Index: head/sys/kern/subr_syscall.c =================================================================== --- head/sys/kern/subr_syscall.c (revision 367631) +++ head/sys/kern/subr_syscall.c (revision 367632) @@ -1,272 +1,274 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (C) 1994, David Greenman * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * Copyright (C) 2010 Konstantin Belousov * * This code is derived from software contributed to Berkeley by * the University of Utah, and William Jolitz. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)trap.c 7.4 (Berkeley) 5/13/91 */ #include "opt_capsicum.h" #include "opt_ktrace.h" __FBSDID("$FreeBSD$"); #include #include #include #ifdef KTRACE #include #include #endif #include static inline void syscallenter(struct thread *td) { struct proc *p; struct syscall_args *sa; struct sysent *se; int error, traced; bool sy_thr_static; VM_CNT_INC(v_syscall); p = td->td_proc; sa = &td->td_sa; td->td_pticks = 0; if (__predict_false(td->td_cowgen != p->p_cowgen)) thread_cow_update(td); traced = (p->p_flag & P_TRACED) != 0; if (__predict_false(traced || td->td_dbgflags & TDB_USERWR)) { PROC_LOCK(p); td->td_dbgflags &= ~TDB_USERWR; if (traced) td->td_dbgflags |= TDB_SCE; PROC_UNLOCK(p); } error = (p->p_sysent->sv_fetch_syscall_args)(td); se = sa->callp; #ifdef KTRACE if (KTRPOINT(td, KTR_SYSCALL)) ktrsyscall(sa->code, se->sy_narg, sa->args); #endif KTR_START4(KTR_SYSC, "syscall", syscallname(p, sa->code), (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "arg0:%p", sa->args[0], "arg1:%p", sa->args[1], "arg2:%p", sa->args[2]); if (__predict_false(error != 0)) { td->td_errno = error; goto retval; } if (__predict_false(traced)) { PROC_LOCK(p); if (p->p_ptevents & PTRACE_SCE) ptracestop((td), SIGTRAP, NULL); PROC_UNLOCK(p); if ((td->td_dbgflags & TDB_USERWR) != 0) { /* * Reread syscall number and arguments if debugger * modified registers or memory. */ error = (p->p_sysent->sv_fetch_syscall_args)(td); se = sa->callp; #ifdef KTRACE if (KTRPOINT(td, KTR_SYSCALL)) ktrsyscall(sa->code, se->sy_narg, sa->args); #endif if (error != 0) { td->td_errno = error; goto retval; } } } #ifdef CAPABILITY_MODE /* * In capability mode, we only allow access to system calls * flagged with SYF_CAPENABLED. */ if (__predict_false(IN_CAPABILITY_MODE(td) && (se->sy_flags & SYF_CAPENABLED) == 0)) { td->td_errno = error = ECAPMODE; goto retval; } #endif /* * Fetch fast sigblock value at the time of syscall entry to * handle sleepqueue primitives which might call cursig(). */ if (__predict_false(sigfastblock_fetch_always)) (void)sigfastblock_fetch(td); /* Let system calls set td_errno directly. */ KASSERT((td->td_pflags & TDP_NERRNO) == 0, ("%s: TDP_NERRNO set", __func__)); sy_thr_static = (se->sy_thrcnt & SY_THR_STATIC) != 0; if (__predict_false(SYSTRACE_ENABLED() || AUDIT_SYSCALL_ENTER(sa->code, td) || !sy_thr_static)) { if (!sy_thr_static) { error = syscall_thread_enter(td, se); if (error != 0) { td->td_errno = error; goto retval; } } #ifdef KDTRACE_HOOKS /* Give the syscall:::entry DTrace probe a chance to fire. */ if (__predict_false(se->sy_entry != 0)) (*systrace_probe_func)(sa, SYSTRACE_ENTRY, 0); #endif error = (se->sy_call)(td, sa->args); /* Save the latest error return value. */ if (__predict_false((td->td_pflags & TDP_NERRNO) != 0)) td->td_pflags &= ~TDP_NERRNO; else td->td_errno = error; /* * Note that some syscall implementations (e.g., sys_execve) * will commit the audit record just before their final return. * These were done under the assumption that nothing of interest * would happen between their return and here, where we would * normally commit the audit record. These assumptions will * need to be revisited should any substantial logic be added * above. */ AUDIT_SYSCALL_EXIT(error, td); #ifdef KDTRACE_HOOKS /* Give the syscall:::return DTrace probe a chance to fire. */ if (__predict_false(se->sy_return != 0)) (*systrace_probe_func)(sa, SYSTRACE_RETURN, error ? -1 : td->td_retval[0]); #endif if (!sy_thr_static) syscall_thread_exit(td, se); } else { error = (se->sy_call)(td, sa->args); /* Save the latest error return value. */ if (__predict_false((td->td_pflags & TDP_NERRNO) != 0)) td->td_pflags &= ~TDP_NERRNO; else td->td_errno = error; } retval: KTR_STOP4(KTR_SYSC, "syscall", syscallname(p, sa->code), (uintptr_t)td, "pid:%d", td->td_proc->p_pid, "error:%d", error, "retval0:%#lx", td->td_retval[0], "retval1:%#lx", td->td_retval[1]); if (__predict_false(traced)) { PROC_LOCK(p); td->td_dbgflags &= ~TDB_SCE; PROC_UNLOCK(p); } (p->p_sysent->sv_set_syscall_retval)(td, error); } static inline void syscallret(struct thread *td) { struct proc *p; struct syscall_args *sa; ksiginfo_t ksi; int traced; KASSERT((td->td_pflags & TDP_FORKING) == 0, ("fork() did not clear TDP_FORKING upon completion")); + KASSERT(td->td_errno != ERELOOKUP, + ("ERELOOKUP not consumed syscall %d", td->td_sa.code)); p = td->td_proc; sa = &td->td_sa; if (__predict_false(td->td_errno == ENOTCAPABLE || td->td_errno == ECAPMODE)) { if ((trap_enotcap || (p->p_flag2 & P2_TRAPCAP) != 0) && IN_CAPABILITY_MODE(td)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGTRAP; ksi.ksi_errno = td->td_errno; ksi.ksi_code = TRAP_CAP; trapsignal(td, &ksi); } } /* * Handle reschedule and other end-of-syscall issues */ userret(td, td->td_frame); #ifdef KTRACE if (KTRPOINT(td, KTR_SYSRET)) { ktrsysret(sa->code, td->td_errno, td->td_retval[0]); } #endif traced = 0; if (__predict_false(p->p_flag & P_TRACED)) { traced = 1; PROC_LOCK(p); td->td_dbgflags |= TDB_SCX; PROC_UNLOCK(p); } if (__predict_false(traced || (td->td_dbgflags & (TDB_EXEC | TDB_FORK)) != 0)) { PROC_LOCK(p); /* * If tracing the execed process, trap to the debugger * so that breakpoints can be set before the program * executes. If debugger requested tracing of syscall * returns, do it now too. */ if (traced && ((td->td_dbgflags & (TDB_FORK | TDB_EXEC)) != 0 || (p->p_ptevents & PTRACE_SCX) != 0)) ptracestop(td, SIGTRAP, NULL); td->td_dbgflags &= ~(TDB_SCX | TDB_EXEC | TDB_FORK); PROC_UNLOCK(p); } if (__predict_false(td->td_pflags & TDP_RFPPWAIT)) fork_rfppwait(td); } Index: head/sys/kern/uipc_usrreq.c =================================================================== --- head/sys/kern/uipc_usrreq.c (revision 367631) +++ head/sys/kern/uipc_usrreq.c (revision 367632) @@ -1,2965 +1,2967 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1991, 1993 * The Regents of the University of California. All Rights Reserved. * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved. * Copyright (c) 2018 Matthew Macy * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94 */ /* * UNIX Domain (Local) Sockets * * This is an implementation of UNIX (local) domain sockets. Each socket has * an associated struct unpcb (UNIX protocol control block). Stream sockets * may be connected to 0 or 1 other socket. Datagram sockets may be * connected to 0, 1, or many other sockets. Sockets may be created and * connected in pairs (socketpair(2)), or bound/connected to using the file * system name space. For most purposes, only the receive socket buffer is * used, as sending on one socket delivers directly to the receive socket * buffer of a second socket. * * The implementation is substantially complicated by the fact that * "ancillary data", such as file descriptors or credentials, may be passed * across UNIX domain sockets. The potential for passing UNIX domain sockets * over other UNIX domain sockets requires the implementation of a simple * garbage collector to find and tear down cycles of disconnected sockets. * * TODO: * RDM * rethink name space problems * need a proper out-of-band */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include MALLOC_DECLARE(M_FILECAPS); /* * See unpcb.h for the locking key. */ static uma_zone_t unp_zone; static unp_gen_t unp_gencnt; /* (l) */ static u_int unp_count; /* (l) Count of local sockets. */ static ino_t unp_ino; /* Prototype for fake inode numbers. */ static int unp_rights; /* (g) File descriptors in flight. */ static struct unp_head unp_shead; /* (l) List of stream sockets. */ static struct unp_head unp_dhead; /* (l) List of datagram sockets. */ static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */ struct unp_defer { SLIST_ENTRY(unp_defer) ud_link; struct file *ud_fp; }; static SLIST_HEAD(, unp_defer) unp_defers; static int unp_defers_count; static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL }; /* * Garbage collection of cyclic file descriptor/socket references occurs * asynchronously in a taskqueue context in order to avoid recursion and * reentrance in the UNIX domain socket, file descriptor, and socket layer * code. See unp_gc() for a full description. */ static struct timeout_task unp_gc_task; /* * The close of unix domain sockets attached as SCM_RIGHTS is * postponed to the taskqueue, to avoid arbitrary recursion depth. * The attached sockets might have another sockets attached. */ static struct task unp_defer_task; /* * Both send and receive buffers are allocated PIPSIZ bytes of buffering for * stream sockets, although the total for sender and receiver is actually * only PIPSIZ. * * Datagram sockets really use the sendspace as the maximum datagram size, * and don't really want to reserve the sendspace. Their recvspace should be * large enough for at least one max-size datagram plus address. */ #ifndef PIPSIZ #define PIPSIZ 8192 #endif static u_long unpst_sendspace = PIPSIZ; static u_long unpst_recvspace = PIPSIZ; static u_long unpdg_sendspace = 2*1024; /* really max datagram size */ static u_long unpdg_recvspace = 4*1024; static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */ static u_long unpsp_recvspace = PIPSIZ; static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Local domain"); static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_STREAM"); static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_DGRAM"); static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "SOCK_SEQPACKET"); SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW, &unpst_sendspace, 0, "Default stream send space."); SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW, &unpst_recvspace, 0, "Default stream receive space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW, &unpdg_sendspace, 0, "Default datagram send space."); SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW, &unpdg_recvspace, 0, "Default datagram receive space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW, &unpsp_sendspace, 0, "Default seqpacket send space."); SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW, &unpsp_recvspace, 0, "Default seqpacket receive space."); SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "File descriptors in flight."); SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD, &unp_defers_count, 0, "File descriptors deferred to taskqueue for close."); /* * Locking and synchronization: * * Several types of locks exist in the local domain socket implementation: * - a global linkage lock * - a global connection list lock * - the mtxpool lock * - per-unpcb mutexes * * The linkage lock protects the global socket lists, the generation number * counter and garbage collector state. * * The connection list lock protects the list of referring sockets in a datagram * socket PCB. This lock is also overloaded to protect a global list of * sockets whose buffers contain socket references in the form of SCM_RIGHTS * messages. To avoid recursion, such references are released by a dedicated * thread. * * The mtxpool lock protects the vnode from being modified while referenced. * Lock ordering rules require that it be acquired before any PCB locks. * * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the * unpcb. This includes the unp_conn field, which either links two connected * PCBs together (for connected socket types) or points at the destination * socket (for connectionless socket types). The operations of creating or * destroying a connection therefore involve locking multiple PCBs. To avoid * lock order reversals, in some cases this involves dropping a PCB lock and * using a reference counter to maintain liveness. * * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer, * allocated in pru_attach() and freed in pru_detach(). The validity of that * pointer is an invariant, so no lock is required to dereference the so_pcb * pointer if a valid socket reference is held by the caller. In practice, * this is always true during operations performed on a socket. Each unpcb * has a back-pointer to its socket, unp_socket, which will be stable under * the same circumstances. * * This pointer may only be safely dereferenced as long as a valid reference * to the unpcb is held. Typically, this reference will be from the socket, * or from another unpcb when the referring unpcb's lock is held (in order * that the reference not be invalidated during use). For example, to follow * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee * that detach is not run clearing unp_socket. * * Blocking with UNIX domain sockets is a tricky issue: unlike most network * protocols, bind() is a non-atomic operation, and connect() requires * potential sleeping in the protocol, due to potentially waiting on local or * distributed file systems. We try to separate "lookup" operations, which * may sleep, and the IPC operations themselves, which typically can occur * with relative atomicity as locks can be held over the entire operation. * * Another tricky issue is simultaneous multi-threaded or multi-process * access to a single UNIX domain socket. These are handled by the flags * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or * binding, both of which involve dropping UNIX domain socket locks in order * to perform namei() and other file system operations. */ static struct rwlock unp_link_rwlock; static struct mtx unp_defers_lock; #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \ "unp_link_rwlock") #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_LOCKED) #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_UNLOCKED) #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock) #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock) #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock) #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock) #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \ RA_WLOCKED) #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock) #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \ "unp_defer", NULL, MTX_DEF) #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock) #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock) #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK(); #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK(); #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \ "unp", "unp", \ MTX_DUPOK|MTX_DEF) #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx) #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx) #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx) #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx) #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx) #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx) #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED) #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED) static int uipc_connect2(struct socket *, struct socket *); static int uipc_ctloutput(struct socket *, struct sockopt *); static int unp_connect(struct socket *, struct sockaddr *, struct thread *); static int unp_connectat(int, struct socket *, struct sockaddr *, struct thread *); static int unp_connect2(struct socket *so, struct socket *so2, int); static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2); static void unp_dispose(struct socket *so); static void unp_dispose_mbuf(struct mbuf *); static void unp_shutdown(struct unpcb *); static void unp_drop(struct unpcb *); static void unp_gc(__unused void *, int); static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int)); static void unp_discard(struct file *); static void unp_freerights(struct filedescent **, int); static void unp_init(void); static int unp_internalize(struct mbuf **, struct thread *); static void unp_internalize_fp(struct file *); static int unp_externalize(struct mbuf *, struct mbuf **, int); static int unp_externalize_fp(struct file *); static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *); static void unp_process_defers(void * __unused, int); static void unp_pcb_hold(struct unpcb *unp) { u_int old __unused; old = refcount_acquire(&unp->unp_refcount); KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp)); } static __result_use_check bool unp_pcb_rele(struct unpcb *unp) { bool ret; UNP_PCB_LOCK_ASSERT(unp); if ((ret = refcount_release(&unp->unp_refcount))) { UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK_DESTROY(unp); uma_zfree(unp_zone, unp); } return (ret); } static void unp_pcb_rele_notlast(struct unpcb *unp) { bool ret __unused; ret = refcount_release(&unp->unp_refcount); KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp)); } static void unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (unp == unp2) { UNP_PCB_LOCK(unp); } else if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp); UNP_PCB_LOCK(unp2); } else { UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); } } static void unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2) { UNP_PCB_UNLOCK(unp); if (unp != unp2) UNP_PCB_UNLOCK(unp2); } /* * Try to lock the connected peer of an already locked socket. In some cases * this requires that we unlock the current socket. The pairbusy counter is * used to block concurrent connection attempts while the lock is dropped. The * caller must be careful to revalidate PCB state. */ static struct unpcb * unp_pcb_lock_peer(struct unpcb *unp) { struct unpcb *unp2; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if (__predict_false(unp2 == NULL)) return (NULL); if (__predict_false(unp == unp2)) return (unp); UNP_PCB_UNLOCK_ASSERT(unp2); if (__predict_true(UNP_PCB_TRYLOCK(unp2))) return (unp2); if ((uintptr_t)unp2 > (uintptr_t)unp) { UNP_PCB_LOCK(unp2); return (unp2); } unp->unp_pairbusy++; unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); UNP_PCB_LOCK(unp2); UNP_PCB_LOCK(unp); KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL, ("%s: socket %p was reconnected", __func__, unp)); if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) { unp->unp_flags &= ~UNP_WAITING; wakeup(unp); } if (unp_pcb_rele(unp2)) { /* unp2 is unlocked. */ return (NULL); } if (unp->unp_conn == NULL) { UNP_PCB_UNLOCK(unp2); return (NULL); } return (unp2); } /* * Definitions of protocols supported in the LOCAL domain. */ static struct domain localdomain; static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream; static struct pr_usrreqs uipc_usrreqs_seqpacket; static struct protosw localsw[] = { { .pr_type = SOCK_STREAM, .pr_domain = &localdomain, .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_stream }, { .pr_type = SOCK_DGRAM, .pr_domain = &localdomain, .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_dgram }, { .pr_type = SOCK_SEQPACKET, .pr_domain = &localdomain, /* * XXXRW: For now, PR_ADDR because soreceive will bump into them * due to our use of sbappendaddr. A new sbappend variants is needed * that supports both atomic record writes and control data. */ .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD| PR_RIGHTS, .pr_ctloutput = &uipc_ctloutput, .pr_usrreqs = &uipc_usrreqs_seqpacket, }, }; static struct domain localdomain = { .dom_family = AF_LOCAL, .dom_name = "local", .dom_init = unp_init, .dom_externalize = unp_externalize, .dom_dispose = unp_dispose, .dom_protosw = localsw, .dom_protoswNPROTOSW = &localsw[nitems(localsw)] }; DOMAIN_SET(local); static void uipc_abort(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_abort: unp == NULL")); UNP_PCB_UNLOCK_ASSERT(unp); UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 != NULL) { unp_pcb_hold(unp2); UNP_PCB_UNLOCK(unp); unp_drop(unp2); } else UNP_PCB_UNLOCK(unp); } static int uipc_accept(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; /* * Pass back name of connected socket, if it was bound and we are * still connected (our peer may have closed already!). */ unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_accept: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); unp2 = unp_pcb_lock_peer(unp); if (unp2 != NULL && unp2->unp_addr != NULL) sa = (struct sockaddr *)unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); if (unp2 != NULL) unp_pcb_unlock_pair(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_attach(struct socket *so, int proto, struct thread *td) { u_long sendspace, recvspace; struct unpcb *unp; int error; bool locked; KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL")); if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { switch (so->so_type) { case SOCK_STREAM: sendspace = unpst_sendspace; recvspace = unpst_recvspace; break; case SOCK_DGRAM: sendspace = unpdg_sendspace; recvspace = unpdg_recvspace; break; case SOCK_SEQPACKET: sendspace = unpsp_sendspace; recvspace = unpsp_recvspace; break; default: panic("uipc_attach"); } error = soreserve(so, sendspace, recvspace); if (error) return (error); } unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO); if (unp == NULL) return (ENOBUFS); LIST_INIT(&unp->unp_refs); UNP_PCB_LOCK_INIT(unp); unp->unp_socket = so; so->so_pcb = unp; refcount_init(&unp->unp_refcount, 1); if ((locked = UNP_LINK_WOWNED()) == false) UNP_LINK_WLOCK(); unp->unp_gencnt = ++unp_gencnt; unp->unp_ino = ++unp_ino; unp_count++; switch (so->so_type) { case SOCK_STREAM: LIST_INSERT_HEAD(&unp_shead, unp, unp_link); break; case SOCK_DGRAM: LIST_INSERT_HEAD(&unp_dhead, unp, unp_link); break; case SOCK_SEQPACKET: LIST_INSERT_HEAD(&unp_sphead, unp, unp_link); break; default: panic("uipc_attach"); } if (locked == false) UNP_LINK_WUNLOCK(); return (0); } static int uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct sockaddr_un *soun = (struct sockaddr_un *)nam; struct vattr vattr; int error, namelen; struct nameidata nd; struct unpcb *unp; struct vnode *vp; struct mount *mp; cap_rights_t rights; char *buf; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_bind: unp == NULL")); if (soun->sun_len > sizeof(struct sockaddr_un)) return (EINVAL); namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path); if (namelen <= 0) return (EINVAL); /* * We don't allow simultaneous bind() calls on a single UNIX domain * socket, so flag in-progress operations, and return an error if an * operation is already in progress. * * Historically, we have not allowed a socket to be rebound, so this * also returns an error. Not allowing re-binding simplifies the * implementation and avoids a great many possible failure modes. */ UNP_PCB_LOCK(unp); if (unp->unp_vnode != NULL) { UNP_PCB_UNLOCK(unp); return (EINVAL); } if (unp->unp_flags & UNP_BINDING) { UNP_PCB_UNLOCK(unp); return (EALREADY); } unp->unp_flags |= UNP_BINDING; UNP_PCB_UNLOCK(unp); buf = malloc(namelen + 1, M_TEMP, M_WAITOK); bcopy(soun->sun_path, buf, namelen); buf[namelen] = 0; restart: NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE, UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td); /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ error = namei(&nd); if (error) goto error; vp = nd.ni_vp; if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (vp != NULL) { vrele(vp); error = EADDRINUSE; goto error; } error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error) goto error; goto restart; } VATTR_NULL(&vattr); vattr.va_type = VSOCK; vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask); #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error) { vn_finished_write(mp); + if (error == ERELOOKUP) + goto restart; goto error; } vp = nd.ni_vp; ASSERT_VOP_ELOCKED(vp, "uipc_bind"); soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK); UNP_PCB_LOCK(unp); VOP_UNP_BIND(vp, unp); unp->unp_vnode = vp; unp->unp_addr = soun; unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); VOP_UNLOCK(vp); vn_finished_write(mp); free(buf, M_TEMP); return (0); error: UNP_PCB_LOCK(unp); unp->unp_flags &= ~UNP_BINDING; UNP_PCB_UNLOCK(unp); free(buf, M_TEMP); return (error); } static int uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td) { return (uipc_bindat(AT_FDCWD, so, nam, td)); } static int uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connect: td != curthread")); error = unp_connect(so, nam, td); return (error); } static int uipc_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { int error; KASSERT(td == curthread, ("uipc_connectat: td != curthread")); error = unp_connectat(fd, so, nam, td); return (error); } static void uipc_close(struct socket *so) { struct unpcb *unp, *unp2; struct vnode *vp = NULL; struct mtx *vplock; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_close: unp == NULL")); vplock = NULL; if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (vp && unp->unp_vnode == NULL) { mtx_unlock(vplock); vp = NULL; } if (vp != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } } static int uipc_connect2(struct socket *so1, struct socket *so2) { struct unpcb *unp, *unp2; int error; unp = so1->so_pcb; KASSERT(unp != NULL, ("uipc_connect2: unp == NULL")); unp2 = so2->so_pcb; KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL")); unp_pcb_lock_pair(unp, unp2); error = unp_connect2(so1, so2, PRU_CONNECT2); unp_pcb_unlock_pair(unp, unp2); return (error); } static void uipc_detach(struct socket *so) { struct unpcb *unp, *unp2; struct mtx *vplock; struct vnode *vp; int local_unp_rights; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_detach: unp == NULL")); vp = NULL; vplock = NULL; SOCK_LOCK(so); if (!SOLISTENING(so)) { /* * Once the socket is removed from the global lists, * uipc_ready() will not be able to locate its socket buffer, so * clear the buffer now. At this point internalized rights have * already been disposed of. */ sbrelease(&so->so_rcv, so); } SOCK_UNLOCK(so); UNP_LINK_WLOCK(); LIST_REMOVE(unp, unp_link); if (unp->unp_gcflag & UNPGC_DEAD) LIST_REMOVE(unp, unp_dead); unp->unp_gencnt = ++unp_gencnt; --unp_count; UNP_LINK_WUNLOCK(); UNP_PCB_UNLOCK_ASSERT(unp); restart: if ((vp = unp->unp_vnode) != NULL) { vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); } UNP_PCB_LOCK(unp); if (unp->unp_vnode != vp && unp->unp_vnode != NULL) { if (vplock) mtx_unlock(vplock); UNP_PCB_UNLOCK(unp); goto restart; } if ((vp = unp->unp_vnode) != NULL) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; } if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); UNP_REF_LIST_LOCK(); while (!LIST_EMPTY(&unp->unp_refs)) { struct unpcb *ref = LIST_FIRST(&unp->unp_refs); unp_pcb_hold(ref); UNP_REF_LIST_UNLOCK(); MPASS(ref != unp); UNP_PCB_UNLOCK_ASSERT(ref); unp_drop(ref); UNP_REF_LIST_LOCK(); } UNP_REF_LIST_UNLOCK(); UNP_PCB_LOCK(unp); local_unp_rights = unp_rights; unp->unp_socket->so_pcb = NULL; unp->unp_socket = NULL; free(unp->unp_addr, M_SONAME); unp->unp_addr = NULL; if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (vp) { mtx_unlock(vplock); vrele(vp); } if (local_unp_rights) taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1); } static int uipc_disconnect(struct socket *so) { struct unpcb *unp, *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL")); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) unp_disconnect(unp, unp2); else UNP_PCB_UNLOCK(unp); return (0); } static int uipc_listen(struct socket *so, int backlog, struct thread *td) { struct unpcb *unp; int error; if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET) return (EOPNOTSUPP); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_listen: unp == NULL")); UNP_PCB_LOCK(unp); if (unp->unp_vnode == NULL) { /* Already connected or not bound to an address. */ error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ; UNP_PCB_UNLOCK(unp); return (error); } SOCK_LOCK(so); error = solisten_proto_check(so); if (error == 0) { cru2xt(td, &unp->unp_peercred); solisten_proto(so, backlog); } SOCK_UNLOCK(so); UNP_PCB_UNLOCK(unp); return (error); } static int uipc_peeraddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp, *unp2; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_LINK_RLOCK(); /* * XXX: It seems that this test always fails even when connection is * established. So, this else clause is added as workaround to * return PF_LOCAL sockaddr. */ unp2 = unp->unp_conn; if (unp2 != NULL) { UNP_PCB_LOCK(unp2); if (unp2->unp_addr != NULL) sa = (struct sockaddr *) unp2->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp2); } else { sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); } UNP_LINK_RUNLOCK(); return (0); } static int uipc_rcvd(struct socket *so, int flags) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); /* * Adjust backpressure on sender and wakeup any waiting to write. * * The unp lock is acquired to maintain the validity of the unp_conn * pointer; no lock on unp2 is required as unp2->unp_socket will be * static as long as we don't permit unp2 to disconnect from unp, * which is prevented by the lock on unp. We cache values from * so_rcv to avoid holding the so_rcv lock over the entire * transaction on the remote so_snd. */ SOCKBUF_LOCK(&so->so_rcv); mbcnt = so->so_rcv.sb_mbcnt; sbcc = sbavail(&so->so_rcv); SOCKBUF_UNLOCK(&so->so_rcv); /* * There is a benign race condition at this point. If we're planning to * clear SB_STOP, but uipc_send is called on the connected socket at * this instant, it might add data to the sockbuf and set SB_STOP. Then * we would erroneously clear SB_STOP below, even though the sockbuf is * full. The race is benign because the only ill effect is to allow the * sockbuf to exceed its size limit, and the size limits are not * strictly guaranteed anyway. */ UNP_PCB_LOCK(unp); unp2 = unp->unp_conn; if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); return (0); } so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_snd); if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax) so2->so_snd.sb_flags &= ~SB_STOP; sowwakeup_locked(so2); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, struct mbuf *control, struct thread *td) { struct unpcb *unp, *unp2; struct socket *so2; u_int mbcnt, sbcc; int freed, error; unp = sotounpcb(so); KASSERT(unp != NULL, ("%s: unp == NULL", __func__)); KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM || so->so_type == SOCK_SEQPACKET, ("%s: socktype %d", __func__, so->so_type)); freed = error = 0; if (flags & PRUS_OOB) { error = EOPNOTSUPP; goto release; } if (control != NULL && (error = unp_internalize(&control, td))) goto release; unp2 = NULL; switch (so->so_type) { case SOCK_DGRAM: { const struct sockaddr *from; if (nam != NULL) { error = unp_connect(so, nam, td); if (error != 0) break; } UNP_PCB_LOCK(unp); /* * Because connect() and send() are non-atomic in a sendto() * with a target address, it's possible that the socket will * have disconnected before the send() can run. In that case * return the slightly counter-intuitive but otherwise * correct error that the socket is not connected. */ unp2 = unp_pcb_lock_peer(unp); if (unp2 == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; break; } if (unp2->unp_flags & UNP_WANTCRED_MASK) control = unp_addsockcred(td, control); if (unp->unp_addr != NULL) from = (struct sockaddr *)unp->unp_addr; else from = &sun_noname; so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) { sorwakeup_locked(so2); m = NULL; control = NULL; } else { SOCKBUF_UNLOCK(&so2->so_rcv); error = ENOBUFS; } if (nam != NULL) unp_disconnect(unp, unp2); else unp_pcb_unlock_pair(unp, unp2); break; } case SOCK_SEQPACKET: case SOCK_STREAM: if ((so->so_state & SS_ISCONNECTED) == 0) { if (nam != NULL) { error = unp_connect(so, nam, td); if (error != 0) break; } else { error = ENOTCONN; break; } } UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) { UNP_PCB_UNLOCK(unp); error = ENOTCONN; break; } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) { unp_pcb_unlock_pair(unp, unp2); error = EPIPE; break; } UNP_PCB_UNLOCK(unp); if ((so2 = unp2->unp_socket) == NULL) { UNP_PCB_UNLOCK(unp2); error = ENOTCONN; break; } SOCKBUF_LOCK(&so2->so_rcv); if (unp2->unp_flags & UNP_WANTCRED_MASK) { /* * Credentials are passed only once on SOCK_STREAM and * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS). */ unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT; control = unp_addsockcred(td, control); } /* * Send to paired receive port and wake up readers. Don't * check for space available in the receive buffer if we're * attaching ancillary data; Unix domain sockets only check * for space in the sending sockbuf, and that check is * performed one level up the stack. At that level we cannot * precisely account for the amount of buffer space used * (e.g., because control messages are not yet internalized). */ switch (so->so_type) { case SOCK_STREAM: if (control != NULL) { sbappendcontrol_locked(&so2->so_rcv, m, control, flags); control = NULL; } else sbappend_locked(&so2->so_rcv, m, flags); break; case SOCK_SEQPACKET: if (sbappendaddr_nospacecheck_locked(&so2->so_rcv, &sun_noname, m, control)) control = NULL; break; } mbcnt = so2->so_rcv.sb_mbcnt; sbcc = sbavail(&so2->so_rcv); if (sbcc) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); /* * The PCB lock on unp2 protects the SB_STOP flag. Without it, * it would be possible for uipc_rcvd to be called at this * point, drain the receiving sockbuf, clear SB_STOP, and then * we would set SB_STOP below. That could lead to an empty * sockbuf having SB_STOP set */ SOCKBUF_LOCK(&so->so_snd); if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax) so->so_snd.sb_flags |= SB_STOP; SOCKBUF_UNLOCK(&so->so_snd); UNP_PCB_UNLOCK(unp2); m = NULL; break; } /* * PRUS_EOF is equivalent to pru_send followed by pru_shutdown. */ if (flags & PRUS_EOF) { UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); } if (control != NULL && error != 0) unp_dispose_mbuf(control); release: if (control != NULL) m_freem(control); /* * In case of PRUS_NOTREADY, uipc_ready() is responsible * for freeing memory. */ if (m != NULL && (flags & PRUS_NOTREADY) == 0) m_freem(m); return (error); } static bool uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp) { struct mbuf *mb, *n; struct sockbuf *sb; SOCK_LOCK(so); if (SOLISTENING(so)) { SOCK_UNLOCK(so); return (false); } mb = NULL; sb = &so->so_rcv; SOCKBUF_LOCK(sb); if (sb->sb_fnrdy != NULL) { for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) { if (mb == m) { *errorp = sbready(sb, m, count); break; } mb = mb->m_next; if (mb == NULL) { mb = n; if (mb != NULL) n = mb->m_nextpkt; } } } SOCKBUF_UNLOCK(sb); SOCK_UNLOCK(so); return (mb != NULL); } static int uipc_ready(struct socket *so, struct mbuf *m, int count) { struct unpcb *unp, *unp2; struct socket *so2; int error, i; unp = sotounpcb(so); KASSERT(so->so_type == SOCK_STREAM, ("%s: unexpected socket type for %p", __func__, so)); UNP_PCB_LOCK(unp); if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { UNP_PCB_UNLOCK(unp); so2 = unp2->unp_socket; SOCKBUF_LOCK(&so2->so_rcv); if ((error = sbready(&so2->so_rcv, m, count)) == 0) sorwakeup_locked(so2); else SOCKBUF_UNLOCK(&so2->so_rcv); UNP_PCB_UNLOCK(unp2); return (error); } UNP_PCB_UNLOCK(unp); /* * The receiving socket has been disconnected, but may still be valid. * In this case, the now-ready mbufs are still present in its socket * buffer, so perform an exhaustive search before giving up and freeing * the mbufs. */ UNP_LINK_RLOCK(); LIST_FOREACH(unp, &unp_shead, unp_link) { if (uipc_ready_scan(unp->unp_socket, m, count, &error)) break; } UNP_LINK_RUNLOCK(); if (unp == NULL) { for (i = 0; i < count; i++) m = m_free(m); error = ECONNRESET; } return (error); } static int uipc_sense(struct socket *so, struct stat *sb) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sense: unp == NULL")); sb->st_blksize = so->so_snd.sb_hiwat; sb->st_dev = NODEV; sb->st_ino = unp->unp_ino; return (0); } static int uipc_shutdown(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL")); UNP_PCB_LOCK(unp); socantsendmore(so); unp_shutdown(unp); UNP_PCB_UNLOCK(unp); return (0); } static int uipc_sockaddr(struct socket *so, struct sockaddr **nam) { struct unpcb *unp; const struct sockaddr *sa; unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL")); *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); UNP_PCB_LOCK(unp); if (unp->unp_addr != NULL) sa = (struct sockaddr *) unp->unp_addr; else sa = &sun_noname; bcopy(sa, *nam, sa->sa_len); UNP_PCB_UNLOCK(unp); return (0); } static struct pr_usrreqs uipc_usrreqs_dgram = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_dgram, .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_seqpacket = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, /* XXX: or...? */ .pru_close = uipc_close, }; static struct pr_usrreqs uipc_usrreqs_stream = { .pru_abort = uipc_abort, .pru_accept = uipc_accept, .pru_attach = uipc_attach, .pru_bind = uipc_bind, .pru_bindat = uipc_bindat, .pru_connect = uipc_connect, .pru_connectat = uipc_connectat, .pru_connect2 = uipc_connect2, .pru_detach = uipc_detach, .pru_disconnect = uipc_disconnect, .pru_listen = uipc_listen, .pru_peeraddr = uipc_peeraddr, .pru_rcvd = uipc_rcvd, .pru_send = uipc_send, .pru_ready = uipc_ready, .pru_sense = uipc_sense, .pru_shutdown = uipc_shutdown, .pru_sockaddr = uipc_sockaddr, .pru_soreceive = soreceive_generic, .pru_close = uipc_close, }; static int uipc_ctloutput(struct socket *so, struct sockopt *sopt) { struct unpcb *unp; struct xucred xu; int error, optval; if (sopt->sopt_level != SOL_LOCAL) return (EINVAL); unp = sotounpcb(so); KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL")); error = 0; switch (sopt->sopt_dir) { case SOPT_GET: switch (sopt->sopt_name) { case LOCAL_PEERCRED: UNP_PCB_LOCK(unp); if (unp->unp_flags & UNP_HAVEPC) xu = unp->unp_peercred; else { if (so->so_type == SOCK_STREAM) error = ENOTCONN; else error = EINVAL; } UNP_PCB_UNLOCK(unp); if (error == 0) error = sooptcopyout(sopt, &xu, sizeof(xu)); break; case LOCAL_CREDS: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CREDS_PERSISTENT: /* Unlocked read. */ optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; case LOCAL_CONNWAIT: /* Unlocked read. */ optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0; error = sooptcopyout(sopt, &optval, sizeof(optval)); break; default: error = EOPNOTSUPP; break; } break; case SOPT_SET: switch (sopt->sopt_name) { case LOCAL_CREDS: case LOCAL_CREDS_PERSISTENT: case LOCAL_CONNWAIT: error = sooptcopyin(sopt, &optval, sizeof(optval), sizeof(optval)); if (error) break; #define OPTSET(bit, exclusive) do { \ UNP_PCB_LOCK(unp); \ if (optval) { \ if ((unp->unp_flags & (exclusive)) != 0) { \ UNP_PCB_UNLOCK(unp); \ error = EINVAL; \ break; \ } \ unp->unp_flags |= (bit); \ } else \ unp->unp_flags &= ~(bit); \ UNP_PCB_UNLOCK(unp); \ } while (0) switch (sopt->sopt_name) { case LOCAL_CREDS: OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS); break; case LOCAL_CREDS_PERSISTENT: OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT); break; case LOCAL_CONNWAIT: OPTSET(UNP_CONNWAIT, 0); break; default: break; } break; #undef OPTSET default: error = ENOPROTOOPT; break; } break; default: error = EOPNOTSUPP; break; } return (error); } static int unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td) { return (unp_connectat(AT_FDCWD, so, nam, td)); } static int unp_connectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td) { struct mtx *vplock; struct sockaddr_un *soun; struct vnode *vp; struct socket *so2; struct unpcb *unp, *unp2, *unp3; struct nameidata nd; char buf[SOCK_MAXADDRLEN]; struct sockaddr *sa; cap_rights_t rights; int error, len; bool connreq; if (nam->sa_family != AF_UNIX) return (EAFNOSUPPORT); if (nam->sa_len > sizeof(struct sockaddr_un)) return (EINVAL); len = nam->sa_len - offsetof(struct sockaddr_un, sun_path); if (len <= 0) return (EINVAL); soun = (struct sockaddr_un *)nam; bcopy(soun->sun_path, buf, len); buf[len] = 0; unp = sotounpcb(so); UNP_PCB_LOCK(unp); for (;;) { /* * Wait for connection state to stabilize. If a connection * already exists, give up. For datagram sockets, which permit * multiple consecutive connect(2) calls, upper layers are * responsible for disconnecting in advance of a subsequent * connect(2), but this is not synchronized with PCB connection * state. * * Also make sure that no threads are currently attempting to * lock the peer socket, to ensure that unp_conn cannot * transition between two valid sockets while locks are dropped. */ if (unp->unp_conn != NULL) { UNP_PCB_UNLOCK(unp); return (EISCONN); } if ((unp->unp_flags & UNP_CONNECTING) != 0) { UNP_PCB_UNLOCK(unp); return (EALREADY); } if (unp->unp_pairbusy > 0) { unp->unp_flags |= UNP_WAITING; mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0); continue; } break; } unp->unp_flags |= UNP_CONNECTING; UNP_PCB_UNLOCK(unp); connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0; if (connreq) sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK); else sa = NULL; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF, UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td); error = namei(&nd); if (error) vp = NULL; else vp = nd.ni_vp; ASSERT_VOP_LOCKED(vp, "unp_connect"); NDFREE(&nd, NDF_ONLY_PNBUF); if (error) goto bad; if (vp->v_type != VSOCK) { error = ENOTSOCK; goto bad; } #ifdef MAC error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD); if (error) goto bad; #endif error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td); if (error) goto bad; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect: unp == NULL")); vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp2); if (unp2 == NULL) { error = ECONNREFUSED; goto bad2; } so2 = unp2->unp_socket; if (so->so_type != so2->so_type) { error = EPROTOTYPE; goto bad2; } if (connreq) { if (so2->so_options & SO_ACCEPTCONN) { CURVNET_SET(so2->so_vnet); so2 = sonewconn(so2, 0); CURVNET_RESTORE(); } else so2 = NULL; if (so2 == NULL) { error = ECONNREFUSED; goto bad2; } unp3 = sotounpcb(so2); unp_pcb_lock_pair(unp2, unp3); if (unp2->unp_addr != NULL) { bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len); unp3->unp_addr = (struct sockaddr_un *) sa; sa = NULL; } unp_copy_peercred(td, unp3, unp, unp2); UNP_PCB_UNLOCK(unp2); unp2 = unp3; /* * It is safe to block on the PCB lock here since unp2 is * nascent and cannot be connected to any other sockets. */ UNP_PCB_LOCK(unp); #ifdef MAC mac_socketpeer_set_from_socket(so, so2); mac_socketpeer_set_from_socket(so2, so); #endif } else { unp_pcb_lock_pair(unp, unp2); } KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 && sotounpcb(so2) == unp2, ("%s: unp2 %p so2 %p", __func__, unp2, so2)); error = unp_connect2(so, so2, PRU_CONNECT); unp_pcb_unlock_pair(unp, unp2); bad2: mtx_unlock(vplock); bad: if (vp != NULL) { vput(vp); } free(sa, M_SONAME); UNP_PCB_LOCK(unp); KASSERT((unp->unp_flags & UNP_CONNECTING) != 0, ("%s: unp %p has UNP_CONNECTING clear", __func__, unp)); unp->unp_flags &= ~UNP_CONNECTING; UNP_PCB_UNLOCK(unp); return (error); } /* * Set socket peer credentials at connection time. * * The client's PCB credentials are copied from its process structure. The * server's PCB credentials are copied from the socket on which it called * listen(2). uipc_listen cached that process's credentials at the time. */ void unp_copy_peercred(struct thread *td, struct unpcb *client_unp, struct unpcb *server_unp, struct unpcb *listen_unp) { cru2xt(td, &client_unp->unp_peercred); client_unp->unp_flags |= UNP_HAVEPC; memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred, sizeof(server_unp->unp_peercred)); server_unp->unp_flags |= UNP_HAVEPC; client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK); } static int unp_connect2(struct socket *so, struct socket *so2, int req) { struct unpcb *unp; struct unpcb *unp2; unp = sotounpcb(so); KASSERT(unp != NULL, ("unp_connect2: unp == NULL")); unp2 = sotounpcb(so2); KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL")); UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp)); if (so2->so_type != so->so_type) return (EPROTOTYPE); unp->unp_conn = unp2; unp_pcb_hold(unp2); unp_pcb_hold(unp); switch (so->so_type) { case SOCK_DGRAM: UNP_REF_LIST_LOCK(); LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink); UNP_REF_LIST_UNLOCK(); soisconnected(so); break; case SOCK_STREAM: case SOCK_SEQPACKET: KASSERT(unp2->unp_conn == NULL, ("%s: socket %p is already connected", __func__, unp2)); unp2->unp_conn = unp; if (req == PRU_CONNECT && ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) soisconnecting(so); else soisconnected(so); soisconnected(so2); break; default: panic("unp_connect2"); } return (0); } static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2) { struct socket *so, *so2; #ifdef INVARIANTS struct unpcb *unptmp; #endif UNP_PCB_LOCK_ASSERT(unp); UNP_PCB_LOCK_ASSERT(unp2); KASSERT(unp->unp_conn == unp2, ("%s: unpcb %p is not connected to %p", __func__, unp, unp2)); unp->unp_conn = NULL; so = unp->unp_socket; so2 = unp2->unp_socket; switch (unp->unp_socket->so_type) { case SOCK_DGRAM: UNP_REF_LIST_LOCK(); #ifdef INVARIANTS LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) { if (unptmp == unp) break; } KASSERT(unptmp != NULL, ("%s: %p not found in reflist of %p", __func__, unp, unp2)); #endif LIST_REMOVE(unp, unp_reflink); UNP_REF_LIST_UNLOCK(); if (so) { SOCK_LOCK(so); so->so_state &= ~SS_ISCONNECTED; SOCK_UNLOCK(so); } break; case SOCK_STREAM: case SOCK_SEQPACKET: if (so) soisdisconnected(so); MPASS(unp2->unp_conn == unp); unp2->unp_conn = NULL; if (so2) soisdisconnected(so2); break; } if (unp == unp2) { unp_pcb_rele_notlast(unp); if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); } else { if (!unp_pcb_rele(unp)) UNP_PCB_UNLOCK(unp); if (!unp_pcb_rele(unp2)) UNP_PCB_UNLOCK(unp2); } } /* * unp_pcblist() walks the global list of struct unpcb's to generate a * pointer list, bumping the refcount on each unpcb. It then copies them out * sequentially, validating the generation number on each to see if it has * been detached. All of this is necessary because copyout() may sleep on * disk I/O. */ static int unp_pcblist(SYSCTL_HANDLER_ARGS) { struct unpcb *unp, **unp_list; unp_gen_t gencnt; struct xunpgen *xug; struct unp_head *head; struct xunpcb *xu; u_int i; int error, n; switch ((intptr_t)arg1) { case SOCK_STREAM: head = &unp_shead; break; case SOCK_DGRAM: head = &unp_dhead; break; case SOCK_SEQPACKET: head = &unp_sphead; break; default: panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1); } /* * The process of preparing the PCB list is too time-consuming and * resource-intensive to repeat twice on every request. */ if (req->oldptr == NULL) { n = unp_count; req->oldidx = 2 * (sizeof *xug) + (n + n/8) * sizeof(struct xunpcb); return (0); } if (req->newptr != NULL) return (EPERM); /* * OK, now we're committed to doing something. */ xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO); UNP_LINK_RLOCK(); gencnt = unp_gencnt; n = unp_count; UNP_LINK_RUNLOCK(); xug->xug_len = sizeof *xug; xug->xug_count = n; xug->xug_gen = gencnt; xug->xug_sogen = so_gencnt; error = SYSCTL_OUT(req, xug, sizeof *xug); if (error) { free(xug, M_TEMP); return (error); } unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK); UNP_LINK_RLOCK(); for (unp = LIST_FIRST(head), i = 0; unp && i < n; unp = LIST_NEXT(unp, unp_link)) { UNP_PCB_LOCK(unp); if (unp->unp_gencnt <= gencnt) { if (cr_cansee(req->td->td_ucred, unp->unp_socket->so_cred)) { UNP_PCB_UNLOCK(unp); continue; } unp_list[i++] = unp; unp_pcb_hold(unp); } UNP_PCB_UNLOCK(unp); } UNP_LINK_RUNLOCK(); n = i; /* In case we lost some during malloc. */ error = 0; xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO); for (i = 0; i < n; i++) { unp = unp_list[i]; UNP_PCB_LOCK(unp); if (unp_pcb_rele(unp)) continue; if (unp->unp_gencnt <= gencnt) { xu->xu_len = sizeof *xu; xu->xu_unpp = (uintptr_t)unp; /* * XXX - need more locking here to protect against * connect/disconnect races for SMP. */ if (unp->unp_addr != NULL) bcopy(unp->unp_addr, &xu->xu_addr, unp->unp_addr->sun_len); else bzero(&xu->xu_addr, sizeof(xu->xu_addr)); if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL) bcopy(unp->unp_conn->unp_addr, &xu->xu_caddr, unp->unp_conn->unp_addr->sun_len); else bzero(&xu->xu_caddr, sizeof(xu->xu_caddr)); xu->unp_vnode = (uintptr_t)unp->unp_vnode; xu->unp_conn = (uintptr_t)unp->unp_conn; xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs); xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink); xu->unp_gencnt = unp->unp_gencnt; sotoxsocket(unp->unp_socket, &xu->xu_socket); UNP_PCB_UNLOCK(unp); error = SYSCTL_OUT(req, xu, sizeof *xu); } else { UNP_PCB_UNLOCK(unp); } } free(xu, M_TEMP); if (!error) { /* * Give the user an updated idea of our state. If the * generation differs from what we told her before, she knows * that something happened while we were processing this * request, and it might be necessary to retry. */ xug->xug_gen = unp_gencnt; xug->xug_sogen = so_gencnt; xug->xug_count = unp_count; error = SYSCTL_OUT(req, xug, sizeof *xug); } free(unp_list, M_TEMP); free(xug, M_TEMP); return (error); } SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb", "List of active local datagram sockets"); SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb", "List of active local stream sockets"); SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb", "List of active local seqpacket sockets"); static void unp_shutdown(struct unpcb *unp) { struct unpcb *unp2; struct socket *so; UNP_PCB_LOCK_ASSERT(unp); unp2 = unp->unp_conn; if ((unp->unp_socket->so_type == SOCK_STREAM || (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) { so = unp2->unp_socket; if (so != NULL) socantrcvmore(so); } } static void unp_drop(struct unpcb *unp) { struct socket *so = unp->unp_socket; struct unpcb *unp2; /* * Regardless of whether the socket's peer dropped the connection * with this socket by aborting or disconnecting, POSIX requires * that ECONNRESET is returned. */ UNP_PCB_LOCK(unp); if (so) so->so_error = ECONNRESET; if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) { /* Last reference dropped in unp_disconnect(). */ unp_pcb_rele_notlast(unp); unp_disconnect(unp, unp2); } else if (!unp_pcb_rele(unp)) { UNP_PCB_UNLOCK(unp); } } static void unp_freerights(struct filedescent **fdep, int fdcount) { struct file *fp; int i; KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount)); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; filecaps_free(&fdep[i]->fde_caps); unp_discard(fp); } free(fdep[0], M_FILECAPS); } static int unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags) { struct thread *td = curthread; /* XXX */ struct cmsghdr *cm = mtod(control, struct cmsghdr *); int i; int *fdp; struct filedesc *fdesc = td->td_proc->p_fd; struct filedescent **fdep; void *data; socklen_t clen = control->m_len, datalen; int error, newfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); error = 0; if (controlp != NULL) /* controlp == NULL => free control messages */ *controlp = NULL; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) { error = EINVAL; break; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { newfds = datalen / sizeof(*fdep); if (newfds == 0) goto next; fdep = data; /* If we're not outputting the descriptors free them. */ if (error || controlp == NULL) { unp_freerights(fdep, newfds); goto next; } FILEDESC_XLOCK(fdesc); /* * Now change each pointer to an fd in the global * table to an integer that is the index to the local * fd table entry that we set up to point to the * global one we are transferring. */ newlen = newfds * sizeof(int); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_XUNLOCK(fdesc); error = E2BIG; unp_freerights(fdep, newfds); goto next; } fdp = (int *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); if (fdallocn(td, 0, fdp, newfds) != 0) { FILEDESC_XUNLOCK(fdesc); error = EMSGSIZE; unp_freerights(fdep, newfds); m_freem(*controlp); *controlp = NULL; goto next; } for (i = 0; i < newfds; i++, fdp++) { _finstall(fdesc, fdep[i]->fde_file, *fdp, (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0, &fdep[i]->fde_caps); unp_externalize_fp(fdep[i]->fde_file); } /* * The new type indicates that the mbuf data refers to * kernel resources that may need to be released before * the mbuf is freed. */ m_chtype(*controlp, MT_EXTCONTROL); FILEDESC_XUNLOCK(fdesc); free(fdep[0], M_FILECAPS); } else { /* We can just copy anything else across. */ if (error || controlp == NULL) goto next; *controlp = sbcreatecontrol(NULL, datalen, cm->cmsg_type, cm->cmsg_level); if (*controlp == NULL) { error = ENOBUFS; goto next; } bcopy(data, CMSG_DATA(mtod(*controlp, struct cmsghdr *)), datalen); } controlp = &(*controlp)->m_next; next: if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } m_freem(control); return (error); } static void unp_zone_change(void *tag) { uma_zone_set_max(unp_zone, maxsockets); } #ifdef INVARIANTS static void unp_zdtor(void *mem, int size __unused, void *arg __unused) { struct unpcb *unp; unp = mem; KASSERT(LIST_EMPTY(&unp->unp_refs), ("%s: unpcb %p has lingering refs", __func__, unp)); KASSERT(unp->unp_socket == NULL, ("%s: unpcb %p has socket backpointer", __func__, unp)); KASSERT(unp->unp_vnode == NULL, ("%s: unpcb %p has vnode references", __func__, unp)); KASSERT(unp->unp_conn == NULL, ("%s: unpcb %p is still connected", __func__, unp)); KASSERT(unp->unp_addr == NULL, ("%s: unpcb %p has leaked addr", __func__, unp)); } #endif static void unp_init(void) { uma_dtor dtor; #ifdef VIMAGE if (!IS_DEFAULT_VNET(curvnet)) return; #endif #ifdef INVARIANTS dtor = unp_zdtor; #else dtor = NULL; #endif unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor, NULL, NULL, UMA_ALIGN_CACHE, 0); uma_zone_set_max(unp_zone, maxsockets); uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached"); EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change, NULL, EVENTHANDLER_PRI_ANY); LIST_INIT(&unp_dhead); LIST_INIT(&unp_shead); LIST_INIT(&unp_sphead); SLIST_INIT(&unp_defers); TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL); TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL); UNP_LINK_LOCK_INIT(); UNP_DEFERRED_LOCK_INIT(); } static void unp_internalize_cleanup_rights(struct mbuf *control) { struct cmsghdr *cp; struct mbuf *m; void *data; socklen_t datalen; for (m = control; m != NULL; m = m->m_next) { cp = mtod(m, struct cmsghdr *); if (cp->cmsg_level != SOL_SOCKET || cp->cmsg_type != SCM_RIGHTS) continue; data = CMSG_DATA(cp); datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data; unp_freerights(data, datalen / sizeof(struct filedesc *)); } } static int unp_internalize(struct mbuf **controlp, struct thread *td) { struct mbuf *control, **initial_controlp; struct proc *p; struct filedesc *fdesc; struct bintime *bt; struct cmsghdr *cm; struct cmsgcred *cmcred; struct filedescent *fde, **fdep, *fdev; struct file *fp; struct timeval *tv; struct timespec *ts; void *data; socklen_t clen, datalen; int i, j, error, *fdp, oldfds; u_int