Index: projects/fuse2/sys/fs/fuse/fuse_internal.c =================================================================== --- projects/fuse2/sys/fs/fuse/fuse_internal.c (revision 346100) +++ projects/fuse2/sys/fs/fuse/fuse_internal.c (revision 346101) @@ -1,750 +1,745 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fuse.h" #include "fuse_file.h" #include "fuse_internal.h" #include "fuse_ipc.h" #include "fuse_node.h" #include "fuse_file.h" SDT_PROVIDER_DECLARE(fuse); /* * Fuse trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(fuse, , internal, trace, "int", "char*"); #ifdef ZERO_PAD_INCOMPLETE_BUFS static int isbzero(void *buf, size_t len); #endif /* Synchronously send a FUSE_ACCESS operation */ int fuse_internal_access(struct vnode *vp, accmode_t mode, - struct fuse_access_param *facp, struct thread *td, struct ucred *cred) { int err = 0; uint32_t mask = F_OK; int dataflags; int vtype; struct mount *mp; struct fuse_dispatcher fdi; struct fuse_access_in *fai; struct fuse_data *data; mp = vnode_mount(vp); vtype = vnode_vtype(vp); data = fuse_get_mpdata(mp); dataflags = data->dataflags; if (mode & VMODIFY_PERMS && vfs_isrdonly(mp)) { switch (vp->v_type) { case VDIR: /* FALLTHROUGH */ case VLNK: /* FALLTHROUGH */ case VREG: return EROFS; default: break; } } /* Unless explicitly permitted, deny everyone except the fs owner. */ - if (!(facp->facc_flags)) { - if (!(dataflags & FSESS_DAEMON_CAN_SPY)) { - int denied = fuse_match_cred(data->daemoncred, cred); - - if (denied) - return EPERM; - } + if (!(dataflags & FSESS_DAEMON_CAN_SPY)) { + if (fuse_match_cred(data->daemoncred, cred)) + return EPERM; } if (dataflags & FSESS_DEFAULT_PERMISSIONS) { struct vattr va; fuse_internal_getattr(vp, &va, cred, td); return vaccess(vp->v_type, va.va_mode, va.va_uid, va.va_gid, mode, cred, NULL); } if (!fsess_isimpl(mp, FUSE_ACCESS)) return 0; if ((mode & (VWRITE | VAPPEND | VADMIN)) != 0) mask |= W_OK; if ((mode & VREAD) != 0) mask |= R_OK; if ((mode & VEXEC) != 0) mask |= X_OK; fdisp_init(&fdi, sizeof(*fai)); fdisp_make_vp(&fdi, FUSE_ACCESS, vp, td, cred); fai = fdi.indata; fai->mask = mask; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_ACCESS); err = 0; } return err; } /* * Cache FUSE attributes from attr, in attribute cache associated with vnode * 'vp'. Optionally, if argument 'vap' is not NULL, store a copy of the * converted attributes there as well. * * If the nominal attribute cache TTL is zero, do not cache on the 'vp' (but do * return the result to the caller). */ void fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr, uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap) { struct mount *mp; struct fuse_vnode_data *fvdat; struct vattr *vp_cache_at; mp = vnode_mount(vp); fvdat = VTOFUD(vp); fuse_validity_2_bintime(attr_valid, attr_valid_nsec, &fvdat->attr_cache_timeout); vp_cache_at = VTOVA(vp); if (vap == NULL && vp_cache_at == NULL) return; if (vap == NULL) vap = vp_cache_at; vattr_null(vap); vap->va_fsid = mp->mnt_stat.f_fsid.val[0]; vap->va_fileid = attr->ino; vap->va_mode = attr->mode & ~S_IFMT; vap->va_nlink = attr->nlink; vap->va_uid = attr->uid; vap->va_gid = attr->gid; vap->va_rdev = attr->rdev; vap->va_size = attr->size; /* XXX on i386, seconds are truncated to 32 bits */ vap->va_atime.tv_sec = attr->atime; vap->va_atime.tv_nsec = attr->atimensec; vap->va_mtime.tv_sec = attr->mtime; vap->va_mtime.tv_nsec = attr->mtimensec; vap->va_ctime.tv_sec = attr->ctime; vap->va_ctime.tv_nsec = attr->ctimensec; vap->va_blocksize = PAGE_SIZE; vap->va_type = IFTOVT(attr->mode); vap->va_bytes = attr->blocks * S_BLKSIZE; vap->va_flags = 0; if (vap != vp_cache_at && vp_cache_at != NULL) memcpy(vp_cache_at, vap, sizeof(*vap)); } /* fsync */ int fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio) { if (tick->tk_aw_ohead.error == ENOSYS) { fsess_set_notimpl(tick->tk_data->mp, fticket_opcode(tick)); } return 0; } int fuse_internal_fsync(struct vnode *vp, struct thread *td, int waitfor, bool datasync) { struct fuse_fsync_in *ffsi = NULL; struct fuse_dispatcher fdi; struct fuse_filehandle *fufh; struct fuse_vnode_data *fvdat = VTOFUD(vp); int op = FUSE_FSYNC; int err = 0; if (!fsess_isimpl(vnode_mount(vp), (vnode_vtype(vp) == VDIR ? FUSE_FSYNCDIR : FUSE_FSYNC))) { return 0; } if (vnode_isdir(vp)) op = FUSE_FSYNCDIR; fdisp_init(&fdi, sizeof(*ffsi)); /* * fsync every open file handle for this file, because we can't be sure * which file handle the caller is really referring to. */ LIST_FOREACH(fufh, &fvdat->handles, next) { if (ffsi == NULL) fdisp_make_vp(&fdi, op, vp, td, NULL); else fdisp_refresh_vp(&fdi, op, vp, td, NULL); ffsi = fdi.indata; ffsi->fh = fufh->fh_id; ffsi->fsync_flags = 0; if (datasync) ffsi->fsync_flags = 1; if (waitfor == MNT_WAIT) { err = fdisp_wait_answ(&fdi); } else { fuse_insert_callback(fdi.tick, fuse_internal_fsync_callback); fuse_insert_message(fdi.tick); } } fdisp_destroy(&fdi); return err; } /* readdir */ int fuse_internal_readdir(struct vnode *vp, struct uio *uio, struct fuse_filehandle *fufh, struct fuse_iov *cookediov) { int err = 0; struct fuse_dispatcher fdi; struct fuse_read_in *fri = NULL; if (uio_resid(uio) == 0) { return 0; } fdisp_init(&fdi, 0); /* * Note that we DO NOT have a UIO_SYSSPACE here (so no need for p2p * I/O). */ while (uio_resid(uio) > 0) { fdi.iosize = sizeof(*fri); if (fri == NULL) fdisp_make_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); else fdisp_refresh_vp(&fdi, FUSE_READDIR, vp, NULL, NULL); fri = fdi.indata; fri->fh = fufh->fh_id; fri->offset = uio_offset(uio); fri->size = MIN(uio->uio_resid, fuse_get_mpdata(vp->v_mount)->max_read); if ((err = fdisp_wait_answ(&fdi))) { break; } if ((err = fuse_internal_readdir_processdata(uio, fri->size, fdi.answ, fdi.iosize, cookediov))) { break; } } fdisp_destroy(&fdi); return ((err == -1) ? 0 : err); } int fuse_internal_readdir_processdata(struct uio *uio, size_t reqsize, void *buf, size_t bufsize, void *param) { int err = 0; int cou = 0; int bytesavail; size_t freclen; struct dirent *de; struct fuse_dirent *fudge; struct fuse_iov *cookediov = param; if (bufsize < FUSE_NAME_OFFSET) { return -1; } for (;;) { if (bufsize < FUSE_NAME_OFFSET) { err = -1; break; } fudge = (struct fuse_dirent *)buf; freclen = FUSE_DIRENT_SIZE(fudge); cou++; if (bufsize < freclen) { err = ((cou == 1) ? -1 : 0); break; } #ifdef ZERO_PAD_INCOMPLETE_BUFS if (isbzero(buf, FUSE_NAME_OFFSET)) { err = -1; break; } #endif if (!fudge->namelen || fudge->namelen > MAXNAMLEN) { err = EINVAL; break; } bytesavail = GENERIC_DIRSIZ((struct pseudo_dirent *) &fudge->namelen); if (bytesavail > uio_resid(uio)) { err = -1; break; } fiov_adjust(cookediov, bytesavail); bzero(cookediov->base, bytesavail); de = (struct dirent *)cookediov->base; de->d_fileno = fudge->ino; de->d_reclen = bytesavail; de->d_type = fudge->type; de->d_namlen = fudge->namelen; memcpy((char *)cookediov->base + sizeof(struct dirent) - MAXNAMLEN - 1, (char *)buf + FUSE_NAME_OFFSET, fudge->namelen); dirent_terminate(de); err = uiomove(cookediov->base, cookediov->len, uio); if (err) { break; } buf = (char *)buf + freclen; bufsize -= freclen; uio_setoffset(uio, fudge->off); } return err; } /* remove */ int fuse_internal_remove(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, enum fuse_opcode op) { struct fuse_dispatcher fdi; int err = 0; fdisp_init(&fdi, cnp->cn_namelen + 1); fdisp_make_vp(&fdi, op, dvp, cnp->cn_thread, cnp->cn_cred); memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); return err; } /* rename */ int fuse_internal_rename(struct vnode *fdvp, struct componentname *fcnp, struct vnode *tdvp, struct componentname *tcnp) { struct fuse_dispatcher fdi; struct fuse_rename_in *fri; int err = 0; fdisp_init(&fdi, sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 2); fdisp_make_vp(&fdi, FUSE_RENAME, fdvp, tcnp->cn_thread, tcnp->cn_cred); fri = fdi.indata; fri->newdir = VTOI(tdvp); memcpy((char *)fdi.indata + sizeof(*fri), fcnp->cn_nameptr, fcnp->cn_namelen); ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen] = '\0'; memcpy((char *)fdi.indata + sizeof(*fri) + fcnp->cn_namelen + 1, tcnp->cn_nameptr, tcnp->cn_namelen); ((char *)fdi.indata)[sizeof(*fri) + fcnp->cn_namelen + tcnp->cn_namelen + 1] = '\0'; err = fdisp_wait_answ(&fdi); fdisp_destroy(&fdi); return err; } /* strategy */ /* entity creation */ void fuse_internal_newentry_makerequest(struct mount *mp, uint64_t dnid, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, struct fuse_dispatcher *fdip) { fdip->iosize = bufsize + cnp->cn_namelen + 1; fdisp_make(fdip, op, mp, dnid, cnp->cn_thread, cnp->cn_cred); memcpy(fdip->indata, buf, bufsize); memcpy((char *)fdip->indata + bufsize, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[bufsize + cnp->cn_namelen] = '\0'; } int fuse_internal_newentry_core(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum vtype vtyp, struct fuse_dispatcher *fdip) { int err = 0; struct fuse_entry_out *feo; struct mount *mp = vnode_mount(dvp); if ((err = fdisp_wait_answ(fdip))) { return err; } feo = fdip->answ; if ((err = fuse_internal_checkentry(feo, vtyp))) { return err; } err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, vtyp); if (err) { fuse_internal_forget_send(mp, cnp->cn_thread, cnp->cn_cred, feo->nodeid, 1); return err; } fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL); return err; } int fuse_internal_newentry(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, enum vtype vtype) { int err; struct fuse_dispatcher fdi; struct mount *mp = vnode_mount(dvp); fdisp_init(&fdi, 0); fuse_internal_newentry_makerequest(mp, VTOI(dvp), cnp, op, buf, bufsize, &fdi); err = fuse_internal_newentry_core(dvp, vpp, cnp, vtype, &fdi); fdisp_destroy(&fdi); return err; } /* entity destruction */ int fuse_internal_forget_callback(struct fuse_ticket *ftick, struct uio *uio) { fuse_internal_forget_send(ftick->tk_data->mp, curthread, NULL, ((struct fuse_in_header *)ftick->tk_ms_fiov.base)->nodeid, 1); return 0; } void fuse_internal_forget_send(struct mount *mp, struct thread *td, struct ucred *cred, uint64_t nodeid, uint64_t nlookup) { struct fuse_dispatcher fdi; struct fuse_forget_in *ffi; /* * KASSERT(nlookup > 0, ("zero-times forget for vp #%llu", * (long long unsigned) nodeid)); */ fdisp_init(&fdi, sizeof(*ffi)); fdisp_make(&fdi, FUSE_FORGET, mp, nodeid, td, cred); ffi = fdi.indata; ffi->nlookup = nlookup; fuse_insert_message(fdi.tick); fdisp_destroy(&fdi); } /* Read a vnode's attributes from cache or fetch them from the fuse daemon */ int fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { struct fuse_dispatcher fdi; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct vattr *attrs; struct fuse_attr_out *fao; int err = 0; if ((attrs = VTOVA(vp)) != NULL) { /* struct copy */ *vap = *attrs; if ((fvdat->flag & FN_SIZECHANGE) != 0) vap->va_size = fvdat->filesize; return 0; } fdisp_init(&fdi, 0); if ((err = fdisp_simple_putget_vp(&fdi, FUSE_GETATTR, vp, td, cred))) { if (err == ENOENT) { fuse_internal_vnode_disappear(vp); } goto out; } fao = (struct fuse_attr_out *)fdi.answ; fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, fao->attr_valid_nsec, vap); if (vap->va_type != vnode_vtype(vp)) { fuse_internal_vnode_disappear(vp); err = ENOENT; goto out; } if ((fvdat->flag & FN_SIZECHANGE) != 0) vap->va_size = fvdat->filesize; if (vnode_isreg(vp) && (fvdat->flag & FN_SIZECHANGE) == 0) { /* * This is for those cases when the file size changed without us * knowing, and we want to catch up. */ off_t new_filesize = fao->attr.size; if (fvdat->filesize != new_filesize) { fuse_vnode_setsize(vp, cred, new_filesize); fvdat->flag &= ~FN_SIZECHANGE; } } out: fdisp_destroy(&fdi); return err; } void fuse_internal_vnode_disappear(struct vnode *vp) { struct fuse_vnode_data *fvdat = VTOFUD(vp); ASSERT_VOP_ELOCKED(vp, "fuse_internal_vnode_disappear"); fvdat->flag |= FN_REVOKED; bintime_clear(&fvdat->attr_cache_timeout); cache_purge(vp); } /* fuse start/stop */ int fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio) { int err = 0; struct fuse_data *data = tick->tk_data; struct fuse_init_out *fiio; if ((err = tick->tk_aw_ohead.error)) { goto out; } if ((err = fticket_pull(tick, uio))) { goto out; } fiio = fticket_resp(tick)->base; /* XXX: Do we want to check anything further besides this? */ if (fiio->major < 7) { SDT_PROBE2(fuse, , internal, trace, 1, "userpace version too low"); err = EPROTONOSUPPORT; goto out; } data->fuse_libabi_major = fiio->major; data->fuse_libabi_minor = fiio->minor; if (fuse_libabi_geq(data, 7, 5)) { if (fticket_resp(tick)->len == sizeof(struct fuse_init_out)) { data->max_write = fiio->max_write; } else { err = EINVAL; } } else { /* Old fix values */ data->max_write = 4096; } out: if (err) { fdata_set_dead(data); } FUSE_LOCK(); data->dataflags |= FSESS_INITED; wakeup(&data->ticketer); FUSE_UNLOCK(); return 0; } void fuse_internal_send_init(struct fuse_data *data, struct thread *td) { struct fuse_init_in *fiii; struct fuse_dispatcher fdi; fdisp_init(&fdi, sizeof(*fiii)); fdisp_make(&fdi, FUSE_INIT, data->mp, 0, td, NULL); fiii = fdi.indata; fiii->major = FUSE_KERNEL_VERSION; fiii->minor = FUSE_KERNEL_MINOR_VERSION; /* * fusefs currently doesn't do any readahead other than fetching whole * buffer cache block sized regions at once. So the max readahead is * the size of a buffer cache block. */ fiii->max_readahead = maxbcachebuf; fiii->flags = 0; fuse_insert_callback(fdi.tick, fuse_internal_init_callback); fuse_insert_message(fdi.tick); fdisp_destroy(&fdi); } #ifdef ZERO_PAD_INCOMPLETE_BUFS static int isbzero(void *buf, size_t len) { int i; for (i = 0; i < len; i++) { if (((char *)buf)[i]) return (0); } return (1); } #endif Index: projects/fuse2/sys/fs/fuse/fuse_internal.h =================================================================== --- projects/fuse2/sys/fs/fuse/fuse_internal.h (revision 346100) +++ projects/fuse2/sys/fs/fuse/fuse_internal.h (revision 346101) @@ -1,321 +1,300 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef _FUSE_INTERNAL_H_ #define _FUSE_INTERNAL_H_ #include #include #include #include #include "fuse_ipc.h" #include "fuse_node.h" static inline bool vfs_isrdonly(struct mount *mp) { return ((mp->mnt_flag & MNT_RDONLY) != 0); } static inline struct mount * vnode_mount(struct vnode *vp) { return (vp->v_mount); } static inline bool vnode_mountedhere(struct vnode *vp) { return (vp->v_mountedhere != NULL); } static inline enum vtype vnode_vtype(struct vnode *vp) { return (vp->v_type); } static inline bool vnode_isvroot(struct vnode *vp) { return ((vp->v_vflag & VV_ROOT) != 0); } static inline bool vnode_isreg(struct vnode *vp) { return (vp->v_type == VREG); } static inline bool vnode_isdir(struct vnode *vp) { return (vp->v_type == VDIR); } static inline bool vnode_islnk(struct vnode *vp) { return (vp->v_type == VLNK); } static inline ssize_t uio_resid(struct uio *uio) { return (uio->uio_resid); } static inline off_t uio_offset(struct uio *uio) { return (uio->uio_offset); } static inline void uio_setoffset(struct uio *uio, off_t offset) { uio->uio_offset = offset; } /* miscellaneous */ static inline bool fuse_isdeadfs(struct vnode *vp) { struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); return (data->dataflags & FSESS_DEAD); } static inline uint64_t fuse_iosize(struct vnode *vp) { return (vp->v_mount->mnt_stat.f_iosize); } /* * Make a cacheable timeout in bintime format value based on a fuse_attr_out * response */ static inline void fuse_validity_2_bintime(uint64_t attr_valid, uint32_t attr_valid_nsec, struct bintime *timeout) { struct timespec now, duration, timeout_ts; getnanouptime(&now); /* "+ 2" is the bound of attr_valid_nsec + now.tv_nsec */ /* Why oh why isn't there a TIME_MAX defined? */ if (attr_valid >= INT_MAX || attr_valid + now.tv_sec + 2 >= INT_MAX) { timeout->sec = INT_MAX; } else { duration.tv_sec = attr_valid; duration.tv_nsec = attr_valid_nsec; timespecadd(&duration, &now, &timeout_ts); timespec2bintime(&timeout_ts, timeout); } } /* * Make a cacheable timeout value in timespec format based on the fuse_entry_out * response */ static inline void fuse_validity_2_timespec(const struct fuse_entry_out *feo, struct timespec *timeout) { struct timespec duration, now; getnanouptime(&now); /* "+ 2" is the bound of entry_valid_nsec + now.tv_nsec */ if (feo->entry_valid >= INT_MAX || feo->entry_valid + now.tv_sec + 2 >= INT_MAX) { timeout->tv_sec = INT_MAX; } else { duration.tv_sec = feo->entry_valid; duration.tv_nsec = feo->entry_valid_nsec; timespecadd(&duration, &now, timeout); } } /* access */ - -#define FVP_ACCESS_NOOP 0x01 - -#define FACCESS_VA_VALID 0x01 -/* - * Caller must be the directory's owner, or the superuser, or the sticky bit - * must not be set - */ -#define FACCESS_STICKY 0x04 -/* Caller requires access to change file's owner */ -#define FACCESS_CHOWN 0x08 -#define FACCESS_SETGID 0x12 - -#define FACCESS_XQUERIES (FACCESS_STICKY | FACCESS_CHOWN | FACCESS_SETGID) - -struct fuse_access_param { - uid_t xuid; - gid_t xgid; - uint32_t facc_flags; -}; - static inline int fuse_match_cred(struct ucred *basecred, struct ucred *usercred) { if (basecred->cr_uid == usercred->cr_uid && basecred->cr_uid == usercred->cr_ruid && basecred->cr_uid == usercred->cr_svuid && basecred->cr_groups[0] == usercred->cr_groups[0] && basecred->cr_groups[0] == usercred->cr_rgid && basecred->cr_groups[0] == usercred->cr_svgid) return (0); return (EPERM); } int fuse_internal_access(struct vnode *vp, accmode_t mode, - struct fuse_access_param *facp, struct thread *td, struct ucred *cred); + struct thread *td, struct ucred *cred); /* attributes */ void fuse_internal_cache_attrs(struct vnode *vp, struct fuse_attr *attr, uint64_t attr_valid, uint32_t attr_valid_nsec, struct vattr *vap); /* fsync */ int fuse_internal_fsync(struct vnode *vp, struct thread *td, int waitfor, bool datasync); int fuse_internal_fsync_callback(struct fuse_ticket *tick, struct uio *uio); /* getattr */ int fuse_internal_getattr(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td); /* readdir */ struct pseudo_dirent { uint32_t d_namlen; }; int fuse_internal_readdir(struct vnode *vp, struct uio *uio, struct fuse_filehandle *fufh, struct fuse_iov *cookediov); int fuse_internal_readdir_processdata(struct uio *uio, size_t reqsize, void *buf, size_t bufsize, void *param); /* remove */ int fuse_internal_remove(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, enum fuse_opcode op); /* rename */ int fuse_internal_rename(struct vnode *fdvp, struct componentname *fcnp, struct vnode *tdvp, struct componentname *tcnp); /* revoke */ void fuse_internal_vnode_disappear(struct vnode *vp); /* strategy */ /* entity creation */ static inline int fuse_internal_checkentry(struct fuse_entry_out *feo, enum vtype vtyp) { if (vtyp != IFTOVT(feo->attr.mode)) { return (EINVAL); } if (feo->nodeid == FUSE_NULL_ID) { return (EINVAL); } if (feo->nodeid == FUSE_ROOT_ID) { return (EINVAL); } return (0); } int fuse_internal_newentry(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, enum vtype vtyp); void fuse_internal_newentry_makerequest(struct mount *mp, uint64_t dnid, struct componentname *cnp, enum fuse_opcode op, void *buf, size_t bufsize, struct fuse_dispatcher *fdip); int fuse_internal_newentry_core(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, enum vtype vtyp, struct fuse_dispatcher *fdip); /* entity destruction */ int fuse_internal_forget_callback(struct fuse_ticket *tick, struct uio *uio); void fuse_internal_forget_send(struct mount *mp, struct thread *td, struct ucred *cred, uint64_t nodeid, uint64_t nlookup); /* fuse start/stop */ int fuse_internal_init_callback(struct fuse_ticket *tick, struct uio *uio); void fuse_internal_send_init(struct fuse_data *data, struct thread *td); #endif /* _FUSE_INTERNAL_H_ */ Index: projects/fuse2/sys/fs/fuse/fuse_vnops.c =================================================================== --- projects/fuse2/sys/fs/fuse/fuse_vnops.c (revision 346100) +++ projects/fuse2/sys/fs/fuse/fuse_vnops.c (revision 346101) @@ -1,2401 +1,2416 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Google Inc. and Amit Singh * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are * met: * * * Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * Redistributions in binary form must reproduce the above * copyright notice, this list of conditions and the following disclaimer * in the documentation and/or other materials provided with the * distribution. * * Neither the name of Google Inc. nor the names of its * contributors may be used to endorse or promote products derived from * this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * Copyright (C) 2005 Csaba Henk. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "fuse.h" #include "fuse_file.h" #include "fuse_internal.h" #include "fuse_ipc.h" #include "fuse_node.h" #include "fuse_io.h" #include /* Maximum number of hardlinks to a single FUSE file */ #define FUSE_LINK_MAX UINT32_MAX SDT_PROVIDER_DECLARE(fuse); /* * Fuse trace probe: * arg0: verbosity. Higher numbers give more verbose messages * arg1: Textual message */ SDT_PROBE_DEFINE2(fuse, , vnops, trace, "int", "char*"); /* vnode ops */ static vop_access_t fuse_vnop_access; static vop_close_t fuse_vnop_close; static vop_create_t fuse_vnop_create; static vop_deleteextattr_t fuse_vnop_deleteextattr; static vop_fdatasync_t fuse_vnop_fdatasync; static vop_fsync_t fuse_vnop_fsync; static vop_getattr_t fuse_vnop_getattr; static vop_getextattr_t fuse_vnop_getextattr; static vop_inactive_t fuse_vnop_inactive; static vop_link_t fuse_vnop_link; static vop_listextattr_t fuse_vnop_listextattr; static vop_lookup_t fuse_vnop_lookup; static vop_mkdir_t fuse_vnop_mkdir; static vop_mknod_t fuse_vnop_mknod; static vop_open_t fuse_vnop_open; static vop_pathconf_t fuse_vnop_pathconf; static vop_read_t fuse_vnop_read; static vop_readdir_t fuse_vnop_readdir; static vop_readlink_t fuse_vnop_readlink; static vop_reclaim_t fuse_vnop_reclaim; static vop_remove_t fuse_vnop_remove; static vop_rename_t fuse_vnop_rename; static vop_rmdir_t fuse_vnop_rmdir; static vop_setattr_t fuse_vnop_setattr; static vop_setextattr_t fuse_vnop_setextattr; static vop_strategy_t fuse_vnop_strategy; static vop_symlink_t fuse_vnop_symlink; static vop_write_t fuse_vnop_write; static vop_getpages_t fuse_vnop_getpages; static vop_putpages_t fuse_vnop_putpages; static vop_print_t fuse_vnop_print; struct vop_vector fuse_vnops = { .vop_default = &default_vnodeops, .vop_access = fuse_vnop_access, .vop_close = fuse_vnop_close, .vop_create = fuse_vnop_create, .vop_deleteextattr = fuse_vnop_deleteextattr, .vop_fsync = fuse_vnop_fsync, .vop_fdatasync = fuse_vnop_fdatasync, .vop_getattr = fuse_vnop_getattr, .vop_getextattr = fuse_vnop_getextattr, .vop_inactive = fuse_vnop_inactive, .vop_link = fuse_vnop_link, .vop_listextattr = fuse_vnop_listextattr, .vop_lookup = fuse_vnop_lookup, .vop_mkdir = fuse_vnop_mkdir, .vop_mknod = fuse_vnop_mknod, .vop_open = fuse_vnop_open, .vop_pathconf = fuse_vnop_pathconf, .vop_read = fuse_vnop_read, .vop_readdir = fuse_vnop_readdir, .vop_readlink = fuse_vnop_readlink, .vop_reclaim = fuse_vnop_reclaim, .vop_remove = fuse_vnop_remove, .vop_rename = fuse_vnop_rename, .vop_rmdir = fuse_vnop_rmdir, .vop_setattr = fuse_vnop_setattr, .vop_setextattr = fuse_vnop_setextattr, .vop_strategy = fuse_vnop_strategy, .vop_symlink = fuse_vnop_symlink, .vop_write = fuse_vnop_write, .vop_getpages = fuse_vnop_getpages, .vop_putpages = fuse_vnop_putpages, .vop_print = fuse_vnop_print, }; static u_long fuse_lookup_cache_hits = 0; SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_hits, CTLFLAG_RD, &fuse_lookup_cache_hits, 0, "number of positive cache hits in lookup"); static u_long fuse_lookup_cache_misses = 0; SYSCTL_ULONG(_vfs_fusefs, OID_AUTO, lookup_cache_misses, CTLFLAG_RD, &fuse_lookup_cache_misses, 0, "number of cache misses in lookup"); /* * XXX: This feature is highly experimental and can bring to instabilities, * needs revisiting before to be enabled by default. */ static int fuse_reclaim_revoked = 0; SYSCTL_INT(_vfs_fusefs, OID_AUTO, reclaim_revoked, CTLFLAG_RW, &fuse_reclaim_revoked, 0, ""); uma_zone_t fuse_pbuf_zone; #define fuse_vm_page_lock(m) vm_page_lock((m)); #define fuse_vm_page_unlock(m) vm_page_unlock((m)); #define fuse_vm_page_lock_queues() ((void)0) #define fuse_vm_page_unlock_queues() ((void)0) +/* Check permission for extattr operations, much like extattr_check_cred */ +static int +fuse_extattr_check_cred(struct vnode *vp, int ns, struct ucred *cred, + struct thread *td, accmode_t accmode) +{ + struct mount *mp = vnode_mount(vp); + struct fuse_data *data = fuse_get_mpdata(mp); + + /* + * Kernel-invoked always succeeds. + */ + if (cred == NOCRED) + return (0); + + /* + * Do not allow privileged processes in jail to directly manipulate + * system attributes. + */ + switch (ns) { + case EXTATTR_NAMESPACE_SYSTEM: + if (data->dataflags & FSESS_DEFAULT_PERMISSIONS) { + return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); + } + /* FALLTHROUGH */ + case EXTATTR_NAMESPACE_USER: + return (fuse_internal_access(vp, accmode, td, cred)); + default: + return (EPERM); + } +} + /* Get a filehandle for a directory */ static int fuse_filehandle_get_dir(struct vnode *vp, struct fuse_filehandle **fufhp, struct ucred *cred, pid_t pid) { if (fuse_filehandle_get(vp, FREAD, fufhp, cred, pid) == 0) return 0; return fuse_filehandle_get(vp, FEXEC, fufhp, cred, pid); } /* Send FUSE_FLUSH for this vnode */ static int fuse_flush(struct vnode *vp, struct ucred *cred, pid_t pid, int fflag) { struct fuse_flush_in *ffi; struct fuse_filehandle *fufh; struct fuse_dispatcher fdi; struct thread *td = curthread; struct mount *mp = vnode_mount(vp); int err; if (!fsess_isimpl(vnode_mount(vp), FUSE_FLUSH)) return 0; err = fuse_filehandle_get(vp, fflag, &fufh, cred, pid); if (err) return err; fdisp_init(&fdi, sizeof(*ffi)); fdisp_make_vp(&fdi, FUSE_FLUSH, vp, td, cred); ffi = fdi.indata; ffi->fh = fufh->fh_id; err = fdisp_wait_answ(&fdi); if (err == ENOSYS) { fsess_set_notimpl(mp, FUSE_FLUSH); err = 0; } fdisp_destroy(&fdi); return err; } /* struct vnop_access_args { struct vnode *a_vp; #if VOP_ACCESS_TAKES_ACCMODE_T accmode_t a_accmode; #else int a_mode; #endif struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; int accmode = ap->a_accmode; struct ucred *cred = ap->a_cred; - struct fuse_access_param facp; struct fuse_data *data = fuse_get_mpdata(vnode_mount(vp)); int err; if (fuse_isdeadfs(vp)) { if (vnode_isvroot(vp)) { return 0; } return ENXIO; } if (!