+ if (nmp->nm_vers == NFS_VER2) {
+ NFS_BITMAP_SET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXNAME);
+ nmp->nm_fsattr.nfsa_maxname = NFS_MAXNAMLEN;
+ } else if (nmp->nm_vers == NFS_VER3) {
+ /* get the NFSv3 FSINFO */
+ error = nfs3_fsinfo(nmp, *npp, ctx);
+ if (error)
+ goto out;
+ /* If the server indicates all pathconf info is */
+ /* the same, grab a copy of that info now */
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_HOMOGENEOUS) &&
+ (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS)) {
+ struct nfs_fsattr nfsa;
+ if (!nfs3_pathconf_rpc(*npp, &nfsa, ctx)) {
+ /* cache a copy of the results */
+ lck_mtx_lock(&nmp->nm_lock);
+ nfs3_pathconf_cache(nmp, &nfsa);
+ lck_mtx_unlock(&nmp->nm_lock);
+ }
+ }
+ }
+out:
+ if (*npp && error) {
+ vnode_put(NFSTOV(*npp));
+ vnode_recycle(NFSTOV(*npp));
+ *npp = NULL;
+ }
+ return (error);
+}
+
+/*
+ * Update an NFSv4 mount path with the contents of the symlink.
+ *
+ * Read the link for the given file handle.
+ * Insert the link's components into the path.
+ */
+int
+nfs4_mount_update_path_with_symlink(struct nfsmount *nmp, struct nfs_fs_path *nfsp, uint32_t curcomp, fhandle_t *dirfhp, int *depthp, fhandle_t *fhp, vfs_context_t ctx)
+{
+ int error = 0, status, numops;
+ uint32_t len = 0, comp, newcomp, linkcompcount;
+ u_int64_t xid;
+ struct nfsm_chain nmreq, nmrep;
+ struct nfsreq rq, *req = &rq;
+ struct nfsreq_secinfo_args si;
+ char *link = NULL, *p, *q, ch;
+ struct nfs_fs_path nfsp2;
+
+ bzero(&nfsp2, sizeof(nfsp2));
+ if (dirfhp->fh_len)
+ NFSREQ_SECINFO_SET(&si, NULL, dirfhp->fh_data, dirfhp->fh_len, nfsp->np_components[curcomp], 0);
+ else
+ NFSREQ_SECINFO_SET(&si, NULL, NULL, 0, nfsp->np_components[curcomp], 0);
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ MALLOC_ZONE(link, char *, MAXPATHLEN, M_NAMEI, M_WAITOK);
+ if (!link)
+ error = ENOMEM;
+
+ // PUTFH, READLINK
+ numops = 2;
+ nfsm_chain_build_alloc_init(error, &nmreq, 12 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "readlink", nmp->nm_minor_vers, numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, fhp->fh_data, fhp->fh_len);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_READLINK);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+
+ error = nfs_request_async(NULL, nmp->nm_mountp, &nmreq, NFSPROC4_COMPOUND,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), &si, 0, NULL, &req);
+ if (!error)
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTFH);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_READLINK);
+ nfsm_chain_get_32(error, &nmrep, len);
+ nfsmout_if(error);
+ if (len == 0)
+ error = ENOENT;
+ else if (len >= MAXPATHLEN)
+ len = MAXPATHLEN - 1;
+ nfsm_chain_get_opaque(error, &nmrep, len, link);
+ nfsmout_if(error);
+ /* make sure link string is terminated properly */
+ link[len] = '\0';
+
+ /* count the number of components in link */
+ p = link;
+ while (*p && (*p == '/'))
+ p++;
+ linkcompcount = 0;
+ while (*p) {
+ linkcompcount++;
+ while (*p && (*p != '/'))
+ p++;
+ while (*p && (*p == '/'))
+ p++;
+ }
+
+ /* free up used components */
+ for (comp=0; comp <= curcomp; comp++) {
+ if (nfsp->np_components[comp]) {
+ FREE(nfsp->np_components[comp], M_TEMP);
+ nfsp->np_components[comp] = NULL;
+ }
+ }
+
+ /* set up new path */
+ nfsp2.np_compcount = nfsp->np_compcount - curcomp - 1 + linkcompcount;
+ MALLOC(nfsp2.np_components, char **, nfsp2.np_compcount*sizeof(char*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!nfsp2.np_components) {
+ error = ENOMEM;
+ goto nfsmout;
+ }
+
+ /* add link components */
+ p = link;
+ while (*p && (*p == '/'))
+ p++;
+ for (newcomp=0; newcomp < linkcompcount; newcomp++) {
+ /* find end of component */
+ q = p;
+ while (*q && (*q != '/'))
+ q++;
+ MALLOC(nfsp2.np_components[newcomp], char *, q-p+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!nfsp2.np_components[newcomp]) {
+ error = ENOMEM;
+ break;
+ }
+ ch = *q;
+ *q = '\0';
+ strlcpy(nfsp2.np_components[newcomp], p, q-p+1);
+ *q = ch;
+ p = q;
+ while (*p && (*p == '/'))
+ p++;
+ }
+ nfsmout_if(error);
+
+ /* add remaining components */
+ for(comp = curcomp + 1; comp < nfsp->np_compcount; comp++,newcomp++) {
+ nfsp2.np_components[newcomp] = nfsp->np_components[comp];
+ nfsp->np_components[comp] = NULL;
+ }
+
+ /* move new path into place */
+ FREE(nfsp->np_components, M_TEMP);
+ nfsp->np_components = nfsp2.np_components;
+ nfsp->np_compcount = nfsp2.np_compcount;
+ nfsp2.np_components = NULL;
+
+ /* for absolute link, let the caller now that the next dirfh is root */
+ if (link[0] == '/') {
+ dirfhp->fh_len = 0;
+ *depthp = 0;
+ }
+nfsmout:
+ if (link)
+ FREE_ZONE(link, MAXPATHLEN, M_NAMEI);
+ if (nfsp2.np_components) {
+ for (comp=0; comp < nfsp2.np_compcount; comp++)
+ if (nfsp2.np_components[comp])
+ FREE(nfsp2.np_components[comp], M_TEMP);
+ FREE(nfsp2.np_components, M_TEMP);
+ }
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/* Set up an NFSv4 mount */
+int
+nfs4_mount(
+ struct nfsmount *nmp,
+ vfs_context_t ctx,
+ nfsnode_t *npp)
+{
+ struct nfsm_chain nmreq, nmrep;
+ int error = 0, numops, status, interval, isdotdot, loopcnt = 0, depth = 0;
+ struct nfs_fs_path fspath, *nfsp, fspath2;
+ uint32_t bitmap[NFS_ATTR_BITMAP_LEN], comp, comp2;
+ fhandle_t fh, dirfh;
+ struct nfs_vattr nvattr;
+ u_int64_t xid;
+ struct nfsreq rq, *req = &rq;
+ struct nfsreq_secinfo_args si;
+ struct nfs_sec sec;
+ struct nfs_fs_locations nfsls;
+
+ *npp = NULL;
+ fh.fh_len = dirfh.fh_len = 0;
+ TAILQ_INIT(&nmp->nm_open_owners);
+ TAILQ_INIT(&nmp->nm_delegations);
+ TAILQ_INIT(&nmp->nm_dreturnq);
+ nmp->nm_stategenid = 1;
+ NVATTR_INIT(&nvattr);
+ bzero(&nfsls, sizeof(nfsls));
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ /*
+ * If no security flavors were specified we'll want to default to the server's
+ * preferred flavor. For NFSv4.0 we need a file handle and name to get that via
+ * SECINFO, so we'll do that on the last component of the server path we are
+ * mounting. If we are mounting the server's root, we'll need to defer the
+ * SECINFO call to the first successful LOOKUP request.
+ */
+ if (!nmp->nm_sec.count)
+ nmp->nm_state |= NFSSTA_NEEDSECINFO;
+
+ /* make a copy of the current location's path */
+ nfsp = &nmp->nm_locations.nl_locations[nmp->nm_locations.nl_current.nli_loc]->nl_path;
+ bzero(&fspath, sizeof(fspath));
+ fspath.np_compcount = nfsp->np_compcount;
+ if (fspath.np_compcount > 0) {
+ MALLOC(fspath.np_components, char **, fspath.np_compcount*sizeof(char*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fspath.np_components) {
+ error = ENOMEM;
+ goto nfsmout;
+ }
+ for (comp=0; comp < nfsp->np_compcount; comp++) {
+ int slen = strlen(nfsp->np_components[comp]);
+ MALLOC(fspath.np_components[comp], char *, slen+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!fspath.np_components[comp]) {
+ error = ENOMEM;
+ break;
+ }
+ strlcpy(fspath.np_components[comp], nfsp->np_components[comp], slen+1);
+ }
+ if (error)
+ goto nfsmout;
+ }
+
+ /* for mirror mounts, we can just use the file handle passed in */
+ if (nmp->nm_fh) {
+ dirfh.fh_len = nmp->nm_fh->fh_len;
+ bcopy(nmp->nm_fh->fh_data, dirfh.fh_data, dirfh.fh_len);
+ NFSREQ_SECINFO_SET(&si, NULL, dirfh.fh_data, dirfh.fh_len, NULL, 0);
+ goto gotfh;
+ }
+
+ /* otherwise, we need to get the fh for the directory we are mounting */
+
+ /* if no components, just get root */
+ if (fspath.np_compcount == 0) {
+nocomponents:
+ // PUTROOTFH + GETATTR(FH)
+ NFSREQ_SECINFO_SET(&si, NULL, NULL, 0, NULL, 0);
+ numops = 2;
+ nfsm_chain_build_alloc_init(error, &nmreq, 9 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "mount", nmp->nm_minor_vers, numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTROOTFH);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ NFS_CLEAR_ATTRIBUTES(bitmap);
+ NFS4_DEFAULT_ATTRIBUTES(bitmap);
+ NFS_BITMAP_SET(bitmap, NFS_FATTR_FILEHANDLE);
+ nfsm_chain_add_bitmap(error, &nmreq, bitmap, NFS_ATTR_BITMAP_LEN);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+ error = nfs_request_async(NULL, nmp->nm_mountp, &nmreq, NFSPROC4_COMPOUND,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), &si, 0, NULL, &req);
+ if (!error)
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTROOTFH);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsmout_if(error);
+ NFS_CLEAR_ATTRIBUTES(nmp->nm_fsattr.nfsa_bitmap);
+ error = nfs4_parsefattr(&nmrep, &nmp->nm_fsattr, &nvattr, &dirfh, NULL, NULL);
