Merge commit 'ea01a15a654b9e1c7b37d958f4d1911882ed7781'

[unleashed.git] / kernel / fs / namefs / namevfs.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2017 by Delphix. All rights reserved.
 */

/*      Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T */
/*        All Rights Reserved   */


/*
 * This file supports the vfs operations for the NAMEFS file system.
 */

#include <sys/types.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/debug.h>
#include <sys/errno.h>
#include <sys/kmem.h>
#include <sys/inline.h>
#include <sys/file.h>
#include <sys/proc.h>
#include <sys/stat.h>
#include <sys/statvfs.h>
#include <sys/mount.h>
#include <sys/sysmacros.h>
#include <sys/var.h>
#include <sys/vfs.h>
#include <sys/vnode.h>
#include <sys/mode.h>
#include <sys/pcb.h>
#include <sys/signal.h>
#include <sys/user.h>
#include <sys/uio.h>
#include <sys/cred.h>
#include <sys/fs/namenode.h>
#include <sys/stream.h>
#include <sys/strsubr.h>
#include <sys/cmn_err.h>
#include <sys/modctl.h>
#include <sys/fs_subr.h>
#include <sys/policy.h>
#include <sys/vmem.h>
#include <sys/fs/sdev_impl.h>

#define NM_INOQUANT             (64 * 1024)

/*
 * Define global data structures.
 */
dev_t   namedev;
int     namefstype;
struct  namenode *nm_filevp_hash[NM_FILEVP_HASH_SIZE];
struct  vfs namevfs;
kmutex_t ntable_lock;

static vmem_t   *nm_inoarena;   /* vmem arena to allocate inode no's from */
static kmutex_t nm_inolock;

static const struct vfsops namefs_vfsops;

/*
 * Functions to allocate node id's starting from 1. Based on vmem routines.
 * The vmem arena is extended in NM_INOQUANT chunks.
 */
uint64_t
namenodeno_alloc(void)
{
        uint64_t nno;

        mutex_enter(&nm_inolock);
        nno = (uint64_t)(uintptr_t)
            vmem_alloc(nm_inoarena, 1, VM_NOSLEEP + VM_FIRSTFIT);
        if (nno == 0) {
                (void) vmem_add(nm_inoarena, (void *)(vmem_size(nm_inoarena,
                    VMEM_ALLOC | VMEM_FREE) + 1), NM_INOQUANT, VM_SLEEP);
                nno = (uint64_t)(uintptr_t)
                    vmem_alloc(nm_inoarena, 1, VM_SLEEP + VM_FIRSTFIT);
                ASSERT(nno != 0);
        }
        mutex_exit(&nm_inolock);
        ASSERT32(nno <= ULONG_MAX);
        return (nno);
}

static void
namenodeno_init(void)
{
        nm_inoarena = vmem_create("namefs_inodes", (void *)1, NM_INOQUANT, 1,
            NULL, NULL, NULL, 1, VM_SLEEP);
        mutex_init(&nm_inolock, NULL, MUTEX_DEFAULT, NULL);
}

void
namenodeno_free(uint64_t nn)
{
        void *vaddr = (void *)(uintptr_t)nn;

        ASSERT32((uint64_t)(uintptr_t)vaddr == nn);

        mutex_enter(&nm_inolock);
        vmem_free(nm_inoarena, vaddr, 1);
        mutex_exit(&nm_inolock);
}

/*
 * Insert a namenode into the nm_filevp_hash table.
 *
 * Each link has a unique namenode with a unique nm_mountvp field.
 * The nm_filevp field of the namenode need not be unique, since a
 * file descriptor may be mounted to multiple nodes at the same time.
 * We hash on nm_filevp since that's what discriminates the searches
 * in namefind() and nm_unmountall().
 */
void
nameinsert(struct namenode *nodep)
{
        struct namenode **bucket;

        ASSERT(MUTEX_HELD(&ntable_lock));

        bucket = NM_FILEVP_HASH(nodep->nm_filevp);
        nodep->nm_nextp = *bucket;
        *bucket = nodep;
}