(data->dataflags & FSESS_INITED)) { if (vnode_isvroot(vp)) { if (priv_check_cred(cred, PRIV_VFS_ADMIN) || (fuse_match_cred(data->daemoncred, cred) == 0)) { return 0; } } return EBADF; } if (vnode_islnk(vp)) { return 0; } - bzero(&facp, sizeof(facp)); - err = fuse_internal_access(vp, accmode, &facp, ap->a_td, ap->a_cred); + err = fuse_internal_access(vp, accmode, ap->a_td, ap->a_cred); return err; } /* struct vop_close_args { struct vnode *a_vp; int a_fflag; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; struct ucred *cred = ap->a_cred; int fflag = ap->a_fflag; struct thread *td = ap->a_td; pid_t pid = td->td_proc->p_pid; int err = 0; if (fuse_isdeadfs(vp)) return 0; if (vnode_isdir(vp)) return 0; if (fflag & IO_NDELAY) return 0; err = fuse_flush(vp, cred, pid, fflag); /* TODO: close the file handle, if we're sure it's no longer used */ if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { fuse_vnode_savesize(vp, cred, td->td_proc->p_pid); } return err; } static void fdisp_make_mknod_for_fallback( struct fuse_dispatcher *fdip, struct componentname *cnp, struct vnode *dvp, uint64_t parentnid, struct thread *td, struct ucred *cred, mode_t mode, enum fuse_opcode *op) { struct fuse_mknod_in *fmni; fdisp_init(fdip, sizeof(*fmni) + cnp->cn_namelen + 1); *op = FUSE_MKNOD; fdisp_make(fdip, *op, vnode_mount(dvp), parentnid, td, cred); fmni = fdip->indata; fmni->mode = mode; fmni->rdev = 0; memcpy((char *)fdip->indata + sizeof(*fmni), cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[sizeof(*fmni) + cnp->cn_namelen] = '\0'; } /* struct vnop_create_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_create(struct vop_create_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; struct thread *td = cnp->cn_thread; struct ucred *cred = cnp->cn_cred; struct fuse_open_in *foi; struct fuse_entry_out *feo; struct fuse_open_out *foo; struct fuse_dispatcher fdi, fdi2; struct fuse_dispatcher *fdip = &fdi; struct fuse_dispatcher *fdip2 = NULL; int err; struct mount *mp = vnode_mount(dvp); uint64_t parentnid = VTOFUD(dvp)->nid; mode_t mode = MAKEIMODE(vap->va_type, vap->va_mode); enum fuse_opcode op; int flags; /* * VOP_CREATE doesn't tell us the open(2) flags, so we guess. Only a * writable mode makes sense, and we might as well include readability * too. */ flags = O_RDWR; if (fuse_isdeadfs(dvp)) { return ENXIO; } bzero(&fdi, sizeof(fdi)); if ((vap->va_type != VREG)) return (EINVAL); if (!fsess_isimpl(mp, FUSE_CREATE)) { /* Fallback to FUSE_MKNOD/FUSE_OPEN */ fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, cred, mode, &op); } else { /* Use FUSE_CREATE */ op = FUSE_CREATE; fdisp_init(fdip, sizeof(*foi) + cnp->cn_namelen + 1); fdisp_make(fdip, op, vnode_mount(dvp), parentnid, td, cred); foi = fdip->indata; foi->mode = mode; foi->flags = O_CREAT | flags; memcpy((char *)fdip->indata + sizeof(*foi), cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdip->indata)[sizeof(*foi) + cnp->cn_namelen] = '\0'; } err = fdisp_wait_answ(fdip); if (err) { if (err == ENOSYS && op == FUSE_CREATE) { fsess_set_notimpl(mp, FUSE_CREATE); fdisp_make_mknod_for_fallback(fdip, cnp, dvp, parentnid, td, cred, mode, &op); err = fdisp_wait_answ(fdip); } if (err) goto out; } feo = fdip->answ; if ((err = fuse_internal_checkentry(feo, VREG))) { goto out; } if (op == FUSE_CREATE) { foo = (struct fuse_open_out*)(feo + 1); } else { /* Issue a separate FUSE_OPEN */ fdip2 = &fdi2; fdisp_init(fdip2, sizeof(*foi)); fdisp_make(fdip2, FUSE_OPEN, vnode_mount(dvp), feo->nodeid, td, cred); foi = fdip2->indata; foi->mode = mode; foi->flags = flags; err = fdisp_wait_answ(fdip2); if (err) goto out; foo = fdip2->answ; } err = fuse_vnode_get(mp, feo, feo->nodeid, dvp, vpp, cnp, VREG); if (err) { struct fuse_release_in *fri; uint64_t nodeid = feo->nodeid; uint64_t fh_id = foo->fh; fdisp_init(fdip, sizeof(*fri)); fdisp_make(fdip, FUSE_RELEASE, mp, nodeid, td, cred); fri = fdip->indata; fri->fh = fh_id; fri->flags = flags; fuse_insert_callback(fdip->tick, fuse_internal_forget_callback); fuse_insert_message(fdip->tick); goto out; } ASSERT_VOP_ELOCKED(*vpp, "fuse_vnop_create"); fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL); fuse_filehandle_init(*vpp, FUFH_RDWR, NULL, td, cred, foo); fuse_vnode_open(*vpp, foo->open_flags, td); cache_purge_negative(dvp); out: if (fdip2) fdisp_destroy(fdip2); fdisp_destroy(fdip); return err; } /* struct vnop_fdatasync_args { struct vop_generic_args a_gen; struct vnode * a_vp; struct thread * a_td; }; */ static int fuse_vnop_fdatasync(struct vop_fdatasync_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; int waitfor = MNT_WAIT; int err = 0; if (fuse_isdeadfs(vp)) { return 0; } if ((err = vop_stdfdatasync_buf(ap))) return err; return fuse_internal_fsync(vp, td, waitfor, true); } /* struct vnop_fsync_args { struct vop_generic_args a_gen; struct vnode * a_vp; int a_waitfor; struct thread * a_td; }; */ static int fuse_vnop_fsync(struct vop_fsync_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; int waitfor = ap->a_waitfor; int err = 0; if (fuse_isdeadfs(vp)) { return 0; } if ((err = vop_stdfsync(ap))) return err; return fuse_internal_fsync(vp, td, waitfor, false); } /* struct vnop_getattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct ucred *cred = ap->a_cred; struct thread *td = curthread; int err = 0; int dataflags; dataflags = fuse_get_mpdata(vnode_mount(vp))->dataflags; /* Note that we are not bailing out on a dead file system just yet. */ if (!(dataflags & FSESS_INITED)) { if (!vnode_isvroot(vp)) { fdata_set_dead(fuse_get_mpdata(vnode_mount(vp))); err = ENOTCONN; return err; } else { goto fake; } } err = fuse_internal_getattr(vp, vap, cred, td); if (err == ENOTCONN && vnode_isvroot(vp)) { /* see comment in fuse_vfsop_statfs() */ goto fake; } else { return err; } fake: bzero(vap, sizeof(*vap)); vap->va_type = vnode_vtype(vp); return 0; } /* struct vnop_inactive_args { struct vnode *a_vp; struct thread *a_td; }; */ static int fuse_vnop_inactive(struct vop_inactive_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_filehandle *fufh, *fufh_tmp; int need_flush = 1; LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { if (need_flush && vp->v_type == VREG) { if ((VTOFUD(vp)->flag & FN_SIZECHANGE) != 0) { fuse_vnode_savesize(vp, NULL, 0); } if (fuse_data_cache_invalidate || (fvdat->flag & FN_REVOKED) != 0) fuse_io_invalbuf(vp, td); else fuse_io_flushbuf(vp, MNT_WAIT, td); need_flush = 0; } fuse_filehandle_close(vp, fufh, td, NULL); } if ((fvdat->flag & FN_REVOKED) != 0 && fuse_reclaim_revoked) { vrecycle(vp); } return 0; } /* struct vnop_link_args { struct vnode *a_tdvp; struct vnode *a_vp; struct componentname *a_cnp; }; */ static int fuse_vnop_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = VTOVA(vp); struct fuse_dispatcher fdi; struct fuse_entry_out *feo; struct fuse_link_in fli; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (vnode_mount(tdvp) != vnode_mount(vp)) { return EXDEV; } /* * This is a seatbelt check to protect naive userspace filesystems from * themselves and the limitations of the FUSE IPC protocol. If a * filesystem does not allow attribute caching, assume it is capable of * validating that nlink does not overflow. */ if (vap != NULL && vap->va_nlink >= FUSE_LINK_MAX) return EMLINK; fli.oldnodeid = VTOI(vp); fdisp_init(&fdi, 0); fuse_internal_newentry_makerequest(vnode_mount(tdvp), VTOI(tdvp), cnp, FUSE_LINK, &fli, sizeof(fli), &fdi); if ((err = fdisp_wait_answ(&fdi))) { goto out; } feo = fdi.answ; err = fuse_internal_checkentry(feo, vnode_vtype(vp)); out: fdisp_destroy(&fdi); return err; } SDT_PROBE_DEFINE3(fuse, , vnops, cache_lookup, "int", "struct timespec*", "struct timespec*"); /* struct vnop_lookup_args { struct vnodeop_desc *a_desc; struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; }; */ int fuse_vnop_lookup(struct vop_lookup_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct thread *td = cnp->cn_thread; struct ucred *cred = cnp->cn_cred; int nameiop = cnp->cn_nameiop; int flags = cnp->cn_flags; int wantparent = flags & (LOCKPARENT | WANTPARENT); int islastcn = flags & ISLASTCN; struct mount *mp = vnode_mount(dvp); int err = 0; int lookup_err = 0; struct vnode *vp = NULL; struct fuse_dispatcher fdi; enum fuse_opcode op; struct fuse_entry_out *feo = NULL; struct fuse_attr_out *fao = NULL; struct fuse_attr *fattr = NULL; uint64_t nid; - struct fuse_access_param facp; if (fuse_isdeadfs(dvp)) { *vpp = NULL; return ENXIO; } if (!vnode_isdir(dvp)) { return ENOTDIR; } if (islastcn && vfs_isrdonly(mp) && (nameiop != LOOKUP)) { return EROFS; } /* * We do access check prior to doing anything else only in the case * when we are at fs root (we'd like to say, "we are at the first * component", but that's not exactly the same... nevermind). * See further comments at further access checks. */ - bzero(&facp, sizeof(facp)); + /* TODO: consider eliminating this. Is there any good reason for it? */ if (vnode_isvroot(dvp)) { /* early permission check hack */ - if ((err = fuse_internal_access(dvp, VEXEC, &facp, td, cred))) { + if ((err = fuse_internal_access(dvp, VEXEC, td, cred))) { return err; } } if (flags & ISDOTDOT) { nid = VTOFUD(dvp)->parent_nid; if (nid == 0) { return ENOENT; } fdisp_init(&fdi, 0); op = FUSE_GETATTR; goto calldaemon; } else if (cnp->cn_namelen == 1 && *(cnp->cn_nameptr) == '.') { nid = VTOI(dvp); fdisp_init(&fdi, 0); op = FUSE_GETATTR; goto calldaemon; } else { struct timespec now, timeout; err = cache_lookup(dvp, vpp, cnp, &timeout, NULL); getnanouptime(&now); SDT_PROBE3(fuse, , vnops, cache_lookup, err, &timeout, &now); switch (err) { case -1: /* positive match */ if (timespeccmp(&timeout, &now, >)) { atomic_add_acq_long(&fuse_lookup_cache_hits, 1); } else { /* Cache timeout */ atomic_add_acq_long(&fuse_lookup_cache_misses, 1); fuse_internal_vnode_disappear(*vpp); if (dvp != *vpp) vput(*vpp); else vrele(*vpp); *vpp = NULL; break; } return 0; case 0: /* no match in cache */ atomic_add_acq_long(&fuse_lookup_cache_misses, 1); break; case ENOENT: /* negative match */ getnanouptime(&now); if (timespeccmp(&timeout, &now, <=)) { /* Cache timeout */ cache_purge_negative(dvp); break; } /* fall through */ default: return err; } } nid = VTOI(dvp); fdisp_init(&fdi, cnp->cn_namelen + 1); op = FUSE_LOOKUP; calldaemon: fdisp_make(&fdi, op, mp, nid, td, cred); if (op == FUSE_LOOKUP) { memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; } lookup_err = fdisp_wait_answ(&fdi); if ((op == FUSE_LOOKUP) && !lookup_err) { /* lookup call succeeded */ nid = ((struct fuse_entry_out *)fdi.answ)->nodeid; if (nid == 0) { /* zero nodeid means ENOENT and cache it */ struct timespec timeout; fdi.answ_stat = ENOENT; lookup_err = ENOENT; if (cnp->cn_flags & MAKEENTRY) { feo = (struct fuse_entry_out *)fdi.answ; fuse_validity_2_timespec(feo, &timeout); cache_enter_time(dvp, *vpp, cnp, &timeout, NULL); } } else if (nid == FUSE_ROOT_ID) { lookup_err = EINVAL; } } if (lookup_err && (!fdi.answ_stat || lookup_err != ENOENT || op != FUSE_LOOKUP)) { fdisp_destroy(&fdi); return lookup_err; } /* lookup_err, if non-zero, must be ENOENT at this point */ if (lookup_err) { /* Entry not found */ if ((nameiop == CREATE || nameiop == RENAME) && islastcn) { - err = fuse_internal_access(dvp, VWRITE, &facp, td, - cred); + err = fuse_internal_access(dvp, VWRITE, td, cred); if (err) goto out; /* * Possibly record the position of a slot in the * directory large enough for the new component name. * This can be recorded in the vnode private data for * dvp. Set the SAVENAME flag to hold onto the * pathname for use later in VOP_CREATE or VOP_RENAME. */ cnp->cn_flags |= SAVENAME; err = EJUSTRETURN; goto out; } err = ENOENT; goto out; } else { /* Entry was found */ if (op == FUSE_GETATTR) { fattr = &((struct fuse_attr_out *)fdi.answ)->attr; } else { feo = (struct fuse_entry_out *)fdi.answ; fattr = &(feo->attr); } - /* TODO: check for ENOTDIR */ - if ((nameiop == DELETE || nameiop == RENAME) && islastcn) { - err = fuse_internal_access(dvp, VWRITE, &facp, - td, cred); + err = fuse_internal_access(dvp, VWRITE, td, cred); if (err != 0) goto out; /* * TODO: if the parent's sticky bit is set, check * whether we're allowed to remove the file. * Need to figure out the vnode locking to make this * work. */ #if defined(NOT_YET) struct vattr dvattr; fuse_internal_getattr(dvp, &dvattr, cred, td); if ((dvattr.va_mode & S_ISTXT) && - fuse_internal_access(dvp, VADMIN, &facp, - cnp->cn_thread, cnp->cn_cred) && - fuse_internal_access(*vpp, VADMIN, &facp, - cnp->cn_thread, cnp->cn_cred)) { + fuse_internal_access(dvp, VADMIN, td, cred) && + fuse_internal_access(*vpp, VADMIN, td, cred)) { err = EPERM; goto out; } #endif } /* * If deleting, and at end of pathname, return parameters * which can be used to remove file. If the wantparent flag * isn't set, we return only the directory, otherwise we go on * and lock the inode, being careful with ".". */ if (nameiop == DELETE && islastcn) { if (nid == VTOI(dvp)) { vref(dvp); *vpp = dvp; } else { err = fuse_vnode_get(dvp->v_mount, feo, nid, dvp, &vp, cnp, IFTOVT(fattr->mode)); if (err) goto out; *vpp = vp; } /* * Save the name for use in VOP_RMDIR and VOP_REMOVE * later. */ cnp->cn_flags |= SAVENAME; goto out; } /* * If rewriting (RENAME), return the inode and the * information required to rewrite the present directory * Must get inode of directory entry to verify it's a * regular file, or empty directory. */ if (nameiop == RENAME && wantparent && islastcn) { /* * Check for "." */ if (nid == VTOI(dvp)) { err = EISDIR; goto out; } err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, &vp, cnp, IFTOVT(fattr->mode)); if (err) { goto out; } - /*err = fuse_lookup_check(dvp, vp, &facp, td, cred,*/ - /*nameiop, islastcn);*/ - /*if (err)*/ - /*goto out;*/ *vpp = vp; /* * Save the name for use in VOP_RENAME later. */ cnp->cn_flags |= SAVENAME; goto out; } if (flags & ISDOTDOT) { struct mount *mp; int ltype; /* * Expanded copy of vn_vget_ino() so that * fuse_vnode_get() can be used. */ mp = dvp->v_mount; ltype = VOP_ISLOCKED(dvp); err = vfs_busy(mp, MBF_NOWAIT); if (err != 0) { vfs_ref(mp); VOP_UNLOCK(dvp, 0); err = vfs_busy(mp, 0); vn_lock(dvp, ltype | LK_RETRY); vfs_rel(mp); if (err) goto out; if ((dvp->v_iflag & VI_DOOMED) != 0) { err = ENOENT; vfs_unbusy(mp); goto out; } } VOP_UNLOCK(dvp, 0); err = fuse_vnode_get(vnode_mount(dvp), feo, nid, NULL, &vp, cnp, IFTOVT(fattr->mode)); vfs_unbusy(mp); vn_lock(dvp, ltype | LK_RETRY); if ((dvp->v_iflag & VI_DOOMED) != 0) { if (err == 0) vput(vp); err = ENOENT; } if (err) goto out; *vpp = vp; } else if (nid == VTOI(dvp)) { vref(dvp); *vpp = dvp; } else { struct fuse_vnode_data *fvdat; err = fuse_vnode_get(vnode_mount(dvp), feo, nid, dvp, &vp, cnp, IFTOVT(fattr->mode)); if (err) { goto out; } fuse_vnode_setparent(vp, dvp); /* * In the case where we are looking up a FUSE node * represented by an existing cached vnode, and the * true size reported by FUSE_LOOKUP doesn't match * the vnode's cached size, fix the vnode cache to * match the real object size. * * This can occur via FUSE distributed filesystems, * irregular files, etc. */ fvdat = VTOFUD(vp); if (vnode_isreg(vp) && fattr->size != fvdat->filesize) { /* * The FN_SIZECHANGE flag reflects a dirty * append. If userspace lets us know our cache * is invalid, that write was lost. (Dirty * writes that do not cause append are also * lost, but we don't detect them here.) * * XXX: Maybe disable WB caching on this mount. */ if (fvdat->flag & FN_SIZECHANGE) printf("%s: WB cache incoherent on " "%s!