+ if (!noop && !newnoop && alloc) {
+ lck_mtx_unlock(&nmp->nm_lock);
+ MALLOC(newnoop, struct nfs_open_owner *, sizeof(struct nfs_open_owner), M_TEMP, M_WAITOK);
+ if (!newnoop)
+ return (NULL);
+ bzero(newnoop, sizeof(*newnoop));
+ lck_mtx_init(&newnoop->noo_lock, nfs_open_grp, LCK_ATTR_NULL);
+ newnoop->noo_mount = nmp;
+ kauth_cred_ref(cred);
+ newnoop->noo_cred = cred;
+ newnoop->noo_name = OSAddAtomic(1, &nfs_open_owner_seqnum);
+ TAILQ_INIT(&newnoop->noo_opens);
+ goto tryagain;
+ }
+ if (!noop && newnoop) {
+ newnoop->noo_flags |= NFS_OPEN_OWNER_LINK;
+ TAILQ_INSERT_HEAD(&nmp->nm_open_owners, newnoop, noo_link);
+ noop = newnoop;
+ }
+ lck_mtx_unlock(&nmp->nm_lock);
+
+ if (newnoop && (noop != newnoop))
+ nfs_open_owner_destroy(newnoop);
+
+ if (noop)
+ nfs_open_owner_ref(noop);
+
+ return (noop);
+}
+
+/*
+ * destroy an open owner that's no longer needed
+ */
+void
+nfs_open_owner_destroy(struct nfs_open_owner *noop)
+{
+ if (noop->noo_cred)
+ kauth_cred_unref(&noop->noo_cred);
+ lck_mtx_destroy(&noop->noo_lock, nfs_open_grp);
+ FREE(noop, M_TEMP);
+}
+
+/*
+ * acquire a reference count on an open owner
+ */
+void
+nfs_open_owner_ref(struct nfs_open_owner *noop)
+{
+ lck_mtx_lock(&noop->noo_lock);
+ noop->noo_refcnt++;
+ lck_mtx_unlock(&noop->noo_lock);
+}
+
+/*
+ * drop a reference count on an open owner and destroy it if
+ * it is no longer referenced and no longer on the mount's list.
+ */
+void
+nfs_open_owner_rele(struct nfs_open_owner *noop)
+{
+ lck_mtx_lock(&noop->noo_lock);
+ if (noop->noo_refcnt < 1)
+ panic("nfs_open_owner_rele: no refcnt");
+ noop->noo_refcnt--;
+ if (!noop->noo_refcnt && (noop->noo_flags & NFS_OPEN_OWNER_BUSY))
+ panic("nfs_open_owner_rele: busy");
+ /* XXX we may potentially want to clean up idle/unused open owner structures */
+ if (noop->noo_refcnt || (noop->noo_flags & NFS_OPEN_OWNER_LINK)) {
+ lck_mtx_unlock(&noop->noo_lock);
+ return;
+ }
+ /* owner is no longer referenced or linked to mount, so destroy it */
+ lck_mtx_unlock(&noop->noo_lock);
+ nfs_open_owner_destroy(noop);
+}
+
+/*
+ * Mark an open owner as busy because we are about to
+ * start an operation that uses and updates open owner state.
+ */
+int
+nfs_open_owner_set_busy(struct nfs_open_owner *noop, thread_t thd)
+{
+ struct nfsmount *nmp;
+ struct timespec ts = {2, 0};
+ int error = 0, slpflag;
+
+ nmp = noop->noo_mount;
+ if (!nmp)
+ return (ENXIO);
+ slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
+
+ lck_mtx_lock(&noop->noo_lock);
+ while (noop->noo_flags & NFS_OPEN_OWNER_BUSY) {
+ if ((error = nfs_sigintr(nmp, NULL, thd, 0)))
+ break;
+ noop->noo_flags |= NFS_OPEN_OWNER_WANT;
+ msleep(noop, &noop->noo_lock, slpflag, "nfs_open_owner_set_busy", &ts);
+ }
+ if (!error)
+ noop->noo_flags |= NFS_OPEN_OWNER_BUSY;
+ lck_mtx_unlock(&noop->noo_lock);
+
+ return (error);
+}
+
+/*
+ * Clear the busy flag on an open owner and wake up anyone waiting
+ * to mark it busy.
+ */
+void
+nfs_open_owner_clear_busy(struct nfs_open_owner *noop)
+{
+ int wanted;
+
+ lck_mtx_lock(&noop->noo_lock);
+ if (!(noop->noo_flags & NFS_OPEN_OWNER_BUSY))
+ panic("nfs_open_owner_clear_busy");
+ wanted = (noop->noo_flags & NFS_OPEN_OWNER_WANT);
+ noop->noo_flags &= ~(NFS_OPEN_OWNER_BUSY|NFS_OPEN_OWNER_WANT);
+ lck_mtx_unlock(&noop->noo_lock);
+ if (wanted)
+ wakeup(noop);
+}
+
+/*
+ * Given an open/lock owner and an error code, increment the
+ * sequence ID if appropriate.
+ */
+void
+nfs_owner_seqid_increment(struct nfs_open_owner *noop, struct nfs_lock_owner *nlop, int error)
+{
+ switch (error) {
+ case NFSERR_STALE_CLIENTID:
+ case NFSERR_STALE_STATEID:
+ case NFSERR_OLD_STATEID:
+ case NFSERR_BAD_STATEID:
+ case NFSERR_BAD_SEQID:
+ case NFSERR_BADXDR:
+ case NFSERR_RESOURCE:
+ case NFSERR_NOFILEHANDLE:
+ /* do not increment the open seqid on these errors */
+ return;
+ }
+ if (noop)
+ noop->noo_seqid++;
+ if (nlop)
+ nlop->nlo_seqid++;
+}
+
+/*
+ * Search a node's open file list for any conflicts with this request.
+ * Also find this open owner's open file structure.
+ * If not found and "alloc" is set, then allocate one.
+ */
+int
+nfs_open_file_find(
+ nfsnode_t np,
+ struct nfs_open_owner *noop,
+ struct nfs_open_file **nofpp,
+ uint32_t accessMode,
+ uint32_t denyMode,
+ int alloc)
+{
+ struct nfs_open_file *nofp = NULL, *nofp2, *newnofp = NULL;
+
+ if (!np)
+ goto alloc;
+tryagain:
+ lck_mtx_lock(&np->n_openlock);
+ TAILQ_FOREACH(nofp2, &np->n_opens, nof_link) {
+ if (nofp2->nof_owner == noop) {
+ nofp = nofp2;
+ if (!accessMode)
+ break;
+ }
+ if ((accessMode & nofp2->nof_deny) || (denyMode & nofp2->nof_access)) {
+ /* This request conflicts with an existing open on this client. */
+ lck_mtx_unlock(&np->n_openlock);
+ *nofpp = NULL;
+ return (EACCES);
+ }
+ }
+
+ /*
+ * If this open owner doesn't have an open
+ * file structure yet, we create one for it.
+ */
+ if (!nofp && !newnofp && alloc) {
+ lck_mtx_unlock(&np->n_openlock);
+alloc:
+ MALLOC(newnofp, struct nfs_open_file *, sizeof(struct nfs_open_file), M_TEMP, M_WAITOK);
+ if (!newnofp) {
+ *nofpp = NULL;
+ return (ENOMEM);
+ }
+ bzero(newnofp, sizeof(*newnofp));
+ lck_mtx_init(&newnofp->nof_lock, nfs_open_grp, LCK_ATTR_NULL);
+ newnofp->nof_owner = noop;
+ nfs_open_owner_ref(noop);
+ newnofp->nof_np = np;
+ lck_mtx_lock(&noop->noo_lock);
+ TAILQ_INSERT_HEAD(&noop->noo_opens, newnofp, nof_oolink);
+ lck_mtx_unlock(&noop->noo_lock);
+ if (np)
+ goto tryagain;
+ }
+ if (!nofp && newnofp) {
+ if (np)
+ TAILQ_INSERT_HEAD(&np->n_opens, newnofp, nof_link);
+ nofp = newnofp;
+ }
+ if (np)
+ lck_mtx_unlock(&np->n_openlock);
+
+ if (newnofp && (nofp != newnofp))
+ nfs_open_file_destroy(newnofp);
+
+ *nofpp = nofp;
+ return (nofp ? 0 : ESRCH);
+}
+
+/*
+ * Destroy an open file structure.
+ */
+void
+nfs_open_file_destroy(struct nfs_open_file *nofp)
+{
+ lck_mtx_lock(&nofp->nof_owner->noo_lock);
+ TAILQ_REMOVE(&nofp->nof_owner->noo_opens, nofp, nof_oolink);
+ lck_mtx_unlock(&nofp->nof_owner->noo_lock);
+ nfs_open_owner_rele(nofp->nof_owner);
+ lck_mtx_destroy(&nofp->nof_lock, nfs_open_grp);
+ FREE(nofp, M_TEMP);
+}
+
+/*
+ * Mark an open file as busy because we are about to
+ * start an operation that uses and updates open file state.
+ */
+int
+nfs_open_file_set_busy(struct nfs_open_file *nofp, thread_t thd)
+{
+ struct nfsmount *nmp;
+ struct timespec ts = {2, 0};
+ int error = 0, slpflag;
+
+ nmp = nofp->nof_owner->noo_mount;
+ if (!nmp)
+ return (ENXIO);
+ slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
+
+ lck_mtx_lock(&nofp->nof_lock);
+ while (nofp->nof_flags & NFS_OPEN_FILE_BUSY) {
+ if ((error = nfs_sigintr(nmp, NULL, thd, 0)))
+ break;
+ nofp->nof_flags |= NFS_OPEN_FILE_WANT;
+ msleep(nofp, &nofp->nof_lock, slpflag, "nfs_open_file_set_busy", &ts);
+ }
+ if (!error)
+ nofp->nof_flags |= NFS_OPEN_FILE_BUSY;
+ lck_mtx_unlock(&nofp->nof_lock);
+
+ return (error);
+}
+
+/*
+ * Clear the busy flag on an open file and wake up anyone waiting
+ * to mark it busy.
+ */
+void
+nfs_open_file_clear_busy(struct nfs_open_file *nofp)
+{
+ int wanted;
+
+ lck_mtx_lock(&nofp->nof_lock);
+ if (!(nofp->nof_flags & NFS_OPEN_FILE_BUSY))
+ panic("nfs_open_file_clear_busy");
+ wanted = (nofp->nof_flags & NFS_OPEN_FILE_WANT);
+ nofp->nof_flags &= ~(NFS_OPEN_FILE_BUSY|NFS_OPEN_FILE_WANT);
+ lck_mtx_unlock(&nofp->nof_lock);
+ if (wanted)
+ wakeup(nofp);
+}
+
+/*
+ * Get the current (delegation, lock, open, default) stateid for this node.
+ * If node has a delegation, use that stateid.
+ * If pid has a lock, use the lockowner's stateid.
+ * Or use the open file's stateid.
+ * If no open file, use a default stateid of all ones.
+ */
+void
+nfs_get_stateid(nfsnode_t np, thread_t thd, kauth_cred_t cred, nfs_stateid *sid)
+{
+ struct nfsmount *nmp = NFSTONMP(np);
+ proc_t p = thd ? get_bsdthreadtask_info(thd) : current_thread(); // XXX async I/O requests don't have a thread
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+ struct nfs_lock_owner *nlop = NULL;
+ nfs_stateid *s = NULL;
+
+ if (np->n_openflags & N_DELEG_MASK)
+ s = &np->n_dstateid;
+ else if (p)
+ nlop = nfs_lock_owner_find(np, p, 0);
+ if (nlop && !TAILQ_EMPTY(&nlop->nlo_locks)) {
+ /* we hold locks, use lock stateid */
+ s = &nlop->nlo_stateid;
+ } else if (((noop = nfs_open_owner_find(nmp, cred, 0))) &&
+ (nfs_open_file_find(np, noop, &nofp, 0, 0, 0) == 0) &&
+ !(nofp->nof_flags & NFS_OPEN_FILE_LOST) &&
+ nofp->nof_access) {
+ /* we (should) have the file open, use open stateid */
+ if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)
+ nfs4_reopen(nofp, thd);
+ if (!(nofp->nof_flags & NFS_OPEN_FILE_LOST))
+ s = &nofp->nof_stateid;
+ }
+
+ if (s) {
+ sid->seqid = s->seqid;
+ sid->other[0] = s->other[0];
+ sid->other[1] = s->other[1];
+ sid->other[2] = s->other[2];
+ } else {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_get_stateid: no stateid for %s\n", vname ? vname : "???");
+ vnode_putname(vname);
+ sid->seqid = sid->other[0] = sid->other[1] = sid->other[2] = 0xffffffff;
+ }
+ if (nlop)
+ nfs_lock_owner_rele(nlop);
+ if (noop)
+ nfs_open_owner_rele(noop);
+}
+
+/*
+ * We always send the open RPC even if this open's mode is a subset of all
+ * the existing opens. This makes sure that we will always be able to do a
+ * downgrade to any of the open modes.