+ if (!error && !NFS_BITMAP_ISSET(&nvattr.nva_bitmap, NFS_FATTR_FILEHANDLE)) {
+ printf("nfs: mount didn't return filehandle?\n");
+ error = EBADRPC;
+ }
+ nfsmout_if(error);
+ nfsm_chain_cleanup(&nmrep);
+ nfsm_chain_null(&nmreq);
+ NVATTR_CLEANUP(&nvattr);
+ goto gotfh;
+ }
+
+ /* look up each path component */
+ for (comp=0; comp < fspath.np_compcount; ) {
+ isdotdot = 0;
+ if (fspath.np_components[comp][0] == '.') {
+ if (fspath.np_components[comp][1] == '\0') {
+ /* skip "." */
+ comp++;
+ continue;
+ }
+ /* treat ".." specially */
+ if ((fspath.np_components[comp][1] == '.') &&
+ (fspath.np_components[comp][2] == '\0'))
+ isdotdot = 1;
+ if (isdotdot && (dirfh.fh_len == 0)) {
+ /* ".." in root directory is same as "." */
+ comp++;
+ continue;
+ }
+ }
+ // PUT(ROOT)FH + LOOKUP(P) + GETFH + GETATTR
+ if (dirfh.fh_len == 0)
+ NFSREQ_SECINFO_SET(&si, NULL, NULL, 0, isdotdot ? NULL : fspath.np_components[comp], 0);
+ else
+ NFSREQ_SECINFO_SET(&si, NULL, dirfh.fh_data, dirfh.fh_len, isdotdot ? NULL : fspath.np_components[comp], 0);
+ numops = 4;
+ nfsm_chain_build_alloc_init(error, &nmreq, 18 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "mount", nmp->nm_minor_vers, numops);
+ numops--;
+ if (dirfh.fh_len) {
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, dirfh.fh_data, dirfh.fh_len);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTROOTFH);
+ }
+ numops--;
+ if (isdotdot) {
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_LOOKUPP);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_LOOKUP);
+ nfsm_chain_add_name(error, &nmreq,
+ fspath.np_components[comp], strlen(fspath.np_components[comp]), nmp);
+ }
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETFH);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ NFS_CLEAR_ATTRIBUTES(bitmap);
+ NFS4_DEFAULT_ATTRIBUTES(bitmap);
+ /* if no namedattr support or component is ".zfs", clear NFS_FATTR_NAMED_ATTR */
+ if (!NMFLAG(nmp, NAMEDATTR) || !strcmp(fspath.np_components[comp], ".zfs"))
+ NFS_BITMAP_CLR(bitmap, NFS_FATTR_NAMED_ATTR);
+ nfsm_chain_add_bitmap(error, &nmreq, bitmap, NFS_ATTR_BITMAP_LEN);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+ error = nfs_request_async(NULL, nmp->nm_mountp, &nmreq, NFSPROC4_COMPOUND,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), &si, 0, NULL, &req);
+ if (!error)
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, dirfh.fh_len ? NFS_OP_PUTFH : NFS_OP_PUTROOTFH);
+ nfsm_chain_op_check(error, &nmrep, isdotdot ? NFS_OP_LOOKUPP : NFS_OP_LOOKUP);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETFH);
+ nfsm_chain_get_32(error, &nmrep, fh.fh_len);
+ nfsm_chain_get_opaque(error, &nmrep, fh.fh_len, fh.fh_data);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ if (!error) {
+ NFS_CLEAR_ATTRIBUTES(nmp->nm_fsattr.nfsa_bitmap);
+ error = nfs4_parsefattr(&nmrep, &nmp->nm_fsattr, &nvattr, NULL, NULL, &nfsls);
+ }
+ nfsm_chain_cleanup(&nmrep);
+ nfsm_chain_null(&nmreq);
+ if (error) {
+ /* LOOKUP succeeded but GETATTR failed? This could be a referral. */
+ /* Try the lookup again with a getattr for fs_locations. */
+ nfs_fs_locations_cleanup(&nfsls);
+ error = nfs4_get_fs_locations(nmp, NULL, dirfh.fh_data, dirfh.fh_len, fspath.np_components[comp], ctx, &nfsls);
+ if (!error && (nfsls.nl_numlocs < 1))
+ error = ENOENT;
+ nfsmout_if(error);
+ if (++loopcnt > MAXSYMLINKS) {
+ /* too many symlink/referral redirections */
+ error = ELOOP;
+ goto nfsmout;
+ }
+ /* tear down the current connection */
+ nfs_disconnect(nmp);
+ /* replace fs locations */
+ nfs_fs_locations_cleanup(&nmp->nm_locations);
+ nmp->nm_locations = nfsls;
+ bzero(&nfsls, sizeof(nfsls));
+ /* initiate a connection using the new fs locations */
+ error = nfs_mount_connect(nmp);
+ if (!error && !(nmp->nm_locations.nl_current.nli_flags & NLI_VALID))
+ error = EIO;
+ nfsmout_if(error);
+ /* add new server's remote path to beginning of our path and continue */
+ nfsp = &nmp->nm_locations.nl_locations[nmp->nm_locations.nl_current.nli_loc]->nl_path;
+ bzero(&fspath2, sizeof(fspath2));
+ fspath2.np_compcount = (fspath.np_compcount - comp - 1) + nfsp->np_compcount;
+ if (fspath2.np_compcount > 0) {
+ MALLOC(fspath2.np_components, char **, fspath2.np_compcount*sizeof(char*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fspath2.np_components) {
+ error = ENOMEM;
+ goto nfsmout;
+ }
+ for (comp2=0; comp2 < nfsp->np_compcount; comp2++) {
+ int slen = strlen(nfsp->np_components[comp2]);
+ MALLOC(fspath2.np_components[comp2], char *, slen+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!fspath2.np_components[comp2]) {
+ /* clean up fspath2, then error out */
+ while (comp2 > 0) {
+ comp2--;
+ FREE(fspath2.np_components[comp2], M_TEMP);
+ }
+ FREE(fspath2.np_components, M_TEMP);
+ error = ENOMEM;
+ goto nfsmout;
+ }
+ strlcpy(fspath2.np_components[comp2], nfsp->np_components[comp2], slen+1);
+ }
+ if ((fspath.np_compcount - comp - 1) > 0)
+ bcopy(&fspath.np_components[comp+1], &fspath2.np_components[nfsp->np_compcount], (fspath.np_compcount - comp - 1)*sizeof(char*));
+ /* free up unused parts of old path (prior components and component array) */
+ do {
+ FREE(fspath.np_components[comp], M_TEMP);
+ } while (comp-- > 0);
+ FREE(fspath.np_components, M_TEMP);
+ /* put new path in place */
+ fspath = fspath2;
+ }
+ /* reset dirfh and component index */
+ dirfh.fh_len = 0;
+ comp = 0;
+ NVATTR_CLEANUP(&nvattr);
+ if (fspath.np_compcount == 0)
+ goto nocomponents;
+ continue;
+ }
+ nfsmout_if(error);
+ /* if file handle is for a symlink, then update the path with the symlink contents */
+ if (NFS_BITMAP_ISSET(&nvattr.nva_bitmap, NFS_FATTR_TYPE) && (nvattr.nva_type == VLNK)) {
+ if (++loopcnt > MAXSYMLINKS)
+ error = ELOOP;
+ else
+ error = nfs4_mount_update_path_with_symlink(nmp, &fspath, comp, &dirfh, &depth, &fh, ctx);
+ nfsmout_if(error);
+ /* directory file handle is either left the same or reset to root (if link was absolute) */
+ /* path traversal starts at beginning of the path again */
+ comp = 0;
+ NVATTR_CLEANUP(&nvattr);
+ nfs_fs_locations_cleanup(&nfsls);
+ continue;
+ }
+ NVATTR_CLEANUP(&nvattr);
+ nfs_fs_locations_cleanup(&nfsls);
+ /* not a symlink... */
+ if ((nmp->nm_state & NFSSTA_NEEDSECINFO) && (comp == (fspath.np_compcount-1)) && !isdotdot) {
+ /* need to get SECINFO for the directory being mounted */
+ if (dirfh.fh_len == 0)
+ NFSREQ_SECINFO_SET(&si, NULL, NULL, 0, isdotdot ? NULL : fspath.np_components[comp], 0);
+ else
+ NFSREQ_SECINFO_SET(&si, NULL, dirfh.fh_data, dirfh.fh_len, isdotdot ? NULL : fspath.np_components[comp], 0);
+ sec.count = NX_MAX_SEC_FLAVORS;
+ error = nfs4_secinfo_rpc(nmp, &si, vfs_context_ucred(ctx), sec.flavors, &sec.count);
+ /* [sigh] some implementations return "illegal" error for unsupported ops */
+ if (error == NFSERR_OP_ILLEGAL)
+ error = 0;
+ nfsmout_if(error);
+ /* set our default security flavor to the first in the list */
+ if (sec.count)
+ nmp->nm_auth = sec.flavors[0];
+ nmp->nm_state &= ~NFSSTA_NEEDSECINFO;
+ }
+ /* advance directory file handle, component index, & update depth */
+ dirfh = fh;
+ comp++;
+ if (!isdotdot) /* going down the hierarchy */
+ depth++;
+ else if (--depth <= 0) /* going up the hierarchy */
+ dirfh.fh_len = 0; /* clear dirfh when we hit root */
+ }
+
+gotfh:
+ /* get attrs for mount point root */
+ numops = NMFLAG(nmp, NAMEDATTR) ? 3 : 2; // PUTFH + GETATTR + OPENATTR
+ nfsm_chain_build_alloc_init(error, &nmreq, 25 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "mount", nmp->nm_minor_vers, numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, dirfh.fh_data, dirfh.fh_len);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ NFS_CLEAR_ATTRIBUTES(bitmap);
+ NFS4_DEFAULT_ATTRIBUTES(bitmap);
+ /* if no namedattr support or last component is ".zfs", clear NFS_FATTR_NAMED_ATTR */
+ if (!NMFLAG(nmp, NAMEDATTR) || ((fspath.np_compcount > 0) && !strcmp(fspath.np_components[fspath.np_compcount-1], ".zfs")))
+ NFS_BITMAP_CLR(bitmap, NFS_FATTR_NAMED_ATTR);
+ nfsm_chain_add_bitmap(error, &nmreq, bitmap, NFS_ATTR_BITMAP_LEN);
+ if (NMFLAG(nmp, NAMEDATTR)) {
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_OPENATTR);
+ nfsm_chain_add_32(error, &nmreq, 0);
+ }
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+ error = nfs_request_async(NULL, nmp->nm_mountp, &nmreq, NFSPROC4_COMPOUND,
+ vfs_context_thread(ctx), vfs_context_ucred(ctx), &si, 0, NULL, &req);
+ if (!error)