newlen; UNP_LINK_UNLOCK_ASSERT(); p = td->td_proc; fdesc = p->p_fd; error = 0; control = *controlp; clen = control->m_len; *controlp = NULL; initial_controlp = controlp; for (cm = mtod(control, struct cmsghdr *); cm != NULL;) { if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) { error = EINVAL; goto out; } data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; switch (cm->cmsg_type) { /* * Fill in credential information. */ case SCM_CREDS: *controlp = sbcreatecontrol(NULL, sizeof(*cmcred), SCM_CREDS, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } cmcred = (struct cmsgcred *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); cmcred->cmcred_pid = p->p_pid; cmcred->cmcred_uid = td->td_ucred->cr_ruid; cmcred->cmcred_gid = td->td_ucred->cr_rgid; cmcred->cmcred_euid = td->td_ucred->cr_uid; cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); for (i = 0; i < cmcred->cmcred_ngroups; i++) cmcred->cmcred_groups[i] = td->td_ucred->cr_groups[i]; break; case SCM_RIGHTS: oldfds = datalen / sizeof (int); if (oldfds == 0) break; /* * Check that all the FDs passed in refer to legal * files. If not, reject the entire operation. */ fdp = data; FILEDESC_SLOCK(fdesc); for (i = 0; i < oldfds; i++, fdp++) { fp = fget_locked(fdesc, *fdp); if (fp == NULL) { FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) { FILEDESC_SUNLOCK(fdesc); error = EOPNOTSUPP; goto out; } } /* * Now replace the integer FDs with pointers to the * file structure and capability rights. */ newlen = oldfds * sizeof(fdep[0]); *controlp = sbcreatecontrol(NULL, newlen, SCM_RIGHTS, SOL_SOCKET); if (*controlp == NULL) { FILEDESC_SUNLOCK(fdesc); error = E2BIG; goto out; } fdp = data; for (i = 0; i < oldfds; i++, fdp++) { if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) { fdp = data; for (j = 0; j < i; j++, fdp++) { fdrop(fdesc->fd_ofiles[*fdp]. fde_file, td); } FILEDESC_SUNLOCK(fdesc); error = EBADF; goto out; } } fdp = data; fdep = (struct filedescent **) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS, M_WAITOK); for (i = 0; i < oldfds; i++, fdev++, fdp++) { fde = &fdesc->fd_ofiles[*fdp]; fdep[i] = fdev; fdep[i]->fde_file = fde->fde_file; filecaps_copy(&fde->fde_caps, &fdep[i]->fde_caps, true); unp_internalize_fp(fdep[i]->fde_file); } FILEDESC_SUNLOCK(fdesc); break; case SCM_TIMESTAMP: *controlp = sbcreatecontrol(NULL, sizeof(*tv), SCM_TIMESTAMP, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } tv = (struct timeval *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); microtime(tv); break; case SCM_BINTIME: *controlp = sbcreatecontrol(NULL, sizeof(*bt), SCM_BINTIME, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } bt = (struct bintime *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); bintime(bt); break; case SCM_REALTIME: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_REALTIME, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanotime(ts); break; case SCM_MONOTONIC: *controlp = sbcreatecontrol(NULL, sizeof(*ts), SCM_MONOTONIC, SOL_SOCKET); if (*controlp == NULL) { error = ENOBUFS; goto out; } ts = (struct timespec *) CMSG_DATA(mtod(*controlp, struct cmsghdr *)); nanouptime(ts); break; default: error = EINVAL; goto out; } if (*controlp != NULL) controlp = &(*controlp)->m_next; if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } out: if (error != 0 && initial_controlp != NULL) unp_internalize_cleanup_rights(*initial_controlp); m_freem(control); return (error); } static struct mbuf * unp_addsockcred(struct thread *td, struct mbuf *control) { struct mbuf *m, *n, *n_prev; struct sockcred *sc; const struct cmsghdr *cm; int ngroups; int i; ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX); m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET); if (m == NULL) return (control); sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *)); sc->sc_uid = td->td_ucred->cr_ruid; sc->sc_euid = td->td_ucred->cr_uid; sc->sc_gid = td->td_ucred->cr_rgid; sc->sc_egid = td->td_ucred->cr_gid; sc->sc_ngroups = ngroups; for (i = 0; i < sc->sc_ngroups; i++) sc->sc_groups[i] = td->td_ucred->cr_groups[i]; /* * Unlink SCM_CREDS control messages (struct cmsgcred), since just * created SCM_CREDS control message (struct sockcred) has another * format. */ if (control != NULL) for (n = control, n_prev = NULL; n != NULL;) { cm = mtod(n, struct cmsghdr *); if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_CREDS) { if (n_prev == NULL) control = n->m_next; else n_prev->m_next = n->m_next; n = m_free(n); } else { n_prev = n; n = n->m_next; } } /* Prepend it to the head. */ m->m_next = control; return (m); } static struct unpcb * fptounp(struct file *fp) { struct socket *so; if (fp->f_type != DTYPE_SOCKET) return (NULL); if ((so = fp->f_data) == NULL) return (NULL); if (so->so_proto->pr_domain != &localdomain) return (NULL); return sotounpcb(so); } static void unp_discard(struct file *fp) { struct unp_defer *dr; if (unp_externalize_fp(fp)) { dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK); dr->ud_fp = fp; UNP_DEFERRED_LOCK(); SLIST_INSERT_HEAD(&unp_defers, dr, ud_link); UNP_DEFERRED_UNLOCK(); atomic_add_int(&unp_defers_count, 1); taskqueue_enqueue(taskqueue_thread, &unp_defer_task); } else (void) closef(fp, (struct thread *)NULL); } static void unp_process_defers(void *arg __unused, int pending) { struct unp_defer *dr; SLIST_HEAD(, unp_defer) drl; int count; SLIST_INIT(&drl); for (;;) { UNP_DEFERRED_LOCK(); if (SLIST_FIRST(&unp_defers) == NULL) { UNP_DEFERRED_UNLOCK(); break; } SLIST_SWAP(&unp_defers, &drl, unp_defer); UNP_DEFERRED_UNLOCK(); count = 0; while ((dr = SLIST_FIRST(&drl)) != NULL) { SLIST_REMOVE_HEAD(&drl, ud_link); closef(dr->ud_fp, NULL); free(dr, M_TEMP); count++; } atomic_add_int(&unp_defers_count, -count); } } static void unp_internalize_fp(struct file *fp) { struct unpcb *unp; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_file = fp; unp->unp_msgcount++; } unp_rights++; UNP_LINK_WUNLOCK(); } static int unp_externalize_fp(struct file *fp) { struct unpcb *unp; int ret; UNP_LINK_WLOCK(); if ((unp = fptounp(fp)) != NULL) { unp->unp_msgcount--; ret = 1; } else ret = 0; unp_rights--; UNP_LINK_WUNLOCK(); return (ret); } /* * unp_defer indicates whether additional work has been defered for a future * pass through unp_gc(). It is thread local and does not require explicit * synchronization. */ static int unp_marked; static void unp_remove_dead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs--; } } static void unp_restore_undead_ref(struct filedescent **fdep, int fdcount) { struct unpcb *unp; struct file *fp; int i; /* * This function can only be called from the gc task. */ KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0, ("%s: not on gc callout", __func__)); UNP_LINK_LOCK_ASSERT(); for (i = 0; i < fdcount; i++) { fp = fdep[i]->fde_file; if ((unp = fptounp(fp)) == NULL) continue; if ((unp->unp_gcflag & UNPGC_DEAD) == 0) continue; unp->unp_gcrefs++; unp_marked++; } } static void unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int)) { struct socket *so, *soa; so = unp->unp_socket; SOCK_LOCK(so); if (SOLISTENING(so)) { /* * Mark all sockets in our accept queue. */ TAILQ_FOREACH(soa, &so->sol_comp, so_list) { if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) continue; SOCKBUF_LOCK(&soa->so_rcv); unp_scan(soa->so_rcv.sb_mb, op); SOCKBUF_UNLOCK(&soa->so_rcv); } } else { /* * Mark all sockets we reference with RIGHTS. */ if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) { SOCKBUF_LOCK(&so->so_rcv); unp_scan(so->so_rcv.sb_mb, op); SOCKBUF_UNLOCK(&so->so_rcv); } } SOCK_UNLOCK(so); } static int unp_recycled; SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "Number of unreachable sockets claimed by the garbage collector."); static int unp_taskcount; SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "Number of times the garbage collector has run."); SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0, "Number of active local sockets."); static void unp_gc(__unused void *arg, int pending) { struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead, NULL }; struct unp_head **head; struct unp_head unp_deadhead; /* List of potentially-dead sockets. */ struct file *f, **unref; struct unpcb *unp, *unptmp; int i, total, unp_unreachable; LIST_INIT(&unp_deadhead); unp_taskcount++; UNP_LINK_RLOCK(); /* * First determine which sockets may be in cycles. */ unp_unreachable = 0; for (head = heads; *head != NULL; head++) LIST_FOREACH(unp, *head, unp_link) { KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0, ("%s: unp %p has unexpected gc flags 0x%x", __func__, unp, (unsigned int)unp->unp_gcflag)); f = unp->unp_file; /* * Check for an unreachable socket potentially in a * cycle. It must be in a queue as indicated by * msgcount, and this must equal the file reference * count. Note that when msgcount is 0 the file is * NULL. */ if (f != NULL && unp->unp_msgcount != 0 && refcount_load(&f->f_count) == unp->unp_msgcount) { LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead); unp->unp_gcflag |= UNPGC_DEAD; unp->unp_gcrefs = unp->unp_msgcount; unp_unreachable++; } } /* * Scan all sockets previously marked as potentially being in a cycle * and remove the references each socket holds on any UNPGC_DEAD * sockets in its queue. After this step, all remaining references on * sockets marked UNPGC_DEAD should not be part of any cycle. */ LIST_FOREACH(unp, &unp_deadhead, unp_dead) unp_gc_scan(unp, unp_remove_dead_ref); /* * If a socket still has a non-negative refcount, it cannot be in a * cycle. In this case increment refcount of all children iteratively. * Stop the scan once we do a complete loop without discovering * a new reachable socket. */ do { unp_marked = 0; LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp) if (unp->unp_gcrefs > 0) { unp->unp_gcflag &= ~UNPGC_DEAD; LIST_REMOVE(unp, unp_dead); KASSERT(unp_unreachable > 0, ("%s: unp_unreachable underflow.", __func__)); unp_unreachable--; unp_gc_scan(unp, unp_restore_undead_ref); } } while (unp_marked); UNP_LINK_RUNLOCK(); if (unp_unreachable == 0) return; /* * Allocate space for a local array of dead unpcbs. * TODO: can this path be simplified by instead using the local * dead list at unp_deadhead, after taking out references * on the file object and/or unpcb and dropping the link lock? */ unref = malloc(unp_unreachable * sizeof(struct file *), M_TEMP, M_WAITOK); /* * Iterate looking for sockets which have been specifically marked * as unreachable and store them locally. */ UNP_LINK_RLOCK(); total = 0; LIST_FOREACH(unp, &unp_deadhead, unp_dead) { KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0, ("%s: unp %p not marked UNPGC_DEAD", __func__, unp)); unp->unp_gcflag &= ~UNPGC_DEAD; f = unp->unp_file; if (unp->unp_msgcount == 0 || f == NULL || refcount_load(&f->f_count) != unp->unp_msgcount || !fhold(f)) continue; unref[total++] = f; KASSERT(total <= unp_unreachable, ("%s: incorrect unreachable count.", __func__)); } UNP_LINK_RUNLOCK(); /* * Now flush all sockets, free'ing rights. This will free the * struct files associated with these sockets but leave each socket * with one remaining ref. */ for (i = 0; i < total; i++) { struct socket *so; so = unref[i]->f_data; CURVNET_SET(so->so_vnet); sorflush(so); CURVNET_RESTORE(); } /* * And finally release the sockets so they can be reclaimed. */ for (i = 0; i < total; i++) fdrop(unref[i], NULL); unp_recycled += total; free(unref, M_TEMP); } static void unp_dispose_mbuf(struct mbuf *m) { if (m) unp_scan(m, unp_freerights); } /* * Synchronize against unp_gc, which can trip over data as we are freeing it. */ static void unp_dispose(struct socket *so) { struct unpcb *unp; unp = sotounpcb(so); UNP_LINK_WLOCK(); unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS; UNP_LINK_WUNLOCK(); if (!SOLISTENING(so)) unp_dispose_mbuf(so->so_rcv.sb_mb); } static void unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int)) { struct mbuf *m; struct cmsghdr *cm; void *data; socklen_t clen, datalen; while (m0 != NULL) { for (m = m0; m; m = m->m_next) { if (m->m_type != MT_CONTROL) continue; cm = mtod(m, struct cmsghdr *); clen = m->m_len; while (cm != NULL) { if (sizeof(*cm) > clen || cm->cmsg_len > clen) break; data = CMSG_DATA(cm); datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data; if (cm->cmsg_level == SOL_SOCKET && cm->cmsg_type == SCM_RIGHTS) { (*op)(data, datalen / sizeof(struct filedescent *)); } if (CMSG_SPACE(datalen) < clen) { clen -= CMSG_SPACE(datalen); cm = (struct cmsghdr *) ((caddr_t)cm + CMSG_SPACE(datalen)); } else { clen = 0; cm = NULL; } } } m0 = m0->m_nextpkt; } } /* * A helper function called by VFS before socket-type vnode reclamation. * For an active vnode it clears unp_vnode pointer and decrements unp_vnode * use count. */ void vfs_unp_reclaim(struct vnode *vp) { struct unpcb *unp; int active; struct mtx *vplock; ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim"); KASSERT(vp->v_type == VSOCK, ("vfs_unp_reclaim: vp->v_type != VSOCK")); active = 0; vplock = mtx_pool_find(mtxpool_sleep, vp); mtx_lock(vplock); VOP_UNP_CONNECT(vp, &unp); if (unp == NULL) goto done; UNP_PCB_LOCK(unp); if (unp->unp_vnode == vp) { VOP_UNP_DETACH(vp); unp->unp_vnode = NULL; active = 1; } UNP_PCB_UNLOCK(unp); done: mtx_unlock(vplock); if (active) vunref(vp); } #ifdef DDB static void db_print_indent(int indent) { int i; for (i = 0; i < indent; i++) db_printf(" "); } static void db_print_unpflags(int unp_flags) { int comma; comma = 0; if (unp_flags & UNP_HAVEPC) { db_printf("%sUNP_HAVEPC", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ALWAYS) { db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_WANTCRED_ONESHOT) { db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNWAIT) { db_printf("%sUNP_CONNWAIT", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_CONNECTING) { db_printf("%sUNP_CONNECTING", comma ? ", " : ""); comma = 1; } if (unp_flags & UNP_BINDING) { db_printf("%sUNP_BINDING", comma ? ", " : ""); comma = 1; } } static void db_print_xucred(int indent, struct xucred *xu) { int comma, i; db_print_indent(indent); db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n", xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups); db_print_indent(indent); db_printf("cr_groups: "); comma = 0; for (i = 0; i < xu->cr_ngroups; i++) { db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]); comma = 1; } db_printf("\n"); } static void db_print_unprefs(int indent, struct unp_head *uh) { struct unpcb *unp; int counter; counter = 0; LIST_FOREACH(unp, uh, unp_reflink) { if (counter % 4 == 0) db_print_indent(indent); db_printf("%p ", unp); if (counter % 4 == 3) db_printf("\n"); counter++; } if (counter != 0 && counter % 4 != 0) db_printf("\n"); } DB_SHOW_COMMAND(unpcb, db_show_unpcb) { struct unpcb *unp; if (!have_addr) { db_printf("usage: show unpcb \n"); return; } unp = (struct unpcb *)addr; db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket, unp->unp_vnode); db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino, unp->unp_conn); db_printf("unp_refs:\n"); db_print_unprefs(2, &unp->unp_refs); /* XXXRW: Would be nice to print the full address, if any. */ db_printf("unp_addr: %p\n", unp->unp_addr); db_printf("unp_gencnt: %llu\n", (unsigned long long)unp->unp_gencnt); db_printf("unp_flags: %x (", unp->unp_flags); db_print_unpflags(unp->unp_flags); db_printf(")\n"); db_printf("unp_peercred:\n"); db_print_xucred(2, &unp->unp_peercred); db_printf("unp_refcount: %u\n", unp->unp_refcount); } #endif Index: head/sys/kern/vfs_subr.c =================================================================== --- head/sys/kern/vfs_subr.c (revision 367631) +++ head/sys/kern/vfs_subr.c (revision 367632) @@ -1,6795 +1,6800 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void vgonel(struct vnode *); static bool vhold_recycle_free(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_lock(void *arg, int what); static void destroy_vpollinfo(struct vpollinfo *vi); static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn); static void vnlru_recalc(void); /* * These fences are intended for cases where some synchronization is * needed between access of v_iflags and lockless vnode refcount (v_holdcnt * and v_usecount) updates. Access to v_iflags is generally synchronized * by the interlock, but we have some internal assertions that check vnode * flags without acquiring the lock. Thus, these fences are INVARIANTS-only * for now. */ #ifdef INVARIANTS #define VNODE_REFCOUNT_FENCE_ACQ() atomic_thread_fence_acq() #define VNODE_REFCOUNT_FENCE_REL() atomic_thread_fence_rel() #else #define VNODE_REFCOUNT_FENCE_ACQ() #define VNODE_REFCOUNT_FENCE_REL() #endif /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VIRF_DOOMED vnode. */ static u_long __exclusive_cache_line numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of allocates vnodes in the system. */ static TAILQ_HEAD(freelst, vnode) vnode_list; static struct vnode *vnode_list_free_marker; static struct vnode *vnode_list_reclaim_marker; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static long wantfreevnodes; static long __exclusive_cache_line freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static long freevnodes_old; static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); static counter_u64_t recycles_free_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles_free, CTLFLAG_RD, &recycles_free_count, "Number of free vnodes recycled to meet vnode cache targets"); static counter_u64_t deferred_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, deferred_inact, CTLFLAG_RD, &deferred_inact, "Number of times inactive processing was deferred"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_list * numvnodes * freevnodes */ static struct mtx __exclusive_cache_line vnode_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; static smr_t buf_trie_smr; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; MALLOC_DEFINE(M_VNODEPOLL, "VN POLL", "vnode poll"); __read_frequently smr_t vfs_smr; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); #define VDBATCH_SIZE 8 struct vdbatch { u_int index; long freevnodes; struct mtx lock; struct vnode *tab[VDBATCH_SIZE]; }; DPCPU_DEFINE_STATIC(struct vdbatch, vd); static void vdbatch_dequeue(struct vnode *vp); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ u_long desiredvnodes; static u_long gapvnodes; /* gap between wanted and desired */ static u_long vhiwat; /* enough extras after expansion */ static u_long vlowat; /* minimal extras before expansion */ static u_long vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static u_long vnlru_read_freevnodes(void); /* * Note that no attempt is made to sanitize these parameters. */ static int sysctl_maxvnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = desiredvnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == desiredvnodes) return (0); mtx_lock(&vnode_list_mtx); desiredvnodes = val; wantfreevnodes = desiredvnodes / 4; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); /* * XXX There is no protection against multiple threads changing * desiredvnodes at the same time. Locking above only helps vnlru and * getnewvnode. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_maxvnodes, "LU", "Target for maximum number of vnodes"); static int sysctl_wantfreevnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = wantfreevnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == wantfreevnodes) return (0); mtx_lock(&vnode_list_mtx); wantfreevnodes = val; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); return (0); } SYSCTL_PROC(_vfs, OID_AUTO, wantfreevnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_wantfreevnodes, "LU", "Target for minimum number of \"free\" vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); static int sysctl_try_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct vnode *vp; struct nameidata nd; char *buf; unsigned long ndflags; int error; if (req->newptr == NULL) return (EINVAL); if (req->newlen >= PATH_MAX) return (E2BIG); buf = malloc(PATH_MAX, M_TEMP, M_WAITOK); error = SYSCTL_IN(req, buf, req->newlen); if (error != 0) goto out; buf[req->newlen] = '\0'; ndflags = LOCKLEAF | NOFOLLOW | AUDITVNODE1 | SAVENAME; NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, buf, curthread); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; if (VN_IS_DOOMED(vp)) { /* * This vnode is being recycled. Return != 0 to let the caller * know that the sysctl had no effect. Return EAGAIN because a * subsequent call will likely succeed (since namei will create * a new vnode if necessary) */ error = EAGAIN; goto putvnode; } counter_u64_add(recycles_count, 1); vgone(vp); putvnode: NDFREE(&nd, 0); out: free(buf, M_TEMP); return (error); } static int sysctl_ftry_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct vnode *vp; struct file *fp; int error; int fd; if (req->newptr == NULL) return (EBADF); error = sysctl_handle_int(oidp, &fd, 0, req); if (error != 0) return (error); error = getvnode(curthread, fd, &cap_fcntl_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) goto drop; counter_u64_add(recycles_count, 1); vgone(vp); VOP_UNLOCK(vp); drop: fdrop(fp, td); return (error); } SYSCTL_PROC(_debug, OID_AUTO, try_reclaim_vnode, CTLTYPE_STRING | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_try_reclaim_vnode, "A", "Try to reclaim a vnode by its pathname"); SYSCTL_PROC(_debug, OID_AUTO, ftry_reclaim_vnode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_ftry_reclaim_vnode, "I", "Try to reclaim a vnode by its file descriptor"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return (uma_zalloc_smr(buf_trie_zone, M_NOWAIT)); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree_smr(buf_trie_zone, node); } PCTRIE_DEFINE_SMR(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free, buf_trie_smr); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512UL * 1024 * 1024 / 64) /* 8M */ #endif static MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); static struct vnode * vn_alloc_marker(struct mount *mp) { struct vnode *vp; vp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); vp->v_type = VMARKER; vp->v_mount = mp; return (vp); } static void vn_free_marker(struct vnode *vp) { MPASS(vp->v_type == VMARKER); free(vp, M_VNODE_MARKER); } /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ cache_vnode_init(vp); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); vp->v_dbatchcpu = NOCPU; /* * Check vhold_recycle_free for an explanation. */ vp->v_holdcnt = VHOLD_NO_SMR; vp->v_type = VNON; mtx_lock(&vnode_list_mtx); TAILQ_INSERT_BEFORE(vnode_list_free_marker, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; vdbatch_dequeue(vp); mtx_lock(&vnode_list_mtx); TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { struct vdbatch *vd; int cpu, physvnodes, virtvnodes; u_int i; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %lu -> %lu\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_list); mtx_init(&vnode_list_mtx, "vnode_list", NULL, MTX_DEF); /* * The lock is taken to appease WITNESS. */ mtx_lock(&vnode_list_mtx); vnlru_recalc(); mtx_unlock(&vnode_list_mtx); vnode_list_free_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_free_marker, v_vnodelist); vnode_list_reclaim_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_reclaim_marker, v_vnodelist); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); uma_zone_set_smr(vnode_zone, vfs_smr); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE | UMA_ZONE_SMR); buf_trie_smr = uma_zone_get_smr(buf_trie_zone); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); recycles_free_count = counter_u64_alloc(M_WAITOK); deferred_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); bzero(vd, sizeof(*vd)); mtx_init(&vd->lock, "vdbatch", NULL, MTX_DEF); } } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { struct mount_pcpu *mpcpu; MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); if (vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MPASS((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0); MPASS((mp->mnt_kern_flag & MNTK_REFEXPIRE) == 0); vfs_mp_count_add_pcpu(mpcpu, ref, 1); vfs_mp_count_add_pcpu(mpcpu, lockref, 1); vfs_op_thread_exit(mp, mpcpu); if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); return (0); } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { struct mount_pcpu *mpcpu; int c; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if (vfs_op_thread_enter(mp, mpcpu)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); vfs_mp_count_sub_pcpu(mpcpu, lockref, 1); vfs_mp_count_sub_pcpu(mpcpu, ref, 1); vfs_op_thread_exit(mp, mpcpu); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_lockref; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if (c == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || fsidcmp(&mp->mnt_stat.f_fsid, fsid) != 0) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (fsidcmp(&mp->mnt_stat.f_fsid, fsid) == 0) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * Try to reduce the total number of vnodes. * * This routine (and its user) are buggy in at least the following ways: * - all parameters were picked years ago when RAM sizes were significantly * smaller * - it can pick vnodes based on pages used by the vm object, but filesystems * like ZFS don't use it making the pick broken * - since ZFS has its own aging policy it gets partially combated by this one * - a dedicated method should be provided for filesystems to let them decide * whether the vnode should be recycled * * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. * * @param reclaim_nc_src Only reclaim directories with outgoing namecache * entries if this argument is strue * @param trigger Only reclaim vnodes with fewer than this many resident * pages. * @param target How many vnodes to reclaim. * @return The number of vnodes that were reclaimed. */ static int vlrureclaim(bool reclaim_nc_src, int trigger, u_long target) { struct vnode *vp, *mvp; struct mount *mp; struct vm_object *object; u_long done; bool retried; mtx_assert(&vnode_list_mtx, MA_OWNED); retried = false; done = 0; mvp = vnode_list_reclaim_marker; restart: vp = mvp; while (done < target) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) break; if (__predict_false(vp->v_type == VMARKER)) continue; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src))) goto next_iter; if (vp->v_type == VBAD || vp->v_type == VNON) goto next_iter; object = atomic_load_ptr(&vp->v_object); if (object == NULL || object->resident_page_count > trigger) { goto next_iter; } /* * Handle races against vnode allocation. Filesystems lock the * vnode some time after it gets returned from getnewvnode, * despite type and hold count being manipulated earlier. * Resorting to checking v_mount restores guarantees present * before the global list was reworked to contain all vnodes. */ if (!VI_TRYLOCK(vp)) goto next_iter; if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { VI_UNLOCK(vp); goto next_iter; } if (vp->v_mount == NULL) { VI_UNLOCK(vp); goto next_iter; } vholdl(vp); VI_UNLOCK(vp); TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); goto next_iter_unlocked; } if (VOP_LOCK(vp, LK_EXCLUSIVE|LK_NOWAIT) != 0) { vdrop(vp); vn_finished_write(mp); goto next_iter_unlocked; } VI_LOCK(vp); if (vp->v_usecount > 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); goto next_iter_unlocked; } counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); done++; next_iter_unlocked: if (should_yield()) kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; next_iter: MPASS(vp->v_type != VMARKER); if (!should_yield()) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; } if (done == 0 && !retried) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_HEAD(&vnode_list, mvp, v_vnodelist); retried = true; goto restart; } return (done); } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static int vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp, *mvp; struct mount *mp; int ocount; mtx_assert(&vnode_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; ocount = count; mvp = vnode_list_free_marker; vp = mvp; for (;;) { if (count == 0) { break; } vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, mvp, v_vnodelist); break; } if (__predict_false(vp->v_type == VMARKER)) continue; if (vp->v_holdcnt > 0) continue; /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. */ if (mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) { continue; } if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { continue; } if (!vhold_recycle_free(vp)) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vtryrecycle(vp) == 0) count--; mtx_lock(&vnode_list_mtx); vp = mvp; } return (ocount - count); } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_list_mtx); } static void vnlru_recalc(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; /* * The main freevnodes counter is only updated when threads requeue their vnode * batches. CPUs are conditionally walked to compute a more accurate total. * * Limit how much of a slop are we willing to tolerate. Note: the actual value * at any given moment can still exceed slop, but it should not be by significant * margin in practice. */ #define VNLRU_FREEVNODES_SLOP 128 static __inline void vn_freevnodes_inc(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes++; critical_exit(); } static __inline void vn_freevnodes_dec(void) { struct vdbatch *vd; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes--; critical_exit(); } static u_long vnlru_read_freevnodes(void) { struct vdbatch *vd; long slop; int cpu; mtx_assert(&vnode_list_mtx, MA_OWNED); if (freevnodes > freevnodes_old) slop = freevnodes - freevnodes_old; else slop = freevnodes_old - freevnodes; if (slop < VNLRU_FREEVNODES_SLOP) return (freevnodes >= 0 ? freevnodes : 0); freevnodes_old = freevnodes; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); freevnodes_old += vd->freevnodes; } return (freevnodes_old >= 0 ? freevnodes_old : 0); } static bool vnlru_under(u_long rnumvnodes, u_long limit) { u_long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = vnlru_read_freevnodes(); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static bool vnlru_under_unlocked(u_long rnumvnodes, u_long limit) { long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = atomic_load_long(&freevnodes); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static void vnlru_kick(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); if (vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } static void vnlru_proc(void) { u_long rnumvnodes, rfreevnodes, target; unsigned long onumvnodes; int done, force, trigger, usevnodes; bool reclaim_nc_src, want_reread; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; want_reread = false; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (want_reread) { force = vnlru_under(numvnodes, vhiwat) ? 