Index: head/sys/kern/kern_descrip.c =================================================================== --- head/sys/kern/kern_descrip.c (revision 366596) +++ head/sys/kern/kern_descrip.c (revision 366597) @@ -1,4630 +1,4735 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1991, 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. * * @(#)kern_descrip.c 8.6 (Berkeley) 4/19/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_ddb.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 #include #include #include #ifdef KTRACE #include #endif #include #include #include #include #include static MALLOC_DEFINE(M_FILEDESC, "filedesc", "Open file descriptor table"); static MALLOC_DEFINE(M_PWD, "pwd", "Descriptor table vnodes"); static MALLOC_DEFINE(M_FILEDESC_TO_LEADER, "filedesc_to_leader", "file desc to leader structures"); static MALLOC_DEFINE(M_SIGIO, "sigio", "sigio structures"); MALLOC_DEFINE(M_FILECAPS, "filecaps", "descriptor capabilities"); MALLOC_DECLARE(M_FADVISE); static __read_mostly uma_zone_t file_zone; static __read_mostly uma_zone_t filedesc0_zone; __read_mostly uma_zone_t pwd_zone; VFS_SMR_DECLARE; static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders); static int fd_first_free(struct filedesc *fdp, int low, int size); static void fdgrowtable(struct filedesc *fdp, int nfd); static void fdgrowtable_exp(struct filedesc *fdp, int nfd); static void fdunused(struct filedesc *fdp, int fd); static void fdused(struct filedesc *fdp, int fd); static int getmaxfd(struct thread *td); static u_long *filecaps_copy_prep(const struct filecaps *src); static void filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, u_long *ioctls); static u_long *filecaps_free_prep(struct filecaps *fcaps); static void filecaps_free_finish(u_long *ioctls); static struct pwd *pwd_alloc(void); /* * Each process has: * * - An array of open file descriptors (fd_ofiles) * - An array of file flags (fd_ofileflags) * - A bitmap recording which descriptors are in use (fd_map) * * A process starts out with NDFILE descriptors. The value of NDFILE has * been selected based the historical limit of 20 open files, and an * assumption that the majority of processes, especially short-lived * processes like shells, will never need more. * * If this initial allocation is exhausted, a larger descriptor table and * map are allocated dynamically, and the pointers in the process's struct * filedesc are updated to point to those. This is repeated every time * the process runs out of file descriptors (provided it hasn't hit its * resource limit). * * Since threads may hold references to individual descriptor table * entries, the tables are never freed. Instead, they are placed on a * linked list and freed only when the struct filedesc is released. */ #define NDFILE 20 #define NDSLOTSIZE sizeof(NDSLOTTYPE) #define NDENTRIES (NDSLOTSIZE * __CHAR_BIT) #define NDSLOT(x) ((x) / NDENTRIES) #define NDBIT(x) ((NDSLOTTYPE)1 << ((x) % NDENTRIES)) #define NDSLOTS(x) (((x) + NDENTRIES - 1) / NDENTRIES) /* * SLIST entry used to keep track of ofiles which must be reclaimed when * the process exits. */ struct freetable { struct fdescenttbl *ft_table; SLIST_ENTRY(freetable) ft_next; }; /* * Initial allocation: a filedesc structure + the head of SLIST used to * keep track of old ofiles + enough space for NDFILE descriptors. */ struct fdescenttbl0 { int fdt_nfiles; struct filedescent fdt_ofiles[NDFILE]; }; struct filedesc0 { struct filedesc fd_fd; SLIST_HEAD(, freetable) fd_free; struct fdescenttbl0 fd_dfiles; NDSLOTTYPE fd_dmap[NDSLOTS(NDFILE)]; }; /* * Descriptor management. */ static int __exclusive_cache_line openfiles; /* actual number of open files */ struct mtx sigio_lock; /* mtx to protect pointers to sigio */ void __read_mostly (*mq_fdclose)(struct thread *td, int fd, struct file *fp); /* * If low >= size, just return low. Otherwise find the first zero bit in the * given bitmap, starting at low and not exceeding size - 1. Return size if * not found. */ static int fd_first_free(struct filedesc *fdp, int low, int size) { NDSLOTTYPE *map = fdp->fd_map; NDSLOTTYPE mask; int off, maxoff; if (low >= size) return (low); off = NDSLOT(low); if (low % NDENTRIES) { mask = ~(~(NDSLOTTYPE)0 >> (NDENTRIES - (low % NDENTRIES))); if ((mask &= ~map[off]) != 0UL) return (off * NDENTRIES + ffsl(mask) - 1); ++off; } for (maxoff = NDSLOTS(size); off < maxoff; ++off) if (map[off] != ~0UL) return (off * NDENTRIES + ffsl(~map[off]) - 1); return (size); } /* * Find the last used fd. * * Call this variant if fdp can't be modified by anyone else (e.g, during exec). * Otherwise use fdlastfile. */ int fdlastfile_single(struct filedesc *fdp) { NDSLOTTYPE *map = fdp->fd_map; int off, minoff; off = NDSLOT(fdp->fd_nfiles - 1); for (minoff = NDSLOT(0); off >= minoff; --off) if (map[off] != 0) return (off * NDENTRIES + flsl(map[off]) - 1); return (-1); } int fdlastfile(struct filedesc *fdp) { FILEDESC_LOCK_ASSERT(fdp); return (fdlastfile_single(fdp)); } static int fdisused(struct filedesc *fdp, int fd) { KASSERT(fd >= 0 && fd < fdp->fd_nfiles, ("file descriptor %d out of range (0, %d)", fd, fdp->fd_nfiles)); return ((fdp->fd_map[NDSLOT(fd)] & NDBIT(fd)) != 0); } /* * Mark a file descriptor as used. */ static void fdused_init(struct filedesc *fdp, int fd) { KASSERT(!fdisused(fdp, fd), ("fd=%d is already used", fd)); fdp->fd_map[NDSLOT(fd)] |= NDBIT(fd); } static void fdused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); fdused_init(fdp, fd); if (fd == fdp->fd_freefile) fdp->fd_freefile++; } /* * Mark a file descriptor as unused. */ static void fdunused(struct filedesc *fdp, int fd) { FILEDESC_XLOCK_ASSERT(fdp); KASSERT(fdisused(fdp, fd), ("fd=%d is already unused", fd)); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("fd=%d is still in use", fd)); fdp->fd_map[NDSLOT(fd)] &= ~NDBIT(fd); if (fd < fdp->fd_freefile) fdp->fd_freefile = fd; } /* * Free a file descriptor. * * Avoid some work if fdp is about to be destroyed. */ static inline void fdefree_last(struct filedescent *fde) { filecaps_free(&fde->fde_caps); } static inline void fdfree(struct filedesc *fdp, int fd) { struct filedescent *fde; fde = &fdp->fd_ofiles[fd]; #ifdef CAPABILITIES seqc_write_begin(&fde->fde_seqc); #endif fde->fde_file = NULL; #ifdef CAPABILITIES seqc_write_end(&fde->fde_seqc); #endif fdefree_last(fde); fdunused(fdp, fd); } /* * System calls on descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct getdtablesize_args { int dummy; }; #endif /* ARGSUSED */ int sys_getdtablesize(struct thread *td, struct getdtablesize_args *uap) { #ifdef RACCT uint64_t lim; #endif td->td_retval[0] = getmaxfd(td); #ifdef RACCT PROC_LOCK(td->td_proc); lim = racct_get_limit(td->td_proc, RACCT_NOFILE); PROC_UNLOCK(td->td_proc); if (lim < td->td_retval[0]) td->td_retval[0] = lim; #endif return (0); } /* * Duplicate a file descriptor to a particular value. * * Note: keep in mind that a potential race condition exists when closing * descriptors from a shared descriptor table (via rfork). */ #ifndef _SYS_SYSPROTO_H_ struct dup2_args { u_int from; u_int to; }; #endif /* ARGSUSED */ int sys_dup2(struct thread *td, struct dup2_args *uap) { return (kern_dup(td, FDDUP_FIXED, 0, (int)uap->from, (int)uap->to)); } /* * Duplicate a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct dup_args { u_int fd; }; #endif /* ARGSUSED */ int sys_dup(struct thread *td, struct dup_args *uap) { return (kern_dup(td, FDDUP_NORMAL, 0, (int)uap->fd, 0)); } /* * The file control system call. */ #ifndef _SYS_SYSPROTO_H_ struct fcntl_args { int fd; int cmd; long arg; }; #endif /* ARGSUSED */ int sys_fcntl(struct thread *td, struct fcntl_args *uap) { return (kern_fcntl_freebsd(td, uap->fd, uap->cmd, uap->arg)); } int kern_fcntl_freebsd(struct thread *td, int fd, int cmd, long arg) { struct flock fl; struct __oflock ofl; intptr_t arg1; int error, newcmd; error = 0; newcmd = cmd; switch (cmd) { case F_OGETLK: case F_OSETLK: case F_OSETLKW: /* * Convert old flock structure to new. */ error = copyin((void *)(intptr_t)arg, &ofl, sizeof(ofl)); fl.l_start = ofl.l_start; fl.l_len = ofl.l_len; fl.l_pid = ofl.l_pid; fl.l_type = ofl.l_type; fl.l_whence = ofl.l_whence; fl.l_sysid = 0; switch (cmd) { case F_OGETLK: newcmd = F_GETLK; break; case F_OSETLK: newcmd = F_SETLK; break; case F_OSETLKW: newcmd = F_SETLKW; break; } arg1 = (intptr_t)&fl; break; case F_GETLK: case F_SETLK: case F_SETLKW: case F_SETLK_REMOTE: error = copyin((void *)(intptr_t)arg, &fl, sizeof(fl)); arg1 = (intptr_t)&fl; break; default: arg1 = arg; break; } if (error) return (error); error = kern_fcntl(td, fd, newcmd, arg1); if (error) return (error); if (cmd == F_OGETLK) { ofl.l_start = fl.l_start; ofl.l_len = fl.l_len; ofl.l_pid = fl.l_pid; ofl.l_type = fl.l_type; ofl.l_whence = fl.l_whence; error = copyout(&ofl, (void *)(intptr_t)arg, sizeof(ofl)); } else if (cmd == F_GETLK) { error = copyout(&fl, (void *)(intptr_t)arg, sizeof(fl)); } return (error); } int kern_fcntl(struct thread *td, int fd, int cmd, intptr_t arg) { struct filedesc *fdp; struct flock *flp; struct file *fp, *fp2; struct filedescent *fde; struct proc *p; struct vnode *vp; struct mount *mp; int error, flg, seals, tmp; uint64_t bsize; off_t foffset; error = 0; flg = F_POSIX; p = td->td_proc; fdp = p->p_fd; AUDIT_ARG_FD(cmd); AUDIT_ARG_CMD(cmd); switch (cmd) { case F_DUPFD: tmp = arg; error = kern_dup(td, FDDUP_FCNTL, 0, fd, tmp); break; case F_DUPFD_CLOEXEC: tmp = arg; error = kern_dup(td, FDDUP_FCNTL, FDDUP_FLAG_CLOEXEC, fd, tmp); break; case F_DUP2FD: tmp = arg; error = kern_dup(td, FDDUP_FIXED, 0, fd, tmp); break; case F_DUP2FD_CLOEXEC: tmp = arg; error = kern_dup(td, FDDUP_FIXED, FDDUP_FLAG_CLOEXEC, fd, tmp); break; case F_GETFD: error = EBADF; FILEDESC_SLOCK(fdp); fde = fdeget_locked(fdp, fd); if (fde != NULL) { td->td_retval[0] = (fde->fde_flags & UF_EXCLOSE) ? FD_CLOEXEC : 0; error = 0; } FILEDESC_SUNLOCK(fdp); break; case F_SETFD: error = EBADF; FILEDESC_XLOCK(fdp); fde = fdeget_locked(fdp, fd); if (fde != NULL) { fde->fde_flags = (fde->fde_flags & ~UF_EXCLOSE) | (arg & FD_CLOEXEC ? UF_EXCLOSE : 0); error = 0; } FILEDESC_XUNLOCK(fdp); break; case F_GETFL: error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETFL, &fp); if (error != 0) break; td->td_retval[0] = OFLAGS(fp->f_flag); fdrop(fp, td); break; case F_SETFL: error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETFL, &fp); if (error != 0) break; do { tmp = flg = fp->f_flag; tmp &= ~FCNTLFLAGS; tmp |= FFLAGS(arg & ~O_ACCMODE) & FCNTLFLAGS; } while(atomic_cmpset_int(&fp->f_flag, flg, tmp) == 0); tmp = fp->f_flag & FNONBLOCK; error = fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); if (error != 0) { fdrop(fp, td); break; } tmp = fp->f_flag & FASYNC; error = fo_ioctl(fp, FIOASYNC, &tmp, td->td_ucred, td); if (error == 0) { fdrop(fp, td); break; } atomic_clear_int(&fp->f_flag, FNONBLOCK); tmp = 0; (void)fo_ioctl(fp, FIONBIO, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_GETOWN: error = fget_fcntl(td, fd, &cap_fcntl_rights, F_GETOWN, &fp); if (error != 0) break; error = fo_ioctl(fp, FIOGETOWN, &tmp, td->td_ucred, td); if (error == 0) td->td_retval[0] = tmp; fdrop(fp, td); break; case F_SETOWN: error = fget_fcntl(td, fd, &cap_fcntl_rights, F_SETOWN, &fp); if (error != 0) break; tmp = arg; error = fo_ioctl(fp, FIOSETOWN, &tmp, td->td_ucred, td); fdrop(fp, td); break; case F_SETLK_REMOTE: error = priv_check(td, PRIV_NFS_LOCKD); if (error != 0) return (error); flg = F_REMOTE; goto do_setlk; case F_SETLKW: flg |= F_WAIT; /* FALLTHROUGH F_SETLK */ case F_SETLK: do_setlk: flp = (struct flock *)arg; if ((flg & F_REMOTE) != 0 && flp->l_sysid == 0) { error = EINVAL; break; } error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if (foffset < 0 || (flp->l_start > 0 && foffset > OFF_MAX - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; switch (flp->l_type) { case F_RDLCK: if ((fp->f_flag & FREAD) == 0) { error = EBADF; break; } if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); } error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_WRLCK: if ((fp->f_flag & FWRITE) == 0) { error = EBADF; break; } if ((p->p_leader->p_flag & P_ADVLOCK) == 0) { PROC_LOCK(p->p_leader); p->p_leader->p_flag |= P_ADVLOCK; PROC_UNLOCK(p->p_leader); } error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_SETLK, flp, flg); break; case F_UNLCK: error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, flg); break; case F_UNLCKSYS: if (flg != F_REMOTE) { error = EINVAL; break; } error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCKSYS, flp, flg); break; default: error = EINVAL; break; } if (error != 0 || flp->l_type == F_UNLCK || flp->l_type == F_UNLCKSYS) { fdrop(fp, td); break; } /* * Check for a race with close. * * The vnode is now advisory locked (or unlocked, but this case * is not really important) as the caller requested. * We had to drop the filedesc lock, so we need to recheck if * the descriptor is still valid, because if it was closed * in the meantime we need to remove advisory lock from the * vnode - close on any descriptor leading to an advisory * locked vnode, removes that lock. * We will return 0 on purpose in that case, as the result of * successful advisory lock might have been externally visible * already. This is fine - effectively we pretend to the caller * that the closing thread was a bit slower and that the * advisory lock succeeded before the close. */ error = fget_unlocked(fdp, fd, &cap_no_rights, &fp2); if (error != 0) { fdrop(fp, td); break; } if (fp != fp2) { flp->l_whence = SEEK_SET; flp->l_start = 0; flp->l_len = 0; flp->l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, flp, F_POSIX); } fdrop(fp, td); fdrop(fp2, td); break; case F_GETLK: error = fget_unlocked(fdp, fd, &cap_flock_rights, &fp); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { error = EBADF; fdrop(fp, td); break; } flp = (struct flock *)arg; if (flp->l_type != F_RDLCK && flp->l_type != F_WRLCK && flp->l_type != F_UNLCK) { error = EINVAL; fdrop(fp, td); break; } if (flp->l_whence == SEEK_CUR) { foffset = foffset_get(fp); if ((flp->l_start > 0 && foffset > OFF_MAX - flp->l_start) || (flp->l_start < 0 && foffset < OFF_MIN - flp->l_start)) { error = EOVERFLOW; fdrop(fp, td); break; } flp->l_start += foffset; } vp = fp->f_vnode; error = VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_GETLK, flp, F_POSIX); fdrop(fp, td); break; case F_ADD_SEALS: error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); if (error != 0) break; error = fo_add_seals(fp, arg); fdrop(fp, td); break; case F_GET_SEALS: error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); if (error != 0) break; if (fo_get_seals(fp, &seals) == 0) td->td_retval[0] = seals; else error = EINVAL; fdrop(fp, td); break; case F_RDAHEAD: arg = arg ? 128 * 1024: 0; /* FALLTHROUGH */ case F_READAHEAD: error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); error = EBADF; break; } vp = fp->f_vnode; if (vp->v_type != VREG) { fdrop(fp, td); error = ENOTTY; break; } /* * Exclusive lock synchronizes against f_seqcount reads and * writes in sequential_heuristic(). */ error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) { fdrop(fp, td); break; } if (arg >= 0) { bsize = fp->f_vnode->v_mount->mnt_stat.f_iosize; arg = MIN(arg, INT_MAX - bsize + 1); fp->f_seqcount[UIO_READ] = MIN(IO_SEQMAX, (arg + bsize - 1) / bsize); atomic_set_int(&fp->f_flag, FRDAHEAD); } else { atomic_clear_int(&fp->f_flag, FRDAHEAD); } VOP_UNLOCK(vp); fdrop(fp, td); break; case F_ISUNIONSTACK: /* * Check if the vnode is part of a union stack (either the * "union" flag from mount(2) or unionfs). * * Prior to introduction of this op libc's readdir would call * fstatfs(2), in effect unnecessarily copying kilobytes of * data just to check fs name and a mount flag. * * Fixing the code to handle everything in the kernel instead * is a non-trivial endeavor and has low priority, thus this * horrible kludge facilitates the current behavior in a much * cheaper manner until someone(tm) sorts this out. */ error = fget_unlocked(fdp, fd, &cap_no_rights, &fp); if (error != 0) break; if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); error = EBADF; break; } vp = fp->f_vnode; /* * Since we don't prevent dooming the vnode even non-null mp * found can become immediately stale. This is tolerable since * mount points are type-stable (providing safe memory access) * and any vfs op on this vnode going forward will return an * error (meaning return value in this case is meaningless). */ mp = atomic_load_ptr(&vp->v_mount); if (__predict_false(mp == NULL)) { fdrop(fp, td); error = EBADF; break; } td->td_retval[0] = 0; if (mp->mnt_kern_flag & MNTK_UNIONFS || mp->mnt_flag & MNT_UNION) td->td_retval[0] = 1; fdrop(fp, td); break; default: error = EINVAL; break; } return (error); } static int getmaxfd(struct thread *td) { return (min((int)lim_cur(td, RLIMIT_NOFILE), maxfilesperproc)); } /* * Common code for dup, dup2, fcntl(F_DUPFD) and fcntl(F_DUP2FD). */ int kern_dup(struct thread *td, u_int mode, int flags, int old, int new) { struct filedesc *fdp; struct filedescent *oldfde, *newfde; struct proc *p; struct file *delfp; u_long *oioctls, *nioctls; int error, maxfd; p = td->td_proc; fdp = p->p_fd; oioctls = NULL; MPASS((flags & ~(FDDUP_FLAG_CLOEXEC)) == 0); MPASS(mode < FDDUP_LASTMODE); AUDIT_ARG_FD(old); /* XXXRW: if (flags & FDDUP_FIXED) AUDIT_ARG_FD2(new); */ /* * Verify we have a valid descriptor to dup from and possibly to * dup to. Unlike dup() and dup2(), fcntl()'s F_DUPFD should * return EINVAL when the new descriptor is out of bounds. */ if (old < 0) return (EBADF); if (new < 0) return (mode == FDDUP_FCNTL ? EINVAL : EBADF); maxfd = getmaxfd(td); if (new >= maxfd) return (mode == FDDUP_FCNTL ? EINVAL : EBADF); error = EBADF; FILEDESC_XLOCK(fdp); if (fget_locked(fdp, old) == NULL) goto unlock; if ((mode == FDDUP_FIXED || mode == FDDUP_MUSTREPLACE) && old == new) { td->td_retval[0] = new; if (flags & FDDUP_FLAG_CLOEXEC) fdp->fd_ofiles[new].fde_flags |= UF_EXCLOSE; error = 0; goto unlock; } oldfde = &fdp->fd_ofiles[old]; if (!fhold(oldfde->fde_file)) goto unlock; /* * If the caller specified a file descriptor, make sure the file * table is large enough to hold it, and grab it. Otherwise, just * allocate a new descriptor the usual way. */ switch (mode) { case FDDUP_NORMAL: case FDDUP_FCNTL: if ((error = fdalloc(td, new, &new)) != 0) { fdrop(oldfde->fde_file, td); goto unlock; } break; case FDDUP_MUSTREPLACE: /* Target file descriptor must exist. */ if (fget_locked(fdp, new) == NULL) { fdrop(oldfde->fde_file, td); goto unlock; } break; case FDDUP_FIXED: if (new >= fdp->fd_nfiles) { /* * The resource limits are here instead of e.g. * fdalloc(), because the file descriptor table may be * shared between processes, so we can't really use * racct_add()/racct_sub(). Instead of counting the * number of actually allocated descriptors, just put * the limit on the size of the file descriptor table. */ #ifdef RACCT if (RACCT_ENABLED()) { error = racct_set_unlocked(p, RACCT_NOFILE, new + 1); if (error != 0) { error = EMFILE; fdrop(oldfde->fde_file, td); goto unlock; } } #endif fdgrowtable_exp(fdp, new + 1); } if (!fdisused(fdp, new)) fdused(fdp, new); break; default: KASSERT(0, ("%s unsupported mode %d", __func__, mode)); } KASSERT(old != new, ("new fd is same as old")); newfde = &fdp->fd_ofiles[new]; delfp = newfde->fde_file; nioctls = filecaps_copy_prep(&oldfde->fde_caps); /* * Duplicate the source descriptor. */ #ifdef CAPABILITIES seqc_write_begin(&newfde->fde_seqc); #endif oioctls = filecaps_free_prep(&newfde->fde_caps); memcpy(newfde, oldfde, fde_change_size); filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, nioctls); if ((flags & FDDUP_FLAG_CLOEXEC) != 0) newfde->fde_flags = oldfde->fde_flags | UF_EXCLOSE; else newfde->fde_flags = oldfde->fde_flags & ~UF_EXCLOSE; #ifdef CAPABILITIES seqc_write_end(&newfde->fde_seqc); #endif td->td_retval[0] = new; error = 0; if (delfp != NULL) { (void) closefp(fdp, new, delfp, td, 1); FILEDESC_UNLOCK_ASSERT(fdp); } else { unlock: FILEDESC_XUNLOCK(fdp); } filecaps_free_finish(oioctls); return (error); } /* * If sigio is on the list associated with a process or process group, * disable signalling from the device, remove sigio from the list and * free sigio. */ void funsetown(struct sigio **sigiop) { struct sigio *sigio; if (*sigiop == NULL) return; SIGIO_LOCK(); sigio = *sigiop; if (sigio == NULL) { SIGIO_UNLOCK(); return; } *(sigio->sio_myref) = NULL; if ((sigio)->sio_pgid < 0) { struct pgrp *pg = (sigio)->sio_pgrp; PGRP_LOCK(pg); SLIST_REMOVE(&sigio->sio_pgrp->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else { struct proc *p = (sigio)->sio_proc; PROC_LOCK(p); SLIST_REMOVE(&sigio->sio_proc->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); } /* * Free a list of sigio structures. * We only need to lock the SIGIO_LOCK because we have made ourselves * inaccessible to callers of fsetown and therefore do not need to lock * the proc or pgrp struct for the list manipulation. */ void funsetownlst(struct sigiolst *sigiolst) { struct proc *p; struct pgrp *pg; struct sigio *sigio; sigio = SLIST_FIRST(sigiolst); if (sigio == NULL) return; p = NULL; pg = NULL; /* * Every entry of the list should belong * to a single proc or pgrp. */ if (sigio->sio_pgid < 0) { pg = sigio->sio_pgrp; PGRP_LOCK_ASSERT(pg, MA_NOTOWNED); } else /* if (sigio->sio_pgid > 0) */ { p = sigio->sio_proc; PROC_LOCK_ASSERT(p, MA_NOTOWNED); } SIGIO_LOCK(); while ((sigio = SLIST_FIRST(sigiolst)) != NULL) { *(sigio->sio_myref) = NULL; if (pg != NULL) { KASSERT(sigio->sio_pgid < 0, ("Proc sigio in pgrp sigio list")); KASSERT(sigio->sio_pgrp == pg, ("Bogus pgrp in sigio list")); PGRP_LOCK(pg); SLIST_REMOVE(&pg->pg_sigiolst, sigio, sigio, sio_pgsigio); PGRP_UNLOCK(pg); } else /* if (p != NULL) */ { KASSERT(sigio->sio_pgid > 0, ("Pgrp sigio in proc sigio list")); KASSERT(sigio->sio_proc == p, ("Bogus proc in sigio list")); PROC_LOCK(p); SLIST_REMOVE(&p->p_sigiolst, sigio, sigio, sio_pgsigio); PROC_UNLOCK(p); } SIGIO_UNLOCK(); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); SIGIO_LOCK(); } SIGIO_UNLOCK(); } /* * This is common code for FIOSETOWN ioctl called by fcntl(fd, F_SETOWN, arg). * * After permission checking, add a sigio structure to the sigio list for * the process or process group. */ int fsetown(pid_t pgid, struct sigio **sigiop) { struct proc *proc; struct pgrp *pgrp; struct sigio *sigio; int ret; if (pgid == 0) { funsetown(sigiop); return (0); } ret = 0; /* Allocate and fill in the new sigio out of locks. */ sigio = malloc(sizeof(struct sigio), M_SIGIO, M_WAITOK); sigio->sio_pgid = pgid; sigio->sio_ucred = crhold(curthread->td_ucred); sigio->sio_myref = sigiop; sx_slock(&proctree_lock); if (pgid > 0) { proc = pfind(pgid); if (proc == NULL) { ret = ESRCH; goto fail; } /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ PROC_UNLOCK(proc); if (proc->p_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } pgrp = NULL; } else /* if (pgid < 0) */ { pgrp = pgfind(-pgid); if (pgrp == NULL) { ret = ESRCH; goto fail; } PGRP_UNLOCK(pgrp); /* * Policy - Don't allow a process to FSETOWN a process * in another session. * * Remove this test to allow maximum flexibility or * restrict FSETOWN to the current process or process * group for maximum safety. */ if (pgrp->pg_session != curthread->td_proc->p_session) { ret = EPERM; goto fail; } proc = NULL; } funsetown(sigiop); if (pgid > 0) { PROC_LOCK(proc); /* * Since funsetownlst() is called without the proctree * locked, we need to check for P_WEXIT. * XXX: is ESRCH correct? */ if ((proc->p_flag & P_WEXIT) != 0) { PROC_UNLOCK(proc); ret = ESRCH; goto fail; } SLIST_INSERT_HEAD(&proc->p_sigiolst, sigio, sio_pgsigio); sigio->sio_proc = proc; PROC_UNLOCK(proc); } else { PGRP_LOCK(pgrp); SLIST_INSERT_HEAD(&pgrp->pg_sigiolst, sigio, sio_pgsigio); sigio->sio_pgrp = pgrp; PGRP_UNLOCK(pgrp); } sx_sunlock(&proctree_lock); SIGIO_LOCK(); *sigiop = sigio; SIGIO_UNLOCK(); return (0); fail: sx_sunlock(&proctree_lock); crfree(sigio->sio_ucred); free(sigio, M_SIGIO); return (ret); } /* * This is common code for FIOGETOWN ioctl called by fcntl(fd, F_GETOWN, arg). */ pid_t fgetown(struct sigio **sigiop) { pid_t pgid; SIGIO_LOCK(); pgid = (*sigiop != NULL) ? (*sigiop)->sio_pgid : 0; SIGIO_UNLOCK(); return (pgid); } /* * Function drops the filedesc lock on return. */ static int closefp(struct filedesc *fdp, int fd, struct file *fp, struct thread *td, int holdleaders) { int error; FILEDESC_XLOCK_ASSERT(fdp); if (holdleaders) { if (td->td_proc->p_fdtol != NULL) { /* * Ask fdfree() to sleep to ensure that all relevant * process leaders can be traversed in closef(). */ fdp->fd_holdleaderscount++; } else { holdleaders = 0; } } /* * We now hold the fp reference that used to be owned by the * descriptor array. We have to unlock the FILEDESC *AFTER* * knote_fdclose to prevent a race of the fd getting opened, a knote * added, and deleteing a knote for the new fd. */ if (__predict_false(!TAILQ_EMPTY(&fdp->fd_kqlist))) knote_fdclose(td, fd); /* * We need to notify mqueue if the object is of type mqueue. */ if (__predict_false(fp->f_type == DTYPE_MQUEUE)) mq_fdclose(td, fd, fp); FILEDESC_XUNLOCK(fdp); error = closef(fp, td); if (holdleaders) { FILEDESC_XLOCK(fdp); fdp->fd_holdleaderscount--; if (fdp->fd_holdleaderscount == 0 && fdp->fd_holdleaderswakeup != 0) { fdp->fd_holdleaderswakeup = 0; wakeup(&fdp->fd_holdleaderscount); } FILEDESC_XUNLOCK(fdp); } return (error); } /* * Close a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct close_args { int fd; }; #endif /* ARGSUSED */ int sys_close(struct thread *td, struct close_args *uap) { return (kern_close(td, uap->fd)); } int kern_close(struct thread *td, int fd) { struct filedesc *fdp; struct file *fp; fdp = td->td_proc->p_fd; AUDIT_SYSCLOSE(td, fd); FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, fd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } fdfree(fdp, fd); /* closefp() drops the FILEDESC lock for us. */ return (closefp(fdp, fd, fp, td, 1)); } int kern_close_range(struct thread *td, u_int lowfd, u_int highfd) { struct filedesc *fdp; int fd, ret, lastfile; ret = 0; fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); /* * Check this prior to clamping; closefrom(3) with only fd 0, 1, and 2 * open should not be a usage error. From a close_range() perspective, * close_range(3, ~0U, 0) in the same scenario should also likely not * be a usage error as all fd above 3 are in-fact already closed. */ if (highfd < lowfd) { ret = EINVAL; goto out; } /* * If lastfile == -1, we're dealing with either a fresh file * table or one in which every fd has been closed. Just return * successful; there's nothing left to do. */ lastfile = fdlastfile(fdp); if (lastfile == -1) goto out; /* Clamped to [lowfd, lastfile] */ highfd = MIN(highfd, lastfile); for (fd = lowfd; fd <= highfd; fd++) { if (fdp->fd_ofiles[fd].fde_file != NULL) { FILEDESC_SUNLOCK(fdp); (void)kern_close(td, fd); FILEDESC_SLOCK(fdp); } } out: FILEDESC_SUNLOCK(fdp); return (ret); } #ifndef _SYS_SYSPROTO_H_ struct close_range_args { u_int lowfd; u_int highfd; int flags; }; #endif int sys_close_range(struct thread *td, struct close_range_args *uap) { /* No flags currently defined */ if (uap->flags != 0) return (EINVAL); return (kern_close_range(td, uap->lowfd, uap->highfd)); } #ifdef COMPAT_FREEBSD12 /* * Close open file descriptors. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd12_closefrom_args { int lowfd; }; #endif /* ARGSUSED */ int freebsd12_closefrom(struct thread *td, struct freebsd12_closefrom_args *uap) { u_int lowfd; AUDIT_ARG_FD(uap->lowfd); /* * Treat negative starting file descriptor values identical to * closefrom(0) which closes all files. */ lowfd = MAX(0, uap->lowfd); return (kern_close_range(td, lowfd, ~0U)); } #endif /* COMPAT_FREEBSD12 */ #if defined(COMPAT_43) /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct ofstat_args { int fd; struct ostat *sb; }; #endif /* ARGSUSED */ int ofstat(struct thread *td, struct ofstat_args *uap) { struct ostat oub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { cvtstat(&ub, &oub); error = copyout(&oub, uap->sb, sizeof(oub)); } return (error); } #endif /* COMPAT_43 */ #if defined(COMPAT_FREEBSD11) int freebsd11_fstat(struct thread *td, struct freebsd11_fstat_args *uap) { struct stat sb; struct freebsd11_stat osb; int error; error = kern_fstat(td, uap->fd, &sb); if (error != 0) return (error); error = freebsd11_cvtstat(&sb, &osb); if (error == 0) error = copyout(&osb, uap->sb, sizeof(osb)); return (error); } #endif /* COMPAT_FREEBSD11 */ /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fstat_args { int fd; struct stat *sb; }; #endif /* ARGSUSED */ int sys_fstat(struct thread *td, struct fstat_args *uap) { struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) error = copyout(&ub, uap->sb, sizeof(ub)); return (error); } int kern_fstat(struct thread *td, int fd, struct stat *sbp) { struct file *fp; int error; AUDIT_ARG_FD(fd); error = fget(td, fd, &cap_fstat_rights, &fp); if (__predict_false(error != 0)) return (error); AUDIT_ARG_FILE(td->td_proc, fp); error = fo_stat(fp, sbp, td->td_ucred, td); fdrop(fp, td); #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) /* * Return status information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct freebsd11_nfstat_args { int fd; struct nstat *sb; }; #endif /* ARGSUSED */ int freebsd11_nfstat(struct thread *td, struct freebsd11_nfstat_args *uap) { struct nstat nub; struct stat ub; int error; error = kern_fstat(td, uap->fd, &ub); if (error == 0) { freebsd11_cvtnstat(&ub, &nub); error = copyout(&nub, uap->sb, sizeof(nub)); } return (error); } #endif /* COMPAT_FREEBSD11 */ /* * Return pathconf information about a file descriptor. */ #ifndef _SYS_SYSPROTO_H_ struct fpathconf_args { int fd; int name; }; #endif /* ARGSUSED */ int sys_fpathconf(struct thread *td, struct fpathconf_args *uap) { long value; int error; error = kern_fpathconf(td, uap->fd, uap->name, &value); if (error == 0) td->td_retval[0] = value; return (error); } int kern_fpathconf(struct thread *td, int fd, int name, long *valuep) { struct file *fp; struct vnode *vp; int error; error = fget(td, fd, &cap_fpathconf_rights, &fp); if (error != 0) return (error); if (name == _PC_ASYNC_IO) { *valuep = _POSIX_ASYNCHRONOUS_IO; goto out; } vp = fp->f_vnode; if (vp != NULL) { vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_PATHCONF(vp, name, valuep); VOP_UNLOCK(vp); } else if (fp->f_type == DTYPE_PIPE || fp->f_type == DTYPE_SOCKET) { if (name != _PC_PIPE_BUF) { error = EINVAL; } else { *valuep = PIPE_BUF; error = 0; } } else { error = EOPNOTSUPP; } out: fdrop(fp, td); return (error); } /* * Copy filecaps structure allocating memory for ioctls array if needed. * * The last parameter indicates whether the fdtable is locked. If it is not and * ioctls are encountered, copying fails and the caller must lock the table. * * Note that if the table was not locked, the caller has to check the relevant * sequence counter to determine whether the operation was successful. */ bool filecaps_copy(const struct filecaps *src, struct filecaps *dst, bool locked) { size_t size; if (src->fc_ioctls != NULL && !locked) return (false); memcpy(dst, src, sizeof(*src)); if (src->fc_ioctls == NULL) return (true); KASSERT(src->fc_nioctls > 0, ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; dst->fc_ioctls = malloc(size, M_FILECAPS, M_WAITOK); memcpy(dst->fc_ioctls, src->fc_ioctls, size); return (true); } static u_long * filecaps_copy_prep(const struct filecaps *src) { u_long *ioctls; size_t size; if (__predict_true(src->fc_ioctls == NULL)) return (NULL); KASSERT(src->fc_nioctls > 0, ("fc_ioctls != NULL, but fc_nioctls=%hd", src->fc_nioctls)); size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; ioctls = malloc(size, M_FILECAPS, M_WAITOK); return (ioctls); } static void filecaps_copy_finish(const struct filecaps *src, struct filecaps *dst, u_long *ioctls) { size_t size; *dst = *src; if (__predict_true(src->fc_ioctls == NULL)) { MPASS(ioctls == NULL); return; } size = sizeof(src->fc_ioctls[0]) * src->fc_nioctls; dst->fc_ioctls = ioctls; bcopy(src->fc_ioctls, dst->fc_ioctls, size); } /* * Move filecaps structure to the new place and clear the old place. */ void filecaps_move(struct filecaps *src, struct filecaps *dst) { *dst = *src; bzero(src, sizeof(*src)); } /* * Fill the given filecaps structure with full rights. */ static void filecaps_fill(struct filecaps *fcaps) { CAP_ALL(&fcaps->fc_rights); fcaps->fc_ioctls = NULL; fcaps->fc_nioctls = -1; fcaps->fc_fcntls = CAP_FCNTL_ALL; } /* * Free memory allocated within filecaps structure. */ void filecaps_free(struct filecaps *fcaps) { free(fcaps->fc_ioctls, M_FILECAPS); bzero(fcaps, sizeof(*fcaps)); } static u_long * filecaps_free_prep(struct filecaps *fcaps) { u_long *ioctls; ioctls = fcaps->fc_ioctls; bzero(fcaps, sizeof(*fcaps)); return (ioctls); } static void filecaps_free_finish(u_long *ioctls) { free(ioctls, M_FILECAPS); } /* * Validate the given filecaps structure. */ static void filecaps_validate(const struct filecaps *fcaps, const char *func) { KASSERT(cap_rights_is_valid(&fcaps->fc_rights), ("%s: invalid rights", func)); KASSERT((fcaps->fc_fcntls & ~CAP_FCNTL_ALL) == 0, ("%s: invalid fcntls", func)); KASSERT(fcaps->fc_fcntls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_FCNTL), ("%s: fcntls without CAP_FCNTL", func)); KASSERT(fcaps->fc_ioctls != NULL ? fcaps->fc_nioctls > 0 : (fcaps->fc_nioctls == -1 || fcaps->fc_nioctls == 0), ("%s: invalid ioctls", func)); KASSERT(fcaps->fc_nioctls == 0 || cap_rights_is_set(&fcaps->fc_rights, CAP_IOCTL), ("%s: ioctls without CAP_IOCTL", func)); } static void fdgrowtable_exp(struct filedesc *fdp, int nfd) { int nfd1; FILEDESC_XLOCK_ASSERT(fdp); nfd1 = fdp->fd_nfiles * 2; if (nfd1 < nfd) nfd1 = nfd; fdgrowtable(fdp, nfd1); } /* * Grow the file table to accommodate (at least) nfd descriptors. */ static void fdgrowtable(struct filedesc *fdp, int nfd) { struct filedesc0 *fdp0; struct freetable *ft; struct fdescenttbl *ntable; struct fdescenttbl *otable; int nnfiles, onfiles; NDSLOTTYPE *nmap, *omap; KASSERT(fdp->fd_nfiles > 0, ("zero-length file table")); /* save old values */ onfiles = fdp->fd_nfiles; otable = fdp->fd_files; omap = fdp->fd_map; /* compute the size of the new table */ nnfiles = NDSLOTS(nfd) * NDENTRIES; /* round up */ if (nnfiles <= onfiles) /* the table is already large enough */ return; /* * Allocate a new table. We need enough space for the number of * entries, file entries themselves and the struct freetable we will use * when we decommission the table and place it on the freelist. * We place the struct freetable in the middle so we don't have * to worry about padding. */ ntable = malloc(offsetof(struct fdescenttbl, fdt_ofiles) + nnfiles * sizeof(ntable->fdt_ofiles[0]) + sizeof(struct freetable), M_FILEDESC, M_ZERO | M_WAITOK); /* copy the old data */ ntable->fdt_nfiles = nnfiles; memcpy(ntable->fdt_ofiles, otable->fdt_ofiles, onfiles * sizeof(ntable->fdt_ofiles[0])); /* * Allocate a new map only if the old is not large enough. It will * grow at a slower rate than the table as it can map more * entries than the table can hold. */ if (NDSLOTS(nnfiles) > NDSLOTS(onfiles)) { nmap = malloc(NDSLOTS(nnfiles) * NDSLOTSIZE, M_FILEDESC, M_ZERO | M_WAITOK); /* copy over the old data and update the pointer */ memcpy(nmap, omap, NDSLOTS(onfiles) * sizeof(*omap)); fdp->fd_map = nmap; } /* * Make sure that ntable is correctly initialized before we replace * fd_files poiner. Otherwise fget_unlocked() may see inconsistent * data. */ atomic_store_rel_ptr((volatile void *)&fdp->fd_files, (uintptr_t)ntable); /* * Do not free the old file table, as some threads may still * reference entries within it. Instead, place it on a freelist * which will be processed when the struct filedesc is released. * * Note that if onfiles == NDFILE, we're dealing with the original * static allocation contained within (struct filedesc0 *)fdp, * which must not be freed. */ if (onfiles > NDFILE) { ft = (struct freetable *)&otable->fdt_ofiles[onfiles]; fdp0 = (struct filedesc0 *)fdp; ft->ft_table = otable; SLIST_INSERT_HEAD(&fdp0->fd_free, ft, ft_next); } /* * The map does not have the same possibility of threads still * holding references to it. So always free it as long as it * does not reference the original static allocation. */ if (NDSLOTS(onfiles) > NDSLOTS(NDFILE)) free(omap, M_FILEDESC); } /* * Allocate a file descriptor for the process. */ int fdalloc(struct thread *td, int minfd, int *result) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int fd, maxfd, allocfd; #ifdef RACCT int error; #endif FILEDESC_XLOCK_ASSERT(fdp); if (fdp->fd_freefile > minfd) minfd = fdp->fd_freefile; maxfd = getmaxfd(td); /* * Search the bitmap for a free descriptor starting at minfd. * If none is found, grow the file table. */ fd = fd_first_free(fdp, minfd, fdp->fd_nfiles); if (__predict_false(fd >= maxfd)) return (EMFILE); if (__predict_false(fd >= fdp->fd_nfiles)) { allocfd = min(fd * 2, maxfd); #ifdef RACCT if (RACCT_ENABLED()) { error = racct_set_unlocked(p, RACCT_NOFILE, allocfd); if (error != 0) return (EMFILE); } #endif /* * fd is already equal to first free descriptor >= minfd, so * we only need to grow the table and we are done. */ fdgrowtable_exp(fdp, allocfd); } /* * Perform some sanity checks, then mark the file descriptor as * used and return it to the caller. */ KASSERT(fd >= 0 && fd < min(maxfd, fdp->fd_nfiles), ("invalid descriptor %d", fd)); KASSERT(!fdisused(fdp, fd), ("fd_first_free() returned non-free descriptor")); KASSERT(fdp->fd_ofiles[fd].fde_file == NULL, ("file descriptor isn't free")); fdused(fdp, fd); *result = fd; return (0); } /* * Allocate n file descriptors for the process. */ int fdallocn(struct thread *td, int minfd, int *fds, int n) { struct proc *p = td->td_proc; struct filedesc *fdp = p->p_fd; int i; FILEDESC_XLOCK_ASSERT(fdp); for (i = 0; i < n; i++) if (fdalloc(td, 0, &fds[i]) != 0) break; if (i < n) { for (i--; i >= 0; i--) fdunused(fdp, fds[i]); return (EMFILE); } return (0); } /* * Create a new open file structure and allocate a file descriptor for the * process that refers to it. We add one reference to the file for the * descriptor table and one reference for resultfp. This is to prevent us * being preempted and the entry in the descriptor table closed after we * release the FILEDESC lock. */ int falloc_caps(struct thread *td, struct file **resultfp, int *resultfd, int flags, struct filecaps *fcaps) { struct file *fp; int error, fd; error = falloc_noinstall(td, &fp); if (error) return (error); /* no reference held on error */ error = finstall(td, fp, &fd, flags, fcaps); if (error) { fdrop(fp, td); /* one reference (fp only) */ return (error); } if (resultfp != NULL) *resultfp = fp; /* copy out result */ else fdrop(fp, td); /* release local reference */ if (resultfd != NULL) *resultfd = fd; return (0); } /* * Create a new open file structure without allocating a file descriptor. */ int falloc_noinstall(struct thread *td, struct file **resultfp) { struct file *fp; int maxuserfiles = maxfiles - (maxfiles / 20); int openfiles_new; static struct timeval lastfail; static int curfail; KASSERT(resultfp != NULL, ("%s: resultfp == NULL", __func__)); openfiles_new = atomic_fetchadd_int(&openfiles, 1) + 1; if ((openfiles_new >= maxuserfiles && priv_check(td, PRIV_MAXFILES) != 0) || openfiles_new >= maxfiles) { atomic_subtract_int(&openfiles, 1); if (ppsratecheck(&lastfail, &curfail, 1)) { printf("kern.maxfiles limit exceeded by uid %i, (%s) " "please see tuning(7).\n", td->td_ucred->cr_ruid, td->td_proc->p_comm); } return (ENFILE); } fp = uma_zalloc(file_zone, M_WAITOK); bzero(fp, sizeof(*fp)); refcount_init(&fp->f_count, 1); fp->f_cred = crhold(td->td_ucred); fp->f_ops = &badfileops; *resultfp = fp; return (0); } /* * Install a file in a file descriptor table. */ void _finstall(struct filedesc *fdp, struct file *fp, int fd, int flags, struct filecaps *fcaps) { struct filedescent *fde; MPASS(fp != NULL); if (fcaps != NULL) filecaps_validate(fcaps, __func__); FILEDESC_XLOCK_ASSERT(fdp); fde = &fdp->fd_ofiles[fd]; #ifdef CAPABILITIES seqc_write_begin(&fde->fde_seqc); #endif fde->fde_file = fp; fde->fde_flags = (flags & O_CLOEXEC) != 0 ? UF_EXCLOSE : 0; if (fcaps != NULL) filecaps_move(fcaps, &fde->fde_caps); else filecaps_fill(&fde->fde_caps); #ifdef CAPABILITIES seqc_write_end(&fde->fde_seqc); #endif } int finstall(struct thread *td, struct file *fp, int *fd, int flags, struct filecaps *fcaps) { struct filedesc *fdp = td->td_proc->p_fd; int error; MPASS(fd != NULL); if (!fhold(fp)) return (EBADF); FILEDESC_XLOCK(fdp); error = fdalloc(td, 0, fd); if (__predict_false(error != 0)) { FILEDESC_XUNLOCK(fdp); fdrop(fp, td); return (error); } _finstall(fdp, fp, *fd, flags, fcaps); FILEDESC_XUNLOCK(fdp); return (0); } /* * Build a new filedesc structure from another. * Copy the current, root, and jail root vnode references. * * If fdp is not NULL, return with it shared locked. */ struct filedesc * fdinit(struct filedesc *fdp, bool prepfiles, int *lastfile) { struct filedesc0 *newfdp0; struct filedesc *newfdp; struct pwd *newpwd; if (prepfiles) MPASS(lastfile != NULL); else MPASS(lastfile == NULL); newfdp0 = uma_zalloc(filedesc0_zone, M_WAITOK | M_ZERO); newfdp = &newfdp0->fd_fd; /* Create the file descriptor table. */ FILEDESC_LOCK_INIT(newfdp); refcount_init(&newfdp->fd_refcnt, 1); refcount_init(&newfdp->fd_holdcnt, 1); newfdp->fd_cmask = CMASK; newfdp->fd_map = newfdp0->fd_dmap; newfdp->fd_files = (struct fdescenttbl *)&newfdp0->fd_dfiles; newfdp->fd_files->fdt_nfiles = NDFILE; if (fdp == NULL) { newpwd = pwd_alloc(); smr_serialized_store(&newfdp->fd_pwd, newpwd, true); return (newfdp); } FILEDESC_SLOCK(fdp); newpwd = pwd_hold_filedesc(fdp); smr_serialized_store(&newfdp->fd_pwd, newpwd, true); if (!prepfiles) { FILEDESC_SUNLOCK(fdp); return (newfdp); } for (;;) { *lastfile = fdlastfile(fdp); if (*lastfile < newfdp->fd_nfiles) break; FILEDESC_SUNLOCK(fdp); fdgrowtable(newfdp, *lastfile + 1); FILEDESC_SLOCK(fdp); } return (newfdp); } static struct filedesc * fdhold(struct proc *p) { struct filedesc *fdp; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = p->p_fd; if (fdp != NULL) refcount_acquire(&fdp->fd_holdcnt); return (fdp); } static void fddrop(struct filedesc *fdp) { if (fdp->fd_holdcnt > 1) { if (refcount_release(&fdp->fd_holdcnt) == 0) return; } FILEDESC_LOCK_DESTROY(fdp); uma_zfree(filedesc0_zone, fdp); } /* * Share a filedesc structure. */ struct filedesc * fdshare(struct filedesc *fdp) { refcount_acquire(&fdp->fd_refcnt); return (fdp); } /* * Unshare a filedesc structure, if necessary by making a copy */ void fdunshare(struct thread *td) { struct filedesc *tmp; struct proc *p = td->td_proc; if (p->p_fd->fd_refcnt == 1) return; tmp = fdcopy(p->p_fd); fdescfree(td); p->p_fd = tmp; } void fdinstall_remapped(struct thread *td, struct filedesc *fdp) { fdescfree(td); td->td_proc->p_fd = fdp; } /* * Copy a filedesc structure. A NULL pointer in returns a NULL reference, * this is to ease callers, not catch errors. */ struct filedesc * fdcopy(struct filedesc *fdp) { struct filedesc *newfdp; struct filedescent *nfde, *ofde; int i, lastfile; MPASS(fdp != NULL); newfdp = fdinit(fdp, true, &lastfile); /* copy all passable descriptors (i.e. not kqueue) */ newfdp->fd_freefile = -1; for (i = 0; i <= lastfile; ++i) { ofde = &fdp->fd_ofiles[i]; if (ofde->fde_file == NULL || (ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0 || !fhold(ofde->fde_file)) { if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; continue; } nfde = &newfdp->fd_ofiles[i]; *nfde = *ofde; filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); fdused_init(newfdp, i); } if (newfdp->fd_freefile == -1) newfdp->fd_freefile = i; newfdp->fd_cmask = fdp->fd_cmask; FILEDESC_SUNLOCK(fdp); return (newfdp); } /* * Copies a filedesc structure, while remapping all file descriptors * stored inside using a translation table. * * File descriptors are copied over to the new file descriptor table, * regardless of whether the close-on-exec flag is set. */ int fdcopy_remapped(struct filedesc *fdp, const int *fds, size_t nfds, struct filedesc **ret) { struct filedesc *newfdp; struct filedescent *nfde, *ofde; int error, i, lastfile; MPASS(fdp != NULL); newfdp = fdinit(fdp, true, &lastfile); if (nfds > lastfile + 1) { /* New table cannot be larger than the old one. */ error = E2BIG; goto bad; } /* Copy all passable descriptors (i.e. not kqueue). */ newfdp->fd_freefile = nfds; for (i = 0; i < nfds; ++i) { if (fds[i] < 0 || fds[i] > lastfile) { /* File descriptor out of bounds. */ error = EBADF; goto bad; } ofde = &fdp->fd_ofiles[fds[i]]; if (ofde->fde_file == NULL) { /* Unused file descriptor. */ error = EBADF; goto bad; } if ((ofde->fde_file->f_ops->fo_flags & DFLAG_PASSABLE) == 0) { /* File descriptor cannot be passed. */ error = EINVAL; goto bad; } if (!fhold(ofde->fde_file)) { error = EBADF; goto bad; } nfde = &newfdp->fd_ofiles[i]; *nfde = *ofde; filecaps_copy(&ofde->fde_caps, &nfde->fde_caps, true); fdused_init(newfdp, i); } newfdp->fd_cmask = fdp->fd_cmask; FILEDESC_SUNLOCK(fdp); *ret = newfdp; return (0); bad: FILEDESC_SUNLOCK(fdp); fdescfree_remapped(newfdp); return (error); } /* * Clear POSIX style locks. This is only used when fdp looses a reference (i.e. * one of processes using it exits) and the table used to be shared. */ static void fdclearlocks(struct thread *td) { struct filedesc *fdp; struct