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTFH);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsmout_if(error);
+ NFS_CLEAR_ATTRIBUTES(nmp->nm_fsattr.nfsa_bitmap);
+ error = nfs4_parsefattr(&nmrep, &nmp->nm_fsattr, &nvattr, NULL, NULL, NULL);
+ nfsmout_if(error);
+ if (NMFLAG(nmp, NAMEDATTR)) {
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_OPENATTR);
+ if (error == ENOENT)
+ error = 0;
+ /* [sigh] some implementations return "illegal" error for unsupported ops */
+ if (error || !NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_supp_attr, NFS_FATTR_NAMED_ATTR)) {
+ nmp->nm_fsattr.nfsa_flags &= ~NFS_FSFLAG_NAMED_ATTR;
+ } else {
+ nmp->nm_fsattr.nfsa_flags |= NFS_FSFLAG_NAMED_ATTR;
+ }
+ } else {
+ nmp->nm_fsattr.nfsa_flags &= ~NFS_FSFLAG_NAMED_ATTR;
+ }
+ if (NMFLAG(nmp, NOACL)) /* make sure ACL support is turned off */
+ nmp->nm_fsattr.nfsa_flags &= ~NFS_FSFLAG_ACL;
+ if (NMFLAG(nmp, ACLONLY) && !(nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_ACL))
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_ACLONLY);
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_supp_attr, NFS_FATTR_FH_EXPIRE_TYPE)) {
+ uint32_t fhtype = ((nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_FHTYPE_MASK) >> NFS_FSFLAG_FHTYPE_SHIFT);
+ if (fhtype != NFS_FH_PERSISTENT)
+ printf("nfs: warning: non-persistent file handles! for %s\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
+ }
+
+ /* make sure it's a directory */
+ if (!NFS_BITMAP_ISSET(&nvattr.nva_bitmap, NFS_FATTR_TYPE) || (nvattr.nva_type != VDIR)) {
+ error = ENOTDIR;
+ goto nfsmout;
+ }
+
+ /* save the NFS fsid */
+ nmp->nm_fsid = nvattr.nva_fsid;
+
+ /* create the root node */
+ error = nfs_nget(nmp->nm_mountp, NULL, NULL, dirfh.fh_data, dirfh.fh_len, &nvattr, &xid, rq.r_auth, NG_MARKROOT, npp);
+ nfsmout_if(error);
+
+ if (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_ACL)
+ vfs_setextendedsecurity(nmp->nm_mountp);
+
+ /* adjust I/O sizes to server limits */
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXREAD) && (nmp->nm_fsattr.nfsa_maxread > 0)) {
+ if (nmp->nm_fsattr.nfsa_maxread < (uint64_t)nmp->nm_rsize) {
+ nmp->nm_rsize = nmp->nm_fsattr.nfsa_maxread & ~(NFS_FABLKSIZE - 1);
+ if (nmp->nm_rsize == 0)
+ nmp->nm_rsize = nmp->nm_fsattr.nfsa_maxread;
+ }
+ }
+ if (NFS_BITMAP_ISSET(nmp->nm_fsattr.nfsa_bitmap, NFS_FATTR_MAXWRITE) && (nmp->nm_fsattr.nfsa_maxwrite > 0)) {
+ if (nmp->nm_fsattr.nfsa_maxwrite < (uint64_t)nmp->nm_wsize) {
+ nmp->nm_wsize = nmp->nm_fsattr.nfsa_maxwrite & ~(NFS_FABLKSIZE - 1);
+ if (nmp->nm_wsize == 0)
+ nmp->nm_wsize = nmp->nm_fsattr.nfsa_maxwrite;
+ }
+ }
+
+ /* set up lease renew timer */
+ nmp->nm_renew_timer = thread_call_allocate(nfs4_renew_timer, nmp);
+ interval = nmp->nm_fsattr.nfsa_lease / 2;
+ if (interval < 1)
+ interval = 1;
+ nfs_interval_timer_start(nmp->nm_renew_timer, interval * 1000);
+
+nfsmout:
+ if (fspath.np_components) {
+ for (comp=0; comp < fspath.np_compcount; comp++)
+ if (fspath.np_components[comp])
+ FREE(fspath.np_components[comp], M_TEMP);
+ FREE(fspath.np_components, M_TEMP);
+ }
+ NVATTR_CLEANUP(&nvattr);
+ nfs_fs_locations_cleanup(&nfsls);
+ if (*npp)
+ nfs_node_unlock(*npp);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/*
+ * Thread to handle initial NFS mount connection.
+ */
+void
+nfs_mount_connect_thread(void *arg, __unused wait_result_t wr)
+{
+ struct nfsmount *nmp = arg;
+ int error = 0, savederror = 0, slpflag = (NMFLAG(nmp, INTR) ? PCATCH : 0);
+ int done = 0, timeo, tries, maxtries;
+
+ if (NM_OMFLAG(nmp, MNTQUICK)) {
+ timeo = 8;
+ maxtries = 1;
+ } else {
+ timeo = 30;
+ maxtries = 2;
+ }
+
+ for (tries = 0; tries < maxtries; tries++) {
+ error = nfs_connect(nmp, 1, timeo);
+ switch (error) {
+ case ETIMEDOUT:
+ case EAGAIN:
+ case EPIPE:
+ case EADDRNOTAVAIL:
+ case ENETDOWN:
+ case ENETUNREACH:
+ case ENETRESET:
+ case ECONNABORTED:
+ case ECONNRESET:
+ case EISCONN:
+ case ENOTCONN:
+ case ESHUTDOWN:
+ case ECONNREFUSED:
+ case EHOSTDOWN:
+ case EHOSTUNREACH:
+ /* just keep retrying on any of these errors */
+ break;
+ case 0:
+ default:
+ /* looks like we got an answer... */
+ done = 1;
+ break;
+ }
+
+ /* save the best error */
+ if (nfs_connect_error_class(error) >= nfs_connect_error_class(savederror))
+ savederror = error;
+ if (done) {
+ error = savederror;
+ break;
+ }
+
+ /* pause before next attempt */
+ if ((error = nfs_sigintr(nmp, NULL, current_thread(), 0)))
+ break;
+ error = tsleep(nmp, PSOCK|slpflag, "nfs_mount_connect_retry", 2*hz);
+ if (error && (error != EWOULDBLOCK))
+ break;
+ error = savederror;
+ }
+
+ /* update status of mount connect */
+ lck_mtx_lock(&nmp->nm_lock);
+ if (!nmp->nm_mounterror)
+ nmp->nm_mounterror = error;
+ nmp->nm_state &= ~NFSSTA_MOUNT_THREAD;
+ lck_mtx_unlock(&nmp->nm_lock);
+ wakeup(&nmp->nm_nss);
+}
+
+int
+nfs_mount_connect(struct nfsmount *nmp)
+{
+ int error = 0, slpflag;
+ thread_t thd;
+ struct timespec ts = { 2, 0 };
+
+ /*
+ * Set up the socket. Perform initial search for a location/server/address to
+ * connect to and negotiate any unspecified mount parameters. This work is
+ * done on a kernel thread to satisfy reserved port usage needs.
+ */
+ slpflag = NMFLAG(nmp, INTR) ? PCATCH : 0;
+ lck_mtx_lock(&nmp->nm_lock);
+ /* set flag that the thread is running */
+ nmp->nm_state |= NFSSTA_MOUNT_THREAD;
+ if (kernel_thread_start(nfs_mount_connect_thread, nmp, &thd) != KERN_SUCCESS) {
+ nmp->nm_state &= ~NFSSTA_MOUNT_THREAD;
+ nmp->nm_mounterror = EIO;
+ printf("nfs mount %s start socket connect thread failed\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname);
+ } else {
+ thread_deallocate(thd);
+ }
+
+ /* wait until mount connect thread is finished/gone */
+ while (nmp->nm_state & NFSSTA_MOUNT_THREAD) {
+ error = msleep(&nmp->nm_nss, &nmp->nm_lock, slpflag|PSOCK, "nfsconnectthread", &ts);
+ if ((error && (error != EWOULDBLOCK)) || ((error = nfs_sigintr(nmp, NULL, current_thread(), 1)))) {
+ /* record error */
+ if (!nmp->nm_mounterror)
+ nmp->nm_mounterror = error;
+ /* signal the thread that we are aborting */
+ nmp->nm_sockflags |= NMSOCK_UNMOUNT;
+ if (nmp->nm_nss)
+ wakeup(nmp->nm_nss);
+ /* and continue waiting on it to finish */
+ slpflag = 0;
+ }
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+
+ /* grab mount connect status */
+ error = nmp->nm_mounterror;
+
+ return (error);
+}
+
+/* Table of maximum minor version for a given version */
+uint32_t maxminorverstab[] = {
+ 0, /* Version 0 (does not exist) */
+ 0, /* Version 1 (does not exist) */
+ 0, /* Version 2 */
+ 0, /* Version 3 */
+ 0, /* Version 4 */
+};
+
+#define NFS_MAX_SUPPORTED_VERSION ((long)(sizeof (maxminorverstab) / sizeof (uint32_t) - 1))
+#define NFS_MAX_SUPPORTED_MINOR_VERSION(v) ((long)(maxminorverstab[(v)]))
+
+#define DEFAULT_NFS_MIN_VERS VER2PVER(2, 0)
+#define DEFAULT_NFS_MAX_VERS VER2PVER(3, 0)
+
+/*
+ * Common code to mount an NFS file system.
+ */
+int
+mountnfs(
+ char *xdrbuf,
+ mount_t mp,
+ vfs_context_t ctx,
+ vnode_t *vpp)
+{
+ struct nfsmount *nmp;
+ nfsnode_t np;
+ int error = 0;
+ struct vfsstatfs *sbp;
+ struct xdrbuf xb;
+ uint32_t i, val, maxio, iosize, len;
+ uint32_t *mattrs;
+ uint32_t *mflags_mask;
+ uint32_t *mflags;
+ uint32_t argslength, attrslength;
+ struct nfs_location_index firstloc = { NLI_VALID, 0, 0, 0 };
+ static const struct nfs_etype nfs_default_etypes = {
+ .count = NFS_MAX_ETYPES,
+ .selected = NFS_MAX_ETYPES,
+ .etypes = { NFS_AES256_CTS_HMAC_SHA1_96,
+ NFS_AES128_CTS_HMAC_SHA1_96,
+ NFS_DES3_CBC_SHA1_KD
+ }
+ };
+ /* make sure mbuf constants are set up */
+ if (!nfs_mbuf_mhlen)
+ nfs_mbuf_init();
+
+ if (vfs_flags(mp) & MNT_UPDATE) {
+ nmp = VFSTONFS(mp);
+ /* update paths, file handles, etc, here XXX */
+ xb_free(xdrbuf);
+ return (0);
+ } else {
+ /* allocate an NFS mount structure for this mount */
+ MALLOC_ZONE(nmp, struct nfsmount *,
+ sizeof (struct nfsmount), M_NFSMNT, M_WAITOK);
+ if (!nmp) {
+ xb_free(xdrbuf);
+ return (ENOMEM);
+ }
+ bzero((caddr_t)nmp, sizeof (struct nfsmount));
+ lck_mtx_init(&nmp->nm_lock, nfs_mount_grp, LCK_ATTR_NULL);
+ TAILQ_INIT(&nmp->nm_resendq);
+ TAILQ_INIT(&nmp->nm_iodq);
+ TAILQ_INIT(&nmp->nm_gsscl);
+ LIST_INIT(&nmp->nm_monlist);
+ vfs_setfsprivate(mp, nmp);
+ vfs_getnewfsid(mp);
+ nmp->nm_mountp = mp;
+ vfs_setauthopaque(mp);
+ /*
+ * Disable cache_lookup_path for NFS. NFS lookup always needs
+ * to be called to check if the directory attribute cache is
+ * valid and possibly purge the directory before calling
+ * cache_lookup.