\n", __func__, vnode_mount(vp)->mnt_stat.f_mntonname); (void)fuse_vnode_setsize(vp, cred, fattr->size); fvdat->flag &= ~FN_SIZECHANGE; } *vpp = vp; } if (op == FUSE_GETATTR) { fao = (struct fuse_attr_out*)fdi.answ; fuse_internal_cache_attrs(*vpp, &fao->attr, fao->attr_valid, fao->attr_valid_nsec, NULL); } else { fuse_internal_cache_attrs(*vpp, &feo->attr, feo->attr_valid, feo->attr_valid_nsec, NULL); } } out: if (!lookup_err) { /* No lookup error; need to clean up. */ if (err) { /* Found inode; exit with no vnode. */ if (op == FUSE_LOOKUP) { fuse_internal_forget_send(vnode_mount(dvp), td, cred, nid, 1); } fdisp_destroy(&fdi); return err; } else { #ifndef NO_EARLY_PERM_CHECK_HACK if (!islastcn) { /* * We have the attributes of the next item * *now*, and it's a fact, and we do not * have to do extra work for it (ie, beg the * daemon), and it neither depends on such * accidental things like attr caching. So * the big idea: check credentials *now*, * not at the beginning of the next call to * lookup. * * The first item of the lookup chain (fs root) * won't be checked then here, of course, as * its never "the next". But go and see that * the root is taken care about at the very * beginning of this function. * * Now, given we want to do the access check * this way, one might ask: so then why not * do the access check just after fetching * the inode and its attributes from the * daemon? Why bother with producing the * corresponding vnode at all if something * is not OK? We know what's the deal as * soon as we get those attrs... There is * one bit of info though not given us by * the daemon: whether his response is * authoritative or not... His response should * be ignored if something is mounted over * the dir in question. But that can be * known only by having the vnode... */ int tmpvtype = vnode_vtype(*vpp); - bzero(&facp, sizeof(facp)); - /*the early perm check hack */ - facp.facc_flags |= FACCESS_VA_VALID; - if ((tmpvtype != VDIR) && (tmpvtype != VLNK)) { err = ENOTDIR; } if (!err && !vnode_mountedhere(*vpp)) { - err = fuse_internal_access(*vpp, VEXEC, &facp, td, cred); + err = fuse_internal_access(*vpp, VEXEC, + td, cred); } if (err) { if (tmpvtype == VLNK) SDT_PROBE2(fuse, , vnops, trace, 1, "weird, permission " "error with a symlink?"); vput(*vpp); *vpp = NULL; } } #endif } } fdisp_destroy(&fdi); return err; } /* struct vnop_mkdir_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; struct fuse_mkdir_in fmdi; if (fuse_isdeadfs(dvp)) { return ENXIO; } fmdi.mode = MAKEIMODE(vap->va_type, vap->va_mode); return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKDIR, &fmdi, sizeof(fmdi), VDIR)); } /* struct vnop_mknod_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; }; */ static int fuse_vnop_mknod(struct vop_mknod_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; struct vattr *vap = ap->a_vap; struct fuse_mknod_in fmni; if (fuse_isdeadfs(dvp)) return ENXIO; fmni.mode = MAKEIMODE(vap->va_type, vap->va_mode); fmni.rdev = vap->va_rdev; return (fuse_internal_newentry(dvp, vpp, cnp, FUSE_MKNOD, &fmni, sizeof(fmni), vap->va_type)); } /* struct vnop_open_args { struct vnode *a_vp; int a_mode; struct ucred *a_cred; struct thread *a_td; int a_fdidx; / struct file *a_fp; }; */ static int fuse_vnop_open(struct vop_open_args *ap) { struct vnode *vp = ap->a_vp; int a_mode = ap->a_mode; struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; pid_t pid = td->td_proc->p_pid; struct fuse_vnode_data *fvdat; if (fuse_isdeadfs(vp)) return ENXIO; if (vp->v_type == VCHR || vp->v_type == VBLK || vp->v_type == VFIFO) return (EOPNOTSUPP); if ((a_mode & (FREAD | FWRITE | FEXEC)) == 0) return EINVAL; fvdat = VTOFUD(vp); if (fuse_filehandle_validrw(vp, a_mode, cred, pid)) { fuse_vnode_open(vp, 0, td); return 0; } return fuse_filehandle_open(vp, a_mode, NULL, td, cred); } static int fuse_vnop_pathconf(struct vop_pathconf_args *ap) { switch (ap->a_name) { case _PC_FILESIZEBITS: *ap->a_retval = 64; return (0); case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; return (0); case _PC_LINK_MAX: *ap->a_retval = MIN(LONG_MAX, FUSE_LINK_MAX); return (0); case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; return (0); case _PC_NO_TRUNC: *ap->a_retval = 1; return (0); default: return (vop_stdpathconf(ap)); } } /* struct vnop_read_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; */ static int fuse_vnop_read(struct vop_read_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; int ioflag = ap->a_ioflag; struct ucred *cred = ap->a_cred; pid_t pid = curthread->td_proc->p_pid; if (fuse_isdeadfs(vp)) { return ENXIO; } if (VTOFUD(vp)->flag & FN_DIRECTIO) { ioflag |= IO_DIRECT; } return fuse_io_dispatch(vp, uio, ioflag, cred, pid); } /* struct vnop_readdir_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; int *a_eofflag; int *ncookies; u_long **a_cookies; }; */ static int fuse_vnop_readdir(struct vop_readdir_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct ucred *cred = ap->a_cred; struct fuse_filehandle *fufh = NULL; struct fuse_iov cookediov; int err = 0; pid_t pid = curthread->td_proc->p_pid; if (fuse_isdeadfs(vp)) { return ENXIO; } if ( /* XXXIP ((uio_iovcnt(uio) > 1)) || */ (uio_resid(uio) < sizeof(struct dirent))) { return EINVAL; } err = fuse_filehandle_get_dir(vp, &fufh, cred, pid); if (err) return (err); #define DIRCOOKEDSIZE FUSE_DIRENT_ALIGN(FUSE_NAME_OFFSET + MAXNAMLEN + 1) fiov_init(&cookediov, DIRCOOKEDSIZE); err = fuse_internal_readdir(vp, uio, fufh, &cookediov); fiov_teardown(&cookediov); return err; } /* struct vnop_readlink_args { struct vnode *a_vp; struct uio *a_uio; struct ucred *a_cred; }; */ static int fuse_vnop_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct ucred *cred = ap->a_cred; struct fuse_dispatcher fdi; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (!vnode_islnk(vp)) { return EINVAL; } fdisp_init(&fdi, 0); err = fdisp_simple_putget_vp(&fdi, FUSE_READLINK, vp, curthread, cred); if (err) { goto out; } if (((char *)fdi.answ)[0] == '/' && fuse_get_mpdata(vnode_mount(vp))->dataflags & FSESS_PUSH_SYMLINKS_IN) { char *mpth = vnode_mount(vp)->mnt_stat.f_mntonname; err = uiomove(mpth, strlen(mpth), uio); } if (!err) { err = uiomove(fdi.answ, fdi.iosize, uio); } out: fdisp_destroy(&fdi); return err; } /* struct vnop_reclaim_args { struct vnode *a_vp; struct thread *a_td; }; */ static int fuse_vnop_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct thread *td = ap->a_td; struct fuse_vnode_data *fvdat = VTOFUD(vp); struct fuse_filehandle *fufh, *fufh_tmp; if (!fvdat) { panic("FUSE: no vnode data during recycling"); } LIST_FOREACH_SAFE(fufh, &fvdat->handles, next, fufh_tmp) { printf("FUSE: vnode being reclaimed with open fufh " "(type=%#x)", fufh->fufh_type); fuse_filehandle_close(vp, fufh, td, NULL); } if ((!fuse_isdeadfs(vp)) && (fvdat->nlookup)) { fuse_internal_forget_send(vnode_mount(vp), td, NULL, VTOI(vp), fvdat->nlookup); } fuse_vnode_setparent(vp, NULL); cache_purge(vp); vfs_hash_remove(vp); vnode_destroy_vobject(vp); fuse_vnode_destroy(vp); return 0; } /* struct vnop_remove_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; }; */ static int fuse_vnop_remove(struct vop_remove_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode *vp = ap->a_vp; struct componentname *cnp = ap->a_cnp; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (vnode_isdir(vp)) { return EPERM; } cache_purge(vp); err = fuse_internal_remove(dvp, vp, cnp, FUSE_UNLINK); if (err == 0) fuse_internal_vnode_disappear(vp); return err; } /* struct vnop_rename_args { struct vnode *a_fdvp; struct vnode *a_fvp; struct componentname *a_fcnp; struct vnode *a_tdvp; struct vnode *a_tvp; struct componentname *a_tcnp; }; */ static int fuse_vnop_rename(struct vop_rename_args *ap) { struct vnode *fdvp = ap->a_fdvp; struct vnode *fvp = ap->a_fvp; struct componentname *fcnp = ap->a_fcnp; struct vnode *tdvp = ap->a_tdvp; struct vnode *tvp = ap->a_tvp; struct componentname *tcnp = ap->a_tcnp; struct fuse_data *data; int err = 0; if (fuse_isdeadfs(fdvp)) { return ENXIO; } if (fvp->v_mount != tdvp->v_mount || (tvp && fvp->v_mount != tvp->v_mount)) { SDT_PROBE2(fuse, , vnops, trace, 1, "cross-device rename"); err = EXDEV; goto out; } cache_purge(fvp); /* * FUSE library is expected to check if target directory is not * under the source directory in the file system tree. * Linux performs this check at VFS level. */ data = fuse_get_mpdata(vnode_mount(tdvp)); sx_xlock(&data->rename_lock); err = fuse_internal_rename(fdvp, fcnp, tdvp, tcnp); if (err == 0) { if (tdvp != fdvp) fuse_vnode_setparent(fvp, tdvp); if (tvp != NULL) fuse_vnode_setparent(tvp, NULL); } sx_unlock(&data->rename_lock); if (tvp != NULL && tvp != fvp) { cache_purge(tvp); } if (vnode_isdir(fvp)) { if ((tvp != NULL) && vnode_isdir(tvp)) { cache_purge(tdvp); } cache_purge(fdvp); } out: if (tdvp == tvp) { vrele(tdvp); } else { vput(tdvp); } if (tvp != NULL) { vput(tvp); } vrele(fdvp); vrele(fvp); return err; } /* struct vnop_rmdir_args { struct vnode *a_dvp; struct vnode *a_vp; struct componentname *a_cnp; } *ap; */ static int fuse_vnop_rmdir(struct vop_rmdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode *vp = ap->a_vp; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } if (VTOFUD(vp) == VTOFUD(dvp)) { return EINVAL; } err = fuse_internal_remove(dvp, vp, ap->a_cnp, FUSE_RMDIR); if (err == 0) fuse_internal_vnode_disappear(vp); return err; } /* struct vnop_setattr_args { struct vnode *a_vp; struct vattr *a_vap; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_setattr(struct vop_setattr_args *ap) { struct vnode *vp = ap->a_vp; struct vattr *vap = ap->a_vap; struct ucred *cred = ap->a_cred; struct thread *td = curthread; struct fuse_dispatcher fdi; struct fuse_setattr_in *fsai; - struct fuse_access_param facp; pid_t pid = td->td_proc->p_pid; int err = 0; enum vtype vtyp; int sizechanged = 0; uint64_t newsize = 0; accmode_t accmode = 0; if (fuse_isdeadfs(vp)) { return ENXIO; } fdisp_init(&fdi, sizeof(*fsai)); fdisp_make_vp(&fdi, FUSE_SETATTR, vp, td, cred); fsai = fdi.indata; fsai->valid = 0; - bzero(&facp, sizeof(facp)); - - facp.xuid = vap->va_uid; - facp.xgid = vap->va_gid; - if (vap->va_uid != (uid_t)VNOVAL) { - facp.facc_flags |= FACCESS_CHOWN; fsai->uid = vap->va_uid; fsai->valid |= FATTR_UID; accmode |= VADMIN; } if (vap->va_gid != (gid_t)VNOVAL) { - facp.facc_flags |= FACCESS_CHOWN; fsai->gid = vap->va_gid; fsai->valid |= FATTR_GID; accmode |= VADMIN; } if (vap->va_size != VNOVAL) { struct fuse_filehandle *fufh = NULL; /*Truncate to a new value. */ fsai->size = vap->va_size; sizechanged = 1; newsize = vap->va_size; fsai->valid |= FATTR_SIZE; accmode |= VWRITE; fuse_filehandle_getrw(vp, FWRITE, &fufh, cred, pid); if (fufh) { fsai->fh = fufh->fh_id; fsai->valid |= FATTR_FH; } } if (vap->va_atime.tv_sec != VNOVAL) { fsai->atime = vap->va_atime.tv_sec; fsai->atimensec = vap->va_atime.tv_nsec; fsai->valid |= FATTR_ATIME; accmode |= VADMIN; /* * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181 */ } if (vap->va_mtime.tv_sec != VNOVAL) { fsai->mtime = vap->va_mtime.tv_sec; fsai->mtimensec = vap->va_mtime.tv_nsec; fsai->valid |= FATTR_MTIME; accmode |= VADMIN; /* * TODO: only require VWRITE if UTIMENS_NULL is set. PR 237181 */ } if (vap->va_mode != (mode_t)VNOVAL) { fsai->mode = vap->va_mode & ALLPERMS; fsai->valid |= FATTR_MODE; accmode |= VADMIN; } if (!fsai->valid) { goto out; } vtyp = vnode_vtype(vp); if (fsai->valid & FATTR_SIZE && vtyp == VDIR) { err = EISDIR; goto out; } if (vfs_isrdonly(vnode_mount(vp)) && (fsai->valid & ~FATTR_SIZE || vtyp == VREG)) { err = EROFS; goto out; } - err = fuse_internal_access(vp, accmode, &facp, td, cred); + err = fuse_internal_access(vp, accmode, td, cred); if (err) goto out; if ((err = fdisp_wait_answ(&fdi))) goto out; vtyp = IFTOVT(((struct fuse_attr_out *)fdi.answ)->attr.mode); if (vnode_vtype(vp) != vtyp) { if (vnode_vtype(vp) == VNON && vtyp != VNON) { SDT_PROBE2(fuse, , vnops, trace, 1, "FUSE: Dang! " "vnode_vtype is VNON and vtype isn't."); } else { /* * STALE vnode, ditch * * The vnode has changed its type "behind our back". * There's nothing really we can do, so let us just * force an internal revocation and tell the caller to * try again, if interested. */ fuse_internal_vnode_disappear(vp); err = EAGAIN; } } if (err == 0) { struct fuse_attr_out *fao = (struct fuse_attr_out*)fdi.answ; fuse_internal_cache_attrs(vp, &fao->attr, fao->attr_valid, fao->attr_valid_nsec, NULL); } out: fdisp_destroy(&fdi); if (!err && sizechanged) { fuse_vnode_setsize(vp, cred, newsize); VTOFUD(vp)->flag &= ~FN_SIZECHANGE; } return err; } /* struct vnop_strategy_args { struct vnode *a_vp; struct buf *a_bp; }; */ static int fuse_vnop_strategy(struct vop_strategy_args *ap) { struct vnode *vp = ap->a_vp; struct buf *bp = ap->a_bp; if (!vp || fuse_isdeadfs(vp)) { bp->b_ioflags |= BIO_ERROR; bp->b_error = ENXIO; bufdone(bp); return 0; } if (bp->b_iocmd == BIO_WRITE) { int err; err = fuse_vnode_refreshsize(vp, NOCRED); if (err) { bp->b_ioflags |= BIO_ERROR; bp->b_error = err; bufdone(bp); return 0; } } /* * VOP_STRATEGY always returns zero and signals error via bp->b_ioflags. * fuse_io_strategy sets bp's error fields */ (void)fuse_io_strategy(vp, bp); return 0; } /* struct vnop_symlink_args { struct vnode *a_dvp; struct vnode **a_vpp; struct componentname *a_cnp; struct vattr *a_vap; char *a_target; }; */ static int fuse_vnop_symlink(struct vop_symlink_args *ap) { struct vnode *dvp = ap->a_dvp; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; const char *target = ap->a_target; struct fuse_dispatcher fdi; int err; size_t len; if (fuse_isdeadfs(dvp)) { return ENXIO; } /* * Unlike the other creator type calls, here we have to create a message * where the name of the new entry comes first, and the data describing * the entry comes second. * Hence we can't rely on our handy fuse_internal_newentry() routine, * but put together the message manually and just call the core part. */ len = strlen(target) + 1; fdisp_init(&fdi, len + cnp->cn_namelen + 1); fdisp_make_vp(&fdi, FUSE_SYMLINK, dvp, curthread, NULL); memcpy(fdi.indata, cnp->cn_nameptr, cnp->cn_namelen); ((char *)fdi.indata)[cnp->cn_namelen] = '\0'; memcpy((char *)fdi.indata + cnp->cn_namelen + 1, target, len); err = fuse_internal_newentry_core(dvp, vpp, cnp, VLNK, &fdi); fdisp_destroy(&fdi); return err; } /* struct vnop_write_args { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; }; */ static int fuse_vnop_write(struct vop_write_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; int ioflag = ap->a_ioflag; struct ucred *cred = ap->a_cred; pid_t pid = curthread->td_proc->p_pid; int err; if (fuse_isdeadfs(vp)) { return ENXIO; } err = fuse_vnode_refreshsize(vp, cred); if (err) return err; if (VTOFUD(vp)->flag & FN_DIRECTIO) { ioflag |= IO_DIRECT; } return fuse_io_dispatch(vp, uio, ioflag, cred, pid); } SDT_PROBE_DEFINE1(fuse, , vnops, vnop_getpages_error, "int"); /* struct vnop_getpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_reqpage; }; */ static int fuse_vnop_getpages(struct vop_getpages_args *ap) { int i, error, nextoff, size, toff, count, npages; struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; struct vnode *vp; struct thread *td; struct ucred *cred; vm_page_t *pages; pid_t pid = curthread->td_proc->p_pid; vp = ap->a_vp; KASSERT(vp->v_object, ("objectless vp passed to getpages")); td = curthread; /* XXX */ cred = curthread->td_ucred; /* XXX */ pages = ap->a_m; npages = ap->a_count; if (!fsess_opt_mmap(vnode_mount(vp))) { SDT_PROBE2(fuse, , vnops, trace, 1, "called on non-cacheable vnode??