filedesc_to_leader *fdtol; struct flock lf; struct file *fp; struct proc *p; struct vnode *vp; int i, lastfile; p = td->td_proc; fdp = p->p_fd; fdtol = p->p_fdtol; MPASS(fdtol != NULL); FILEDESC_XLOCK(fdp); KASSERT(fdtol->fdl_refcount > 0, ("filedesc_to_refcount botch: fdl_refcount=%d", fdtol->fdl_refcount)); if (fdtol->fdl_refcount == 1 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { lastfile = fdlastfile(fdp); for (i = 0; i <= lastfile; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp == NULL || fp->f_type != DTYPE_VNODE || !fhold(fp)) continue; FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)p->p_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdrop(fp, td); } } retry: if (fdtol->fdl_refcount == 1) { if (fdp->fd_holdleaderscount > 0 && (p->p_leader->p_flag & P_ADVLOCK) != 0) { /* * close() or kern_dup() has cleared a reference * in a shared file descriptor table. */ fdp->fd_holdleaderswakeup = 1; sx_sleep(&fdp->fd_holdleaderscount, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } if (fdtol->fdl_holdcount > 0) { /* * Ensure that fdtol->fdl_leader remains * valid in closef(). */ fdtol->fdl_wakeup = 1; sx_sleep(fdtol, FILEDESC_LOCK(fdp), PLOCK, "fdlhold", 0); goto retry; } } fdtol->fdl_refcount--; if (fdtol->fdl_refcount == 0 && fdtol->fdl_holdcount == 0) { fdtol->fdl_next->fdl_prev = fdtol->fdl_prev; fdtol->fdl_prev->fdl_next = fdtol->fdl_next; } else fdtol = NULL; p->p_fdtol = NULL; FILEDESC_XUNLOCK(fdp); if (fdtol != NULL) free(fdtol, M_FILEDESC_TO_LEADER); } /* * Release a filedesc structure. */ static void fdescfree_fds(struct thread *td, struct filedesc *fdp, bool needclose) { struct filedesc0 *fdp0; struct freetable *ft, *tft; struct filedescent *fde; struct file *fp; int i, lastfile; lastfile = fdlastfile_single(fdp); for (i = 0; i <= lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL) { fdefree_last(fde); if (needclose) (void) closef(fp, td); else fdrop(fp, td); } } if (NDSLOTS(fdp->fd_nfiles) > NDSLOTS(NDFILE)) free(fdp->fd_map, M_FILEDESC); if (fdp->fd_nfiles > NDFILE) free(fdp->fd_files, M_FILEDESC); fdp0 = (struct filedesc0 *)fdp; SLIST_FOREACH_SAFE(ft, &fdp0->fd_free, ft_next, tft) free(ft->ft_table, M_FILEDESC); fddrop(fdp); } void fdescfree(struct thread *td) { struct proc *p; struct filedesc *fdp; struct pwd *pwd; p = td->td_proc; fdp = p->p_fd; MPASS(fdp != NULL); #ifdef RACCT if (RACCT_ENABLED()) racct_set_unlocked(p, RACCT_NOFILE, 0); #endif if (p->p_fdtol != NULL) fdclearlocks(td); PROC_LOCK(p); p->p_fd = NULL; PROC_UNLOCK(p); if (refcount_release(&fdp->fd_refcnt) == 0) return; FILEDESC_XLOCK(fdp); pwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); pwd_set(fdp, NULL); FILEDESC_XUNLOCK(fdp); pwd_drop(pwd); fdescfree_fds(td, fdp, 1); } void fdescfree_remapped(struct filedesc *fdp) { pwd_drop(smr_serialized_load(&fdp->fd_pwd, true)); fdescfree_fds(curthread, fdp, 0); } /* * For setugid programs, we don't want to people to use that setugidness * to generate error messages which write to a file which otherwise would * otherwise be off-limits to the process. We check for filesystems where * the vnode can change out from under us after execve (like [lin]procfs). * * Since fdsetugidsafety calls this only for fd 0, 1 and 2, this check is * sufficient. We also don't check for setugidness since we know we are. */ static bool is_unsafe(struct file *fp) { struct vnode *vp; if (fp->f_type != DTYPE_VNODE) return (false); vp = fp->f_vnode; return ((vp->v_vflag & VV_PROCDEP) != 0); } /* * Make this setguid thing safe, if at all possible. */ void fdsetugidsafety(struct thread *td) { struct filedesc *fdp; struct file *fp; int i; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); for (i = 0; i <= 2; i++) { fp = fdp->fd_ofiles[i].fde_file; if (fp != NULL && is_unsafe(fp)) { FILEDESC_XLOCK(fdp); knote_fdclose(td, i); /* * NULL-out descriptor prior to close to avoid * a race while close blocks. */ fdfree(fdp, i); FILEDESC_XUNLOCK(fdp); (void) closef(fp, td); } } } /* * If a specific file object occupies a specific file descriptor, close the * file descriptor entry and drop a reference on the file object. This is a * convenience function to handle a subsequent error in a function that calls * falloc() that handles the race that another thread might have closed the * file descriptor out from under the thread creating the file object. */ void fdclose(struct thread *td, struct file *fp, int idx) { struct filedesc *fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); if (fdp->fd_ofiles[idx].fde_file == fp) { fdfree(fdp, idx); FILEDESC_XUNLOCK(fdp); fdrop(fp, td); } else FILEDESC_XUNLOCK(fdp); } /* * Close any files on exec? */ void fdcloseexec(struct thread *td) { struct filedesc *fdp; struct filedescent *fde; struct file *fp; int i, lastfile; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); lastfile = fdlastfile_single(fdp); for (i = 0; i <= lastfile; i++) { fde = &fdp->fd_ofiles[i]; fp = fde->fde_file; if (fp != NULL && (fp->f_type == DTYPE_MQUEUE || (fde->fde_flags & UF_EXCLOSE))) { FILEDESC_XLOCK(fdp); fdfree(fdp, i); (void) closefp(fdp, i, fp, td, 0); FILEDESC_UNLOCK_ASSERT(fdp); } } } /* * It is unsafe for set[ug]id processes to be started with file * descriptors 0..2 closed, as these descriptors are given implicit * significance in the Standard C library. fdcheckstd() will create a * descriptor referencing /dev/null for each of stdin, stdout, and * stderr that is not already open. */ int fdcheckstd(struct thread *td) { struct filedesc *fdp; register_t save; int i, error, devnull; fdp = td->td_proc->p_fd; KASSERT(fdp->fd_refcnt == 1, ("the fdtable should not be shared")); MPASS(fdp->fd_nfiles >= 3); devnull = -1; for (i = 0; i <= 2; i++) { if (fdp->fd_ofiles[i].fde_file != NULL) continue; save = td->td_retval[0]; if (devnull != -1) { error = kern_dup(td, FDDUP_FIXED, 0, devnull, i); } else { error = kern_openat(td, AT_FDCWD, "/dev/null", UIO_SYSSPACE, O_RDWR, 0); if (error == 0) { devnull = td->td_retval[0]; KASSERT(devnull == i, ("we didn't get our fd")); } } td->td_retval[0] = save; if (error != 0) return (error); } return (0); } /* * Internal form of close. Decrement reference count on file structure. * Note: td may be NULL when closing a file that was being passed in a * message. */ int closef(struct file *fp, struct thread *td) { struct vnode *vp; struct flock lf; struct filedesc_to_leader *fdtol; struct filedesc *fdp; /* * POSIX record locking dictates that any close releases ALL * locks owned by this process. This is handled by setting * a flag in the unlock to free ONLY locks obeying POSIX * semantics, and not to free BSD-style file locks. * If the descriptor was in a message, POSIX-style locks * aren't passed with the descriptor, and the thread pointer * will be NULL. Callers should be careful only to pass a * NULL thread pointer when there really is no owning * context that might have locks, or the locks will be * leaked. */ if (fp->f_type == DTYPE_VNODE && td != NULL) { vp = fp->f_vnode; if ((td->td_proc->p_leader->p_flag & P_ADVLOCK) != 0) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, (caddr_t)td->td_proc->p_leader, F_UNLCK, &lf, F_POSIX); } fdtol = td->td_proc->p_fdtol; if (fdtol != NULL) { /* * Handle special case where file descriptor table is * shared between multiple process leaders. */ fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); for (fdtol = fdtol->fdl_next; fdtol != td->td_proc->p_fdtol; fdtol = fdtol->fdl_next) { if ((fdtol->fdl_leader->p_flag & P_ADVLOCK) == 0) continue; fdtol->fdl_holdcount++; FILEDESC_XUNLOCK(fdp); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; vp = fp->f_vnode; (void) VOP_ADVLOCK(vp, (caddr_t)fdtol->fdl_leader, F_UNLCK, &lf, F_POSIX); FILEDESC_XLOCK(fdp); fdtol->fdl_holdcount--; if (fdtol->fdl_holdcount == 0 && fdtol->fdl_wakeup != 0) { fdtol->fdl_wakeup = 0; wakeup(fdtol); } } FILEDESC_XUNLOCK(fdp); } } return (fdrop(fp, td)); } /* * Initialize the file pointer with the specified properties. * * The ops are set with release semantics to be certain that the flags, type, * and data are visible when ops is. This is to prevent ops methods from being * called with bad data. */ void finit(struct file *fp, u_int flag, short type, void *data, struct fileops *ops) { fp->f_data = data; fp->f_flag = flag; fp->f_type = type; atomic_store_rel_ptr((volatile uintptr_t *)&fp->f_ops, (uintptr_t)ops); } void finit_vnode(struct file *fp, u_int flag, void *data, struct fileops *ops) { fp->f_seqcount[UIO_READ] = 1; fp->f_seqcount[UIO_WRITE] = 1; finit(fp, (flag & FMASK) | (fp->f_flag & FHASLOCK), DTYPE_VNODE, data, ops); } int fget_cap_locked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp, struct filecaps *havecapsp) { struct filedescent *fde; int error; FILEDESC_LOCK_ASSERT(fdp); fde = fdeget_locked(fdp, fd); if (fde == NULL) { error = EBADF; goto out; } #ifdef CAPABILITIES error = cap_check(cap_rights_fde_inline(fde), needrightsp); if (error != 0) goto out; #endif if (havecapsp != NULL) filecaps_copy(&fde->fde_caps, havecapsp, true); *fpp = fde->fde_file; error = 0; out: return (error); } int fget_cap(struct thread *td, int fd, cap_rights_t *needrightsp, struct file **fpp, struct filecaps *havecapsp) { struct filedesc *fdp = td->td_proc->p_fd; int error; #ifndef CAPABILITIES error = fget_unlocked(fdp, fd, needrightsp, fpp); if (havecapsp != NULL && error == 0) filecaps_fill(havecapsp); #else struct file *fp; seqc_t seq; *fpp = NULL; for (;;) { error = fget_unlocked_seq(fdp, fd, needrightsp, &fp, &seq); if (error != 0) return (error); if (havecapsp != NULL) { if (!filecaps_copy(&fdp->fd_ofiles[fd].fde_caps, havecapsp, false)) { fdrop(fp, td); goto get_locked; } } if (!fd_modified(fdp, fd, seq)) break; fdrop(fp, td); } *fpp = fp; return (0); get_locked: FILEDESC_SLOCK(fdp); error = fget_cap_locked(fdp, fd, needrightsp, fpp, havecapsp); if (error == 0 && !fhold(*fpp)) error = EBADF; FILEDESC_SUNLOCK(fdp); #endif return (error); } +#ifdef CAPABILITIES +int +fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) +{ + const struct filedescent *fde; + const struct fdescenttbl *fdt; + struct filedesc *fdp; + struct file *fp; + struct vnode *vp; + const cap_rights_t *haverights; + cap_rights_t rights; + seqc_t seq; + + VFS_SMR_ASSERT_ENTERED(); + + rights = *ndp->ni_rightsneeded; + cap_rights_set_one(&rights, CAP_LOOKUP); + + fdp = curproc->p_fd; + fdt = fdp->fd_files; + if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) + return (EBADF); + seq = seqc_read_any(fd_seqc(fdt, fd)); + if (__predict_false(seqc_in_modify(seq))) + return (EAGAIN); + fde = &fdt->fdt_ofiles[fd]; + haverights = cap_rights_fde_inline(fde); + fp = fde->fde_file; + if (__predict_false(fp == NULL)) + return (EAGAIN); + if (__predict_false(cap_check_inline_transient(haverights, &rights))) + return (EAGAIN); + *fsearch = ((fp->f_flag & FSEARCH) != 0); + vp = fp->f_vnode; + if (__predict_false(vp == NULL || vp->v_type != VDIR)) { + return (EAGAIN); + } + if (!filecaps_copy(&fde->fde_caps, &ndp->ni_filecaps, false)) { + return (EAGAIN); + } + /* + * Use an acquire barrier to force re-reading of fdt so it is + * refreshed for verification. + */ + atomic_thread_fence_acq(); + fdt = fdp->fd_files; + if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq))) + return (EAGAIN); + /* + * If file descriptor doesn't have all rights, + * all lookups relative to it must also be + * strictly relative. + * + * Not yet supported by fast path. + */ + CAP_ALL(&rights); + if (!cap_rights_contains(&ndp->ni_filecaps.fc_rights, &rights) || + ndp->ni_filecaps.fc_fcntls != CAP_FCNTL_ALL || + ndp->ni_filecaps.fc_nioctls != -1) { +#ifdef notyet + ndp->ni_lcf |= NI_LCF_STRICTRELATIVE; +#else + return (EAGAIN); +#endif + } + *vpp = vp; + return (0); +} +#else +int +fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch) +{ + const struct fdescenttbl *fdt; + struct filedesc *fdp; + struct file *fp; + struct vnode *vp; + + VFS_SMR_ASSERT_ENTERED(); + + fdp = curproc->p_fd; + fdt = fdp->fd_files; + if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) + return (EBADF); + fp = fdt->fdt_ofiles[fd].fde_file; + if (__predict_false(fp == NULL)) + return (EAGAIN); + *fsearch = ((fp->f_flag & FSEARCH) != 0); + vp = fp->f_vnode; + if (__predict_false(vp == NULL || vp->v_type != VDIR)) { + return (EAGAIN); + } + /* + * Use an acquire barrier to force re-reading of fdt so it is + * refreshed for verification. + */ + atomic_thread_fence_acq(); + fdt = fdp->fd_files; + if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) + return (EAGAIN); + filecaps_fill(&ndp->ni_filecaps); + *vpp = vp; + return (0); +} +#endif + int fget_unlocked_seq(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp, seqc_t *seqp) { #ifdef CAPABILITIES const struct filedescent *fde; #endif const struct fdescenttbl *fdt; struct file *fp; #ifdef CAPABILITIES seqc_t seq; cap_rights_t haverights; int error; #endif fdt = fdp->fd_files; if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) return (EBADF); /* * Fetch the descriptor locklessly. We avoid fdrop() races by * never raising a refcount above 0. To accomplish this we have * to use a cmpset loop rather than an atomic_add. The descriptor * must be re-verified once we acquire a reference to be certain * that the identity is still correct and we did not lose a race * due to preemption. */ for (;;) { #ifdef CAPABILITIES seq = seqc_read(fd_seqc(fdt, fd)); fde = &fdt->fdt_ofiles[fd]; haverights = *cap_rights_fde_inline(fde); fp = fde->fde_file; if (!seqc_consistent(fd_seqc(fdt, fd), seq)) continue; #else fp = fdt->fdt_ofiles[fd].fde_file; #endif if (fp == NULL) return (EBADF); #ifdef CAPABILITIES error = cap_check_inline(&haverights, needrightsp); if (error != 0) return (error); #endif if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) { /* * The count was found either saturated or zero. * This re-read is not any more racy than using the * return value from fcmpset. */ if (fp->f_count != 0) return (EBADF); /* * Force a reload. Other thread could reallocate the * table before this fd was closed, so it is possible * that there is a stale fp pointer in cached version. */ fdt = atomic_load_ptr(&fdp->fd_files); continue; } /* * Use an acquire barrier to force re-reading of fdt so it is * refreshed for verification. */ atomic_thread_fence_acq(); fdt = fdp->fd_files; #ifdef CAPABILITIES if (seqc_consistent_nomb(fd_seqc(fdt, fd), seq)) #else if (fp == fdt->fdt_ofiles[fd].fde_file) #endif break; fdrop(fp, curthread); } *fpp = fp; if (seqp != NULL) { #ifdef CAPABILITIES *seqp = seq; #endif } return (0); } /* * See the comments in fget_unlocked_seq for an explanation of how this works. * * This is a simplified variant which bails out to the aforementioned routine * if anything goes wrong. In practice this only happens when userspace is * racing with itself. */ int fget_unlocked(struct filedesc *fdp, int fd, cap_rights_t *needrightsp, struct file **fpp) { #ifdef CAPABILITIES const struct filedescent *fde; #endif const struct fdescenttbl *fdt; struct file *fp; #ifdef CAPABILITIES seqc_t seq; const cap_rights_t *haverights; #endif fdt = fdp->fd_files; if (__predict_false((u_int)fd >= fdt->fdt_nfiles)) return (EBADF); #ifdef CAPABILITIES seq = seqc_read_any(fd_seqc(fdt, fd)); if (__predict_false(seqc_in_modify(seq))) goto out_fallback; fde = &fdt->fdt_ofiles[fd]; haverights = cap_rights_fde_inline(fde); fp = fde->fde_file; #else fp = fdt->fdt_ofiles[fd].fde_file; #endif if (__predict_false(fp == NULL)) goto out_fallback; #ifdef CAPABILITIES if (__predict_false(cap_check_inline_transient(haverights, needrightsp))) goto out_fallback; #endif if (__predict_false(!refcount_acquire_if_not_zero(&fp->f_count))) goto out_fallback; /* * Use an acquire barrier to force re-reading of fdt so it is * refreshed for verification. */ atomic_thread_fence_acq(); fdt = fdp->fd_files; #ifdef CAPABILITIES if (__predict_false(!seqc_consistent_nomb(fd_seqc(fdt, fd), seq))) #else if (__predict_false(fp != fdt->fdt_ofiles[fd].fde_file)) #endif goto out_fdrop; *fpp = fp; return (0); out_fdrop: fdrop(fp, curthread); out_fallback: return (fget_unlocked_seq(fdp, fd, needrightsp, fpp, NULL)); } /* * Extract the file pointer associated with the specified descriptor for the * current user process. * * If the descriptor doesn't exist or doesn't match 'flags', EBADF is * returned. * * File's rights will be checked against the capability rights mask. * * If an error occurred the non-zero error is returned and *fpp is set to * NULL. Otherwise *fpp is held and set and zero is returned. Caller is * responsible for fdrop(). */ static __inline int _fget(struct thread *td, int fd, struct file **fpp, int flags, cap_rights_t *needrightsp) { struct filedesc *fdp; struct file *fp; int error; *fpp = NULL; fdp = td->td_proc->p_fd; error = fget_unlocked(fdp, fd, needrightsp, &fp); if (__predict_false(error != 0)) return (error); if (__predict_false(fp->f_ops == &badfileops)) { fdrop(fp, td); return (EBADF); } /* * FREAD and FWRITE failure return EBADF as per POSIX. */ error = 0; switch (flags) { case FREAD: case FWRITE: if ((fp->f_flag & flags) == 0) error = EBADF; break; case FEXEC: if ((fp->f_flag & (FREAD | FEXEC)) == 0 || ((fp->f_flag & FWRITE) != 0)) error = EBADF; break; case 0: break; default: KASSERT(0, ("wrong flags")); } if (error != 0) { fdrop(fp, td); return (error); } *fpp = fp; return (0); } int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, 0, rightsp)); } int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, struct file **fpp) { int error; #ifndef CAPABILITIES error = _fget(td, fd, fpp, 0, rightsp); if (maxprotp != NULL) *maxprotp = VM_PROT_ALL; return (error); #else cap_rights_t fdrights; struct filedesc *fdp; struct file *fp; seqc_t seq; *fpp = NULL; fdp = td->td_proc->p_fd; MPASS(cap_rights_is_set(rightsp, CAP_MMAP)); for (;;) { error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); if (__predict_false(error != 0)) return (error); if (__predict_false(fp->f_ops == &badfileops)) { fdrop(fp, td); return (EBADF); } if (maxprotp != NULL) fdrights = *cap_rights(fdp, fd); if (!fd_modified(fdp, fd, seq)) break; fdrop(fp, td); } /* * If requested, convert capability rights to access flags. */ if (maxprotp != NULL) *maxprotp = cap_rights_to_vmprot(&fdrights); *fpp = fp; return (0); #endif } int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FREAD, rightsp)); } int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp) { return (_fget(td, fd, fpp, FWRITE, rightsp)); } int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, struct file **fpp) { struct filedesc *fdp = td->td_proc->p_fd; #ifndef CAPABILITIES return (fget_unlocked(fdp, fd, rightsp, fpp)); #else struct file *fp; int error; seqc_t seq; *fpp = NULL; MPASS(cap_rights_is_set(rightsp, CAP_FCNTL)); for (;;) { error = fget_unlocked_seq(fdp, fd, rightsp, &fp, &seq); if (error != 0) return (error); error = cap_fcntl_check(fdp, fd, needfcntl); if (!fd_modified(fdp, fd, seq)) break; fdrop(fp, td); } if (error != 0) { fdrop(fp, td); return (error); } *fpp = fp; return (0); #endif } /* * Like fget() but loads the underlying vnode, or returns an error if the * descriptor does not represent a vnode. Note that pipes use vnodes but * never have VM objects. The returned vnode will be vref()'d. * * XXX: what about the unused flags ? */ static __inline int _fgetvp(struct thread *td, int fd, int flags, cap_rights_t *needrightsp, struct vnode **vpp) { struct file *fp; int error; *vpp = NULL; error = _fget(td, fd, &fp, flags, needrightsp); if (error != 0) return (error); if (fp->f_vnode == NULL) { error = EINVAL; } else { *vpp = fp->f_vnode; vrefact(*vpp); } fdrop(fp, td); return (error); } int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, 0, rightsp, vpp)); } int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp) { struct filecaps caps; struct file *fp; int error; error = fget_cap(td, fd, needrightsp, &fp, &caps); if (error != 0) return (error); if (fp->f_ops == &badfileops) { error = EBADF; goto out; } if (fp->f_vnode == NULL) { error = EINVAL; goto out; } *havecaps = caps; *vpp = fp->f_vnode; vrefact(*vpp); fdrop(fp, td); return (0); out: filecaps_free(&caps); fdrop(fp, td); return (error); } int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FREAD, rightsp, vpp)); } int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FEXEC, rightsp, vpp)); } #ifdef notyet int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp) { return (_fgetvp(td, fd, FWRITE, rightsp, vpp)); } #endif /* * Handle the last reference to a file being closed. * * Without the noinline attribute clang keeps inlining the func thorough this * file when fdrop is used. */ int __noinline _fdrop(struct file *fp, struct thread *td) { int error; if (fp->f_count != 0) panic("fdrop: count %d", fp->f_count); error = fo_close(fp, td); atomic_subtract_int(&openfiles, 1); crfree(fp->f_cred); free(fp->f_advice, M_FADVISE); uma_zfree(file_zone, fp); return (error); } /* * Apply an advisory lock on a file descriptor. * * Just attempt to get a record lock of the requested type on the entire file * (l_whence = SEEK_SET, l_start = 0, l_len = 0). */ #ifndef _SYS_SYSPROTO_H_ struct flock_args { int fd; int how; }; #endif /* ARGSUSED */ int sys_flock(struct thread *td, struct flock_args *uap) { struct file *fp; struct vnode *vp; struct flock lf; int error; error = fget(td, uap->fd, &cap_flock_rights, &fp); if (error != 0) return (error); if (fp->f_type != DTYPE_VNODE) { fdrop(fp, td); return (EOPNOTSUPP); } vp = fp->f_vnode; lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (uap->how & LOCK_UN) { lf.l_type = F_UNLCK; atomic_clear_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_UNLCK, &lf, F_FLOCK); goto done2; } if (uap->how & LOCK_EX) lf.l_type = F_WRLCK; else if (uap->how & LOCK_SH) lf.l_type = F_RDLCK; else { error = EBADF; goto done2; } atomic_set_int(&fp->f_flag, FHASLOCK); error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, (uap->how & LOCK_NB) ? F_FLOCK : F_FLOCK | F_WAIT); done2: fdrop(fp, td); return (error); } /* * Duplicate the specified descriptor to a free descriptor. */ int dupfdopen(struct thread *td, struct filedesc *fdp, int dfd, int mode, int openerror, int *indxp) { struct filedescent *newfde, *oldfde; struct file *fp; u_long *ioctls; int error, indx; KASSERT(openerror == ENODEV || openerror == ENXIO, ("unexpected error %d in %s", openerror, __func__)); /* * If the to-be-dup'd fd number is greater than the allowed number * of file descriptors, or the fd to be dup'd has already been * closed, then reject. */ FILEDESC_XLOCK(fdp); if ((fp = fget_locked(fdp, dfd)) == NULL) { FILEDESC_XUNLOCK(fdp); return (EBADF); } error = fdalloc(td, 0, &indx); if (error != 0) { FILEDESC_XUNLOCK(fdp); return (error); } /* * There are two cases of interest here. * * For