diff --git a/sys/kern/kern_jail.c.ident2 b/sys/kern/kern_jail.c --- a/sys/kern/kern_jail.c.ident2 +++ b/sys/kern/kern_jail.c @@ -3328,6 +3328,7 @@ { sx_assert(&allprison_lock, SA_XLOCKED); mtx_assert(&pr->pr_mtx, MA_NOTOWNED); + vfs_exjail_delete(pr); shm_remove_prison(pr); (void)osd_jail_call(pr, PR_METHOD_REMOVE, NULL); } diff --git a/sys/kern/vfs_export.c.ident b/sys/kern/vfs_export.c --- a/sys/kern/vfs_export.c.ident +++ b/sys/kern/vfs_export.c @@ -52,6 +52,7 @@ #include #include #include +#include #include #include #include @@ -70,11 +71,11 @@ static struct radix_node_head *vfs_create_addrlist_af( struct radix_node_head **prnh, int off); #endif -static void vfs_free_addrlist(struct netexport *nep); static int vfs_free_netcred(struct radix_node *rn, void *w); static void vfs_free_addrlist_af(struct radix_node_head **prnh); static int vfs_hang_addrlist(struct mount *mp, struct netexport *nep, struct export_args *argp); +static struct ucred *vfs_free_exports(struct mount *); static struct netcred *vfs_export_lookup(struct mount *, struct sockaddr *); /* @@ -274,7 +275,7 @@ /* * Free the net address hash lists that are hanging off the mount points. */ -static void +void vfs_free_addrlist(struct netexport *nep) { struct ucred *cred; @@ -285,8 +286,10 @@ vfs_free_addrlist_af(&nep->ne6); cred = nep->ne_defexported.netc_anon; - if (cred != NULL) + if (cred != NULL) { crfree(cred); + nep->ne_defexported.netc_anon = NULL; + } } @@ -295,12 +298,18 @@ * and the passed in netexport. * Struct export_args *argp is the variable used to twiddle options, * the structure is described in sys/mount.h + * The do_exjail argument should be true if *mp is in the mountlist + * and false if not. It is not in the mountlist for the NFSv4 rootfs + * fake mount point just used for exports. */ int -vfs_export(struct mount *mp, struct export_args *argp) +vfs_export(struct mount *mp, struct export_args *argp, bool do_exjail) { struct netexport *nep; + struct ucred *cr; + struct prison *pr; int error; + bool new_nep; if ((argp->ex_flags & (MNT_DELEXPORT | MNT_EXPORTED)) == 0) return (EINVAL); @@ -311,6 +320,7 @@ return (EINVAL); error = 0; + pr = curthread->td_ucred->cr_prison; lockmgr(&mp->mnt_explock, LK_EXCLUSIVE, NULL); nep = mp->mnt_export; if (argp->ex_flags & MNT_DELEXPORT) { @@ -318,29 +328,69 @@ error = ENOENT; goto out; } + MNT_ILOCK(mp); + if (mp->mnt_exjail != NULL && mp->mnt_exjail->cr_prison != pr && + pr == &prison0) { + MNT_IUNLOCK(mp); + /* EXDEV will not get logged by mountd(8). */ + error = EXDEV; + goto out; + } else if (mp->mnt_exjail != NULL && + mp->mnt_exjail->cr_prison != pr) { + MNT_IUNLOCK(mp); + /* EPERM will get logged by mountd(8). */ + error = EPERM; + goto out; + } + MNT_IUNLOCK(mp); if (mp->mnt_flag & MNT_EXPUBLIC) { vfs_setpublicfs(NULL, NULL, NULL); MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_EXPUBLIC; MNT_IUNLOCK(mp); } - vfs_free_addrlist(nep); - mp->mnt_export = NULL; - free(nep, M_MOUNT); - nep = NULL; - MNT_ILOCK(mp); - mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); - MNT_IUNLOCK(mp); + cr = vfs_free_exports(mp); + if (cr != NULL) { + atomic_add_int(&pr->pr_exportcnt, -1); + crfree(cr); + } } if (argp->ex_flags & MNT_EXPORTED) { + new_nep = false; + MNT_ILOCK(mp); + if (mp->mnt_exjail == NULL) { + MNT_IUNLOCK(mp); + if (do_exjail && nep != NULL) { + vfs_free_addrlist(nep); + memset(nep, 0, sizeof(*nep)); + new_nep = true; + } + } else if (mp->mnt_exjail->cr_prison != pr) { + MNT_IUNLOCK(mp); + error = EPERM; + goto out; + } else + MNT_IUNLOCK(mp); if (nep == NULL) { - nep = malloc(sizeof(struct netexport), M_MOUNT, M_WAITOK | M_ZERO); + nep = malloc(sizeof(struct netexport), M_MOUNT, + M_WAITOK | M_ZERO); mp->mnt_export = nep; + new_nep = true; } if (argp->ex_flags & MNT_EXPUBLIC) { - if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) + if ((error = vfs_setpublicfs(mp, nep, argp)) != 0) { + if (new_nep) { + mp->mnt_export = NULL; + free(nep, M_MOUNT); + } goto out; + } + new_nep = false; MNT_ILOCK(mp); + if (do_exjail && mp->mnt_exjail == NULL) { + mp->mnt_exjail = crhold(curthread->td_ucred); + atomic_add_int(&pr->pr_exportcnt, 1); + } mp->mnt_flag |= MNT_EXPUBLIC; MNT_IUNLOCK(mp); } @@ -348,9 +398,18 @@ argp->ex_numsecflavors = 1; argp->ex_secflavors[0] = AUTH_SYS; } - if ((error = vfs_hang_addrlist(mp, nep, argp))) + if ((error = vfs_hang_addrlist(mp, nep, argp))) { + if (new_nep) { + mp->mnt_export = NULL; + free(nep, M_MOUNT); + } goto out; + } MNT_ILOCK(mp); + if (do_exjail && mp->mnt_exjail == NULL) { + mp->mnt_exjail = crhold(curthread->td_ucred); + atomic_add_int(&pr->pr_exportcnt, 1); + } mp->mnt_flag |= MNT_EXPORTED; MNT_IUNLOCK(mp); } @@ -368,6 +427,69 @@ vfs_deleteopt(mp->mnt_optnew, "export"); vfs_deleteopt(mp->mnt_opt, "export"); return (error); +} + +static struct ucred * +vfs_free_exports(struct mount *mp) +{ + struct ucred *cr; + + vfs_free_addrlist(mp->mnt_export); + free(mp->mnt_export, M_MOUNT); + mp->mnt_export = NULL; + MNT_ILOCK(mp); + cr = mp->mnt_exjail; + mp->mnt_exjail = NULL; + mp->mnt_flag &= ~(MNT_EXPORTED | MNT_DEFEXPORTED); + MNT_IUNLOCK(mp); + return (cr); +} + +/* + * Get rid of credential references for this prison. + */ +void +vfs_exjail_delete(struct prison *pr) +{ + struct mount *mp; + struct ucred *cr; + int error, i; + + i = atomic_load_int(&pr->pr_exportcnt); + KASSERT(i >= 0, ("vfs_exjail_delete: pr_exportcnt negative")); + if (i == 0) + return; + mtx_lock(&mountlist_mtx); +tryagain: + TAILQ_FOREACH(mp, &mountlist, mnt_list) { + cr = NULL; + MNT_ILOCK(mp); + if (mp->mnt_exjail != NULL && + mp->mnt_exjail->cr_prison == pr) { + cr = mp->mnt_exjail; + mp->mnt_exjail = NULL; + } + MNT_IUNLOCK(mp); + if (cr != NULL) { + crfree(cr); + i--; + error = vfs_busy(mp, MBF_MNTLSTLOCK | MBF_NOWAIT); + if (error != 0) { + if (i == 0) + break; + continue; + } + lockmgr(&mp->mnt_explock, LK_EXCLUSIVE, NULL); + vfs_free_exports(mp); + lockmgr(&mp->mnt_explock, LK_RELEASE, NULL); + mtx_lock(&mountlist_mtx); + vfs_unbusy(mp); + if (i == 0) + break; + goto tryagain; + } + } + mtx_unlock(&mountlist_mtx); } /* diff --git a/sys/kern/vfs_mount.c.ident b/sys/kern/vfs_mount.c --- a/sys/kern/vfs_mount.c.ident +++ b/sys/kern/vfs_mount.c @@ -761,6 +761,8 @@ #endif if (mp->mnt_opt != NULL) vfs_freeopts(mp->mnt_opt); + if (mp->mnt_exjail != NULL) + crfree(mp->mnt_exjail); crfree(mp->mnt_cred); uma_zfree(mount_zone, mp); } @@ -1471,7 +1473,7 @@ } else export_error = EINVAL; if (export_error == 0) - export_error = vfs_export(mp, &export); + export_error = vfs_export(mp, &export, true); free(export.ex_groups, M_TEMP); break; case (sizeof(export)): @@ -1493,7 +1495,7 @@ else export_error = EINVAL; if (export_error == 0) - export_error = vfs_export(mp, &export); + export_error = vfs_export(mp, &export, true); free(grps, M_TEMP); break; default: diff --git a/sys/sys/jail.h.ident b/sys/sys/jail.h --- a/sys/sys/jail.h.ident +++ b/sys/sys/jail.h @@ -175,6 +175,7 @@ int pr_id; /* (c) prison id */ volatile u_int pr_ref; /* (r) refcount */ volatile u_int pr_uref; /* (r) user (alive) refcount */ + volatile int pr_exportcnt; /* (r) count of mount exports */ unsigned pr_flags; /* (p) PR_* flags */ LIST_HEAD(, prison) pr_children; /* (a) list of child jails */ LIST_HEAD(, proc) pr_proclist; /* (A) list of jailed processes */ diff --git a/sys/sys/mount.h.ident b/sys/sys/mount.h --- a/sys/sys/mount.h.ident +++ b/sys/sys/mount.h @@ -216,6 +216,7 @@ * i - interlock * v - vnode freelist mutex * d - deferred unmount list mutex + * e - mnt_explock * * Unmarked fields are considered stable as long as a ref is held. * @@ -245,13 +246,14 @@ void * mnt_data; /* private data */ time_t mnt_time; /* last time written*/ int mnt_iosize_max; /* max size for clusters, etc */ - struct netexport *mnt_export; /* export list */ + struct netexport *mnt_export; /* (e) export list */ struct label *mnt_label; /* MAC label for the fs */ u_int mnt_hashseed; /* Random seed for vfs_hash */ int mnt_lockref; /* (i) Lock reference count */ int mnt_secondary_writes; /* (i) # of secondary writes */ int mnt_secondary_accwrites;/* (i) secondary wr. starts */ struct thread *mnt_susp_owner; /* (i) thread owning suspension */ + struct ucred *mnt_exjail; /* (i) jail which did exports */ #define mnt_endzero mnt_gjprovider char *mnt_gjprovider; /* gjournal provider name */ struct mtx mnt_listmtx; @@ -1016,8 +1018,10 @@ (struct mount *, struct netexport *, struct export_args *); void vfs_periodic(struct mount *, int); int vfs_busy(struct mount *, int); +void vfs_exjail_delete(struct prison *); int vfs_export /* process mount export info */ - (struct mount *, struct export_args *); + (struct mount *, struct export_args *, bool); +void vfs_free_addrlist(struct netexport *); void vfs_allocate_syncvnode(struct mount *); void vfs_deallocate_syncvnode(struct mount *); int vfs_donmount(struct thread *td, uint64_t fsflags,