\n"); return (VM_PAGER_ERROR); } /* * If the last page is partially valid, just return it and allow * the pager to zero-out the blanks. Partially valid pages can * only occur at the file EOF. * * XXXGL: is that true for FUSE, which is a local filesystem, * but still somewhat disconnected from the kernel? */ VM_OBJECT_WLOCK(vp->v_object); if (pages[npages - 1]->valid != 0 && --npages == 0) goto out; VM_OBJECT_WUNLOCK(vp->v_object); /* * We use only the kva address for the buffer, but this is extremely * convenient and fast. */ bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK); kva = (vm_offset_t)bp->b_data; pmap_qenter(kva, pages, npages); VM_CNT_INC(v_vnodein); VM_CNT_ADD(v_vnodepgsin, npages); count = npages << PAGE_SHIFT; iov.iov_base = (caddr_t)kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = IDX_TO_OFF(pages[0]->pindex); uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = td; error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid); pmap_qremove(kva, npages); uma_zfree(fuse_pbuf_zone, bp); if (error && (uio.uio_resid == count)) { SDT_PROBE1(fuse, , vnops, vnop_getpages_error, error); return VM_PAGER_ERROR; } /* * Calculate the number of bytes read and validate only that number * of bytes. Note that due to pending writes, size may be 0. This * does not mean that the remaining data is invalid! */ size = count - uio.uio_resid; VM_OBJECT_WLOCK(vp->v_object); fuse_vm_page_lock_queues(); for (i = 0, toff = 0; i < npages; i++, toff = nextoff) { vm_page_t m; nextoff = toff + PAGE_SIZE; m = pages[i]; if (nextoff <= size) { /* * Read operation filled an entire page */ m->valid = VM_PAGE_BITS_ALL; KASSERT(m->dirty == 0, ("fuse_getpages: page %p is dirty", m)); } else if (size > toff) { /* * Read operation filled a partial page. */ m->valid = 0; vm_page_set_valid_range(m, 0, size - toff); KASSERT(m->dirty == 0, ("fuse_getpages: page %p is dirty", m)); } else { /* * Read operation was short. If no error occurred * we may have hit a zero-fill section. We simply * leave valid set to 0. */ ; } } fuse_vm_page_unlock_queues(); out: VM_OBJECT_WUNLOCK(vp->v_object); if (ap->a_rbehind) *ap->a_rbehind = 0; if (ap->a_rahead) *ap->a_rahead = 0; return (VM_PAGER_OK); } /* struct vnop_putpages_args { struct vnode *a_vp; vm_page_t *a_m; int a_count; int a_sync; int *a_rtvals; vm_ooffset_t a_offset; }; */ static int fuse_vnop_putpages(struct vop_putpages_args *ap) { struct uio uio; struct iovec iov; vm_offset_t kva; struct buf *bp; int i, error, npages, count; off_t offset; int *rtvals; struct vnode *vp; struct thread *td; struct ucred *cred; vm_page_t *pages; vm_ooffset_t fsize; pid_t pid = curthread->td_proc->p_pid; vp = ap->a_vp; KASSERT(vp->v_object, ("objectless vp passed to putpages")); fsize = vp->v_object->un_pager.vnp.vnp_size; td = curthread; /* XXX */ cred = curthread->td_ucred; /* XXX */ pages = ap->a_m; count = ap->a_count; rtvals = ap->a_rtvals; npages = btoc(count); offset = IDX_TO_OFF(pages[0]->pindex); if (!fsess_opt_mmap(vnode_mount(vp))) { SDT_PROBE2(fuse, , vnops, trace, 1, "called on non-cacheable vnode??\n"); } for (i = 0; i < npages; i++) rtvals[i] = VM_PAGER_AGAIN; /* * When putting pages, do not extend file past EOF. */ if (offset + count > fsize) { count = fsize - offset; if (count < 0) count = 0; } /* * We use only the kva address for the buffer, but this is extremely * convenient and fast. */ bp = uma_zalloc(fuse_pbuf_zone, M_WAITOK); kva = (vm_offset_t)bp->b_data; pmap_qenter(kva, pages, npages); VM_CNT_INC(v_vnodeout); VM_CNT_ADD(v_vnodepgsout, count); iov.iov_base = (caddr_t)kva; iov.iov_len = count; uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = offset; uio.uio_resid = count; uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_WRITE; uio.uio_td = td; error = fuse_io_dispatch(vp, &uio, IO_DIRECT, cred, pid); pmap_qremove(kva, npages); uma_zfree(fuse_pbuf_zone, bp); if (!error) { int nwritten = round_page(count - uio.uio_resid) / PAGE_SIZE; for (i = 0; i < nwritten; i++) { rtvals[i] = VM_PAGER_OK; VM_OBJECT_WLOCK(pages[i]->object); vm_page_undirty(pages[i]); VM_OBJECT_WUNLOCK(pages[i]->object); } } return rtvals[0]; } static const char extattr_namespace_separator = '.'; /* struct vop_getextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct uio *a_uio; size_t *a_size; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_getextattr(struct vop_getextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_getxattr_in *get_xattr_in; struct fuse_getxattr_out *get_xattr_out; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; char *attr_str; size_t len; int err; if (fuse_isdeadfs(vp)) return (ENXIO); - err = extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); + err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len + sizeof(*get_xattr_in)); fdisp_make_vp(&fdi, FUSE_GETXATTR, vp, td, cred); get_xattr_in = fdi.indata; /* * Check to see whether we're querying the available size or * issuing the actual request. If we pass in 0, we get back struct * fuse_getxattr_out. If we pass in a non-zero size, we get back * that much data, without the struct fuse_getxattr_out header. */ if (uio == NULL) get_xattr_in->size = 0; else get_xattr_in->size = uio->uio_resid; attr_str = (char *)fdi.indata + sizeof(*get_xattr_in); snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) fsess_set_notimpl(mp, FUSE_GETXATTR); goto out; } get_xattr_out = fdi.answ; if (ap->a_size != NULL) *ap->a_size = get_xattr_out->size; if (uio != NULL) err = uiomove(fdi.answ, fdi.iosize, uio); out: fdisp_destroy(&fdi); return (err); } /* struct vop_setextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct uio *a_uio; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_setextattr(struct vop_setextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_setxattr_in *set_xattr_in; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; size_t len; char *attr_str; int err; if (fuse_isdeadfs(vp)) return (ENXIO); + if (vfs_isrdonly(mp)) + return EROFS; + /* Deleting xattrs must use VOP_DELETEEXTATTR instead */ if (ap->a_uio == NULL) return (EINVAL); - err = extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VWRITE); + err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, + VWRITE); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len + sizeof(*set_xattr_in) + uio->uio_resid); fdisp_make_vp(&fdi, FUSE_SETXATTR, vp, td, cred); set_xattr_in = fdi.indata; set_xattr_in->size = uio->uio_resid; attr_str = (char *)fdi.indata + sizeof(*set_xattr_in); snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = uiomove((char *)fdi.indata + sizeof(*set_xattr_in) + len, uio->uio_resid, uio); if (err != 0) { goto out; } err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) fsess_set_notimpl(mp, FUSE_SETXATTR); goto out; } out: fdisp_destroy(&fdi); return (err); } /* * The Linux / FUSE extended attribute list is simply a collection of * NUL-terminated strings. The FreeBSD extended attribute list is a single * byte length followed by a non-NUL terminated string. So, this allows * conversion of the Linux / FUSE format to the FreeBSD format in place. * Linux attribute names are reported with the namespace as a prefix (e.g. * "user.attribute_name"), but in FreeBSD they are reported without the * namespace prefix (e.g. "attribute_name"). So, we're going from: * * user.attr_name1\0user.attr_name2\0 * * to: * * attr_name1attr_name2 * * Where "" is a single byte number of characters in the attribute name. * * Args: * prefix - exattr namespace prefix string * list, list_len - input list with namespace prefixes * bsd_list, bsd_list_len - output list compatible with bsd vfs */ static int fuse_xattrlist_convert(char *prefix, const char *list, int list_len, char *bsd_list, int *bsd_list_len) { int len, pos, dist_to_next, prefix_len; pos = 0; *bsd_list_len = 0; prefix_len = strlen(prefix); while (pos < list_len && list[pos] != '\0') { dist_to_next = strlen(&list[pos]) + 1; if (bcmp(&list[pos], prefix, prefix_len) == 0 && list[pos + prefix_len] == extattr_namespace_separator) { len = dist_to_next - (prefix_len + sizeof(extattr_namespace_separator)) - 1; if (len >= EXTATTR_MAXNAMELEN) return (ENAMETOOLONG); bsd_list[*bsd_list_len] = len; memcpy(&bsd_list[*bsd_list_len + 1], &list[pos + prefix_len + sizeof(extattr_namespace_separator)], len); *bsd_list_len += len + 1; } pos += dist_to_next; } return (0); } /* struct vop_listextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; struct uio *a_uio; size_t *a_size; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_listextattr(struct vop_listextattr_args *ap) { struct vnode *vp = ap->a_vp; struct uio *uio = ap->a_uio; struct fuse_dispatcher fdi; struct fuse_listxattr_in *list_xattr_in; struct fuse_listxattr_out *list_xattr_out; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; size_t len; char *prefix; char *attr_str; char *bsd_list = NULL; char *linux_list; int bsd_list_len; int linux_list_len; int err; if (fuse_isdeadfs(vp)) return (ENXIO); - err = extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); + err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VREAD); if (err) return err; /* * Add space for a NUL and the period separator if enabled. * Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + 1; fdisp_init(&fdi, sizeof(*list_xattr_in) + len); fdisp_make_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); /* * Retrieve Linux / FUSE compatible list size. */ list_xattr_in = fdi.indata; list_xattr_in->size = 0; attr_str = (char *)fdi.indata + sizeof(*list_xattr_in); snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator); err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) fsess_set_notimpl(mp, FUSE_LISTXATTR); goto out; } list_xattr_out = fdi.answ; linux_list_len = list_xattr_out->size; if (linux_list_len == 0) { if (ap->a_size != NULL) *ap->a_size = linux_list_len; goto out; } /* * Retrieve Linux / FUSE compatible list values. */ fdisp_refresh_vp(&fdi, FUSE_LISTXATTR, vp, td, cred); list_xattr_in = fdi.indata; list_xattr_in->size = linux_list_len + sizeof(*list_xattr_out); list_xattr_in->flags = 0; attr_str = (char *)fdi.indata + sizeof(*list_xattr_in); snprintf(attr_str, len, "%s%c", prefix, extattr_namespace_separator); err = fdisp_wait_answ(&fdi); if (err != 0) goto out; linux_list = fdi.answ; linux_list_len = fdi.iosize; /* * Retrieve the BSD compatible list values. * The Linux / FUSE attribute list format isn't the same * as FreeBSD's format. So we need to transform it into * FreeBSD's format before giving it to the user. */ bsd_list = malloc(linux_list_len, M_TEMP, M_WAITOK); err = fuse_xattrlist_convert(prefix, linux_list, linux_list_len, bsd_list, &bsd_list_len); if (err != 0) goto out; if (ap->a_size != NULL) *ap->a_size = bsd_list_len; if (uio != NULL) err = uiomove(bsd_list, bsd_list_len, uio); out: free(bsd_list, M_TEMP); fdisp_destroy(&fdi); return (err); } /* struct vop_deleteextattr_args { struct vop_generic_args a_gen; struct vnode *a_vp; int a_attrnamespace; const char *a_name; struct ucred *a_cred; struct thread *a_td; }; */ static int fuse_vnop_deleteextattr(struct vop_deleteextattr_args *ap) { struct vnode *vp = ap->a_vp; struct fuse_dispatcher fdi; struct mount *mp = vnode_mount(vp); struct thread *td = ap->a_td; struct ucred *cred = ap->a_cred; char *prefix; size_t len; char *attr_str; int err; if (fuse_isdeadfs(vp)) return (ENXIO); - err = extattr_check_cred(vp, ap->a_attrnamespace, cred, td, VWRITE); + if (vfs_isrdonly(mp)) + return EROFS; + + err = fuse_extattr_check_cred(vp, ap->a_attrnamespace, cred, td, + VWRITE); if (err) return err; /* Default to looking for user attributes. */ if (ap->a_attrnamespace == EXTATTR_NAMESPACE_SYSTEM) prefix = EXTATTR_NAMESPACE_SYSTEM_STRING; else prefix = EXTATTR_NAMESPACE_USER_STRING; len = strlen(prefix) + sizeof(extattr_namespace_separator) + strlen(ap->a_name) + 1; fdisp_init(&fdi, len); fdisp_make_vp(&fdi, FUSE_REMOVEXATTR, vp, td, cred); attr_str = fdi.indata; snprintf(attr_str, len, "%s%c%s", prefix, extattr_namespace_separator, ap->a_name); err = fdisp_wait_answ(&fdi); if (err != 0) { if (err == ENOSYS) fsess_set_notimpl(mp, FUSE_REMOVEXATTR); } fdisp_destroy(&fdi); return (err); } /* struct vnop_print_args { struct vnode *a_vp; }; */ static int fuse_vnop_print(struct vop_print_args *ap) { struct fuse_vnode_data *fvdat = VTOFUD(ap->a_vp); printf("nodeid: %ju, parent nodeid: %ju, nlookup: %ju, flag: %#x\n", (uintmax_t)VTOILLU(ap->a_vp), (uintmax_t)fvdat->parent_nid, (uintmax_t)fvdat->nlookup, fvdat->flag); return 0; } Index: projects/fuse2/tests/sys/fs/fusefs/allow_other.cc =================================================================== --- projects/fuse2/tests/sys/fs/fusefs/allow_other.cc (revision 346100) +++ projects/fuse2/tests/sys/fs/fusefs/allow_other.cc (revision 346101) @@ -1,224 +1,271 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 The FreeBSD Foundation * * This software was developed by BFF Storage Systems, LLC under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Tests for the "allow_other" mount option. They must be in their own * file so they can be run as root */ extern "C" { #include +#include #include #include } #include "mockfs.hh" #include "utils.hh" using namespace testing; const static char FULLPATH[] = "mountpoint/some_file.txt"; const static char RELPATH[] = "some_file.txt"; class NoAllowOther: public FuseTest { public: /* Unprivileged user id */ int m_uid; virtual void SetUp() { if (geteuid() != 0) { GTEST_SKIP() << "This test must be run as root"; } FuseTest::SetUp(); } }; class AllowOther: public NoAllowOther { public: virtual void SetUp() { m_allow_other = true; NoAllowOther::SetUp(); } }; TEST_F(AllowOther, allowed) { fork(true, [&] { uint64_t ino = 42; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_open(ino, 0, 1); expect_flush(ino, 1, ReturnErrno(0)); expect_release(ino, FH); }, []() { int fd; fd = open(FULLPATH, O_RDONLY); if (fd < 0) { perror("open"); return(1); } return 0; } ); } /* * A variation of the Open.multiple_creds test showing how the bug can lead to a * privilege elevation. The first process is privileged and opens a file only * visible to root. The second process is unprivileged and shouldn't be able * to open the file, but does thanks to the bug */ TEST_F(AllowOther, privilege_escalation) { const static char FULLPATH[] = "mountpoint/some_file.txt"; const static char RELPATH[] = "some_file.txt"; int fd1; const static uint64_t ino = 42; const static uint64_t fh = 100; /* Fork a child to open the file with different credentials */ fork(true, [&] { expect_lookup(RELPATH, ino, S_IFREG | 0600, 0, 2); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_OPEN && in->header.pid == (uint32_t)getpid() && in->header.uid == (uint32_t)geteuid() && in->header.nodeid == ino); }, Eq(true)), _) ).WillOnce(Invoke( ReturnImmediate([](auto in __unused, auto out) { out->body.open.fh = fh; out->header.len = sizeof(out->header); SET_OUT_HEADER_LEN(out, open); }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_OPEN && in->header.pid != (uint32_t)getpid() && in->header.uid != (uint32_t)geteuid() && in->header.nodeid == ino); }, Eq(true)), _) ).Times(AnyNumber()) .WillRepeatedly(Invoke(ReturnErrno(EPERM))); fd1 = open(FULLPATH, O_RDONLY); EXPECT_LE(0, fd1) << strerror(errno); }, [] { int fd0; fd0 = open(FULLPATH, O_RDONLY); if (fd0 >= 0) { fprintf(stderr, "Privilege escalation!\n"); return 1; } if (errno != EPERM) { fprintf(stderr, "Unexpected error %s\n", strerror(errno)); return 1; } return 0; } ); /* Deliberately leak fd1. close(2) will be tested in release.cc */ } TEST_F(NoAllowOther, disallowed) { fork(true, [] { }, []() { int fd; fd = open(FULLPATH, O_RDONLY); if (fd >= 0) { fprintf(stderr, "open should've failed\n"); return(1); } else if (errno != EPERM) { fprintf(stderr, "Unexpected error: %s\n", strerror(errno)); return(1); } return 0; } ); } /* * When -o allow_other is not used, users other than the owner aren't allowed * to open anything inside of the mount point, not just the mountpoint itself * This is a regression test for bug 237052 */ TEST_F(NoAllowOther, disallowed_beneath_root) { const static char FULLPATH[] = "mountpoint/some_dir"; const static char RELPATH[] = "some_dir"; const static char RELPATH2[] = "other_dir"; const static uint64_t ino = 42; const static uint64_t ino2 = 43; int dfd; expect_lookup(RELPATH, ino, S_IFDIR | 0755, 0, 1); EXPECT_LOOKUP(ino, RELPATH2) .WillRepeatedly(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino2; out->body.entry.attr.nlink = 1; out->body.entry.attr_valid = UINT64_MAX; }))); expect_opendir(ino); dfd = open(FULLPATH, O_DIRECTORY); ASSERT_LE(0, dfd) << strerror(errno); fork(true, [] { }, [&]() { int fd; fd = openat(dfd, RELPATH2, O_RDONLY); if (fd >= 0) { fprintf(stderr, "openat should've failed\n"); + return(1); + } else if (errno != EPERM) { + fprintf(stderr, "Unexpected error: %s\n", + strerror(errno)); + return(1); + } + return 0; + } + ); +} + +/* + * Provide coverage for the extattr methods, which have a slightly different + * code path + */ +TEST_F(NoAllowOther, setextattr) +{ + int ino = 42; + + fork(true, [&] { + EXPECT_LOOKUP(1, RELPATH) + .WillOnce(Invoke( + ReturnImmediate([=](auto in __unused, auto out) { + SET_OUT_HEADER_LEN(out, entry); + out->body.entry.attr_valid = UINT64_MAX; + out->body.entry.entry_valid = UINT64_MAX; + out->body.entry.attr.mode = S_IFREG | 0644; + out->body.entry.nodeid = ino; + }))); + + /* + * lookup the file to get it into the cache. + * Otherwise, the unprivileged lookup will fail with + * EACCES + */ + ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); + }, [&]() { + const char value[] = "whatever"; + ssize_t value_len = strlen(value) + 1; + int ns = EXTATTR_NAMESPACE_USER; + ssize_t r; + + r = extattr_set_file(FULLPATH, ns, "foo", + (void*)value, value_len); + if (r >= 0) { + fprintf(stderr, "should've failed\n"); return(1); } else if (errno != EPERM) { fprintf(stderr, "Unexpected error: %s\n", strerror(errno)); return(1); } return 0; } ); } Index: projects/fuse2/tests/sys/fs/fusefs/lookup.cc =================================================================== --- projects/fuse2/tests/sys/fs/fusefs/lookup.cc (revision 346100) +++ projects/fuse2/tests/sys/fs/fusefs/lookup.cc (revision 346101) @@ -1,295 +1,308 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 The FreeBSD Foundation * * This software was developed by BFF Storage Systems, LLC under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ extern "C" { #include } #include "mockfs.hh" #include "utils.hh" using namespace testing; class Lookup: public FuseTest {}; /* * If lookup returns a non-zero cache timeout, then subsequent VOP_GETATTRs * should use the cached attributes, rather than query the daemon */ TEST_F(Lookup, attr_cache) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const uint64_t generation = 13; struct stat sb; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.nodeid = ino; out->body.entry.attr_valid = UINT64_MAX; out->body.entry.attr.ino = ino; // Must match nodeid out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.attr.size = 1; out->body.entry.attr.blocks = 2; out->body.entry.attr.atime = 3; out->body.entry.attr.mtime = 4; out->body.entry.attr.ctime = 5; out->body.entry.attr.atimensec = 6; out->body.entry.attr.mtimensec = 7; out->body.entry.attr.ctimensec = 8; out->body.entry.attr.nlink = 9; out->body.entry.attr.uid = 10; out->body.entry.attr.gid = 11; out->body.entry.attr.rdev = 12; out->body.entry.generation = generation; }))); /* stat(2) issues a VOP_LOOKUP followed by a VOP_GETATTR */ ASSERT_EQ(0, stat(FULLPATH, &sb)) << strerror(errno); EXPECT_EQ(1, sb.st_size); EXPECT_EQ(2, sb.st_blocks); EXPECT_EQ(3, sb.st_atim.tv_sec); EXPECT_EQ(6, sb.st_atim.tv_nsec); EXPECT_EQ(4, sb.st_mtim.tv_sec); EXPECT_EQ(7, sb.st_mtim.tv_nsec); EXPECT_EQ(5, sb.st_ctim.tv_sec); EXPECT_EQ(8, sb.st_ctim.tv_nsec); EXPECT_EQ(9ull, sb.st_nlink); EXPECT_EQ(10ul, sb.st_uid); EXPECT_EQ(11ul, sb.st_gid); EXPECT_EQ(12ul, sb.st_rdev); EXPECT_EQ(ino, sb.st_ino); EXPECT_EQ(S_IFREG | 0644, sb.st_mode); // fuse(4) does not _yet_ support inode generations //EXPECT_EQ(generation, sb.st_gen); //st_birthtim and st_flags are not supported by protocol 7.8. They're //only supported as OS-specific extensions to OSX. //EXPECT_EQ(, sb.st_birthtim); //EXPECT_EQ(, sb.st_flags); //FUSE can't set st_blksize until protocol 7.9 } /* * If lookup returns a finite but non-zero cache timeout, then we should discard * the cached attributes and requery the daemon. */ TEST_F(Lookup, attr_cache_timeout) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; struct stat sb; /* * The timeout should be longer than the longest plausible time the * daemon would take to complete a write(2) to /dev/fuse, but no longer. */ long timeout_ns = 250'000'000; EXPECT_LOOKUP(1, RELPATH) .Times(2) .WillRepeatedly(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.nodeid = ino; out->body.entry.attr_valid_nsec = timeout_ns; out->body.entry.attr.ino = ino; // Must match nodeid out->body.entry.attr.mode = S_IFREG | 0644; }))); /* access(2) will issue a VOP_LOOKUP and fill the attr cache */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); /* Next access(2) will use the cached attributes */ usleep(2 * timeout_ns / 1000); /* The cache has timed out; VOP_GETATTR should query the daemon*/ ASSERT_EQ(0, stat(FULLPATH, &sb)) << strerror(errno); } TEST_F(Lookup, enoent) { const char FULLPATH[] = "mountpoint/does_not_exist"; const char RELPATH[] = "does_not_exist"; EXPECT_LOOKUP(1, RELPATH).WillOnce(Invoke(ReturnErrno(ENOENT))); EXPECT_NE(0, access(FULLPATH, F_OK)); EXPECT_EQ(ENOENT, errno); } -//TODO: test ENOTDIR +TEST_F(Lookup, enotdir) +{ + const char FULLPATH[] = "mountpoint/not_a_dir/some_file.txt"; + const char RELPATH[] = "not_a_dir"; + EXPECT_LOOKUP(1, RELPATH) + .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { + SET_OUT_HEADER_LEN(out, entry); + out->body.entry.entry_valid = UINT64_MAX; + out->body.entry.attr.mode = S_IFREG | 0644; + out->body.entry.nodeid = 42; + }))); + + ASSERT_EQ(-1, access(FULLPATH, F_OK)); + ASSERT_EQ(ENOTDIR, errno); +} + /* * If lookup returns a non-zero entry timeout, then subsequent VOP_LOOKUPs * should use the cached inode rather than requery the daemon */ TEST_F(Lookup, entry_cache) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.entry_valid = UINT64_MAX; out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = 14; }))); ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); /* The second access(2) should use the cache */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); } /* * If the daemon returns an error of 0 and an inode of 0, that's a flag for * "ENOENT and cache it" with the given entry_timeout */ TEST_F(Lookup, entry_cache_negative) { struct timespec entry_valid = {.tv_sec = TIME_T_MAX, .tv_nsec = 0}; EXPECT_LOOKUP(1, "does_not_exist").Times(1) .WillOnce(Invoke(ReturnNegativeCache(&entry_valid))); EXPECT_NE(0, access("mountpoint/does_not_exist", F_OK)); EXPECT_EQ(ENOENT, errno); EXPECT_NE(0, access("mountpoint/does_not_exist", F_OK)); EXPECT_EQ(ENOENT, errno); } /* Negative entry caches should timeout, too */ TEST_F(Lookup, entry_cache_negative_timeout) { const char *RELPATH = "does_not_exist"; const char *FULLPATH = "mountpoint/does_not_exist"; /* * The timeout should be longer than the longest plausible time the * daemon would take to complete a write(2) to /dev/fuse, but no longer. */ struct timespec entry_valid = {.tv_sec = 0, .tv_nsec = 250'000'000}; EXPECT_LOOKUP(1, RELPATH).Times(2) .WillRepeatedly(Invoke(ReturnNegativeCache(&entry_valid))); EXPECT_NE(0, access(FULLPATH, F_OK)); EXPECT_EQ(ENOENT, errno); usleep(2 * entry_valid.tv_nsec / 1000); /* The cache has timed out; VOP_LOOKUP should requery the daemon*/ EXPECT_NE(0, access(FULLPATH, F_OK)); EXPECT_EQ(ENOENT, errno); } /* * If lookup returns a finite but non-zero entry cache timeout, then we should * discard the cached inode and requery the daemon */ TEST_F(Lookup, entry_cache_timeout) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; /* * The timeout should be longer than the longest plausible time the * daemon would take to complete a write(2) to /dev/fuse, but no longer. */ long timeout_ns = 250'000'000; EXPECT_LOOKUP(1, RELPATH) .Times(2) .WillRepeatedly(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.entry_valid_nsec = timeout_ns; out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = 14; }))); /* access(2) will issue a VOP_LOOKUP and fill the entry cache */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); /* Next access(2) will use the cached entry */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); usleep(2 * timeout_ns / 1000); /* The cache has timed out; VOP_LOOKUP should requery the daemon*/ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); } // TODO: export_support // After upgrading the protocol to 7.10, check that the kernel will only // attempt to lookup "." and ".." if the filesystem sets FUSE_EXPORT_SUPPORT in // the init flags. If not, then all lookups for those entries will return // ESTALE. TEST_F(Lookup, ok) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = 14; }))); /* * access(2) is one of the few syscalls that will not (always) follow * up a successful VOP_LOOKUP with another VOP. */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); } // Lookup in a subdirectory of the fuse mount TEST_F(Lookup, subdir) { const char FULLPATH[] = "mountpoint/some_dir/some_file.txt"; const char DIRPATH[] = "some_dir"; const char RELPATH[] = "some_file.txt"; uint64_t dir_ino = 2; uint64_t file_ino = 3; EXPECT_LOOKUP(1, DIRPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFDIR | 0755; out->body.entry.nodeid = dir_ino; }))); EXPECT_LOOKUP(dir_ino, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = file_ino; }))); /* * access(2) is one of the