+ */
+ vfs_setauthcache_ttl(mp, 0);
+
+ nfs_nhinit_finish();
+
+ nmp->nm_args = xdrbuf;
+
+ /* set up defaults */
+ nmp->nm_ref = 0;
+ nmp->nm_vers = 0;
+ nmp->nm_min_vers = DEFAULT_NFS_MIN_VERS;
+ nmp->nm_max_vers = DEFAULT_NFS_MAX_VERS;
+ nmp->nm_timeo = NFS_TIMEO;
+ nmp->nm_retry = NFS_RETRANS;
+ nmp->nm_sotype = 0;
+ nmp->nm_sofamily = 0;
+ nmp->nm_nfsport = 0;
+ nmp->nm_wsize = NFS_WSIZE;
+ nmp->nm_rsize = NFS_RSIZE;
+ nmp->nm_readdirsize = NFS_READDIRSIZE;
+ nmp->nm_numgrps = NFS_MAXGRPS;
+ nmp->nm_readahead = NFS_DEFRAHEAD;
+ nmp->nm_tprintf_delay = nfs_tprintf_delay;
+ if (nmp->nm_tprintf_delay < 0)
+ nmp->nm_tprintf_delay = 0;
+ nmp->nm_tprintf_initial_delay = nfs_tprintf_initial_delay;
+ if (nmp->nm_tprintf_initial_delay < 0)
+ nmp->nm_tprintf_initial_delay = 0;
+ nmp->nm_acregmin = NFS_MINATTRTIMO;
+ nmp->nm_acregmax = NFS_MAXATTRTIMO;
+ nmp->nm_acdirmin = NFS_MINDIRATTRTIMO;
+ nmp->nm_acdirmax = NFS_MAXDIRATTRTIMO;
+ nmp->nm_etype = nfs_default_etypes;
+ nmp->nm_auth = RPCAUTH_SYS;
+ nmp->nm_iodlink.tqe_next = NFSNOLIST;
+ nmp->nm_deadtimeout = 0;
+ nmp->nm_curdeadtimeout = 0;
+ NFS_BITMAP_SET(nmp->nm_flags, NFS_MFLAG_NOACL);
+ nmp->nm_realm = NULL;
+ nmp->nm_principal = NULL;
+ nmp->nm_sprinc = NULL;
+ }
+
+ mattrs = nmp->nm_mattrs;
+ mflags = nmp->nm_mflags;
+ mflags_mask = nmp->nm_mflags_mask;
+
+ /* set up NFS mount with args */
+ xb_init_buffer(&xb, xdrbuf, 2*XDRWORD);
+ xb_get_32(error, &xb, val); /* version */
+ xb_get_32(error, &xb, argslength); /* args length */
+ nfsmerr_if(error);
+ xb_init_buffer(&xb, xdrbuf, argslength); /* restart parsing with actual buffer length */
+ xb_get_32(error, &xb, val); /* version */
+ xb_get_32(error, &xb, argslength); /* args length */
+ xb_get_32(error, &xb, val); /* XDR args version */
+ if (val != NFS_XDRARGS_VERSION_0)
+ error = EINVAL;
+ len = NFS_MATTR_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, mattrs, len); /* mount attribute bitmap */
+ attrslength = 0;
+ xb_get_32(error, &xb, attrslength); /* attrs length */
+ if (!error && (attrslength > (argslength - ((4+NFS_MATTR_BITMAP_LEN+1)*XDRWORD))))
+ error = EINVAL;
+ nfsmerr_if(error);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FLAGS)) {
+ len = NFS_MFLAG_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, mflags_mask, len); /* mount flag mask */
+ len = NFS_MFLAG_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, mflags, len); /* mount flag values */
+ if (!error) {
+ /* clear all mask bits and OR in all the ones that are set */
+ nmp->nm_flags[0] &= ~mflags_mask[0];
+ nmp->nm_flags[0] |= (mflags_mask[0] & mflags[0]);
+ }
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION)) {
+ /* Can't specify a single version and a range */
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION_RANGE))
+ error = EINVAL;
+ xb_get_32(error, &xb, nmp->nm_vers);
+ if (nmp->nm_vers > NFS_MAX_SUPPORTED_VERSION ||
+ nmp->nm_vers < NFS_VER2)
+ error = EINVAL;
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_MINOR_VERSION))
+ xb_get_32(error, &xb, nmp->nm_minor_vers);
+ else
+ nmp->nm_minor_vers = maxminorverstab[nmp->nm_vers];
+ if (nmp->nm_minor_vers > maxminorverstab[nmp->nm_vers])
+ error = EINVAL;
+ nmp->nm_max_vers = nmp->nm_min_vers =
+ VER2PVER(nmp->nm_vers, nmp->nm_minor_vers);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_MINOR_VERSION)) {
+ /* should have also gotten NFS version (and already gotten minor version) */
+ if (!NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION))
+ error = EINVAL;
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION_RANGE)) {
+ xb_get_32(error, &xb, nmp->nm_min_vers);
+ xb_get_32(error, &xb, nmp->nm_max_vers);
+ if ((nmp->nm_min_vers > nmp->nm_max_vers) ||
+ (PVER2MAJOR(nmp->nm_max_vers) > NFS_MAX_SUPPORTED_VERSION) ||
+ (PVER2MINOR(nmp->nm_min_vers) > maxminorverstab[PVER2MAJOR(nmp->nm_min_vers)]) ||
+ (PVER2MINOR(nmp->nm_max_vers) > maxminorverstab[PVER2MAJOR(nmp->nm_max_vers)]))
+ error = EINVAL;
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READ_SIZE))
+ xb_get_32(error, &xb, nmp->nm_rsize);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_WRITE_SIZE))
+ xb_get_32(error, &xb, nmp->nm_wsize);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READDIR_SIZE))
+ xb_get_32(error, &xb, nmp->nm_readdirsize);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READAHEAD))
+ xb_get_32(error, &xb, nmp->nm_readahead);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_REG_MIN)) {
+ xb_get_32(error, &xb, nmp->nm_acregmin);
+ xb_skip(error, &xb, XDRWORD);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_REG_MAX)) {
+ xb_get_32(error, &xb, nmp->nm_acregmax);
+ xb_skip(error, &xb, XDRWORD);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_DIR_MIN)) {
+ xb_get_32(error, &xb, nmp->nm_acdirmin);
+ xb_skip(error, &xb, XDRWORD);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_DIR_MAX)) {
+ xb_get_32(error, &xb, nmp->nm_acdirmax);
+ xb_skip(error, &xb, XDRWORD);
+ }
+ nfsmerr_if(error);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_LOCK_MODE)) {
+ xb_get_32(error, &xb, val);
+ switch (val) {
+ case NFS_LOCK_MODE_DISABLED:
+ case NFS_LOCK_MODE_LOCAL:
+ if (nmp->nm_vers >= NFS_VER4) {
+ /* disabled/local lock mode only allowed on v2/v3 */
+ error = EINVAL;
+ break;
+ }
+ /* FALLTHROUGH */
+ case NFS_LOCK_MODE_ENABLED:
+ nmp->nm_lockmode = val;
+ break;
+ default:
+ error = EINVAL;
+ }
+ }
+ nfsmerr_if(error);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SECURITY)) {
+ uint32_t seccnt;
+ xb_get_32(error, &xb, seccnt);
+ if (!error && ((seccnt < 1) || (seccnt > NX_MAX_SEC_FLAVORS)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ nmp->nm_sec.count = seccnt;
+ for (i=0; i < seccnt; i++) {
+ xb_get_32(error, &xb, nmp->nm_sec.flavors[i]);
+ /* Check for valid security flavor */
+ switch (nmp->nm_sec.flavors[i]) {
+ case RPCAUTH_NONE:
+ case RPCAUTH_SYS:
+ case RPCAUTH_KRB5:
+ case RPCAUTH_KRB5I:
+ case RPCAUTH_KRB5P:
+ break;
+ default:
+ error = EINVAL;
+ }
+ }
+ /* start with the first flavor */
+ nmp->nm_auth = nmp->nm_sec.flavors[0];
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_KERB_ETYPE)) {
+ uint32_t etypecnt;
+ xb_get_32(error, &xb, etypecnt);
+ if (!error && ((etypecnt < 1) || (etypecnt > NFS_MAX_ETYPES)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ nmp->nm_etype.count = etypecnt;
+ xb_get_32(error, &xb, nmp->nm_etype.selected);
+ nfsmerr_if(error);
+ if (etypecnt) {
+ nmp->nm_etype.selected = etypecnt; /* Nothing is selected yet, so set selected to count */
+ for (i=0; i < etypecnt; i++) {
+ xb_get_32(error, &xb, nmp->nm_etype.etypes[i]);
+ /* Check for valid encryption type */
+ switch (nmp->nm_etype.etypes[i]) {
+ case NFS_DES3_CBC_SHA1_KD:
+ case NFS_AES128_CTS_HMAC_SHA1_96:
+ case NFS_AES256_CTS_HMAC_SHA1_96:
+ break;
+ default:
+ error = EINVAL;
+ }
+ }
+ }
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MAX_GROUP_LIST))
+ xb_get_32(error, &xb, nmp->nm_numgrps);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SOCKET_TYPE)) {
+ char sotype[6];
+
+ xb_get_32(error, &xb, val);
+ if (!error && ((val < 3) || (val > 5)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, sotype, val, 0);
+ nfsmerr_if(error);
+ sotype[val] = '\0';
+ if (!strcmp(sotype, "tcp")) {
+ nmp->nm_sotype = SOCK_STREAM;
+ } else if (!strcmp(sotype, "udp")) {
+ nmp->nm_sotype = SOCK_DGRAM;
+ } else if (!strcmp(sotype, "tcp4")) {
+ nmp->nm_sotype = SOCK_STREAM;
+ nmp->nm_sofamily = AF_INET;
+ } else if (!strcmp(sotype, "udp4")) {
+ nmp->nm_sotype = SOCK_DGRAM;
+ nmp->nm_sofamily = AF_INET;
+ } else if (!strcmp(sotype, "tcp6")) {
+ nmp->nm_sotype = SOCK_STREAM;
+ nmp->nm_sofamily = AF_INET6;
+ } else if (!strcmp(sotype, "udp6")) {
+ nmp->nm_sotype = SOCK_DGRAM;
+ nmp->nm_sofamily = AF_INET6;
+ } else if (!strcmp(sotype, "inet4")) {
+ nmp->nm_sofamily = AF_INET;
+ } else if (!strcmp(sotype, "inet6")) {
+ nmp->nm_sofamily = AF_INET6;
+ } else if (!strcmp(sotype, "inet")) {
+ nmp->nm_sofamily = 0; /* ok */
+ } else {
+ error = EINVAL;
+ }
+ if (!error && (nmp->nm_vers >= NFS_VER4) && nmp->nm_sotype &&
+ (nmp->nm_sotype != SOCK_STREAM))
+ error = EINVAL; /* NFSv4 is only allowed over TCP. */
+ nfsmerr_if(error);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_PORT))
+ xb_get_32(error, &xb, nmp->nm_nfsport);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MOUNT_PORT))
+ xb_get_32(error, &xb, nmp->nm_mountport);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_REQUEST_TIMEOUT)) {
+ /* convert from time to 0.1s units */
+ xb_get_32(error, &xb, nmp->nm_timeo);
+ xb_get_32(error, &xb, val);
