--- sys/compat/cloudabi/cloudabi_fd.c.orig +++ sys/compat/cloudabi/cloudabi_fd.c @@ -389,7 +389,7 @@ int error, oflags; /* Obtain file descriptor properties. */ - error = fget_cap(td, uap->fd, cap_rights_init(&rights), &fp, + error = fget_cap(td, uap->fd, cap_rights_init(&rights), NULL, &fp, &fcaps); if (error != 0) return (error); --- sys/fs/fdescfs/fdesc_vnops.c.orig +++ sys/fs/fdescfs/fdesc_vnops.c @@ -515,7 +515,7 @@ cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); } else { error = getvnode_path(td, fd, - cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_SET), NULL, &fp); } if (error) { /* @@ -652,7 +652,7 @@ VOP_UNLOCK(vn); td = curthread; - error = fget_cap(td, fd_fd, &cap_no_rights, &fp, NULL); + error = fget_cap(td, fd_fd, &cap_no_rights, NULL, &fp, NULL); if (error != 0) goto out; --- sys/kern/kern_descrip.c.orig +++ sys/kern/kern_descrip.c @@ -114,7 +114,8 @@ static void fdunused(struct filedesc *fdp, int fd); static void fdused(struct filedesc *fdp, int fd); static int fget_unlocked_seq(struct filedesc *fdp, int fd, - cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp); + const cap_rights_t *needrightsp, uint8_t *flagsp, + struct file **fpp, seqc_t *seqp); static int getmaxfd(struct thread *td); static u_long *filecaps_copy_prep(const struct filecaps *src); static void filecaps_copy_finish(const struct filecaps *src, @@ -470,6 +471,8 @@ return (error); } +#define FD_RESOLVE_BENEATH 2 + int kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) { @@ -519,7 +522,9 @@ fde = fdeget_locked(fdp, fd); if (fde != NULL) { td->td_retval[0] = - (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; + ((fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0) | + ((fde->fde_flags & UF_RESOLVE_BENEATH) ? + FD_RESOLVE_BENEATH : 0); error = 0; } FILEDESC_SUNLOCK(fdp); @@ -530,8 +535,13 @@ FILEDESC_XLOCK(fdp); fde = fdeget_locked(fdp, fd); if (fde != NULL) { + /* + * UF_RESOLVE_BENEATH is sticky and cannot be cleared. + */ fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | - (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); + ((arg & FD_CLOEXEC) != 0 ? UF_EXCLOSE : 0) | + ((arg & FD_RESOLVE_BENEATH) != 0 ? + UF_RESOLVE_BENEATH : 0); error = 0; } FILEDESC_XUNLOCK(fdp); @@ -2158,7 +2168,8 @@ seqc_write_begin(&fde->fde_seqc); #endif fde->fde_file = fp; - fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; + fde->fde_flags = ((flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0) | + ((flags & O_RESOLVE_BENEATH) != 0 ? UF_RESOLVE_BENEATH : 0); if (fcaps != NULL) filecaps_move(fcaps, &fde->fde_caps); else @@ -2978,7 +2989,7 @@ } int -fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, +fget_cap_locked(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, struct file **fpp, struct filecaps *havecapsp) { struct filedescent *fde; @@ -3010,8 +3021,8 @@ } int -fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, - struct file **fpp, struct filecaps *havecapsp) +fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, + uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp) { struct filedesc *fdp = td->td_proc->p_fd; int error; @@ -3025,7 +3036,8 @@ *fpp = NULL; for (;;) { - error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq); + error = fget_unlocked_seq(fdp, fd, needrightsp, flagsp, &fp, + &seq); if (error != 0) return (error); @@ -3089,7 +3101,7 @@ #ifdef CAPABILITIES int -fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) +fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, int *flagsp) { const struct filedescent *fde; const struct fdescenttbl *fdt; @@ -3099,6 +3111,7 @@ const cap_rights_t *haverights; cap_rights_t rights; seqc_t seq; + int flags; VFS_SMR_ASSERT_ENTERED(); @@ -3117,7 +3130,9 @@ return (EAGAIN); if (__predict_false(cap_check_inline_transient(haverights, &rights))) return (EAGAIN); - *fsearch = ((fp->f_flag & FSEARCH) != 0); + flags = fp->f_flag & FSEARCH; + flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? + O_RESOLVE_BENEATH : 0; vp = fp->f_vnode; if (__predict_false(vp == NULL)) { return (EAGAIN); @@ -3151,16 +3166,19 @@ #endif } *vpp = vp; + *flagsp = flags; return (0); } #else int -fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) +fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, int *flagsp) { + const struct filedescent *fde; const struct fdescenttbl *fdt; struct filedesc *fdp; struct file *fp; struct vnode *vp; + int flags; VFS_SMR_ASSERT_ENTERED(); @@ -3168,10 +3186,13 @@ fdt = fdp->fd_files; if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) return (EBADF); - fp = fdt->fdt_ofiles[fd].fde_file; + fde = &fdt->fdt_ofiles[fd]; + fp = fde->fde_file; if (__predict_false(fp == NULL)) return (EAGAIN); - *fsearch = ((fp->f_flag & FSEARCH) != 0); + flags = fp->f_flag & FSEARCH; + flags |= (fde->fde_flags & UF_RESOLVE_BENEATH) != 0 ? + O_RESOLVE_BENEATH : 0; vp = fp->f_vnode; if (__predict_false(vp == NULL || vp->v_type != VDIR)) { return (EAGAIN); @@ -3186,6 +3207,7 @@ return (EAGAIN); filecaps_fill(&ndp->ni_filecaps); *vpp = vp; + *flagsp = flags; return (0); } #endif @@ -3199,13 +3221,15 @@ struct componentname *cnp; cap_rights_t rights; int error; + uint8_t flags; td = curthread; rights = *ndp->ni_rightsneeded; cap_rights_set_one(&rights, CAP_LOOKUP); cnp = &ndp->ni_cnd; - error = fget_cap(td, ndp->ni_dirfd, &rights, &fp, &ndp->ni_filecaps); + error = fget_cap(td, ndp->ni_dirfd, &rights, &flags, &fp, + &ndp->ni_filecaps); if (__predict_false(error != 0)) return (error); if (__predict_false(fp->f_ops == &badfileops)) { @@ -3223,6 +3247,10 @@ */ if ((fp->f_flag & FSEARCH) != 0) cnp->cn_flags |= NOEXECCHECK; + if ((flags & UF_RESOLVE_BENEATH) != 0) { + cnp->cn_flags |= RBENEATH; + ndp->ni_resflags |= NIRES_BENEATH; + } fdrop(fp, td); #ifdef CAPABILITIES @@ -3256,12 +3284,10 @@ } static int -fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, - struct file **fpp, seqc_t *seqp) +fget_unlocked_seq(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, + uint8_t *flagsp, struct file **fpp, seqc_t *seqp) { -#ifdef CAPABILITIES const struct filedescent *fde; -#endif const struct fdescenttbl *fdt; struct file *fp; #ifdef CAPABILITIES @@ -3269,6 +3295,7 @@ cap_rights_t haverights; int error; #endif + uint8_t flags; fdt = fdp->fd_files; if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) @@ -3287,10 +3314,13 @@ fde = &fdt->fdt_ofiles[fd]; haverights = *cap_rights_fde_inline(fde); fp = fde->fde_file; + flags = fde->fde_flags; if (!seqc_consistent(fd_seqc(fdt, fd), seq)) continue; #else - fp = fdt->fdt_ofiles[fd].fde_file; + fde = &fdt->fdt_ofiles[fd]; + flags = fde->fde_flags; + fp = fde->fde_file; #endif if (fp == NULL) return (EBADF); @@ -3323,6 +3353,8 @@ fdrop(fp, curthread); } *fpp = fp; + if (flagsp != NULL) + *flagsp = flags; if (seqp != NULL) { #ifdef CAPABILITIES *seqp = seq; @@ -3339,8 +3371,8 @@ * racing with itself. */ int -fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, - struct file **fpp) +fget_unlocked_flags(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, + uint8_t *flagsp, struct file **fpp) { #ifdef CAPABILITIES const struct filedescent *fde; @@ -3351,6 +3383,7 @@ seqc_t seq; const cap_rights_t *haverights; #endif + uint8_t flags; fdt = fdp->fd_files; if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) { @@ -3362,8 +3395,10 @@ fde = &fdt->fdt_ofiles[fd]; haverights = cap_rights_fde_inline(fde); fp = fde->fde_file; + flags = fde->fde_flags; #else fp = fdt->fdt_ofiles[fd].fde_file; + flags = fdt->fdt_ofiles[fd].fde_flags; #endif if (__predict_false(fp == NULL)) goto out_fallback; @@ -3387,12 +3422,21 @@ #endif goto out_fdrop; *fpp = fp; + if (flagsp != NULL) + *flagsp = flags; return (0); out_fdrop: fdrop(fp, curthread); out_fallback: *fpp = NULL; - return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL)); + return (fget_unlocked_seq(fdp, fd, needrightsp, flagsp, fpp, NULL)); +} + +int +fget_unlocked(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, + struct file **fpp) +{ + return (fget_unlocked_flags(fdp, fd, needrightsp, NULL, fpp)); } /* @@ -3406,7 +3450,7 @@ */ #ifdef CAPABILITIES int -fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, +fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, struct file **fpp) { const struct filedescent *fde; @@ -3436,7 +3480,7 @@ } #else int -fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, +fget_only_user(struct filedesc *fdp, int fd, const cap_rights_t *needrightsp, struct file **fpp) { struct file *fp; @@ -3472,7 +3516,7 @@ */ static __inline int _fget(struct thread *td, int fd, struct file **fpp, int flags, - cap_rights_t *needrightsp) + const cap_rights_t *needrightsp) { struct filedesc *fdp; struct file *fp; @@ -3520,15 +3564,15 @@ } int -fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) +fget(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, 0, rightsp)); } int -fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, - struct file **fpp) +fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp, + vm_prot_t *maxprotp, struct file **fpp) { int error; #ifndef CAPABILITIES @@ -3546,7 +3590,7 @@ fdp = td->td_proc->p_fd; MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); for (;;) { - error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); + error = fget_unlocked_seq(fdp, fd, rightsp, NULL, &fp, &seq); if (__predict_false(error != 0)) return (error); if (__predict_false(fp->f_ops == &badfileops)) { @@ -3571,22 +3615,24 @@ } int -fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) +fget_read(struct thread *td, int fd, const cap_rights_t *rightsp, + struct file **fpp) { return (_fget(td, fd, fpp, FREAD, rightsp)); } int -fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) +fget_write(struct thread *td, int fd, const cap_rights_t *rightsp, + struct file **fpp) { return (_fget(td, fd, fpp, FWRITE, rightsp)); } int -fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, - struct file **fpp) +fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp, + int needfcntl, struct file **fpp) { struct filedesc *fdp = td->td_proc->p_fd; #ifndef CAPABILITIES @@ -3599,7 +3645,7 @@ *fpp = NULL; MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); for (;;) { - error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); + error = fget_unlocked_seq(fdp, fd, rightsp, NULL, &fp, &seq); if (error != 0) return (error); error = cap_fcntl_check(fdp, fd, needfcntl); @@ -3624,7 +3670,7 @@ * XXX: what about the unused flags ? */ static __inline int -_fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, +_fgetvp(struct thread *td, int fd, int flags, const cap_rights_t *needrightsp, struct vnode **vpp) { struct file *fp; @@ -3646,21 +3692,22 @@ } int -fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) +fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp, + struct vnode **vpp) { return (_fgetvp(td, fd, 0, rightsp, vpp)); } int -fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, +fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp) { struct filecaps caps; struct file *fp; int error; - error = fget_cap(td, fd, needrightsp, &fp, &caps); + error = fget_cap(td, fd, needrightsp, NULL, &fp, &caps); if (error != 0) return (error); if (fp->f_ops == &badfileops) { @@ -3685,14 +3732,16 @@ } int -fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) +fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp, + struct vnode **vpp) { return (_fgetvp(td, fd, FREAD, rightsp, vpp)); } int -fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) +fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp, + struct vnode **vpp) { return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); @@ -3700,7 +3749,7 @@ #ifdef notyet int -fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, +fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp, struct vnode **vpp) { --- sys/kern/sys_procdesc.c.orig +++ sys/kern/sys_procdesc.c @@ -121,7 +121,7 @@ * died. */ int -procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp, +procdesc_find(struct thread *td, int fd, const cap_rights_t *rightsp, struct proc **p) { struct procdesc *pd; @@ -168,7 +168,8 @@ * Retrieve the PID associated with a process descriptor. */ int -kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp) +kern_pdgetpid(struct thread *td, int fd, const cap_rights_t *rightsp, + pid_t *pidp) { struct file *fp; int error; --- sys/kern/uipc_mqueue.c.orig +++ sys/kern/uipc_mqueue.c @@ -2160,13 +2160,14 @@ return (error); } -typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **); +typedef int (*_fgetf)(struct thread *, int, const cap_rights_t *, + struct file **); /* * Get message queue by giving file slot */ static int -_getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func, +_getmq(struct thread *td, int fd, const cap_rights_t *rightsp, _fgetf func, struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq) { struct mqfs_node *pn; --- sys/kern/uipc_sem.c.orig +++ sys/kern/uipc_sem.c @@ -123,8 +123,8 @@ semid_t *semidp, mode_t mode, unsigned int value, int flags, int compat32); static void ksem_drop(struct ksem *ks); -static int ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp, - struct file **fpp); +static int ksem_get(struct thread *td, semid_t id, + const cap_rights_t *rightsp, struct file **fpp); static struct ksem *ksem_hold(struct ksem *ks); static void ksem_insert(char *path, Fnv32_t fnv, struct ksem *ks); static struct ksem *ksem_lookup(char *path, Fnv32_t fnv); @@ -588,7 +588,7 @@ } static int -ksem_get(struct thread *td, semid_t id, cap_rights_t *rightsp, +ksem_get(struct thread *td, semid_t id, const cap_rights_t *rightsp, struct file **fpp) { struct ksem *ks; --- sys/kern/uipc_syscalls.c.orig +++ sys/kern/uipc_syscalls.c @@ -91,13 +91,13 @@ * A reference on the file entry is held upon returning. */ int -getsock_cap(struct thread *td, int fd, cap_rights_t *rightsp, +getsock_cap(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp, u_int *fflagp, struct filecaps *havecapsp) { struct file *fp; int error; - error = fget_cap(td, fd, rightsp, &fp, havecapsp); + error = fget_cap(td, fd, rightsp, NULL, &fp, havecapsp); if (error != 0) return (error); if (fp->f_type != DTYPE_SOCKET) { @@ -727,7 +727,7 @@ struct uio auio; struct iovec *iov; struct socket *so; - cap_rights_t *rights; + const cap_rights_t *rights; #ifdef KTRACE struct uio *ktruio = NULL; #endif --- sys/kern/uipc_usrreq.c.orig +++ sys/kern/uipc_usrreq.c @@ -57,7 +57,6 @@ * need a proper out-of-band */ -#include #include "opt_ddb.h" #include @@ -67,6 +66,7 @@ #include #include #include +#include #include #include #include @@ -1993,22 +1993,34 @@ free(fdep[0], M_FILECAPS); } +static bool +restrict_rights(struct file *fp, struct thread *td) +{ + struct prison *prison1, *prison2; + + prison1 = fp->f_cred->cr_prison; + prison2 = td->td_ucred->cr_prison; + return (prison1 != prison2 && prison1->pr_root != prison2->pr_root && + prison2 != &prison0); +} + static int unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) { struct thread *td = curthread; /* XXX */ struct cmsghdr *cm = mtod(control, struct cmsghdr *); - int i; int *fdp; struct filedesc *fdesc = td->td_proc->p_fd; struct filedescent **fdep; void *data; socklen_t clen = control->m_len, datalen; - int error, newfds; + int error, fdflags, newfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); + fdflags = (flags & MSG_CMSG_CLOEXEC) ? O_CLOEXEC : 0; + error = 0; if (controlp != NULL) /* controlp == NULL => free control messages */ *controlp = NULL; @@ -2059,11 +2071,14 @@ *controlp = NULL; goto next; } - for (i = 0; i < newfds; i++, fdp++) { - _finstall(fdesc, fdep[i]->fde_file, *fdp, - (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0, - &fdep[i]->fde_caps); - unp_externalize_fp(fdep[i]->fde_file); + for (int i = 0; i < newfds; i++, fdp++) { + struct file *fp; + + fp = fdep[i]->fde_file; + _finstall(fdesc, fp, *fdp, fdflags | + (restrict_rights(fp, td) ? + O_RESOLVE_BENEATH : 0), &fdep[i]->fde_caps); + unp_externalize_fp(fp); } /* --- sys/kern/vfs_acl.c.orig +++ sys/kern/vfs_acl.c @@ -433,7 +433,7 @@ AUDIT_ARG_FD(uap->filedes); error = getvnode_path(td, uap->filedes, - cap_rights_init_one(&rights, CAP_ACL_GET), &fp); + cap_rights_init_one(&rights, CAP_ACL_GET), NULL, &fp); if (error == 0) { error = vacl_get_acl(td, fp->f_vnode, uap->type, uap->aclp); fdrop(fp, td); @@ -566,7 +566,7 @@ AUDIT_ARG_FD(uap->filedes); error = getvnode_path(td, uap->filedes, - cap_rights_init_one(&rights, CAP_ACL_CHECK), &fp); + cap_rights_init_one(&rights, CAP_ACL_CHECK), NULL, &fp); if (error == 0) { error = vacl_aclcheck(td, fp->f_vnode, uap->type, uap->aclp); fdrop(fp, td); --- sys/kern/vfs_cache.c.orig +++ sys/kern/vfs_cache.c @@ -4364,17 +4364,23 @@ { struct nameidata *ndp; struct componentname *cnp; - int error; - bool fsearch; + int error, flags; ndp = fpl->ndp; cnp = fpl->cnp; - error = fgetvp_lookup_smr(ndp->ni_dirfd, ndp, vpp, &fsearch); + error = fgetvp_lookup_smr(ndp->ni_dirfd, ndp, vpp, &flags); if (__predict_false(error != 0)) { return (cache_fpl_aborted(fpl)); } - fpl->fsearch = fsearch; + if (__predict_false((flags & O_RESOLVE_BENEATH) != 0)) { + _Static_assert((CACHE_FPL_SUPPORTED_CN_FLAGS & RBENEATH) == 0, + "RBENEATH supported by fplookup"); + cache_fpl_smr_exit(fpl); + cache_fpl_aborted(fpl); + return (EOPNOTSUPP); + } + fpl->fsearch = (flags & FSEARCH) != 0; if ((*vpp)->v_type != VDIR) { if (!((cnp->cn_flags & EMPTYPATH) != 0 && cnp->cn_pnbuf[0] == '\0')) { cache_fpl_smr_exit(fpl); --- sys/kern/vfs_extattr.c.orig +++ sys/kern/vfs_extattr.c @@ -241,7 +241,7 @@ AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, - cap_rights_init_one(&rights, CAP_EXTATTR_SET), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_SET), NULL, &fp); if (error) return (error); @@ -408,7 +408,7 @@ AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, - cap_rights_init_one(&rights, CAP_EXTATTR_GET), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_GET), NULL, &fp); if (error) return (error); @@ -543,7 +543,7 @@ AUDIT_ARG_TEXT(attrname); error = getvnode_path(td, uap->fd, - cap_rights_init_one(&rights, CAP_EXTATTR_DELETE), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_DELETE), NULL, &fp); if (error) return (error); @@ -689,7 +689,7 @@ AUDIT_ARG_FD(uap->fd); AUDIT_ARG_VALUE(uap->attrnamespace); error = getvnode_path(td, uap->fd, - cap_rights_init_one(&rights, CAP_EXTATTR_LIST), &fp); + cap_rights_init_one(&rights, CAP_EXTATTR_LIST), NULL, &fp); if (error) return (error); --- sys/kern/vfs_syscalls.c.orig +++ sys/kern/vfs_syscalls.c @@ -373,7 +373,7 @@ int error; AUDIT_ARG_FD(fd); - error = getvnode_path(td, fd, &cap_fstatfs_rights, &fp); + error = getvnode_path(td, fd, &cap_fstatfs_rights, NULL, &fp); if (error != 0) return (error); vp = fp->f_vnode; @@ -887,12 +887,17 @@ struct mount *mp; struct file *fp; int error; + uint8_t fdflags; AUDIT_ARG_FD(uap->fd); - error = getvnode_path(td, uap->fd, &cap_fchdir_rights, + error = getvnode_path(td, uap->fd, &cap_fchdir_rights, &fdflags, &fp); if (error != 0) return (error); + if ((fdflags & UF_RESOLVE_BENEATH) != 0) { + fdrop(fp, td); + return (ENOTCAPABLE); + } vp = fp->f_vnode; vrefact(vp); fdrop(fp, td); @@ -1243,6 +1248,10 @@ else #endif fcaps = NULL; + if ((nd.ni_resflags & NIRES_BENEATH) != 0) + flags |= O_RESOLVE_BENEATH; + else + flags &= ~O_RESOLVE_BENEATH; error = finstall_refed(td, fp, &indx, flags, fcaps); /* On success finstall_refed() consumes fcaps. */ if (error != 0) { @@ -1933,7 +1942,7 @@ fp = NULL; if (fd != FD_NONE) { - error = getvnode_path(td, fd, &cap_no_rights, &fp); + error = getvnode_path(td, fd, &cap_no_rights, NULL, &fp); if (error != 0) return (error); } @@ -4315,13 +4324,14 @@ * semantics. */ int -getvnode_path(struct thread *td, int fd, cap_rights_t *rightsp, - struct file **fpp) +getvnode_path(struct thread *td, int fd, const cap_rights_t *rightsp, + uint8_t *flagsp, struct file **fpp) { struct file *fp; int error; - error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp); + error = fget_unlocked_flags(td->td_proc->p_fd, fd, rightsp, flagsp, + &fp); if (error != 0) return (error); @@ -4353,11 +4363,12 @@ * A reference on the file entry is held upon returning. */ int -getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) +getvnode(struct thread *td, int fd, const cap_rights_t *rightsp, + struct file **fpp) { int error; - error = getvnode_path(td, fd, rightsp, fpp); + error = getvnode_path(td, fd, rightsp, NULL, fpp); /* * Filter out O_PATH file descriptors, most getvnode() callers --- sys/sys/file.h.orig +++ sys/sys/file.h @@ -251,14 +251,15 @@ extern int maxfiles; /* kernel limit on number of open files */ extern int maxfilesperproc; /* per process limit on number of open files */ -int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); -int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, +int fget(struct thread *td, int fd, const cap_rights_t *rightsp, + struct file **fpp); +int fget_mmap(struct thread *td, int fd, const cap_rights_t *rightsp, vm_prot_t *maxprotp, struct file **fpp); -int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, +int fget_read(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp); -int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, +int fget_write(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp); -int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, +int fget_fcntl(struct thread *td, int fd, const cap_rights_t *rightsp, int needfcntl, struct file **fpp); int _fdrop(struct file *fp, struct thread *td); int fget_remote(struct thread *td, struct proc *p, int fd, struct file **fpp); @@ -281,17 +282,17 @@ void finit(struct file *, u_int, short, void *, struct fileops *); void finit_vnode(struct file *, u_int, void *, struct fileops *); -int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, +int fgetvp(struct thread *td, int fd, const cap_rights_t *rightsp, struct vnode **vpp); -int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, +int fgetvp_exec(struct thread *td, int fd, const cap_rights_t *rightsp, struct vnode **vpp); -int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, +int fgetvp_rights(struct thread *td, int fd, const cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp); -int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, +int fgetvp_read(struct thread *td, int fd, const cap_rights_t *rightsp, struct vnode **vpp); -int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, +int fgetvp_write(struct thread *td, int fd, const cap_rights_t *rightsp, struct vnode **vpp); -int fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch); +int fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, int *flagsp); int fgetvp_lookup(int fd, struct nameidata *ndp, struct vnode **vpp); static __inline __result_use_check bool --- sys/sys/filedesc.h.orig +++ sys/sys/filedesc.h @@ -136,6 +136,7 @@ * Per-process open flags. */ #define UF_EXCLOSE 0x01 /* auto-close on exec */ +#define UF_RESOLVE_BENEATH 0x02 /* lookups must be beneath this dir */ #ifdef _KERNEL @@ -267,22 +268,26 @@ struct filedesc_to_leader * filedesc_to_leader_share(struct filedesc_to_leader *fdtol, struct filedesc *fdp); -int getvnode(struct thread *td, int fd, cap_rights_t *rightsp, - struct file **fpp); -int getvnode_path(struct thread *td, int fd, cap_rights_t *rightsp, +int getvnode(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp); +int getvnode_path(struct thread *td, int fd, const cap_rights_t *rightsp, + uint8_t *flagsp, struct file **fpp); void mountcheckdirs(struct vnode *olddp, struct vnode *newdp); -int fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, - struct file **fpp, struct filecaps *havecapsp); -int fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, - struct file **fpp, struct filecaps *havecapsp); +int fget_cap_locked(struct filedesc *fdp, int fd, + const cap_rights_t *needrightsp, struct file **fpp, + struct filecaps *havecapsp); +int fget_cap(struct thread *td, int fd, const cap_rights_t *needrightsp, + uint8_t *flagsp, struct file **fpp, struct filecaps *havecapsp); /* Return a referenced file from an unlocked descriptor. */ -int fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, +int fget_unlocked(struct filedesc *fdp, int fd, + const cap_rights_t *needrightsp, struct file **fpp); +int fget_unlocked_flags(struct filedesc *fdp, int fd, + const cap_rights_t *needrightsp, uint8_t *flagsp, struct file **fpp); /* Return a file pointer without a ref. FILEDESC_IS_ONLY_USER must be true. */ -int fget_only_user(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, - struct file **fpp); +int fget_only_user(struct filedesc *fdp, int fd, + const cap_rights_t *needrightsp, struct file **fpp); #define fput_only_user(fdp, fp) ({ \ MPASS(FILEDESC_IS_ONLY_USER(fdp)); \ MPASS(refcount_load(&fp->f_count) > 0); \ --- sys/sys/namei.h.orig +++ sys/sys/namei.h @@ -72,7 +72,7 @@ */ const char *ni_dirp; /* pathname pointer */ enum uio_seg ni_segflg; /* location of pathname */ - cap_rights_t *ni_rightsneeded; /* rights required to look up vnode */ + const cap_rights_t *ni_rightsneeded; /* rights needed to look up vnode */ /* * Arguments to lookup. */ @@ -208,6 +208,7 @@ #define NIRES_ABS 0x00000001 /* Path was absolute */ #define NIRES_STRICTREL 0x00000002 /* Restricted lookup result */ #define NIRES_EMPTYPATH 0x00000004 /* EMPTYPATH used */ +#define NIRES_BENEATH 0x00000008 /* O_RESOLVE_BENEATH is to be inherited */ /* * Flags in ni_lcf, valid for the duration of the namei call. @@ -250,7 +251,7 @@ #define NDINIT_ALL(ndp, op, flags, segflg, namep, dirfd, startdir, rightsp, td) \ do { \ struct nameidata *_ndp = (ndp); \ - cap_rights_t *_rightsp = (rightsp); \ + const cap_rights_t *_rightsp = (rightsp); \ MPASS(_rightsp != NULL); \ NDINIT_PREFILL(_ndp); \ NDINIT_DBG(_ndp); \ --- sys/sys/procdesc.h.orig +++ sys/sys/procdesc.h @@ -94,8 +94,10 @@ * In-kernel interfaces to process descriptors. */ int procdesc_exit(struct proc *); -int procdesc_find(struct thread *, int fd, cap_rights_t *, struct proc **); -int kern_pdgetpid(struct thread *, int fd, cap_rights_t *, pid_t *pidp); +int procdesc_find(struct thread *, int fd, const cap_rights_t *, + struct proc **); +int kern_pdgetpid(struct thread *, int fd, const cap_rights_t *, + pid_t *pidp); void procdesc_new(struct proc *, int); void procdesc_finit(struct procdesc *, struct file *); pid_t procdesc_pid(struct file *); --- sys/sys/socketvar.h.orig +++ sys/sys/socketvar.h @@ -418,7 +418,7 @@ */ int getsockaddr(struct sockaddr **namp, const struct sockaddr *uaddr, size_t len); -int getsock_cap(struct thread *td, int fd, cap_rights_t *rightsp, +int getsock_cap(struct thread *td, int fd, const cap_rights_t *rightsp, struct file **fpp, u_int *fflagp, struct filecaps *havecaps); void soabort(struct socket *so); int soaccept(struct socket *so, struct sockaddr **nam); --- tests/sys/kern/Makefile.orig +++ tests/sys/kern/Makefile @@ -75,6 +75,7 @@ LIBADD.sendfile_helper+= pthread LIBADD.fdgrowtable_test+= util pthread kvm procstat LIBADD.sigwait+= rt +LIBADD.unix_passfd_test+= jail NETBSD_ATF_TESTS_C+= lockf_test NETBSD_ATF_TESTS_C+= mqueue_test --- tests/sys/kern/unix_passfd_test.c.orig +++ tests/sys/kern/unix_passfd_test.c @@ -25,15 +25,18 @@ * SUCH DAMAGE. */ -#include -#include +#include +#include #include #include #include #include +#include +#include #include #include +#include #include #include #include @@ -713,6 +716,132 @@ (void)close(putfd); } +ATF_TC_WITH_CLEANUP(cross_jail_dirfd); +ATF_TC_HEAD(cross_jail_dirfd, tc) +{ + atf_tc_set_md_var(tc, "require.user", "root"); +} +ATF_TC_BODY(cross_jail_dirfd, tc) +{ + int error, sock[2], jid1, jid2, status; + pid_t pid1, pid2; + + domainsocketpair(sock); + + error = mkdir("./a", 0755); + ATF_REQUIRE(error == 0); + error = mkdir("./b", 0755); + ATF_REQUIRE(error == 0); + error = mkdir("./c", 0755); + ATF_REQUIRE(error == 0); + error = mkdir("./a/c", 0755); + ATF_REQUIRE(error == 0); + + jid1 = jail_setv(JAIL_CREATE, + "name", "passfd_test_cross_jail_dirfd1", + "path", "./a", + "persist", NULL, + NULL); + ATF_REQUIRE_MSG(jid1 >= 0, "jail_setv: %s", jail_errmsg); + + jid2 = jail_setv(JAIL_CREATE, + "name", "passfd_test_cross_jail_dirfd2", + "path", "./b", + "persist", NULL, + NULL); + ATF_REQUIRE_MSG(jid2 >= 0, "jail_setv: %s", jail_errmsg); + + pid1 = fork(); + ATF_REQUIRE(pid1 >= 0); + if (pid1 == 0) { + ssize_t len; + int dfd, error; + char ch; + + error = jail_attach(jid1); + if (error != 0) + err(1, "jail_attach"); + + dfd = open(".", O_RDONLY | O_DIRECTORY); + if (dfd < 0) + err(1, "open(\".\") in jail %d", jid1); + + ch = 0; + len = sendfd_payload(sock[0], dfd, &ch, sizeof(ch)); + if (len == -1) + err(1, "sendmsg"); + + _exit(0); + } + + pid2 = fork(); + ATF_REQUIRE(pid2 >= 0); + if (pid2 == 0) { + int dfd, dfd2, error, fd; + char ch; + + error = jail_attach(jid2); + if (error != 0) + err(1, "jail_attach"); + + /* Get a directory from outside the jail root. */ + recvfd_payload(sock[1], &dfd, &ch, sizeof(ch), + CMSG_SPACE(sizeof(int)), 0); + + if ((fcntl(dfd, F_GETFD) & 2) == 0) + errx(1, "dfd does not have FD_RESOLVE_BENEATH set"); + + /* Make sure we can't chdir. */ + error = fchdir(dfd); + if (error == 0) + errx(1, "fchdir succeeded"); + if (errno != ENOTCAPABLE) + err(1, "fchdir"); + + /* Make sure a dotdot access fails. */ + fd = openat(dfd, "../c", O_RDONLY | O_DIRECTORY); + if (fd >= 0) + errx(1, "openat(\"../c\") succeeded"); + if (errno != ENOTCAPABLE) + err(1, "openat"); + + /* Accesses within the sender's jail root are ok. */ + fd = openat(dfd, "c", O_RDONLY | O_DIRECTORY); + if (fd < 0) + err(1, "openat(\"c\")"); + + dfd2 = openat(dfd, "", O_EMPTY_PATH | O_RDONLY | O_DIRECTORY); + if (dfd2 < 0) + err(1, "openat(\"\")"); + if ((fcntl(dfd2, F_GETFD) & 2) == 0) + errx(1, "dfd2 does not have FD_RESOLVE_BENEATH set"); + + _exit(0); + } + + error = waitpid(pid1, &status, 0); + ATF_REQUIRE(error != -1); + ATF_REQUIRE(WIFEXITED(status)); + ATF_REQUIRE(WEXITSTATUS(status) == 0); + error = waitpid(pid2, &status, 0); + ATF_REQUIRE(error != -1); + ATF_REQUIRE(WIFEXITED(status)); + ATF_REQUIRE(WEXITSTATUS(status) == 0); + + closesocketpair(sock); +} +ATF_TC_CLEANUP(cross_jail_dirfd, tc) +{ + int jid; + + jid = jail_getid("passfd_test_cross_jail_dirfd1"); + if (jid >= 0 && jail_remove(jid) != 0) + err(1, "jail_remove"); + jid = jail_getid("passfd_test_cross_jail_dirfd2"); + if (jid >= 0 && jail_remove(jid) != 0) + err(1, "jail_remove"); +} + ATF_TP_ADD_TCS(tp) { @@ -728,6 +857,7 @@ ATF_TP_ADD_TC(tp, truncated_rights); ATF_TP_ADD_TC(tp, copyout_rights_error); ATF_TP_ADD_TC(tp, empty_rights_message); + ATF_TP_ADD_TC(tp, cross_jail_dirfd); return (atf_no_error()); }