/*
 * Remove a namenode from the hash table, if present.
 */
void
nameremove(struct namenode *nodep)
{
        struct namenode *np, **npp;

        ASSERT(MUTEX_HELD(&ntable_lock));

        for (npp = NM_FILEVP_HASH(nodep->nm_filevp); (np = *npp) != NULL;
            npp = &np->nm_nextp) {
                if (np == nodep) {
                        *npp = np->nm_nextp;
                        return;
                }
        }
}

/*
 * Search for a namenode that has a nm_filevp == vp and nm_mountpt == mnt.
 * If mnt is NULL, return the first link with nm_filevp of vp.
 * Returns namenode pointer on success, NULL on failure.
 */
struct namenode *
namefind(vnode_t *vp, vnode_t *mnt)
{
        struct namenode *np;

        ASSERT(MUTEX_HELD(&ntable_lock));
        for (np = *NM_FILEVP_HASH(vp); np != NULL; np = np->nm_nextp)
                if (np->nm_filevp == vp &&
                    (mnt == NULL || np->nm_mountpt == mnt))
                        break;
        return (np);
}

/*
 * Force the unmouting of a file descriptor from ALL of the nodes
 * that it was mounted to.
 * At the present time, the only usage for this routine is in the
 * event one end of a pipe was mounted. At the time the unmounted
 * end gets closed down, the mounted end is forced to be unmounted.
 *
 * This routine searches the namenode hash list for all namenodes
 * that have a nm_filevp field equal to vp. Each time one is found,
 * the dounmount() routine is called. This causes the nm_unmount()
 * routine to be called and thus, the file descriptor is unmounted
 * from the node.
 *
 * At the start of this routine, the reference count for vp is
 * incremented to protect the vnode from being released in the
 * event the mount was the only thing keeping the vnode active.
 * If that is the case, the fop_close operation is applied to
 * the vnode, prior to it being released.
 */
static int
nm_umountall(vnode_t *vp, cred_t *crp)
{
        vfs_t *vfsp;
        struct namenode *nodep;
        int error = 0;
        int realerr = 0;

        /*
         * For each namenode that is associated with the file:
         * If the v_vfsp field is not namevfs, dounmount it.  Otherwise,
         * it was created in nm_open() and will be released in time.
         * The following loop replicates some code from nm_find.  That
         * routine can't be used as is since the list isn't strictly
         * consumed as it is traversed.
         */
        mutex_enter(&ntable_lock);
        nodep = *NM_FILEVP_HASH(vp);
        while (nodep) {
                if (nodep->nm_filevp == vp &&
                    (vfsp = NMTOV(nodep)->v_vfsp) != NULL &&
                    vfsp != &namevfs && (NMTOV(nodep)->v_flag & VROOT)) {

                        /*
                         * If the vn_vfswlock fails, skip the vfs since
                         * somebody else may be unmounting it.
                         */
                        if (vn_vfswlock(vfsp->vfs_vnodecovered)) {
                                realerr = EBUSY;
                                nodep = nodep->nm_nextp;
                                continue;
                        }

                        /*
                         * Can't hold ntable_lock across call to do_unmount
                         * because nm_unmount tries to acquire it.  This means
                         * there is a window where another mount of vp can
                         * happen so it is possible that after nm_unmountall
                         * there are still some mounts.  This situation existed
                         * without MT locking because dounmount can sleep
                         * so another mount could happen during that time.
                         * This situation is unlikely and doesn't really cause
                         * any problems.
                         */
                        mutex_exit(&ntable_lock);
                        if ((error = dounmount(vfsp, 0, crp)) != 0)
                                realerr = error;
                        mutex_enter(&ntable_lock);
                        /*
                         * Since we dropped the ntable_lock, we
                         * have to start over from the beginning.
                         * If for some reasons dounmount() fails,
                         * start from beginning means that we will keep on
                         * trying unless another thread unmounts it for us.
                         */
                        nodep = *NM_FILEVP_HASH(vp);
                } else
                        nodep = nodep->nm_nextp;
        }
        mutex_exit(&ntable_lock);
        return (realerr);
}