ENODEV simply dup (dfd) to file descriptor (indx) and return. * * For ENXIO steal away the file structure from (dfd) and store it in * (indx). (dfd) is effectively closed by this operation. */ switch (openerror) { case ENODEV: /* * Check that the mode the file is being opened for is a * subset of the mode of the existing descriptor. */ if (((mode & (FREAD|FWRITE)) | fp->f_flag) != fp->f_flag) { fdunused(fdp, indx); FILEDESC_XUNLOCK(fdp); return (EACCES); } if (!fhold(fp)) { fdunused(fdp, indx); FILEDESC_XUNLOCK(fdp); return (EBADF); } newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; ioctls = filecaps_copy_prep(&oldfde->fde_caps); #ifdef CAPABILITIES seqc_write_begin(&newfde->fde_seqc); #endif memcpy(newfde, oldfde, fde_change_size); filecaps_copy_finish(&oldfde->fde_caps, &newfde->fde_caps, ioctls); #ifdef CAPABILITIES seqc_write_end(&newfde->fde_seqc); #endif break; case ENXIO: /* * Steal away the file pointer from dfd and stuff it into indx. */ newfde = &fdp->fd_ofiles[indx]; oldfde = &fdp->fd_ofiles[dfd]; #ifdef CAPABILITIES seqc_write_begin(&newfde->fde_seqc); #endif memcpy(newfde, oldfde, fde_change_size); oldfde->fde_file = NULL; fdunused(fdp, dfd); #ifdef CAPABILITIES seqc_write_end(&newfde->fde_seqc); #endif break; } FILEDESC_XUNLOCK(fdp); *indxp = indx; return (0); } /* * This sysctl determines if we will allow a process to chroot(2) if it * has a directory open: * 0: disallowed for all processes. * 1: allowed for processes that were not already chroot(2)'ed. * 2: allowed for all processes. */ static int chroot_allow_open_directories = 1; SYSCTL_INT(_kern, OID_AUTO, chroot_allow_open_directories, CTLFLAG_RW, &chroot_allow_open_directories, 0, "Allow a process to chroot(2) if it has a directory open"); /* * Helper function for raised chroot(2) security function: Refuse if * any filedescriptors are open directories. */ static int chroot_refuse_vdir_fds(struct filedesc *fdp) { struct vnode *vp; struct file *fp; int fd, lastfile; FILEDESC_LOCK_ASSERT(fdp); lastfile = fdlastfile(fdp); for (fd = 0; fd <= lastfile; fd++) { fp = fget_locked(fdp, fd); if (fp == NULL) continue; if (fp->f_type == DTYPE_VNODE) { vp = fp->f_vnode; if (vp->v_type == VDIR) return (EPERM); } } return (0); } static void pwd_fill(struct pwd *oldpwd, struct pwd *newpwd) { if (newpwd->pwd_cdir == NULL && oldpwd->pwd_cdir != NULL) { vrefact(oldpwd->pwd_cdir); newpwd->pwd_cdir = oldpwd->pwd_cdir; } if (newpwd->pwd_rdir == NULL && oldpwd->pwd_rdir != NULL) { vrefact(oldpwd->pwd_rdir); newpwd->pwd_rdir = oldpwd->pwd_rdir; } if (newpwd->pwd_jdir == NULL && oldpwd->pwd_jdir != NULL) { vrefact(oldpwd->pwd_jdir); newpwd->pwd_jdir = oldpwd->pwd_jdir; } } struct pwd * pwd_hold_filedesc(struct filedesc *fdp) { struct pwd *pwd; FILEDESC_LOCK_ASSERT(fdp); pwd = FILEDESC_LOCKED_LOAD_PWD(fdp); if (pwd != NULL) refcount_acquire(&pwd->pwd_refcount); return (pwd); } bool pwd_hold_smr(struct pwd *pwd) { MPASS(pwd != NULL); if (__predict_true(refcount_acquire_if_not_zero(&pwd->pwd_refcount))) { return (true); } return (false); } struct pwd * pwd_hold(struct thread *td) { struct filedesc *fdp; struct pwd *pwd; fdp = td->td_proc->p_fd; vfs_smr_enter(); pwd = vfs_smr_entered_load(&fdp->fd_pwd); if (pwd_hold_smr(pwd)) { vfs_smr_exit(); return (pwd); } vfs_smr_exit(); FILEDESC_SLOCK(fdp); pwd = pwd_hold_filedesc(fdp); MPASS(pwd != NULL); FILEDESC_SUNLOCK(fdp); return (pwd); } struct pwd * pwd_get_smr(void) { struct pwd *pwd; pwd = vfs_smr_entered_load(&curproc->p_fd->fd_pwd); MPASS(pwd != NULL); return (pwd); } static struct pwd * pwd_alloc(void) { struct pwd *pwd; pwd = uma_zalloc_smr(pwd_zone, M_WAITOK); bzero(pwd, sizeof(*pwd)); refcount_init(&pwd->pwd_refcount, 1); return (pwd); } void pwd_drop(struct pwd *pwd) { if (!refcount_release(&pwd->pwd_refcount)) return; if (pwd->pwd_cdir != NULL) vrele(pwd->pwd_cdir); if (pwd->pwd_rdir != NULL) vrele(pwd->pwd_rdir); if (pwd->pwd_jdir != NULL) vrele(pwd->pwd_jdir); uma_zfree_smr(pwd_zone, pwd); } /* * Common routine for kern_chroot() and jail_attach(). The caller is * responsible for invoking priv_check() and mac_vnode_check_chroot() to * authorize this operation. */ int pwd_chroot(struct thread *td, struct vnode *vp) { struct filedesc *fdp; struct pwd *newpwd, *oldpwd; int error; fdp = td->td_proc->p_fd; newpwd = pwd_alloc(); FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); if (chroot_allow_open_directories == 0 || (chroot_allow_open_directories == 1 && oldpwd->pwd_rdir != rootvnode)) { error = chroot_refuse_vdir_fds(fdp); if (error != 0) { FILEDESC_XUNLOCK(fdp); pwd_drop(newpwd); return (error); } } vrefact(vp); newpwd->pwd_rdir = vp; if (oldpwd->pwd_jdir == NULL) { vrefact(vp); newpwd->pwd_jdir = vp; } pwd_fill(oldpwd, newpwd); pwd_set(fdp, newpwd); FILEDESC_XUNLOCK(fdp); pwd_drop(oldpwd); return (0); } void pwd_chdir(struct thread *td, struct vnode *vp) { struct filedesc *fdp; struct pwd *newpwd, *oldpwd; VNPASS(vp->v_usecount > 0, vp); newpwd = pwd_alloc(); fdp = td->td_proc->p_fd; FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); newpwd->pwd_cdir = vp; pwd_fill(oldpwd, newpwd); pwd_set(fdp, newpwd); FILEDESC_XUNLOCK(fdp); pwd_drop(oldpwd); } void pwd_ensure_dirs(void) { struct filedesc *fdp; struct pwd *oldpwd, *newpwd; fdp = curproc->p_fd; FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); if (oldpwd->pwd_cdir != NULL && oldpwd->pwd_rdir != NULL) { FILEDESC_XUNLOCK(fdp); return; } FILEDESC_XUNLOCK(fdp); newpwd = pwd_alloc(); FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); pwd_fill(oldpwd, newpwd); if (newpwd->pwd_cdir == NULL) { vrefact(rootvnode); newpwd->pwd_cdir = rootvnode; } if (newpwd->pwd_rdir == NULL) { vrefact(rootvnode); newpwd->pwd_rdir = rootvnode; } pwd_set(fdp, newpwd); FILEDESC_XUNLOCK(fdp); pwd_drop(oldpwd); } void pwd_set_rootvnode(void) { struct filedesc *fdp; struct pwd *oldpwd, *newpwd; fdp = curproc->p_fd; newpwd = pwd_alloc(); FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); vrefact(rootvnode); newpwd->pwd_cdir = rootvnode; vrefact(rootvnode); newpwd->pwd_rdir = rootvnode; pwd_fill(oldpwd, newpwd); pwd_set(fdp, newpwd); FILEDESC_XUNLOCK(fdp); pwd_drop(oldpwd); } /* * Scan all active processes and prisons to see if any of them have a current * or root directory of `olddp'. If so, replace them with the new mount point. */ void mountcheckdirs(struct vnode *olddp, struct vnode *newdp) { struct filedesc *fdp; struct pwd *newpwd, *oldpwd; struct prison *pr; struct proc *p; int nrele; if (vrefcnt(olddp) == 1) return; nrele = 0; newpwd = pwd_alloc(); sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_XLOCK(fdp); oldpwd = FILEDESC_XLOCKED_LOAD_PWD(fdp); if (oldpwd == NULL || (oldpwd->pwd_cdir != olddp && oldpwd->pwd_rdir != olddp && oldpwd->pwd_jdir != olddp)) { FILEDESC_XUNLOCK(fdp); fddrop(fdp); continue; } if (oldpwd->pwd_cdir == olddp) { vrefact(newdp); newpwd->pwd_cdir = newdp; } if (oldpwd->pwd_rdir == olddp) { vrefact(newdp); newpwd->pwd_rdir = newdp; } if (oldpwd->pwd_jdir == olddp) { vrefact(newdp); newpwd->pwd_jdir = newdp; } pwd_fill(oldpwd, newpwd); pwd_set(fdp, newpwd); FILEDESC_XUNLOCK(fdp); pwd_drop(oldpwd); fddrop(fdp); newpwd = pwd_alloc(); } sx_sunlock(&allproc_lock); pwd_drop(newpwd); if (rootvnode == olddp) { vrefact(newdp); rootvnode = newdp; nrele++; } mtx_lock(&prison0.pr_mtx); if (prison0.pr_root == olddp) { vrefact(newdp); prison0.pr_root = newdp; nrele++; } mtx_unlock(&prison0.pr_mtx); sx_slock(&allprison_lock); TAILQ_FOREACH(pr, &allprison, pr_list) { mtx_lock(&pr->pr_mtx); if (pr->pr_root == olddp) { vrefact(newdp); pr->pr_root = newdp; nrele++; } mtx_unlock(&pr->pr_mtx); } sx_sunlock(&allprison_lock); while (nrele--) vrele(olddp); } struct filedesc_to_leader * filedesc_to_leader_alloc(struct filedesc_to_leader *old, struct filedesc *fdp, struct proc *leader) { struct filedesc_to_leader *fdtol; fdtol = malloc(sizeof(struct filedesc_to_leader), M_FILEDESC_TO_LEADER, M_WAITOK); fdtol->fdl_refcount = 1; fdtol->fdl_holdcount = 0; fdtol->fdl_wakeup = 0; fdtol->fdl_leader = leader; if (old != NULL) { FILEDESC_XLOCK(fdp); fdtol->fdl_next = old->fdl_next; fdtol->fdl_prev = old; old->fdl_next = fdtol; fdtol->fdl_next->fdl_prev = fdtol; FILEDESC_XUNLOCK(fdp); } else { fdtol->fdl_next = fdtol; fdtol->fdl_prev = fdtol; } return (fdtol); } static int sysctl_kern_proc_nfds(SYSCTL_HANDLER_ARGS) { NDSLOTTYPE *map; struct filedesc *fdp; int count, off, minoff; if (*(int *)arg1 != 0) return (EINVAL); fdp = curproc->p_fd; count = 0; FILEDESC_SLOCK(fdp); map = fdp->fd_map; off = NDSLOT(fdp->fd_nfiles - 1); for (minoff = NDSLOT(0); off >= minoff; --off) count += bitcountl(map[off]); FILEDESC_SUNLOCK(fdp); return (SYSCTL_OUT(req, &count, sizeof(count))); } static SYSCTL_NODE(_kern_proc, KERN_PROC_NFDS, nfds, CTLFLAG_RD|CTLFLAG_CAPRD|CTLFLAG_MPSAFE, sysctl_kern_proc_nfds, "Number of open file descriptors"); /* * Get file structures globally. */ static int sysctl_kern_file(SYSCTL_HANDLER_ARGS) { struct xfile xf; struct filedesc *fdp; struct file *fp; struct proc *p; int error, n, lastfile; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); if (req->oldptr == NULL) { n = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; /* overestimates sparse tables. */ n += fdp->fd_nfiles; fddrop(fdp); } sx_sunlock(&allproc_lock); return (SYSCTL_OUT(req, 0, n * sizeof(xf))); } error = 0; bzero(&xf, sizeof(xf)); xf.xf_size = sizeof(xf); sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { PROC_LOCK(p); if (p->p_state == PRS_NEW) { PROC_UNLOCK(p); continue; } if (p_cansee(req->td, p) != 0) { PROC_UNLOCK(p); continue; } xf.xf_pid = p->p_pid; xf.xf_uid = p->p_ucred->cr_uid; fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) continue; FILEDESC_SLOCK(fdp); lastfile = fdlastfile(fdp); for (n = 0; fdp->fd_refcnt > 0 && n <= lastfile; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; xf.xf_fd = n; xf.xf_file = (uintptr_t)fp; xf.xf_data = (uintptr_t)fp->f_data; xf.xf_vnode = (uintptr_t)fp->f_vnode; xf.xf_type = (uintptr_t)fp->f_type; xf.xf_count = fp->f_count; xf.xf_msgcount = 0; xf.xf_offset = foffset_get(fp); xf.xf_flag = fp->f_flag; error = SYSCTL_OUT(req, &xf, sizeof(xf)); if (error) break; } FILEDESC_SUNLOCK(fdp); fddrop(fdp); if (error) break; } sx_sunlock(&allproc_lock); return (error); } SYSCTL_PROC(_kern, KERN_FILE, file, CTLTYPE_OPAQUE|CTLFLAG_RD|CTLFLAG_MPSAFE, 0, 0, sysctl_kern_file, "S,xfile", "Entire file table"); #ifdef KINFO_FILE_SIZE CTASSERT(sizeof(struct kinfo_file) == KINFO_FILE_SIZE); #endif static int xlate_fflags(int fflags) { static const struct { int fflag; int kf_fflag; } fflags_table[] = { { FAPPEND, KF_FLAG_APPEND }, { FASYNC, KF_FLAG_ASYNC }, { FFSYNC, KF_FLAG_FSYNC }, { FHASLOCK, KF_FLAG_HASLOCK }, { FNONBLOCK, KF_FLAG_NONBLOCK }, { FREAD, KF_FLAG_READ }, { FWRITE, KF_FLAG_WRITE }, { O_CREAT, KF_FLAG_CREAT }, { O_DIRECT, KF_FLAG_DIRECT }, { O_EXCL, KF_FLAG_EXCL }, { O_EXEC, KF_FLAG_EXEC }, { O_EXLOCK, KF_FLAG_EXLOCK }, { O_NOFOLLOW, KF_FLAG_NOFOLLOW }, { O_SHLOCK, KF_FLAG_SHLOCK }, { O_TRUNC, KF_FLAG_TRUNC } }; unsigned int i; int kflags; kflags = 0; for (i = 0; i < nitems(fflags_table); i++) if (fflags & fflags_table[i].fflag) kflags |= fflags_table[i].kf_fflag; return (kflags); } /* Trim unused data from kf_path by truncating the structure size. */ void pack_kinfo(struct kinfo_file *kif) { kif->kf_structsize = offsetof(struct kinfo_file, kf_path) + strlen(kif->kf_path) + 1; kif->kf_structsize = roundup(kif->kf_structsize, sizeof(uint64_t)); } static void export_file_to_kinfo(struct file *fp, int fd, cap_rights_t *rightsp, struct kinfo_file *kif, struct filedesc *fdp, int flags) { int error; bzero(kif, sizeof(*kif)); /* Set a default type to allow for empty fill_kinfo() methods. */ kif->kf_type = KF_TYPE_UNKNOWN; kif->kf_flags = xlate_fflags(fp->f_flag); if (rightsp != NULL) kif->kf_cap_rights = *rightsp; else cap_rights_init_zero(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = fp->f_count; kif->kf_offset = foffset_get(fp); /* * This may drop the filedesc lock, so the 'fp' cannot be * accessed after this call. */ error = fo_fill_kinfo(fp, kif, fdp); if (error == 0) kif->kf_status |= KF_ATTR_VALID; if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) pack_kinfo(kif); else kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); } static void export_vnode_to_kinfo(struct vnode *vp, int fd, int fflags, struct kinfo_file *kif, int flags) { int error; bzero(kif, sizeof(*kif)); kif->kf_type = KF_TYPE_VNODE; error = vn_fill_kinfo_vnode(vp, kif); if (error == 0) kif->kf_status |= KF_ATTR_VALID; kif->kf_flags = xlate_fflags(fflags); cap_rights_init_zero(&kif->kf_cap_rights); kif->kf_fd = fd; kif->kf_ref_count = -1; kif->kf_offset = -1; if ((flags & KERN_FILEDESC_PACK_KINFO) != 0) pack_kinfo(kif); else kif->kf_structsize = roundup2(sizeof(*kif), sizeof(uint64_t)); vrele(vp); } struct export_fd_buf { struct filedesc *fdp; struct sbuf *sb; ssize_t remainder; struct kinfo_file kif; int flags; }; static int export_kinfo_to_sb(struct export_fd_buf *efbuf) { struct kinfo_file *kif; kif = &efbuf->kif; if (efbuf->remainder != -1) { if (efbuf->remainder < kif->kf_structsize) { /* Terminate export. */ efbuf->remainder = 0; return (0); } efbuf->remainder -= kif->kf_structsize; } return (sbuf_bcat(efbuf->sb, kif, kif->kf_structsize) == 0 ? 0 : ENOMEM); } static int export_file_to_sb(struct file *fp, int fd, cap_rights_t *rightsp, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); export_file_to_kinfo(fp, fd, rightsp, &efbuf->kif, efbuf->fdp, efbuf->flags); FILEDESC_SUNLOCK(efbuf->fdp); error = export_kinfo_to_sb(efbuf); FILEDESC_SLOCK(efbuf->fdp); return (error); } static int export_vnode_to_sb(struct vnode *vp, int fd, int fflags, struct export_fd_buf *efbuf) { int error; if (efbuf->remainder == 0) return (0); if (efbuf->fdp != NULL) FILEDESC_SUNLOCK(efbuf->fdp); export_vnode_to_kinfo(vp, fd, fflags, &efbuf->kif, efbuf->flags); error = export_kinfo_to_sb(efbuf); if (efbuf->fdp != NULL) FILEDESC_SLOCK(efbuf->fdp); return (error); } /* * Store a process file descriptor information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_filedesc_out(struct proc *p, struct sbuf *sb, ssize_t maxlen, int flags) { struct file *fp; struct filedesc *fdp; struct export_fd_buf *efbuf; struct vnode *cttyvp, *textvp, *tracevp; struct pwd *pwd; int error, i, lastfile; cap_rights_t rights; PROC_LOCK_ASSERT(p, MA_OWNED); /* ktrace vnode */ tracevp = p->p_tracevp; if (tracevp != NULL) vrefact(tracevp); /* text vnode */ textvp = p->p_textvp; if (textvp != NULL) vrefact(textvp); /* Controlling tty. */ cttyvp = NULL; if (p->p_pgrp != NULL && p->p_pgrp->pg_session != NULL) { cttyvp = p->p_pgrp->pg_session->s_ttyvp; if (cttyvp != NULL) vrefact(cttyvp); } fdp = fdhold(p); PROC_UNLOCK(p); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = NULL; efbuf->sb = sb; efbuf->remainder = maxlen; efbuf->flags = flags; if (tracevp != NULL) export_vnode_to_sb(tracevp, KF_FD_TYPE_TRACE, FREAD | FWRITE, efbuf); if (textvp != NULL) export_vnode_to_sb(textvp, KF_FD_TYPE_TEXT, FREAD, efbuf); if (cttyvp != NULL) export_vnode_to_sb(cttyvp, KF_FD_TYPE_CTTY, FREAD | FWRITE, efbuf); error = 0; if (fdp == NULL) goto fail; efbuf->fdp = fdp; FILEDESC_SLOCK(fdp); pwd = pwd_hold_filedesc(fdp); if (pwd != NULL) { /* working directory */ if (pwd->pwd_cdir != NULL) { vrefact(pwd->pwd_cdir); export_vnode_to_sb(pwd->pwd_cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } /* root directory */ if (pwd->pwd_rdir != NULL) { vrefact(pwd->pwd_rdir); export_vnode_to_sb(pwd->pwd_rdir, KF_FD_TYPE_ROOT, FREAD, efbuf); } /* jail directory */ if (pwd->pwd_jdir != NULL) { vrefact(pwd->pwd_jdir); export_vnode_to_sb(pwd->pwd_jdir, KF_FD_TYPE_JAIL, FREAD, efbuf); } } lastfile = fdlastfile(fdp); for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; #ifdef CAPABILITIES rights = *cap_rights(fdp, i); #else /* !CAPABILITIES */ rights = cap_no_rights; #endif /* * Create sysctl entry. It is OK to drop the filedesc * lock inside of export_file_to_sb() as we will * re-validate and re-evaluate its properties when the * loop continues. */ error = export_file_to_sb(fp, i, &rights, efbuf); if (error != 0 || efbuf->remainder == 0) break; } FILEDESC_SUNLOCK(fdp); if (pwd != NULL) pwd_drop(pwd); fddrop(fdp); fail: free(efbuf, M_TEMP); return (error); } #define FILEDESC_SBUF_SIZE (sizeof(struct kinfo_file) * 5) /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_filedesc(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, FILEDESC_SBUF_SIZE, req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_filedesc_out(p, &sb, maxlen, KERN_FILEDESC_PACK_KINFO); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } #ifdef COMPAT_FREEBSD7 #ifdef KINFO_OFILE_SIZE CTASSERT(sizeof(struct kinfo_ofile) == KINFO_OFILE_SIZE); #endif static void kinfo_to_okinfo(struct kinfo_file *kif, struct kinfo_ofile *okif) { okif->kf_structsize = sizeof(*okif); okif->kf_type = kif->kf_type; okif->kf_fd = kif->kf_fd; okif->kf_ref_count = kif->kf_ref_count; okif->kf_flags = kif->kf_flags & (KF_FLAG_READ | KF_FLAG_WRITE | KF_FLAG_APPEND | KF_FLAG_ASYNC | KF_FLAG_FSYNC | KF_FLAG_NONBLOCK | KF_FLAG_DIRECT | KF_FLAG_HASLOCK); okif->kf_offset = kif->kf_offset; if (kif->kf_type == KF_TYPE_VNODE) okif->kf_vnode_type = kif->kf_un.kf_file.kf_file_type; else okif->kf_vnode_type = KF_VTYPE_VNON; strlcpy(okif->kf_path, kif->kf_path, sizeof(okif->kf_path)); if (kif->kf_type == KF_TYPE_SOCKET) { okif->kf_sock_domain = kif->kf_un.kf_sock.kf_sock_domain0; okif->kf_sock_type = kif->kf_un.kf_sock.kf_sock_type0; okif->kf_sock_protocol = kif->kf_un.kf_sock.kf_sock_protocol0; okif->kf_sa_local = kif->kf_un.kf_sock.kf_sa_local; okif->kf_sa_peer = kif->kf_un.kf_sock.kf_sa_peer; } else { okif->kf_sa_local.ss_family = AF_UNSPEC; okif->kf_sa_peer.ss_family = AF_UNSPEC; } } static int export_vnode_for_osysctl(struct vnode *vp, int type, struct kinfo_file *kif, struct kinfo_ofile *okif, struct filedesc *fdp, struct sysctl_req *req) { int error; vrefact(vp); FILEDESC_SUNLOCK(fdp); export_vnode_to_kinfo(vp, type, 0, kif, KERN_FILEDESC_PACK_KINFO); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); return (error); } /* * Get per-process file descriptors for use by procstat(1), et al. */ static int sysctl_kern_proc_ofiledesc(SYSCTL_HANDLER_ARGS) { struct kinfo_ofile *okif; struct kinfo_file *kif; struct filedesc *fdp; struct pwd *pwd; int error, i, lastfile, *name; struct file *fp; struct proc *p; name = (int *)arg1; error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) return (error); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (ENOENT); kif = malloc(sizeof(*kif), M_TEMP, M_WAITOK); okif = malloc(sizeof(*okif), M_TEMP, M_WAITOK); FILEDESC_SLOCK(fdp); pwd = pwd_hold_filedesc(fdp); if (pwd != NULL) { if (pwd->pwd_cdir != NULL) export_vnode_for_osysctl(pwd->pwd_cdir, KF_FD_TYPE_CWD, kif, okif, fdp, req); if (pwd->pwd_rdir != NULL) export_vnode_for_osysctl(pwd->pwd_rdir, KF_FD_TYPE_ROOT, kif, okif, fdp, req); if (pwd->pwd_jdir != NULL) export_vnode_for_osysctl(pwd->pwd_jdir, KF_FD_TYPE_JAIL, kif, okif, fdp, req); } lastfile = fdlastfile(fdp); for (i = 0; fdp->fd_refcnt > 0 && i <= lastfile; i++) { if ((fp = fdp->fd_ofiles[i].fde_file) == NULL) continue; export_file_to_kinfo(fp, i, NULL, kif, fdp, KERN_FILEDESC_PACK_KINFO); FILEDESC_SUNLOCK(fdp); kinfo_to_okinfo(kif, okif); error = SYSCTL_OUT(req, okif, sizeof(*okif)); FILEDESC_SLOCK(fdp); if (error) break; } FILEDESC_SUNLOCK(fdp); if (pwd != NULL) pwd_drop(pwd); fddrop(fdp); free(kif, M_TEMP); free(okif, M_TEMP); return (0); } static SYSCTL_NODE(_kern_proc, KERN_PROC_OFILEDESC, ofiledesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_ofiledesc, "Process ofiledesc entries"); #endif /* COMPAT_FREEBSD7 */ int vntype_to_kinfo(int vtype) { struct { int vtype; int kf_vtype; } vtypes_table[] = { { VBAD, KF_VTYPE_VBAD }, { VBLK, KF_VTYPE_VBLK }, { VCHR, KF_VTYPE_VCHR }, { VDIR, KF_VTYPE_VDIR }, { VFIFO, KF_VTYPE_VFIFO }, { VLNK, KF_VTYPE_VLNK }, { VNON, KF_VTYPE_VNON }, { VREG, KF_VTYPE_VREG }, { VSOCK, KF_VTYPE_VSOCK } }; unsigned int i; /* * Perform vtype translation. */ for (i = 0; i < nitems(vtypes_table); i++) if (vtypes_table[i].vtype == vtype) return (vtypes_table[i].kf_vtype); return (KF_VTYPE_UNKNOWN); } static SYSCTL_NODE(_kern_proc, KERN_PROC_FILEDESC, filedesc, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_filedesc, "Process filedesc entries"); /* * Store a process current working directory information to sbuf. * * Takes a locked proc as argument, and returns with the proc unlocked. */ int kern_proc_cwd_out(struct proc *p, struct sbuf *sb, ssize_t maxlen) { struct filedesc *fdp; struct pwd *pwd; struct export_fd_buf *efbuf; struct vnode *cdir; int error; PROC_LOCK_ASSERT(p, MA_OWNED); fdp = fdhold(p); PROC_UNLOCK(p); if (fdp == NULL) return (EINVAL); efbuf = malloc(sizeof(*efbuf), M_TEMP, M_WAITOK); efbuf->fdp = fdp; efbuf->sb = sb; efbuf->remainder = maxlen; FILEDESC_SLOCK(fdp); pwd = FILEDESC_LOCKED_LOAD_PWD(fdp); cdir = pwd->pwd_cdir; if (cdir == NULL) { error = EINVAL; } else { vrefact(cdir); error = export_vnode_to_sb(cdir, KF_FD_TYPE_CWD, FREAD, efbuf); } FILEDESC_SUNLOCK(fdp); fddrop(fdp); free(efbuf, M_TEMP); return (error); } /* * Get per-process current working directory. */ static int sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) { struct sbuf sb; struct proc *p; ssize_t maxlen; int error, error2, *name; name = (int *)arg1; sbuf_new_for_sysctl(&sb, NULL, sizeof(struct kinfo_file), req); sbuf_clear_flags(&sb, SBUF_INCLUDENUL); error = pget((pid_t)name[0], PGET_CANDEBUG | PGET_NOTWEXIT, &p); if (error != 0) { sbuf_delete(&sb); return (error); } maxlen = req->oldptr != NULL ? req->oldlen : -1; error = kern_proc_cwd_out(p, &sb, maxlen); error2 = sbuf_finish(&sb); sbuf_delete(&sb); return (error != 0 ? error : error2); } static SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, CTLFLAG_RD|CTLFLAG_MPSAFE, sysctl_kern_proc_cwd, "Process current working directory"); #ifdef DDB /* * For the purposes of debugging, generate a human-readable string for the * file type. */ static const char * file_type_to_name(short type) { switch (type) { case 0: return ("zero"); case DTYPE_VNODE: return ("vnode"); case DTYPE_SOCKET: return ("socket"); case DTYPE_PIPE: return ("pipe"); case DTYPE_FIFO: return ("fifo"); case DTYPE_KQUEUE: return ("kqueue"); case DTYPE_CRYPTO: return ("crypto"); case DTYPE_MQUEUE: return ("mqueue"); case DTYPE_SHM: return ("shm"); case DTYPE_SEM: return ("ksem"); case DTYPE_PTS: return ("pts"); case DTYPE_DEV: return ("dev"); case DTYPE_PROCDESC: return ("proc"); case DTYPE_LINUXEFD: return ("levent"); case DTYPE_LINUXTFD: return ("ltimer"); default: return ("unkn"); } } /* * For the purposes of debugging, identify a process (if any, perhaps one of * many) that