+ nfsmerr_if(error);
+ if (val >= 1000000000)
+ error = EINVAL;
+ nfsmerr_if(error);
+ nmp->nm_timeo *= 10;
+ nmp->nm_timeo += (val+100000000-1)/100000000;
+ /* now convert to ticks */
+ nmp->nm_timeo = (nmp->nm_timeo * NFS_HZ + 5) / 10;
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SOFT_RETRY_COUNT)) {
+ xb_get_32(error, &xb, val);
+ if (!error && (val > 1))
+ nmp->nm_retry = val;
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_DEAD_TIMEOUT)) {
+ xb_get_32(error, &xb, nmp->nm_deadtimeout);
+ xb_skip(error, &xb, XDRWORD);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FH)) {
+ nfsmerr_if(error);
+ MALLOC(nmp->nm_fh, fhandle_t *, sizeof(fhandle_t), M_TEMP, M_WAITOK|M_ZERO);
+ if (!nmp->nm_fh)
+ error = ENOMEM;
+ xb_get_32(error, &xb, nmp->nm_fh->fh_len);
+ nfsmerr_if(error);
+ if (nmp->nm_fh->fh_len < 0 ||
+ (size_t)nmp->nm_fh->fh_len > sizeof(nmp->nm_fh->fh_data))
+ error = EINVAL;
+ else
+ error = xb_get_bytes(&xb, (char*)&nmp->nm_fh->fh_data[0], nmp->nm_fh->fh_len, 0);
+ }
+ nfsmerr_if(error);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FS_LOCATIONS)) {
+ uint32_t loc, serv, addr, comp;
+ struct nfs_fs_location *fsl;
+ struct nfs_fs_server *fss;
+ struct nfs_fs_path *fsp;
+
+ xb_get_32(error, &xb, nmp->nm_locations.nl_numlocs); /* fs location count */
+ /* sanity check location count */
+ if (!error && ((nmp->nm_locations.nl_numlocs < 1) || (nmp->nm_locations.nl_numlocs > 256)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ MALLOC(nmp->nm_locations.nl_locations, struct nfs_fs_location **, nmp->nm_locations.nl_numlocs * sizeof(struct nfs_fs_location*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!nmp->nm_locations.nl_locations)
+ error = ENOMEM;
+ for (loc = 0; loc < nmp->nm_locations.nl_numlocs; loc++) {
+ nfsmerr_if(error);
+ MALLOC(fsl, struct nfs_fs_location *, sizeof(struct nfs_fs_location), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fsl)
+ error = ENOMEM;
+ nmp->nm_locations.nl_locations[loc] = fsl;
+ xb_get_32(error, &xb, fsl->nl_servcount); /* server count */
+ /* sanity check server count */
+ if (!error && ((fsl->nl_servcount < 1) || (fsl->nl_servcount > 256)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ MALLOC(fsl->nl_servers, struct nfs_fs_server **, fsl->nl_servcount * sizeof(struct nfs_fs_server*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fsl->nl_servers)
+ error = ENOMEM;
+ for (serv = 0; serv < fsl->nl_servcount; serv++) {
+ nfsmerr_if(error);
+ MALLOC(fss, struct nfs_fs_server *, sizeof(struct nfs_fs_server), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fss)
+ error = ENOMEM;
+ fsl->nl_servers[serv] = fss;
+ xb_get_32(error, &xb, val); /* server name length */
+ /* sanity check server name length */
+ if (!error && ((val < 1) || (val > MAXPATHLEN)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ MALLOC(fss->ns_name, char *, val+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!fss->ns_name)
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, fss->ns_name, val, 0); /* server name */
+ xb_get_32(error, &xb, fss->ns_addrcount); /* address count */
+ /* sanity check address count (OK to be zero) */
+ if (!error && (fss->ns_addrcount > 256))
+ error = EINVAL;
+ nfsmerr_if(error);
+ if (fss->ns_addrcount > 0) {
+ MALLOC(fss->ns_addresses, char **, fss->ns_addrcount * sizeof(char *), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fss->ns_addresses)
+ error = ENOMEM;
+ for (addr = 0; addr < fss->ns_addrcount; addr++) {
+ xb_get_32(error, &xb, val); /* address length */
+ /* sanity check address length */
+ if (!error && ((val < 1) || (val > 128)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ MALLOC(fss->ns_addresses[addr], char *, val+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!fss->ns_addresses[addr])
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, fss->ns_addresses[addr], val, 0); /* address */
+ }
+ }
+ xb_get_32(error, &xb, val); /* server info length */
+ xb_skip(error, &xb, val); /* skip server info */
+ }
+ /* get pathname */
+ fsp = &fsl->nl_path;
+ xb_get_32(error, &xb, fsp->np_compcount); /* component count */
+ /* sanity check component count */
+ if (!error && (fsp->np_compcount > MAXPATHLEN))
+ error = EINVAL;
+ nfsmerr_if(error);
+ if (fsp->np_compcount) {
+ MALLOC(fsp->np_components, char **, fsp->np_compcount * sizeof(char*), M_TEMP, M_WAITOK|M_ZERO);
+ if (!fsp->np_components)
+ error = ENOMEM;
+ }
+ for (comp = 0; comp < fsp->np_compcount; comp++) {
+ xb_get_32(error, &xb, val); /* component length */
+ /* sanity check component length */
+ if (!error && (val == 0)) {
+ /*
+ * Apparently some people think a path with zero components should
+ * be encoded with one zero-length component. So, just ignore any
+ * zero length components.
+ */
+ comp--;
+ fsp->np_compcount--;
+ if (fsp->np_compcount == 0) {
+ FREE(fsp->np_components, M_TEMP);
+ fsp->np_components = NULL;
+ }
+ continue;
+ }
+ if (!error && ((val < 1) || (val > MAXPATHLEN)))
+ error = EINVAL;
+ nfsmerr_if(error);
+ MALLOC(fsp->np_components[comp], char *, val+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!fsp->np_components[comp])
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, fsp->np_components[comp], val, 0); /* component */
+ }
+ xb_get_32(error, &xb, val); /* fs location info length */
+ xb_skip(error, &xb, val); /* skip fs location info */
+ }
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MNTFLAGS))
+ xb_skip(error, &xb, XDRWORD);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MNTFROM)) {
+ xb_get_32(error, &xb, len);
+ nfsmerr_if(error);
+ val = len;
+ if (val >= sizeof(vfs_statfs(mp)->f_mntfromname))
+ val = sizeof(vfs_statfs(mp)->f_mntfromname) - 1;
+ error = xb_get_bytes(&xb, vfs_statfs(mp)->f_mntfromname, val, 0);
+ if ((len - val) > 0)
+ xb_skip(error, &xb, len - val);
+ nfsmerr_if(error);
+ vfs_statfs(mp)->f_mntfromname[val] = '\0';
+ }
+ nfsmerr_if(error);
+
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_REALM)) {
+ xb_get_32(error, &xb, len);
+ if (!error && ((len < 1) || (len > MAXPATHLEN)))
+ error=EINVAL;
+ nfsmerr_if(error);
+ /* allocate an extra byte for a leading '@' if its not already prepended to the realm */
+ MALLOC(nmp->nm_realm, char *, len+2, M_TEMP, M_WAITOK|M_ZERO);
+ if (!nmp->nm_realm)
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, nmp->nm_realm, len, 0);
+ if (error == 0 && *nmp->nm_realm != '@') {
+ bcopy(nmp->nm_realm, &nmp->nm_realm[1], len);
+ nmp->nm_realm[0] = '@';
+ }
+ }
+ nfsmerr_if(error);
+
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_PRINCIPAL)) {
+ xb_get_32(error, &xb, len);
+ if (!error && ((len < 1) || (len > MAXPATHLEN)))
+ error=EINVAL;
+ nfsmerr_if(error);
+ MALLOC(nmp->nm_principal, char *, len+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!nmp->nm_principal)
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, nmp->nm_principal, len, 0);
+ }
+ nfsmerr_if(error);
+
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SVCPRINCIPAL)) {
+ xb_get_32(error, &xb, len);
+ if (!error && ((len < 1) || (len > MAXPATHLEN)))
+ error=EINVAL;
+ nfsmerr_if(error);
+ MALLOC(nmp->nm_sprinc, char *, len+1, M_TEMP, M_WAITOK|M_ZERO);
+ if (!nmp->nm_sprinc)
+ error = ENOMEM;
+ nfsmerr_if(error);
+ error = xb_get_bytes(&xb, nmp->nm_sprinc, len, 0);
+ }
+ nfsmerr_if(error);
+
+ /*
+ * Sanity check/finalize settings.
+ */
+
+ if (nmp->nm_timeo < NFS_MINTIMEO)
+ nmp->nm_timeo = NFS_MINTIMEO;
+ else if (nmp->nm_timeo > NFS_MAXTIMEO)
+ nmp->nm_timeo = NFS_MAXTIMEO;
+ if (nmp->nm_retry > NFS_MAXREXMIT)
+ nmp->nm_retry = NFS_MAXREXMIT;
+
+ if (nmp->nm_numgrps > NFS_MAXGRPS)
+ nmp->nm_numgrps = NFS_MAXGRPS;
+ if (nmp->nm_readahead > NFS_MAXRAHEAD)
+ nmp->nm_readahead = NFS_MAXRAHEAD;
+ if (nmp->nm_acregmin > nmp->nm_acregmax)
+ nmp->nm_acregmin = nmp->nm_acregmax;
+ if (nmp->nm_acdirmin > nmp->nm_acdirmax)
+ nmp->nm_acdirmin = nmp->nm_acdirmax;
+
+ /* need at least one fs location */
+ if (nmp->nm_locations.nl_numlocs < 1)
+ error = EINVAL;
+ nfsmerr_if(error);
+
+ /* init mount's mntfromname to first location */
+ if (!NM_OMATTR_GIVEN(nmp, MNTFROM))
+ nfs_location_mntfromname(&nmp->nm_locations, firstloc,
+ vfs_statfs(mp)->f_mntfromname, sizeof(vfs_statfs(mp)->f_mntfromname), 0);
+
+ /* Need to save the mounting credential for v4. */
+ nmp->nm_mcred = vfs_context_ucred(ctx);
+ if (IS_VALID_CRED(nmp->nm_mcred))
+ kauth_cred_ref(nmp->nm_mcred);
+
+ /*
+ * If a reserved port is required, check for that privilege.
+ * (Note that mirror mounts are exempt because the privilege was
+ * already checked for the original mount.)