/*
 * Force the unmouting of a file descriptor from ALL of the nodes
 * that it was mounted to.  XXX: fifo_close() calls this routine.
 *
 * nm_umountall() may return EBUSY.
 * nm_unmountall() will keep on trying until it succeeds.
 */
int
nm_unmountall(vnode_t *vp, cred_t *crp)
{
        int error;

        /*
         * Nm_umuontall() returns only if it succeeds or
         * return with error EBUSY.  If EBUSY, that means
         * it cannot acquire the lock on the covered vnode,
         * and we will keep on trying.
         */
        for (;;) {
                error = nm_umountall(vp, crp);
                if (error != EBUSY)
                        break;
                delay(1);       /* yield cpu briefly, then try again */
        }
        return (error);
}

/*
 * Mount a file descriptor onto the node in the file system.
 * Create a new vnode, update the attributes with info from the
 * file descriptor and the mount point.  The mask, mode, uid, gid,
 * atime, mtime and ctime are taken from the mountpt.  Link count is
 * set to one, the file system id is namedev and nodeid is unique
 * for each mounted object.  Other attributes are taken from mount point.
 * Make sure user is owner (or root) with write permissions on mount point.
 * Hash the new vnode and return 0.
 * Upon entry to this routine, the file descriptor is in the
 * fd field of a struct namefd.  Copy that structure from user
 * space and retrieve the file descriptor.
 */
static int
nm_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *crp)
{
        struct namefd namefdp;
        struct vnode *filevp;           /* file descriptor vnode */
        struct file *fp;
        struct vnode *newvp;            /* vnode representing this mount */
        struct vnode *rvp;              /* realvp (if any) for the mountpt */
        struct namenode *nodep;         /* namenode for this mount */
        struct vattr filevattr;         /* attributes of file dec.  */
        struct vattr *vattrp;           /* attributes of this mount */
        char *resource_name;
        char *resource_nodetype;
        statvfs64_t *svfsp;
        int error = 0;

        /*
         * Get the file descriptor from user space.
         * Make sure the file descriptor is valid and has an
         * associated file pointer.
         * If so, extract the vnode from the file pointer.
         */
        if (uap->datalen != sizeof (struct namefd))
                return (EINVAL);

        if (copyin(uap->dataptr, &namefdp, uap->datalen))
                return (EFAULT);

        if ((fp = getf(namefdp.fd)) == NULL)
                return (EBADF);

        /*
         * If the mount point already has something mounted
         * on it, disallow this mount.  (This restriction may
         * be removed in a later release).
         * Or unmount has completed but the namefs ROOT vnode
         * count has not decremented to zero, disallow this mount.
         */

        mutex_enter(&mvp->v_lock);
        if ((mvp->v_flag & VROOT) ||
            vfs_matchops(mvp->v_vfsp, &namefs_vfsops)) {
                mutex_exit(&mvp->v_lock);
                releasef(namefdp.fd);
                return (EBUSY);
        }
        mutex_exit(&mvp->v_lock);

        /*
         * Cannot allow users to fattach() in /dev/pts.
         * First, there is no need for doing so and secondly
         * we cannot allow arbitrary users to park on a node in
         * /dev/pts or /dev/vt.
         */
        rvp = NULLVP;
        if (vn_matchops(mvp, spec_getvnodeops()) &&
            fop_realvp(mvp, &rvp, NULL) == 0 && rvp &&
            (vn_matchops(rvp, devpts_getvnodeops()) ||
            vn_matchops(rvp, devvt_getvnodeops()))) {
                releasef(namefdp.fd);
                return (ENOTSUP);
        }

        filevp = fp->f_vnode;
        if (filevp->v_type == VDIR || filevp->v_type == VPORT) {
                releasef(namefdp.fd);
                return (EINVAL);
        }

        /*
         * If the fd being mounted refers to neither a door nor a stream,
         * make sure the caller is privileged.
         */
        if (filevp->v_type != VDOOR && filevp->v_stream == NULL) {
                if (secpolicy_fs_mount(crp, filevp, vfsp) != 0) {
                        /* fd is neither a stream nor a door */
                        releasef(namefdp.fd);
                        return (EINVAL);
                }
        }