references the passed file in its file descriptor array. Return * NULL if none. */ static struct proc * file_to_first_proc(struct file *fp) { struct filedesc *fdp; struct proc *p; int n; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; fdp = p->p_fd; if (fdp == NULL) continue; for (n = 0; n < fdp->fd_nfiles; n++) { if (fp == fdp->fd_ofiles[n].fde_file) return (p); } } return (NULL); } static void db_print_file(struct file *fp, int header) { #define XPTRWIDTH ((int)howmany(sizeof(void *) * NBBY, 4)) struct proc *p; if (header) db_printf("%*s %6s %*s %8s %4s %5s %6s %*s %5s %s\n", XPTRWIDTH, "File", "Type", XPTRWIDTH, "Data", "Flag", "GCFl", "Count", "MCount", XPTRWIDTH, "Vnode", "FPID", "FCmd"); p = file_to_first_proc(fp); db_printf("%*p %6s %*p %08x %04x %5d %6d %*p %5d %s\n", XPTRWIDTH, fp, file_type_to_name(fp->f_type), XPTRWIDTH, fp->f_data, fp->f_flag, 0, fp->f_count, 0, XPTRWIDTH, fp->f_vnode, p != NULL ? p->p_pid : -1, p != NULL ? p->p_comm : "-"); #undef XPTRWIDTH } DB_SHOW_COMMAND(file, db_show_file) { struct file *fp; if (!have_addr) { db_printf("usage: show file \n"); return; } fp = (struct file *)addr; db_print_file(fp, 1); } DB_SHOW_COMMAND(files, db_show_files) { struct filedesc *fdp; struct file *fp; struct proc *p; int header; int n; header = 1; FOREACH_PROC_IN_SYSTEM(p) { if (p->p_state == PRS_NEW) continue; if ((fdp = p->p_fd) == NULL) continue; for (n = 0; n < fdp->fd_nfiles; ++n) { if ((fp = fdp->fd_ofiles[n].fde_file) == NULL) continue; db_print_file(fp, header); header = 0; } } } #endif SYSCTL_INT(_kern, KERN_MAXFILESPERPROC, maxfilesperproc, CTLFLAG_RW, &maxfilesperproc, 0, "Maximum files allowed open per process"); SYSCTL_INT(_kern, KERN_MAXFILES, maxfiles, CTLFLAG_RW, &maxfiles, 0, "Maximum number of files"); SYSCTL_INT(_kern, OID_AUTO, openfiles, CTLFLAG_RD, &openfiles, 0, "System-wide number of open files"); /* ARGSUSED*/ static void filelistinit(void *dummy) { file_zone = uma_zcreate("Files", sizeof(struct file), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); filedesc0_zone = uma_zcreate("filedesc0", sizeof(struct filedesc0), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); pwd_zone = uma_zcreate("PWD", sizeof(struct pwd), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_SMR); /* * XXXMJG this is a temporary hack due to boot ordering issues against * the vnode zone. */ vfs_smr = uma_zone_get_smr(pwd_zone); mtx_init(&sigio_lock, "sigio lock", NULL, MTX_DEF); } SYSINIT(select, SI_SUB_LOCK, SI_ORDER_FIRST, filelistinit, NULL); /*-------------------------------------------------------------------*/ static int badfo_readwrite(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EBADF); } static int badfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } static int badfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (0); } static int badfo_kqfilter(struct file *fp, struct knote *kn) { return (EBADF); } static int badfo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_close(struct file *fp, struct thread *td) { return (0); } static int badfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EBADF); } static int badfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { return (EBADF); } static int badfo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return (0); } struct fileops badfileops = { .fo_read = badfo_readwrite, .fo_write = badfo_readwrite, .fo_truncate = badfo_truncate, .fo_ioctl = badfo_ioctl, .fo_poll = badfo_poll, .fo_kqfilter = badfo_kqfilter, .fo_stat = badfo_stat, .fo_close = badfo_close, .fo_chmod = badfo_chmod, .fo_chown = badfo_chown, .fo_sendfile = badfo_sendfile, .fo_fill_kinfo = badfo_fill_kinfo, }; int invfo_rdwr(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return (EOPNOTSUPP); } int invfo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return (ENOTTY); } int invfo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return (poll_no_poll(events)); } int invfo_kqfilter(struct file *fp, struct knote *kn) { return (EINVAL); } int invfo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return (EINVAL); } int invfo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { return (EINVAL); } /*-------------------------------------------------------------------*/ /* * File Descriptor pseudo-device driver (/dev/fd/). * * Opening minor device N dup()s the file (if any) connected to file * descriptor N belonging to the calling process. Note that this driver * consists of only the ``open()'' routine, because all subsequent * references to this file will be direct to the other driver. * * XXX: we could give this one a cloning event handler if necessary. */ /* ARGSUSED */ static int fdopen(struct cdev *dev, int mode, int type, struct thread *td) { /* * XXX Kludge: set curthread->td_dupfd to contain the value of the * the file descriptor being sought for duplication. The error * return ensures that the vnode for this device will be released * by vn_open. Open will detect this special error and take the * actions in dupfdopen below. Other callers of vn_open or VOP_OPEN * will simply report the error. */ td->td_dupfd = dev2unit(dev); return (ENODEV); } static struct cdevsw fildesc_cdevsw = { .d_version = D_VERSION, .d_open = fdopen, .d_name = "FD", }; static void fildesc_drvinit(void *unused) { struct cdev *dev; dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 0, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/0"); make_dev_alias(dev, "stdin"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 1, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/1"); make_dev_alias(dev, "stdout"); dev = make_dev_credf(MAKEDEV_ETERNAL, &fildesc_cdevsw, 2, NULL, UID_ROOT, GID_WHEEL, 0666, "fd/2"); make_dev_alias(dev, "stderr"); } SYSINIT(fildescdev, SI_SUB_DRIVERS, SI_ORDER_MIDDLE, fildesc_drvinit, NULL); Index: head/sys/kern/vfs_cache.c =================================================================== --- head/sys/kern/vfs_cache.c (revision 366596) +++ head/sys/kern/vfs_cache.c (revision 366597) @@ -1,4351 +1,4401 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Poul-Henning Kamp of the FreeBSD Project. * * 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_cache.c 8.5 (Berkeley) 3/22/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.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 #ifdef KTRACE #include #endif #include #include #include #ifdef DDB #include #endif #include SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, fullpath_smr, hit, "struct vnode *", "const char *"); SDT_PROBE_DEFINE4(vfs, namecache, fullpath_smr, miss, "struct vnode *", "struct namecache *", "int", "int"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, removecnp, hit, "struct vnode *", "struct componentname *"); SDT_PROBE_DEFINE2(vfs, namecache, removecnp, miss, "struct vnode *", "struct componentname *"); SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, shrink_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, fplookup, lookup, done, "struct nameidata", "int", "bool"); SDT_PROBE_DECLARE(vfs, namei, lookup, entry); SDT_PROBE_DECLARE(vfs, namei, lookup, return); /* * This structure describes the elements in the cache of recent * names looked up by namei. */ struct negstate { u_char neg_flag; }; _Static_assert(sizeof(struct negstate) <= sizeof(struct vnode *), "the state must fit in a union with a pointer without growing it"); struct namecache { LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ CK_SLIST_ENTRY(namecache) nc_hash;/* hash chain */ struct vnode *nc_dvp; /* vnode of parent of name */ union { struct vnode *nu_vp; /* vnode the name refers to */ struct negstate nu_neg;/* negative entry state */ } n_un; u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ char nc_name[0]; /* segment name + nul */ }; /* * struct namecache_ts repeats struct namecache layout up to the * nc_nlen member. * struct namecache_ts is used in place of struct namecache when time(s) need * to be stored. The nc_dotdottime field is used when a cache entry is mapping * both a non-dotdot directory name plus dotdot for the directory's * parent. * * See below for alignment requirement. */ struct namecache_ts { struct timespec nc_time; /* timespec provided by fs */ struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ int nc_ticks; /* ticks value when entry was added */ struct namecache nc_nc; }; /* * At least mips n32 performs 64-bit accesses to timespec as found * in namecache_ts and requires them to be aligned. Since others * may be in the same spot suffer a little bit and enforce the * alignment for everyone. Note this is a nop for 64-bit platforms. */ #define CACHE_ZONE_ALIGNMENT UMA_ALIGNOF(time_t) #define CACHE_PATH_CUTOFF 39 #define CACHE_ZONE_SMALL_SIZE (sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1) #define CACHE_ZONE_SMALL_TS_SIZE (sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1) #define CACHE_ZONE_LARGE_SIZE (sizeof(struct namecache) + NAME_MAX + 1) #define CACHE_ZONE_LARGE_TS_SIZE (sizeof(struct namecache_ts) + NAME_MAX + 1) _Static_assert((CACHE_ZONE_SMALL_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_SMALL_TS_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_LARGE_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); _Static_assert((CACHE_ZONE_LARGE_TS_SIZE % (CACHE_ZONE_ALIGNMENT + 1)) == 0, "bad zone size"); #define nc_vp n_un.nu_vp #define nc_neg n_un.nu_neg /* * Flags in namecache.nc_flag */ #define NCF_WHITE 0x01 #define NCF_ISDOTDOT 0x02 #define NCF_TS 0x04 #define NCF_DTS 0x08 #define NCF_DVDROP 0x10 #define NCF_NEGATIVE 0x20 #define NCF_INVALID 0x40 #define NCF_WIP 0x80 /* * Flags in negstate.neg_flag */ #define NEG_HOT 0x01 /* * Mark an entry as invalid. * * This is called before it starts getting deconstructed. */ static void cache_ncp_invalidate(struct namecache *ncp) { KASSERT((ncp->nc_flag & NCF_INVALID) == 0, ("%s: entry %p already invalid", __func__, ncp)); atomic_store_char(&ncp->nc_flag, ncp->nc_flag | NCF_INVALID); atomic_thread_fence_rel(); } /* * Check whether the entry can be safely used. * * All places which elide locks are supposed to call this after they are * done with reading from an entry. */ static bool cache_ncp_canuse(struct namecache *ncp) { atomic_thread_fence_acq(); return ((atomic_load_char(&ncp->nc_flag) & (NCF_INVALID | NCF_WIP)) == 0); } /* * Name caching works as follows: * * Names found by directory scans are retained in a cache * for future reference. It is managed LRU, so frequently * used names will hang around. Cache is indexed by hash value * obtained from (dvp, name) where dvp refers to the directory * containing name. * * If it is a "negative" entry, (i.e. for a name that is known NOT to * exist) the vnode pointer will be NULL. * * Upon reaching the last segment of a path, if the reference * is for DELETE, or NOCACHE is set (rewrite), and the * name is located in the cache, it will be dropped. * * These locks are used (in the order in which they can be taken): * NAME TYPE ROLE * vnodelock mtx vnode lists and v_cache_dd field protection * bucketlock mtx for access to given set of hash buckets * neglist mtx negative entry LRU management * * Additionally, ncneg_shrink_lock mtx is used to have at most one thread * shrinking the LRU list. * * It is legal to take multiple vnodelock and bucketlock locks. The locking * order is lower address first. Both are recursive. * * "." lookups are lockless. * * ".." and vnode -> name lookups require vnodelock. * * name -> vnode lookup requires the relevant bucketlock to be held for reading. * * Insertions and removals of entries require involved vnodes and bucketlocks * to be locked to provide safe operation against other threads modifying the * cache. * * Some lookups result in removal of the found entry (e.g. getting rid of a * negative entry with the intent to create a positive one), which poses a * problem when multiple threads reach the state. Similarly, two different * threads can purge two different vnodes and try to remove the same name. * * If the already held vnode lock is lower than the second required lock, we * can just take the other lock. However, in the opposite case, this could * deadlock. As such, this is resolved by trylocking and if that fails unlocking * the first node, locking everything in order and revalidating the state. */ VFS_SMR_DECLARE; /* * Structures associated with name caching. */ #define NCHHASH(hash) \ (&nchashtbl[(hash) & nchash]) static __read_mostly CK_SLIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */ static u_long __read_mostly nchash; /* size of hash table */ SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "Size of namecache hash table"); static u_long __read_mostly ncnegfactor = 5; /* ratio of negative entries */ SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "Ratio of negative namecache entries"); static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */ static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */ u_int ncsizefactor = 2; SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, "Size factor for namecache"); static u_int __read_mostly ncsize; /* the size as computed on creation or resizing */ struct nchstats nchstats; /* cache effectiveness statistics */ static bool __read_frequently cache_fast_revlookup = true; SYSCTL_BOOL(_vfs, OID_AUTO, cache_fast_revlookup, CTLFLAG_RW, &cache_fast_revlookup, 0, ""); static struct mtx __exclusive_cache_line ncneg_shrink_lock; struct neglist { struct mtx nl_lock; TAILQ_HEAD(, namecache) nl_list; } __aligned(CACHE_LINE_SIZE); static struct neglist __read_mostly *neglists; static struct neglist ncneg_hot; static u_long numhotneg; #define ncneghash 3 #define numneglists (ncneghash + 1) static inline struct neglist * NCP2NEGLIST(struct namecache *ncp) { return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]); } static inline struct negstate * NCP2NEGSTATE(struct namecache *ncp) { MPASS(ncp->nc_flag & NCF_NEGATIVE); return (&ncp->nc_neg); } #define numbucketlocks (ncbuckethash + 1) static u_int __read_mostly ncbuckethash; static struct mtx_padalign __read_mostly *bucketlocks; #define HASH2BUCKETLOCK(hash) \ ((struct mtx *)(&bucketlocks[((hash) & ncbuckethash)])) #define numvnodelocks (ncvnodehash + 1) static u_int __read_mostly ncvnodehash; static struct mtx __read_mostly *vnodelocks; static inline struct mtx * VP2VNODELOCK(struct vnode *vp) { return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]); } /* * UMA zones for the VFS cache. * * The small cache is used for entries with short names, which are the * most common. The large cache is used for entries which are too big to * fit in the small cache. */ static uma_zone_t __read_mostly cache_zone_small; static uma_zone_t __read_mostly cache_zone_small_ts; static uma_zone_t __read_mostly cache_zone_large; static uma_zone_t __read_mostly cache_zone_large_ts; static struct namecache * cache_alloc(int len, int ts) { struct namecache_ts *ncp_ts; struct namecache *ncp; if (__predict_false(ts)) { if (len <= CACHE_PATH_CUTOFF) ncp_ts = uma_zalloc_smr(cache_zone_small_ts, M_WAITOK); else ncp_ts = uma_zalloc_smr(cache_zone_large_ts, M_WAITOK); ncp = &ncp_ts->nc_nc; } else { if (len <= CACHE_PATH_CUTOFF) ncp = uma_zalloc_smr(cache_zone_small, M_WAITOK); else ncp = uma_zalloc_smr(cache_zone_large, M_WAITOK); } return (ncp); } static void cache_free(struct namecache *ncp) { struct namecache_ts *ncp_ts; MPASS(ncp != NULL); if ((ncp->nc_flag & NCF_DVDROP) != 0) vdrop(ncp->nc_dvp); if (__predict_false(ncp->nc_flag & NCF_TS)) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree_smr(cache_zone_small_ts, ncp_ts); else uma_zfree_smr(cache_zone_large_ts, ncp_ts); } else { if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree_smr(cache_zone_small, ncp); else uma_zfree_smr(cache_zone_large, ncp); } } static void cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; KASSERT((ncp->nc_flag & NCF_TS) != 0 || (tsp == NULL && ticksp == NULL), ("No NCF_TS")); if (tsp == NULL) return; ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_time; *ticksp = ncp_ts->nc_ticks; } #ifdef DEBUG_CACHE static int __read_mostly doingcache = 1; /* 1 => enable the cache */ SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "VFS namecache enabled"); #endif /* Export size information to userland */ SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct namecache), "sizeof(struct namecache)"); /* * The new name cache statistics */ static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache statistics"); #define STATNODE_ULONG(name, descr) \ SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr); #define STATNODE_COUNTER(name, descr) \ static COUNTER_U64_DEFINE_EARLY(name); \ SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, \ descr); STATNODE_ULONG(numneg, "Number of negative cache entries"); STATNODE_ULONG(numcache, "Number of cache entries"); STATNODE_COUNTER(numcachehv, "Number of namecache entries with vnodes held"); STATNODE_COUNTER(numdrops, "Number of dropped entries due to reaching the limit"); STATNODE_COUNTER(dothits, "Number of '.' hits"); STATNODE_COUNTER(dotdothits, "Number of '..' hits"); STATNODE_COUNTER(nummiss, "Number of cache misses"); STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache"); STATNODE_COUNTER(numposzaps, "Number of cache hits (positive) we do not want to cache"); STATNODE_COUNTER(numposhits, "Number of cache hits (positive)"); STATNODE_COUNTER(numnegzaps, "Number of cache hits (negative) we do not want to cache"); STATNODE_COUNTER(numneghits, "Number of cache hits (negative)"); /* These count for vn_getcwd(), too. */ STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls"); STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); STATNODE_COUNTER(numfullpathfail2, "Number of fullpath search errors (VOP_VPTOCNP failures)"); STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls"); STATNODE_COUNTER(zap_and_exit_bucket_relock_success, "Number of successful removals after relocking"); static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail, "Number of times zap_and_exit failed to lock"); static long zap_and_exit_bucket_fail2; STATNODE_ULONG(zap_and_exit_bucket_fail2, "Number of times zap_and_exit failed to lock"); static long cache_lock_vnodes_cel_3_failures; STATNODE_ULONG(cache_lock_vnodes_cel_3_failures, "Number of times 3-way vnode locking failed"); STATNODE_ULONG(numhotneg, "Number of hot negative entries"); STATNODE_COUNTER(numneg_evicted, "Number of negative entries evicted when adding a new entry"); STATNODE_COUNTER(shrinking_skipped, "Number of times shrinking was already in progress"); static void cache_zap_locked(struct namecache *ncp); static int vn_fullpath_hardlink(struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen); static int vn_fullpath_any_smr(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen, bool slash_prefixed, size_t addend); static int vn_fullpath_any(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen); static int vn_fullpath_dir(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, bool slash_prefixed, size_t addend); static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); static inline void cache_assert_vlp_locked(struct mtx *vlp) { if (vlp != NULL) mtx_assert(vlp, MA_OWNED); } static inline void cache_assert_vnode_locked(struct vnode *vp) { struct mtx *vlp; vlp = VP2VNODELOCK(vp); cache_assert_vlp_locked(vlp); } /* * TODO: With the value stored we can do better than computing the hash based * on the address. The choice of FNV should also be revisited. */ static void cache_prehash(struct vnode *vp) { vp->v_nchash = fnv_32_buf(&vp, sizeof(vp), FNV1_32_INIT); } static uint32_t cache_get_hash(char *name, u_char len, struct vnode *dvp) { return (fnv_32_buf(name, len, dvp->v_nchash)); } static inline struct nchashhead * NCP2BUCKET(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (NCHHASH(hash)); } static inline struct mtx * NCP2BUCKETLOCK(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (HASH2BUCKETLOCK(hash)); } #ifdef INVARIANTS static void cache_assert_bucket_locked(struct namecache *ncp) { struct mtx *blp; blp = NCP2BUCKETLOCK(ncp); mtx_assert(blp, MA_OWNED); } static void cache_assert_bucket_unlocked(struct namecache *ncp) { struct mtx *blp; blp = NCP2BUCKETLOCK(ncp); mtx_assert(blp, MA_NOTOWNED); } #else #define cache_assert_bucket_locked(x) do { } while (0) #define cache_assert_bucket_unlocked(x) do { } while (0) #endif #define cache_sort_vnodes(x, y) _cache_sort_vnodes((void **)(x), (void **)(y)) static void _cache_sort_vnodes(void **p1, void **p2) { void *tmp; MPASS(*p1 != NULL || *p2 != NULL); if (*p1 > *p2) { tmp = *p2; *p2 = *p1; *p1 = tmp; } } static void cache_lock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) mtx_lock(&bucketlocks[i]); } static void cache_unlock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) mtx_unlock(&bucketlocks[i]); } static void cache_lock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_lock(&vnodelocks[i]); } static void cache_unlock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_unlock(&vnodelocks[i]); } static int cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { if (!mtx_trylock(vlp1)) return (EAGAIN); } if (!mtx_trylock(vlp2)) { if (vlp1 != NULL) mtx_unlock(vlp1); return (EAGAIN); } return (0); } static void cache_lock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); MPASS(vlp1 <= vlp2); if (vlp1 != NULL) mtx_lock(vlp1); if (vlp2 != NULL) mtx_lock(vlp2); } static void cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); } static int sysctl_nchstats(SYSCTL_HANDLER_ARGS) { struct nchstats snap; if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, sizeof(snap))); snap = nchstats; snap.ncs_goodhits = counter_u64_fetch(numposhits); snap.ncs_neghits = counter_u64_fetch(numneghits); snap.ncs_badhits = counter_u64_fetch(numposzaps) + counter_u64_fetch(numnegzaps); snap.ncs_miss = counter_u64_fetch(nummisszap) + counter_u64_fetch(nummiss); return (SYSCTL_OUT(req, &snap, sizeof(snap))); } SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", "VFS cache effectiveness statistics"); #ifdef DIAGNOSTIC /* * Grab an atomic snapshot of the name cache hash chain lengths */ static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "hash table stats"); static int sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) { struct nchashhead *ncpp; struct namecache *ncp; int i, error, n_nchash, *cntbuf; retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); cache_lock_all_buckets(); if (n_nchash != nchash + 1) { cache_unlock_all_buckets(); free(cntbuf, M_TEMP); goto retry; } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) CK_SLIST_FOREACH(ncp, ncpp, nc_hash) cntbuf[i]++; cache_unlock_all_buckets(); for (error = 0, i = 0; i < n_nchash; i++) if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) break; free(cntbuf, M_TEMP); return (error); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); static int sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) { int error; struct nchashhead *ncpp; struct namecache *ncp; int n_nchash; int count, maxlength, used, pct; if (!req->oldptr) return SYSCTL_OUT(req, 0, 4 * sizeof(int)); cache_lock_all_buckets(); n_nchash = nchash + 1; /* nchash is max index, not count */ used = 0; maxlength = 0; /* Scan hash tables for applicable entries */ for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { count = 0; CK_SLIST_FOREACH(ncp, ncpp, nc_hash) { count++; } if (count) used++; if (maxlength < count) maxlength = count; } n_nchash = nchash + 1; cache_unlock_all_buckets(); pct = (used * 100) / (n_nchash / 100); error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); if (error) return (error); error = SYSCTL_OUT(req, &used, sizeof(used)); if (error) return (error); error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); if (error) return (error); error = SYSCTL_OUT(req, &pct, sizeof(pct)); if (error) return (error); return (0); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); #endif /* * Negative entries management * * A variation of LRU scheme is used. New entries are hashed into one of * numneglists cold lists. Entries get promoted to the hot list on first hit. * * The shrinker will demote hot list head and evict from the cold list in a * round-robin manner. */ static void cache_negative_init(struct namecache *ncp) { struct negstate *negstate; ncp->nc_flag |= NCF_NEGATIVE; negstate = NCP2NEGSTATE(ncp); negstate->neg_flag = 0; } static void cache_negative_hit(struct namecache *ncp) { struct neglist *neglist; struct negstate *negstate; negstate = NCP2NEGSTATE(ncp); if ((negstate->neg_flag & NEG_HOT) != 0) return; neglist = NCP2NEGLIST(ncp); mtx_lock(&ncneg_hot.nl_lock); mtx_lock(&neglist->nl_lock); if ((negstate->neg_flag & NEG_HOT) == 0) { numhotneg++; TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); negstate->neg_flag |= NEG_HOT; } mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); } static void cache_negative_insert(struct namecache *ncp) { struct neglist *neglist; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp); neglist = NCP2NEGLIST(ncp); mtx_lock(&neglist->nl_lock); TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); mtx_unlock(&neglist->nl_lock); atomic_add_long(&numneg, 1); } static void cache_negative_remove(struct namecache *ncp) { struct neglist *neglist; struct negstate *negstate; bool hot_locked = false; bool list_locked = false; cache_assert_bucket_locked(ncp); neglist = NCP2NEGLIST(ncp); negstate = NCP2NEGSTATE(ncp); if ((negstate->neg_flag & NEG_HOT) != 0) { hot_locked = true; mtx_lock(&ncneg_hot.nl_lock); if ((negstate->neg_flag & NEG_HOT) == 0) { list_locked = true; mtx_lock(&neglist->nl_lock); } } else { list_locked = true; mtx_lock(&neglist->nl_lock); /* * We may be racing against promotion in lockless lookup. */ if ((negstate->neg_flag & NEG_HOT) != 0) { mtx_unlock(&neglist->nl_lock); hot_locked = true; mtx_lock(&ncneg_hot.nl_lock); mtx_lock(&neglist->nl_lock); } } if ((negstate->neg_flag & NEG_HOT) != 0) { mtx_assert(&ncneg_hot.nl_lock, MA_OWNED); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); numhotneg--; } else { mtx_assert(&neglist->nl_lock, MA_OWNED); TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); } if (list_locked) mtx_unlock(&neglist->nl_lock); if (hot_locked) mtx_unlock(&ncneg_hot.nl_lock); atomic_subtract_long(&numneg, 1); } static void cache_negative_shrink_select(struct namecache **ncpp, struct neglist **neglistpp) { struct neglist *neglist; struct namecache *ncp; static u_int cycle; u_int i; *ncpp = ncp = NULL; for (i = 0; i < numneglists; i++) { neglist = &neglists[(cycle + i) % numneglists]; if (TAILQ_FIRST(&neglist->nl_list) == NULL) continue; mtx_lock(&neglist->nl_lock); ncp = TAILQ_FIRST(&neglist->nl_list); if (ncp != NULL) break; mtx_unlock(&neglist->nl_lock); } *neglistpp = neglist; *ncpp = ncp; cycle++; } static void cache_negative_zap_one(void) { struct namecache *ncp, *ncp2; struct neglist *neglist; struct negstate *negstate; struct mtx *dvlp; struct mtx *blp; if (mtx_owner(&ncneg_shrink_lock) != NULL || !mtx_trylock(&ncneg_shrink_lock)) { counter_u64_add(shrinking_skipped, 1); return; } mtx_lock(&ncneg_hot.nl_lock); ncp = TAILQ_FIRST(&ncneg_hot.nl_list); if (ncp != NULL) { neglist = NCP2NEGLIST(ncp); negstate = NCP2NEGSTATE(ncp); mtx_lock(&neglist->nl_lock); MPASS((negstate->neg_flag & NEG_HOT) != 0); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); negstate->neg_flag &= ~NEG_HOT; numhotneg--; mtx_unlock(&neglist->nl_lock); } mtx_unlock(&ncneg_hot.nl_lock); cache_negative_shrink_select(&ncp, &neglist); mtx_unlock(&ncneg_shrink_lock); if (ncp == NULL) return; MPASS(ncp->nc_flag & NCF_NEGATIVE); dvlp = VP2VNODELOCK(ncp->nc_dvp); blp = NCP2BUCKETLOCK(ncp); mtx_unlock(&neglist->nl_lock); mtx_lock(dvlp); mtx_lock(blp); /* * Enter SMR to safely check the negative list. * Even if the found pointer matches, the entry may now be reallocated * and used by a different vnode. */ vfs_smr_enter(); ncp2 = TAILQ_FIRST(&neglist->nl_list); if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) || blp != NCP2BUCKETLOCK(ncp2)) { vfs_smr_exit(); ncp = NULL; } else { vfs_smr_exit(); SDT_PROBE2(vfs, namecache, shrink_negative, done, ncp->nc_dvp, ncp->nc_name); cache_zap_locked(ncp); counter_u64_add(numneg_evicted, 1); } mtx_unlock(blp); mtx_unlock(dvlp); if (ncp != NULL) cache_free(ncp); } /* * cache_zap_locked(): * * Removes a namecache entry from cache, whether it contains an actual * pointer to a vnode or if it is just a negative cache entry. */ static void cache_zap_locked(struct namecache *ncp) { struct nchashhead *ncpp; if (!(ncp->nc_flag & NCF_NEGATIVE)) cache_assert_vnode_locked(ncp->nc_vp); cache_assert_vnode_locked(ncp->nc_dvp); cache_assert_bucket_locked(ncp); cache_ncp_invalidate(ncp); ncpp = NCP2BUCKET(ncp); CK_SLIST_REMOVE(ncpp, ncp, namecache, nc_hash); if (!(ncp->nc_flag & NCF_NEGATIVE)) { SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, ncp->nc_name, ncp->nc_vp); TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); if (ncp == ncp->nc_vp->v_cache_dd) { vn_seqc_write_begin_unheld(ncp->nc_vp); ncp->nc_vp->v_cache_dd = NULL; vn_seqc_write_end(ncp->nc_vp); } } else { SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp, ncp->nc_name); cache_negative_remove(ncp); } if (ncp->nc_flag & NCF_ISDOTDOT) { if (ncp == ncp->nc_dvp->v_cache_dd) { vn_seqc_write_begin_unheld(ncp->nc_dvp); ncp->nc_dvp->v_cache_dd = NULL; vn_seqc_write_end(ncp->nc_dvp); } } else { LIST_REMOVE(ncp, nc_src); if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { ncp->nc_flag |= NCF_DVDROP; counter_u64_add(numcachehv, -1); } } atomic_subtract_long(&numcache, 1); } static void cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp) { struct mtx *blp; MPASS(ncp->nc_dvp == vp); MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_vnode_locked(vp); blp = NCP2BUCKETLOCK(ncp); mtx_lock(blp); cache_zap_locked(ncp); mtx_unlock(blp); } static bool cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp, struct mtx **vlpp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct mtx *blp; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); if (ncp->nc_flag & NCF_NEGATIVE) { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_zap_negative_locked_vnode_kl(ncp, vp); return (true); } pvlp = VP2VNODELOCK(vp); blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); if (*vlpp == vlp1 || *vlpp == vlp2) { to_unlock = *vlpp; *vlpp = NULL; } else { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) goto out_relock; to_unlock = vlp1; } } mtx_lock(blp); cache_zap_locked(ncp); mtx_unlock(blp); if (to_unlock != NULL) mtx_unlock(to_unlock); return (true); out_relock: mtx_unlock(vlp2); mtx_lock(vlp1); mtx_lock(vlp2); MPASS(*vlpp == NULL); *vlpp = vlp1; return (false); } /* * If trylocking failed we can get here. We know enough to take all needed locks * in the right order and re-lookup the entry. */ static int cache_zap_unlocked_bucket(struct namecache *ncp, struct componentname *cnp, struct vnode *dvp, struct mtx *dvlp, struct mtx *vlp, uint32_t hash, struct mtx *blp) { struct namecache *rncp; cache_assert_bucket_unlocked(ncp); cache_sort_vnodes(&dvlp, &vlp); cache_lock_vnodes(dvlp, vlp); mtx_lock(blp); CK_SLIST_FOREACH(rncp, (NCHHASH(hash)), nc_hash) { if (rncp == ncp && rncp->nc_dvp == dvp && rncp->nc_nlen == cnp->cn_namelen && !bcmp(rncp->nc_name, cnp->cn_nameptr, rncp->nc_nlen)) break; } if (rncp != NULL) { cache_zap_locked(rncp); mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); counter_u64_add(zap_and_exit_bucket_relock_success, 1); return (0); } mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); return (EAGAIN); } static int __noinline cache_zap_locked_bucket(struct namecache *ncp, struct componentname *cnp, uint32_t hash, struct mtx *blp) { struct mtx *dvlp, *vlp; struct vnode *dvp; cache_assert_bucket_locked(ncp); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); if (cache_trylock_vnodes(dvlp, vlp) == 0) { cache_zap_locked(ncp); mtx_unlock(blp); cache_unlock_vnodes(dvlp, vlp); return (0); } dvp = ncp->nc_dvp; mtx_unlock(blp); return (cache_zap_unlocked_bucket(ncp, cnp, dvp, dvlp, vlp, hash, blp)); } static __noinline int cache_remove_cnp(struct vnode *dvp, struct componentname *cnp) { struct namecache *ncp; struct mtx *blp; struct mtx *dvlp, *dvlp2; uint32_t hash; int error; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { dvlp = VP2VNODELOCK(dvp); dvlp2 = NULL; mtx_lock(dvlp); retry_dotdot: ncp = dvp->v_cache_dd; if (ncp == NULL) { mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); SDT_PROBE2(vfs, namecache, removecnp, miss, dvp, cnp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (!cache_zap_locked_vnode_kl2(ncp, dvp, &dvlp2)) goto retry_dotdot; MPASS(dvp->v_cache_dd == NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); cache_free(ncp); } else { vn_seqc_write_begin(dvp); dvp->v_cache_dd = NULL; vn_seqc_write_end(dvp); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); } SDT_PROBE2(vfs, namecache, removecnp, hit, dvp, cnp); return (1); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); retry: if (CK_SLIST_EMPTY(NCHHASH(hash))) goto out_no_entry; mtx_lock(blp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (ncp == NULL) { mtx_unlock(blp); goto out_no_entry; } error = cache_zap_locked_bucket(ncp, cnp, hash, blp); if (__predict_false(error != 0)) { zap_and_exit_bucket_fail++; goto retry; } counter_u64_add(numposzaps, 1); SDT_PROBE2(vfs, namecache, removecnp, hit, dvp, cnp); cache_free(ncp); return (1); out_no_entry: counter_u64_add(nummisszap, 1); SDT_PROBE2(vfs, namecache, removecnp, miss, dvp, cnp); return (0); } static int __noinline cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { int ltype; *vpp = dvp; counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); if (tsp != NULL) timespecclear(tsp); if (ticksp != NULL) *ticksp = ticks; vrefact(*vpp); /* * When we lookup "." we still can be asked to lock it * differently... */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(*vpp)) { if (ltype == LK_EXCLUSIVE) { vn_lock(*vpp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED((*vpp))) { /* forced unmount */ vrele(*vpp); *vpp = NULL; return (ENOENT); } } else vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); } return (-1); } static int __noinline cache_lookup_dotdot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; struct namecache *ncp; struct mtx *dvlp; enum vgetstate vs; int error, ltype; bool whiteout; MPASS((cnp->cn_flags & ISDOTDOT) != 0); if ((cnp->cn_flags & MAKEENTRY) == 0) { cache_remove_cnp(dvp, cnp); return (0); } counter_u64_add(dotdothits, 1); retry: dvlp = VP2VNODELOCK(dvp); mtx_lock(dvlp); ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); mtx_unlock(dvlp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_flag & NCF_NEGATIVE) *vpp = NULL; else *vpp = ncp->nc_vp; } else *vpp = ncp->nc_dvp; if (*vpp == NULL) goto negative_success; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", *vpp); cache_out_ts(ncp, tsp, ticksp); if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == NCF_DTS && tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_dotdottime; } MPASS(dvp != *vpp); ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp); vs = vget_prep(*vpp); mtx_unlock(dvlp); error = vget_finish(*vpp, cnp->cn_lkflags, vs); vn_lock(dvp, ltype | LK_RETRY); if (VN_IS_DOOMED(dvp)) { if (error == 0) vput(*vpp); *vpp = NULL; return (ENOENT); } if (error) { *vpp = NULL; goto retry; } return (-1); negative_success: if (__predict_false(cnp->cn_nameiop == CREATE)) { if (cnp->cn_flags & ISLASTCN) { counter_u64_add(numnegzaps, 1); cache_zap_negative_locked_vnode_kl(ncp, dvp); mtx_unlock(dvlp); cache_free(ncp); return (0); } } SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); cache_out_ts(ncp, tsp, ticksp); counter_u64_add(numneghits, 1); whiteout = (ncp->nc_flag & NCF_WHITE); cache_negative_hit(ncp); mtx_unlock(dvlp); if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); } /** * Lookup a name in the name cache * * # Arguments * * - dvp: Parent directory in which to search. * - vpp: Return argument. Will contain desired vnode on cache hit. * - cnp: Parameters of the name search. The most interesting bits of * the cn_flags field have the following meanings: * - MAKEENTRY: If clear, free an entry from the cache rather than look * it up. * - ISDOTDOT: Must be set if and only if cn_nameptr == ".." * - tsp: Return storage for cache timestamp. On a successful (positive * or negative) lookup, tsp will be filled with any timespec that * was stored when this cache entry was created. However, it will * be clear for "." entries. * - ticks: Return storage for alternate cache timestamp. On a successful * (positive or negative) lookup, it will contain the ticks value * that was current when the cache entry was created, unless cnp * was ".". * * Either both tsp and ticks have to be provided or neither of them. * * # Returns * * - -1: A positive cache hit. vpp will contain the desired vnode. * - ENOENT: A negative cache hit, or dvp was recycled out from under us due * to a forced unmount. vpp will not be modified. If the entry * is a whiteout, then the ISWHITEOUT flag will be set in * cnp->cn_flags. * - 0: A cache miss. vpp will not be modified. * * # Locking * * On a cache hit, vpp will be returned locked and ref'd. If we're looking up * .., dvp is unlocked. If we're looking up . an extra ref is taken, but the * lock is not recursively acquired. */ static int __noinline cache_lookup_fallback(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; struct mtx *blp; uint32_t hash; enum vgetstate vs; int error; bool whiteout; MPASS((cnp->cn_flags & (MAKEENTRY | ISDOTDOT)) == MAKEENTRY); retry: hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); mtx_lock(blp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (__predict_false(ncp == NULL)) { mtx_unlock(blp); SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummiss, 1); return (0); } if (ncp->nc_flag & NCF_NEGATIVE) goto negative_success; counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); MPASS(dvp != *vpp); vs = vget_prep(*vpp); mtx_unlock(blp); error = vget_finish(*vpp, cnp->cn_lkflags, vs); if (error) { *vpp = NULL; goto retry; } return (-1); negative_success: if (__predict_false(cnp->cn_nameiop == CREATE)) { if (cnp->cn_flags & ISLASTCN) { counter_u64_add(numnegzaps, 1); error = cache_zap_locked_bucket(ncp, cnp, hash, blp); if (__predict_false(error != 0)) { zap_and_exit_bucket_fail2++; goto retry; } cache_free(ncp); return (0); } } SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); cache_out_ts(ncp, tsp, ticksp); counter_u64_add(numneghits, 1); whiteout = (ncp->nc_flag & NCF_WHITE); cache_negative_hit(ncp); mtx_unlock(blp); if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); } int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; struct negstate *negstate; uint32_t hash; enum vgetstate vs; int error; bool whiteout; u_short nc_flag; MPASS((tsp == NULL && ticksp == NULL) || (tsp != NULL && ticksp != NULL)); #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) { cnp->cn_flags &= ~MAKEENTRY; return (0); } #endif if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') return (cache_lookup_dotdot(dvp, vpp, cnp, tsp, ticksp)); } MPASS((cnp->cn_flags & ISDOTDOT) == 0); if ((cnp->cn_flags & MAKEENTRY) == 0) { cache_remove_cnp(dvp, cnp); return (0); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); vfs_smr_enter(); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } if (__predict_false(ncp == NULL)) { vfs_smr_exit(); SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummiss, 1); return (0); } nc_flag = atomic_load_char(&ncp->nc_flag); if (nc_flag & NCF_NEGATIVE) goto negative_success; counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); MPASS(dvp != *vpp); if (!cache_ncp_canuse(ncp)) { vfs_smr_exit(); *vpp = NULL; goto out_fallback; } vs = vget_prep_smr(*vpp); vfs_smr_exit(); if (__predict_false(vs == VGET_NONE)) { *vpp = NULL; goto out_fallback; } error = vget_finish(*vpp, cnp->cn_lkflags, vs); if (error) { *vpp = NULL; goto out_fallback; } return (-1); negative_success: if (__predict_false(cnp->cn_nameiop == CREATE)) { if (cnp->cn_flags & ISLASTCN) { vfs_smr_exit(); goto out_fallback; } } SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); cache_out_ts(ncp, tsp, ticksp); counter_u64_add(numneghits, 1); whiteout = (ncp->nc_flag & NCF_WHITE); /* * TODO: We need to take locks to promote an entry. Code doing it * in SMR lookup can be modified to be shared. */ negstate = NCP2NEGSTATE(ncp); if ((negstate->neg_flag & NEG_HOT) == 0 || !cache_ncp_canuse(ncp)) { vfs_smr_exit(); goto out_fallback; } vfs_smr_exit(); if (whiteout) cnp->cn_flags |= ISWHITEOUT; return (ENOENT); out_fallback: return (cache_lookup_fallback(dvp, vpp, cnp, tsp, ticksp)); } struct celockstate { struct mtx *vlp[3]; struct mtx *blp[2]; }; CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); static inline void cache_celockstate_init(struct celockstate *cel) { bzero(cel, sizeof(*cel)); } static void cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, struct vnode *dvp) { struct mtx *vlp1, *vlp2; MPASS(cel->vlp[0] == NULL); MPASS(cel->vlp[1] == NULL); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL || dvp != NULL); vlp1 = VP2VNODELOCK(vp); vlp2 = VP2VNODELOCK(dvp); cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { mtx_lock(vlp1); cel->vlp[0] = vlp1; } mtx_lock(vlp2); cel->vlp[1] = vlp2; } static void cache_unlock_vnodes_cel(struct celockstate *cel) { MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); if (cel->vlp[0] != NULL) mtx_unlock(cel->vlp[0]); if (cel->vlp[1] != NULL) mtx_unlock(cel->vlp[1]); if (cel->vlp[2] != NULL) mtx_unlock(cel->vlp[2]); } static bool cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) { struct mtx *vlp; bool ret; cache_assert_vlp_locked(cel->vlp[0]); cache_assert_vlp_locked(cel->vlp[1]); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL); vlp = VP2VNODELOCK(vp); ret = true; if (vlp >= cel->vlp[1]) { mtx_lock(vlp); } else { if (mtx_trylock(vlp)) goto out; cache_lock_vnodes_cel_3_failures++; cache_unlock_vnodes_cel(cel); if (vlp < cel->vlp[0]) { mtx_lock(vlp); mtx_lock(cel->vlp[0]); mtx_lock(cel->vlp[1]); } else { if (cel->vlp[0] != NULL) mtx_lock(cel->vlp[0]); mtx_lock(vlp); mtx_lock(cel->vlp[1]); } ret = false; } out: cel->vlp[2] = vlp; return (ret); } static void cache_lock_buckets_cel(struct celockstate *cel, struct mtx *blp1, struct mtx *blp2) { MPASS(cel->blp[0] == NULL); MPASS(cel->blp[1] == NULL); cache_sort_vnodes(&blp1, &blp2); if (blp1 != NULL) { mtx_lock(blp1); cel->blp[0] = blp1; } mtx_lock(blp2); cel->blp[1] = blp2; } static void cache_unlock_buckets_cel(struct celockstate *cel) { if (cel->blp[0] != NULL) mtx_unlock(cel->blp[0]); mtx_unlock(cel->blp[1]); } /* * Lock part of the cache affected by the insertion. * * This means vnodelocks for dvp, vp and the relevant bucketlock. * However, insertion can result in removal of an old entry. In this * case we have an additional vnode and bucketlock pair to lock. * * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while * preserving the locking order (smaller address first). */ static void cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct mtx *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); if (vp == NULL || vp->v_type != VDIR) break; ncp = vp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == vp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; /* * All vnodes got re-locked. Re-validate the state and if * nothing changed we are done. Otherwise restart. */ if (ncp == vp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct mtx *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); ncp = dvp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == dvp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; if (ncp == dvp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_unlock(struct celockstate *cel) { cache_unlock_buckets_cel(cel); cache_unlock_vnodes_cel(cel); } static void __noinline cache_enter_dotdot_prep(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct celockstate cel; struct namecache *ncp; uint32_t hash; int len; if (dvp->v_cache_dd == NULL) return; len = cnp->cn_namelen; cache_celockstate_init(&cel); hash = cache_get_hash(cnp->cn_nameptr, len, dvp); cache_enter_lock_dd(&cel, dvp, vp, hash); vn_seqc_write_begin(dvp); ncp = dvp->v_cache_dd; if (ncp != NULL && (ncp->nc_flag & NCF_ISDOTDOT)) { KASSERT(ncp->nc_dvp == dvp, ("wrong isdotdot parent")); cache_zap_locked(ncp); } else { ncp = NULL; } dvp->v_cache_dd = NULL; vn_seqc_write_end(dvp); cache_enter_unlock(&cel); if (ncp != NULL) cache_free(ncp); } /* * Add an entry to the cache. */ void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp) { struct celockstate cel; struct namecache *ncp, *n2, *ndd; struct namecache_ts *ncp_ts; struct nchashhead *ncpp; uint32_t hash; int flag; int len; u_long lnumcache; VNPASS(!VN_IS_DOOMED(dvp), dvp); VNPASS(dvp->v_type != VNON, dvp); if (vp != NULL) { VNPASS(!VN_IS_DOOMED(vp), vp); VNPASS(vp->v_type != VNON, vp); } #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) return; #endif flag = 0; if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) return; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { cache_enter_dotdot_prep(dvp, vp, cnp); flag = NCF_ISDOTDOT; } } /* * Avoid blowout in namecache entries. */ lnumcache = atomic_fetchadd_long(&numcache, 1) + 1; if (__predict_false(lnumcache >= ncsize)) { atomic_subtract_long(&numcache, 1); counter_u64_add(numdrops, 1); return; } cache_celockstate_init(&cel); ndd = NULL; ncp_ts = NULL; /* * Calculate the hash key and setup as much of the new * namecache entry as possible before acquiring the lock. */ ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); ncp->nc_flag = flag | NCF_WIP; ncp->nc_vp = vp; if (vp == NULL) cache_negative_init(ncp); ncp->nc_dvp = dvp; if (tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); ncp_ts->nc_time = *tsp; ncp_ts->nc_ticks = ticks; ncp_ts->nc_nc.nc_flag |= NCF_TS; if (dtsp != NULL) { ncp_ts->nc_dotdottime = *dtsp; ncp_ts->nc_nc.nc_flag |= NCF_DTS; } } len = ncp->nc_nlen = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); memcpy(ncp->nc_name, cnp->cn_nameptr, len); ncp->nc_name[len] = '\0'; cache_enter_lock(&cel, dvp, vp, hash); /* * See if this vnode or negative entry is already in the cache * with this name. This can happen with concurrent lookups of * the same path name. */ ncpp = NCHHASH(hash); CK_SLIST_FOREACH(n2, ncpp, nc_hash) { if (n2->nc_dvp == dvp && n2->nc_nlen == cnp->cn_namelen && !