+ */
+ if (NMFLAG(nmp, RESVPORT) && !vfs_iskernelmount(mp))
+ error = priv_check_cred(nmp->nm_mcred, PRIV_NETINET_RESERVEDPORT, 0);
+ nfsmerr_if(error);
+
+ /* do mount's initial socket connection */
+ error = nfs_mount_connect(nmp);
+ nfsmerr_if(error);
+
+ /* set up the version-specific function tables */
+ if (nmp->nm_vers < NFS_VER4)
+ nmp->nm_funcs = &nfs3_funcs;
+ else
+ nmp->nm_funcs = &nfs4_funcs;
+
+ /* sanity check settings now that version/connection is set */
+ if (nmp->nm_vers == NFS_VER2) /* ignore RDIRPLUS on NFSv2 */
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_RDIRPLUS);
+ if (nmp->nm_vers >= NFS_VER4) {
+ if (NFS_BITMAP_ISSET(nmp->nm_flags, NFS_MFLAG_ACLONLY)) /* aclonly trumps noacl */
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_NOACL);
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_CALLUMNT);
+ if (nmp->nm_lockmode != NFS_LOCK_MODE_ENABLED)
+ error = EINVAL; /* disabled/local lock mode only allowed on v2/v3 */
+ } else {
+ /* ignore these if not v4 */
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_NOCALLBACK);
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_NAMEDATTR);
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_NOACL);
+ NFS_BITMAP_CLR(nmp->nm_flags, NFS_MFLAG_ACLONLY);
+ }
+ nfsmerr_if(error);
+
+ if (nmp->nm_sotype == SOCK_DGRAM) {
+ /* I/O size defaults for UDP are different */
+ if (!NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READ_SIZE))
+ nmp->nm_rsize = NFS_DGRAM_RSIZE;
+ if (!NFS_BITMAP_ISSET(mattrs, NFS_MATTR_WRITE_SIZE))
+ nmp->nm_wsize = NFS_DGRAM_WSIZE;
+ }
+
+ /* round down I/O sizes to multiple of NFS_FABLKSIZE */
+ nmp->nm_rsize &= ~(NFS_FABLKSIZE - 1);
+ if (nmp->nm_rsize <= 0)
+ nmp->nm_rsize = NFS_FABLKSIZE;
+ nmp->nm_wsize &= ~(NFS_FABLKSIZE - 1);
+ if (nmp->nm_wsize <= 0)
+ nmp->nm_wsize = NFS_FABLKSIZE;
+
+ /* and limit I/O sizes to maximum allowed */
+ maxio = (nmp->nm_vers == NFS_VER2) ? NFS_V2MAXDATA :
+ (nmp->nm_sotype == SOCK_DGRAM) ? NFS_MAXDGRAMDATA : NFS_MAXDATA;
+ if (maxio > NFS_MAXBSIZE)
+ maxio = NFS_MAXBSIZE;
+ if (nmp->nm_rsize > maxio)
+ nmp->nm_rsize = maxio;
+ if (nmp->nm_wsize > maxio)
+ nmp->nm_wsize = maxio;
+
+ if (nmp->nm_readdirsize > maxio)
+ nmp->nm_readdirsize = maxio;
+ if (nmp->nm_readdirsize > nmp->nm_rsize)
+ nmp->nm_readdirsize = nmp->nm_rsize;
+
+ /* Set up the sockets and related info */
+ if (nmp->nm_sotype == SOCK_DGRAM)
+ TAILQ_INIT(&nmp->nm_cwndq);
+
+ /*
+ * Get the root node/attributes from the NFS server and
+ * do any basic, version-specific setup.
+ */
+ error = nmp->nm_funcs->nf_mount(nmp, ctx, &np);
+ nfsmerr_if(error);
+
+ /*
+ * A reference count is needed on the node representing the
+ * remote root. If this object is not persistent, then backward
+ * traversals of the mount point (i.e. "..") will not work if
+ * the node gets flushed out of the cache.
+ */
+ nmp->nm_dnp = np;
+ *vpp = NFSTOV(np);
+ /* get usecount and drop iocount */
+ error = vnode_ref(*vpp);
+ vnode_put(*vpp);
+ if (error) {
+ vnode_recycle(*vpp);
+ goto nfsmerr;
+ }
+
+ /*
+ * Do statfs to ensure static info gets set to reasonable values.
+ */
+ if ((error = nmp->nm_funcs->nf_update_statfs(nmp, ctx))) {
+ int error2 = vnode_getwithref(*vpp);
+ vnode_rele(*vpp);
+ if (!error2)
+ vnode_put(*vpp);
+ vnode_recycle(*vpp);
+ goto nfsmerr;
+ }
+ sbp = vfs_statfs(mp);
+ sbp->f_bsize = nmp->nm_fsattr.nfsa_bsize;
+ sbp->f_blocks = nmp->nm_fsattr.nfsa_space_total / sbp->f_bsize;
+ sbp->f_bfree = nmp->nm_fsattr.nfsa_space_free / sbp->f_bsize;
+ sbp->f_bavail = nmp->nm_fsattr.nfsa_space_avail / sbp->f_bsize;
+ sbp->f_bused = (nmp->nm_fsattr.nfsa_space_total / sbp->f_bsize) -
+ (nmp->nm_fsattr.nfsa_space_free / sbp->f_bsize);
+ sbp->f_files = nmp->nm_fsattr.nfsa_files_total;
+ sbp->f_ffree = nmp->nm_fsattr.nfsa_files_free;
+ sbp->f_iosize = nfs_iosize;
+
+ /*
+ * Calculate the size used for I/O buffers. Use the larger
+ * of the two sizes to minimise NFS requests but make sure
+ * that it is at least one VM page to avoid wasting buffer
+ * space and to allow easy mmapping of I/O buffers.
+ * The read/write RPC calls handle the splitting up of
+ * buffers into multiple requests if the buffer size is
+ * larger than the I/O size.
+ */
+#ifndef CONFIG_EMBEDDED
+ iosize = max(nmp->nm_rsize, nmp->nm_wsize);
+ if (iosize < PAGE_SIZE)
+ iosize = PAGE_SIZE;
+#else
+ iosize = PAGE_SIZE;
+#endif
+ nmp->nm_biosize = trunc_page_32(iosize);
+
+ /* For NFSv3 and greater, there is a (relatively) reliable ACCESS call. */
+ if (nmp->nm_vers > NFS_VER2)
+ vfs_setauthopaqueaccess(mp);
+
+ switch (nmp->nm_lockmode) {
+ case NFS_LOCK_MODE_DISABLED:
+ break;
+ case NFS_LOCK_MODE_LOCAL:
+ vfs_setlocklocal(nmp->nm_mountp);
+ break;
+ case NFS_LOCK_MODE_ENABLED:
+ default:
+ if (nmp->nm_vers <= NFS_VER3)
+ nfs_lockd_mount_register(nmp);
+ break;
+ }
+
+ /* success! */
+ lck_mtx_lock(&nmp->nm_lock);
+ nmp->nm_state |= NFSSTA_MOUNTED;
+ lck_mtx_unlock(&nmp->nm_lock);
+ return (0);
+nfsmerr:
+ nfs_mount_drain_and_cleanup(nmp);
+ return (error);
+}
+
+#if CONFIG_TRIGGERS
+
+/*
+ * We've detected a file system boundary on the server and
+ * need to mount a new file system so that our file systems
+ * MIRROR the file systems on the server.
+ *
+ * Build the mount arguments for the new mount and call kernel_mount().
+ */
+int
+nfs_mirror_mount_domount(vnode_t dvp, vnode_t vp, vfs_context_t ctx)
+{
+ nfsnode_t np = VTONFS(vp);
+ nfsnode_t dnp = VTONFS(dvp);
+ struct nfsmount *nmp = NFSTONMP(np);
+ char fstype[MFSTYPENAMELEN], *mntfromname = NULL, *path = NULL, *relpath, *p, *cp;
+ int error = 0, pathbuflen = MAXPATHLEN, i, mntflags = 0, referral, skipcopy = 0;
+ size_t nlen;
+ struct xdrbuf xb, xbnew;
+ uint32_t mattrs[NFS_MATTR_BITMAP_LEN];
+ uint32_t newmattrs[NFS_MATTR_BITMAP_LEN];
+ uint32_t newmflags[NFS_MFLAG_BITMAP_LEN];
+ uint32_t newmflags_mask[NFS_MFLAG_BITMAP_LEN];
+ uint32_t argslength = 0, val, count, mlen, mlen2, rlen, relpathcomps;
+ uint32_t argslength_offset, attrslength_offset, end_offset;
+ uint32_t numlocs, loc, numserv, serv, numaddr, addr, numcomp, comp;
+ char buf[XDRWORD];
+ struct nfs_fs_locations nfsls;
+
+ referral = (np->n_vattr.nva_flags & NFS_FFLAG_TRIGGER_REFERRAL);
+ if (referral)
+ bzero(&nfsls, sizeof(nfsls));
+
+ xb_init(&xbnew, 0);
+
+ if (!nmp || (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)))
+ return (ENXIO);
+
+ /* allocate a couple path buffers we need */
+ MALLOC_ZONE(mntfromname, char *, pathbuflen, M_NAMEI, M_WAITOK);
+ if (!mntfromname) {
+ error = ENOMEM;
+ goto nfsmerr;
+ }
+ MALLOC_ZONE(path, char *, pathbuflen, M_NAMEI, M_WAITOK);
+ if (!path) {
+ error = ENOMEM;
+ goto nfsmerr;
+ }
+
+ /* get the path for the directory being mounted on */
+ error = vn_getpath(vp, path, &pathbuflen);
+ if (error) {
+ error = ENOMEM;
+ goto nfsmerr;
+ }
+
+ /*
+ * Set up the mntfromname for the new mount based on the
+ * current mount's mntfromname and the directory's path
+ * relative to the current mount's mntonname.
+ * Set up relpath to point at the relative path on the current mount.
+ * Also, count the number of components in relpath.
+ * We'll be adding those to each fs location path in the new args.