+ *
+ * Note: local conflicts should have already been checked. (nfs_open_file_find)
+ */
+int
+nfs4_open(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ uint32_t accessMode,
+ uint32_t denyMode,
+ vfs_context_t ctx)
+{
+ vnode_t vp = NFSTOV(np);
+ vnode_t dvp = NULL;
+ struct componentname cn;
+ const char *vname = NULL;
+ size_t namelen;
+ char smallname[128];
+ char *filename = NULL;
+ int error = 0, readtoo = 0;
+
+ dvp = vnode_getparent(vp);
+ vname = vnode_getname(vp);
+ if (!dvp || !vname) {
+ error = EIO;
+ goto out;
+ }
+ filename = &smallname[0];
+ namelen = snprintf(filename, sizeof(smallname), "%s", vname);
+ if (namelen >= sizeof(smallname)) {
+ namelen++; /* snprintf result doesn't include '\0' */
+ MALLOC(filename, char *, namelen, M_TEMP, M_WAITOK);
+ if (!filename) {
+ error = ENOMEM;
+ goto out;
+ }
+ snprintf(filename, namelen, "%s", vname);
+ }
+ bzero(&cn, sizeof(cn));
+ cn.cn_nameptr = filename;
+ cn.cn_namelen = namelen;
+
+ if (!(accessMode & NFS_OPEN_SHARE_ACCESS_READ)) {
+ /*
+ * Try to open it for read access too,
+ * so the buffer cache can read data.
+ */
+ readtoo = 1;
+ accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
+ }
+tryagain:
+ error = nfs4_open_rpc(nofp, ctx, &cn, NULL, dvp, &vp, NFS_OPEN_NOCREATE, accessMode, denyMode);
+ if (error) {
+ if (!nfs_mount_state_error_should_restart(error) && readtoo) {
+ /* try again without the extra read access */
+ accessMode &= ~NFS_OPEN_SHARE_ACCESS_READ;
+ readtoo = 0;
+ goto tryagain;
+ }
+ goto out;
+ }
+ nofp->nof_access |= accessMode;
+ nofp->nof_deny |= denyMode;
+
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE) {
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ)
+ nofp->nof_r++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE)
+ nofp->nof_w++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH)
+ nofp->nof_rw++;
+ } else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE) {
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ)
+ nofp->nof_r_dw++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE)
+ nofp->nof_w_dw++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH)
+ nofp->nof_rw_dw++;
+ } else { /* NFS_OPEN_SHARE_DENY_BOTH */
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ)
+ nofp->nof_r_drw++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE)
+ nofp->nof_w_drw++;
+ else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH)
+ nofp->nof_rw_drw++;
+ }
+ nofp->nof_opencnt++;
+out:
+ if (filename && (filename != &smallname[0]))
+ FREE(filename, M_TEMP);
+ if (vname)
+ vnode_putname(vname);
+ if (dvp != NULLVP)
+ vnode_put(dvp);
+ return (error);
+}
+
+
+int
+nfs4_vnop_open(
+ struct vnop_open_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ int a_mode;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vfs_context_t ctx = ap->a_context;
+ vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
+ int error, accessMode, denyMode, opened = 0;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+
+ if (!(ap->a_mode & (FREAD|FWRITE)))
+ return (EINVAL);
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+
+ /* First, call the common code */
+ if ((error = nfs3_vnop_open(ap)))
+ return (error);
+
+ if (!vnode_isreg(vp)) {
+ /* Just mark that it was opened */
+ lck_mtx_lock(&np->n_openlock);
+ np->n_openrefcnt++;
+ lck_mtx_unlock(&np->n_openlock);
+ return (0);
+ }
+
+ /* mode contains some combination of: FREAD, FWRITE, O_SHLOCK, O_EXLOCK */
+ accessMode = 0;
+ if (ap->a_mode & FREAD)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
+ if (ap->a_mode & FWRITE)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
+ if (ap->a_mode & O_EXLOCK)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else if (ap->a_mode & O_SHLOCK)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 1);
+ if (!noop)
+ return (ENOMEM);
+
+restart:
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+
+ error = nfs_open_file_find(np, noop, &nofp, accessMode, denyMode, 1);
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_open: LOST %s\n", vname);
+ vnode_putname(vname);
+ error = EIO;
+ }
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ goto restart;
+ }
+ if (!error)
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ nofp = NULL;
+ goto out;
+ }
+
+ /*
+ * If we just created the file and the modes match, then we simply use
+ * the open performed in the create. Otherwise, send the request.
+ */
+ if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread()) &&
+ (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) &&
+ (denyMode == NFS_OPEN_SHARE_DENY_NONE)) {
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
+ nofp->nof_creator = NULL;
+ } else {
+ if (!opened)
+ error = nfs4_open(np, nofp, accessMode, denyMode, ctx);
+ if ((error == EACCES) && (nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread())) {
+ /*
+ * Ugh. This can happen if we just created the file with read-only
+ * perms and we're trying to open it for real with different modes
+ * (e.g. write-only or with a deny mode) and the server decides to
+ * not allow the second open because of the read-only perms.
+ * The best we can do is to just use the create's open.
+ * We may have access we don't need or we may not have a requested
+ * deny mode. We may log complaints later, but we'll try to avoid it.
+ */
+ if (denyMode != NFS_OPEN_SHARE_DENY_NONE) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs4_vnop_open: deny mode foregone on create, %s\n", vname);
+ vnode_putname(vname);
+ }
+ nofp->nof_creator = NULL;
+ error = 0;
+ }
+ if (error)
+ goto out;
+ opened = 1;
+ /*
+ * If we had just created the file, we already had it open.
+ * If the actual open mode is less than what we grabbed at
+ * create time, then we'll downgrade the open here.
+ */
+ if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) &&
+ (nofp->nof_creator == current_thread())) {
+ error = nfs4_close(np, nofp, NFS_OPEN_SHARE_ACCESS_BOTH, NFS_OPEN_SHARE_DENY_NONE, ctx);
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_open: create close error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ }
+ if (!nfs_mount_state_error_should_restart(error)) {
+ error = 0;
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
+ }
+ }
+ }
+
+out:
+ if (nofp)
+ nfs_open_file_clear_busy(nofp);
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ nofp = NULL;
+ goto restart;
+ }
+ if (noop)
+ nfs_open_owner_rele(noop);
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_open: error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ }
+ return (error);
+}
+
+int
+nfs4_close(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ uint32_t accessMode,
+ uint32_t denyMode,
+ vfs_context_t ctx)
+{
+ struct nfs_lock_owner *nlop;
+ int error = 0, changed = 0, closed = 0;
+ uint32_t newAccessMode, newDenyMode;
+
+ /* warn if modes don't match current state */
+ if (((accessMode & nofp->nof_access) != accessMode) || ((denyMode & nofp->nof_deny) != denyMode)) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs4_close: mode mismatch %d %d, current %d %d, %s\n",
+ accessMode, denyMode, nofp->nof_access, nofp->nof_deny, vname);
+ vnode_putname(vname);
+ }
+
+ /*
+ * If we're closing a write-only open, we may not have a write-only count
+ * if we also grabbed read access. So, check the read-write count.
+ */
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE) {
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w == 0) && nofp->nof_rw)
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ } else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE) {
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w_dw == 0) && nofp->nof_rw_dw)
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ } else { /* NFS_OPEN_SHARE_DENY_BOTH */
+ if ((accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (nofp->nof_w_drw == 0) && nofp->nof_rw_drw)
+ accessMode = NFS_OPEN_SHARE_ACCESS_BOTH;
+ }
+
+ /*
+ * Calculate new modes: a mode bit gets removed when there's only
+ * one count in all the corresponding counts
+ */
+ newAccessMode = nofp->nof_access;
+ newDenyMode = nofp->nof_deny;
+ if ((accessMode & NFS_OPEN_SHARE_ACCESS_READ) &&
+ (newAccessMode & NFS_OPEN_SHARE_ACCESS_READ) &&
+ ((nofp->nof_r + nofp->nof_rw + nofp->nof_r_dw +
+ nofp->nof_rw_dw + nofp->nof_r_drw + nofp->nof_rw_dw) == 1)) {
+ newAccessMode &= ~NFS_OPEN_SHARE_ACCESS_READ;
+ changed = 1;
+ }
+ if ((accessMode & NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ (newAccessMode & NFS_OPEN_SHARE_ACCESS_WRITE) &&
+ ((nofp->nof_w + nofp->nof_rw + nofp->nof_w_dw +
+ nofp->nof_rw_dw + nofp->nof_w_drw + nofp->nof_rw_dw) == 1)) {
+ newAccessMode &= ~NFS_OPEN_SHARE_ACCESS_WRITE;
+ changed = 1;
+ }
+ if ((denyMode & NFS_OPEN_SHARE_DENY_READ) &&
+ (newDenyMode & NFS_OPEN_SHARE_DENY_READ) &&
+ ((nofp->nof_r_drw + nofp->nof_w_drw + nofp->nof_rw_drw) == 1)) {
+ newDenyMode &= ~NFS_OPEN_SHARE_DENY_READ;
+ changed = 1;
+ }
+ if ((denyMode & NFS_OPEN_SHARE_DENY_WRITE) &&
+ (newDenyMode & NFS_OPEN_SHARE_DENY_WRITE) &&
+ ((nofp->nof_r_drw + nofp->nof_w_drw + nofp->nof_rw_drw +
+ nofp->nof_r_dw + nofp->nof_w_dw + nofp->nof_rw_dw) == 1)) {
+ newDenyMode &= ~NFS_OPEN_SHARE_DENY_WRITE;
+ changed = 1;
+ }
+
+
+ if ((newAccessMode == 0) || (nofp->nof_opencnt == 1)) {
+ /*
+ * No more access after this close, so clean up and close it.
+ */
+ closed = 1;
+ if (!(nofp->nof_flags & NFS_OPEN_FILE_LOST))
+ error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
+ if (error == NFSERR_LOCKS_HELD) {
+ /*
+ * Hmm... the server says we have locks we need to release first
+ * Find the lock owner and try to unlock everything.
+ */
+ nlop = nfs_lock_owner_find(np, vfs_context_proc(ctx), 0);
+ if (nlop) {
+ nfs4_unlock_rpc(np, nlop, F_WRLCK, 0, UINT64_MAX, ctx);
+ nfs_lock_owner_rele(nlop);
+ }
+ error = nfs4_close_rpc(np, nofp, vfs_context_thread(ctx), vfs_context_ucred(ctx), 0);
+ }
+ } else if (changed) {
+ /*
+ * File is still open but with less access, so downgrade the open.