        /*
         * Make sure the file descriptor is not the root of some
         * file system.
         * If it's not, create a reference and allocate a namenode
         * to represent this mount request.
         */
        if (filevp->v_flag & VROOT) {
                releasef(namefdp.fd);
                return (EBUSY);
        }

        nodep = kmem_zalloc(sizeof (struct namenode), KM_SLEEP);

        mutex_init(&nodep->nm_lock, NULL, MUTEX_DEFAULT, NULL);
        vattrp = &nodep->nm_vattr;
        vattrp->va_mask = AT_ALL;
        if (error = fop_getattr(mvp, vattrp, 0, crp, NULL))
                goto out;

        filevattr.va_mask = AT_ALL;
        if (error = fop_getattr(filevp, &filevattr, 0, crp, NULL))
                goto out;
        /*
         * Make sure the user is the owner of the mount point
         * or has sufficient privileges.
         */
        if (error = secpolicy_vnode_owner(crp, vattrp->va_uid))
                goto out;

        /*
         * Make sure the user has write permissions on the
         * mount point (or has sufficient privileges).
         */
        if (secpolicy_vnode_access2(crp, mvp, vattrp->va_uid, vattrp->va_mode,
            VWRITE) != 0) {
                error = EACCES;
                goto out;
        }

        /*
         * If the file descriptor has file/record locking, don't
         * allow the mount to succeed.
         */
        if (vn_has_flocks(filevp)) {
                error = EACCES;
                goto out;
        }

        /*
         * Initialize the namenode.
         */
        if (filevp->v_stream) {
                struct stdata *stp = filevp->v_stream;
                mutex_enter(&stp->sd_lock);
                stp->sd_flag |= STRMOUNT;
                mutex_exit(&stp->sd_lock);
        }
        nodep->nm_filevp = filevp;
        mutex_enter(&fp->f_tlock);
        fp->f_count++;
        mutex_exit(&fp->f_tlock);

        releasef(namefdp.fd);
        nodep->nm_filep = fp;
        nodep->nm_mountpt = mvp;

        /*
         * The attributes for the mounted file descriptor were initialized
         * above by applying fop_getattr to the mount point.  Some of
         * the fields of the attributes structure will be overwritten
         * by the attributes from the file descriptor.
         */
        vattrp->va_type    = filevattr.va_type;
        vattrp->va_fsid    = namedev;
        vattrp->va_nodeid  = namenodeno_alloc();
        vattrp->va_nlink   = 1;
        vattrp->va_size    = filevattr.va_size;
        vattrp->va_rdev    = filevattr.va_rdev;
        vattrp->va_blksize = filevattr.va_blksize;
        vattrp->va_nblocks = filevattr.va_nblocks;
        vattrp->va_seq     = 0;

        /*
         * Initialize new vnode structure for the mounted file descriptor.
         */
        nodep->nm_vnode = vn_alloc(KM_SLEEP);
        newvp = NMTOV(nodep);

        newvp->v_flag = filevp->v_flag | VROOT | VNOMAP | VNOSWAP;
        vn_setops(newvp, &nm_vnodeops);
        newvp->v_vfsp = vfsp;
        newvp->v_stream = filevp->v_stream;
        newvp->v_type = filevp->v_type;
        newvp->v_rdev = filevp->v_rdev;
        newvp->v_data = (caddr_t)nodep;
        VFS_HOLD(vfsp);
        vn_exists(newvp);

        /*
         * Initialize the vfs structure.
         */
        vfsp->vfs_vnodecovered = NULL;
        vfsp->vfs_flag |= VFS_UNLINKABLE;
        vfsp->vfs_bsize = 1024;
        vfsp->vfs_fstype = namefstype;
        vfs_make_fsid(&vfsp->vfs_fsid, namedev, namefstype);
        vfsp->vfs_data = (caddr_t)nodep;
        vfsp->vfs_dev = namedev;
        vfsp->vfs_bcount = 0;