+ */
+ nlen = strlcpy(mntfromname, vfs_statfs(nmp->nm_mountp)->f_mntfromname, MAXPATHLEN);
+ if ((nlen > 0) && (mntfromname[nlen-1] == '/')) { /* avoid double '/' in new name */
+ mntfromname[nlen-1] = '\0';
+ nlen--;
+ }
+ relpath = mntfromname + nlen;
+ nlen = strlcat(mntfromname, path + strlen(vfs_statfs(nmp->nm_mountp)->f_mntonname), MAXPATHLEN);
+ if (nlen >= MAXPATHLEN) {
+ error = ENAMETOOLONG;
+ goto nfsmerr;
+ }
+ /* count the number of components in relpath */
+ p = relpath;
+ while (*p && (*p == '/'))
+ p++;
+ relpathcomps = 0;
+ while (*p) {
+ relpathcomps++;
+ while (*p && (*p != '/'))
+ p++;
+ while (*p && (*p == '/'))
+ p++;
+ }
+
+ /* grab a copy of the file system type */
+ vfs_name(vnode_mount(vp), fstype);
+
+ /* for referrals, fetch the fs locations */
+ if (referral) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ if (!vname) {
+ error = ENOENT;
+ } else {
+ error = nfs4_get_fs_locations(nmp, dnp, NULL, 0, vname, ctx, &nfsls);
+ vnode_putname(vname);
+ if (!error && (nfsls.nl_numlocs < 1))
+ error = ENOENT;
+ }
+ nfsmerr_if(error);
+ }
+
+ /* set up NFS mount args based on current mount args */
+
+#define xb_copy_32(E, XBSRC, XBDST, V) \
+ do { \
+ if (E) break; \
+ xb_get_32((E), (XBSRC), (V)); \
+ if (skipcopy) break; \
+ xb_add_32((E), (XBDST), (V)); \
+ } while (0)
+#define xb_copy_opaque(E, XBSRC, XBDST) \
+ do { \
+ uint32_t __count, __val; \
+ xb_copy_32((E), (XBSRC), (XBDST), __count); \
+ if (E) break; \
+ __count = nfsm_rndup(__count); \
+ __count /= XDRWORD; \
+ while (__count-- > 0) \
+ xb_copy_32((E), (XBSRC), (XBDST), __val); \
+ } while (0)
+
+ xb_init_buffer(&xb, nmp->nm_args, 2*XDRWORD);
+ xb_get_32(error, &xb, val); /* version */
+ xb_get_32(error, &xb, argslength); /* args length */
+ xb_init_buffer(&xb, nmp->nm_args, argslength);
+
+ xb_init_buffer(&xbnew, NULL, 0);
+ xb_copy_32(error, &xb, &xbnew, val); /* version */
+ argslength_offset = xb_offset(&xbnew);
+ xb_copy_32(error, &xb, &xbnew, val); /* args length */
+ xb_copy_32(error, &xb, &xbnew, val); /* XDR args version */
+ count = NFS_MATTR_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, mattrs, count); /* mount attribute bitmap */
+ nfsmerr_if(error);
+ for (i = 0; i < NFS_MATTR_BITMAP_LEN; i++)
+ newmattrs[i] = mattrs[i];
+ if (referral)
+ NFS_BITMAP_SET(newmattrs, NFS_MATTR_FS_LOCATIONS);
+ else
+ NFS_BITMAP_SET(newmattrs, NFS_MATTR_FH);
+ NFS_BITMAP_SET(newmattrs, NFS_MATTR_FLAGS);
+ NFS_BITMAP_SET(newmattrs, NFS_MATTR_MNTFLAGS);
+ NFS_BITMAP_CLR(newmattrs, NFS_MATTR_MNTFROM);
+ xb_add_bitmap(error, &xbnew, newmattrs, NFS_MATTR_BITMAP_LEN);
+ attrslength_offset = xb_offset(&xbnew);
+ xb_copy_32(error, &xb, &xbnew, val); /* attrs length */
+ NFS_BITMAP_ZERO(newmflags_mask, NFS_MFLAG_BITMAP_LEN);
+ NFS_BITMAP_ZERO(newmflags, NFS_MFLAG_BITMAP_LEN);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FLAGS)) {
+ count = NFS_MFLAG_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, newmflags_mask, count); /* mount flag mask bitmap */
+ count = NFS_MFLAG_BITMAP_LEN;
+ xb_get_bitmap(error, &xb, newmflags, count); /* mount flag bitmap */
+ }
+ NFS_BITMAP_SET(newmflags_mask, NFS_MFLAG_EPHEMERAL);
+ NFS_BITMAP_SET(newmflags, NFS_MFLAG_EPHEMERAL);
+ xb_add_bitmap(error, &xbnew, newmflags_mask, NFS_MFLAG_BITMAP_LEN);
+ xb_add_bitmap(error, &xbnew, newmflags, NFS_MFLAG_BITMAP_LEN);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_MINOR_VERSION))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_VERSION_RANGE)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READ_SIZE))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_WRITE_SIZE))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READDIR_SIZE))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_READAHEAD))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_REG_MIN)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_REG_MAX)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_DIR_MIN)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_ATTRCACHE_DIR_MAX)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_LOCK_MODE))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SECURITY)) {
+ xb_copy_32(error, &xb, &xbnew, count);
+ while (!error && (count-- > 0))
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_KERB_ETYPE)) {
+ xb_copy_32(error, &xb, &xbnew, count);
+ xb_add_32(error, &xbnew, -1);
+ while (!error && (count-- > 0))
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MAX_GROUP_LIST))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SOCKET_TYPE))
+ xb_copy_opaque(error, &xb, &xbnew);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_NFS_PORT))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MOUNT_PORT))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_REQUEST_TIMEOUT)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_SOFT_RETRY_COUNT))
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_DEAD_TIMEOUT)) {
+ xb_copy_32(error, &xb, &xbnew, val);
+ xb_copy_32(error, &xb, &xbnew, val);
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FH)) {
+ xb_get_32(error, &xb, count);
+ xb_skip(error, &xb, count);
+ }
+ if (!referral) {
+ /* set the initial file handle to the directory's file handle */
+ xb_add_fh(error, &xbnew, np->n_fhp, np->n_fhsize);
+ }
+ /* copy/extend/skip fs locations */
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_FS_LOCATIONS)) {
+ numlocs = numserv = numaddr = numcomp = 0;
+ if (referral) /* don't copy the fs locations for a referral */
+ skipcopy = 1;
+ xb_copy_32(error, &xb, &xbnew, numlocs); /* location count */
+ for (loc = 0; !error && (loc < numlocs); loc++) {
+ xb_copy_32(error, &xb, &xbnew, numserv); /* server count */
+ for (serv = 0; !error && (serv < numserv); serv++) {
+ xb_copy_opaque(error, &xb, &xbnew); /* server name */
+ xb_copy_32(error, &xb, &xbnew, numaddr); /* address count */
+ for (addr = 0; !error && (addr < numaddr); addr++)
+ xb_copy_opaque(error, &xb, &xbnew); /* address */
+ xb_copy_opaque(error, &xb, &xbnew); /* server info */
+ }
+ /* pathname */
+ xb_get_32(error, &xb, numcomp); /* component count */
+ if (!skipcopy)
+ xb_add_32(error, &xbnew, numcomp+relpathcomps); /* new component count */
+ for (comp = 0; !error && (comp < numcomp); comp++)
+ xb_copy_opaque(error, &xb, &xbnew); /* component */
+ /* add additional components */
+ for (comp = 0; !skipcopy && !error && (comp < relpathcomps); comp++) {
+ p = relpath;
+ while (*p && (*p == '/'))
+ p++;
+ while (*p && !error) {
+ cp = p;
+ while (*p && (*p != '/'))
+ p++;
+ xb_add_string(error, &xbnew, cp, (p - cp)); /* component */
+ while (*p && (*p == '/'))
+ p++;
+ }
+ }
+ xb_copy_opaque(error, &xb, &xbnew); /* fs location info */
+ }
+ if (referral)
+ skipcopy = 0;
+ }
+ if (referral) {
+ /* add referral's fs locations */
+ xb_add_32(error, &xbnew, nfsls.nl_numlocs); /* FS_LOCATIONS */
+ for (loc = 0; !error && (loc < nfsls.nl_numlocs); loc++) {
+ xb_add_32(error, &xbnew, nfsls.nl_locations[loc]->nl_servcount);
+ for (serv = 0; !error && (serv < nfsls.nl_locations[loc]->nl_servcount); serv++) {
+ xb_add_string(error, &xbnew, nfsls.nl_locations[loc]->nl_servers[serv]->ns_name,
+ strlen(nfsls.nl_locations[loc]->nl_servers[serv]->ns_name));
+ xb_add_32(error, &xbnew, nfsls.nl_locations[loc]->nl_servers[serv]->ns_addrcount);
+ for (addr = 0; !error && (addr < nfsls.nl_locations[loc]->nl_servers[serv]->ns_addrcount); addr++)
+ xb_add_string(error, &xbnew, nfsls.nl_locations[loc]->nl_servers[serv]->ns_addresses[addr],
+ strlen(nfsls.nl_locations[loc]->nl_servers[serv]->ns_addresses[addr]));
+ xb_add_32(error, &xbnew, 0); /* empty server info */
+ }
+ xb_add_32(error, &xbnew, nfsls.nl_locations[loc]->nl_path.np_compcount);
+ for (comp = 0; !error && (comp < nfsls.nl_locations[loc]->nl_path.np_compcount); comp++)
+ xb_add_string(error, &xbnew, nfsls.nl_locations[loc]->nl_path.np_components[comp],
+ strlen(nfsls.nl_locations[loc]->nl_path.np_components[comp]));
+ xb_add_32(error, &xbnew, 0); /* empty fs location info */
+ }
+ }
+ if (NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MNTFLAGS))
+ xb_get_32(error, &xb, mntflags);
+ /*
+ * We add the following mount flags to the ones for the mounted-on mount:
+ * MNT_DONTBROWSE - to keep the mount from showing up as a separate volume
+ * MNT_AUTOMOUNTED - to keep DiskArb from retriggering the mount after
+ * an unmount (looking for /.autodiskmounted)
+ */
+ mntflags |= (MNT_AUTOMOUNTED | MNT_DONTBROWSE);
+ xb_add_32(error, &xbnew, mntflags);
+ if (!referral && NFS_BITMAP_ISSET(mattrs, NFS_MATTR_MNTFROM)) {
+ /* copy mntfrom string and add relpath */
+ rlen = strlen(relpath);
+ xb_get_32(error, &xb, mlen);
+ nfsmerr_if(error);
+ mlen2 = mlen + ((relpath[0] != '/') ? 1 : 0) + rlen;
+ xb_add_32(error, &xbnew, mlen2);
+ count = mlen/XDRWORD;
+ /* copy the original string */
+ while (count-- > 0)
+ xb_copy_32(error, &xb, &xbnew, val);
+ if (!error && (mlen % XDRWORD)) {
+ error = xb_get_bytes(&xb, buf, mlen%XDRWORD, 0);
+ if (!error)
+ error = xb_add_bytes(&xbnew, buf, mlen%XDRWORD, 1);
+ }
+ /* insert a '/' if the relative path doesn't start with one */
+ if (!error && (relpath[0] != '/')) {
+ buf[0] = '/';
+ error = xb_add_bytes(&xbnew, buf, 1, 1);
+ }
+ /* add the additional relative path */
+ if (!error)
+ error = xb_add_bytes(&xbnew, relpath, rlen, 1);
+ /* make sure the resulting string has the right number of pad bytes */
+ if (!error && (mlen2 != nfsm_rndup(mlen2))) {
+ bzero(buf, sizeof(buf));
+ count = nfsm_rndup(mlen2) - mlen2;
+ error = xb_add_bytes(&xbnew, buf, count, 1);
+ }
+ }
+ xb_build_done(error, &xbnew);
+
+ /* update opaque counts */
+ end_offset = xb_offset(&xbnew);
+ if (!error) {
+ error = xb_seek(&xbnew, argslength_offset);
+ argslength = end_offset - argslength_offset + XDRWORD/*version*/;
+ xb_add_32(error, &xbnew, argslength);
+ }
+ if (!error) {
+ error = xb_seek(&xbnew, attrslength_offset);
+ xb_add_32(error, &xbnew, end_offset - attrslength_offset - XDRWORD/*don't include length field*/);
+ }
+ nfsmerr_if(error);
+
+ /*
+ * For kernel_mount() call, use the existing mount flags (instead of the
+ * original flags) because flags like MNT_NOSUID and MNT_NODEV may have
+ * been silently enforced.