+ */
+ if (!(nofp->nof_flags & NFS_OPEN_FILE_LOST))
+ error = nfs4_open_downgrade_rpc(np, nofp, ctx);
+ }
+
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs4_close: error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ return (error);
+ }
+
+ /* Decrement the corresponding open access/deny mode counter. */
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE) {
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ) {
+ if (nofp->nof_r == 0)
+ printf("nfs4_close: open(R) count underrun\n");
+ else
+ nofp->nof_r--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) {
+ if (nofp->nof_w == 0)
+ printf("nfs4_close: open(W) count underrun\n");
+ else
+ nofp->nof_w--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) {
+ if (nofp->nof_rw == 0)
+ printf("nfs4_close: open(RW) count underrun\n");
+ else
+ nofp->nof_rw--;
+ }
+ } else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE) {
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ) {
+ if (nofp->nof_r_dw == 0)
+ printf("nfs4_close: open(R,DW) count underrun\n");
+ else
+ nofp->nof_r_dw--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) {
+ if (nofp->nof_w_dw == 0)
+ printf("nfs4_close: open(W,DW) count underrun\n");
+ else
+ nofp->nof_w_dw--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) {
+ if (nofp->nof_rw_dw == 0)
+ printf("nfs4_close: open(RW,DW) count underrun\n");
+ else
+ nofp->nof_rw_dw--;
+ }
+ } else { /* NFS_OPEN_SHARE_DENY_BOTH */
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_READ) {
+ if (nofp->nof_r_drw == 0)
+ printf("nfs4_close: open(R,DRW) count underrun\n");
+ else
+ nofp->nof_r_drw--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_WRITE) {
+ if (nofp->nof_w_drw == 0)
+ printf("nfs4_close: open(W,DRW) count underrun\n");
+ else
+ nofp->nof_w_drw--;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) {
+ if (nofp->nof_rw_drw == 0)
+ printf("nfs4_close: open(RW,DRW) count underrun\n");
+ else
+ nofp->nof_rw_drw--;
+ }
+ }
+ /* update the modes */
+ nofp->nof_access = newAccessMode;
+ nofp->nof_deny = newDenyMode;
+ if (closed) {
+ if (nofp->nof_r || nofp->nof_w ||
+ (nofp->nof_rw && !((nofp->nof_flags & NFS_OPEN_FILE_CREATE) && !nofp->nof_creator && (nofp->nof_rw == 1))) ||
+ nofp->nof_r_dw || nofp->nof_w_dw || nofp->nof_rw_dw ||
+ nofp->nof_r_drw || nofp->nof_w_drw || nofp->nof_rw_drw)
+ printf("nfs4_close: unexpected count: %u %u %u dw %u %u %u drw %u %u %u flags 0x%x\n",
+ nofp->nof_r, nofp->nof_w, nofp->nof_rw,
+ nofp->nof_r_dw, nofp->nof_w_dw, nofp->nof_rw_dw,
+ nofp->nof_r_drw, nofp->nof_w_drw, nofp->nof_rw_drw,
+ nofp->nof_flags);
+ /* clear out all open info, just to be safe */
+ nofp->nof_access = nofp->nof_deny = 0;
+ nofp->nof_mmap_access = nofp->nof_mmap_deny = 0;
+ nofp->nof_r = nofp->nof_w = nofp->nof_rw = 0;
+ nofp->nof_r_dw = nofp->nof_w_dw = nofp->nof_rw_dw = 0;
+ nofp->nof_r_drw = nofp->nof_w_drw = nofp->nof_rw_drw = 0;
+ nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
+ /* XXX we may potentially want to clean up idle/unused open file structures */
+ }
+ nofp->nof_opencnt--;
+ if (nofp->nof_flags & NFS_OPEN_FILE_LOST) {
+ error = EIO;
+ if (!nofp->nof_opencnt)
+ nofp->nof_flags &= ~NFS_OPEN_FILE_LOST;
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_close: LOST%s, %s\n", !(nofp->nof_flags & NFS_OPEN_FILE_LOST) ? " (last)" : "", vname);
+ vnode_putname(vname);
+ }
+ return (error);
+}
+
+int
+nfs4_vnop_close(
+ struct vnop_close_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ int a_fflag;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vfs_context_t ctx = ap->a_context;
+ vnode_t vp = ap->a_vp;
+ int fflag = ap->a_fflag;
+ int error, common_error, accessMode, denyMode;
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+
+ /* First, call the common code */
+ common_error = nfs3_vnop_close(ap);
+
+ if (!vnode_isreg(vp)) {
+ /* Just mark that it was closed */
+ lck_mtx_lock(&np->n_openlock);
+ np->n_openrefcnt--;
+ lck_mtx_unlock(&np->n_openlock);
+ return (common_error);
+ }
+
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 0);
+ if (!noop) {
+ printf("nfs4_vnop_close: can't get open owner!\n");
+ return (EIO);
+ }
+
+restart:
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+
+ error = nfs_open_file_find(np, noop, &nofp, 0, 0, 0);
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ goto restart;
+ }
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs4_vnop_close: no open file for owner %d, %s\n", error, vname);
+ vnode_putname(vname);
+ error = EBADF;
+ goto out;
+ }
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ nofp = NULL;
+ goto out;
+ }
+
+ /* fflag contains some combination of: FREAD, FWRITE, FHASLOCK */
+ accessMode = 0;
+ if (fflag & FREAD)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_READ;
+ if (fflag & FWRITE)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
+// XXX It would be nice if we still had the O_EXLOCK/O_SHLOCK flags that were on the open
+// if (fflag & O_EXLOCK)
+// denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+// else if (fflag & O_SHLOCK)
+// denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+// else
+// denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ if (fflag & FHASLOCK) {
+ /* XXX assume FHASLOCK is for the deny mode and not flock */
+ /* FHASLOCK flock will be unlocked in the close path, but the flag is not cleared. */
+ if (nofp->nof_deny & NFS_OPEN_SHARE_DENY_READ)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else if (nofp->nof_deny & NFS_OPEN_SHARE_DENY_WRITE)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ } else {
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ }
+
+ if (!accessMode) {
+ error = EINVAL;
+ goto out;
+ }
+
+ error = nfs4_close(np, nofp, accessMode, denyMode, ctx);
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_close: close error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ }
+
+out:
+ if (nofp)
+ nfs_open_file_clear_busy(nofp);
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ nofp = NULL;
+ goto restart;
+ }
+ if (noop)
+ nfs_open_owner_rele(noop);
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_close: error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ }
+ if (!error)
+ error = common_error;
+ return (error);
+}
+
+int
+nfs4_vnop_mmap(
+ struct vnop_mmap_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ int a_fflags;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vfs_context_t ctx = ap->a_context;
+ vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(vp);
+ int error = 0, accessMode, denyMode;
+ struct nfsmount *nmp;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+
+ if (!vnode_isreg(vp) || !(ap->a_fflags & (PROT_READ|PROT_WRITE)))
+ return (EINVAL);
+
+ /*
+ * fflags contains some combination of: PROT_READ, PROT_WRITE
+ * Since it's not possible to mmap() without having the file open for reading,
+ * read access is always there (regardless if PROT_READ is not set).
+ */
+ accessMode = NFS_OPEN_SHARE_ACCESS_READ;
+ if (ap->a_fflags & PROT_WRITE)
+ accessMode |= NFS_OPEN_SHARE_ACCESS_WRITE;
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 0);
+ if (!noop) {
+ printf("nfs4_vnop_mmap: no open owner\n");
+ return (EPERM);
+ }
+
+restart:
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error) {
+ nfs_open_owner_rele(noop);
+ return (error);
+ }
+
+ error = nfs_open_file_find(np, noop, &nofp, 0, 0, 1);
+ if (error || (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST))) {
+ printf("nfs4_vnop_mmap: no open file for owner %d\n", error);
+ error = EPERM;
+ }
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ goto restart;
+ }
+ if (!error)
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ nofp = NULL;
+ goto out;
+ }
+
+ /*
+ * The open reference for mmap must mirror an existing open because
+ * we may need to reclaim it after the file is closed.
+ * So grab another open count matching the accessMode passed in.
+ * If we already had an mmap open, prefer read/write without deny mode.
+ * This means we may have to drop the current mmap open first.
+ */
+
+ /* determine deny mode for open */
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) {
+ if (nofp->nof_rw)
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ else if (nofp->nof_rw_dw)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else if (nofp->nof_rw_drw)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else
+ error = EPERM;
+ } else { /* NFS_OPEN_SHARE_ACCESS_READ */
+ if (nofp->nof_r)
+ denyMode = NFS_OPEN_SHARE_DENY_NONE;
+ else if (nofp->nof_r_dw)
+ denyMode = NFS_OPEN_SHARE_DENY_WRITE;
+ else if (nofp->nof_r_drw)
+ denyMode = NFS_OPEN_SHARE_DENY_BOTH;
+ else
+ error = EPERM;
+ }
+ if (error) /* mmap mode without proper open mode */
+ goto out;
+
+ /*
+ * If the existing mmap access is more than the new access OR the
+ * existing access is the same and the existing deny mode is less,
+ * then we'll stick with the existing mmap open mode.
+ */
+ if ((nofp->nof_mmap_access > accessMode) ||
+ ((nofp->nof_mmap_access == accessMode) && (nofp->nof_mmap_deny <= denyMode)))
+ goto out;
+
+ /* update mmap open mode */
+ if (nofp->nof_mmap_access) {
+ error = nfs4_close(np, nofp, nofp->nof_mmap_access, nofp->nof_mmap_deny, ctx);
+ if (error) {
+ if (!nfs_mount_state_error_should_restart(error))
+ printf("nfs_vnop_mmap: close of previous mmap mode failed: %d\n", error);
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_mmap: update, close error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ goto out;
+ }
+ nofp->nof_mmap_access = nofp->nof_mmap_deny = 0;
+ }
+
+ if (accessMode == NFS_OPEN_SHARE_ACCESS_BOTH) {
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE)
+ nofp->nof_rw++;
+ else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE)
+ nofp->nof_rw_dw++;
+ else /* NFS_OPEN_SHARE_DENY_BOTH */
+ nofp->nof_rw_drw++;
+ } else if (accessMode == NFS_OPEN_SHARE_ACCESS_READ) {
+ if (denyMode == NFS_OPEN_SHARE_DENY_NONE)
+ nofp->nof_r++;
+ else if (denyMode == NFS_OPEN_SHARE_DENY_WRITE)
+ nofp->nof_r_dw++;
+ else /* NFS_OPEN_SHARE_DENY_BOTH */
+ nofp->nof_r_drw++;
+ }
+ nofp->nof_mmap_access = accessMode;
+ nofp->nof_mmap_deny = denyMode;
+ nofp->nof_opencnt++;
+
+out:
+ if (nofp)
+ nfs_open_file_clear_busy(nofp);
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ nofp = NULL;
+ goto restart;
+ }
+ if (noop)
+ nfs_open_owner_rele(noop);
+ return (error);
+}
+
+
+int
+nfs4_vnop_mnomap(
+ struct vnop_mnomap_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vfs_context_t ctx = ap->a_context;
+ vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(vp);
+ struct nfsmount *nmp;
+ struct nfs_open_file *nofp = NULL;
+ int error;
+
+ nmp = VTONMP(vp);
+ if (!nmp)
+ return (ENXIO);
+
+ /* walk all open files and close all mmap opens */
+loop:
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error)
+ return (error);
+ lck_mtx_lock(&np->n_openlock);
+ TAILQ_FOREACH(nofp, &np->n_opens, nof_link) {
+ if (!nofp->nof_mmap_access)
+ continue;
+ lck_mtx_unlock(&np->n_openlock);
+ if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ goto loop;
+ }
+ error = nfs_open_file_set_busy(nofp, vfs_context_thread(ctx));
+ if (error) {
+ lck_mtx_lock(&np->n_openlock);
+ break;
+ }
+ if (nofp->nof_mmap_access) {
+ error = nfs4_close(np, nofp, nofp->nof_mmap_access, nofp->nof_mmap_deny, ctx);
+ if (!nfs_mount_state_error_should_restart(error)) {
+ if (error) /* not a state-operation-restarting error, so just clear the access */
+ printf("nfs_vnop_mnomap: close of mmap mode failed: %d\n", error);
+ nofp->nof_mmap_access = nofp->nof_mmap_deny = 0;
+ }
+ if (error) {
+ const char *vname = vnode_getname(NFSTOV(np));
+ printf("nfs_vnop_mnomap: error %d, %s\n", error, vname);
+ vnode_putname(vname);
+ }
+ }
+ nfs_open_file_clear_busy(nofp);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto loop;
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_mount_state_in_use_end(nmp, error);
+ return (error);
+}
+
+/*
+ * Search a node's lock owner list for the owner for this process.
+ * If not found and "alloc" is set, then allocate a new one.