        /*
         * Set the name we mounted from.
         */
        switch (filevp->v_type) {
        case VPROC:     /* fop_getattr() translates this to VREG */
        case VREG:      resource_nodetype = "file"; break;
        case VDIR:      resource_nodetype = "directory"; break;
        case VBLK:      resource_nodetype = "device"; break;
        case VCHR:      resource_nodetype = "device"; break;
        case VLNK:      resource_nodetype = "link"; break;
        case VFIFO:     resource_nodetype = "fifo"; break;
        case VDOOR:     resource_nodetype = "door"; break;
        case VSOCK:     resource_nodetype = "socket"; break;
        default:        resource_nodetype = "resource"; break;
        }

#define RESOURCE_NAME_SZ 128 /* Maximum length of the resource name */
        resource_name = kmem_alloc(RESOURCE_NAME_SZ, KM_SLEEP);
        svfsp = kmem_alloc(sizeof (statvfs64_t), KM_SLEEP);

        error = VFS_STATVFS(filevp->v_vfsp, svfsp);
        if (error == 0) {
                (void) snprintf(resource_name, RESOURCE_NAME_SZ,
                    "unspecified_%s_%s", svfsp->f_basetype, resource_nodetype);
        } else {
                (void) snprintf(resource_name, RESOURCE_NAME_SZ,
                    "unspecified_%s", resource_nodetype);
        }

        vfs_setresource(vfsp, resource_name, 0);

        kmem_free(svfsp, sizeof (statvfs64_t));
        kmem_free(resource_name, RESOURCE_NAME_SZ);
#undef RESOURCE_NAME_SZ

        /*
         * Insert the namenode.
         */
        mutex_enter(&ntable_lock);
        nameinsert(nodep);
        mutex_exit(&ntable_lock);
        return (0);
out:
        releasef(namefdp.fd);
        kmem_free(nodep, sizeof (struct namenode));
        return (error);
}

/*
 * Unmount a file descriptor from a node in the file system.
 * If the user is not the owner of the file and is not privileged,
 * the request is denied.
 * Otherwise, remove the namenode from the hash list.
 * If the mounted file descriptor was that of a stream and this
 * was the last mount of the stream, turn off the STRMOUNT flag.
 * If the rootvp is referenced other than through the mount,
 * nm_inactive will clean up.
 */
static int
nm_unmount(vfs_t *vfsp, int flag, cred_t *crp)
{
        struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
        vnode_t *vp, *thisvp;
        struct file *fp = NULL;

        ASSERT((nodep->nm_flag & NMNMNT) == 0);

        /*
         * forced unmount is not supported by this file system
         * and thus, ENOTSUP, is being returned.
         */
        if (flag & MS_FORCE) {
                return (ENOTSUP);
        }

        vp = nodep->nm_filevp;
        mutex_enter(&nodep->nm_lock);
        if (secpolicy_vnode_owner(crp, nodep->nm_vattr.va_uid) != 0) {
                mutex_exit(&nodep->nm_lock);
                return (EPERM);
        }

        mutex_exit(&nodep->nm_lock);

        mutex_enter(&ntable_lock);
        nameremove(nodep);
        thisvp = NMTOV(nodep);
        mutex_enter(&thisvp->v_lock);
        VN_RELE_LOCKED(thisvp);
        if (thisvp->v_count == 0) {
                fp = nodep->nm_filep;
                mutex_exit(&thisvp->v_lock);
                vn_invalid(thisvp);
                vn_free(thisvp);
                VFS_RELE(vfsp);
                namenodeno_free(nodep->nm_vattr.va_nodeid);
                kmem_free(nodep, sizeof (struct namenode));
        } else {
                thisvp->v_flag &= ~VROOT;
                mutex_exit(&thisvp->v_lock);
        }
        if (namefind(vp, NULLVP) == NULL && vp->v_stream) {
                struct stdata *stp = vp->v_stream;
                mutex_enter(&stp->sd_lock);
                stp->sd_flag &= ~STRMOUNT;
                mutex_exit(&stp->sd_lock);
        }
        mutex_exit(&ntable_lock);
        if (fp != NULL)
                (void) closef(fp);
        return (0);
}