+ */
+ mntflags = vnode_vfsvisflags(vp);
+ mntflags |= (MNT_AUTOMOUNTED | MNT_DONTBROWSE);
+
+ /* do the mount */
+ error = kernel_mount(fstype, dvp, vp, path, xb_buffer_base(&xbnew), argslength,
+ mntflags, KERNEL_MOUNT_PERMIT_UNMOUNT | KERNEL_MOUNT_NOAUTH, ctx);
+
+nfsmerr:
+ if (error)
+ printf("nfs: mirror mount of %s on %s failed (%d)\n",
+ mntfromname, path, error);
+ /* clean up */
+ xb_cleanup(&xbnew);
+ if (referral)
+ nfs_fs_locations_cleanup(&nfsls);
+ if (path)
+ FREE_ZONE(path, MAXPATHLEN, M_NAMEI);
+ if (mntfromname)
+ FREE_ZONE(mntfromname, MAXPATHLEN, M_NAMEI);
+ if (!error)
+ nfs_ephemeral_mount_harvester_start();
+ return (error);
+}
+
+/*
+ * trigger vnode functions
+ */
+
+resolver_result_t
+nfs_mirror_mount_trigger_resolve(
+ vnode_t vp,
+ const struct componentname *cnp,
+ enum path_operation pop,
+ __unused int flags,
+ __unused void *data,
+ vfs_context_t ctx)
+{
+ nfsnode_t np = VTONFS(vp);
+ vnode_t pvp = NULLVP;
+ int error = 0;
+ resolver_result_t result;
+
+ /*
+ * We have a trigger node that doesn't have anything mounted on it yet.
+ * We'll do the mount if either:
+ * (a) this isn't the last component of the path OR
+ * (b) this is an op that looks like it should trigger the mount.
+ */
+ if (cnp->cn_flags & ISLASTCN) {
+ switch (pop) {
+ case OP_MOUNT:
+ case OP_UNMOUNT:
+ case OP_STATFS:
+ case OP_LINK:
+ case OP_UNLINK:
+ case OP_RENAME:
+ case OP_MKNOD:
+ case OP_MKFIFO:
+ case OP_SYMLINK:
+ case OP_ACCESS:
+ case OP_GETATTR:
+ case OP_MKDIR:
+ case OP_RMDIR:
+ case OP_REVOKE:
+ case OP_GETXATTR:
+ case OP_LISTXATTR:
+ /* don't perform the mount for these operations */
+ result = vfs_resolver_result(np->n_trigseq, RESOLVER_NOCHANGE, 0);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger RESOLVE: no change, last %d nameiop %d, seq %d",
+ (cnp->cn_flags & ISLASTCN) ? 1 : 0, cnp->cn_nameiop, np->n_trigseq);
+#endif
+ return (result);
+ case OP_OPEN:
+ case OP_CHDIR:
+ case OP_CHROOT:
+ case OP_TRUNCATE:
+ case OP_COPYFILE:
+ case OP_PATHCONF:
+ case OP_READLINK:
+ case OP_SETATTR:
+ case OP_EXCHANGEDATA:
+ case OP_SEARCHFS:
+ case OP_FSCTL:
+ case OP_SETXATTR:
+ case OP_REMOVEXATTR:
+ default:
+ /* go ahead and do the mount */
+ break;
+ }
+ }
+
+ if (vnode_mountedhere(vp) != NULL) {
+ /*
+ * Um... there's already something mounted.
+ * Been there. Done that. Let's just say it succeeded.
+ */
+ error = 0;
+ goto skipmount;
+ }
+
+ if ((error = nfs_node_set_busy(np, vfs_context_thread(ctx)))) {
+ result = vfs_resolver_result(np->n_trigseq, RESOLVER_ERROR, error);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger RESOLVE: busy error %d, last %d nameiop %d, seq %d",
+ error, (cnp->cn_flags & ISLASTCN) ? 1 : 0, cnp->cn_nameiop, np->n_trigseq);
+#endif
+ return (result);
+ }
+
+ pvp = vnode_getparent(vp);
+ if (pvp == NULLVP)
+ error = EINVAL;
+ if (!error)
+ error = nfs_mirror_mount_domount(pvp, vp, ctx);
+skipmount:
+ if (!error)
+ np->n_trigseq++;
+ result = vfs_resolver_result(np->n_trigseq, error ? RESOLVER_ERROR : RESOLVER_RESOLVED, error);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger RESOLVE: %s %d, last %d nameiop %d, seq %d",
+ error ? "error" : "resolved", error,
+ (cnp->cn_flags & ISLASTCN) ? 1 : 0, cnp->cn_nameiop, np->n_trigseq);
+#endif
+
+ if (pvp != NULLVP)
+ vnode_put(pvp);
+ nfs_node_clear_busy(np);
+ return (result);
+}
+
+resolver_result_t
+nfs_mirror_mount_trigger_unresolve(
+ vnode_t vp,
+ int flags,
+ __unused void *data,
+ vfs_context_t ctx)
+{
+ nfsnode_t np = VTONFS(vp);
+ mount_t mp;
+ int error;
+ resolver_result_t result;
+
+ if ((error = nfs_node_set_busy(np, vfs_context_thread(ctx)))) {
+ result = vfs_resolver_result(np->n_trigseq, RESOLVER_ERROR, error);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger UNRESOLVE: busy error %d, seq %d", error, np->n_trigseq);
+#endif
+ return (result);
+ }
+
+ mp = vnode_mountedhere(vp);
+ if (!mp)
+ error = EINVAL;
+ if (!error)
+ error = vfs_unmountbyfsid(&(vfs_statfs(mp)->f_fsid), flags, ctx);
+ if (!error)
+ np->n_trigseq++;
+ result = vfs_resolver_result(np->n_trigseq, error ? RESOLVER_ERROR : RESOLVER_UNRESOLVED, error);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger UNRESOLVE: %s %d, seq %d",
+ error ? "error" : "unresolved", error, np->n_trigseq);
+#endif
+ nfs_node_clear_busy(np);
+ return (result);
+}
+
+resolver_result_t
+nfs_mirror_mount_trigger_rearm(
+ vnode_t vp,
+ __unused int flags,
+ __unused void *data,
+ vfs_context_t ctx)
+{
+ nfsnode_t np = VTONFS(vp);
+ int error;
+ resolver_result_t result;
+
+ if ((error = nfs_node_set_busy(np, vfs_context_thread(ctx)))) {
+ result = vfs_resolver_result(np->n_trigseq, RESOLVER_ERROR, error);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger REARM: busy error %d, seq %d", error, np->n_trigseq);
+#endif
+ return (result);
+ }
+
+ np->n_trigseq++;
+ result = vfs_resolver_result(np->n_trigseq,
+ vnode_mountedhere(vp) ? RESOLVER_RESOLVED : RESOLVER_UNRESOLVED, 0);
+#ifdef NFS_TRIGGER_DEBUG
+ NP(np, "nfs trigger REARM: %s, seq %d",
+ vnode_mountedhere(vp) ? "resolved" : "unresolved", np->n_trigseq);
+#endif
+ nfs_node_clear_busy(np);
+ return (result);
+}
+
+/*
+ * Periodically attempt to unmount ephemeral (mirror) mounts in an attempt to limit
+ * the number of unused mounts.
+ */
+
+#define NFS_EPHEMERAL_MOUNT_HARVEST_INTERVAL 120 /* how often the harvester runs */
+struct nfs_ephemeral_mount_harvester_info {
+ fsid_t fsid; /* FSID that we need to try to unmount */
+ uint32_t mountcount; /* count of ephemeral mounts seen in scan */
+ };
+/* various globals for the harvester */
+static thread_call_t nfs_ephemeral_mount_harvester_timer = NULL;
+static int nfs_ephemeral_mount_harvester_on = 0;
+
+kern_return_t thread_terminate(thread_t);
+
+static int
+nfs_ephemeral_mount_harvester_callback(mount_t mp, void *arg)
+{
+ struct nfs_ephemeral_mount_harvester_info *hinfo = arg;
+ struct nfsmount *nmp;
+ struct timeval now;
+
+ if (strcmp(mp->mnt_vfsstat.f_fstypename, "nfs"))
+ return (VFS_RETURNED);
+ nmp = VFSTONFS(mp);
+ if (!nmp || !NMFLAG(nmp, EPHEMERAL))
+ return (VFS_RETURNED);
+ hinfo->mountcount++;
+
+ /* avoid unmounting mounts that have been triggered within the last harvest interval */
+ microtime(&now);
+ if ((nmp->nm_mounttime >> 32) > ((uint32_t)now.tv_sec - NFS_EPHEMERAL_MOUNT_HARVEST_INTERVAL))
+ return (VFS_RETURNED);
+
+ if (hinfo->fsid.val[0] || hinfo->fsid.val[1]) {
+ /* attempt to unmount previously-found ephemeral mount */
+ vfs_unmountbyfsid(&hinfo->fsid, 0, vfs_context_kernel());
+ hinfo->fsid.val[0] = hinfo->fsid.val[1] = 0;
+ }
+
+ /*
+ * We can't call unmount here since we hold a mount iter ref
+ * on mp so save its fsid for the next call iteration to unmount.
+ */
+ hinfo->fsid.val[0] = mp->mnt_vfsstat.f_fsid.val[0];
+ hinfo->fsid.val[1] = mp->mnt_vfsstat.f_fsid.val[1];
+
+ return (VFS_RETURNED);
+}
+
+/*
+ * Spawn a thread to do the ephemeral mount harvesting.
+ */
+static void
+nfs_ephemeral_mount_harvester_timer_func(void)
+{
+ thread_t thd;
+
+ if (kernel_thread_start(nfs_ephemeral_mount_harvester, NULL, &thd) == KERN_SUCCESS)
+ thread_deallocate(thd);
+}
+
+/*
+ * Iterate all mounts looking for NFS ephemeral mounts to try to unmount.
+ */
+void
+nfs_ephemeral_mount_harvester(__unused void *arg, __unused wait_result_t wr)
+{
+ struct nfs_ephemeral_mount_harvester_info hinfo;
+ uint64_t deadline;
+
+ hinfo.mountcount = 0;
+ hinfo.fsid.val[0] = hinfo.fsid.val[1] = 0;
+ vfs_iterate(VFS_ITERATE_TAIL_FIRST, nfs_ephemeral_mount_harvester_callback, &hinfo);
+ if (hinfo.fsid.val[0] || hinfo.fsid.val[1]) {
+ /* attempt to unmount last found ephemeral mount */
+ vfs_unmountbyfsid(&hinfo.fsid, 0, vfs_context_kernel());
projects / apple / xnu.git / blobdiff