+ */
+struct nfs_lock_owner *
+nfs_lock_owner_find(nfsnode_t np, proc_t p, int alloc)
+{
+ pid_t pid = proc_pid(p);
+ struct nfs_lock_owner *nlop, *newnlop = NULL;
+
+tryagain:
+ lck_mtx_lock(&np->n_openlock);
+ TAILQ_FOREACH(nlop, &np->n_lock_owners, nlo_link) {
+ if (nlop->nlo_pid != pid)
+ continue;
+ if (timevalcmp(&nlop->nlo_pid_start, &p->p_start, ==))
+ break;
+ /* stale lock owner... reuse it if we can */
+ if (nlop->nlo_refcnt) {
+ TAILQ_REMOVE(&np->n_lock_owners, nlop, nlo_link);
+ nlop->nlo_flags &= ~NFS_LOCK_OWNER_LINK;
+ lck_mtx_unlock(&np->n_openlock);
+ goto tryagain;
+ }
+ nlop->nlo_pid_start = p->p_start;
+ nlop->nlo_seqid = 0;
+ nlop->nlo_stategenid = 0;
+ break;
+ }
+
+ if (!nlop && !newnlop && alloc) {
+ lck_mtx_unlock(&np->n_openlock);
+ MALLOC(newnlop, struct nfs_lock_owner *, sizeof(struct nfs_lock_owner), M_TEMP, M_WAITOK);
+ if (!newnlop)
+ return (NULL);
+ bzero(newnlop, sizeof(*newnlop));
+ lck_mtx_init(&newnlop->nlo_lock, nfs_open_grp, LCK_ATTR_NULL);
+ newnlop->nlo_pid = pid;
+ newnlop->nlo_pid_start = p->p_start;
+ newnlop->nlo_name = OSAddAtomic(1, &nfs_lock_owner_seqnum);
+ TAILQ_INIT(&newnlop->nlo_locks);
+ goto tryagain;
+ }
+ if (!nlop && newnlop) {
+ newnlop->nlo_flags |= NFS_LOCK_OWNER_LINK;
+ TAILQ_INSERT_HEAD(&np->n_lock_owners, newnlop, nlo_link);
+ nlop = newnlop;
+ }
+ lck_mtx_unlock(&np->n_openlock);
+
+ if (newnlop && (nlop != newnlop))
+ nfs_lock_owner_destroy(newnlop);
+
+ if (nlop)
+ nfs_lock_owner_ref(nlop);
+
+ return (nlop);
+}
+
+/*
+ * destroy a lock owner that's no longer needed
+ */
+void
+nfs_lock_owner_destroy(struct nfs_lock_owner *nlop)
+{
+ if (nlop->nlo_open_owner) {
+ nfs_open_owner_rele(nlop->nlo_open_owner);
+ nlop->nlo_open_owner = NULL;
+ }
+ lck_mtx_destroy(&nlop->nlo_lock, nfs_open_grp);
+ FREE(nlop, M_TEMP);
+}
+
+/*
+ * acquire a reference count on a lock owner
+ */
+void
+nfs_lock_owner_ref(struct nfs_lock_owner *nlop)
+{
+ lck_mtx_lock(&nlop->nlo_lock);
+ nlop->nlo_refcnt++;
+ lck_mtx_unlock(&nlop->nlo_lock);
+}
+
+/*
+ * drop a reference count on a lock owner and destroy it if
+ * it is no longer referenced and no longer on the mount's list.
+ */
+void
+nfs_lock_owner_rele(struct nfs_lock_owner *nlop)
+{
+ lck_mtx_lock(&nlop->nlo_lock);
+ if (nlop->nlo_refcnt < 1)
+ panic("nfs_lock_owner_rele: no refcnt");
+ nlop->nlo_refcnt--;
+ if (!nlop->nlo_refcnt && (nlop->nlo_flags & NFS_LOCK_OWNER_BUSY))
+ panic("nfs_lock_owner_rele: busy");
+ /* XXX we may potentially want to clean up idle/unused lock owner structures */
+ if (nlop->nlo_refcnt || (nlop->nlo_flags & NFS_LOCK_OWNER_LINK)) {
+ lck_mtx_unlock(&nlop->nlo_lock);
+ return;
+ }
+ /* owner is no longer referenced or linked to mount, so destroy it */
+ lck_mtx_unlock(&nlop->nlo_lock);
+ nfs_lock_owner_destroy(nlop);
+}
+
+/*
+ * Mark a lock owner as busy because we are about to
+ * start an operation that uses and updates lock owner state.
+ */
+int
+nfs_lock_owner_set_busy(struct nfs_lock_owner *nlop, thread_t thd)
+{
+ struct nfsmount *nmp;
+ struct timespec ts = {2, 0};
+ int error = 0, slpflag;
+
+ nmp = nlop->nlo_open_owner->noo_mount;
+ if (!nmp)
+ return (ENXIO);
+ slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
+
+ lck_mtx_lock(&nlop->nlo_lock);
+ while (nlop->nlo_flags & NFS_LOCK_OWNER_BUSY) {
+ if ((error = nfs_sigintr(nmp, NULL, thd, 0)))
+ break;
+ nlop->nlo_flags |= NFS_LOCK_OWNER_WANT;
+ msleep(nlop, &nlop->nlo_lock, slpflag, "nfs_lock_owner_set_busy", &ts);
+ }
+ if (!error)
+ nlop->nlo_flags |= NFS_LOCK_OWNER_BUSY;
+ lck_mtx_unlock(&nlop->nlo_lock);
+
+ return (error);
+}
+
+/*
+ * Clear the busy flag on a lock owner and wake up anyone waiting
+ * to mark it busy.
+ */
+void
+nfs_lock_owner_clear_busy(struct nfs_lock_owner *nlop)
+{
+ int wanted;
+
+ lck_mtx_lock(&nlop->nlo_lock);
+ if (!(nlop->nlo_flags & NFS_LOCK_OWNER_BUSY))
+ panic("nfs_lock_owner_clear_busy");
+ wanted = (nlop->nlo_flags & NFS_LOCK_OWNER_WANT);
+ nlop->nlo_flags &= ~(NFS_LOCK_OWNER_BUSY|NFS_LOCK_OWNER_WANT);
+ lck_mtx_unlock(&nlop->nlo_lock);
+ if (wanted)
+ wakeup(nlop);
+}
+
+/*
+ * Insert a held lock into a lock owner's sorted list.
+ * (flock locks are always inserted at the head the list)
+ */
+void
+nfs_lock_owner_insert_held_lock(struct nfs_lock_owner *nlop, struct nfs_file_lock *newnflp)
+{
+ struct nfs_file_lock *nflp;
+
+ /* insert new lock in lock owner's held lock list */
+ lck_mtx_lock(&nlop->nlo_lock);
+ if ((newnflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_FLOCK) {
+ TAILQ_INSERT_HEAD(&nlop->nlo_locks, newnflp, nfl_lolink);
+ } else {
+ TAILQ_FOREACH(nflp, &nlop->nlo_locks, nfl_lolink) {
+ if (newnflp->nfl_start < nflp->nfl_start)
+ break;
+ }
+ if (nflp)
+ TAILQ_INSERT_BEFORE(nflp, newnflp, nfl_lolink);
+ else
+ TAILQ_INSERT_TAIL(&nlop->nlo_locks, newnflp, nfl_lolink);
+ }
+ lck_mtx_unlock(&nlop->nlo_lock);
+}
+
+/*
+ * Get a file lock structure for this lock owner.
+ */
+struct nfs_file_lock *
+nfs_file_lock_alloc(struct nfs_lock_owner *nlop)
+{
+ struct nfs_file_lock *nflp = NULL;
+
+ lck_mtx_lock(&nlop->nlo_lock);
+ if (!nlop->nlo_alock.nfl_owner) {
+ nflp = &nlop->nlo_alock;
+ nflp->nfl_owner = nlop;
+ }
+ lck_mtx_unlock(&nlop->nlo_lock);
+ if (!nflp) {
+ MALLOC(nflp, struct nfs_file_lock *, sizeof(struct nfs_file_lock), M_TEMP, M_WAITOK);
+ if (!nflp)
+ return (NULL);
+ bzero(nflp, sizeof(*nflp));
+ nflp->nfl_flags |= NFS_FILE_LOCK_ALLOC;
+ nflp->nfl_owner = nlop;
+ }
+ nfs_lock_owner_ref(nlop);
+ return (nflp);
+}
+
+/*
+ * destroy the given NFS file lock structure
+ */
+void
+nfs_file_lock_destroy(struct nfs_file_lock *nflp)
+{
+ struct nfs_lock_owner *nlop = nflp->nfl_owner;
+
+ if (nflp->nfl_flags & NFS_FILE_LOCK_ALLOC) {
+ nflp->nfl_owner = NULL;
+ FREE(nflp, M_TEMP);
+ } else {
+ lck_mtx_lock(&nlop->nlo_lock);
+ bzero(nflp, sizeof(nflp));
+ lck_mtx_unlock(&nlop->nlo_lock);
+ }
+ nfs_lock_owner_rele(nlop);
+}
+
+/*
+ * Check if one file lock conflicts with another.
+ * (nflp1 is the new lock. nflp2 is the existing lock.)
+ */
+int
+nfs_file_lock_conflict(struct nfs_file_lock *nflp1, struct nfs_file_lock *nflp2, int *willsplit)
+{
+ /* no conflict if lock is dead */
+ if ((nflp1->nfl_flags & NFS_FILE_LOCK_DEAD) || (nflp2->nfl_flags & NFS_FILE_LOCK_DEAD))
+ return (0);
+ /* no conflict if it's ours - unless the lock style doesn't match */
+ if ((nflp1->nfl_owner == nflp2->nfl_owner) &&
+ ((nflp1->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == (nflp2->nfl_flags & NFS_FILE_LOCK_STYLE_MASK))) {
+ if (willsplit && (nflp1->nfl_type != nflp2->nfl_type) &&
+ (nflp1->nfl_start > nflp2->nfl_start) &&
+ (nflp1->nfl_end < nflp2->nfl_end))
+ *willsplit = 1;
+ return (0);
+ }
+ /* no conflict if ranges don't overlap */
+ if ((nflp1->nfl_start > nflp2->nfl_end) || (nflp1->nfl_end < nflp2->nfl_start))
+ return (0);
+ /* no conflict if neither lock is exclusive */
+ if ((nflp1->nfl_type != F_WRLCK) && (nflp2->nfl_type != F_WRLCK))
+ return (0);
+ /* conflict */
+ return (1);
+}
+
+/*
+ * Send an NFSv4 LOCK RPC to the server.
+ */
+int
+nfs4_lock_rpc(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ struct nfs_file_lock *nflp,
+ int reclaim,
+ thread_t thd,
+ kauth_cred_t cred)
+{
+ struct nfs_lock_owner *nlop = nflp->nfl_owner;
+ struct nfsmount *nmp;
+ struct nfsm_chain nmreq, nmrep;
+ uint64_t xid;
+ uint32_t locktype;
+ int error = 0, lockerror = ENOENT, newlocker, numops, status;
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+
+ newlocker = (nlop->nlo_stategenid != nmp->nm_stategenid);
+ locktype = (nflp->nfl_flags & NFS_FILE_LOCK_WAIT) ?
+ ((nflp->nfl_type == F_WRLCK) ?
+ NFS_LOCK_TYPE_WRITEW :
+ NFS_LOCK_TYPE_READW) :
+ ((nflp->nfl_type == F_WRLCK) ?
+ NFS_LOCK_TYPE_WRITE :
+ NFS_LOCK_TYPE_READ);
+ if (newlocker) {
+ error = nfs_open_file_set_busy(nofp, thd);
+ if (error)
+ return (error);
+ error = nfs_open_owner_set_busy(nofp->nof_owner, thd);
+ if (error) {
+ nfs_open_file_clear_busy(nofp);
+ return (error);
+ }
+ if (!nlop->nlo_open_owner) {
+ nfs_open_owner_ref(nofp->nof_owner);
+ nlop->nlo_open_owner = nofp->nof_owner;
+ }
+ }
+ error = nfs_lock_owner_set_busy(nlop, thd);
+ if (error) {
+ if (newlocker) {
+ nfs_open_owner_clear_busy(nofp->nof_owner);
+ nfs_open_file_clear_busy(nofp);
+ }
+ return (error);
+ }
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ // PUTFH, GETATTR, LOCK
+ numops = 3;
+ nfsm_chain_build_alloc_init(error, &nmreq, 33 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "lock", numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, np->n_fhp, np->n_fhsize);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ nfsm_chain_add_bitmap_masked(error, &nmreq, nfs_getattr_bitmap,
+ NFS_ATTR_BITMAP_LEN, nmp->nm_fsattr.nfsa_supp_attr);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_LOCK);
+ nfsm_chain_add_32(error, &nmreq, locktype);
+ nfsm_chain_add_32(error, &nmreq, reclaim);
+ nfsm_chain_add_64(error, &nmreq, nflp->nfl_start);
+ nfsm_chain_add_64(error, &nmreq, NFS_LOCK_LENGTH(nflp->nfl_start, nflp->nfl_end));
+ nfsm_chain_add_32(error, &nmreq, newlocker);
+ if (newlocker) {
+ nfsm_chain_add_32(error, &nmreq, nofp->nof_owner->noo_seqid);
+ nfsm_chain_add_stateid(error, &nmreq, &nofp->nof_stateid);
+ nfsm_chain_add_32(error, &nmreq, nlop->nlo_seqid);
+ nfsm_chain_add_lock_owner4(error, &nmreq, nmp, nlop);
+ } else {
+ nfsm_chain_add_stateid(error, &nmreq, &nlop->nlo_stateid);
+ nfsm_chain_add_32(error, &nmreq, nlop->nlo_seqid);
+ }
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+
+ error = nfs_request2(np, NULL, &nmreq, NFSPROC4_COMPOUND, thd, cred, (reclaim ? R_RECOVER : 0), &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTFH);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsm_chain_loadattr(error, &nmrep, np, NFS_VER4, NULL, &xid);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_LOCK);
+ nfs_owner_seqid_increment(newlocker ? nofp->nof_owner : NULL, nlop, error);
+ nfsm_chain_get_stateid(error, &nmrep, &nlop->nlo_stateid);
+
+ /* Update the lock owner's stategenid once it appears the server has state for it. */
+ /* We determine this by noting the request was successful (we got a stateid). */
+ if (newlocker && !error)
+ nlop->nlo_stategenid = nmp->nm_stategenid;
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfs_lock_owner_clear_busy(nlop);
+ if (newlocker) {
+ nfs_open_owner_clear_busy(nofp->nof_owner);
+ nfs_open_file_clear_busy(nofp);
+ }
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/*
+ * Send an NFSv4 LOCKU RPC to the server.