/*
 * Create a reference to the root of a mounted file descriptor.
 * This routine is called from lookupname() in the event a path
 * is being searched that has a mounted file descriptor in it.
 */
static int
nm_root(vfs_t *vfsp, vnode_t **vpp)
{
        struct namenode *nodep = (struct namenode *)vfsp->vfs_data;
        struct vnode *vp = NMTOV(nodep);

        VN_HOLD(vp);
        *vpp = vp;
        return (0);
}

/*
 * Return in sp the status of this file system.
 */
static int
nm_statvfs(vfs_t *vfsp, struct statvfs64 *sp)
{
        dev32_t d32;

        bzero(sp, sizeof (*sp));
        sp->f_bsize     = 1024;
        sp->f_frsize    = 1024;
        (void) cmpldev(&d32, vfsp->vfs_dev);
        sp->f_fsid = d32;
        (void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
        sp->f_flag      = vf_to_stf(vfsp->vfs_flag);
        return (0);
}

/*
 * Since this file system has no disk blocks of its own, apply
 * the fop_fsync operation on the mounted file descriptor.
 */
static int
nm_sync(vfs_t *vfsp, short flag, cred_t *crp)
{
        struct namenode *nodep;

        if (vfsp == NULL)
                return (0);

        nodep = (struct namenode *)vfsp->vfs_data;
        if (flag & SYNC_CLOSE)
                return (nm_umountall(nodep->nm_filevp, crp));

        return (fop_fsync(nodep->nm_filevp, FSYNC, crp, NULL));
}

static const struct vfsops namefs_vfsops = {
        .vfs_mount = nm_mount,
        .vfs_unmount = nm_unmount,
        .vfs_root = nm_root,
        .vfs_statvfs = nm_statvfs,
        .vfs_sync = nm_sync,
};

static const struct vfsops namefs_dummy_vfsops = {
        .vfs_statvfs = nm_statvfs,
        .vfs_sync = nm_sync,
};

/*
 * File system initialization routine. Save the file system type,
 * establish a file system device number and initialize nm_filevp_hash[].
 */
int
nameinit(int fstype, char *name)
{
        int error;
        int dev;

        error = vfs_setfsops(fstype, &namefs_vfsops);
        if (error != 0) {
                cmn_err(CE_WARN, "nameinit: bad fsytpe");
                return (error);
        }

        namefstype = fstype;

        if ((dev = getudev()) == (major_t)-1) {
                cmn_err(CE_WARN, "nameinit: can't get unique device");
                dev = 0;
        }
        mutex_init(&ntable_lock, NULL, MUTEX_DEFAULT, NULL);
        namedev = makedevice(dev, 0);
        bzero(nm_filevp_hash, sizeof (nm_filevp_hash));
        vfs_setops(&namevfs, &namefs_dummy_vfsops);
        namevfs.vfs_vnodecovered = NULL;
        namevfs.vfs_bsize = 1024;
        namevfs.vfs_fstype = namefstype;
        vfs_make_fsid(&namevfs.vfs_fsid, namedev, namefstype);
        namevfs.vfs_dev = namedev;
        return (0);
}

static mntopts_t nm_mntopts = {
        0,
        NULL
};

static vfsdef_t vfw = {
        VFSDEF_VERSION,
        "namefs",
        nameinit,
        VSW_HASPROTO | VSW_ZMOUNT,
        &nm_mntopts
};

/*
 * Module linkage information for the kernel.
 */
static struct modlfs modlfs = {
        &mod_fsops, "filesystem for namefs", &vfw
};

static struct modlinkage modlinkage = {
        MODREV_1, (void *)&modlfs, NULL
};

int
_init(void)
{
        namenodeno_init();
        return (mod_install(&modlinkage));
}

int
_fini(void)
{
        return (EBUSY);
}

int
_info(struct modinfo *modinfop)
{
        return (mod_info(&modlinkage, modinfop));
}