   1.1    pooka puffs_init()
    601   1.1    pooka {
    602   1.1    pooka 
    603   1.5    pooka 	malloc_type_attach(M_PUFFS);
    604   1.5    pooka 
    605  1.31    pooka 	pool_init(&puffs_pnpool, sizeof(struct puffs_node), 0, 0, 0,
    606  1.31    pooka 	    "puffpnpl", &pool_allocator_nointr, IPL_NONE);
    607  1.31    pooka 	puffs_transport_init();
    608  1.31    pooka 	puffs_msgif_init();
    609   1.1    pooka }
    610   1.1    pooka 
    611   1.1    pooka void
    612   1.1    pooka puffs_done()
    613   1.1    pooka {
    614   1.1    pooka 
    615  1.31    pooka 	puffs_msgif_destroy();
    616  1.31    pooka 	puffs_transport_destroy();
    617  1.31    pooka 	pool_destroy(&puffs_pnpool);
    618  1.31    pooka 
    619   1.5    pooka 	malloc_type_detach(M_PUFFS);
    620   1.1    pooka }
    621   1.1    pooka 
    622   1.1    pooka int
    623   1.1    pooka puffs_snapshot(struct mount *mp, struct vnode *vp, struct timespec *ts)
    624   1.1    pooka {
    625   1.1    pooka 
    626   1.1    pooka 	return EOPNOTSUPP;
    627   1.1    pooka }
    628   1.1    pooka 
    629  1.26    pooka int
    630  1.26    pooka puffs_suspendctl(struct mount *mp, int cmd)
    631  1.26    pooka {
    632  1.26    pooka 	struct puffs_mount *pmp;
    633  1.26    pooka 	int error;
    634  1.26    pooka 
    635  1.26    pooka 	pmp = MPTOPUFFSMP(mp);
    636  1.26    pooka 	switch (cmd) {
    637  1.26    pooka 	case SUSPEND_SUSPEND:
    638  1.26    pooka 		DPRINTF(("puffs_suspendctl: suspending\n"));
    639  1.27  hannken 		if ((error = fstrans_setstate(mp, FSTRANS_SUSPENDING)) != 0)
    640  1.26    pooka 			break;
    641  1.26    pooka 		puffs_suspendtouser(pmp, PUFFS_SUSPEND_START);
    642  1.26    pooka 
    643  1.30    pooka 		error = pageflush(mp, FSCRED, 0, 1, curlwp);
    644  1.26    pooka 		if (error == 0)
    645  1.27  hannken 			error = fstrans_setstate(mp, FSTRANS_SUSPENDED);
    646  1.26    pooka 
    647  1.26    pooka 		if (error != 0) {
    648  1.26    pooka 			puffs_suspendtouser(pmp, PUFFS_SUSPEND_ERROR);
    649  1.27  hannken 			(void) fstrans_setstate(mp, FSTRANS_NORMAL);
    650  1.26    pooka 			break;
    651  1.26    pooka 		}
    652  1.26    pooka 
    653  1.26    pooka 		puffs_suspendtouser(pmp, PUFFS_SUSPEND_SUSPENDED);
    654  1.26    pooka 
    655  1.26    pooka 		break;
    656  1.26    pooka 
    657  1.26    pooka 	case SUSPEND_RESUME:
    658  1.26    pooka 		DPRINTF(("puffs_suspendctl: resume\n"));
    659  1.26    pooka 		error = 0;
    660  1.27  hannken 		(void) fstrans_setstate(mp, FSTRANS_NORMAL);
    661  1.26    pooka 		puffs_suspendtouser(pmp, PUFFS_SUSPEND_RESUME);
    662  1.26    pooka 		break;
    663  1.26    pooka 
    664  1.26    pooka 	default:
    665  1.26    pooka 		error = EINVAL;
    666  1.26    pooka 		break;
    667  1.26    pooka 	}
    668  1.26    pooka 
    669  1.26    pooka 	DPRINTF(("puffs_suspendctl: return %d\n", error));
    670  1.26    pooka 	return error;
    671  1.26    pooka }
    672  1.26    pooka 
    673   1.1    pooka const struct vnodeopv_desc * const puffs_vnodeopv_descs[] = {
    674   1.1    pooka 	&puffs_vnodeop_opv_desc,
    675   1.3    pooka 	&puffs_specop_opv_desc,
    676   1.4    pooka 	&puffs_fifoop_opv_desc,
    677  1.12    pooka 	&puffs_msgop_opv_desc,
    678   1.1    pooka 	NULL,
    679   1.1    pooka };
    680   1.1    pooka 
    681   1.1    pooka struct vfsops puffs_vfsops = {
    682   1.1    pooka 	MOUNT_PUFFS,
    683   1.1    pooka 	puffs_mount,		/* mount	*/
    684   1.1    pooka 	puffs_start,		/* start	*/
    685   1.1    pooka 	puffs_unmount,		/* unmount	*/
    686   1.1    pooka 	puffs_root,		/* root		*/
    687   1.1    pooka 	puffs_quotactl,		/* quotactl	*/
    688   1.1    pooka 	puffs_statvfs,		/* statvfs	*/
    689   1.1    pooka 	puffs_sync,		/* sync		*/
    690   1.1    pooka 	puffs_vget,		/* vget		*/
    691  1.15      chs 	(void *)eopnotsupp,	/* fhtovp	*/
    692  1.15      chs 	(void *)eopnotsupp,	/* vptofh	*/
    693   1.1    pooka 	puffs_init,		/* init		*/
    694   1.1    pooka 	NULL,			/* reinit	*/
    695   1.1    pooka 	puffs_done,		/* done		*/
    696   1.1    pooka 	NULL,			/* mountroot	*/
    697   1.1    pooka 	puffs_snapshot,		/* snapshot	*/
    698   1.1    pooka 	vfs_stdextattrctl,	/* extattrctl	*/
    699  1.26    pooka 	puffs_suspendctl,	/* suspendctl	*/
    700   1.1    pooka 	puffs_vnodeopv_descs,	/* vnodeops	*/
    701   1.1    pooka 	0,			/* refcount	*/
    702   1.1    pooka 	{ NULL, NULL }
    703   1.1    pooka };
    704   1.1    pooka VFS_ATTACH(puffs_vfsops);
    705