+ */
+int
+nfs4_unlock_rpc(
+ nfsnode_t np,
+ struct nfs_lock_owner *nlop,
+ int type,
+ uint64_t start,
+ uint64_t end,
+ vfs_context_t ctx)
+{
+ struct nfsmount *nmp;
+ struct nfsm_chain nmreq, nmrep;
+ uint64_t xid;
+ int error = 0, lockerror = ENOENT, numops, status;
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+
+ error = nfs_lock_owner_set_busy(nlop, vfs_context_thread(ctx));
+ if (error)
+ return (error);
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ // PUTFH, GETATTR, LOCKU
+ numops = 3;
+ nfsm_chain_build_alloc_init(error, &nmreq, 26 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "unlock", numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, np->n_fhp, np->n_fhsize);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ nfsm_chain_add_bitmap_masked(error, &nmreq, nfs_getattr_bitmap,
+ NFS_ATTR_BITMAP_LEN, nmp->nm_fsattr.nfsa_supp_attr);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_LOCKU);
+ nfsm_chain_add_32(error, &nmreq, (type == F_WRLCK) ? NFS_LOCK_TYPE_WRITE : NFS_LOCK_TYPE_READ);
+ nfsm_chain_add_32(error, &nmreq, nlop->nlo_seqid);
+ nfsm_chain_add_stateid(error, &nmreq, &nlop->nlo_stateid);
+ nfsm_chain_add_64(error, &nmreq, start);
+ nfsm_chain_add_64(error, &nmreq, NFS_LOCK_LENGTH(start, end));
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+
+ error = nfs_request(np, NULL, &nmreq, NFSPROC4_COMPOUND, ctx, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTFH);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsm_chain_loadattr(error, &nmrep, np, NFS_VER4, NULL, &xid);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_LOCKU);
+ nfs_owner_seqid_increment(NULL, nlop, error);
+ nfsm_chain_get_stateid(error, &nmrep, &nlop->nlo_stateid);
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfs_lock_owner_clear_busy(nlop);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/*
+ * Check for any conflicts with the given lock.
+ *
+ * Checking for a lock doesn't require the file to be opened.
+ * So we skip all the open owner, open file, lock owner work
+ * and just check for a conflicting lock.
+ */
+int
+nfs4_getlock(
+ nfsnode_t np,
+ struct nfs_lock_owner *nlop,
+ struct flock *fl,
+ uint64_t start,
+ uint64_t end,
+ vfs_context_t ctx)
+{
+ struct nfsmount *nmp;
+ struct nfs_file_lock *nflp;
+ struct nfsm_chain nmreq, nmrep;
+ uint64_t xid, val64 = 0;
+ uint32_t val = 0;
+ int error = 0, lockerror = ENOENT, numops, status;
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+
+ lck_mtx_lock(&np->n_openlock);
+ /* scan currently held locks for conflict */
+ TAILQ_FOREACH(nflp, &np->n_locks, nfl_link) {
+ if (nflp->nfl_flags & NFS_FILE_LOCK_BLOCKED)
+ continue;
+ if ((start <= nflp->nfl_end) && (end >= nflp->nfl_start) &&
+ ((fl->l_type == F_WRLCK) || (nflp->nfl_type == F_WRLCK)))
+ break;
+ }
+ if (nflp) {
+ /* found a conflicting lock */
+ fl->l_type = nflp->nfl_type;
+ fl->l_pid = (nflp->nfl_flags & NFS_FILE_LOCK_STYLE_FLOCK) ? -1 : nflp->nfl_owner->nlo_pid;
+ fl->l_start = nflp->nfl_start;
+ fl->l_len = NFS_FLOCK_LENGTH(nflp->nfl_start, nflp->nfl_end);
+ fl->l_whence = SEEK_SET;
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ if (nflp)
+ return (0);
+
+ /* no conflict found locally, so ask the server */
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ // PUTFH, GETATTR, LOCKT
+ numops = 3;
+ nfsm_chain_build_alloc_init(error, &nmreq, 26 * NFSX_UNSIGNED);
+ nfsm_chain_add_compound_header(error, &nmreq, "locktest", numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, NFS_VER4, np->n_fhp, np->n_fhsize);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ nfsm_chain_add_bitmap_masked(error, &nmreq, nfs_getattr_bitmap,
+ NFS_ATTR_BITMAP_LEN, nmp->nm_fsattr.nfsa_supp_attr);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_LOCKT);
+ nfsm_chain_add_32(error, &nmreq, (fl->l_type == F_WRLCK) ? NFS_LOCK_TYPE_WRITE : NFS_LOCK_TYPE_READ);
+ nfsm_chain_add_64(error, &nmreq, start);
+ nfsm_chain_add_64(error, &nmreq, NFS_LOCK_LENGTH(start, end));
+ nfsm_chain_add_lock_owner4(error, &nmreq, nmp, nlop);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ nfsmout_if(error);
+
+ error = nfs_request(np, NULL, &nmreq, NFSPROC4_COMPOUND, ctx, &nmrep, &xid, &status);
+
+ if ((lockerror = nfs_node_lock(np)))
+ error = lockerror;
+ nfsm_chain_skip_tag(error, &nmrep);
+ nfsm_chain_get_32(error, &nmrep, numops);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_PUTFH);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsm_chain_loadattr(error, &nmrep, np, NFS_VER4, NULL, &xid);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_LOCKT);
+ if (error == NFSERR_DENIED) {
+ error = 0;
+ nfsm_chain_get_64(error, &nmrep, fl->l_start);
+ nfsm_chain_get_64(error, &nmrep, val64);
+ fl->l_len = (val64 == UINT64_MAX) ? 0 : val64;
+ nfsm_chain_get_32(error, &nmrep, val);
+ fl->l_type = (val == NFS_LOCK_TYPE_WRITE) ? F_WRLCK : F_RDLCK;
+ fl->l_pid = 0;
+ fl->l_whence = SEEK_SET;
+ } else if (!error) {
+ fl->l_type = F_UNLCK;
+ }
+nfsmout:
+ if (!lockerror)
+ nfs_node_unlock(np);
+ nfsm_chain_cleanup(&nmreq);
+ nfsm_chain_cleanup(&nmrep);
+ return (error);
+}
+
+/*
+ * Acquire a file lock for the given range.
+ *
+ * Add the lock (request) to the lock queue.
+ * Scan the lock queue for any conflicting locks.
+ * If a conflict is found, block or return an error.
+ * Once end of queue is reached, send request to the server.
+ * If the server grants the lock, scan the lock queue and
+ * update any existing locks. Then (optionally) scan the
+ * queue again to coalesce any locks adjacent to the new one.
+ */
+int
+nfs4_setlock(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ struct nfs_lock_owner *nlop,
+ int op,
+ uint64_t start,
+ uint64_t end,
+ int style,
+ short type,
+ vfs_context_t ctx)
+{
+ struct nfsmount *nmp;
+ struct nfs_file_lock *newnflp, *nflp, *nflp2 = NULL, *nextnflp, *flocknflp = NULL;
+ struct nfs_file_lock *coalnflp;
+ int error = 0, error2, willsplit = 0, delay, slpflag, busy = 0, inuse = 0, restart, inqueue = 0;
+ struct timespec ts = {1, 0};
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+ slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
+
+ /* allocate a new lock */
+ newnflp = nfs_file_lock_alloc(nlop);
+ if (!newnflp)
+ return (ENOLCK);
+ newnflp->nfl_start = start;
+ newnflp->nfl_end = end;
+ newnflp->nfl_type = type;
+ if (op == F_SETLKW)
+ newnflp->nfl_flags |= NFS_FILE_LOCK_WAIT;
+ newnflp->nfl_flags |= style;
+ newnflp->nfl_flags |= NFS_FILE_LOCK_BLOCKED;
+
+ if ((style == NFS_FILE_LOCK_STYLE_FLOCK) && (type == F_WRLCK)) {
+ /*
+ * For exclusive flock-style locks, if we block waiting for the
+ * lock, we need to first release any currently held shared
+ * flock-style lock. So, the first thing we do is check if we
+ * have a shared flock-style lock.
+ */
+ nflp = TAILQ_FIRST(&nlop->nlo_locks);
+ if (nflp && ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) != NFS_FILE_LOCK_STYLE_FLOCK))
+ nflp = NULL;
+ if (nflp && (nflp->nfl_type != F_RDLCK))
+ nflp = NULL;
+ flocknflp = nflp;
+ }
+
+restart:
+ restart = 0;
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error)
+ goto error_out;
+ inuse = 1;
+ if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ inuse = 0;
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ goto restart;
+ }
+
+ lck_mtx_lock(&np->n_openlock);
+ if (!inqueue) {
+ /* insert new lock at beginning of list */
+ TAILQ_INSERT_HEAD(&np->n_locks, newnflp, nfl_link);
+ inqueue = 1;
+ }
+
+ /* scan current list of locks (held and pending) for conflicts */
+ for (nflp = TAILQ_NEXT(newnflp, nfl_link); nflp; nflp = TAILQ_NEXT(nflp, nfl_link)) {
+ if (!nfs_file_lock_conflict(newnflp, nflp, &willsplit))
+ continue;
+ /* Conflict */
+ if (!(newnflp->nfl_flags & NFS_FILE_LOCK_WAIT)) {
+ error = EAGAIN;
+ break;
+ }
+ /* Block until this lock is no longer held. */
+ if (nflp->nfl_blockcnt == UINT_MAX) {
+ error = ENOLCK;
+ break;
+ }
+ nflp->nfl_blockcnt++;
+ do {
+ if (flocknflp) {
+ /* release any currently held shared lock before sleeping */
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_mount_state_in_use_end(nmp, 0);
+ inuse = 0;
+ error = nfs4_unlock(np, nofp, nlop, 0, UINT64_MAX, NFS_FILE_LOCK_STYLE_FLOCK, ctx);
+ flocknflp = NULL;
+ if (!error)
+ error = nfs_mount_state_in_use_start(nmp);
+ if (error) {
+ lck_mtx_lock(&np->n_openlock);
+ break;
+ }
+ inuse = 1;
+ lck_mtx_lock(&np->n_openlock);
+ /* no need to block/sleep if the conflict is gone */
+ if (!nfs_file_lock_conflict(newnflp, nflp, NULL))
+ break;
+ }
+ msleep(nflp, &np->n_openlock, slpflag, "nfs4_setlock_blocked", &ts);
+ error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0);
+ if (!error && (nmp->nm_state & NFSSTA_RECOVER)) {
+ /* looks like we have a recover pending... restart */
+ restart = 1;
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_mount_state_in_use_end(nmp, 0);
+ inuse = 0;
+ lck_mtx_lock(&np->n_openlock);
+ break;
+ }
+ } while (!error && nfs_file_lock_conflict(newnflp, nflp, NULL));
+ nflp->nfl_blockcnt--;
+ if ((nflp->nfl_flags & NFS_FILE_LOCK_DEAD) && !nflp->nfl_blockcnt) {
+ TAILQ_REMOVE(&np->n_locks, nflp, nfl_link);
+ nfs_file_lock_destroy(nflp);
+ }
+ if (error || restart)
+ break;
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ if (restart)
+ goto restart;
+ if (error)
+ goto error_out;
+
+ if (willsplit) {
+ /*
+ * It looks like this operation is splitting a lock.