1 : 0; want_reread = false; } /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (rnumvnodes > desiredvnodes) { vnlru_free_locked(rnumvnodes - desiredvnodes, NULL); rnumvnodes = atomic_load_long(&numvnodes); } /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (force == 0 && !vnlru_under(rnumvnodes, vlowat)) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } rfreevnodes = vnlru_read_freevnodes(); onumvnodes = rnumvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (rnumvnodes <= desiredvnodes) usevnodes = rnumvnodes - rfreevnodes; else usevnodes = rnumvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; target = rnumvnodes * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; done = vlrureclaim(reclaim_nc_src, trigger, target); mtx_unlock(&vnode_list_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(UMA_RECLAIM_DRAIN); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } want_reread = true; force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else { want_reread = true; kern_yield(PRI_USER); } } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); vdrop(vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); vdrop(vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if (!VN_IS_DOOMED(vp)) { counter_u64_add(recycles_free_count, 1); vgonel(vp); } VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(vnmp); return (0); } /* * Allocate a new vnode. * * The operation never returns an error. Returning an error was disabled * in r145385 (dated 2005) with the following comment: * * XXX Not all VFS_VGET/ffs_vget callers check returns. * * Given the age of this commit (almost 15 years at the time of writing this * comment) restoring the ability to fail requires a significant audit of * all codepaths. * * The routine can try to free a vnode or stall for up to 1 second waiting for * vnlru to clear things up, but ultimately always performs a M_WAITOK allocation. */ static u_long vn_alloc_cyclecount; static struct vnode * __noinline vn_alloc_hard(struct mount *mp) { u_long rnumvnodes, rfreevnodes; mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (rnumvnodes + 1 < desiredvnodes) { vn_alloc_cyclecount = 0; goto alloc; } rfreevnodes = vnlru_read_freevnodes(); if (vn_alloc_cyclecount++ >= rfreevnodes) { vn_alloc_cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (vnlru_free_locked(1, NULL) > 0) goto alloc; if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPEND) == 0) { /* * Wait for space for a new vnode. */ vnlru_kick(); msleep(&vnlruproc_sig, &vnode_list_mtx, PVFS, "vlruwk", hz); if (atomic_load_long(&numvnodes) + 1 > desiredvnodes && vnlru_read_freevnodes() > 1) vnlru_free_locked(1, NULL); } alloc: rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (vnlru_under(rnumvnodes, vlowat)) vnlru_kick(); mtx_unlock(&vnode_list_mtx); return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static struct vnode * vn_alloc(struct mount *mp) { u_long rnumvnodes; if (__predict_false(vn_alloc_cyclecount != 0)) return (vn_alloc_hard(mp)); rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (__predict_false(vnlru_under_unlocked(rnumvnodes, vlowat))) { atomic_subtract_long(&numvnodes, 1); return (vn_alloc_hard(mp)); } return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } static void vn_free(struct vnode *vp) { atomic_subtract_long(&numvnodes, 1); uma_zfree_smr(vnode_zone, vp); } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); KASSERT(vops->registered, ("%s: not registered vector op %p\n", __func__, vops)); td = curthread; if (td->td_vp_reserved != NULL) { vp = td->td_vp_reserved; td->td_vp_reserved = NULL; } else { vp = vn_alloc(mp); } counter_u64_add(vnodes_created, 1); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; #ifdef WITNESS if (lo->lo_name != tag) { #endif lo->lo_name = tag; #ifdef WITNESS WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } #endif /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } void getnewvnode_reserve(void) { struct thread *td; td = curthread; MPASS(td->td_vp_reserved == NULL); td->td_vp_reserved = vn_alloc(NULL); } void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; if (td->td_vp_reserved != NULL) { vn_free(td->td_vp_reserved); td->td_vp_reserved = NULL; } } static void __noinline freevnode(struct vnode *vp) { struct bufobj *bo; /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); /* * Paired with vgone. */ vn_seqc_write_end_locked(vp); VNPASS(vp->v_seqc_users == 0, vp); bo = &vp->v_bufobj; VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNPASS(vp->v_holdcnt == VHOLD_NO_SMR, vp); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VNASSERT((vp->v_iflag & (VI_DOINGINACT | VI_OWEINACT)) == 0, vp, ("Leaked inactivation")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_irflag = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; vn_free(vp); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); - if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) + do { + error = BO_SYNC(bo, MNT_WAIT); + } while (error == ERELOOKUP); + if (error != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); if ((flags & V_VMIO) == 0 && bo->bo_object != NULL) { BO_UNLOCK(bo); vm_object_pip_wait_unlocked(bo->bo_object, "bovlbx"); BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { /* * If we are flushing both V_NORMAL and V_ALT buffers then * do not skip any buffers. If we are flushing only V_NORMAL * buffers then skip buffers marked as BX_ALTDATA. If we are * flushing only V_ALT buffers then skip buffers not marked * as BX_ALTDATA. */ if (((flags & (V_NORMAL | V_ALT)) != (V_NORMAL | V_ALT)) && (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA) != 0) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, off_t length, int blksize) { struct buf *bp, *nbp; struct bufobj *bo; daddr_t startlbn; CTR4(KTR_VFS, "%s: vp %p with block %d:%ju", __func__, vp, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ startlbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); bo = &vp->v_bufobj; restart_unlocked: BO_LOCK(bo); while (v_inval_buf_range_locked(vp, bo, startlbn, INT64_MAX) == EAGAIN) ; if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart_unlocked; VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } /* * Invalidate the cached pages of a file's buffer within the range of block * numbers [startlbn, endlbn). */ void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize) { struct bufobj *bo; off_t start, end; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range"); start = blksize * startlbn; end = blksize * endlbn; bo = &vp->v_bufobj; BO_LOCK(bo); MPASS(blksize == bo->bo_bsize); while (v_inval_buf_range_locked(vp, bo, startlbn, endlbn) == EAGAIN) ; BO_UNLOCK(bo); vn_pages_remove(vp, OFF_TO_IDX(start), OFF_TO_IDX(end + PAGE_SIZE - 1)); } static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn) { struct buf *bp, *nbp; bool anyfreed; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range_locked"); ASSERT_BO_LOCKED(bo); do { anyfreed = false; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || nbp->b_vp != vp || (nbp->b_flags & B_DELWRI) != 0)) return (EAGAIN); } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) return (EAGAIN); } } while (anyfreed); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; b_xflags_t flags; flags = bp->b_xflags; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((flags & (BX_VNDIRTY | BX_VNCLEAN)) != 0 && (flags & (BX_VNDIRTY | BX_VNCLEAN)) != (BX_VNDIRTY | BX_VNCLEAN), ("%s: buffer %p has invalid queue state", __func__, bp)); if ((flags & BX_VNDIRTY) != 0) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); KASSERT((bo->bo_flag & BO_NOBUFS) == 0, ("buf_vlist_add: bo %p does not allow bufs", bo)); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno)); } /* * Look up a buf using the buffer tries, without the bufobj lock. This relies * on SMR for safe lookup, and bufs being in a no-free zone to provide type * stability of the result. Like other lockless lookups, the found buf may * already be invalid by the time this function returns. */ struct buf * gbincore_unlocked(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_UNLOCKED(bo); bp = BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return (BUF_PCTRIE_LOOKUP_UNLOCKED(&bo->bo_dirty.bv_root, lblkno)); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); buf_vlist_remove(bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_MPSAFE| CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Move the buffer between the clean and dirty lists of its vnode. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; KASSERT((bp->b_flags & B_PAGING) == 0, ("%s: cannot reassign paging buffer %p", __func__, bp)); CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); BO_LOCK(bo); buf_vlist_remove(bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VIRF_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * usecount is manipulated using atomics without holding any locks. * * holdcnt can be manipulated using atomics without holding any locks, * except when transitioning 1<->0, in which case the interlock is held. * * Consumers which don't guarantee liveness of the vnode can use SMR to * try to get a reference. Note this operation can fail since the vnode * may be awaiting getting freed by the time they get to it. */ enum vgetstate vget_prep_smr(struct vnode *vp) { enum vgetstate vs; VFS_SMR_ASSERT_ENTERED(); if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { if (vhold_smr(vp)) vs = VGET_HOLDCNT; else vs = VGET_NONE; } return (vs); } enum vgetstate vget_prep(struct vnode *vp) { enum vgetstate vs; if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { vhold(vp); vs = VGET_HOLDCNT; } return (vs); } void vget_abort(struct vnode *vp, enum vgetstate vs) { switch (vs) { case VGET_USECOUNT: vrele(vp); break; case VGET_HOLDCNT: vdrop(vp); break; default: __assert_unreachable(); } } int vget(struct vnode *vp, int flags) { enum vgetstate vs; vs = vget_prep(vp); return (vget_finish(vp, flags, vs)); } int vget_finish(struct vnode *vp, int flags, enum vgetstate vs) { int error; if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); error = vn_lock(vp, flags); if (__predict_false(error != 0)) { vget_abort(vp, vs); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } vget_finish_ref(vp, vs); return (0); } void vget_finish_ref(struct vnode *vp, enum vgetstate vs) { int old; VNPASS(vs == VGET_HOLDCNT || vs == VGET_USECOUNT, vp); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); if (vs == VGET_USECOUNT) return; /* * We hold the vnode. If the usecount is 0 it will be utilized to keep * the vnode around. Otherwise someone else lended their hold count and * we have to drop ours. */ old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old >= 0, vp, ("%s: wrong use count %d", __func__, old)); if (old != 0) { #ifdef INVARIANTS old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif } } void vref(struct vnode *vp) { enum vgetstate vs; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vs = vget_prep(vp); vget_finish_ref(vp, vs); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count %d", __func__, old)); #else refcount_acquire(&vp->v_usecount); #endif } void vlazy(struct vnode *vp) { struct mount *mp; VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode not held", __func__)); if ((vp->v_mflag & VMP_LAZYLIST) != 0) return; /* * We may get here for inactive routines after the vnode got doomed. */ if (VN_IS_DOOMED(vp)) return; mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_LAZYLIST) == 0) { vp->v_mflag |= VMP_LAZYLIST; TAILQ_INSERT_TAIL(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize++; } mtx_unlock(&mp->mnt_listmtx); } /* * This routine is only meant to be called from vgonel prior to dooming * the vnode. */ static void vunlazy_gone(struct vnode *vp) { struct mount *mp; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); VNPASS(!VN_IS_DOOMED(vp), vp); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } } static void vdefer_inactive(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode without hold count", __func__)); if (VN_IS_DOOMED(vp)) { vdropl(vp); return; } if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vdropl(vp); return; } if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; vdropl(vp); return; } vlazy(vp); vp->v_iflag |= VI_DEFINACT; VI_UNLOCK(vp); counter_u64_add(deferred_inact, 1); } static void vdefer_inactive_unlocked(struct vnode *vp) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } vdefer_inactive(vp); } enum vput_op { VRELE, VPUT, VUNREF }; /* * Handle ->v_usecount transitioning to 0. * * By releasing the last usecount we take ownership of the hold count which * provides liveness of the vnode, meaning we have to vdrop. * * For all vnodes we may need to perform inactive processing. It requires an * exclusive lock on the vnode, while it is legal to call here with only a * shared lock (or no locks). If locking the vnode in an expected manner fails, * inactive processing gets deferred to the syncer. * * XXX Some filesystems pass in an exclusively locked vnode and strongly depend * on the lock being held all the way until VOP_INACTIVE. This in particular * happens with UFS which adds half-constructed vnodes to the hash, where they * can be found by other code. */ static void vput_final(struct vnode *vp, enum vput_op func) { int error; bool want_unlock; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNPASS(vp->v_holdcnt > 0, vp); VI_LOCK(vp); /* * By the time we got here someone else might have transitioned * the count back to > 0. */ if (vp->v_usecount > 0) goto out; /* * If the vnode is doomed vgone already performed inactive processing * (if needed). */ if (VN_IS_DOOMED(vp)) goto out; if (__predict_true(VOP_NEED_INACTIVE(vp) == 0)) goto out; if (vp->v_iflag & VI_DOINGINACT) goto out; /* * Locking operations here will drop the interlock and possibly the * vnode lock, opening a window where the vnode can get doomed all the * while ->v_usecount is 0. Set VI_OWEINACT to let vgone know to * perform inactive. */ vp->v_iflag |= VI_OWEINACT; want_unlock = false; error = 0; switch (func) { case VRELE: switch (VOP_ISLOCKED(vp)) { case LK_EXCLUSIVE: break; case LK_EXCLOTHER: case 0: want_unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; default: /* * The lock has at least one sharer, but we have no way * to conclude whether this is us. Play it safe and * defer processing. */ error = EAGAIN; break; } break; case VPUT: want_unlock = true; if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } if (error == 0) { vinactive(vp); if (want_unlock) VOP_UNLOCK(vp); vdropl(vp); } else { vdefer_inactive(vp); } return; out: if (func == VPUT) VOP_UNLOCK(vp); vdropl(vp); } /* * Decrement ->v_usecount for a vnode. * * Releasing the last use count requires additional processing, see vput_final * above for details. * * Comment above each variant denotes lock state on entry and exit. */ /* * in: any * out: same as passed in */ void vrele(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VRELE); } /* * in: locked * out: unlocked */ void vput(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) { VOP_UNLOCK(vp); return; } vput_final(vp, VPUT); } /* * in: locked * out: locked */ void vunref(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VUNREF); } void vhold(struct vnode *vp) { int old; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old >= 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); if (old == 0) vn_freevnodes_dec(); } void vholdnz(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0 && (old & VHOLD_ALL_FLAGS) == 0, vp, ("%s: wrong hold count %d", __func__, old)); #else atomic_add_int(&vp->v_holdcnt, 1); #endif } /* * Grab a hold count unless the vnode is freed. * * Only use this routine if vfs smr is the only protection you have against * freeing the vnode. * * The code loops trying to add a hold count as long as the VHOLD_NO_SMR flag * is not set. After the flag is set the vnode becomes immutable to anyone but * the thread which managed to set the flag. * * It may be tempting to replace the loop with: * count = atomic_fetchadd_int(&vp->v_holdcnt, 1); * if (count & VHOLD_NO_SMR) { * backpedal and error out; * } * * However, while this is more performant, it hinders debugging by eliminating * the previously mentioned invariant. */ bool vhold_smr(struct vnode *vp) { int count; VFS_SMR_ASSERT_ENTERED(); count = atomic_load_int(&vp->v_holdcnt); for (;;) { if (count & VHOLD_NO_SMR) { VNASSERT((count & ~VHOLD_NO_SMR) == 0, vp, ("non-zero hold count with flags %d\n", count)); return (false); } VNASSERT(count >= 0, vp, ("invalid hold count %d\n", count)); if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { if (count == 0) vn_freevnodes_dec(); return (true); } } } /* * Hold a free vnode for recycling. * * Note: vnode_init references this comment. * * Attempts to recycle only need the global vnode list lock and have no use for * SMR. * * However, vnodes get inserted into the global list before they get fully * initialized and stay there until UMA decides to free the memory. This in * particular means the target can be found before it becomes usable and after * it becomes recycled. Picking up such vnodes is guarded with v_holdcnt set to * VHOLD_NO_SMR. * * Note: the vnode may gain more references after we transition the count 0->1. */ static bool vhold_recycle_free(struct vnode *vp) { int count; mtx_assert(&vnode_list_mtx, MA_OWNED); count = atomic_load_int(&vp->v_holdcnt); for (;;) { if (count & VHOLD_NO_SMR) { VNASSERT((count & ~VHOLD_NO_SMR) == 0, vp, ("non-zero hold count with flags %d\n", count)); return (false); } VNASSERT(count >= 0, vp, ("invalid hold count %d\n", count)); if (count > 0) { return (false); } if (atomic_fcmpset_int(&vp->v_holdcnt, &count, count + 1)) { vn_freevnodes_dec(); return (true); } } } static void __noinline vdbatch_process(struct vdbatch *vd) { struct vnode *vp; int i; mtx_assert(&vd->lock, MA_OWNED); MPASS(curthread->td_pinned > 0); MPASS(vd->index == VDBATCH_SIZE); mtx_lock(&vnode_list_mtx); critical_enter(); freevnodes += vd->freevnodes; for (i = 0; i < VDBATCH_SIZE; i++) { vp = vd->tab[i]; TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist); MPASS(vp->v_dbatchcpu != NOCPU); vp->v_dbatchcpu = NOCPU; } mtx_unlock(&vnode_list_mtx); vd->freevnodes = 0; bzero(vd->tab, sizeof(vd->tab)); vd->index = 0; critical_exit(); } static void vdbatch_enqueue(struct vnode *vp) { struct vdbatch *vd; ASSERT_VI_LOCKED(vp, __func__); VNASSERT(!VN_IS_DOOMED(vp), vp, ("%s: deferring requeue of a doomed vnode", __func__)); if (vp->v_dbatchcpu != NOCPU) { VI_UNLOCK(vp); return; } sched_pin(); vd = DPCPU_PTR(vd); mtx_lock(&vd->lock); MPASS(vd->index < VDBATCH_SIZE); MPASS(vd->tab[vd->index] == NULL); /* * A hack: we depend on being pinned so that we know what to put in * ->v_dbatchcpu. */ vp->v_dbatchcpu = curcpu; vd->tab[vd->index] = vp; vd->index++; VI_UNLOCK(vp); if (vd->index == VDBATCH_SIZE) vdbatch_process(vd); mtx_unlock(&vd->lock); sched_unpin(); } /* * This routine must only be called for vnodes which are about to be * deallocated. Supporting dequeue for arbitrary vndoes would require * validating that the locked batch matches. */ static void vdbatch_dequeue(struct vnode *vp) { struct vdbatch *vd; int i; short cpu; VNASSERT(vp->v_type == VBAD || vp->v_type == VNON, vp, ("%s: called for a used vnode\n", __func__)); cpu = vp->v_dbatchcpu; if (cpu == NOCPU) return; vd = DPCPU_ID_PTR(cpu, vd); mtx_lock(&vd->lock); for (i = 0; i < vd->index; i++) { if (vd->tab[i] != vp) continue; vp->v_dbatchcpu = NOCPU; vd->index--; vd->tab[i] = vd->tab[vd->index]; vd->tab[vd->index] = NULL; break; } mtx_unlock(&vd->lock); /* * Either we dequeued the vnode above or the target CPU beat us to it. */ MPASS(vp->v_dbatchcpu == NOCPU); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VIRF_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ static void vdrop_deactivate(struct vnode *vp) { struct mount *mp; ASSERT_VI_LOCKED(vp, __func__); /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(!VN_IS_DOOMED(vp), vp, ("vdrop: returning doomed vnode")); VNASSERT(vp->v_op != NULL, vp, ("vdrop: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with VI_OWEINACT set")); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("vnode with VI_DEFINACT set")); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNASSERT(vp->v_mflag & VMP_LAZYLIST, vp, ("lost VMP_LAZYLIST")); /* * Don't remove the vnode from the lazy list if another thread * has increased the hold count. It may have re-enqueued the * vnode to the lazy list and is now responsible for its * removal. */ if (vp->v_holdcnt == 0) { vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; } mtx_unlock(&mp->mnt_listmtx); } vdbatch_enqueue(vp); } static void __noinline vdropl_final(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(VN_IS_DOOMED(vp), vp); /* * Set the VHOLD_NO_SMR flag. * * We may be racing against vhold_smr. If they win we can just pretend * we never got this far, they will vdrop later. */ if (__predict_false(!atomic_cmpset_int(&vp->v_holdcnt, 0, VHOLD_NO_SMR))) { vn_freevnodes_inc(); VI_UNLOCK(vp); /* * We lost the aforementioned race. Any subsequent access is * invalid as they might have managed to vdropl on their own. */ return; } /* * Don't bump freevnodes as this one is going away. */ freevnode(vp); } void vdrop(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); vdropl(vp); } void vdropl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!refcount_release(&vp->v_holdcnt)) { VI_UNLOCK(vp); return; } if (!VN_IS_DOOMED(vp)) { vn_freevnodes_inc(); vdrop_deactivate(vp); /* * Also unlocks the interlock. We can't assert on it as we * released our hold and by now the vnode might have been * freed. */ return; } vdropl_final(vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. */ static void vinactivef(struct vnode *vp) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } void vinactive(struct vnode *vp) { ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((vp->v_iflag & VI_OWEINACT) == 0) return; if (vp->v_iflag & VI_DOINGINACT) return; if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; return; } vinactivef(vp); } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } - error = VOP_FSYNC(vp, MNT_WAIT, td); + do { + error = VOP_FSYNC(vp, MNT_WAIT, td); + } while (error == ERELOOKUP); if (error != 0) { VOP_UNLOCK(vp); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount <= 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; if (TAILQ_EMPTY(&mp->mnt_uppers)) return; mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; struct mount *mp; vm_object_t object; bool active, doinginact, oweinact; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_irflag & VIRF_DOOMED) return; /* * Paired with freevnode. */ vn_seqc_write_begin_locked(vp); vunlazy_gone(vp); vp->v_irflag |= VIRF_DOOMED; /* * Check to see if the vnode is in use. If so, we have to * call VOP_CLOSE() and VOP_INACTIVE(). * * It could be that VOP_INACTIVE() requested reclamation, in * which case we should avoid recursion, so check * VI_DOINGINACT. This is not precise but good enough. */ active = vp->v_usecount > 0; oweinact = (vp->v_iflag & VI_OWEINACT) != 0; doinginact = (vp->v_iflag & VI_DOINGINACT) != 0; /* * If we need to do inactive VI_OWEINACT will be set. */ if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vp->v_iflag &= ~VI_DEFINACT; vdropl(vp); } else { VNASSERT(vp->v_holdcnt > 0, vp, ("vnode without hold count")); VI_UNLOCK(vp); } cache_purge_vgone(vp); vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if ((oweinact || active) && !doinginact) { VI_LOCK(vp); vinactivef(vp); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set later, or in * vm_object_terminate() after the object's page queue is * flushed. */ object = vp->v_bufobj.bo_object; if (object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Handle the VM part. Tmpfs handles v_object on its own (the * OBJT_VNODE check). Nullfs or other bypassing filesystems * should not touch the object borrowed from the lower vnode * (the handle check). */ if (object != NULL && object->type == OBJT_VNODE && object->handle == vp) vnode_destroy_vobject(vp); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p", vp)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_type = VBAD; } /* * Print out a description of a vnode. */ static const char * const typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; _Static_assert((VHOLD_ALL_FLAGS & ~VHOLD_NO_SMR) == 0, "new hold count flag not added to vn_printf"); void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; u_int holdcnt; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("type %s\n", typename[vp->v_type]); holdcnt = atomic_load_int(&vp->v_holdcnt); printf(" usecount %d, writecount %d, refcount %d seqc users %d", vp->v_usecount, vp->v_writecount, holdcnt & ~VHOLD_ALL_FLAGS, vp->v_seqc_users); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (holdcnt & VHOLD_NO_SMR) strlcat(buf, "|VHOLD_NO_SMR", sizeof(buf)); printf(" hold count flags (%s)\n", buf + 1); buf[0] = '\0'; buf[1] = '\0'; if (vp->v_irflag & VIRF_DOOMED) strlcat(buf, "|VIRF_DOOMED", sizeof(buf)); if (vp->v_irflag & VIRF_PGREAD) strlcat(buf, "|VIRF_PGREAD", sizeof(buf)); flags = vp->v_irflag & ~(VIRF_DOOMED | VIRF_PGREAD); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VIRF(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_VMSIZEVNLOCK) strlcat(buf, "|VV_VMSIZEVNLOCK", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); if (vp->v_vflag & VV_READLINK) strlcat(buf, "|VV_READLINK", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_TEXT_REF) strlcat(buf, "|VI_TEXT_REF", sizeof(buf)); if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); if (vp->v_iflag & VI_DEFINACT) strlcat(buf, "|VI_DEFINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_TEXT_REF | VI_MOUNT | VI_DOINGINACT | VI_OWEINACT | VI_DEFINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_mflag & VMP_LAZYLIST) strlcat(buf, "|VMP_LAZYLIST", sizeof(buf)); flags = vp->v_mflag & ~(VMP_LAZYLIST); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VMP(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_FPLOOKUP); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_REF), mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_lazyvnodelistsize = %d\n", mp->mnt_lazyvnodelistsize); db_printf(" mnt_writeopcount = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_WRITEOPCOUNT), mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_LOCKREF), mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf(" mnt_vfs_ops = %d\n", mp->mnt_vfs_ops); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt > 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } static void vfs_deferred_inactive(struct vnode *vp, int lkflags) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("VI_DEFINACT still set")); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } if (vn_lock(vp, lkflags) == 0) { VI_LOCK(vp); vinactive(vp); VOP_UNLOCK(vp); vdropl(vp); return; } vdefer_inactive_unlocked(vp); } static int vfs_periodic_inactive_filter(struct vnode *vp, void *arg) { return (vp->v_iflag & VI_DEFINACT); } static void __noinline vfs_periodic_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; int lkflags; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) lkflags |= LK_NOWAIT; MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_inactive_filter, NULL) { if ((vp->v_iflag & VI_DEFINACT) == 0) { VI_UNLOCK(vp); continue; } vp->v_iflag &= ~VI_DEFINACT; vfs_deferred_inactive(vp, lkflags); } } static inline bool vfs_want_msync(struct vnode *vp) { struct vm_object *obj; /* * This test may be performed without any locks held. * We rely on vm_object's type stability. */ if (vp->v_vflag & VV_NOSYNC) return (false); obj = vp->v_object; return (obj != NULL && vm_object_mightbedirty(obj)); } static int vfs_periodic_msync_inactive_filter(struct vnode *vp, void *arg __unused) { if (vp->v_vflag & VV_NOSYNC) return (false); if (vp->v_iflag & VI_DEFINACT) return (true); return (vfs_want_msync(vp)); } static void __noinline vfs_periodic_msync_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; int lkflags, objflags; bool seen_defer; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) { lkflags |= LK_NOWAIT; objflags = OBJPC_NOSYNC; } else { objflags = OBJPC_SYNC; } MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_msync_inactive_filter, NULL) { seen_defer = false; if (vp->v_iflag & VI_DEFINACT) { vp->v_iflag &= ~VI_DEFINACT; seen_defer = true; } if (!vfs_want_msync(vp)) { if (seen_defer) vfs_deferred_inactive(vp, lkflags); else VI_UNLOCK(vp); continue; } if (vget(vp, lkflags) == 0) { obj = vp->v_object; if (obj != NULL && (vp->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, objflags); VM_OBJECT_WUNLOCK(obj); } vput(vp); if (seen_defer) vdrop(vp); } else { if (seen_defer) vdefer_inactive_unlocked(vp); } } } void vfs_periodic(struct mount *mp, int flags) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if ((mp->mnt_kern_flag & MNTK_NOMSYNC) != 0) vfs_periodic_inactive(mp, flags); else vfs_periodic_msync_inactive(mp, flags); } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); free(vi, M_VNODEPOLL); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = malloc(sizeof(*vi), M_VNODEPOLL, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_lock); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_need_inactive = vop_stdneed_inactive, /* need_inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; VFS_VOP_VECTOR_REGISTER(sync_vnodeops); /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); /* * The filesystem at hand may be idle with free vnodes stored in the * batch. Return them instead of letting them stay there indefinitely. */ vfs_periodic(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } int vn_need_pageq_flush(struct vnode *vp) { struct vm_object *obj; int need; MPASS(mtx_owned(VI_MTX(vp))); need = 0; if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) need = 1; return (need); } /* * Check if vnode represents a disk device */ bool vn_isdisk_error(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: *errp = error; return (error == 0); } bool vn_isdisk(struct vnode *vp) { int error; return (vn_isdisk_error(vp, &error)); } /* * VOP_FPLOOKUP_VEXEC routines are subject to special circumstances, see * the comment above cache_fplookup for details. */ int vaccess_vexec_smr(mode_t file_mode, uid_t file_uid, gid_t file_gid, struct ucred *cred) { int error; VFS_SMR_ASSERT_ENTERED(); /* Check the owner. */ if (cred->cr_uid == file_uid) { if (file_mode & S_IXUSR) return (0); goto out_error; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) return (0); goto out_error; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) return (0); out_error: /* * Permission check failed, but it is possible denial will get overwritten * (e.g., when root is traversing through a 700 directory owned by someone * else). * * vaccess() calls priv_check_cred which in turn can descent into MAC * modules overriding this result. It's quite unclear what semantics * are allowed for them to operate, thus for safety we don't call them * from within the SMR section. This also means if any such modules * are present, we have to let the regular lookup decide. */ error = priv_check_cred_vfs_lookup_nomac(cred); switch (error) { case 0: return (0); case EAGAIN: /* * MAC modules present. */ return (EAGAIN); case EPERM: return (EACCES); default: return (error); } } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, and credentials. * Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (KERNEL_PANICKED() || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif /* * It may be tempting to add vn_seqc_write_begin/end calls here and * in vop_rename_post but that's not going to work out since some * filesystems relookup vnodes mid-rename. This is probably a bug. * * For now filesystems are expected to do the relevant calls after they * decide what vnodes to operate on. */ if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_fplookup_vexec_debugpre(void *ap __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_fplookup_vexec_debugpost(void *ap __unused, int rc __unused) { VFS_SMR_ASSERT_ENTERED(); } void vop_strategy_debugpre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (!KERNEL_PANICKED() && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_debugpre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_debugpost(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_debugpre(void *ap) { struct vop_unlock_args *a = ap; ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_need_inactive_debugpre(void *ap) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } void vop_need_inactive_debugpost(void *ap, int rc) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } #endif void vop_create_pre(void *ap) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_create_post(void *ap, int rc) { struct vop_create_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_whiteout_pre(void *ap) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_whiteout_post(void *ap, int rc) { struct vop_whiteout_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); } void vop_deleteextattr_pre(void *ap) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_pre(void *ap) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_begin(vp); vn_seqc_write_begin(tdvp); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a; struct vnode *vp, *tdvp; a = ap; vp = a->a_vp; tdvp = a->a_tdvp; vn_seqc_write_end(vp); vn_seqc_write_end(tdvp); if (!rc) { VFS_KNOTE_LOCKED(vp, NOTE_LINK); VFS_KNOTE_LOCKED(tdvp, NOTE_WRITE); } } void vop_mkdir_pre(void *ap) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); } #ifdef DEBUG_VFS_LOCKS void vop_mkdir_debugpost(void *ap, int rc) { struct vop_mkdir_args *a; a = ap; if (!rc) cache_validate(a->a_dvp, *a->a_vpp, a->a_cnp); } #endif void vop_mknod_pre(void *ap) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a; struct vnode *vp; a = ap; vp = a->a_vp; ASSERT_VOP_IN_SEQC(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_REVOKE); } void vop_remove_pre(void *ap) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_pre(void *ap) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_begin(dvp); vn_seqc_write_begin(vp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a; struct vnode *dvp, *vp; a = ap; dvp = a->a_dvp; vp = a->a_vp; vn_seqc_write_end(dvp); vn_seqc_write_end(vp); if (!rc) { VFS_KNOTE_LOCKED(dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(vp, NOTE_DELETE); } } void vop_setattr_pre(void *ap) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_setacl_pre(void *ap) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setacl_post(void *ap, int rc __unused) { struct vop_setacl_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); } void vop_setextattr_pre(void *ap) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_begin(vp); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a; struct vnode *vp; a = ap; vp = a->a_vp; vn_seqc_write_end(vp); if (!rc) VFS_KNOTE_LOCKED(vp, NOTE_ATTRIB); } void vop_symlink_pre(void *ap) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_begin(dvp); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a; struct vnode *dvp; a = ap; dvp = a->a_dvp; vn_seqc_write_end(dvp); if (!rc) VFS_KNOTE_LOCKED(dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ !VN_IS_DOOMED(a->a_vp))) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_read_pgcache_post(void *ap, int rc) { struct vop_read_pgcache_args *a = ap; if (!rc) VFS_KNOTE_UNLOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp); } static void vfs_knl_assert_lock(void *arg, int what) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; if (what == LA_LOCKED) ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); else ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } /* * Returns whether the directory is empty or not. * If it is empty, the return value is 0; otherwise * the return value is an error value (which may * be ENOTEMPTY). */ int vfs_emptydir(struct vnode *vp) { struct uio uio; struct iovec iov; struct dirent *dirent, *dp, *endp; int error, eof; error = 0; eof = 0; ASSERT_VOP_LOCKED(vp, "vfs_emptydir"); dirent = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK); iov.iov_base = dirent; iov.iov_len = sizeof(struct dirent); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = sizeof(struct dirent); uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = curthread; while (eof == 0 && error == 0) { error = VOP_READDIR(vp, &uio, curthread->td_ucred, &eof, NULL, NULL); if (error != 0) break; endp = (void *)((uint8_t *)dirent + sizeof(struct dirent) - uio.uio_resid); for (dp = dirent; dp < endp; dp = (void *)((uint8_t *)dp + GENERIC_DIRSIZ(dp))) { if (dp->d_type == DT_WHT) continue; if (dp->d_namlen == 0) continue; if (dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) { error = ENOTEMPTY; break; } if (dp->d_namlen > 2) { error = ENOTEMPTY; break; } if (dp->d_namlen == 1 && dp->d_name[0] != '.') { error = ENOTEMPTY; break; } if (dp->d_namlen == 2 && dp->d_name[1] != '.') { error = ENOTEMPTY; break; } uio.uio_resid = sizeof(struct dirent); } } free(dirent, M_TEMP); return (error); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * Clear out a doomed vnode (if any) and replace it with a new one as long * as the fs is not being unmounted. Return the root vnode to the caller. */ static int __noinline vfs_cache_root_fallback(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; restart: if (mp->mnt_rootvnode != NULL) { MNT_ILOCK(mp); vp = mp->mnt_rootvnode; if (vp != NULL) { if (!VN_IS_DOOMED(vp)) { vrefact(vp); MNT_IUNLOCK(mp); error = vn_lock(vp, flags); if (error == 0) { *vpp = vp; return (0); } vrele(vp); goto restart; } /* * Clear the old one. */ mp->mnt_rootvnode = NULL; } MNT_IUNLOCK(mp); if (vp != NULL) { vfs_op_barrier_wait(mp); vrele(vp); } } error = VFS_CACHEDROOT(mp, flags, vpp); if (error != 0) return (error); if (mp->mnt_vfs_ops == 0) { MNT_ILOCK(mp); if (mp->mnt_vfs_ops != 0) { MNT_IUNLOCK(mp); return (0); } if (mp->mnt_rootvnode == NULL) { vrefact(*vpp); mp->mnt_rootvnode = *vpp; } else { if (mp->mnt_rootvnode != *vpp) { if (!VN_IS_DOOMED(mp->mnt_rootvnode)) { panic("%s: mismatch between vnode returned " " by VFS_CACHEDROOT and the one cached " " (%p != %p)", __func__, *vpp, mp->mnt_rootvnode); } } } MNT_IUNLOCK(mp); } return (0); } int vfs_cache_root(struct mount *mp, int flags, struct vnode **vpp) { struct mount_pcpu *mpcpu; struct vnode *vp; int error; if (!vfs_op_thread_enter(mp, mpcpu)) return (vfs_cache_root_fallback(mp, flags, vpp)); vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL || VN_IS_DOOMED(vp)) { vfs_op_thread_exit(mp, mpcpu); return (vfs_cache_root_fallback(mp, flags, vpp)); } vrefact(vp); vfs_op_thread_exit(mp, mpcpu); error = vn_lock(vp, flags); if (error != 0) { vrele(vp); return (vfs_cache_root_fallback(mp, flags, vpp)); } *vpp = vp; return (0); } struct vnode * vfs_cache_root_clear(struct mount *mp) { struct vnode *vp; /* * ops > 0 guarantees there is nobody who can see this vnode */ MPASS(mp->mnt_vfs_ops > 0); vp = mp->mnt_rootvnode; if (vp != NULL) vn_seqc_write_begin(vp); mp->mnt_rootvnode = NULL; return (vp); } void vfs_cache_root_set(struct mount *mp, struct vnode *vp) { MPASS(mp->mnt_vfs_ops > 0); vrefact(vp); mp->mnt_rootvnode = vp; } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * lazy vnodes. See MNT_VNODE_FOREACH_LAZY() in sys/mount.h */ static void mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_lazy iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_lazy_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_lazylist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, mvp, v_lazylist); TAILQ_INSERT_BEFORE(vp, mvp, v_lazylist); /* * Note we may be racing against vdrop which transitioned the hold * count to 0 and now waits for the ->mnt_listmtx lock. This is fine, * if we are the only user after we get the interlock we will just * vdrop. */ vhold(vp); mtx_unlock(&mp->mnt_listmtx); VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); goto out_lost; } VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); /* * There is nothing to do if we are the last user. */ if (!refcount_release_if_not_last(&vp->v_holdcnt)) goto out_lost; mtx_lock(&mp->mnt_listmtx); return (true); out_lost: vdropl(vp); maybe_yield(); mtx_lock(&mp->mnt_listmtx); return (false); } static struct vnode * mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_lazylist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } /* * See if we want to process the vnode. Note we may encounter a * long string of vnodes we don't care about and hog the list * as a result. Check for it and requeue the marker. */ VNPASS(!VN_IS_DOOMED(vp), vp); if (!cb(vp, cbarg)) { if (!should_yield()) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); goto restart; } /* * Try-lock because this is the wrong lock order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_lazy_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the lazy list %p %p", vp, mp)); VNPASS(vp->v_mount == mp, vp); VNPASS(!VN_IS_DOOMED(vp), vp); break; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "lazy iter"); return (vp); } struct vnode * __mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } struct vnode * __mnt_vnode_first_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; if (TAILQ_EMPTY(&mp->mnt_lazyvnodelist)) return (NULL); *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_lazyvnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_lazylist); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } void __mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); } int vn_dir_check_exec(struct vnode *vp, struct componentname *cnp) { if ((cnp->cn_flags & NOEXECCHECK) != 0) { cnp->cn_flags &= ~NOEXECCHECK; return (0); } return (VOP_ACCESS(vp, VEXEC, cnp->cn_cred, cnp->cn_thread)); } /* * Do not use this variant unless you have means other than the hold count * to prevent the vnode from getting freed. */ void vn_seqc_write_begin_unheld_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_seqc_users >= 0, vp); vp->v_seqc_users++; if (vp->v_seqc_users == 1) seqc_sleepable_write_begin(&vp->v_seqc); } void vn_seqc_write_begin_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_holdcnt > 0, vp); vn_seqc_write_begin_unheld_locked(vp); } void vn_seqc_write_begin(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_begin_locked(vp); VI_UNLOCK(vp); } void vn_seqc_write_begin_unheld(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_begin_unheld_locked(vp); VI_UNLOCK(vp); } void vn_seqc_write_end_locked(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNPASS(vp->v_seqc_users > 0, vp); vp->v_seqc_users--; if (vp->v_seqc_users == 0) seqc_sleepable_write_end(&vp->v_seqc); } void vn_seqc_write_end(struct vnode *vp) { VI_LOCK(vp); vn_seqc_write_end_locked(vp); VI_UNLOCK(vp); } Index: head/sys/kern/vfs_syscalls.c =================================================================== --- head/sys/kern/vfs_syscalls.c (revision 367631) +++ head/sys/kern/vfs_syscalls.c (revision 367632) @@ -1,4930 +1,4953 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_syscalls.c 8.13 (Berkeley) 4/15/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include #include #include #include #include MALLOC_DEFINE(M_FADVISE, "fadvise", "posix_fadvise(2) information"); static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag); static int setfflags(struct thread *td, struct vnode *, u_long); static int getutimes(const struct timeval *, enum uio_seg, struct timespec *); static int getutimens(const struct timespec *, enum uio_seg, struct timespec *, int *); static int setutimes(struct thread *td, struct vnode *, const struct timespec *, int, int); static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td); static int kern_fhlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp); static int kern_getfhat(struct thread *td, int flags, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp); static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count, struct thread *td); static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path, enum uio_seg segflag); static uint64_t at2cnpflags(u_int at_flags, u_int mask) { u_int64_t res; MPASS((at_flags & (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)) != (AT_SYMLINK_FOLLOW | AT_SYMLINK_NOFOLLOW)); res = 0; at_flags &= mask; if ((at_flags & AT_BENEATH) != 0) res |= BENEATH; if ((at_flags & AT_RESOLVE_BENEATH) != 0) res |= RBENEATH; if ((at_flags & AT_SYMLINK_FOLLOW) != 0) res |= FOLLOW; /* NOFOLLOW is pseudo flag */ if ((mask & AT_SYMLINK_NOFOLLOW) != 0) { res |= (at_flags & AT_SYMLINK_NOFOLLOW) != 0 ? NOFOLLOW : FOLLOW; } return (res); } int kern_sync(struct thread *td) { struct mount *mp, *nmp; int save; mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } if ((mp->mnt_flag & MNT_RDONLY) == 0 && vn_start_write(NULL, &mp, V_NOWAIT) == 0) { save = curthread_pflags_set(TDP_SYNCIO); vfs_periodic(mp, MNT_NOWAIT); VFS_SYNC(mp, MNT_NOWAIT); curthread_pflags_restore(save); vn_finished_write(mp); } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); return (0); } /* * Sync each mounted filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct sync_args { int dummy; }; #endif /* ARGSUSED */ int sys_sync(struct thread *td, struct sync_args *uap) { return (kern_sync(td)); } /* * Change filesystem quotas. */ #ifndef _SYS_SYSPROTO_H_ struct quotactl_args { char *path; int cmd; int uid; caddr_t arg; }; #endif int sys_quotactl(struct thread *td, struct quotactl_args *uap) { struct mount *mp; struct nameidata nd; int error; AUDIT_ARG_CMD(uap->cmd); AUDIT_ARG_UID(uap->uid); if (!prison_allow(td->td_ucred, PR_ALLOW_QUOTAS)) return (EPERM); NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); mp = nd.ni_vp->v_mount; vfs_ref(mp); vput(nd.ni_vp); error = vfs_busy(mp, 0); if (error != 0) { vfs_rel(mp); return (error); } error = VFS_QUOTACTL(mp, uap->cmd, uap->uid, uap->arg); /* * Since quota on operation typically needs to open quota * file, the Q_QUOTAON handler needs to unbusy the mount point * before calling into namei. Otherwise, unmount might be * started between two vfs_busy() invocations (first is our, * second is from mount point cross-walk code in lookup()), * causing deadlock. * * Require that Q_QUOTAON handles the vfs_busy() reference on * its own, always returning with ubusied mount point. */ if ((uap->cmd >> SUBCMDSHIFT) != Q_QUOTAON && (uap->cmd >> SUBCMDSHIFT) != Q_QUOTAOFF) vfs_unbusy(mp); vfs_rel(mp); return (error); } /* * Used by statfs conversion routines to scale the block size up if * necessary so that all of the block counts are <= 'max_size'. Note * that 'max_size' should be a bitmask, i.e. 2^n - 1 for some non-zero * value of 'n'. */ void statfs_scale_blocks(struct statfs *sf, long max_size) { uint64_t count; int shift; KASSERT(powerof2(max_size + 1), ("%s: invalid max_size", __func__)); /* * Attempt to scale the block counts to give a more accurate * overview to userland of the ratio of free space to used * space. To do this, find the largest block count and compute * a divisor that lets it fit into a signed integer <= max_size. */ if (sf->f_bavail < 0) count = -sf->f_bavail; else count = sf->f_bavail; count = MAX(sf->f_blocks, MAX(sf->f_bfree, count)); if (count <= max_size) return; count >>= flsl(max_size); shift = 0; while (count > 0) { shift++; count >>=1; } sf->f_bsize <<= shift; sf->f_blocks >>= shift; sf->f_bfree >>= shift; sf->f_bavail >>= shift; } static int kern_do_statfs(struct thread *td, struct mount *mp, struct statfs *buf) { int error; if (mp == NULL) return (EBADF); error = vfs_busy(mp, 0); vfs_rel(mp); if (error != 0) return (error); #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif error = VFS_STATFS(mp, buf); if (error != 0) goto out; if (priv_check_cred_vfs_generation(td->td_ucred)) { buf->f_fsid.val[0] = buf->f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, buf); } out: vfs_unbusy(mp); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct statfs_args { char *path; struct statfs *buf; }; #endif int sys_statfs(struct thread *td, struct statfs_args *uap) { struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(struct statfs)); free(sfp, M_STATFS); return (error); } int kern_statfs(struct thread *td, const char *path, enum uio_seg pathseg, struct statfs *buf) { struct mount *mp; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); error = namei(&nd); if (error != 0) return (error); mp = nd.ni_vp->v_mount; vfs_ref(mp); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); return (kern_do_statfs(td, mp, buf)); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct fstatfs_args { int fd; struct statfs *buf; }; #endif int sys_fstatfs(struct thread *td, struct fstatfs_args *uap) { struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(struct statfs)); free(sfp, M_STATFS); return (error); } int kern_fstatfs(struct thread *td, int fd, struct statfs *buf) { struct file *fp; struct mount *mp; struct vnode *vp; int error; AUDIT_ARG_FD(fd); error = getvnode(td, fd, &cap_fstatfs_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; vn_lock(vp, LK_SHARED | LK_RETRY); #ifdef AUDIT AUDIT_ARG_VNODE1(vp); #endif mp = vp->v_mount; if (mp != NULL) vfs_ref(mp); VOP_UNLOCK(vp); fdrop(fp, td); return (kern_do_statfs(td, mp, buf)); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct getfsstat_args { struct statfs *buf; long bufsize; int mode; }; #endif int sys_getfsstat(struct thread *td, struct getfsstat_args *uap) { size_t count; int error; if (uap->bufsize < 0 || uap->bufsize > SIZE_MAX) return (EINVAL); error = kern_getfsstat(td, &uap->buf, uap->bufsize, &count, UIO_USERSPACE, uap->mode); if (error == 0) td->td_retval[0] = count; return (error); } /* * If (bufsize > 0 && bufseg == UIO_SYSSPACE) * The caller is responsible for freeing memory which will be allocated * in '*buf'. */ int kern_getfsstat(struct thread *td, struct statfs **buf, size_t bufsize, size_t *countp, enum uio_seg bufseg, int mode) { struct mount *mp, *nmp; struct statfs *sfsp, *sp, *sptmp, *tofree; size_t count, maxcount; int error; switch (mode) { case MNT_WAIT: case MNT_NOWAIT: break; default: if (bufseg == UIO_SYSSPACE) *buf = NULL; return (EINVAL); } restart: maxcount = bufsize / sizeof(struct statfs); if (bufsize == 0) { sfsp = NULL; tofree = NULL; } else if (bufseg == UIO_USERSPACE) { sfsp = *buf; tofree = NULL; } else /* if (bufseg == UIO_SYSSPACE) */ { count = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { count++; } mtx_unlock(&mountlist_mtx); if (maxcount > count) maxcount = count; tofree = sfsp = *buf = malloc(maxcount * sizeof(struct statfs), M_STATFS, M_WAITOK); } count = 0; /* * If there is no target buffer they only want the count. * * This could be TAILQ_FOREACH but it is open-coded to match the original * code below. */ if (sfsp == NULL) { mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif count++; nmp = TAILQ_NEXT(mp, mnt_list); } mtx_unlock(&mountlist_mtx); *countp = count; return (0); } /* * They want the entire thing. * * Short-circuit the corner case of no room for anything, avoids * relocking below. */ if (maxcount < 1) { goto out; } mtx_lock(&mountlist_mtx); for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { if (prison_canseemount(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #ifdef MAC if (mac_mount_check_stat(td->td_ucred, mp) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } #endif if (mode == MNT_WAIT) { if (vfs_busy(mp, MBF_MNTLSTLOCK) != 0) { /* * If vfs_busy() failed, and MBF_NOWAIT * wasn't passed, then the mp is gone. * Furthermore, because of MBF_MNTLSTLOCK, * the mountlist_mtx was dropped. We have * no other choice than to start over. */ mtx_unlock(&mountlist_mtx); free(tofree, M_STATFS); goto restart; } } else { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { nmp = TAILQ_NEXT(mp, mnt_list); continue; } } sp = &mp->mnt_stat; /* * If MNT_NOWAIT is specified, do not refresh * the fsstat cache. */ if (mode != MNT_NOWAIT) { error = VFS_STATFS(mp, sp); if (error != 0) { mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); continue; } } if (priv_check_cred_vfs_generation(td->td_ucred)) { sptmp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); *sptmp = *sp; sptmp->f_fsid.val[0] = sptmp->f_fsid.val[1] = 0; prison_enforce_statfs(td->td_ucred, mp, sptmp); sp = sptmp; } else sptmp = NULL; if (bufseg == UIO_SYSSPACE) { bcopy(sp, sfsp, sizeof(*sp)); free(sptmp, M_STATFS); } else /* if (bufseg == UIO_USERSPACE) */ { error = copyout(sp, sfsp, sizeof(*sp)); free(sptmp, M_STATFS); if (error != 0) { vfs_unbusy(mp); return (error); } } sfsp++; count++; if (count == maxcount) { vfs_unbusy(mp); goto out; } mtx_lock(&mountlist_mtx); nmp = TAILQ_NEXT(mp, mnt_list); vfs_unbusy(mp); } mtx_unlock(&mountlist_mtx); out: *countp = count; return (0); } #ifdef COMPAT_FREEBSD4 /* * Get old format filesystem statistics. */ static void freebsd4_cvtstatfs(struct statfs *, struct ostatfs *); #ifndef _SYS_SYSPROTO_H_ struct freebsd4_statfs_args { char *path; struct ostatfs *buf; }; #endif int freebsd4_statfs(struct thread *td, struct freebsd4_statfs_args *uap) { struct ostatfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get filesystem statistics. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fstatfs_args { int fd; struct ostatfs *buf; }; #endif int freebsd4_fstatfs(struct thread *td, struct freebsd4_fstatfs_args *uap) { struct ostatfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get statistics on all filesystems. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_getfsstat_args { struct ostatfs *buf; long bufsize; int mode; }; #endif int freebsd4_getfsstat(struct thread *td, struct freebsd4_getfsstat_args *uap) { struct statfs *buf, *sp; struct ostatfs osb; size_t count, size; int error; if (uap->bufsize < 0) return (EINVAL); count = uap->bufsize / sizeof(struct ostatfs); if (count > SIZE_MAX / sizeof(struct statfs)) return (EINVAL); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->mode); if (error == 0) td->td_retval[0] = count; if (size != 0) { sp = buf; while (count != 0 && error == 0) { freebsd4_cvtstatfs(sp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); sp++; uap->buf++; count--; } free(buf, M_STATFS); } return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd4_fhstatfs_args { struct fhandle *u_fhp; struct ostatfs *buf; }; #endif int freebsd4_fhstatfs(struct thread *td, struct freebsd4_fhstatfs_args *uap) { struct ostatfs osb; struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) { freebsd4_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void freebsd4_cvtstatfs(struct statfs *nsp, struct ostatfs *osp) { statfs_scale_blocks(nsp, LONG_MAX); bzero(osp, sizeof(*osp)); osp->f_bsize = nsp->f_bsize; osp->f_iosize = MIN(nsp->f_iosize, LONG_MAX); osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = MIN(nsp->f_files, LONG_MAX); osp->f_ffree = MIN(nsp->f_ffree, LONG_MAX); osp->f_owner = nsp->f_owner; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_syncwrites = MIN(nsp->f_syncwrites, LONG_MAX); osp->f_asyncwrites = MIN(nsp->f_asyncwrites, LONG_MAX); osp->f_syncreads = MIN(nsp->f_syncreads, LONG_MAX); osp->f_asyncreads = MIN(nsp->f_asyncreads, LONG_MAX); strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, OMFSNAMELEN)); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, OMNAMELEN)); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, OMNAMELEN)); osp->f_fsid = nsp->f_fsid; } #endif /* COMPAT_FREEBSD4 */ #if defined(COMPAT_FREEBSD11) /* * Get old format filesystem statistics. */ static void freebsd11_cvtstatfs(struct statfs *, struct freebsd11_statfs *); int freebsd11_statfs(struct thread *td, struct freebsd11_statfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_statfs(td, uap->path, UIO_USERSPACE, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get filesystem statistics. */ int freebsd11_fstatfs(struct thread *td, struct freebsd11_fstatfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; int error; sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fstatfs(td, uap->fd, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Get statistics on all filesystems. */ int freebsd11_getfsstat(struct thread *td, struct freebsd11_getfsstat_args *uap) { struct freebsd11_statfs osb; struct statfs *buf, *sp; size_t count, size; int error; count = uap->bufsize / sizeof(struct ostatfs); size = count * sizeof(struct statfs); error = kern_getfsstat(td, &buf, size, &count, UIO_SYSSPACE, uap->mode); if (error == 0) td->td_retval[0] = count; if (size > 0) { sp = buf; while (count > 0 && error == 0) { freebsd11_cvtstatfs(sp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); sp++; uap->buf++; count--; } free(buf, M_STATFS); } return (error); } /* * Implement fstatfs() for (NFS) file handles. */ int freebsd11_fhstatfs(struct thread *td, struct freebsd11_fhstatfs_args *uap) { struct freebsd11_statfs osb; struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) { freebsd11_cvtstatfs(sfp, &osb); error = copyout(&osb, uap->buf, sizeof(osb)); } free(sfp, M_STATFS); return (error); } /* * Convert a new format statfs structure to an old format statfs structure. */ static void freebsd11_cvtstatfs(struct statfs *nsp, struct freebsd11_statfs *osp) { bzero(osp, sizeof(*osp)); osp->f_version = FREEBSD11_STATFS_VERSION; osp->f_type = nsp->f_type; osp->f_flags = nsp->f_flags; osp->f_bsize = nsp->f_bsize; osp->f_iosize = nsp->f_iosize; osp->f_blocks = nsp->f_blocks; osp->f_bfree = nsp->f_bfree; osp->f_bavail = nsp->f_bavail; osp->f_files = nsp->f_files; osp->f_ffree = nsp->f_ffree; osp->f_syncwrites = nsp->f_syncwrites; osp->f_asyncwrites = nsp->f_asyncwrites; osp->f_syncreads = nsp->f_syncreads; osp->f_asyncreads = nsp->f_asyncreads; osp->f_namemax = nsp->f_namemax; osp->f_owner = nsp->f_owner; osp->f_fsid = nsp->f_fsid; strlcpy(osp->f_fstypename, nsp->f_fstypename, MIN(MFSNAMELEN, sizeof(osp->f_fstypename))); strlcpy(osp->f_mntonname, nsp->f_mntonname, MIN(MNAMELEN, sizeof(osp->f_mntonname))); strlcpy(osp->f_mntfromname, nsp->f_mntfromname, MIN(MNAMELEN, sizeof(osp->f_mntfromname))); } #endif /* COMPAT_FREEBSD11 */ /* * Change current working directory to a given file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchdir_args { int fd; }; #endif int sys_fchdir(struct thread *td, struct fchdir_args *uap) { struct vnode *vp, *tdp; struct mount *mp; struct file *fp; int error; AUDIT_ARG_FD(uap->fd); error = getvnode(td, uap->fd, &cap_fchdir_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; vrefact(vp); fdrop(fp, td); vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); error = change_dir(vp, td); while (!error && (mp = vp->v_mountedhere) != NULL) { if (vfs_busy(mp, 0)) continue; error = VFS_ROOT(mp, LK_SHARED, &tdp); vfs_unbusy(mp); if (error != 0) break; vput(vp); vp = tdp; } if (error != 0) { vput(vp); return (error); } VOP_UNLOCK(vp); pwd_chdir(td, vp); return (0); } /* * Change current working directory (``.''). */ #ifndef _SYS_SYSPROTO_H_ struct chdir_args { char *path; }; #endif int sys_chdir(struct thread *td, struct chdir_args *uap) { return (kern_chdir(td, uap->path, UIO_USERSPACE)); } int kern_chdir(struct thread *td, const char *path, enum uio_seg pathseg) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); if ((error = change_dir(nd.ni_vp, td)) != 0) { vput(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } VOP_UNLOCK(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); pwd_chdir(td, nd.ni_vp); return (0); } /* * Change notion of root (``/'') directory. */ #ifndef _SYS_SYSPROTO_H_ struct chroot_args { char *path; }; #endif int sys_chroot(struct thread *td, struct chroot_args *uap) { struct nameidata nd; int error; error = priv_check(td, PRIV_VFS_CHROOT); if (error != 0) return (error); NDINIT(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) goto error; error = change_dir(nd.ni_vp, td); if (error != 0) goto e_vunlock; #ifdef MAC error = mac_vnode_check_chroot(td->td_ucred, nd.ni_vp); if (error != 0) goto e_vunlock; #endif VOP_UNLOCK(nd.ni_vp); error = pwd_chroot(td, nd.ni_vp); vrele(nd.ni_vp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); e_vunlock: vput(nd.ni_vp); error: NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } /* * Common routine for chroot and chdir. Callers must provide a locked vnode * instance. */ int change_dir(struct vnode *vp, struct thread *td) { #ifdef MAC int error; #endif ASSERT_VOP_LOCKED(vp, "change_dir(): vp not locked"); if (vp->v_type != VDIR) return (ENOTDIR); #ifdef MAC error = mac_vnode_check_chdir(td->td_ucred, vp); if (error != 0) return (error); #endif return (VOP_ACCESS(vp, VEXEC, td->td_ucred, td)); } static __inline void flags_to_rights(int flags, cap_rights_t *rightsp) { if (flags & O_EXEC) { cap_rights_set_one(rightsp, CAP_FEXECVE); } else { switch ((flags & O_ACCMODE)) { case O_RDONLY: cap_rights_set_one(rightsp, CAP_READ); break; case O_RDWR: cap_rights_set_one(rightsp, CAP_READ); /* FALLTHROUGH */ case O_WRONLY: cap_rights_set_one(rightsp, CAP_WRITE); if (!(flags & (O_APPEND | O_TRUNC))) cap_rights_set_one(rightsp, CAP_SEEK); break; } } if (flags & O_CREAT) cap_rights_set_one(rightsp, CAP_CREATE); if (flags & O_TRUNC) cap_rights_set_one(rightsp, CAP_FTRUNCATE); if (flags & (O_SYNC | O_FSYNC)) cap_rights_set_one(rightsp, CAP_FSYNC); if (flags & (O_EXLOCK | O_SHLOCK)) cap_rights_set_one(rightsp, CAP_FLOCK); } /* * Check permissions, allocate an open file structure, and call the device * open routine if any. */ #ifndef _SYS_SYSPROTO_H_ struct open_args { char *path; int flags; int mode; }; #endif int sys_open(struct thread *td, struct open_args *uap) { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct openat_args { int fd; char *path; int flag; int mode; }; #endif int sys_openat(struct thread *td, struct openat_args *uap) { AUDIT_ARG_FD(uap->fd); return (kern_openat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->mode)); } int kern_openat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int flags, int mode) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; struct file *fp; struct vnode *vp; struct nameidata nd; cap_rights_t rights; int cmode, error, indx; indx = -1; AUDIT_ARG_FFLAGS(flags); AUDIT_ARG_MODE(mode); cap_rights_init_one(&rights, CAP_LOOKUP); flags_to_rights(flags, &rights); /* * Only one of the O_EXEC, O_RDONLY, O_WRONLY and O_RDWR flags * may be specified. */ if (flags & O_EXEC) { if (flags & O_ACCMODE) return (EINVAL); } else if ((flags & O_ACCMODE) == O_ACCMODE) { return (EINVAL); } else { flags = FFLAGS(flags); } /* * Allocate a file structure. The descriptor to reference it * is allocated and set by finstall() below. */ error = falloc_noinstall(td, &fp); if (error != 0) return (error); /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ /* Set the flags early so the finit in devfs can pick them up. */ fp->f_flag = flags & FMASK; cmode = ((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT; NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, &rights, td); td->td_dupfd = -1; /* XXX check for fdopen */ error = vn_open(&nd, &flags, cmode, fp); if (error != 0) { /* * If the vn_open replaced the method vector, something * wonderous happened deep below and we just pass it up * pretending we know what we do. */ if (error == ENXIO && fp->f_ops != &badfileops) goto success; /* * Handle special fdopen() case. bleh. * * Don't do this for relative (capability) lookups; we don't * understand exactly what would happen, and we don't think * that it ever should. */ if ((nd.ni_resflags & NIRES_STRICTREL) == 0 && (error == ENODEV || error == ENXIO) && td->td_dupfd >= 0) { error = dupfdopen(td, fdp, td->td_dupfd, flags, error, &indx); if (error == 0) goto success; } goto bad; } td->td_dupfd = 0; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; /* * Store the vnode, for any f_type. Typically, the vnode use * count is decremented by direct call to vn_closefile() for * files that switched type in the cdevsw fdopen() method. */ fp->f_vnode = vp; /* * If the file wasn't claimed by devfs bind it to the normal * vnode operations here. */ if (fp->f_ops == &badfileops) { KASSERT(vp->v_type != VFIFO, ("Unexpected fifo.")); finit_vnode(fp, flags, NULL, &vnops); } VOP_UNLOCK(vp); if (flags & O_TRUNC) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } success: /* * If we haven't already installed the FD (for dupfdopen), do so now. */ if (indx == -1) { struct filecaps *fcaps; #ifdef CAPABILITIES if ((nd.ni_resflags & NIRES_STRICTREL) != 0) fcaps = &nd.ni_filecaps; else #endif fcaps = NULL; error = finstall(td, fp, &indx, flags, fcaps); /* On success finstall() consumes fcaps. */ if (error != 0) { filecaps_free(&nd.ni_filecaps); goto bad; } } else { filecaps_free(&nd.ni_filecaps); } /* * Release our private reference, leaving the one associated with * the descriptor table intact. */ fdrop(fp, td); td->td_retval[0] = indx; return (0); bad: KASSERT(indx == -1, ("indx=%d, should be -1", indx)); fdrop(fp, td); return (error); } #ifdef COMPAT_43 /* * Create a file. */ #ifndef _SYS_SYSPROTO_H_ struct ocreat_args { char *path; int mode; }; #endif int ocreat(struct thread *td, struct ocreat_args *uap) { return (kern_openat(td, AT_FDCWD, uap->path, UIO_USERSPACE, O_WRONLY | O_CREAT | O_TRUNC, uap->mode)); } #endif /* COMPAT_43 */ /* * Create a special file. */ #ifndef _SYS_SYSPROTO_H_ struct mknodat_args { int fd; char *path; mode_t mode; dev_t dev; }; #endif int sys_mknodat(struct thread *td, struct mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #if defined(COMPAT_FREEBSD11) int freebsd11_mknod(struct thread *td, struct freebsd11_mknod_args *uap) { return (kern_mknodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } int freebsd11_mknodat(struct thread *td, struct freebsd11_mknodat_args *uap) { return (kern_mknodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->dev)); } #endif /* COMPAT_FREEBSD11 */ int kern_mknodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int mode, dev_t dev) { struct vnode *vp; struct mount *mp; struct vattr vattr; struct nameidata nd; int error, whiteout = 0; AUDIT_ARG_MODE(mode); AUDIT_ARG_DEV(dev); switch (mode & S_IFMT) { case S_IFCHR: case S_IFBLK: error = priv_check(td, PRIV_VFS_MKNOD_DEV); if (error == 0 && dev == VNOVAL) error = EINVAL; break; case S_IFWHT: error = priv_check(td, PRIV_VFS_MKNOD_WHT); break; case S_IFIFO: if (dev == 0) return (kern_mkfifoat(td, fd, path, pathseg, mode)); /* FALLTHROUGH */ default: error = EINVAL; break; } if (error != 0) return (error); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, &cap_mknodat_rights, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } else { VATTR_NULL(&vattr); vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; vattr.va_rdev = dev; whiteout = 0; switch (mode & S_IFMT) { case S_IFCHR: vattr.va_type = VCHR; break; case S_IFBLK: vattr.va_type = VBLK; break; case S_IFWHT: whiteout = 1; break; default: panic("kern_mknod: invalid mode"); } } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } #ifdef MAC if (error == 0 && !whiteout) error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); #endif if (error == 0) { if (whiteout) error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, CREATE); else { error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); } } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); + if (error == ERELOOKUP) + goto restart; return (error); } /* * Create a named pipe. */ #ifndef _SYS_SYSPROTO_H_ struct mkfifo_args { char *path; int mode; }; #endif int sys_mkfifo(struct thread *td, struct mkfifo_args *uap) { return (kern_mkfifoat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkfifoat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkfifoat(struct thread *td, struct mkfifoat_args *uap) { return (kern_mkfifoat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkfifoat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int mode) { struct mount *mp; struct vattr vattr; struct nameidata nd; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, pathseg, path, fd, &cap_mkfifoat_rights, td); if ((error = namei(&nd)) != 0) return (error); if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VFIFO; vattr.va_mode = (mode & ALLPERMS) & ~td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); if (error == 0) vput(nd.ni_vp); #ifdef MAC out: #endif vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); + if (error == ERELOOKUP) + goto restart; return (error); } /* * Make a hard file link. */ #ifndef _SYS_SYSPROTO_H_ struct link_args { char *path; char *link; }; #endif int sys_link(struct thread *td, struct link_args *uap) { return (kern_linkat(td, AT_FDCWD, AT_FDCWD, uap->path, uap->link, UIO_USERSPACE, FOLLOW)); } #ifndef _SYS_SYSPROTO_H_ struct linkat_args { int fd1; char *path1; int fd2; char *path2; int flag; }; #endif int sys_linkat(struct thread *td, struct linkat_args *uap) { int flag; flag = uap->flag; if ((flag & ~(AT_SYMLINK_FOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); return (kern_linkat(td, uap->fd1, uap->fd2, uap->path1, uap->path2, UIO_USERSPACE, at2cnpflags(flag, AT_SYMLINK_FOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH))); } int hardlink_check_uid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_uid, CTLFLAG_RW, &hardlink_check_uid, 0, "Unprivileged processes cannot create hard links to files owned by other " "users"); static int hardlink_check_gid = 0; SYSCTL_INT(_security_bsd, OID_AUTO, hardlink_check_gid, CTLFLAG_RW, &hardlink_check_gid, 0, "Unprivileged processes cannot create hard links to files owned by other " "groups"); static int can_hardlink(struct vnode *vp, struct ucred *cred) { struct vattr va; int error; if (!hardlink_check_uid && !hardlink_check_gid) return (0); error = VOP_GETATTR(vp, &va, cred); if (error != 0) return (error); if (hardlink_check_uid && cred->cr_uid != va.va_uid) { error = priv_check_cred(cred, PRIV_VFS_LINK); if (error != 0) return (error); } if (hardlink_check_gid && !groupmember(va.va_gid, cred)) { error = priv_check_cred(cred, PRIV_VFS_LINK); if (error != 0) return (error); } return (0); } int kern_linkat(struct thread *td, int fd1, int fd2, const char *path1, const char *path2, enum uio_seg segflag, int follow) { struct nameidata nd; int error; do { bwillwrite(); NDINIT_ATRIGHTS(&nd, LOOKUP, follow | AUDITVNODE1, segflag, path1, fd1, &cap_linkat_source_rights, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = kern_linkat_vp(td, nd.ni_vp, fd2, path2, segflag); - } while (error == EAGAIN); + } while (error == EAGAIN || error == ERELOOKUP); return (error); } static int kern_linkat_vp(struct thread *td, struct vnode *vp, int fd, const char *path, enum uio_seg segflag) { struct nameidata nd; struct mount *mp; int error; if (vp->v_type == VDIR) { vrele(vp); return (EPERM); /* POSIX */ } NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE2 | NOCACHE, segflag, path, fd, &cap_linkat_target_rights, td); if ((error = namei(&nd)) == 0) { if (nd.ni_vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_dvp == nd.ni_vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); vrele(vp); return (EEXIST); } else if (nd.ni_dvp->v_mount != vp->v_mount) { /* * Cross-device link. No need to recheck * vp->v_type, since it cannot change, except * to VBAD. */ NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vrele(vp); return (EXDEV); } else if ((error = vn_lock(vp, LK_EXCLUSIVE)) == 0) { error = can_hardlink(vp, td->td_ucred); #ifdef MAC if (error == 0) error = mac_vnode_check_link(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); #endif if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); return (error); } error = vn_start_write(vp, &mp, V_NOWAIT); if (error != 0) { vput(vp); vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); return (EAGAIN); } error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd); VOP_UNLOCK(vp); vput(nd.ni_dvp); vn_finished_write(mp); NDFREE(&nd, NDF_ONLY_PNBUF); } else { vput(nd.ni_dvp); NDFREE(&nd, NDF_ONLY_PNBUF); vrele(vp); return (EAGAIN); } } vrele(vp); return (error); } /* * Make a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct symlink_args { char *path; char *link; }; #endif int sys_symlink(struct thread *td, struct symlink_args *uap) { return (kern_symlinkat(td, uap->path, AT_FDCWD, uap->link, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct symlinkat_args { char *path; int fd; char *path2; }; #endif int sys_symlinkat(struct thread *td, struct symlinkat_args *uap) { return (kern_symlinkat(td, uap->path1, uap->fd, uap->path2, UIO_USERSPACE)); } int kern_symlinkat(struct thread *td, const char *path1, int fd, const char *path2, enum uio_seg segflg) { struct mount *mp; struct vattr vattr; const char *syspath; char *tmppath; struct nameidata nd; int error; if (segflg == UIO_SYSSPACE) { syspath = path1; } else { tmppath = uma_zalloc(namei_zone, M_WAITOK); if ((error = copyinstr(path1, tmppath, MAXPATHLEN, NULL)) != 0) goto out; syspath = tmppath; } AUDIT_ARG_TEXT(syspath); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NOCACHE, segflg, path2, fd, &cap_symlinkat_rights, td); if ((error = namei(&nd)) != 0) goto out; if (nd.ni_vp) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(nd.ni_vp); error = EEXIST; goto out; } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) goto out; goto restart; } VATTR_NULL(&vattr); vattr.va_mode = ACCESSPERMS &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC vattr.va_type = VLNK; error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out2; #endif error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr, syspath); if (error == 0) vput(nd.ni_vp); #ifdef MAC out2: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); + if (error == ERELOOKUP) + goto restart; out: if (segflg != UIO_SYSSPACE) uma_zfree(namei_zone, tmppath); return (error); } /* * Delete a whiteout from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct undelete_args { char *path; }; #endif int sys_undelete(struct thread *td, struct undelete_args *uap) { struct mount *mp; struct nameidata nd; int error; restart: bwillwrite(); NDINIT(&nd, DELETE, LOCKPARENT | DOWHITEOUT | AUDITVNODE1, UIO_USERSPACE, uap->path, td); error = namei(&nd); if (error != 0) return (error); if (nd.ni_vp != NULLVP || !(nd.ni_cnd.cn_flags & ISWHITEOUT)) { NDFREE(&nd, NDF_ONLY_PNBUF); if (nd.ni_vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if (nd.ni_vp) vrele(nd.ni_vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } error = VOP_WHITEOUT(nd.ni_dvp, &nd.ni_cnd, DELETE); NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); vn_finished_write(mp); + if (error == ERELOOKUP) + goto restart; return (error); } /* * Delete a name from the filesystem. */ #ifndef _SYS_SYSPROTO_H_ struct unlink_args { char *path; }; #endif int sys_unlink(struct thread *td, struct unlink_args *uap) { return (kern_funlinkat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE, 0, 0)); } static int kern_funlinkat_ex(struct thread *td, int dfd, const char *path, int fd, int flag, enum uio_seg pathseg, ino_t oldinum) { if ((flag & ~AT_REMOVEDIR) != 0) return (EINVAL); if ((flag & AT_REMOVEDIR) != 0) return (kern_frmdirat(td, dfd, path, fd, UIO_USERSPACE, 0)); return (kern_funlinkat(td, dfd, path, fd, UIO_USERSPACE, 0, 0)); } #ifndef _SYS_SYSPROTO_H_ struct unlinkat_args { int fd; char *path; int flag; }; #endif int sys_unlinkat(struct thread *td, struct unlinkat_args *uap) { return (kern_funlinkat_ex(td, uap->fd, uap->path, FD_NONE, uap->flag, UIO_USERSPACE, 0)); } #ifndef _SYS_SYSPROTO_H_ struct funlinkat_args { int dfd; const char *path; int fd; int flag; }; #endif int sys_funlinkat(struct thread *td, struct funlinkat_args *uap) { return (kern_funlinkat_ex(td, uap->dfd, uap->path, uap->fd, uap->flag, UIO_USERSPACE, 0)); } int kern_funlinkat(struct thread *td, int dfd, const char *path, int fd, enum uio_seg pathseg, int flag, ino_t oldinum) { struct mount *mp; struct file *fp; struct vnode *vp; struct nameidata nd; struct stat sb; int error; fp = NULL; if (fd != FD_NONE) { error = getvnode(td, fd, &cap_no_rights, &fp); if (error != 0) return (error); } restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_BENEATH | AT_RESOLVE_BENEATH), pathseg, path, dfd, &cap_unlinkat_rights, td); if ((error = namei(&nd)) != 0) { if (error == EINVAL) error = EPERM; goto fdout; } vp = nd.ni_vp; if (vp->v_type == VDIR && oldinum == 0) { error = EPERM; /* POSIX */ } else if (oldinum != 0 && ((error = VOP_STAT(vp, &sb, td->td_ucred, NOCRED, td)) == 0) && sb.st_ino != oldinum) { error = EIDRM; /* Identifier removed */ } else if (fp != NULL && fp->f_vnode != vp) { if (VN_IS_DOOMED(fp->f_vnode)) error = EBADF; else error = EDEADLK; } else { /* * The root of a mounted filesystem cannot be deleted. * * XXX: can this only be a VDIR case? */ if (vp->v_vflag & VV_ROOT) error = EBUSY; } if (error == 0) { if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) { goto fdout; } goto restart; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_REMOVE(nd.ni_dvp, vp, &nd.ni_cnd); #ifdef MAC out: #endif vn_finished_write(mp); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (vp == nd.ni_dvp) vrele(vp); else vput(vp); + if (error == ERELOOKUP) + goto restart; fdout: if (fp != NULL) fdrop(fp, td); return (error); } /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct lseek_args { int fd; int pad; off_t offset; int whence; }; #endif int sys_lseek(struct thread *td, struct lseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } int kern_lseek(struct thread *td, int fd, off_t offset, int whence) { struct file *fp; int error; AUDIT_ARG_FD(fd); error = fget(td, fd, &cap_seek_rights, &fp); if (error != 0) return (error); error = (fp->f_ops->fo_flags & DFLAG_SEEKABLE) != 0 ? fo_seek(fp, offset, whence, td) : ESPIPE; fdrop(fp, td); return (error); } #if defined(COMPAT_43) /* * Reposition read/write file offset. */ #ifndef _SYS_SYSPROTO_H_ struct olseek_args { int fd; long offset; int whence; }; #endif int olseek(struct thread *td, struct olseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Version with the 'pad' argument */ int freebsd6_lseek(struct thread *td, struct freebsd6_lseek_args *uap) { return (kern_lseek(td, uap->fd, uap->offset, uap->whence)); } #endif /* * Check access permissions using passed credentials. */ static int vn_access(struct vnode *vp, int user_flags, struct ucred *cred, struct thread *td) { accmode_t accmode; int error; /* Flags == 0 means only check for existence. */ if (user_flags == 0) return (0); accmode = 0; if (user_flags & R_OK) accmode |= VREAD; if (user_flags & W_OK) accmode |= VWRITE; if (user_flags & X_OK) accmode |= VEXEC; #ifdef MAC error = mac_vnode_check_access(cred, vp, accmode); if (error != 0) return (error); #endif if ((accmode & VWRITE) == 0 || (error = vn_writechk(vp)) == 0) error = VOP_ACCESS(vp, accmode, cred, td); return (error); } /* * Check access permissions using "real" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct access_args { char *path; int amode; }; #endif int sys_access(struct thread *td, struct access_args *uap) { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, 0, uap->amode)); } #ifndef _SYS_SYSPROTO_H_ struct faccessat_args { int dirfd; char *path; int amode; int flag; } #endif int sys_faccessat(struct thread *td, struct faccessat_args *uap) { return (kern_accessat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flag, uap->amode)); } int kern_accessat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int flag, int amode) { struct ucred *cred, *usecred; struct vnode *vp; struct nameidata nd; int error; if ((flag & ~(AT_EACCESS | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); if (amode != F_OK && (amode & ~(R_OK | W_OK | X_OK)) != 0) return (EINVAL); /* * Create and modify a temporary credential instead of one that * is potentially shared (if we need one). */ cred = td->td_ucred; if ((flag & AT_EACCESS) == 0 && ((cred->cr_uid != cred->cr_ruid || cred->cr_rgid != cred->cr_groups[0]))) { usecred = crdup(cred); usecred->cr_uid = cred->cr_ruid; usecred->cr_groups[0] = cred->cr_rgid; td->td_ucred = usecred; } else usecred = cred; AUDIT_ARG_VALUE(amode); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_BENEATH | AT_RESOLVE_BENEATH), pathseg, path, fd, &cap_fstat_rights, td); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; error = vn_access(vp, amode, usecred, td); NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); out: if (usecred != cred) { td->td_ucred = cred; crfree(usecred); } return (error); } /* * Check access permissions using "effective" credentials. */ #ifndef _SYS_SYSPROTO_H_ struct eaccess_args { char *path; int amode; }; #endif int sys_eaccess(struct thread *td, struct eaccess_args *uap) { return (kern_accessat(td, AT_FDCWD, uap->path, UIO_USERSPACE, AT_EACCESS, uap->amode)); } #if defined(COMPAT_43) /* * Get file status; this version follows links. */ #ifndef _SYS_SYSPROTO_H_ struct ostat_args { char *path; struct ostat *ub; }; #endif int ostat(struct thread *td, struct ostat_args *uap) { struct stat sb; struct ostat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct olstat_args { char *path; struct ostat *ub; }; #endif int olstat(struct thread *td, struct olstat_args *uap) { struct stat sb; struct ostat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); cvtstat(&sb, &osb); return (copyout(&osb, uap->ub, sizeof (osb))); } /* * Convert from an old to a new stat structure. * XXX: many values are blindly truncated. */ void cvtstat(struct stat *st, struct ostat *ost) { bzero(ost, sizeof(*ost)); ost->st_dev = st->st_dev; ost->st_ino = st->st_ino; ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; ost->st_size = MIN(st->st_size, INT32_MAX); ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_blksize = st->st_blksize; ost->st_blocks = st->st_blocks; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; } #endif /* COMPAT_43 */ #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11) int ino64_trunc_error; SYSCTL_INT(_vfs, OID_AUTO, ino64_trunc_error, CTLFLAG_RW, &ino64_trunc_error, 0, "Error on truncation of device, file or inode number, or link count"); int freebsd11_cvtstat(struct stat *st, struct freebsd11_stat *ost) { ost->st_dev = st->st_dev; if (ost->st_dev != st->st_dev) { switch (ino64_trunc_error) { default: /* * Since dev_t is almost raw, don't clamp to the * maximum for case 2, but ignore the error. */ break; case 1: return (EOVERFLOW); } } ost->st_ino = st->st_ino; if (ost->st_ino != st->st_ino) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: ost->st_ino = UINT32_MAX; break; } } ost->st_mode = st->st_mode; ost->st_nlink = st->st_nlink; if (ost->st_nlink != st->st_nlink) { switch (ino64_trunc_error) { default: case 0: break; case 1: return (EOVERFLOW); case 2: ost->st_nlink = UINT16_MAX; break; } } ost->st_uid = st->st_uid; ost->st_gid = st->st_gid; ost->st_rdev = st->st_rdev; if (ost->st_rdev != st->st_rdev) { switch (ino64_trunc_error) { default: break; case 1: return (EOVERFLOW); } } ost->st_atim = st->st_atim; ost->st_mtim = st->st_mtim; ost->st_ctim = st->st_ctim; ost->st_size = st->st_size; ost->st_blocks = st->st_blocks; ost->st_blksize = st->st_blksize; ost->st_flags = st->st_flags; ost->st_gen = st->st_gen; ost->st_lspare = 0; ost->st_birthtim = st->st_birthtim; bzero((char *)&ost->st_birthtim + sizeof(ost->st_birthtim), sizeof(*ost) - offsetof(struct freebsd11_stat, st_birthtim) - sizeof(ost->st_birthtim)); return (0); } int freebsd11_stat(struct thread *td, struct freebsd11_stat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->ub, sizeof(osb)); return (error); } int freebsd11_lstat(struct thread *td, struct freebsd11_lstat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->ub, sizeof(osb)); return (error); } int freebsd11_fhstat(struct thread *td, struct freebsd11_fhstat_args* uap) { struct fhandle fh; struct stat sb; struct freebsd11_stat osb; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->sb, sizeof(osb)); return (error); } int freebsd11_fstatat(struct thread *td, struct freebsd11_fstatat_args* uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->buf, sizeof(osb)); return (error); } #endif /* COMPAT_FREEBSD11 */ /* * Get file status */ #ifndef _SYS_SYSPROTO_H_ struct fstatat_args { int fd; char *path; struct stat *buf; int flag; } #endif int sys_fstatat(struct thread *td, struct fstatat_args *uap) { struct stat sb; int error; error = kern_statat(td, uap->flag, uap->fd, uap->path, UIO_USERSPACE, &sb, NULL); if (error == 0) error = copyout(&sb, uap->buf, sizeof (sb)); return (error); } int kern_statat(struct thread *td, int flag, int fd, const char *path, enum uio_seg pathseg, struct stat *sbp, void (*hook)(struct vnode *vp, struct stat *sbp)) { struct nameidata nd; int error; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_BENEATH | AT_RESOLVE_BENEATH | AT_SYMLINK_NOFOLLOW) | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights, td); if ((error = namei(&nd)) != 0) return (error); error = VOP_STAT(nd.ni_vp, sbp, td->td_ucred, NOCRED, td); if (error == 0) { if (__predict_false(hook != NULL)) hook(nd.ni_vp, sbp); } NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_vp); #ifdef __STAT_TIME_T_EXT sbp->st_atim_ext = 0; sbp->st_mtim_ext = 0; sbp->st_ctim_ext = 0; sbp->st_btim_ext = 0; #endif #ifdef KTRACE if (KTRPOINT(td, KTR_STRUCT)) ktrstat_error(sbp, error); #endif return (error); } #if defined(COMPAT_FREEBSD11) /* * Implementation of the NetBSD [l]stat() functions. */ void freebsd11_cvtnstat(struct stat *sb, struct nstat *nsb) { bzero(nsb, sizeof(*nsb)); nsb->st_dev = sb->st_dev; nsb->st_ino = sb->st_ino; nsb->st_mode = sb->st_mode; nsb->st_nlink = sb->st_nlink; nsb->st_uid = sb->st_uid; nsb->st_gid = sb->st_gid; nsb->st_rdev = sb->st_rdev; nsb->st_atim = sb->st_atim; nsb->st_mtim = sb->st_mtim; nsb->st_ctim = sb->st_ctim; nsb->st_size = sb->st_size; nsb->st_blocks = sb->st_blocks; nsb->st_blksize = sb->st_blksize; nsb->st_flags = sb->st_flags; nsb->st_gen = sb->st_gen; nsb->st_birthtim = sb->st_birthtim; } #ifndef _SYS_SYSPROTO_H_ struct freebsd11_nstat_args { char *path; struct nstat *ub; }; #endif int freebsd11_nstat(struct thread *td, struct freebsd11_nstat_args *uap) { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, 0, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); freebsd11_cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } /* * NetBSD lstat. Get file status; this version does not follow links. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd11_nlstat_args { char *path; struct nstat *ub; }; #endif int freebsd11_nlstat(struct thread *td, struct freebsd11_nlstat_args *uap) { struct stat sb; struct nstat nsb; int error; error = kern_statat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->path, UIO_USERSPACE, &sb, NULL); if (error != 0) return (error); freebsd11_cvtnstat(&sb, &nsb); return (copyout(&nsb, uap->ub, sizeof (nsb))); } #endif /* COMPAT_FREEBSD11 */ /* * Get configurable pathname variables. */ #ifndef _SYS_SYSPROTO_H_ struct pathconf_args { char *path; int name; }; #endif int sys_pathconf(struct thread *td, struct pathconf_args *uap) { long value; int error; error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, FOLLOW, &value); if (error == 0) td->td_retval[0] = value; return (error); } #ifndef _SYS_SYSPROTO_H_ struct lpathconf_args { char *path; int name; }; #endif int sys_lpathconf(struct thread *td, struct lpathconf_args *uap) { long value; int error; error = kern_pathconf(td, uap->path, UIO_USERSPACE, uap->name, NOFOLLOW, &value); if (error == 0) td->td_retval[0] = value; return (error); } int kern_pathconf(struct thread *td, const char *path, enum uio_seg pathseg, int name, u_long flags, long *valuep) { struct nameidata nd; int error; NDINIT(&nd, LOOKUP, LOCKSHARED | LOCKLEAF | AUDITVNODE1 | flags, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = VOP_PATHCONF(nd.ni_vp, name, valuep); vput(nd.ni_vp); return (error); } /* * Return target name of a symbolic link. */ #ifndef _SYS_SYSPROTO_H_ struct readlink_args { char *path; char *buf; size_t count; }; #endif int sys_readlink(struct thread *td, struct readlink_args *uap) { return (kern_readlinkat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->count)); } #ifndef _SYS_SYSPROTO_H_ struct readlinkat_args { int fd; char *path; char *buf; size_t bufsize; }; #endif int sys_readlinkat(struct thread *td, struct readlinkat_args *uap) { return (kern_readlinkat(td, uap->fd, uap->path, UIO_USERSPACE, uap->buf, UIO_USERSPACE, uap->bufsize)); } int kern_readlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, char *buf, enum uio_seg bufseg, size_t count) { struct vnode *vp; struct nameidata nd; int error; if (count > IOSIZE_MAX) return (EINVAL); NDINIT_AT(&nd, LOOKUP, NOFOLLOW | LOCKSHARED | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; error = kern_readlink_vp(vp, buf, bufseg, count, td); vput(vp); return (error); } /* * Helper function to readlink from a vnode */ static int kern_readlink_vp(struct vnode *vp, char *buf, enum uio_seg bufseg, size_t count, struct thread *td) { struct iovec aiov; struct uio auio; int error; ASSERT_VOP_LOCKED(vp, "kern_readlink_vp(): vp not locked"); #ifdef MAC error = mac_vnode_check_readlink(td->td_ucred, vp); if (error != 0) return (error); #endif if (vp->v_type != VLNK && (vp->v_vflag & VV_READLINK) == 0) return (EINVAL); aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = 0; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; auio.uio_resid = count; error = VOP_READLINK(vp, &auio, td->td_ucred); td->td_retval[0] = count - auio.uio_resid; return (error); } /* * Common implementation code for chflags() and fchflags(). */ static int setfflags(struct thread *td, struct vnode *vp, u_long flags) { struct mount *mp; struct vattr vattr; int error; /* We can't support the value matching VNOVAL. */ if (flags == VNOVAL) return (EOPNOTSUPP); /* * Prevent non-root users from setting flags on devices. When * a device is reused, users can retain ownership of the device * if they are allowed to set flags and programs assume that * chown can't fail when done as root. */ if (vp->v_type == VCHR || vp->v_type == VBLK) { error = priv_check(td, PRIV_VFS_CHFLAGS_DEV); if (error != 0) return (error); } if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); VATTR_NULL(&vattr); vattr.va_flags = flags; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_vnode_check_setflags(td->td_ucred, vp, vattr.va_flags); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Change flags of a file given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chflags_args { const char *path; u_long flags; }; #endif int sys_chflags(struct thread *td, struct chflags_args *uap) { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, 0)); } #ifndef _SYS_SYSPROTO_H_ struct chflagsat_args { int fd; const char *path; u_long flags; int atflag; } #endif int sys_chflagsat(struct thread *td, struct chflagsat_args *uap) { if ((uap->atflag & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); return (kern_chflagsat(td, uap->fd, uap->path, UIO_USERSPACE, uap->flags, uap->atflag)); } /* * Same as chflags() but doesn't follow symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchflags_args { const char *path; u_long flags; }; #endif int sys_lchflags(struct thread *td, struct lchflags_args *uap) { return (kern_chflagsat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->flags, AT_SYMLINK_NOFOLLOW)); } static int kern_chflagsat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, u_long flags, int atflag) { struct nameidata nd; int error; AUDIT_ARG_FFLAGS(flags); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(atflag, AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchflags_rights, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfflags(td, nd.ni_vp, flags); vrele(nd.ni_vp); return (error); } /* * Change flags of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchflags_args { int fd; u_long flags; }; #endif int sys_fchflags(struct thread *td, struct fchflags_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_FFLAGS(uap->flags); error = getvnode(td, uap->fd, &cap_fchflags_rights, &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); #endif error = setfflags(td, fp->f_vnode, uap->flags); fdrop(fp, td); return (error); } /* * Common implementation code for chmod(), lchmod() and fchmod(). */ int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode) { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_mode = mode & ALLPERMS; #ifdef MAC error = mac_vnode_check_setmode(cred, vp, vattr.va_mode); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Change mode of a file given path name. */ #ifndef _SYS_SYSPROTO_H_ struct chmod_args { char *path; int mode; }; #endif int sys_chmod(struct thread *td, struct chmod_args *uap) { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchmodat_args { int dirfd; char *path; mode_t mode; int flag; } #endif int sys_fchmodat(struct thread *td, struct fchmodat_args *uap) { if ((uap->flag & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); return (kern_fchmodat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode, uap->flag)); } /* * Change mode of a file given path name (don't follow links.) */ #ifndef _SYS_SYSPROTO_H_ struct lchmod_args { char *path; int mode; }; #endif int sys_lchmod(struct thread *td, struct lchmod_args *uap) { return (kern_fchmodat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode, AT_SYMLINK_NOFOLLOW)); } int kern_fchmodat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, mode_t mode, int flag) { struct nameidata nd; int error; AUDIT_ARG_MODE(mode); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchmod_rights, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfmode(td, td->td_ucred, nd.ni_vp, mode); vrele(nd.ni_vp); return (error); } /* * Change mode of a file given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchmod_args { int fd; int mode; }; #endif int sys_fchmod(struct thread *td, struct fchmod_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_MODE(uap->mode); error = fget(td, uap->fd, &cap_fchmod_rights, &fp); if (error != 0) return (error); error = fo_chmod(fp, uap->mode, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation for chown(), lchown(), and fchown() */ int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid) { struct mount *mp; struct vattr vattr; int error; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); VATTR_NULL(&vattr); vattr.va_uid = uid; vattr.va_gid = gid; #ifdef MAC error = mac_vnode_check_setowner(cred, vp, vattr.va_uid, vattr.va_gid); if (error == 0) #endif error = VOP_SETATTR(vp, &vattr, cred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Set ownership given a path name. */ #ifndef _SYS_SYSPROTO_H_ struct chown_args { char *path; int uid; int gid; }; #endif int sys_chown(struct thread *td, struct chown_args *uap) { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, 0)); } #ifndef _SYS_SYSPROTO_H_ struct fchownat_args { int fd; const char * path; uid_t uid; gid_t gid; int flag; }; #endif int sys_fchownat(struct thread *td, struct fchownat_args *uap) { if ((uap->flag & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); return (kern_fchownat(td, uap->fd, uap->path, UIO_USERSPACE, uap->uid, uap->gid, uap->flag)); } int kern_fchownat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, int uid, int gid, int flag) { struct nameidata nd; int error; AUDIT_ARG_OWNER(uid, gid); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH) | AUDITVNODE1, pathseg, path, fd, &cap_fchown_rights, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setfown(td, td->td_ucred, nd.ni_vp, uid, gid); vrele(nd.ni_vp); return (error); } /* * Set ownership given a path name, do not cross symlinks. */ #ifndef _SYS_SYSPROTO_H_ struct lchown_args { char *path; int uid; int gid; }; #endif int sys_lchown(struct thread *td, struct lchown_args *uap) { return (kern_fchownat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->uid, uap->gid, AT_SYMLINK_NOFOLLOW)); } /* * Set ownership given a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fchown_args { int fd; int uid; int gid; }; #endif int sys_fchown(struct thread *td, struct fchown_args *uap) { struct file *fp; int error; AUDIT_ARG_FD(uap->fd); AUDIT_ARG_OWNER(uap->uid, uap->gid); error = fget(td, uap->fd, &cap_fchown_rights, &fp); if (error != 0) return (error); error = fo_chown(fp, uap->uid, uap->gid, td->td_ucred, td); fdrop(fp, td); return (error); } /* * Common implementation code for utimes(), lutimes(), and futimes(). */ static int getutimes(const struct timeval *usrtvp, enum uio_seg tvpseg, struct timespec *tsp) { struct timeval tv[2]; const struct timeval *tvp; int error; if (usrtvp == NULL) { vfs_timestamp(&tsp[0]); tsp[1] = tsp[0]; } else { if (tvpseg == UIO_SYSSPACE) { tvp = usrtvp; } else { if ((error = copyin(usrtvp, tv, sizeof(tv))) != 0) return (error); tvp = tv; } if (tvp[0].tv_usec < 0 || tvp[0].tv_usec >= 1000000 || tvp[1].tv_usec < 0 || tvp[1].tv_usec >= 1000000) return (EINVAL); TIMEVAL_TO_TIMESPEC(&tvp[0], &tsp[0]); TIMEVAL_TO_TIMESPEC(&tvp[1], &tsp[1]); } return (0); } /* * Common implementation code for futimens(), utimensat(). */ #define UTIMENS_NULL 0x1 #define UTIMENS_EXIT 0x2 static int getutimens(const struct timespec *usrtsp, enum uio_seg tspseg, struct timespec *tsp, int *retflags) { struct timespec tsnow; int error; vfs_timestamp(&tsnow); *retflags = 0; if (usrtsp == NULL) { tsp[0] = tsnow; tsp[1] = tsnow; *retflags |= UTIMENS_NULL; return (0); } if (tspseg == UIO_SYSSPACE) { tsp[0] = usrtsp[0]; tsp[1] = usrtsp[1]; } else if ((error = copyin(usrtsp, tsp, sizeof(*tsp) * 2)) != 0) return (error); if (tsp[0].tv_nsec == UTIME_OMIT && tsp[1].tv_nsec == UTIME_OMIT) *retflags |= UTIMENS_EXIT; if (tsp[0].tv_nsec == UTIME_NOW && tsp[1].tv_nsec == UTIME_NOW) *retflags |= UTIMENS_NULL; if (tsp[0].tv_nsec == UTIME_OMIT) tsp[0].tv_sec = VNOVAL; else if (tsp[0].tv_nsec == UTIME_NOW) tsp[0] = tsnow; else if (tsp[0].tv_nsec < 0 || tsp[0].tv_nsec >= 1000000000L) return (EINVAL); if (tsp[1].tv_nsec == UTIME_OMIT) tsp[1].tv_sec = VNOVAL; else if (tsp[1].tv_nsec == UTIME_NOW) tsp[1] = tsnow; else if (tsp[1].tv_nsec < 0 || tsp[1].tv_nsec >= 1000000000L) return (EINVAL); return (0); } /* * Common implementation code for utimes(), lutimes(), futimes(), futimens(), * and utimensat(). */ static int setutimes(struct thread *td, struct vnode *vp, const struct timespec *ts, int numtimes, int nullflag) { struct mount *mp; struct vattr vattr; int error, setbirthtime; if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); setbirthtime = 0; if (numtimes < 3 && !VOP_GETATTR(vp, &vattr, td->td_ucred) && timespeccmp(&ts[1], &vattr.va_birthtime, < )) setbirthtime = 1; VATTR_NULL(&vattr); vattr.va_atime = ts[0]; vattr.va_mtime = ts[1]; if (setbirthtime) vattr.va_birthtime = ts[1]; if (numtimes > 2) vattr.va_birthtime = ts[2]; if (nullflag) vattr.va_vaflags |= VA_UTIMES_NULL; #ifdef MAC error = mac_vnode_check_setutimes(td->td_ucred, vp, vattr.va_atime, vattr.va_mtime); #endif if (error == 0) error = VOP_SETATTR(vp, &vattr, td->td_ucred); VOP_UNLOCK(vp); vn_finished_write(mp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct utimes_args { char *path; struct timeval *tptr; }; #endif int sys_utimes(struct thread *td, struct utimes_args *uap) { return (kern_utimesat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct futimesat_args { int fd; const char * path; const struct timeval * times; }; #endif int sys_futimesat(struct thread *td, struct futimesat_args *uap) { return (kern_utimesat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE)); } int kern_utimesat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct nameidata nd; struct timespec ts[2]; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, fd, &cap_futimes_rights, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct lutimes_args { char *path; struct timeval *tptr; }; #endif int sys_lutimes(struct thread *td, struct lutimes_args *uap) { return (kern_lutimes(td, uap->path, UIO_USERSPACE, uap->tptr, UIO_USERSPACE)); } int kern_lutimes(struct thread *td, const char *path, enum uio_seg pathseg, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct nameidata nd; int error; if ((error = getutimes(tptr, tptrseg, ts)) != 0) return (error); NDINIT(&nd, LOOKUP, NOFOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); error = setutimes(td, nd.ni_vp, ts, 2, tptr == NULL); vrele(nd.ni_vp); return (error); } /* * Set the access and modification times of a file. */ #ifndef _SYS_SYSPROTO_H_ struct futimes_args { int fd; struct timeval *tptr; }; #endif int sys_futimes(struct thread *td, struct futimes_args *uap) { return (kern_futimes(td, uap->fd, uap->tptr, UIO_USERSPACE)); } int kern_futimes(struct thread *td, int fd, struct timeval *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; int error; AUDIT_ARG_FD(fd); error = getutimes(tptr, tptrseg, ts); if (error != 0) return (error); error = getvnode(td, fd, &cap_futimes_rights, &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); #endif error = setutimes(td, fp->f_vnode, ts, 2, tptr == NULL); fdrop(fp, td); return (error); } int sys_futimens(struct thread *td, struct futimens_args *uap) { return (kern_futimens(td, uap->fd, uap->times, UIO_USERSPACE)); } int kern_futimens(struct thread *td, int fd, struct timespec *tptr, enum uio_seg tptrseg) { struct timespec ts[2]; struct file *fp; int error, flags; AUDIT_ARG_FD(fd); error = getutimens(tptr, tptrseg, ts, &flags); if (error != 0) return (error); if (flags & UTIMENS_EXIT) return (0); error = getvnode(td, fd, &cap_futimes_rights, &fp); if (error != 0) return (error); #ifdef AUDIT vn_lock(fp->f_vnode, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(fp->f_vnode); VOP_UNLOCK(fp->f_vnode); #endif error = setutimes(td, fp->f_vnode, ts, 2, flags & UTIMENS_NULL); fdrop(fp, td); return (error); } int sys_utimensat(struct thread *td, struct utimensat_args *uap) { return (kern_utimensat(td, uap->fd, uap->path, UIO_USERSPACE, uap->times, UIO_USERSPACE, uap->flag)); } int kern_utimensat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, struct timespec *tptr, enum uio_seg tptrseg, int flag) { struct nameidata nd; struct timespec ts[2]; int error, flags; if ((flag & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); if ((error = getutimens(tptr, tptrseg, ts, &flags)) != 0) return (error); NDINIT_ATRIGHTS(&nd, LOOKUP, at2cnpflags(flag, AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH) | AUDITVNODE1, pathseg, path, fd, &cap_futimes_rights, td); if ((error = namei(&nd)) != 0) return (error); /* * We are allowed to call namei() regardless of 2xUTIME_OMIT. * POSIX states: * "If both tv_nsec fields are UTIME_OMIT... EACCESS may be detected." * "Search permission is denied by a component of the path prefix." */ NDFREE(&nd, NDF_ONLY_PNBUF); if ((flags & UTIMENS_EXIT) == 0) error = setutimes(td, nd.ni_vp, ts, 2, flags & UTIMENS_NULL); vrele(nd.ni_vp); return (error); } /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct truncate_args { char *path; int pad; off_t length; }; #endif int sys_truncate(struct thread *td, struct truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int kern_truncate(struct thread *td, const char *path, enum uio_seg pathseg, off_t length) { struct mount *mp; struct vnode *vp; void *rl_cookie; struct vattr vattr; struct nameidata nd; int error; if (length < 0) - return(EINVAL); + return (EINVAL); +retry: NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, pathseg, path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); if ((error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) { vn_rangelock_unlock(vp, rl_cookie); vrele(vp); return (error); } NDFREE(&nd, NDF_ONLY_PNBUF); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) error = EISDIR; #ifdef MAC else if ((error = mac_vnode_check_write(td->td_ucred, NOCRED, vp))) { } #endif else if ((error = vn_writechk(vp)) == 0 && (error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td)) == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, td->td_ucred); } VOP_UNLOCK(vp); vn_finished_write(mp); vn_rangelock_unlock(vp, rl_cookie); vrele(vp); + if (error == ERELOOKUP) + goto retry; return (error); } #if defined(COMPAT_43) /* * Truncate a file given its path name. */ #ifndef _SYS_SYSPROTO_H_ struct otruncate_args { char *path; long length; }; #endif int otruncate(struct thread *td, struct otruncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD6) /* Versions with the pad argument */ int freebsd6_truncate(struct thread *td, struct freebsd6_truncate_args *uap) { return (kern_truncate(td, uap->path, UIO_USERSPACE, uap->length)); } int freebsd6_ftruncate(struct thread *td, struct freebsd6_ftruncate_args *uap) { return (kern_ftruncate(td, uap->fd, uap->length)); } #endif int kern_fsync(struct thread *td, int fd, bool fullsync) { struct vnode *vp; struct mount *mp; struct file *fp; int error, lock_flags; AUDIT_ARG_FD(fd); error = getvnode(td, fd, &cap_fsync_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; #if 0 if (!fullsync) /* XXXKIB: compete outstanding aio writes */; #endif +retry: error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error != 0) goto drop; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); AUDIT_ARG_VNODE1(vp); if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = fullsync ? VOP_FSYNC(vp, MNT_WAIT, td) : VOP_FDATASYNC(vp, td); VOP_UNLOCK(vp); vn_finished_write(mp); + if (error == ERELOOKUP) + goto retry; drop: fdrop(fp, td); return (error); } /* * Sync an open file. */ #ifndef _SYS_SYSPROTO_H_ struct fsync_args { int fd; }; #endif int sys_fsync(struct thread *td, struct fsync_args *uap) { return (kern_fsync(td, uap->fd, true)); } int sys_fdatasync(struct thread *td, struct fdatasync_args *uap) { return (kern_fsync(td, uap->fd, false)); } /* * Rename files. Source and destination must either both be directories, or * both not be directories. If target is a directory, it must be empty. */ #ifndef _SYS_SYSPROTO_H_ struct rename_args { char *from; char *to; }; #endif int sys_rename(struct thread *td, struct rename_args *uap) { return (kern_renameat(td, AT_FDCWD, uap->from, AT_FDCWD, uap->to, UIO_USERSPACE)); } #ifndef _SYS_SYSPROTO_H_ struct renameat_args { int oldfd; char *old; int newfd; char *new; }; #endif int sys_renameat(struct thread *td, struct renameat_args *uap) { return (kern_renameat(td, uap->oldfd, uap->old, uap->newfd, uap->new, UIO_USERSPACE)); } #ifdef MAC static int kern_renameat_mac(struct thread *td, int oldfd, const char *old, int newfd, const char *new, enum uio_seg pathseg, struct nameidata *fromnd) { int error; NDINIT_ATRIGHTS(fromnd, DELETE, LOCKPARENT | LOCKLEAF | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights, td); if ((error = namei(fromnd)) != 0) return (error); error = mac_vnode_check_rename_from(td->td_ucred, fromnd->ni_dvp, fromnd->ni_vp, &fromnd->ni_cnd); VOP_UNLOCK(fromnd->ni_dvp); if (fromnd->ni_dvp != fromnd->ni_vp) VOP_UNLOCK(fromnd->ni_vp); if (error != 0) { NDFREE(fromnd, NDF_ONLY_PNBUF); vrele(fromnd->ni_dvp); vrele(fromnd->ni_vp); if (fromnd->ni_startdir) vrele(fromnd->ni_startdir); } return (error); } #endif int kern_renameat(struct thread *td, int oldfd, const char *old, int newfd, const char *new, enum uio_seg pathseg) { struct mount *mp = NULL; struct vnode *tvp, *fvp, *tdvp; struct nameidata fromnd, tond; int error; again: bwillwrite(); #ifdef MAC if (mac_vnode_check_rename_from_enabled()) { error = kern_renameat_mac(td, oldfd, old, newfd, new, pathseg, &fromnd); if (error != 0) return (error); } else { #endif NDINIT_ATRIGHTS(&fromnd, DELETE, WANTPARENT | SAVESTART | AUDITVNODE1, pathseg, old, oldfd, &cap_renameat_source_rights, td); if ((error = namei(&fromnd)) != 0) return (error); #ifdef MAC } #endif fvp = fromnd.ni_vp; NDINIT_ATRIGHTS(&tond, RENAME, LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | AUDITVNODE2, pathseg, new, newfd, &cap_renameat_target_rights, td); if (fromnd.ni_vp->v_type == VDIR) tond.