few syscalls that will not (always) follow * up a successful VOP_LOOKUP with another VOP. */ ASSERT_EQ(0, access(FULLPATH, F_OK)) << strerror(errno); } - - Index: projects/fuse2/tests/sys/fs/fusefs/setattr.cc =================================================================== --- projects/fuse2/tests/sys/fs/fusefs/setattr.cc (revision 346100) +++ projects/fuse2/tests/sys/fs/fusefs/setattr.cc (revision 346101) @@ -1,557 +1,582 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 The FreeBSD Foundation * * This software was developed by BFF Storage Systems, LLC under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ extern "C" { #include #include } #include "mockfs.hh" #include "utils.hh" using namespace testing; class Setattr : public FuseTest {}; +class RofsSetattr: public Setattr { +public: +virtual void SetUp() { + m_ro = true; + Setattr::SetUp(); +} +}; + /* * If setattr returns a non-zero cache timeout, then subsequent VOP_GETATTRs * should use the cached attributes, rather than query the daemon */ TEST_F(Setattr, attr_cache) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; struct stat sb; const mode_t newmode = 0644; EXPECT_LOOKUP(1, RELPATH) .WillRepeatedly(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino; out->body.entry.entry_valid = UINT64_MAX; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | newmode; out->body.attr.attr_valid = UINT64_MAX; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { return (in->header.opcode == FUSE_GETATTR); }, Eq(true)), _) ).Times(0); /* Set an attribute with SETATTR */ ASSERT_EQ(0, chmod(FULLPATH, newmode)) << strerror(errno); /* The stat(2) should use cached attributes */ ASSERT_EQ(0, stat(FULLPATH, &sb)); EXPECT_EQ(S_IFREG | newmode, sb.st_mode); } /* Change the mode of a file */ TEST_F(Setattr, chmod) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const mode_t oldmode = 0755; const mode_t newmode = 0644; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | oldmode; out->body.entry.nodeid = ino; - out->body.entry.attr.mode = S_IFREG | oldmode; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_MODE; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.mode == newmode); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | newmode; }))); EXPECT_EQ(0, chmod(FULLPATH, newmode)) << strerror(errno); } /* Change the owner and group of a file */ TEST_F(Setattr, chown) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const gid_t oldgroup = 66; const gid_t newgroup = 99; const uid_t olduser = 33; const uid_t newuser = 44; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino; out->body.entry.attr.gid = oldgroup; out->body.entry.attr.uid = olduser; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_GID | FATTR_UID; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.uid == newuser && in->body.setattr.gid == newgroup); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | 0644; out->body.attr.attr.uid = newuser; out->body.attr.attr.gid = newgroup; }))); EXPECT_EQ(0, chown(FULLPATH, newuser, newgroup)) << strerror(errno); } /* * FUSE daemons are allowed to check permissions however they like. If the * daemon returns EPERM, even if the file permissions "should" grant access, * then fuse(4) should return EPERM too. */ TEST_F(Setattr, eperm) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0777; out->body.entry.nodeid = ino; out->body.entry.attr.uid = in->header.uid; out->body.entry.attr.gid = in->header.gid; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino); }, Eq(true)), _) ).WillOnce(Invoke(ReturnErrno(EPERM))); EXPECT_NE(0, truncate(FULLPATH, 10)); EXPECT_EQ(EPERM, errno); } /* Change the mode of an open file, by its file descriptor */ TEST_F(Setattr, fchmod) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; uint64_t ino = 42; int fd; const mode_t oldmode = 0755; const mode_t newmode = 0644; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | oldmode; out->body.entry.nodeid = ino; out->body.entry.attr_valid = UINT64_MAX; }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_OPEN && in->header.nodeid == ino); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { out->header.len = sizeof(out->header); SET_OUT_HEADER_LEN(out, open); }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_MODE; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.mode == newmode); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | newmode; }))); fd = open(FULLPATH, O_RDONLY); ASSERT_LE(0, fd) << strerror(errno); ASSERT_EQ(0, fchmod(fd, newmode)) << strerror(errno); /* Deliberately leak fd. close(2) will be tested in release.cc */ } /* Change the size of an open file, by its file descriptor */ TEST_F(Setattr, ftruncate) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; uint64_t ino = 42; int fd; uint64_t fh = 0xdeadbeef1a7ebabe; const off_t oldsize = 99; const off_t newsize = 12345; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0755; out->body.entry.nodeid = ino; out->body.entry.attr_valid = UINT64_MAX; out->body.entry.attr.size = oldsize; }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_OPEN && in->header.nodeid == ino); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { out->header.len = sizeof(out->header); SET_OUT_HEADER_LEN(out, open); out->body.open.fh = fh; }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_SIZE | FATTR_FH; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.fh == fh); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | 0755; out->body.attr.attr.size = newsize; }))); fd = open(FULLPATH, O_RDWR); ASSERT_LE(0, fd) << strerror(errno); ASSERT_EQ(0, ftruncate(fd, newsize)) << strerror(errno); /* Deliberately leak fd. close(2) will be tested in release.cc */ } /* Change the size of the file */ TEST_F(Setattr, truncate) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const uint64_t oldsize = 100'000'000; const uint64_t newsize = 20'000'000; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino; out->body.entry.attr.size = oldsize; }))); EXPECT_CALL(*m_mock, process( ResultOf([](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_SIZE; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.size == newsize); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | 0644; out->body.attr.attr.size = newsize; }))); EXPECT_EQ(0, truncate(FULLPATH, newsize)) << strerror(errno); } /* * Truncating a file should discard cached data past the truncation point. * This is a regression test for bug 233783. The bug only applies when * vfs.fusefs.data_cache_mode=1 or 2, but the test should pass regardless. */ TEST_F(Setattr, truncate_discards_cached_data) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; void *w0buf, *rbuf, *expected; off_t w0_offset = 0x1b8df; size_t w0_size = 0x61e8; off_t r_offset = 0xe1e6; off_t r_size = 0xe229; size_t trunc0_size = 0x10016; size_t trunc1_size = 131072; size_t cur_size = 0; const uint64_t ino = 42; mode_t mode = S_IFREG | 0644; int fd; w0buf = malloc(w0_size); ASSERT_NE(NULL, w0buf) << strerror(errno); memset(w0buf, 'X', w0_size); rbuf = malloc(r_size); ASSERT_NE(NULL, rbuf) << strerror(errno); expected = malloc(r_size); ASSERT_NE(NULL, expected) << strerror(errno); memset(expected, 0, r_size); expect_lookup(RELPATH, ino, mode, 0, 1); expect_open(ino, O_RDWR, 1); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_GETATTR && in->header.nodeid == ino); }, Eq(true)), _) ).WillRepeatedly(Invoke(ReturnImmediate([&](auto i __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; out->body.attr.attr.mode = mode; out->body.attr.attr.size = cur_size; }))); /* * The exact pattern of FUSE_WRITE operations depends on the setting of * vfs.fusefs.data_cache_mode. But it's not important for this test. * Just set the mocks to accept anything */ EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_WRITE); }, Eq(true)), _) ).WillRepeatedly(Invoke(ReturnImmediate([&](auto in, auto out) { SET_OUT_HEADER_LEN(out, write); out->body.attr.attr.ino = ino; out->body.write.size = in->body.write.size; cur_size = std::max(cur_size, in->body.write.size + in->body.write.offset); }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && (in->body.setattr.valid & FATTR_SIZE)); }, Eq(true)), _) ).WillRepeatedly(Invoke(ReturnImmediate([&](auto in, auto out) { auto trunc_size = in->body.setattr.size; SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; out->body.attr.attr.mode = mode; out->body.attr.attr.size = trunc_size; cur_size = trunc_size; }))); /* exact pattern of FUSE_READ depends on vfs.fusefs.data_cache_mode */ EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_READ); }, Eq(true)), _) ).WillRepeatedly(Invoke(ReturnImmediate([&](auto in, auto out) { auto osize = std::min(cur_size - in->body.read.offset, (size_t)in->body.read.size); out->header.len = sizeof(struct fuse_out_header) + osize; bzero(out->body.bytes, osize); }))); fd = open(FULLPATH, O_RDWR, 0644); ASSERT_LE(0, fd) << strerror(errno); ASSERT_EQ((ssize_t)w0_size, pwrite(fd, w0buf, w0_size, w0_offset)); /* 1st truncate should discard cached data */ EXPECT_EQ(0, ftruncate(fd, trunc0_size)) << strerror(errno); /* 2nd truncate extends file into previously cached data */ EXPECT_EQ(0, ftruncate(fd, trunc1_size)) << strerror(errno); /* Read should return all zeros */ ASSERT_EQ((ssize_t)r_size, pread(fd, rbuf, r_size, r_offset)); ASSERT_EQ(0, memcmp(expected, rbuf, r_size)); free(expected); free(rbuf); free(w0buf); } /* Change a file's timestamps */ TEST_F(Setattr, utimensat) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const timespec oldtimes[2] = { {.tv_sec = 1, .tv_nsec = 2}, {.tv_sec = 3, .tv_nsec = 4}, }; const timespec newtimes[2] = { {.tv_sec = 5, .tv_nsec = 6}, {.tv_sec = 7, .tv_nsec = 8}, }; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino; out->body.entry.attr_valid = UINT64_MAX; out->body.entry.attr.atime = oldtimes[0].tv_sec; out->body.entry.attr.atimensec = oldtimes[0].tv_nsec; out->body.entry.attr.mtime = oldtimes[1].tv_sec; out->body.entry.attr.mtimensec = oldtimes[1].tv_nsec; }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_ATIME | FATTR_MTIME; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.atime == newtimes[0].tv_sec && in->body.setattr.atimensec == newtimes[0].tv_nsec && in->body.setattr.mtime == newtimes[1].tv_sec && in->body.setattr.mtimensec == newtimes[1].tv_nsec); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | 0644; out->body.attr.attr.atime = newtimes[0].tv_sec; out->body.attr.attr.atimensec = newtimes[0].tv_nsec; out->body.attr.attr.mtime = newtimes[1].tv_sec; out->body.attr.attr.mtimensec = newtimes[1].tv_nsec; }))); EXPECT_EQ(0, utimensat(AT_FDCWD, FULLPATH, &newtimes[0], 0)) << strerror(errno); } /* Change a file mtime but not its atime */ TEST_F(Setattr, utimensat_mtime_only) { const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; const uint64_t ino = 42; const timespec oldtimes[2] = { {.tv_sec = 1, .tv_nsec = 2}, {.tv_sec = 3, .tv_nsec = 4}, }; const timespec newtimes[2] = { {.tv_sec = 5, .tv_nsec = UTIME_OMIT}, {.tv_sec = 7, .tv_nsec = 8}, }; EXPECT_LOOKUP(1, RELPATH) .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, entry); out->body.entry.attr.mode = S_IFREG | 0644; out->body.entry.nodeid = ino; out->body.entry.attr_valid = UINT64_MAX; out->body.entry.attr.atime = oldtimes[0].tv_sec; out->body.entry.attr.atimensec = oldtimes[0].tv_nsec; out->body.entry.attr.mtime = oldtimes[1].tv_sec; out->body.entry.attr.mtimensec = oldtimes[1].tv_nsec; }))); EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { /* In protocol 7.23, ctime will be changed too */ uint32_t valid = FATTR_MTIME; return (in->header.opcode == FUSE_SETATTR && in->header.nodeid == ino && in->body.setattr.valid == valid && in->body.setattr.mtime == newtimes[1].tv_sec && in->body.setattr.mtimensec == newtimes[1].tv_nsec); }, Eq(true)), _) ).WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, attr); out->body.attr.attr.ino = ino; // Must match nodeid out->body.attr.attr.mode = S_IFREG | 0644; out->body.attr.attr.atime = oldtimes[0].tv_sec; out->body.attr.attr.atimensec = oldtimes[0].tv_nsec; out->body.attr.attr.mtime = newtimes[1].tv_sec; out->body.attr.attr.mtimensec = newtimes[1].tv_nsec; }))); EXPECT_EQ(0, utimensat(AT_FDCWD, FULLPATH, &newtimes[0], 0)) << strerror(errno); } -// TODO: test for erofs +/* On a read-only mount, no attributes may be changed */ +TEST_F(RofsSetattr, erofs) +{ + const char FULLPATH[] = "mountpoint/some_file.txt"; + const char RELPATH[] = "some_file.txt"; + const uint64_t ino = 42; + const mode_t oldmode = 0755; + const mode_t newmode = 0644; + + EXPECT_LOOKUP(1, RELPATH) + .WillOnce(Invoke(ReturnImmediate([=](auto in __unused, auto out) { + SET_OUT_HEADER_LEN(out, entry); + out->body.entry.attr.mode = S_IFREG | oldmode; + out->body.entry.nodeid = ino; + }))); + + ASSERT_EQ(-1, chmod(FULLPATH, newmode)); + ASSERT_EQ(EROFS, errno); +} Index: projects/fuse2/tests/sys/fs/fusefs/xattr.cc =================================================================== --- projects/fuse2/tests/sys/fs/fusefs/xattr.cc (revision 346100) +++ projects/fuse2/tests/sys/fs/fusefs/xattr.cc (revision 346101) @@ -1,607 +1,640 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019 The FreeBSD Foundation * * This software was developed by BFF Storage Systems, LLC under sponsorship * from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ extern "C" { #include #include #include } #include "mockfs.hh" #include "utils.hh" using namespace testing; const char FULLPATH[] = "mountpoint/some_file.txt"; const char RELPATH[] = "some_file.txt"; /* For testing filesystems without posix locking support */ class Xattr: public FuseTest { public: void expect_getxattr(uint64_t ino, const char *attr, ProcessMockerT r) { EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { const char *a = (const char*)in->body.bytes + sizeof(fuse_getxattr_in); return (in->header.opcode == FUSE_GETXATTR && in->header.nodeid == ino && 0 == strcmp(attr, a)); }, Eq(true)), _) ).WillOnce(Invoke(r)); } void expect_listxattr(uint64_t ino, uint32_t size, ProcessMockerT r) { EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { return (in->header.opcode == FUSE_LISTXATTR && in->header.nodeid == ino && in->body.listxattr.size == size); }, Eq(true)), _) ).WillOnce(Invoke(r)) .RetiresOnSaturation(); } void expect_removexattr(uint64_t ino, const char *attr, int error) { EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { const char *a = (const char*)in->body.bytes; return (in->header.opcode == FUSE_REMOVEXATTR && in->header.nodeid == ino && 0 == strcmp(attr, a)); }, Eq(true)), _) ).WillOnce(Invoke(ReturnErrno(error))); } void expect_setxattr(uint64_t ino, const char *attr, const char *value, ProcessMockerT r) { EXPECT_CALL(*m_mock, process( ResultOf([=](auto in) { const char *a = (const char*)in->body.bytes + sizeof(fuse_setxattr_in); const char *v = a + strlen(a) + 1; return (in->header.opcode == FUSE_SETXATTR && in->header.nodeid == ino && 0 == strcmp(attr, a) && 0 == strcmp(value, v)); }, Eq(true)), _) ).WillOnce(Invoke(r)); } }; class Getxattr: public Xattr {}; class Listxattr: public Xattr {}; class Removexattr: public Xattr {}; class Setxattr: public Xattr {}; +class RofsXattr: public Xattr { +public: +virtual void SetUp() { + m_ro = true; + Xattr::SetUp(); +} +}; /* * If the extended attribute does not exist on this file, the daemon should * return ENOATTR (ENODATA on Linux, but it's up to the daemon to choose the * correct errror code) */ TEST_F(Getxattr, enoattr) { char data[80]; uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "user.foo", ReturnErrno(ENOATTR)); r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(-1, r); ASSERT_EQ(ENOATTR, errno); } /* * If the filesystem returns ENOSYS, then it will be treated as a permanent * failure and all future VOP_GETEXTATTR calls will fail with EOPNOTSUPP * without querying the filesystem daemon */ /* https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236557 */ TEST_F(Getxattr, DISABLED_enosys) { char data[80]; uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "user.foo", ReturnErrno(ENOSYS)); r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(-1, r); EXPECT_EQ(EOPNOTSUPP, errno); /* Subsequent attempts should not query the filesystem at all */ r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(-1, r); EXPECT_EQ(EOPNOTSUPP, errno); } /* * On FreeBSD, if the user passes an insufficiently large buffer then the * filesystem is supposed to copy as much of the attribute's value as will fit. * * On Linux, however, the filesystem is supposed to return ERANGE. * * libfuse specifies the Linux behavior. However, that's probably an error. * It would probably be correct for the filesystem to use platform-dependent * behavior. * * This test case covers a filesystem that uses the Linux behavior */ TEST_F(Getxattr, erange) { char data[10]; uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "user.foo", ReturnErrno(ERANGE)); r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(-1, r); ASSERT_EQ(ERANGE, errno); } /* * If the user passes a 0-length buffer, then the daemon should just return the * size of the attribute */ TEST_F(Getxattr, size_only) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "user.foo", ReturnImmediate([](auto in __unused, auto out) { SET_OUT_HEADER_LEN(out, getxattr); out->body.getxattr.size = 99; }) ); ASSERT_EQ(99, extattr_get_file(FULLPATH, ns, "foo", NULL, 0)) << strerror(errno);; } /* * Successfully get an attribute from the system namespace */ TEST_F(Getxattr, system) { uint64_t ino = 42; char data[80]; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_SYSTEM; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "system.foo", ReturnImmediate([&](auto in __unused, auto out) { memcpy((void*)out->body.bytes, value, value_len); out->header.len = sizeof(out->header) + value_len; }) ); r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(value_len, r) << strerror(errno); EXPECT_STREQ(value, data); } /* * Successfully get an attribute from the user namespace */ TEST_F(Getxattr, user) { uint64_t ino = 42; char data[80]; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_getxattr(ino, "user.foo", ReturnImmediate([&](auto in __unused, auto out) { memcpy((void*)out->body.bytes, value, value_len); out->header.len = sizeof(out->header) + value_len; }) ); r = extattr_get_file(FULLPATH, ns, "foo", data, sizeof(data)); ASSERT_EQ(value_len, r) << strerror(errno); EXPECT_STREQ(value, data); } /* * If the filesystem returns ENOSYS, then it will be treated as a permanent * failure and all future VOP_LISTEXTATTR calls will fail with EOPNOTSUPP * without querying the filesystem daemon */ /* https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236557 */ TEST_F(Listxattr, DISABLED_enosys) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnErrno(ENOSYS)); ASSERT_EQ(-1, extattr_list_file(FULLPATH, ns, NULL, 0)); EXPECT_EQ(EOPNOTSUPP, errno); /* Subsequent attempts should not query the filesystem at all */ ASSERT_EQ(-1, extattr_list_file(FULLPATH, ns, NULL, 0)); EXPECT_EQ(EOPNOTSUPP, errno); } /* * Listing extended attributes failed because they aren't configured on this * filesystem */ TEST_F(Listxattr, enotsup) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnErrno(ENOTSUP)); ASSERT_EQ(-1, extattr_list_file(FULLPATH, ns, NULL, 0)); ASSERT_EQ(ENOTSUP, errno); } /* * On FreeBSD, if the user passes an insufficiently large buffer then the * filesystem is supposed to copy as much of the attribute's value as will fit. * * On Linux, however, the filesystem is supposed to return ERANGE. * * libfuse specifies the Linux behavior. However, that's probably an error. * It would probably be correct for the filesystem to use platform-dependent * behavior. * * This test case covers a filesystem that uses the Linux behavior */ TEST_F(Listxattr, erange) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnErrno(ERANGE)); ASSERT_EQ(-1, extattr_list_file(FULLPATH, ns, NULL, 0)); ASSERT_EQ(ERANGE, errno); } /* * Get the size of the list that it would take to list no extended attributes */ TEST_F(Listxattr, size_only_empty) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnImmediate([](auto i __unused, auto out) { out->body.listxattr.size = 0; SET_OUT_HEADER_LEN(out, listxattr); })); ASSERT_EQ(0, extattr_list_file(FULLPATH, ns, NULL, 0)) << strerror(errno); } /* * Get the size of the list that it would take to list some extended * attributes. Due to the format differences between a FreeBSD and a * Linux/FUSE extended attribute list, fuse(4) will actually allocate a buffer * and get the whole list, then convert it, just to figure out its size. */ TEST_F(Listxattr, size_only_nonempty) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnImmediate([](auto i __unused, auto out) { out->body.listxattr.size = 45; SET_OUT_HEADER_LEN(out, listxattr); })); // TODO: fix the expected size after fixing the size calculation bug in // fuse_vnop_listextattr. It should be exactly 45. expect_listxattr(ino, 53, ReturnImmediate([](auto in __unused, auto out) { const char l[] = "user.foo"; strlcpy((char*)out->body.bytes, l, sizeof(out->body.bytes)); out->header.len = sizeof(fuse_out_header) + sizeof(l); }) ); ASSERT_EQ(4, extattr_list_file(FULLPATH, ns, NULL, 0)) << strerror(errno); } TEST_F(Listxattr, size_only_really_big) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnImmediate([](auto i __unused, auto out) { out->body.listxattr.size = 16000; SET_OUT_HEADER_LEN(out, listxattr); })); // TODO: fix the expected size after fixing the size calculation bug in // fuse_vnop_listextattr. It should be exactly 16000. expect_listxattr(ino, 16008, ReturnImmediate([](auto in __unused, auto out) { const char l[16] = "user.foobarbang"; for (int i=0; i < 1000; i++) { memcpy(&out->body.bytes[16 * i], l, 16); } out->header.len = sizeof(fuse_out_header) + 16000; }) ); ASSERT_EQ(11000, extattr_list_file(FULLPATH, ns, NULL, 0)) << strerror(errno); } /* * List all of the user attributes of a file which has both user and system * attributes */ TEST_F(Listxattr, user) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; char data[80]; char expected[9] = {3, 'f', 'o', 'o', 4, 'b', 'a', 'n', 'g'}; char attrs[28] = "user.foo\0system.x\0user.bang"; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnImmediate([&](auto in __unused, auto out) { out->body.listxattr.size = sizeof(attrs); SET_OUT_HEADER_LEN(out, listxattr); }) ); // TODO: fix the expected size after fixing the size calculation bug in // fuse_vnop_listextattr. expect_listxattr(ino, sizeof(attrs) + 8, ReturnImmediate([&](auto in __unused, auto out) { memcpy((void*)out->body.bytes, attrs, sizeof(attrs)); out->header.len = sizeof(fuse_out_header) + sizeof(attrs); })); ASSERT_EQ((ssize_t)sizeof(expected), extattr_list_file(FULLPATH, ns, data, sizeof(data))) << strerror(errno); ASSERT_EQ(0, memcmp(expected, data, sizeof(expected))); } /* * List all of the system attributes of a file which has both user and system * attributes */ TEST_F(Listxattr, system) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_SYSTEM; char data[80]; char expected[2] = {1, 'x'}; char attrs[28] = "user.foo\0system.x\0user.bang"; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_listxattr(ino, 0, ReturnImmediate([&](auto in __unused, auto out) { out->body.listxattr.size = sizeof(attrs); SET_OUT_HEADER_LEN(out, listxattr); }) ); // TODO: fix the expected size after fixing the size calculation bug in // fuse_vnop_listextattr. expect_listxattr(ino, sizeof(attrs) + 8, ReturnImmediate([&](auto in __unused, auto out) { memcpy((void*)out->body.bytes, attrs, sizeof(attrs)); out->header.len = sizeof(fuse_out_header) + sizeof(attrs); })); ASSERT_EQ((ssize_t)sizeof(expected), extattr_list_file(FULLPATH, ns, data, sizeof(data))) << strerror(errno); ASSERT_EQ(0, memcmp(expected, data, sizeof(expected))); } /* Fail to remove a nonexistent attribute */ TEST_F(Removexattr, enoattr) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_removexattr(ino, "user.foo", ENOATTR); ASSERT_EQ(-1, extattr_delete_file(FULLPATH, ns, "foo")); ASSERT_EQ(ENOATTR, errno); } /* * If the filesystem returns ENOSYS, then it will be treated as a permanent * failure and all future VOP_DELETEEXTATTR calls will fail with EOPNOTSUPP * without querying the filesystem daemon */ /* https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236557 */ TEST_F(Removexattr, DISABLED_enosys) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_removexattr(ino, "user.foo", ENOSYS); ASSERT_EQ(-1, extattr_delete_file(FULLPATH, ns, "foo")); EXPECT_EQ(EOPNOTSUPP, errno); /* Subsequent attempts should not query the filesystem at all */ ASSERT_EQ(-1, extattr_delete_file(FULLPATH, ns, "foo")); EXPECT_EQ(EOPNOTSUPP, errno); } /* Successfully remove a user xattr */ TEST_F(Removexattr, user) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_USER; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_removexattr(ino, "user.foo", 0); ASSERT_EQ(0, extattr_delete_file(FULLPATH, ns, "foo")) << strerror(errno); } /* Successfully remove a system xattr */ TEST_F(Removexattr, system) { uint64_t ino = 42; int ns = EXTATTR_NAMESPACE_SYSTEM; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_removexattr(ino, "system.foo", 0); ASSERT_EQ(0, extattr_delete_file(FULLPATH, ns, "foo")) << strerror(errno); } /* * If the filesystem returns ENOSYS, then it will be treated as a permanent * failure and all future VOP_SETEXTATTR calls will fail with EOPNOTSUPP * without querying the filesystem daemon */ /* https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=236557 */ TEST_F(Setxattr, DISABLED_enosys) { uint64_t ino = 42; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_setxattr(ino, "user.foo", value, ReturnErrno(ENOSYS)); r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); ASSERT_EQ(-1, r); EXPECT_EQ(EOPNOTSUPP, errno); /* Subsequent attempts should not query the filesystem at all */ r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); ASSERT_EQ(-1, r); EXPECT_EQ(EOPNOTSUPP, errno); } /* * SETXATTR will return ENOTSUP if the namespace is invalid or the filesystem * as currently configured doesn't support extended attributes. */ TEST_F(Setxattr, enotsup) { uint64_t ino = 42; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_setxattr(ino, "user.foo", value, ReturnErrno(ENOTSUP)); r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); ASSERT_EQ(-1, r); EXPECT_EQ(ENOTSUP, errno); } /* * Successfully set a user attribute. */ TEST_F(Setxattr, user) { uint64_t ino = 42; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_USER; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_setxattr(ino, "user.foo", value, ReturnErrno(0)); r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); ASSERT_EQ(value_len, r) << strerror(errno); } /* * Successfully set a system attribute. */ TEST_F(Setxattr, system) { uint64_t ino = 42; const char value[] = "whatever"; ssize_t value_len = strlen(value) + 1; int ns = EXTATTR_NAMESPACE_SYSTEM; ssize_t r; expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); expect_setxattr(ino, "system.foo", value, ReturnErrno(0)); r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); ASSERT_EQ(value_len, r) << strerror(errno); +} + +TEST_F(RofsXattr, deleteextattr_erofs) +{ + uint64_t ino = 42; + int ns = EXTATTR_NAMESPACE_USER; + + expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); + + ASSERT_EQ(-1, extattr_delete_file(FULLPATH, ns, "foo")); + ASSERT_EQ(EROFS, errno); +} + +TEST_F(RofsXattr, setextattr_erofs) +{ + uint64_t ino = 42; + const char value[] = "whatever"; + ssize_t value_len = strlen(value) + 1; + int ns = EXTATTR_NAMESPACE_USER; + ssize_t r; + + expect_lookup(RELPATH, ino, S_IFREG | 0644, 0, 1); + + r = extattr_set_file(FULLPATH, ns, "foo", (void*)value, value_len); + ASSERT_EQ(-1, r); + EXPECT_EQ(EROFS, errno); } // TODO: EROFS tests