+ * We allocate a new lock now so we don't have to worry
+ * about the allocation failing after we've updated some state.
+ */
+ nflp2 = nfs_file_lock_alloc(nlop);
+ if (!nflp2) {
+ error = ENOLCK;
+ goto error_out;
+ }
+ }
+
+ /* once scan for local conflicts is clear, send request to server */
+ if ((error = nfs_open_state_set_busy(np, ctx)))
+ goto error_out;
+ busy = 1;
+ delay = 0;
+ do {
+ error = nfs4_lock_rpc(np, nofp, newnflp, 0, vfs_context_thread(ctx), vfs_context_ucred(ctx));
+ if (!error || ((error != NFSERR_DENIED) && (error != NFSERR_GRACE)))
+ break;
+ /* request was denied due to either conflict or grace period */
+ if ((error != NFSERR_GRACE) && !(newnflp->nfl_flags & NFS_FILE_LOCK_WAIT)) {
+ error = EAGAIN;
+ break;
+ }
+ if (flocknflp) {
+ /* release any currently held shared lock before sleeping */
+ nfs_open_state_clear_busy(np);
+ busy = 0;
+ nfs_mount_state_in_use_end(nmp, 0);
+ inuse = 0;
+ error2 = nfs4_unlock(np, nofp, nlop, 0, UINT64_MAX, NFS_FILE_LOCK_STYLE_FLOCK, ctx);
+ flocknflp = NULL;
+ if (!error2)
+ error2 = nfs_mount_state_in_use_start(nmp);
+ if (!error2) {
+ inuse = 1;
+ error2 = nfs_open_state_set_busy(np, ctx);
+ }
+ if (error2) {
+ error = error2;
+ break;
+ }
+ busy = 1;
+ }
+ /* wait a little bit and send the request again */
+ if (error == NFSERR_GRACE)
+ delay = 4;
+ if (delay < 4)
+ delay++;
+ tsleep(newnflp, slpflag, "nfs4_setlock_delay", delay * (hz/2));
+ error = nfs_sigintr(NFSTONMP(np), NULL, vfs_context_thread(ctx), 0);
+ if (!error && (nmp->nm_state & NFSSTA_RECOVER)) {
+ /* looks like we have a recover pending... restart */
+ nfs_open_state_clear_busy(np);
+ busy = 0;
+ nfs_mount_state_in_use_end(nmp, 0);
+ inuse = 0;
+ goto restart;
+ }
+ } while (!error);
+
+error_out:
+ if (nfs_mount_state_error_should_restart(error)) {
+ /* looks like we need to restart this operation */
+ if (busy) {
+ nfs_open_state_clear_busy(np);
+ busy = 0;
+ }
+ if (inuse) {
+ nfs_mount_state_in_use_end(nmp, error);
+ inuse = 0;
+ }
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ newnflp->nfl_flags &= ~NFS_FILE_LOCK_BLOCKED;
+ if (error) {
+ newnflp->nfl_flags |= NFS_FILE_LOCK_DEAD;
+ if (newnflp->nfl_blockcnt) {
+ /* wake up anyone blocked on this lock */
+ wakeup(newnflp);
+ } else {
+ /* remove newnflp from lock list and destroy */
+ TAILQ_REMOVE(&np->n_locks, newnflp, nfl_link);
+ nfs_file_lock_destroy(newnflp);
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ if (busy)
+ nfs_open_state_clear_busy(np);
+ if (inuse)
+ nfs_mount_state_in_use_end(nmp, error);
+ if (nflp2)
+ nfs_file_lock_destroy(nflp2);
+ return (error);
+ }
+
+ /* server granted the lock */
+
+ /*
+ * Scan for locks to update.
+ *
+ * Locks completely covered are killed.
+ * At most two locks may need to be clipped.
+ * It's possible that a single lock may need to be split.
+ */
+ TAILQ_FOREACH_SAFE(nflp, &np->n_locks, nfl_link, nextnflp) {
+ if (nflp == newnflp)
+ continue;
+ if (nflp->nfl_flags & (NFS_FILE_LOCK_BLOCKED|NFS_FILE_LOCK_DEAD))
+ continue;
+ if (nflp->nfl_owner != nlop)
+ continue;
+ if ((newnflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) != (nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK))
+ continue;
+ if ((newnflp->nfl_start > nflp->nfl_end) || (newnflp->nfl_end < nflp->nfl_start))
+ continue;
+ /* here's one to update */
+ if ((newnflp->nfl_start <= nflp->nfl_start) && (newnflp->nfl_end >= nflp->nfl_end)) {
+ /* The entire lock is being replaced. */
+ nflp->nfl_flags |= NFS_FILE_LOCK_DEAD;
+ lck_mtx_lock(&nlop->nlo_lock);
+ TAILQ_REMOVE(&nlop->nlo_locks, nflp, nfl_lolink);
+ lck_mtx_unlock(&nlop->nlo_lock);
+ /* lock will be destroyed below, if no waiters */
+ } else if ((newnflp->nfl_start > nflp->nfl_start) && (newnflp->nfl_end < nflp->nfl_end)) {
+ /* We're replacing a range in the middle of a lock. */
+ /* The current lock will be split into two locks. */
+ /* Update locks and insert new lock after current lock. */
+ nflp2->nfl_flags |= (nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK);
+ nflp2->nfl_type = nflp->nfl_type;
+ nflp2->nfl_start = newnflp->nfl_end + 1;
+ nflp2->nfl_end = nflp->nfl_end;
+ nflp->nfl_end = newnflp->nfl_start - 1;
+ TAILQ_INSERT_AFTER(&np->n_locks, nflp, nflp2, nfl_link);
+ nfs_lock_owner_insert_held_lock(nlop, nflp2);
+ nextnflp = nflp2;
+ nflp2 = NULL;
+ } else if (newnflp->nfl_start > nflp->nfl_start) {
+ /* We're replacing the end of a lock. */
+ nflp->nfl_end = newnflp->nfl_start - 1;
+ } else if (newnflp->nfl_end < nflp->nfl_end) {
+ /* We're replacing the start of a lock. */
+ nflp->nfl_start = newnflp->nfl_end + 1;
+ }
+ if (nflp->nfl_blockcnt) {
+ /* wake up anyone blocked on this lock */
+ wakeup(nflp);
+ } else if (nflp->nfl_flags & NFS_FILE_LOCK_DEAD) {
+ /* remove nflp from lock list and destroy */
+ TAILQ_REMOVE(&np->n_locks, nflp, nfl_link);
+ nfs_file_lock_destroy(nflp);
+ }
+ }
+
+ nfs_lock_owner_insert_held_lock(nlop, newnflp);
+
+ /*
+ * POSIX locks should be coalesced when possible.
+ */
+ if ((style == NFS_FILE_LOCK_STYLE_POSIX) && (nofp->nof_flags & NFS_OPEN_FILE_POSIXLOCK)) {
+ /*
+ * Walk through the lock queue and check each of our held locks with
+ * the previous and next locks in the lock owner's "held lock list".
+ * If the two locks can be coalesced, we merge the current lock into
+ * the other (previous or next) lock. Merging this way makes sure that
+ * lock ranges are always merged forward in the lock queue. This is
+ * important because anyone blocked on the lock being "merged away"
+ * will still need to block on that range and it will simply continue
+ * checking locks that are further down the list.
+ */
+ TAILQ_FOREACH_SAFE(nflp, &np->n_locks, nfl_link, nextnflp) {
+ if (nflp->nfl_flags & (NFS_FILE_LOCK_BLOCKED|NFS_FILE_LOCK_DEAD))
+ continue;
+ if (nflp->nfl_owner != nlop)
+ continue;
+ if ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) != NFS_FILE_LOCK_STYLE_POSIX)
+ continue;
+ if (((coalnflp = TAILQ_PREV(nflp, nfs_file_lock_queue, nfl_lolink))) &&
+ ((coalnflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_POSIX) &&
+ (coalnflp->nfl_type == nflp->nfl_type) &&
+ (coalnflp->nfl_end == (nflp->nfl_start - 1))) {
+ coalnflp->nfl_end = nflp->nfl_end;
+ nflp->nfl_flags |= NFS_FILE_LOCK_DEAD;
+ lck_mtx_lock(&nlop->nlo_lock);
+ TAILQ_REMOVE(&nlop->nlo_locks, nflp, nfl_lolink);
+ lck_mtx_unlock(&nlop->nlo_lock);
+ } else if (((coalnflp = TAILQ_NEXT(nflp, nfl_lolink))) &&
+ ((coalnflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_POSIX) &&
+ (coalnflp->nfl_type == nflp->nfl_type) &&
+ (coalnflp->nfl_start == (nflp->nfl_end + 1))) {
+ coalnflp->nfl_start = nflp->nfl_start;
+ nflp->nfl_flags |= NFS_FILE_LOCK_DEAD;
+ lck_mtx_lock(&nlop->nlo_lock);
+ TAILQ_REMOVE(&nlop->nlo_locks, nflp, nfl_lolink);
+ lck_mtx_unlock(&nlop->nlo_lock);
+ }
+ if (!(nflp->nfl_flags & NFS_FILE_LOCK_DEAD))
+ continue;
+ if (nflp->nfl_blockcnt) {
+ /* wake up anyone blocked on this lock */
+ wakeup(nflp);
+ } else {
+ /* remove nflp from lock list and destroy */
+ TAILQ_REMOVE(&np->n_locks, nflp, nfl_link);
+ nfs_file_lock_destroy(nflp);
+ }
+ }
+ }
+
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+
+ if (nflp2)
+ nfs_file_lock_destroy(nflp2);
+ return (error);
+}
+
+int
+nfs4_unlock(
+ nfsnode_t np,
+ struct nfs_open_file *nofp,
+ struct nfs_lock_owner *nlop,
+ uint64_t start,
+ uint64_t end,
+ int style,
+ vfs_context_t ctx)
+{
+ struct nfsmount *nmp;
+ struct nfs_file_lock *nflp, *nextnflp, *newnflp = NULL;
+ int error = 0, willsplit = 0, send_unlock_rpcs = 1;
+
+ nmp = NFSTONMP(np);
+ if (!nmp)
+ return (ENXIO);
+
+restart:
+ if ((error = nfs_mount_state_in_use_start(nmp)))
+ return (error);
+ if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN) {
+ nfs_mount_state_in_use_end(nmp, 0);
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ goto restart;
+ }
+ if ((error = nfs_open_state_set_busy(np, ctx))) {
+ nfs_mount_state_in_use_end(nmp, error);
+ return (error);
+ }
+
+ lck_mtx_lock(&np->n_openlock);
+ if ((start > 0) && (end < UINT64_MAX) && !willsplit) {
+ /*
+ * We may need to allocate a new lock if an existing lock gets split.
+ * So, we first scan the list to check for a split, and if there's
+ * going to be one, we'll allocate one now.
+ */
+ TAILQ_FOREACH_SAFE(nflp, &np->n_locks, nfl_link, nextnflp) {
+ if (nflp->nfl_flags & (NFS_FILE_LOCK_BLOCKED|NFS_FILE_LOCK_DEAD))
+ continue;
+ if (nflp->nfl_owner != nlop)
+ continue;
+ if ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) != style)
+ continue;
+ if ((start > nflp->nfl_end) || (end < nflp->nfl_start))
+ continue;
+ if ((start > nflp->nfl_start) && (end < nflp->nfl_end)) {
+ willsplit = 1;
+ break;
+ }
+ }
+ if (willsplit) {
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, 0);
+ newnflp = nfs_file_lock_alloc(nlop);
+ if (!newnflp)
+ return (ENOMEM);
+ goto restart;
+ }
+ }
+
+ /*
+ * Free all of our locks in the given range.
+ *
+ * Note that this process requires sending requests to the server.
+ * Because of this, we will release the n_openlock while performing
+ * the unlock RPCs. The N_OPENBUSY state keeps the state of *held*
+ * locks from changing underneath us. However, other entries in the
+ * list may be removed. So we need to be careful walking the list.
+ */
+
+ /*
+ * Don't unlock ranges that are held by other-style locks.
+ * If style is posix, don't send any unlock rpcs if flock is held.
+ * If we unlock an flock, don't send unlock rpcs for any posix-style
+ * ranges held - instead send unlocks for the ranges not held.