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&tond)) != 0) { /* Translate error code for rename("dir1", "dir2/."). */ if (error == EISDIR && fvp->v_type == VDIR) error = EINVAL; NDFREE(&fromnd, NDF_ONLY_PNBUF); vrele(fromnd.ni_dvp); vrele(fvp); goto out1; } tdvp = tond.ni_dvp; tvp = tond.ni_vp; error = vn_start_write(fvp, &mp, V_NOWAIT); if (error != 0) { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); vrele(tond.ni_startdir); if (fromnd.ni_startdir != NULL) vrele(fromnd.ni_startdir); error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH); if (error != 0) return (error); goto again; } if (tvp != NULL) { if (fvp->v_type == VDIR && tvp->v_type != VDIR) { error = ENOTDIR; goto out; } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef CAPABILITIES if (newfd != AT_FDCWD && (tond.ni_resflags & NIRES_ABS) == 0) { /* * If the target already exists we require CAP_UNLINKAT * from 'newfd', when newfd was used for the lookup. */ error = cap_check(&tond.ni_filecaps.fc_rights, &cap_unlinkat_rights); if (error != 0) goto out; } #endif } if (fvp == tdvp) { error = EINVAL; goto out; } /* * If the source is the same as the destination (that is, if they * are links to the same vnode), then there is nothing to do. */ if (fvp == tvp) - error = -1; + error = ERESTART; #ifdef MAC else error = mac_vnode_check_rename_to(td->td_ucred, tdvp, tond.ni_vp, fromnd.ni_dvp == tdvp, &tond.ni_cnd); #endif out: if (error == 0) { error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd, tond.ni_dvp, tond.ni_vp, &tond.ni_cnd); NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); } else { NDFREE(&fromnd, NDF_ONLY_PNBUF); NDFREE(&tond, NDF_ONLY_PNBUF); if (tvp != NULL) vput(tvp); if (tdvp == tvp) vrele(tdvp); else vput(tdvp); vrele(fromnd.ni_dvp); vrele(fvp); } vrele(tond.ni_startdir); vn_finished_write(mp); out1: if (fromnd.ni_startdir) vrele(fromnd.ni_startdir); - if (error == -1) + if (error == ERESTART) return (0); + if (error == ERELOOKUP) + goto again; return (error); } /* * Make a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct mkdir_args { char *path; int mode; }; #endif int sys_mkdir(struct thread *td, struct mkdir_args *uap) { return (kern_mkdirat(td, AT_FDCWD, uap->path, UIO_USERSPACE, uap->mode)); } #ifndef _SYS_SYSPROTO_H_ struct mkdirat_args { int fd; char *path; mode_t mode; }; #endif int sys_mkdirat(struct thread *td, struct mkdirat_args *uap) { return (kern_mkdirat(td, uap->fd, uap->path, UIO_USERSPACE, uap->mode)); } int kern_mkdirat(struct thread *td, int fd, const char *path, enum uio_seg segflg, int mode) { struct mount *mp; struct vnode *vp; struct vattr vattr; struct nameidata nd; int error; AUDIT_ARG_MODE(mode); restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, CREATE, LOCKPARENT | SAVENAME | AUDITVNODE1 | NC_NOMAKEENTRY | NC_KEEPPOSENTRY, segflg, path, fd, &cap_mkdirat_rights, td); nd.ni_cnd.cn_flags |= WILLBEDIR; if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; if (vp != NULL) { NDFREE(&nd, NDF_ONLY_PNBUF); /* * XXX namei called with LOCKPARENT but not LOCKLEAF has * the strange behaviour of leaving the vnode unlocked * if the target is the same vnode as the parent. */ if (vp == nd.ni_dvp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); vrele(vp); return (EEXIST); } if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } VATTR_NULL(&vattr); vattr.va_type = VDIR; vattr.va_mode = (mode & ACCESSPERMS) &~ td->td_proc->p_fd->fd_cmask; #ifdef MAC error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd, &vattr); if (error != 0) goto out; #endif error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); #ifdef MAC out: #endif NDFREE(&nd, NDF_ONLY_PNBUF); vput(nd.ni_dvp); if (error == 0) vput(nd.ni_vp); vn_finished_write(mp); + if (error == ERELOOKUP) + goto restart; return (error); } /* * Remove a directory file. */ #ifndef _SYS_SYSPROTO_H_ struct rmdir_args { char *path; }; #endif int sys_rmdir(struct thread *td, struct rmdir_args *uap) { return (kern_frmdirat(td, AT_FDCWD, uap->path, FD_NONE, UIO_USERSPACE, 0)); } int kern_frmdirat(struct thread *td, int dfd, const char *path, int fd, enum uio_seg pathseg, int flag) { struct mount *mp; struct vnode *vp; struct file *fp; struct nameidata nd; cap_rights_t rights; int error; fp = NULL; if (fd != FD_NONE) { error = getvnode(td, fd, cap_rights_init_one(&rights, CAP_LOOKUP), &fp); if (error != 0) return (error); } restart: bwillwrite(); NDINIT_ATRIGHTS(&nd, DELETE, LOCKPARENT | LOCKLEAF | AUDITVNODE1 | at2cnpflags(flag, AT_BENEATH | AT_RESOLVE_BENEATH), pathseg, path, dfd, &cap_unlinkat_rights, td); if ((error = namei(&nd)) != 0) goto fdout; vp = nd.ni_vp; if (vp->v_type != VDIR) { error = ENOTDIR; goto out; } /* * No rmdir "." please. */ if (nd.ni_dvp == vp) { error = EINVAL; goto out; } /* * The root of a mounted filesystem cannot be deleted. */ if (vp->v_vflag & VV_ROOT) { error = EBUSY; goto out; } if (fp != NULL && fp->f_vnode != vp) { if (VN_IS_DOOMED(fp->f_vnode)) error = EBADF; else error = EDEADLK; goto out; } #ifdef MAC error = mac_vnode_check_unlink(td->td_ucred, nd.ni_dvp, vp, &nd.ni_cnd); if (error != 0) goto out; #endif if (vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) goto fdout; goto restart; } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_UNLINK); error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd); vn_finished_write(mp); out: NDFREE(&nd, NDF_ONLY_PNBUF); vput(vp); if (nd.ni_dvp == vp) vrele(nd.ni_dvp); else vput(nd.ni_dvp); + if (error == ERELOOKUP) + goto restart; fdout: if (fp != NULL) fdrop(fp, td); return (error); } #if defined(COMPAT_43) || defined(COMPAT_FREEBSD11) int freebsd11_kern_getdirentries(struct thread *td, int fd, char *ubuf, u_int count, long *basep, void (*func)(struct freebsd11_dirent *)) { struct freebsd11_dirent dstdp; struct dirent *dp, *edp; char *dirbuf; off_t base; ssize_t resid, ucount; int error; /* XXX arbitrary sanity limit on `count'. */ count = min(count, 64 * 1024); dirbuf = malloc(count, M_TEMP, M_WAITOK); error = kern_getdirentries(td, fd, dirbuf, count, &base, &resid, UIO_SYSSPACE); if (error != 0) goto done; if (basep != NULL) *basep = base; ucount = 0; for (dp = (struct dirent *)dirbuf, edp = (struct dirent *)&dirbuf[count - resid]; ucount < count && dp < edp; ) { if (dp->d_reclen == 0) break; MPASS(dp->d_reclen >= _GENERIC_DIRLEN(0)); if (dp->d_namlen >= sizeof(dstdp.d_name)) continue; dstdp.d_type = dp->d_type; dstdp.d_namlen = dp->d_namlen; dstdp.d_fileno = dp->d_fileno; /* truncate */ if (dstdp.d_fileno != dp->d_fileno) { switch (ino64_trunc_error) { default: case 0: break; case 1: error = EOVERFLOW; goto done; case 2: dstdp.d_fileno = UINT32_MAX; break; } } dstdp.d_reclen = sizeof(dstdp) - sizeof(dstdp.d_name) + ((dp->d_namlen + 1 + 3) &~ 3); bcopy(dp->d_name, dstdp.d_name, dstdp.d_namlen); bzero(dstdp.d_name + dstdp.d_namlen, dstdp.d_reclen - offsetof(struct freebsd11_dirent, d_name) - dstdp.d_namlen); MPASS(dstdp.d_reclen <= dp->d_reclen); MPASS(ucount + dstdp.d_reclen <= count); if (func != NULL) func(&dstdp); error = copyout(&dstdp, ubuf + ucount, dstdp.d_reclen); if (error != 0) break; dp = (struct dirent *)((char *)dp + dp->d_reclen); ucount += dstdp.d_reclen; } done: free(dirbuf, M_TEMP); if (error == 0) td->td_retval[0] = ucount; return (error); } #endif /* COMPAT */ #ifdef COMPAT_43 static void ogetdirentries_cvt(struct freebsd11_dirent *dp) { #if (BYTE_ORDER == LITTLE_ENDIAN) /* * The expected low byte of dp->d_namlen is our dp->d_type. * The high MBZ byte of dp->d_namlen is our dp->d_namlen. */ dp->d_type = dp->d_namlen; dp->d_namlen = 0; #else /* * The dp->d_type is the high byte of the expected dp->d_namlen, * so must be zero'ed. */ dp->d_type = 0; #endif } /* * Read a block of directory entries in a filesystem independent format. */ #ifndef _SYS_SYSPROTO_H_ struct ogetdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int ogetdirentries(struct thread *td, struct ogetdirentries_args *uap) { long loff; int error; error = kern_ogetdirentries(td, uap, &loff); if (error == 0) error = copyout(&loff, uap->basep, sizeof(long)); return (error); } int kern_ogetdirentries(struct thread *td, struct ogetdirentries_args *uap, long *ploff) { long base; int error; /* XXX arbitrary sanity limit on `count'. */ if (uap->count > 64 * 1024) return (EINVAL); error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, ogetdirentries_cvt); if (error == 0 && uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD11) #ifndef _SYS_SYSPROTO_H_ struct freebsd11_getdirentries_args { int fd; char *buf; u_int count; long *basep; }; #endif int freebsd11_getdirentries(struct thread *td, struct freebsd11_getdirentries_args *uap) { long base; int error; error = freebsd11_kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL); if (error == 0 && uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(long)); return (error); } int freebsd11_getdents(struct thread *td, struct freebsd11_getdents_args *uap) { struct freebsd11_getdirentries_args ap; ap.fd = uap->fd; ap.buf = uap->buf; ap.count = uap->count; ap.basep = NULL; return (freebsd11_getdirentries(td, &ap)); } #endif /* COMPAT_FREEBSD11 */ /* * Read a block of directory entries in a filesystem independent format. */ int sys_getdirentries(struct thread *td, struct getdirentries_args *uap) { off_t base; int error; error = kern_getdirentries(td, uap->fd, uap->buf, uap->count, &base, NULL, UIO_USERSPACE); if (error != 0) return (error); if (uap->basep != NULL) error = copyout(&base, uap->basep, sizeof(off_t)); return (error); } int kern_getdirentries(struct thread *td, int fd, char *buf, size_t count, off_t *basep, ssize_t *residp, enum uio_seg bufseg) { struct vnode *vp; struct file *fp; struct uio auio; struct iovec aiov; off_t loff; int error, eofflag; off_t foffset; AUDIT_ARG_FD(fd); if (count > IOSIZE_MAX) return (EINVAL); auio.uio_resid = count; error = getvnode(td, fd, &cap_read_rights, &fp); if (error != 0) return (error); if ((fp->f_flag & FREAD) == 0) { fdrop(fp, td); return (EBADF); } vp = fp->f_vnode; foffset = foffset_lock(fp, 0); unionread: if (vp->v_type != VDIR) { error = EINVAL; goto fail; } aiov.iov_base = buf; aiov.iov_len = count; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = bufseg; auio.uio_td = td; vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); loff = auio.uio_offset = foffset; #ifdef MAC error = mac_vnode_check_readdir(td->td_ucred, vp); if (error == 0) #endif error = VOP_READDIR(vp, &auio, fp->f_cred, &eofflag, NULL, NULL); foffset = auio.uio_offset; if (error != 0) { VOP_UNLOCK(vp); goto fail; } if (count == auio.uio_resid && (vp->v_vflag & VV_ROOT) && (vp->v_mount->mnt_flag & MNT_UNION)) { struct vnode *tvp = vp; vp = vp->v_mount->mnt_vnodecovered; VREF(vp); fp->f_vnode = vp; foffset = 0; vput(tvp); goto unionread; } VOP_UNLOCK(vp); *basep = loff; if (residp != NULL) *residp = auio.uio_resid; td->td_retval[0] = count - auio.uio_resid; fail: foffset_unlock(fp, foffset, 0); fdrop(fp, td); return (error); } /* * Set the mode mask for creation of filesystem nodes. */ #ifndef _SYS_SYSPROTO_H_ struct umask_args { int newmask; }; #endif int sys_umask(struct thread *td, struct umask_args *uap) { struct filedesc *fdp; fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); td->td_retval[0] = fdp->fd_cmask; fdp->fd_cmask = uap->newmask & ALLPERMS; FILEDESC_XUNLOCK(fdp); return (0); } /* * Void all references to file by ripping underlying filesystem away from * vnode. */ #ifndef _SYS_SYSPROTO_H_ struct revoke_args { char *path; }; #endif int sys_revoke(struct thread *td, struct revoke_args *uap) { struct vnode *vp; struct vattr vattr; struct nameidata nd; int error; NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_USERSPACE, uap->path, td); if ((error = namei(&nd)) != 0) return (error); vp = nd.ni_vp; NDFREE(&nd, NDF_ONLY_PNBUF); if (vp->v_type != VCHR || vp->v_rdev == NULL) { error = EINVAL; goto out; } #ifdef MAC error = mac_vnode_check_revoke(td->td_ucred, vp); if (error != 0) goto out; #endif error = VOP_GETATTR(vp, &vattr, td->td_ucred); if (error != 0) goto out; if (td->td_ucred->cr_uid != vattr.va_uid) { error = priv_check(td, PRIV_VFS_ADMIN); if (error != 0) goto out; } if (devfs_usecount(vp) > 0) VOP_REVOKE(vp, REVOKEALL); out: vput(vp); return (error); } /* * Convert a user file descriptor to a kernel file entry and check that, if it * is a capability, the correct rights are present. A reference on the file * entry is held upon returning. */ int getvnode(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { struct file *fp; int error; error = fget_unlocked(td->td_proc->p_fd, fd, rightsp, &fp); if (error != 0) return (error); /* * The file could be not of the vnode type, or it may be not * yet fully initialized, in which case the f_vnode pointer * may be set, but f_ops is still badfileops. E.g., * devfs_open() transiently create such situation to * facilitate csw d_fdopen(). * * Dupfdopen() handling in kern_openat() installs the * half-baked file into the process descriptor table, allowing * other thread to dereference it. Guard against the race by * checking f_ops. */ if (fp->f_vnode == NULL || fp->f_ops == &badfileops) { fdrop(fp, td); return (EINVAL); } *fpp = fp; return (0); } /* * Get an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct lgetfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_lgetfh(struct thread *td, struct lgetfh_args *uap) { return (kern_getfhat(td, AT_SYMLINK_NOFOLLOW, AT_FDCWD, uap->fname, UIO_USERSPACE, uap->fhp)); } #ifndef _SYS_SYSPROTO_H_ struct getfh_args { char *fname; fhandle_t *fhp; }; #endif int sys_getfh(struct thread *td, struct getfh_args *uap) { return (kern_getfhat(td, 0, AT_FDCWD, uap->fname, UIO_USERSPACE, uap->fhp)); } /* * syscall for the rpc.lockd to use to translate an open descriptor into * a NFS file handle. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct getfhat_args { int fd; char *path; fhandle_t *fhp; int flags; }; #endif int sys_getfhat(struct thread *td, struct getfhat_args *uap) { if ((uap->flags & ~(AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH)) != 0) return (EINVAL); return (kern_getfhat(td, uap->flags, uap->fd, uap->path, UIO_USERSPACE, uap->fhp)); } static int kern_getfhat(struct thread *td, int flags, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp) { struct nameidata nd; fhandle_t fh; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); NDINIT_AT(&nd, LOOKUP, at2cnpflags(flags, AT_SYMLINK_NOFOLLOW | AT_BENEATH | AT_RESOLVE_BENEATH) | LOCKLEAF | AUDITVNODE1, pathseg, path, fd, td); error = namei(&nd); if (error != 0) return (error); NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; bzero(&fh, sizeof(fh)); fh.fh_fsid = vp->v_mount->mnt_stat.f_fsid; error = VOP_VPTOFH(vp, &fh.fh_fid); vput(vp); if (error == 0) error = copyout(&fh, fhp, sizeof (fh)); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fhlink_args { fhandle_t *fhp; const char *to; }; #endif int sys_fhlink(struct thread *td, struct fhlink_args *uap) { return (kern_fhlinkat(td, AT_FDCWD, uap->to, UIO_USERSPACE, uap->fhp)); } #ifndef _SYS_SYSPROTO_H_ struct fhlinkat_args { fhandle_t *fhp; int tofd; const char *to; }; #endif int sys_fhlinkat(struct thread *td, struct fhlinkat_args *uap) { return (kern_fhlinkat(td, uap->tofd, uap->to, UIO_USERSPACE, uap->fhp)); } static int kern_fhlinkat(struct thread *td, int fd, const char *path, enum uio_seg pathseg, fhandle_t *fhp) { fhandle_t fh; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); error = copyin(fhp, &fh, sizeof(fh)); if (error != 0) return (error); do { bwillwrite(); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp); vfs_unbusy(mp); if (error != 0) return (error); VOP_UNLOCK(vp); - } while ((error = kern_linkat_vp(td, vp, fd, path, pathseg)) == EAGAIN); + error = kern_linkat_vp(td, vp, fd, path, pathseg); + } while (error == EAGAIN || error == ERELOOKUP); return (error); } #ifndef _SYS_SYSPROTO_H_ struct fhreadlink_args { fhandle_t *fhp; char *buf; size_t bufsize; }; #endif int sys_fhreadlink(struct thread *td, struct fhreadlink_args *uap) { fhandle_t fh; struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_GETFH); if (error != 0) return (error); if (uap->bufsize > IOSIZE_MAX) return (EINVAL); error = copyin(uap->fhp, &fh, sizeof(fh)); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_SHARED, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = kern_readlink_vp(vp, uap->buf, UIO_USERSPACE, uap->bufsize, td); vput(vp); return (error); } /* * syscall for the rpc.lockd to use to translate a NFS file handle into an * open descriptor. * * warning: do not remove the priv_check() call or this becomes one giant * security hole. */ #ifndef _SYS_SYSPROTO_H_ struct fhopen_args { const struct fhandle *u_fhp; int flags; }; #endif int sys_fhopen(struct thread *td, struct fhopen_args *uap) { struct mount *mp; struct vnode *vp; struct fhandle fhp; struct file *fp; int fmode, error; int indx; error = priv_check(td, PRIV_VFS_FHOPEN); if (error != 0) return (error); indx = -1; fmode = FFLAGS(uap->flags); /* why not allow a non-read/write open for our lockd? */ if (((fmode & (FREAD | FWRITE)) == 0) || (fmode & O_CREAT)) return (EINVAL); error = copyin(uap->u_fhp, &fhp, sizeof(fhp)); if (error != 0) return(error); /* find the mount point */ mp = vfs_busyfs(&fhp.fh_fsid); if (mp == NULL) return (ESTALE); /* now give me my vnode, it gets returned to me locked */ error = VFS_FHTOVP(mp, &fhp.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = falloc_noinstall(td, &fp); if (error != 0) { vput(vp); return (error); } /* * An extra reference on `fp' has been held for us by * falloc_noinstall(). */ #ifdef INVARIANTS td->td_dupfd = -1; #endif error = vn_open_vnode(vp, fmode, td->td_ucred, td, fp); if (error != 0) { KASSERT(fp->f_ops == &badfileops, ("VOP_OPEN in fhopen() set f_ops")); KASSERT(td->td_dupfd < 0, ("fhopen() encountered fdopen()")); vput(vp); goto bad; } #ifdef INVARIANTS td->td_dupfd = 0; #endif fp->f_vnode = vp; finit_vnode(fp, fmode, NULL, &vnops); VOP_UNLOCK(vp); if ((fmode & O_TRUNC) != 0) { error = fo_truncate(fp, 0, td->td_ucred, td); if (error != 0) goto bad; } error = finstall(td, fp, &indx, fmode, NULL); bad: fdrop(fp, td); td->td_retval[0] = indx; return (error); } /* * Stat an (NFS) file handle. */ #ifndef _SYS_SYSPROTO_H_ struct fhstat_args { struct fhandle *u_fhp; struct stat *sb; }; #endif int sys_fhstat(struct thread *td, struct fhstat_args *uap) { struct stat sb; struct fhandle fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fh)); if (error != 0) return (error); error = kern_fhstat(td, fh, &sb); if (error == 0) error = copyout(&sb, uap->sb, sizeof(sb)); return (error); } int kern_fhstat(struct thread *td, struct fhandle fh, struct stat *sb) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTAT); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); vfs_unbusy(mp); if (error != 0) return (error); error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td); vput(vp); return (error); } /* * Implement fstatfs() for (NFS) file handles. */ #ifndef _SYS_SYSPROTO_H_ struct fhstatfs_args { struct fhandle *u_fhp; struct statfs *buf; }; #endif int sys_fhstatfs(struct thread *td, struct fhstatfs_args *uap) { struct statfs *sfp; fhandle_t fh; int error; error = copyin(uap->u_fhp, &fh, sizeof(fhandle_t)); if (error != 0) return (error); sfp = malloc(sizeof(struct statfs), M_STATFS, M_WAITOK); error = kern_fhstatfs(td, fh, sfp); if (error == 0) error = copyout(sfp, uap->buf, sizeof(*sfp)); free(sfp, M_STATFS); return (error); } int kern_fhstatfs(struct thread *td, fhandle_t fh, struct statfs *buf) { struct mount *mp; struct vnode *vp; int error; error = priv_check(td, PRIV_VFS_FHSTATFS); if (error != 0) return (error); if ((mp = vfs_busyfs(&fh.fh_fsid)) == NULL) return (ESTALE); error = VFS_FHTOVP(mp, &fh.fh_fid, LK_EXCLUSIVE, &vp); if (error != 0) { vfs_unbusy(mp); return (error); } vput(vp); error = prison_canseemount(td->td_ucred, mp); if (error != 0) goto out; #ifdef MAC error = mac_mount_check_stat(td->td_ucred, mp); if (error != 0) goto out; #endif error = VFS_STATFS(mp, buf); out: vfs_unbusy(mp); return (error); } /* * Unlike madvise(2), we do not make a best effort to remember every * possible caching hint. Instead, we remember the last setting with * the exception that we will allow POSIX_FADV_NORMAL to adjust the * region of any current setting. */ int kern_posix_fadvise(struct thread *td, int fd, off_t offset, off_t len, int advice) { struct fadvise_info *fa, *new; struct file *fp; struct vnode *vp; off_t end; int error; if (offset < 0 || len < 0 || offset > OFF_MAX - len) return (EINVAL); AUDIT_ARG_VALUE(advice); switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: new = malloc(sizeof(*fa), M_FADVISE, M_WAITOK); break; case POSIX_FADV_NORMAL: case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: new = NULL; break; default: return (EINVAL); } /* XXX: CAP_POSIX_FADVISE? */ AUDIT_ARG_FD(fd); error = fget(td, fd, &cap_no_rights, &fp); if (error != 0) goto out; AUDIT_ARG_FILE(td->td_proc, fp); if ((fp->f_ops->fo_flags & DFLAG_SEEKABLE) == 0) { error = ESPIPE; goto out; } if (fp->f_type != DTYPE_VNODE) { error = ENODEV; goto out; } vp = fp->f_vnode; if (vp->v_type != VREG) { error = ENODEV; goto out; } if (len == 0) end = OFF_MAX; else end = offset + len - 1; switch (advice) { case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_RANDOM: case POSIX_FADV_NOREUSE: /* * Try to merge any existing non-standard region with * this new region if possible, otherwise create a new * non-standard region for this request. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL && fa->fa_advice == advice && ((fa->fa_start <= end && fa->fa_end >= offset) || (end != OFF_MAX && fa->fa_start == end + 1) || (fa->fa_end != OFF_MAX && fa->fa_end + 1 == offset))) { if (offset < fa->fa_start) fa->fa_start = offset; if (end > fa->fa_end) fa->fa_end = end; } else { new->fa_advice = advice; new->fa_start = offset; new->fa_end = end; fp->f_advice = new; new = fa; } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_NORMAL: /* * If a the "normal" region overlaps with an existing * non-standard region, trim or remove the * non-standard region. */ mtx_pool_lock(mtxpool_sleep, fp); fa = fp->f_advice; if (fa != NULL) { if (offset <= fa->fa_start && end >= fa->fa_end) { new = fa; fp->f_advice = NULL; } else if (offset <= fa->fa_start && end >= fa->fa_start) fa->fa_start = end + 1; else if (offset <= fa->fa_end && end >= fa->fa_end) fa->fa_end = offset - 1; else if (offset >= fa->fa_start && end <= fa->fa_end) { /* * If the "normal" region is a middle * portion of the existing * non-standard region, just remove * the whole thing rather than picking * one side or the other to * preserve. */ new = fa; fp->f_advice = NULL; } } mtx_pool_unlock(mtxpool_sleep, fp); break; case POSIX_FADV_WILLNEED: case POSIX_FADV_DONTNEED: error = VOP_ADVISE(vp, offset, end, advice); break; } out: if (fp != NULL) fdrop(fp, td); free(new, M_FADVISE); return (error); } int sys_posix_fadvise(struct thread *td, struct posix_fadvise_args *uap) { int error; error = kern_posix_fadvise(td, uap->fd, uap->offset, uap->len, uap->advice); return (kern_posix_error(td, error)); } int kern_copy_file_range(struct thread *td, int infd, off_t *inoffp, int outfd, off_t *outoffp, size_t len, unsigned int flags) { struct file *infp, *outfp; struct vnode *invp, *outvp; int error; size_t retlen; void *rl_rcookie, *rl_wcookie; off_t savinoff, savoutoff; infp = outfp = NULL; rl_rcookie = rl_wcookie = NULL; savinoff = -1; error = 0; retlen = 0; if (flags != 0) { error = EINVAL; goto out; } if (len > SSIZE_MAX) /* * Although the len argument is size_t, the return argument * is ssize_t (which is signed). Therefore a size that won't * fit in ssize_t can't be returned. */ len = SSIZE_MAX; /* Get the file structures for the file descriptors. */ error = fget_read(td, infd, &cap_read_rights, &infp); if (error != 0) goto out; if (infp->f_ops == &badfileops) { error = EBADF; goto out; } if (infp->f_vnode == NULL) { error = EINVAL; goto out; } error = fget_write(td, outfd, &cap_write_rights, &outfp); if (error != 0) goto out; if (outfp->f_ops == &badfileops) { error = EBADF; goto out; } if (outfp->f_vnode == NULL) { error = EINVAL; goto out; } /* Set the offset pointers to the correct place. */ if (inoffp == NULL) inoffp = &infp->f_offset; if (outoffp == NULL) outoffp = &outfp->f_offset; savinoff = *inoffp; savoutoff = *outoffp; invp = infp->f_vnode; outvp = outfp->f_vnode; /* Sanity check the f_flag bits. */ if ((outfp->f_flag & (FWRITE | FAPPEND)) != FWRITE || (infp->f_flag & FREAD) == 0) { error = EBADF; goto out; } /* If len == 0, just return 0. */ if (len == 0) goto out; /* * If infp and outfp refer to the same file, the byte ranges cannot * overlap. */ if (invp == outvp && ((savinoff <= savoutoff && savinoff + len > savoutoff) || (savinoff > savoutoff && savoutoff + len > savinoff))) { error = EINVAL; goto out; } /* Range lock the byte ranges for both invp and outvp. */ for (;;) { rl_wcookie = vn_rangelock_wlock(outvp, *outoffp, *outoffp + len); rl_rcookie = vn_rangelock_tryrlock(invp, *inoffp, *inoffp + len); if (rl_rcookie != NULL) break; vn_rangelock_unlock(outvp, rl_wcookie); rl_rcookie = vn_rangelock_rlock(invp, *inoffp, *inoffp + len); vn_rangelock_unlock(invp, rl_rcookie); } retlen = len; error = vn_copy_file_range(invp, inoffp, outvp, outoffp, &retlen, flags, infp->f_cred, outfp->f_cred, td); out: if (rl_rcookie != NULL) vn_rangelock_unlock(invp, rl_rcookie); if (rl_wcookie != NULL) vn_rangelock_unlock(outvp, rl_wcookie); if (savinoff != -1 && (error == EINTR || error == ERESTART)) { *inoffp = savinoff; *outoffp = savoutoff; } if (outfp != NULL) fdrop(outfp, td); if (infp != NULL) fdrop(infp, td); td->td_retval[0] = retlen; return (error); } int sys_copy_file_range(struct thread *td, struct copy_file_range_args *uap) { off_t inoff, outoff, *inoffp, *outoffp; int error; inoffp = outoffp = NULL; if (uap->inoffp != NULL) { error = copyin(uap->inoffp, &inoff, sizeof(off_t)); if (error != 0) return (error); inoffp = &inoff; } if (uap->outoffp != NULL) { error = copyin(uap->outoffp, &outoff, sizeof(off_t)); if (error != 0) return (error); outoffp = &outoff; } error = kern_copy_file_range(td, uap->infd, inoffp, uap->outfd, outoffp, uap->len, uap->flags); if (error == 0 && uap->inoffp != NULL) error = copyout(inoffp, uap->inoffp, sizeof(off_t)); if (error == 0 && uap->outoffp != NULL) error = copyout(outoffp, uap->outoffp, sizeof(off_t)); return (error); }