+ */
+ if ((style == NFS_FILE_LOCK_STYLE_POSIX) &&
+ ((nflp = TAILQ_FIRST(&nlop->nlo_locks))) &&
+ ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_FLOCK))
+ send_unlock_rpcs = 0;
+ if ((style == NFS_FILE_LOCK_STYLE_FLOCK) &&
+ ((nflp = TAILQ_FIRST(&nlop->nlo_locks))) &&
+ ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_FLOCK) &&
+ ((nflp = TAILQ_NEXT(nflp, nfl_lolink))) &&
+ ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_POSIX)) {
+ uint64_t s = 0;
+ int type = TAILQ_FIRST(&nlop->nlo_locks)->nfl_type;
+ while (nflp) {
+ if ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) == NFS_FILE_LOCK_STYLE_POSIX) {
+ /* unlock the range preceding this lock */
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, type, s, nflp->nfl_start-1, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ if (error)
+ goto out;
+ s = nflp->nfl_end+1;
+ }
+ nflp = TAILQ_NEXT(nflp, nfl_lolink);
+ }
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, type, s, end, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ if (error)
+ goto out;
+ send_unlock_rpcs = 0;
+ }
+
+ TAILQ_FOREACH_SAFE(nflp, &np->n_locks, nfl_link, nextnflp) {
+ if (nflp->nfl_flags & (NFS_FILE_LOCK_BLOCKED|NFS_FILE_LOCK_DEAD))
+ continue;
+ if (nflp->nfl_owner != nlop)
+ continue;
+ if ((nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK) != style)
+ continue;
+ if ((start > nflp->nfl_end) || (end < nflp->nfl_start))
+ continue;
+ /* here's one to unlock */
+ if ((start <= nflp->nfl_start) && (end >= nflp->nfl_end)) {
+ /* The entire lock is being unlocked. */
+ if (send_unlock_rpcs) {
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, nflp->nfl_type, nflp->nfl_start, nflp->nfl_end, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ }
+ nextnflp = TAILQ_NEXT(nflp, nfl_link);
+ if (error)
+ break;
+ nflp->nfl_flags |= NFS_FILE_LOCK_DEAD;
+ lck_mtx_lock(&nlop->nlo_lock);
+ TAILQ_REMOVE(&nlop->nlo_locks, nflp, nfl_lolink);
+ lck_mtx_unlock(&nlop->nlo_lock);
+ /* lock will be destroyed below, if no waiters */
+ } else if ((start > nflp->nfl_start) && (end < nflp->nfl_end)) {
+ /* We're unlocking a range in the middle of a lock. */
+ /* The current lock will be split into two locks. */
+ if (send_unlock_rpcs) {
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, nflp->nfl_type, start, end, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ }
+ if (error)
+ break;
+ /* update locks and insert new lock after current lock */
+ newnflp->nfl_flags |= (nflp->nfl_flags & NFS_FILE_LOCK_STYLE_MASK);
+ newnflp->nfl_type = nflp->nfl_type;
+ newnflp->nfl_start = end + 1;
+ newnflp->nfl_end = nflp->nfl_end;
+ nflp->nfl_end = start - 1;
+ TAILQ_INSERT_AFTER(&np->n_locks, nflp, newnflp, nfl_link);
+ nfs_lock_owner_insert_held_lock(nlop, newnflp);
+ nextnflp = newnflp;
+ newnflp = NULL;
+ } else if (start > nflp->nfl_start) {
+ /* We're unlocking the end of a lock. */
+ if (send_unlock_rpcs) {
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, nflp->nfl_type, start, nflp->nfl_end, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ }
+ nextnflp = TAILQ_NEXT(nflp, nfl_link);
+ if (error)
+ break;
+ nflp->nfl_end = start - 1;
+ } else if (end < nflp->nfl_end) {
+ /* We're unlocking the start of a lock. */
+ if (send_unlock_rpcs) {
+ lck_mtx_unlock(&np->n_openlock);
+ error = nfs4_unlock_rpc(np, nlop, nflp->nfl_type, nflp->nfl_start, end, ctx);
+ if (nfs_mount_state_error_should_restart(error)) {
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, error);
+ goto restart;
+ }
+ lck_mtx_lock(&np->n_openlock);
+ }
+ nextnflp = TAILQ_NEXT(nflp, nfl_link);
+ if (error)
+ break;
+ nflp->nfl_start = end + 1;
+ }
+ if (nflp->nfl_blockcnt) {
+ /* wake up anyone blocked on this lock */
+ wakeup(nflp);
+ } else if (nflp->nfl_flags & NFS_FILE_LOCK_DEAD) {
+ /* remove nflp from lock list and destroy */
+ TAILQ_REMOVE(&np->n_locks, nflp, nfl_link);
+ nfs_file_lock_destroy(nflp);
+ }
+ }
+out:
+ lck_mtx_unlock(&np->n_openlock);
+ nfs_open_state_clear_busy(np);
+ nfs_mount_state_in_use_end(nmp, 0);
+
+ if (newnflp)
+ nfs_file_lock_destroy(newnflp);
+ return (error);
+}
+
+/*
+ * NFSv4 advisory file locking
+ */
+int
+nfs4_vnop_advlock(
+ struct vnop_advlock_args /* {
+ struct vnodeop_desc *a_desc;
+ vnode_t a_vp;
+ caddr_t a_id;
+ int a_op;
+ struct flock *a_fl;
+ int a_flags;
+ vfs_context_t a_context;
+ } */ *ap)
+{
+ vnode_t vp = ap->a_vp;
+ nfsnode_t np = VTONFS(ap->a_vp);
+ struct flock *fl = ap->a_fl;
+ int op = ap->a_op;
+ int flags = ap->a_flags;
+ vfs_context_t ctx = ap->a_context;
+ struct nfsmount *nmp;
+ struct nfs_vattr nvattr;
+ struct nfs_open_owner *noop = NULL;
+ struct nfs_open_file *nofp = NULL;
+ struct nfs_lock_owner *nlop = NULL;
+ off_t lstart;
+ uint64_t start, end;
+ int error = 0, modified, style;
+#define OFF_MAX QUAD_MAX
+
+ nmp = VTONMP(ap->a_vp);
+ if (!nmp)
+ return (ENXIO);
+
+ switch (fl->l_whence) {
+ case SEEK_SET:
+ case SEEK_CUR:
+ /*
+ * Caller is responsible for adding any necessary offset
+ * to fl->l_start when SEEK_CUR is used.
+ */
+ lstart = fl->l_start;
+ break;
+ case SEEK_END:
+ /* need to flush, and refetch attributes to make */
+ /* sure we have the correct end of file offset */
+ if ((error = nfs_node_lock(np)))
+ return (error);
+ modified = (np->n_flag & NMODIFIED);
+ nfs_node_unlock(np);
+ if (modified && ((error = nfs_vinvalbuf(vp, V_SAVE, ctx, 1))))
+ return (error);
+ if ((error = nfs_getattr(np, &nvattr, ctx, NGA_UNCACHED)))
+ return (error);
+ nfs_data_lock(np, NFS_DATA_LOCK_SHARED);
+ if ((np->n_size > OFF_MAX) ||
+ ((fl->l_start > 0) && (np->n_size > (u_quad_t)(OFF_MAX - fl->l_start))))
+ error = EOVERFLOW;
+ lstart = np->n_size + fl->l_start;
+ nfs_data_unlock(np);
+ if (error)
+ return (error);
+ break;
+ default:
+ return (EINVAL);
+ }
+ if (lstart < 0)
+ return (EINVAL);
+ start = lstart;
+ if (fl->l_len == 0) {
+ end = UINT64_MAX;
+ } else if (fl->l_len > 0) {
+ if ((fl->l_len - 1) > (OFF_MAX - lstart))
+ return (EOVERFLOW);
+ end = start - 1 + fl->l_len;
+ } else { /* l_len is negative */
+ if ((lstart + fl->l_len) < 0)
+ return (EINVAL);
+ end = start - 1;
+ start += fl->l_len;
+ }
+ if (error)
+ return (error);
+
+ style = (flags & F_FLOCK) ? NFS_FILE_LOCK_STYLE_FLOCK : NFS_FILE_LOCK_STYLE_POSIX;
+ if ((style == NFS_FILE_LOCK_STYLE_FLOCK) && ((start != 0) || (end != UINT64_MAX)))
+ return (EINVAL);
+
+ /* find the lock owner, alloc if not unlock */
+ nlop = nfs_lock_owner_find(np, vfs_context_proc(ctx), (op != F_UNLCK));
+ if (!nlop) {
+ error = (op == F_UNLCK) ? 0 : ENOMEM;
+ if (error)
+ printf("nfs4_vnop_advlock: no lock owner %d\n", error);
+ goto out;
+ }
+
+ if (op == F_GETLK) {
+ error = nfs4_getlock(np, nlop, fl, start, end, ctx);
+ } else {
+ /* find the open owner */
+ noop = nfs_open_owner_find(nmp, vfs_context_ucred(ctx), 0);
+ if (!noop) {
+ printf("nfs4_vnop_advlock: no open owner\n");
+ error = EPERM;
+ goto out;
+ }
+ /* find the open file */
+restart:
+ error = nfs_open_file_find(np, noop, &nofp, 0, 0, 0);
+ if (error)
+ error = EBADF;
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
+ printf("nfs_vnop_advlock: LOST\n");
+ error = EIO;
+ }
+ if (!error && (nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ nfs4_reopen(nofp, vfs_context_thread(ctx));
+ nofp = NULL;
+ goto restart;
+ }
+ if (error) {
+ printf("nfs4_vnop_advlock: no open file %d\n", error);
+ goto out;
+ }
+ if (op == F_UNLCK) {
+ error = nfs4_unlock(np, nofp, nlop, start, end, style, ctx);
+ } else if ((op == F_SETLK) || (op == F_SETLKW)) {
+ if ((op == F_SETLK) && (flags & F_WAIT))
+ op = F_SETLKW;
+ error = nfs4_setlock(np, nofp, nlop, op, start, end, style, fl->l_type, ctx);
+ } else {
+ /* not getlk, unlock or lock? */
+ error = EINVAL;
+ }
+ }
+
+out:
+ if (nlop)
+ nfs_lock_owner_rele(nlop);
+ if (noop)
+ nfs_open_owner_rele(noop);
+ return (error);
+}
+
+/*
+ * Check if an open owner holds any locks on a file.
+ */
+int
+nfs4_check_for_locks(struct nfs_open_owner *noop, struct nfs_open_file *nofp)
+{
+ struct nfs_lock_owner *nlop;
+
+ TAILQ_FOREACH(nlop, &nofp->nof_np->n_lock_owners, nlo_link) {
+ if (nlop->nlo_open_owner != noop)
+ continue;
+ if (!TAILQ_EMPTY(&nlop->nlo_locks))
+ break;
+ }
+ return (nlop ? 1 : 0);
+}
+
+/*
+ * Reopen simple (no deny, no locks) open state that was lost.
+ */
+void
+nfs4_reopen(struct nfs_open_file *nofp, thread_t thd)
+{
+ struct nfs_open_owner *noop = nofp->nof_owner;
+ struct nfsmount *nmp = NFSTONMP(nofp->nof_np);
+ vnode_t vp = NFSTOV(nofp->nof_np);
+ vnode_t dvp = NULL;
+ struct componentname cn;
+ const char *vname = NULL;
+ size_t namelen;
+ char smallname[128];
+ char *filename = NULL;
+ int error = 0, done = 0, slpflag = (nmp->nm_flag & NFSMNT_INT) ? PCATCH : 0;
+ struct timespec ts = { 1, 0 };
+
+ lck_mtx_lock(&nofp->nof_lock);
+ while (nofp->nof_flags & NFS_OPEN_FILE_REOPENING) {
+ if ((error = nfs_sigintr(nmp, NULL, thd, 0)))
+ break;
+ msleep(&nofp->nof_flags, &nofp->nof_lock, slpflag|(PZERO-1), "nfsreopenwait", &ts);
+ }
+ if (!(nofp->nof_flags & NFS_OPEN_FILE_REOPEN)) {
+ lck_mtx_unlock(&nofp->nof_lock);
+ return;
+ }
+ nofp->nof_flags |= NFS_OPEN_FILE_REOPENING;
+ lck_mtx_unlock(&nofp->nof_lock);
+
+ dvp = vnode_getparent(vp);
+ vname = vnode_getname(vp);
+ if (!dvp || !vname) {
+ error = EIO;
+ goto out;
+ }
+ filename = &smallname[0];
+ namelen = snprintf(filename, sizeof(smallname), "%s", vname);
+ if (namelen >= sizeof(smallname)) {
+ namelen++; /* snprintf result doesn't include '\0' */
+ MALLOC(filename, char *, namelen, M_TEMP, M_WAITOK);
+ if (!filename) {
+ error = ENOMEM;
+ goto out;
+ }
+ snprintf(filename, namelen, "%s", vname);
+ }
+ bzero(&cn, sizeof(cn));
+ cn.cn_nameptr = filename;
+ cn.cn_namelen = namelen;
+
+restart:
+ done = 0;
+ if ((error = nfs_mount_state_in_use_start(nmp)))
+ goto out;
+
+ if (nofp->nof_rw)
+ error = nfs4_open_reopen_rpc(nofp, thd, noop->noo_cred, &cn, dvp, &vp, NFS_OPEN_SHARE_ACCESS_BOTH, NFS_OPEN_SHARE_DENY_NONE);
+ if (!error && nofp->nof_w)
+ error = nfs4_open_reopen_rpc(nofp, thd, noop->noo_cred, &cn, dvp, &vp, NFS_OPEN_SHARE_ACCESS_WRITE, NFS_OPEN_SHARE_DENY_NONE);
+ if (!error && nofp->nof_r)
+ error = nfs4_open_reopen_rpc(nofp, thd, noop->noo_cred, &cn, dvp, &vp, NFS_OPEN_SHARE_ACCESS_READ, NFS_OPEN_SHARE_DENY_NONE);
+
+ if (nfs_mount_state_in_use_end(nmp, error)) {
+ if (error == NFSERR_GRACE)
+ goto restart;
+ error = 0;
+ goto out;
+ }
+ done = 1;
+out:
+ lck_mtx_lock(&nofp->nof_lock);
+ nofp->nof_flags &= ~NFS_OPEN_FILE_REOPENING;
+ if (error)
+ nofp->nof_flags |= NFS_OPEN_FILE_LOST;
+ if (done)
+ nofp->nof_flags &= ~NFS_OPEN_FILE_REOPEN;
+ else
+ printf("nfs4_reopen: failed, error %d, lost %d\n", error, (nofp->nof_flags & NFS_OPEN_FILE_LOST) ? 1 : 0);
+ lck_mtx_unlock(&nofp->nof_lock);
+ if (filename && (filename != &smallname[0]))
+ FREE(filename, M_TEMP);
+ if (vname)
+ vnode_putname(vname);
+ if (dvp != NULLVP)
+ vnode_put(dvp);
+}
+
+/*
+ * Send a normal OPEN RPC to open/create a file.
+ */
+int
+nfs4_open_rpc(
+ struct nfs_open_file *nofp,
+ vfs_context_t ctx,
+ struct componentname *cnp,
+ struct vnode_attr *vap,
+ vnode_t dvp,
+ vnode_t *vpp,
+ int create,
+ int share_access,
+ int share_deny)
+{
+ return (nfs4_open_rpc_internal(nofp, ctx, vfs_context_thread(ctx), vfs_context_ucred(ctx),
+ cnp, vap, dvp, vpp, create, share_access, share_deny));
+}
+
+/*
+ * Send an OPEN RPC to reopen a file.
+ */
+int
+nfs4_open_reopen_rpc(
+ struct nfs_open_file *nofp,
+ thread_t thd,
+ kauth_cred_t cred,
+ struct componentname *cnp,
+ vnode_t dvp,
+ vnode_t *vpp,
+ int share_access,
+ int share_deny)
+{
+ return (nfs4_open_rpc_internal(nofp, NULL, thd, cred, cnp, NULL, dvp, vpp, 0, share_access, share_deny));
+}
+
+/*
+ * common OPEN RPC code
+ *
+ * If create is set, ctx must be passed in.
+ */
+int
+nfs4_open_rpc_internal(
+ struct nfs_open_file *nofp,
+ vfs_context_t ctx,
+ thread_t thd,
+ kauth_cred_t cred,
+ struct componentname *cnp,
+ struct vnode_attr *vap,
+ vnode_t dvp,
+ vnode_t *vpp,
+ int create,
+ int share_access,
+ int share_deny)
+{
+ struct nfsmount *nmp;
+ struct nfs_open_owner *noop = nofp->nof_owner;
+ struct nfs_vattr nvattr, dnvattr;
+ int error = 0, open_error = EIO, lockerror = ENOENT, busyerror = ENOENT, status;
+ int nfsvers, numops, exclusive = 0, gotuid, gotgid;
+ u_int64_t xid, savedxid = 0;
+ nfsnode_t dnp = VTONFS(dvp);
+ nfsnode_t np, newnp = NULL;
+ vnode_t newvp = NULL;
+ struct nfsm_chain nmreq, nmrep;
+ uint32_t bitmap[NFS_ATTR_BITMAP_LEN], bmlen;
+ uint32_t rflags, delegation = 0, recall = 0, val;
+ struct nfs_stateid stateid, dstateid, *sid;
+ fhandle_t fh;
+ struct nfsreq *req = NULL;
+ struct nfs_dulookup dul;
+
+ if (create && !ctx)
+ return (EINVAL);
+
+ nmp = VTONMP(dvp);
+ if (!nmp)
+ return (ENXIO);
+ nfsvers = nmp->nm_vers;
+
+ np = *vpp ? VTONFS(*vpp) : NULL;
+ if (create && vap) {
+ exclusive = (vap->va_vaflags & VA_EXCLUSIVE);
+ nfs_avoid_needless_id_setting_on_create(dnp, vap, ctx);
+ gotuid = VATTR_IS_ACTIVE(vap, va_uid);
+ gotgid = VATTR_IS_ACTIVE(vap, va_gid);
+ } else {
+ exclusive = gotuid = gotgid = 0;
+ }
+ if (nofp) {
+ sid = &nofp->nof_stateid;
+ } else {
+ stateid.seqid = stateid.other[0] = stateid.other[1] = stateid.other[2] = 0;
+ sid = &stateid;
+ }
+
+ if ((error = nfs_open_owner_set_busy(noop, thd)))
+ return (error);
+again:
+ rflags = 0;
+
+ nfsm_chain_null(&nmreq);
+ nfsm_chain_null(&nmrep);
+
+ // PUTFH, SAVEFH, OPEN(CREATE?), GETATTR(FH), RESTOREFH, GETATTR
+ numops = 6;
+ nfsm_chain_build_alloc_init(error, &nmreq, 53 * NFSX_UNSIGNED + cnp->cn_namelen);
+ nfsm_chain_add_compound_header(error, &nmreq, create ? "create" : "open", numops);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_PUTFH);
+ nfsm_chain_add_fh(error, &nmreq, nfsvers, dnp->n_fhp, dnp->n_fhsize);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_SAVEFH);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_OPEN);
+ nfsm_chain_add_32(error, &nmreq, noop->noo_seqid);
+ nfsm_chain_add_32(error, &nmreq, share_access);
+ nfsm_chain_add_32(error, &nmreq, share_deny);
+
+ // open owner: clientid + uid
+ nfsm_chain_add_64(error, &nmreq, nmp->nm_clientid); // open_owner4.clientid
+ nfsm_chain_add_32(error, &nmreq, NFSX_UNSIGNED);
+ nfsm_chain_add_32(error, &nmreq, kauth_cred_getuid(noop->noo_cred)); // open_owner4.owner
+
+ // openflag4
+ nfsm_chain_add_32(error, &nmreq, create);
+ if (create) {
+ if (exclusive) {
+ static uint32_t create_verf; // XXX need a better verifier
+ create_verf++;
+ nfsm_chain_add_32(error, &nmreq, NFS_CREATE_EXCLUSIVE);
+ /* insert 64 bit verifier */
+ nfsm_chain_add_32(error, &nmreq, create_verf);
+ nfsm_chain_add_32(error, &nmreq, create_verf);
+ } else {
+ nfsm_chain_add_32(error, &nmreq, NFS_CREATE_UNCHECKED);
+ nfsm_chain_add_fattr4(error, &nmreq, vap, nmp);
+ }
+ }
+
+ // open_claim4
+ nfsm_chain_add_32(error, &nmreq, NFS_CLAIM_NULL);
+ nfsm_chain_add_string(error, &nmreq, cnp->cn_nameptr, cnp->cn_namelen);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ NFS_COPY_ATTRIBUTES(nfs_getattr_bitmap, bitmap);
+ NFS_BITMAP_SET(bitmap, NFS_FATTR_FILEHANDLE);
+ nfsm_chain_add_bitmap_masked(error, &nmreq, bitmap,
+ NFS_ATTR_BITMAP_LEN, nmp->nm_fsattr.nfsa_supp_attr);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_RESTOREFH);
+ numops--;
+ nfsm_chain_add_32(error, &nmreq, NFS_OP_GETATTR);
+ nfsm_chain_add_bitmap_masked(error, &nmreq, nfs_getattr_bitmap,
+ NFS_ATTR_BITMAP_LEN, nmp->nm_fsattr.nfsa_supp_attr);
+ nfsm_chain_build_done(error, &nmreq);
+ nfsm_assert(error, (numops == 0), EPROTO);
+ if (!error)
+ error = busyerror = nfs_node_set_busy(dnp, thd);
+ nfsmout_if(error);
+
+ if (create)
+ nfs_dulookup_init(&dul, dnp, cnp->cn_nameptr, cnp->cn_namelen, ctx);
+
+ error = nfs_request_async(dnp, NULL, &nmreq, NFSPROC4_COMPOUND, thd, cred, NULL, &req);
+ if (!error) {
+ if (create)
+ nfs_dulookup_start(&dul, dnp, ctx);
+ error = nfs_request_async_finish(req, &nmrep, &xid, &status);
+ savedxid = xid;
+ }
+
+ if (create)
+ nfs_dulookup_finish(&dul, dnp, ctx);
+
+ if ((lockerror = nfs_node_lock(dnp)))
+ error = lockerror;
+ 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_SAVEFH);
+ nfsmout_if(error);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_OPEN);
+ nfs_owner_seqid_increment(noop, NULL, error);
+ nfsm_chain_get_stateid(error, &nmrep, sid);
+ nfsm_chain_check_change_info(error, &nmrep, dnp);
+ nfsm_chain_get_32(error, &nmrep, rflags);
+ bmlen = NFS_ATTR_BITMAP_LEN;
+ nfsm_chain_get_bitmap(error, &nmrep, bitmap, bmlen);
+ nfsm_chain_get_32(error, &nmrep, delegation);
+ if (!error)
+ switch (delegation) {
+ case NFS_OPEN_DELEGATE_NONE:
+ break;
+ case NFS_OPEN_DELEGATE_READ:
+ nfsm_chain_get_stateid(error, &nmrep, &dstateid);
+ nfsm_chain_get_32(error, &nmrep, recall);
+ // ACE: (skip) XXX
+ nfsm_chain_adv(error, &nmrep, 3 * NFSX_UNSIGNED);
+ nfsm_chain_get_32(error, &nmrep, val); /* string length */
+ nfsm_chain_adv(error, &nmrep, nfsm_rndup(val));
+ break;
+ case NFS_OPEN_DELEGATE_WRITE:
+ nfsm_chain_get_stateid(error, &nmrep, &dstateid);
+ nfsm_chain_get_32(error, &nmrep, recall);
+ // space (skip) XXX
+ nfsm_chain_adv(error, &nmrep, 3 * NFSX_UNSIGNED);
+ // ACE: (skip) XXX
+ nfsm_chain_adv(error, &nmrep, 3 * NFSX_UNSIGNED);
+ nfsm_chain_get_32(error, &nmrep, val); /* string length */
+ nfsm_chain_adv(error, &nmrep, nfsm_rndup(val));
+ break;
+ default:
+ error = EBADRPC;
+ break;
+ }
+ /* At this point if we have no error, the object was created/opened. */
+ /* if we don't get attributes, then we should lookitup. */
+ open_error = error;
+ nfsmout_if(error);
+ if (create && !exclusive)
+ nfs_vattr_set_supported(bitmap, vap);
+ nfsm_chain_op_check(error, &nmrep, NFS_OP_GETATTR);
+ nfsmout_if(error);
+ NFS_CLEAR_ATTRIBUTES(nvattr.nva_bitmap);
+ error = nfs4_parsefattr(&nmrep, NULL, &nvattr, &fh, NULL);
+ nfsmout_if(error);
+ if (!NFS_BITMAP_ISSET(nvattr.nva_bitmap, NFS_FATTR_FILEHANDLE)) {
+ printf("nfs: open/create didn't return filehandle?\n");
projects / apple / xnu.git / blobdiff