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { MPASS(cache_ncp_canuse(n2)); if ((n2->nc_flag & NCF_NEGATIVE) != 0) KASSERT(vp == NULL, ("%s: found entry pointing to a different vnode (%p != %p)", __func__, NULL, vp)); else KASSERT(n2->nc_vp == vp, ("%s: found entry pointing to a different vnode (%p != %p)", __func__, n2->nc_vp, vp)); /* * Entries are supposed to be immutable unless in the * process of getting destroyed. Accommodating for * changing timestamps is possible but not worth it. * This should be harmless in terms of correctness, in * the worst case resulting in an earlier expiration. * Alternatively, the found entry can be replaced * altogether. */ MPASS((n2->nc_flag & (NCF_TS | NCF_DTS)) == (ncp->nc_flag & (NCF_TS | NCF_DTS))); #if 0 if (tsp != NULL) { KASSERT((n2->nc_flag & NCF_TS) != 0, ("no NCF_TS")); n2_ts = __containerof(n2, struct namecache_ts, nc_nc); n2_ts->nc_time = ncp_ts->nc_time; n2_ts->nc_ticks = ncp_ts->nc_ticks; if (dtsp != NULL) { n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; n2_ts->nc_nc.nc_flag |= NCF_DTS; } } #endif goto out_unlock_free; } } if (flag == NCF_ISDOTDOT) { /* * See if we are trying to add .. entry, but some other lookup * has populated v_cache_dd pointer already. */ if (dvp->v_cache_dd != NULL) goto out_unlock_free; KASSERT(vp == NULL || vp->v_type == VDIR, ("wrong vnode type %p", vp)); vn_seqc_write_begin(dvp); dvp->v_cache_dd = ncp; vn_seqc_write_end(dvp); } if (vp != NULL) { if (flag != NCF_ISDOTDOT) { /* * For this case, the cache entry maps both the * directory name in it and the name ".." for the * directory's parent. */ vn_seqc_write_begin(vp); if ((ndd = vp->v_cache_dd) != NULL) { if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) cache_zap_locked(ndd); else ndd = NULL; } vp->v_cache_dd = ncp; vn_seqc_write_end(vp); } else if (vp->v_type != VDIR) { if (vp->v_cache_dd != NULL) { vn_seqc_write_begin(vp); vp->v_cache_dd = NULL; vn_seqc_write_end(vp); } } } if (flag != NCF_ISDOTDOT) { if (LIST_EMPTY(&dvp->v_cache_src)) { vhold(dvp); counter_u64_add(numcachehv, 1); } LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); } /* * If the entry is "negative", we place it into the * "negative" cache queue, otherwise, we place it into the * destination vnode's cache entries queue. */ if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, vp); } else { if (cnp->cn_flags & ISWHITEOUT) ncp->nc_flag |= NCF_WHITE; cache_negative_insert(ncp); SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, ncp->nc_name); } /* * Insert the new namecache entry into the appropriate chain * within the cache entries table. */ CK_SLIST_INSERT_HEAD(ncpp, ncp, nc_hash); atomic_thread_fence_rel(); /* * Mark the entry as fully constructed. * It is immutable past this point until its removal. */ atomic_store_char(&ncp->nc_flag, ncp->nc_flag & ~NCF_WIP); cache_enter_unlock(&cel); if (numneg * ncnegfactor > lnumcache) cache_negative_zap_one(); if (ndd != NULL) cache_free(ndd); return; out_unlock_free: cache_enter_unlock(&cel); atomic_subtract_long(&numcache, 1); cache_free(ncp); return; } static u_int cache_roundup_2(u_int val) { u_int res; for (res = 1; res <= val; res <<= 1) continue; return (res); } static struct nchashhead * nchinittbl(u_long elements, u_long *hashmask) { struct nchashhead *hashtbl; u_long hashsize, i; hashsize = cache_roundup_2(elements) / 2; hashtbl = malloc((u_long)hashsize * sizeof(*hashtbl), M_VFSCACHE, M_WAITOK); for (i = 0; i < hashsize; i++) CK_SLIST_INIT(&hashtbl[i]); *hashmask = hashsize - 1; return (hashtbl); } static void ncfreetbl(struct nchashhead *hashtbl) { free(hashtbl, M_VFSCACHE); } /* * Name cache initialization, from vfs_init() when we are booting */ static void nchinit(void *dummy __unused) { u_int i; cache_zone_small = uma_zcreate("S VFS Cache", CACHE_ZONE_SMALL_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_small_ts = uma_zcreate("STS VFS Cache", CACHE_ZONE_SMALL_TS_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_large = uma_zcreate("L VFS Cache", CACHE_ZONE_LARGE_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); cache_zone_large_ts = uma_zcreate("LTS VFS Cache", CACHE_ZONE_LARGE_TS_SIZE, NULL, NULL, NULL, NULL, CACHE_ZONE_ALIGNMENT, UMA_ZONE_ZINIT); VFS_SMR_ZONE_SET(cache_zone_small); VFS_SMR_ZONE_SET(cache_zone_small_ts); VFS_SMR_ZONE_SET(cache_zone_large); VFS_SMR_ZONE_SET(cache_zone_large_ts); ncsize = desiredvnodes * ncsizefactor; nchashtbl = nchinittbl(desiredvnodes * 2, &nchash); ncbuckethash = cache_roundup_2(mp_ncpus * mp_ncpus) - 1; if (ncbuckethash < 7) /* arbitrarily chosen to avoid having one lock */ ncbuckethash = 7; if (ncbuckethash > nchash) ncbuckethash = nchash; bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numbucketlocks; i++) mtx_init(&bucketlocks[i], "ncbuc", NULL, MTX_DUPOK | MTX_RECURSE); ncvnodehash = ncbuckethash; vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numvnodelocks; i++) mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numneglists; i++) { mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); TAILQ_INIT(&neglists[i].nl_list); } mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF); TAILQ_INIT(&ncneg_hot.nl_list); mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); void cache_vnode_init(struct vnode *vp) { LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); vp->v_cache_dd = NULL; cache_prehash(vp); } void cache_changesize(u_long newmaxvnodes) { struct nchashhead *new_nchashtbl, *old_nchashtbl; u_long new_nchash, old_nchash; struct namecache *ncp; uint32_t hash; u_long newncsize; int i; newncsize = newmaxvnodes * ncsizefactor; newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); if (newmaxvnodes < numbucketlocks) newmaxvnodes = numbucketlocks; new_nchashtbl = nchinittbl(newmaxvnodes, &new_nchash); /* If same hash table size, nothing to do */ if (nchash == new_nchash) { ncfreetbl(new_nchashtbl); return; } /* * Move everything from the old hash table to the new table. * None of the namecache entries in the table can be removed * because to do so, they have to be removed from the hash table. */ cache_lock_all_vnodes(); cache_lock_all_buckets(); old_nchashtbl = nchashtbl; old_nchash = nchash; nchashtbl = new_nchashtbl; nchash = new_nchash; for (i = 0; i <= old_nchash; i++) { while ((ncp = CK_SLIST_FIRST(&old_nchashtbl[i])) != NULL) { hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); CK_SLIST_REMOVE(&old_nchashtbl[i], ncp, namecache, nc_hash); CK_SLIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); } } ncsize = newncsize; cache_unlock_all_buckets(); cache_unlock_all_vnodes(); ncfreetbl(old_nchashtbl); } /* * Invalidate all entries from and to a particular vnode. */ static void cache_purge_impl(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp, *vlp2; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); vlp2 = NULL; mtx_lock(vlp); retry: while (!LIST_EMPTY(&vp->v_cache_src)) { ncp = LIST_FIRST(&vp->v_cache_src); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } while (!TAILQ_EMPTY(&vp->v_cache_dst)) { ncp = TAILQ_FIRST(&vp->v_cache_dst); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } ncp = vp->v_cache_dd; if (ncp != NULL) { KASSERT(ncp->nc_flag & NCF_ISDOTDOT, ("lost dotdot link")); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); mtx_unlock(vlp); if (vlp2 != NULL) mtx_unlock(vlp2); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Opportunistic check to see if there is anything to do. */ static bool cache_has_entries(struct vnode *vp) { if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && vp->v_cache_dd == NULL) return (false); return (true); } void cache_purge(struct vnode *vp) { SDT_PROBE1(vfs, namecache, purge, done, vp); if (!cache_has_entries(vp)) return; cache_purge_impl(vp); } /* * Only to be used by vgone. */ void cache_purge_vgone(struct vnode *vp) { struct mtx *vlp; VNPASS(VN_IS_DOOMED(vp), vp); if (cache_has_entries(vp)) { cache_purge_impl(vp); return; } /* * Serialize against a potential thread doing cache_purge. */ vlp = VP2VNODELOCK(vp); mtx_wait_unlocked(vlp); if (cache_has_entries(vp)) { cache_purge_impl(vp); return; } return; } /* * Invalidate all negative entries for a particular directory vnode. */ void cache_purge_negative(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp; SDT_PROBE1(vfs, namecache, purge_negative, done, vp); if (LIST_EMPTY(&vp->v_cache_src)) return; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { if (!(ncp->nc_flag & NCF_NEGATIVE)) continue; cache_zap_negative_locked_vnode_kl(ncp, vp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } mtx_unlock(vlp); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } void cache_rename(struct vnode *fdvp, struct vnode *fvp, struct vnode *tdvp, struct vnode *tvp, struct componentname *fcnp, struct componentname *tcnp) { ASSERT_VOP_IN_SEQC(fdvp); ASSERT_VOP_IN_SEQC(fvp); ASSERT_VOP_IN_SEQC(tdvp); if (tvp != NULL) ASSERT_VOP_IN_SEQC(tvp); cache_purge(fvp); if (tvp != NULL) { cache_purge(tvp); KASSERT(!cache_remove_cnp(tdvp, tcnp), ("%s: lingering negative entry", __func__)); } else { cache_remove_cnp(tdvp, tcnp); } } /* * Flush all entries referencing a particular filesystem. */ void cache_purgevfs(struct mount *mp) { struct vnode *vp, *mvp; SDT_PROBE1(vfs, namecache, purgevfs, done, mp); /* * Somewhat wasteful iteration over all vnodes. Would be better to * support filtering and avoid the interlock to begin with. */ MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { if (!cache_has_entries(vp)) { VI_UNLOCK(vp); continue; } vholdl(vp); VI_UNLOCK(vp); cache_purge(vp); vdrop(vp); } } /* * Perform canonical checks and cache lookup and pass on to filesystem * through the vop_cachedlookup only if needed. */ int vfs_cache_lookup(struct vop_lookup_args *ap) { struct vnode *dvp; int error; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; int flags = cnp->cn_flags; *vpp = NULL; dvp = ap->a_dvp; if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); error = cache_lookup(dvp, vpp, cnp, NULL, NULL); if (error == 0) return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); if (error == -1) return (0); return (error); } /* Implementation of the getcwd syscall. */ int sys___getcwd(struct thread *td, struct __getcwd_args *uap) { char *buf, *retbuf; size_t buflen; int error; buflen = uap->buflen; if (__predict_false(buflen < 2)) return (EINVAL); if (buflen > MAXPATHLEN) buflen = MAXPATHLEN; buf = uma_zalloc(namei_zone, M_WAITOK); error = vn_getcwd(buf, &retbuf, &buflen); if (error == 0) error = copyout(retbuf, uap->buf, buflen); uma_zfree(namei_zone, buf); return (error); } int vn_getcwd(char *buf, char **retbuf, size_t *buflen) { struct pwd *pwd; int error; vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(pwd->pwd_cdir, pwd->pwd_rdir, buf, retbuf, buflen, false, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); error = vn_fullpath_any(pwd->pwd_cdir, pwd->pwd_rdir, buf, retbuf, buflen); pwd_drop(pwd); } #ifdef KTRACE if (KTRPOINT(curthread, KTR_NAMEI) && error == 0) ktrnamei(*retbuf); #endif return (error); } static int kern___realpathat(struct thread *td, int fd, const char *path, char *buf, size_t size, int flags, enum uio_seg pathseg) { struct nameidata nd; char *retbuf, *freebuf; int error; if (flags != 0) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | SAVENAME | WANTPARENT | AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights, td); if ((error = namei(&nd)) != 0) return (error); error = vn_fullpath_hardlink(&nd, &retbuf, &freebuf, &size); if (error == 0) { error = copyout(retbuf, buf, size); free(freebuf, M_TEMP); } NDFREE(&nd, 0); return (error); } int sys___realpathat(struct thread *td, struct __realpathat_args *uap) { return (kern___realpathat(td, uap->fd, uap->path, uap->buf, uap->size, uap->flags, UIO_USERSPACE)); } /* * Retrieve the full filesystem path that correspond to a vnode from the name * cache (if available) */ int vn_fullpath(struct vnode *vp, char **retbuf, char **freebuf) { struct pwd *pwd; char *buf; size_t buflen; int error; if (__predict_false(vp == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(vp, pwd->pwd_rdir, buf, retbuf, &buflen, false, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); error = vn_fullpath_any(vp, pwd->pwd_rdir, buf, retbuf, &buflen); pwd_drop(pwd); } if (error == 0) *freebuf = buf; else free(buf, M_TEMP); return (error); } /* * This function is similar to vn_fullpath, but it attempts to lookup the * pathname relative to the global root mount point. This is required for the * auditing sub-system, as audited pathnames must be absolute, relative to the * global root mount point. */ int vn_fullpath_global(struct vnode *vp, char **retbuf, char **freebuf) { char *buf; size_t buflen; int error; if (__predict_false(vp == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); vfs_smr_enter(); error = vn_fullpath_any_smr(vp, rootvnode, buf, retbuf, &buflen, false, 0); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { error = vn_fullpath_any(vp, rootvnode, buf, retbuf, &buflen); } if (error == 0) *freebuf = buf; else free(buf, M_TEMP); return (error); } static struct namecache * vn_dd_from_dst(struct vnode *vp) { struct namecache *ncp; cache_assert_vnode_locked(vp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) return (ncp); } return (NULL); } int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, size_t *buflen) { struct vnode *dvp; struct namecache *ncp; struct mtx *vlp; int error; vlp = VP2VNODELOCK(*vp); mtx_lock(vlp); ncp = (*vp)->v_cache_dd; if (ncp != NULL && (ncp->nc_flag & NCF_ISDOTDOT) == 0) { KASSERT(ncp == vn_dd_from_dst(*vp), ("%s: mismatch for dd entry (%p != %p)", __func__, ncp, vn_dd_from_dst(*vp))); } else { ncp = vn_dd_from_dst(*vp); } if (ncp != NULL) { if (*buflen < ncp->nc_nlen) { mtx_unlock(vlp); vrele(*vp); counter_u64_add(numfullpathfail4, 1); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, ncp->nc_name, vp); dvp = *vp; *vp = ncp->nc_dvp; vref(*vp); mtx_unlock(vlp); vrele(dvp); return (0); } SDT_PROBE1(vfs, namecache, fullpath, miss, vp); mtx_unlock(vlp); vn_lock(*vp, LK_SHARED | LK_RETRY); error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); vput(*vp); if (error) { counter_u64_add(numfullpathfail2, 1); SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *vp = dvp; if (VN_IS_DOOMED(dvp)) { /* forced unmount */ vrele(dvp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } /* * *vp has its use count incremented still. */ return (0); } /* * Resolve a directory to a pathname. * * The name of the directory can always be found in the namecache or fetched * from the filesystem. There is also guaranteed to be only one parent, meaning * we can just follow vnodes up until we find the root. * * The vnode must be referenced. */ static int vn_fullpath_dir(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, bool slash_prefixed, size_t addend) { #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *vp1; size_t buflen; int error; VNPASS(vp->v_type == VDIR || VN_IS_DOOMED(vp), vp); VNPASS(vp->v_usecount > 0, vp); buflen = *len; if (!slash_prefixed) { MPASS(*len >= 2); buflen--; buf[buflen] = '\0'; } error = 0; SDT_PROBE1(vfs, namecache, fullpath, entry, vp); counter_u64_add(numfullpathcalls, 1); while (vp != rdir && vp != rootvnode) { /* * The vp vnode must be already fully constructed, * since it is either found in namecache or obtained * from VOP_VPTOCNP(). We may test for VV_ROOT safely * without obtaining the vnode lock. */ if ((vp->v_vflag & VV_ROOT) != 0) { vn_lock(vp, LK_RETRY | LK_SHARED); /* * With the vnode locked, check for races with * unmount, forced or not. Note that we * already verified that vp is not equal to * the root vnode, which means that * mnt_vnodecovered can be NULL only for the * case of unmount. */ if (VN_IS_DOOMED(vp) || (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || vp1->v_mountedhere != vp->v_mount) { vput(vp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } vref(vp1); vput(vp); vp = vp1; continue; } if (vp->v_type != VDIR) { vrele(vp); counter_u64_add(numfullpathfail1, 1); error = ENOTDIR; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } error = vn_vptocnp(&vp, curthread->td_ucred, buf, &buflen); if (error) break; if (buflen == 0) { vrele(vp); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, startvp, NULL); break; } buf[--buflen] = '/'; slash_prefixed = true; } if (error) return (error); if (!slash_prefixed) { if (buflen == 0) { vrele(vp); counter_u64_add(numfullpathfail4, 1); SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, startvp, NULL); return (ENOMEM); } buf[--buflen] = '/'; } counter_u64_add(numfullpathfound, 1); vrele(vp); *retbuf = buf + buflen; SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, *retbuf); *len -= buflen; *len += addend; return (0); } /* * Resolve an arbitrary vnode to a pathname. * * Note 2 caveats: * - hardlinks are not tracked, thus if the vnode is not a directory this can * resolve to a different path than the one used to find it * - namecache is not mandatory, meaning names are not guaranteed to be added * (in which case resolving fails) */ static void __inline cache_rev_failed_impl(int *reason, int line) { *reason = line; } #define cache_rev_failed(var) cache_rev_failed_impl((var), __LINE__) static int vn_fullpath_any_smr(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen, bool slash_prefixed, size_t addend) { #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *tvp; struct mount *mp; struct namecache *ncp; size_t orig_buflen; int reason; int error; #ifdef KDTRACE_HOOKS int i; #endif seqc_t vp_seqc, tvp_seqc; u_char nc_flag; VFS_SMR_ASSERT_ENTERED(); if (!cache_fast_revlookup) { vfs_smr_exit(); return (-1); } orig_buflen = *buflen; if (!slash_prefixed) { MPASS(*buflen >= 2); *buflen -= 1; buf[*buflen] = '\0'; } if (vp == rdir || vp == rootvnode) { if (!slash_prefixed) { *buflen -= 1; buf[*buflen] = '/'; } goto out_ok; } #ifdef KDTRACE_HOOKS i = 0; #endif error = -1; ncp = NULL; /* for sdt probe down below */ vp_seqc = vn_seqc_read_any(vp); if (seqc_in_modify(vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } for (;;) { #ifdef KDTRACE_HOOKS i++; #endif if ((vp->v_vflag & VV_ROOT) != 0) { mp = atomic_load_ptr(&vp->v_mount); if (mp == NULL) { cache_rev_failed(&reason); goto out_abort; } tvp = atomic_load_ptr(&mp->mnt_vnodecovered); tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(tvp_seqc)) { cache_rev_failed(&reason); goto out_abort; } if (!vn_seqc_consistent(vp, vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } vp = tvp; vp_seqc = tvp_seqc; continue; } ncp = atomic_load_ptr(&vp->v_cache_dd); if (ncp == NULL) { cache_rev_failed(&reason); goto out_abort; } nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) != 0) { cache_rev_failed(&reason); goto out_abort; } if (!cache_ncp_canuse(ncp)) { cache_rev_failed(&reason); goto out_abort; } if (ncp->nc_nlen >= *buflen) { cache_rev_failed(&reason); error = ENOMEM; goto out_abort; } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); *buflen -= 1; buf[*buflen] = '/'; tvp = ncp->nc_dvp; tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(tvp_seqc)) { cache_rev_failed(&reason); goto out_abort; } if (!vn_seqc_consistent(vp, vp_seqc)) { cache_rev_failed(&reason); goto out_abort; } vp = tvp; vp_seqc = tvp_seqc; if (vp == rdir || vp == rootvnode) break; } out_ok: vfs_smr_exit(); *retbuf = buf + *buflen; *buflen = orig_buflen - *buflen + addend; SDT_PROBE2(vfs, namecache, fullpath_smr, hit, startvp, *retbuf); return (0); out_abort: *buflen = orig_buflen; SDT_PROBE4(vfs, namecache, fullpath_smr, miss, startvp, ncp, reason, i); vfs_smr_exit(); return (error); } static int vn_fullpath_any(struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen) { size_t orig_buflen; bool slash_prefixed; int error; if (*buflen < 2) return (EINVAL); orig_buflen = *buflen; vref(vp); slash_prefixed = false; if (vp->v_type != VDIR) { *buflen -= 1; buf[*buflen] = '\0'; error = vn_vptocnp(&vp, curthread->td_ucred, buf, buflen); if (error) return (error); if (*buflen == 0) { vrele(vp); return (ENOMEM); } *buflen -= 1; buf[*buflen] = '/'; slash_prefixed = true; } return (vn_fullpath_dir(vp, rdir, buf, retbuf, buflen, slash_prefixed, orig_buflen - *buflen)); } /* * Resolve an arbitrary vnode to a pathname (taking care of hardlinks). * * Since the namecache does not track handlings, the caller is expected to first * look up the target vnode with SAVENAME | WANTPARENT flags passed to namei. * * Then we have 2 cases: * - if the found vnode is a directory, the path can be constructed just by * fullowing names up the chain * - otherwise we populate the buffer with the saved name and start resolving * from the parent */ static int vn_fullpath_hardlink(struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen) { char *buf, *tmpbuf; struct pwd *pwd; struct componentname *cnp; struct vnode *vp; size_t addend; int error; bool slash_prefixed; enum vtype type; if (*buflen < 2) return (EINVAL); if (*buflen > MAXPATHLEN) *buflen = MAXPATHLEN; slash_prefixed = false; buf = malloc(*buflen, M_TEMP, M_WAITOK); addend = 0; vp = ndp->ni_vp; /* * Check for VBAD to work around the vp_crossmp bug in lookup(). * * For example consider tmpfs on /tmp and realpath /tmp. ni_vp will be * set to mount point's root vnode while ni_dvp will be vp_crossmp. * If the type is VDIR (like in this very case) we can skip looking * at ni_dvp in the first place. However, since vnodes get passed here * unlocked the target may transition to doomed state (type == VBAD) * before we get to evaluate the condition. If this happens, we will * populate part of the buffer and descend to vn_fullpath_dir with * vp == vp_crossmp. Prevent the problem by checking for VBAD. * * This should be atomic_load(&vp->v_type) but it is ilegal to take * an address of a bit field, even if said field is sized to char. * Work around the problem by reading the value into a full-sized enum * and then re-reading it with atomic_load which will still prevent * the compiler from re-reading down the road. */ type = vp->v_type; type = atomic_load_int(&type); if (type == VBAD) { error = ENOENT; goto out_bad; } if (type != VDIR) { cnp = &ndp->ni_cnd; addend = cnp->cn_namelen + 2; if (*buflen < addend) { error = ENOMEM; goto out_bad; } *buflen -= addend; tmpbuf = buf + *buflen; tmpbuf[0] = '/'; memcpy(&tmpbuf[1], cnp->cn_nameptr, cnp->cn_namelen); tmpbuf[addend - 1] = '\0'; slash_prefixed = true; vp = ndp->ni_dvp; } vfs_smr_enter(); pwd = pwd_get_smr(); error = vn_fullpath_any_smr(vp, pwd->pwd_rdir, buf, retbuf, buflen, slash_prefixed, addend); VFS_SMR_ASSERT_NOT_ENTERED(); if (error < 0) { pwd = pwd_hold(curthread); vref(vp); error = vn_fullpath_dir(vp, pwd->pwd_rdir, buf, retbuf, buflen, slash_prefixed, addend); pwd_drop(pwd); if (error != 0) goto out_bad; } *freebuf = buf; return (0); out_bad: free(buf, M_TEMP); return (error); } struct vnode * vn_dir_dd_ino(struct vnode *vp) { struct namecache *ncp; struct vnode *ddvp; struct mtx *vlp; enum vgetstate vs; ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) continue; ddvp = ncp->nc_dvp; vs = vget_prep(ddvp); mtx_unlock(vlp); if (vget_finish(ddvp, LK_SHARED | LK_NOWAIT, vs)) return (NULL); return (ddvp); } mtx_unlock(vlp); return (NULL); } int vn_commname(struct vnode *vp, char *buf, u_int buflen) { struct namecache *ncp; struct mtx *vlp; int l; vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; if (ncp == NULL) { mtx_unlock(vlp); return (ENOENT); } l = min(ncp->nc_nlen, buflen - 1); memcpy(buf, ncp->nc_name, l); mtx_unlock(vlp); buf[l] = '\0'; return (0); } /* * This function updates path string to vnode's full global path * and checks the size of the new path string against the pathlen argument. * * Requires a locked, referenced vnode. * Vnode is re-locked on success or ENODEV, otherwise unlocked. * * If vp is a directory, the call to vn_fullpath_global() always succeeds * because it falls back to the ".." lookup if the namecache lookup fails. */ int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen) { struct nameidata nd; struct vnode *vp1; char *rpath, *fbuf; int error; ASSERT_VOP_ELOCKED(vp, __func__); /* Construct global filesystem path from vp. */ VOP_UNLOCK(vp); error = vn_fullpath_global(vp, &rpath, &fbuf); if (error != 0) { vrele(vp); return (error); } if (strlen(rpath) >= pathlen) { vrele(vp); error = ENAMETOOLONG; goto out; } /* * Re-lookup the vnode by path to detect a possible rename. * As a side effect, the vnode is relocked. * If vnode was renamed, return ENOENT. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, path, td); error = namei(&nd); if (error != 0) { vrele(vp); goto out; } NDFREE(&nd, NDF_ONLY_PNBUF); vp1 = nd.ni_vp; vrele(vp); if (vp1 == vp) strcpy(path, rpath); else { vput(vp1); error = ENOENT; } out: free(fbuf, M_TEMP); return (error); } #ifdef DDB static void db_print_vpath(struct vnode *vp) { while (vp != NULL) { db_printf("%p: ", vp); if (vp == rootvnode) { db_printf("/"); vp = NULL; } else { if (vp->v_vflag & VV_ROOT) { db_printf(""); vp = vp->v_mount->mnt_vnodecovered; } else { struct namecache *ncp; char *ncn; int i; ncp = TAILQ_FIRST(&vp->v_cache_dst); if (ncp != NULL) { ncn = ncp->nc_name; for (i = 0; i < ncp->nc_nlen; i++) db_printf("%c", *ncn++); vp = ncp->nc_dvp; } else { vp = NULL; } } } db_printf("\n"); } return; } DB_SHOW_COMMAND(vpath, db_show_vpath) { struct vnode *vp; if (!have_addr) { db_printf("usage: show vpath \n"); return; } vp = (struct vnode *)addr; db_print_vpath(vp); } #endif static bool __read_frequently cache_fast_lookup = true; SYSCTL_BOOL(_vfs, OID_AUTO, cache_fast_lookup, CTLFLAG_RW, &cache_fast_lookup, 0, ""); #define CACHE_FPL_FAILED -2020 static void cache_fpl_cleanup_cnp(struct componentname *cnp) { uma_zfree(namei_zone, cnp->cn_pnbuf); #ifdef DIAGNOSTIC cnp->cn_pnbuf = NULL; cnp->cn_nameptr = NULL; #endif } static void cache_fpl_handle_root(struct nameidata *ndp, struct vnode **dpp) { struct componentname *cnp; cnp = &ndp->ni_cnd; while (*(cnp->cn_nameptr) == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } *dpp = ndp->ni_rootdir; } /* * Components of nameidata (or objects it can point to) which may * need restoring in case fast path lookup fails. */ struct nameidata_saved { long cn_namelen; char *cn_nameptr; size_t ni_pathlen; int cn_flags; }; struct cache_fpl { struct nameidata *ndp; struct componentname *cnp; struct pwd *pwd; struct vnode *dvp; struct vnode *tvp; seqc_t dvp_seqc; seqc_t tvp_seqc; struct nameidata_saved snd; int line; enum cache_fpl_status status:8; bool in_smr; + bool fsearch; }; static void cache_fpl_checkpoint(struct cache_fpl *fpl, struct nameidata_saved *snd) { snd->cn_flags = fpl->ndp->ni_cnd.cn_flags; snd->cn_namelen = fpl->ndp->ni_cnd.cn_namelen; snd->cn_nameptr = fpl->ndp->ni_cnd.cn_nameptr; snd->ni_pathlen = fpl->ndp->ni_pathlen; } static void cache_fpl_restore(struct cache_fpl *fpl, struct nameidata_saved *snd) { fpl->ndp->ni_cnd.cn_flags = snd->cn_flags; fpl->ndp->ni_cnd.cn_namelen = snd->cn_namelen; fpl->ndp->ni_cnd.cn_nameptr = snd->cn_nameptr; fpl->ndp->ni_pathlen = snd->ni_pathlen; } #ifdef INVARIANTS #define cache_fpl_smr_assert_entered(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == true); \ VFS_SMR_ASSERT_ENTERED(); \ }) #define cache_fpl_smr_assert_not_entered(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == false); \ VFS_SMR_ASSERT_NOT_ENTERED(); \ }) #else #define cache_fpl_smr_assert_entered(fpl) do { } while (0) #define cache_fpl_smr_assert_not_entered(fpl) do { } while (0) #endif #define cache_fpl_smr_enter_initial(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ vfs_smr_enter(); \ _fpl->in_smr = true; \ }) #define cache_fpl_smr_enter(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == false); \ vfs_smr_enter(); \ _fpl->in_smr = true; \ }) #define cache_fpl_smr_exit(fpl) ({ \ struct cache_fpl *_fpl = (fpl); \ MPASS(_fpl->in_smr == true); \ vfs_smr_exit(); \ _fpl->in_smr = false; \ }) static int cache_fpl_aborted_impl(struct cache_fpl *fpl, int line) { if (fpl->status != CACHE_FPL_STATUS_UNSET) { KASSERT(fpl->status == CACHE_FPL_STATUS_PARTIAL, ("%s: converting to abort from %d at %d, set at %d\n", __func__, fpl->status, line, fpl->line)); } fpl->status = CACHE_FPL_STATUS_ABORTED; fpl->line = line; return (CACHE_FPL_FAILED); } #define cache_fpl_aborted(x) cache_fpl_aborted_impl((x), __LINE__) static int cache_fpl_partial_impl(struct cache_fpl *fpl, int line) { KASSERT(fpl->status == CACHE_FPL_STATUS_UNSET, ("%s: setting to partial at %d, but already set to %d at %d\n", __func__, line, fpl->status, fpl->line)); cache_fpl_smr_assert_entered(fpl); fpl->status = CACHE_FPL_STATUS_PARTIAL; fpl->line = line; return (CACHE_FPL_FAILED); } #define cache_fpl_partial(x) cache_fpl_partial_impl((x), __LINE__) static int cache_fpl_handled_impl(struct cache_fpl *fpl, int error, int line) { KASSERT(fpl->status == CACHE_FPL_STATUS_UNSET, ("%s: setting to handled at %d, but already set to %d at %d\n", __func__, line, fpl->status, fpl->line)); cache_fpl_smr_assert_not_entered(fpl); MPASS(error != CACHE_FPL_FAILED); fpl->status = CACHE_FPL_STATUS_HANDLED; fpl->line = line; return (error); } #define cache_fpl_handled(x, e) cache_fpl_handled_impl((x), (e), __LINE__) #define CACHE_FPL_SUPPORTED_CN_FLAGS \ (LOCKLEAF | LOCKPARENT | WANTPARENT | NOCACHE | FOLLOW | LOCKSHARED | SAVENAME | \ SAVESTART | WILLBEDIR | ISOPEN | NOMACCHECK | AUDITVNODE1 | AUDITVNODE2 | NOCAPCHECK) #define CACHE_FPL_INTERNAL_CN_FLAGS \ (ISDOTDOT | MAKEENTRY | ISLASTCN) _Static_assert((CACHE_FPL_SUPPORTED_CN_FLAGS & CACHE_FPL_INTERNAL_CN_FLAGS) == 0, "supported and internal flags overlap"); static bool cache_fpl_islastcn(struct nameidata *ndp) { return (*ndp->ni_next == 0); } static bool cache_fpl_isdotdot(struct componentname *cnp) { if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.' && cnp->cn_nameptr[0] == '.') return (true); return (false); } static bool cache_can_fplookup(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct thread *td; ndp = fpl->ndp; cnp = fpl->cnp; td = cnp->cn_thread; if (!cache_fast_lookup) { cache_fpl_aborted(fpl); return (false); } #ifdef MAC if (mac_vnode_check_lookup_enabled()) { cache_fpl_aborted(fpl); return (false); } #endif if ((cnp->cn_flags & ~CACHE_FPL_SUPPORTED_CN_FLAGS) != 0) { cache_fpl_aborted(fpl); return (false); } - if (ndp->ni_dirfd != AT_FDCWD) { - cache_fpl_aborted(fpl); - return (false); - } if (IN_CAPABILITY_MODE(td)) { cache_fpl_aborted(fpl); return (false); } if (AUDITING_TD(td)) { cache_fpl_aborted(fpl); return (false); } if (ndp->ni_startdir != NULL) { cache_fpl_aborted(fpl); return (false); } return (true); } +static int +cache_fplookup_dirfd(struct cache_fpl *fpl, struct vnode **vpp) +{ + struct nameidata *ndp; + int error; + bool fsearch; + + ndp = fpl->ndp; + error = fgetvp_lookup_smr(ndp->ni_dirfd, ndp, vpp, &fsearch); + if (__predict_false(error != 0)) { + cache_fpl_smr_exit(fpl); + return (cache_fpl_aborted(fpl)); + } + fpl->fsearch = fsearch; + return (0); +} + static bool cache_fplookup_vnode_supported(struct vnode *vp) { return (vp->v_type != VLNK); } /* * Move a negative entry to the hot list. * * We have to take locks, but they may be contended and in the worst * case we may need to go off CPU. We don't want to spin within the * smr section and we can't block with it. Instead we are going to * look up the entry again. */ static int __noinline cache_fplookup_negative_promote(struct cache_fpl *fpl, struct namecache *oncp, uint32_t hash) { struct componentname *cnp; struct namecache *ncp; struct neglist *neglist; struct negstate *negstate; struct vnode *dvp; u_char nc_flag; cnp = fpl->cnp; dvp = fpl->dvp; if (!vhold_smr(dvp)) return (cache_fpl_aborted(fpl)); neglist = NCP2NEGLIST(oncp); cache_fpl_smr_exit(fpl); mtx_lock(&ncneg_hot.nl_lock); mtx_lock(&neglist->nl_lock); /* * For hash iteration. */ cache_fpl_smr_enter(fpl); /* * Avoid all surprises by only succeeding if we got the same entry and * bailing completely otherwise. * * In particular at this point there can be a new ncp which matches the * search but hashes to a different neglist. */ CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp == oncp) break; } /* * No match to begin with. */ if (__predict_false(ncp == NULL)) { goto out_abort; } /* * The newly found entry may be something different... */ if (!(ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen))) { goto out_abort; } /* * ... and not even negative. */ nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_NEGATIVE) == 0) { goto out_abort; } if (__predict_false(!cache_ncp_canuse(ncp))) { goto out_abort; } negstate = NCP2NEGSTATE(ncp); if ((negstate->neg_flag & NEG_HOT) == 0) { numhotneg++; TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); negstate->neg_flag |= NEG_HOT; } SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); counter_u64_add(numneghits, 1); cache_fpl_smr_exit(fpl); mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); vdrop(dvp); return (cache_fpl_handled(fpl, ENOENT)); out_abort: cache_fpl_smr_exit(fpl); mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); vdrop(dvp); return (cache_fpl_aborted(fpl)); } /* * The target vnode is not supported, prepare for the slow path to take over. */ static int __noinline cache_fplookup_partial_setup(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; enum vgetstate dvs; struct vnode *dvp; struct pwd *pwd; seqc_t dvp_seqc; ndp = fpl->ndp; cnp = fpl->cnp; pwd = fpl->pwd; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; if (!pwd_hold_smr(pwd)) { cache_fpl_smr_exit(fpl); return (cache_fpl_aborted(fpl)); } dvs = vget_prep_smr(dvp); cache_fpl_smr_exit(fpl); if (__predict_false(dvs == VGET_NONE)) { pwd_drop(pwd); return (cache_fpl_aborted(fpl)); } vget_finish_ref(dvp, dvs); if (!vn_seqc_consistent(dvp, dvp_seqc)) { vrele(dvp); pwd_drop(pwd); return (cache_fpl_aborted(fpl)); } cache_fpl_restore(fpl, &fpl->snd); ndp->ni_startdir = dvp; cnp->cn_flags |= MAKEENTRY; if (cache_fpl_islastcn(ndp)) cnp->cn_flags |= ISLASTCN; if (cache_fpl_isdotdot(cnp)) cnp->cn_flags |= ISDOTDOT; return (0); } static int cache_fplookup_final_child(struct cache_fpl *fpl, enum vgetstate tvs) { struct componentname *cnp; struct vnode *tvp; seqc_t tvp_seqc; int error, lkflags; cnp = fpl->cnp; tvp = fpl->tvp; tvp_seqc = fpl->tvp_seqc; if ((cnp->cn_flags & LOCKLEAF) != 0) { lkflags = LK_SHARED; if ((cnp->cn_flags & LOCKSHARED) == 0) lkflags = LK_EXCLUSIVE; error = vget_finish(tvp, lkflags, tvs); if (__predict_false(error != 0)) { return (cache_fpl_aborted(fpl)); } } else { vget_finish_ref(tvp, tvs); } if (!vn_seqc_consistent(tvp, tvp_seqc)) { if ((cnp->cn_flags & LOCKLEAF) != 0) vput(tvp); else vrele(tvp); return (cache_fpl_aborted(fpl)); } return (cache_fpl_handled(fpl, 0)); } /* * They want to possibly modify the state of the namecache. * * Don't try to match the API contract, just leave. * TODO: this leaves scalability on the table */ static int cache_fplookup_final_modifying(struct cache_fpl *fpl) { struct componentname *cnp; cnp = fpl->cnp; MPASS(cnp->cn_nameiop != LOOKUP); return (cache_fpl_partial(fpl)); } static int __noinline cache_fplookup_final_withparent(struct cache_fpl *fpl) { struct componentname *cnp; enum vgetstate dvs, tvs; struct vnode *dvp, *tvp; seqc_t dvp_seqc; int error; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; tvp = fpl->tvp; MPASS((cnp->cn_flags & (LOCKPARENT|WANTPARENT)) != 0); /* * This is less efficient than it can be for simplicity. */ dvs = vget_prep_smr(dvp); if (__predict_false(dvs == VGET_NONE)) { return (cache_fpl_aborted(fpl)); } tvs = vget_prep_smr(tvp); if (__predict_false(tvs == VGET_NONE)) { cache_fpl_smr_exit(fpl); vget_abort(dvp, dvs); return (cache_fpl_aborted(fpl)); } cache_fpl_smr_exit(fpl); if ((cnp->cn_flags & LOCKPARENT) != 0) { error = vget_finish(dvp, LK_EXCLUSIVE, dvs); if (__predict_false(error != 0)) { vget_abort(tvp, tvs); return (cache_fpl_aborted(fpl)); } } else { vget_finish_ref(dvp, dvs); } if (!vn_seqc_consistent(dvp, dvp_seqc)) { vget_abort(tvp, tvs); if ((cnp->cn_flags & LOCKPARENT) != 0) vput(dvp); else vrele(dvp); return (cache_fpl_aborted(fpl)); } error = cache_fplookup_final_child(fpl, tvs); if (__predict_false(error != 0)) { MPASS(fpl->status == CACHE_FPL_STATUS_ABORTED); if ((cnp->cn_flags & LOCKPARENT) != 0) vput(dvp); else vrele(dvp); return (error); } MPASS(fpl->status == CACHE_FPL_STATUS_HANDLED); return (0); } static int cache_fplookup_final(struct cache_fpl *fpl) { struct componentname *cnp; enum vgetstate tvs; struct vnode *dvp, *tvp; seqc_t dvp_seqc; cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; tvp = fpl->tvp; VNPASS(cache_fplookup_vnode_supported(dvp), dvp); if (cnp->cn_nameiop != LOOKUP) { return (cache_fplookup_final_modifying(fpl)); } if ((cnp->cn_flags & (LOCKPARENT|WANTPARENT)) != 0) return (cache_fplookup_final_withparent(fpl)); tvs = vget_prep_smr(tvp); if (__predict_false(tvs == VGET_NONE)) { return (cache_fpl_partial(fpl)); } if (!vn_seqc_consistent(dvp, dvp_seqc)) { cache_fpl_smr_exit(fpl); vget_abort(tvp, tvs); return (cache_fpl_aborted(fpl)); } cache_fpl_smr_exit(fpl); return (cache_fplookup_final_child(fpl, tvs)); } static int __noinline cache_fplookup_dot(struct cache_fpl *fpl) { struct vnode *dvp; dvp = fpl->dvp; fpl->tvp = dvp; fpl->tvp_seqc = vn_seqc_read_any(dvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_aborted(fpl)); } counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", dvp); return (0); } static int __noinline cache_fplookup_dotdot(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct namecache *ncp; struct vnode *dvp; struct prison *pr; u_char nc_flag; ndp = fpl->ndp; cnp = fpl->cnp; dvp = fpl->dvp; /* * XXX this is racy the same way regular lookup is */ for (pr = cnp->cn_cred->cr_prison; pr != NULL; pr = pr->pr_parent) if (dvp == pr->pr_root) break; if (dvp == ndp->ni_rootdir || dvp == ndp->ni_topdir || dvp == rootvnode || pr != NULL) { fpl->tvp = dvp; fpl->tvp_seqc = vn_seqc_read_any(dvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_aborted(fpl)); } return (0); } if ((dvp->v_vflag & VV_ROOT) != 0) { /* * TODO * The opposite of climb mount is needed here. */ return (cache_fpl_aborted(fpl)); } ncp = atomic_load_ptr(&dvp->v_cache_dd); if (ncp == NULL) { return (cache_fpl_aborted(fpl)); } nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_ISDOTDOT) != 0) { if ((nc_flag & NCF_NEGATIVE) != 0) return (cache_fpl_aborted(fpl)); fpl->tvp = ncp->nc_vp; } else { fpl->tvp = ncp->nc_dvp; } if (__predict_false(!cache_ncp_canuse(ncp))) { return (cache_fpl_aborted(fpl)); } fpl->tvp_seqc = vn_seqc_read_any(fpl->tvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_partial(fpl)); } counter_u64_add(dotdothits, 1); return (0); } static int cache_fplookup_next(struct cache_fpl *fpl) { struct componentname *cnp; struct namecache *ncp; struct negstate *negstate; struct vnode *dvp, *tvp; u_char nc_flag; uint32_t hash; bool neg_hot; cnp = fpl->cnp; dvp = fpl->dvp; if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')) { return (cache_fplookup_dot(fpl)); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); CK_SLIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } /* * If there is no entry we have to punt to the slow path to perform * actual lookup. Should there be nothing with this name a negative * entry will be created. */ if (__predict_false(ncp == NULL)) { return (cache_fpl_partial(fpl)); } tvp = atomic_load_ptr(&ncp->nc_vp); nc_flag = atomic_load_char(&ncp->nc_flag); if ((nc_flag & NCF_NEGATIVE) != 0) { /* * If they want to create an entry we need to replace this one. */ if (__predict_false(fpl->cnp->cn_nameiop != LOOKUP)) { return (cache_fpl_partial(fpl)); } negstate = NCP2NEGSTATE(ncp); neg_hot = ((negstate->neg_flag & NEG_HOT) != 0); if (__predict_false(!cache_ncp_canuse(ncp))) { return (cache_fpl_partial(fpl)); } if (__predict_false((nc_flag & NCF_WHITE) != 0)) { return (cache_fpl_partial(fpl)); } if (!neg_hot) { return (cache_fplookup_negative_promote(fpl, ncp, hash)); } SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); counter_u64_add(numneghits, 1); cache_fpl_smr_exit(fpl); return (cache_fpl_handled(fpl, ENOENT)); } if (__predict_false(!cache_ncp_canuse(ncp))) { return (cache_fpl_partial(fpl)); } fpl->tvp = tvp; fpl->tvp_seqc = vn_seqc_read_any(tvp); if (seqc_in_modify(fpl->tvp_seqc)) { return (cache_fpl_partial(fpl)); } if (!cache_fplookup_vnode_supported(tvp)) { return (cache_fpl_partial(fpl)); } counter_u64_add(numposhits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, tvp); return (0); } static bool cache_fplookup_mp_supported(struct mount *mp) { if (mp == NULL) return (false); if ((mp->mnt_kern_flag & MNTK_FPLOOKUP) == 0) return (false); return (true); } /* * Walk up the mount stack (if any). * * Correctness is provided in the following ways: * - all vnodes are protected from freeing with SMR * - struct mount objects are type stable making them always safe to access * - stability of the particular mount is provided by busying it * - relationship between the vnode which is mounted on and the mount is * verified with the vnode sequence counter after busying * - association between root vnode of the mount and the mount is protected * by busy * * From that point on we can read the sequence counter of the root vnode * and get the next mount on the stack (if any) using the same protection. * * By the end of successful walk we are guaranteed the reached state was * indeed present at least at some point which matches the regular lookup. */ static int __noinline cache_fplookup_climb_mount(struct cache_fpl *fpl) { struct mount *mp, *prev_mp; struct vnode *vp; seqc_t vp_seqc; vp = fpl->tvp; vp_seqc = fpl->tvp_seqc; VNPASS(vp->v_type == VDIR || vp->v_type == VBAD, vp); mp = atomic_load_ptr(&vp->v_mountedhere); if (mp == NULL) return (0); prev_mp = NULL; for (;;) { if (!vfs_op_thread_enter_crit(mp)) { if (prev_mp != NULL) vfs_op_thread_exit_crit(prev_mp); return (cache_fpl_partial(fpl)); } if (prev_mp != NULL) vfs_op_thread_exit_crit(prev_mp); if (!vn_seqc_consistent(vp, vp_seqc)) { vfs_op_thread_exit_crit(mp); return (cache_fpl_partial(fpl)); } if (!cache_fplookup_mp_supported(mp)) { vfs_op_thread_exit_crit(mp); return (cache_fpl_partial(fpl)); } vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL || VN_IS_DOOMED(vp)) { vfs_op_thread_exit_crit(mp); return (cache_fpl_partial(fpl)); } vp_seqc = vn_seqc_read_any(vp); if (seqc_in_modify(vp_seqc)) { vfs_op_thread_exit_crit(mp); return (cache_fpl_partial(fpl)); } prev_mp = mp; mp = atomic_load_ptr(&vp->v_mountedhere); if (mp == NULL) break; } vfs_op_thread_exit_crit(prev_mp); fpl->tvp = vp; fpl->tvp_seqc = vp_seqc; return (0); } static bool cache_fplookup_need_climb_mount(struct cache_fpl *fpl) { struct mount *mp; struct vnode *vp; vp = fpl->tvp; /* * Hack: while this is a union, the pointer tends to be NULL so save on * a branch. */ mp = atomic_load_ptr(&vp->v_mountedhere); if (mp == NULL) return (false); if (vp->v_type == VDIR) return (true); return (false); } /* * Parse the path. * * The code was originally copy-pasted from regular lookup and despite * clean ups leaves performance on the table. Any modifications here * must take into account that in case off fallback the resulting * nameidata state has to be compatible with the original. */ static int cache_fplookup_parse(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; char *cp; ndp = fpl->ndp; cnp = fpl->cnp; /* * Search a new directory. * * The last component of the filename is left accessible via * cnp->cn_nameptr for callers that need the name. Callers needing * the name set the SAVENAME flag. When done, they assume * responsibility for freeing the pathname buffer. */ for (cp = cnp->cn_nameptr; *cp != 0 && *cp != '/'; cp++) continue; cnp->cn_namelen = cp - cnp->cn_nameptr; if (__predict_false(cnp->cn_namelen > NAME_MAX)) { cache_fpl_smr_exit(fpl); return (cache_fpl_handled(fpl, ENAMETOOLONG)); } ndp->ni_pathlen -= cnp->cn_namelen; KASSERT(ndp->ni_pathlen <= PATH_MAX, ("%s: ni_pathlen underflow to %zd\n", __func__, ndp->ni_pathlen)); ndp->ni_next = cp; /* * Replace multiple slashes by a single slash and trailing slashes * by a null. This must be done before VOP_LOOKUP() because some * fs's don't know about trailing slashes. Remember if there were * trailing slashes to handle symlinks, existing non-directories * and non-existing files that won't be directories specially later. */ while (*cp == '/' && (cp[1] == '/' || cp[1] == '\0')) { cp++; ndp->ni_pathlen--; if (*cp == '\0') { /* * TODO * Regular lookup performs the following: * *ndp->ni_next = '\0'; * cnp->cn_flags |= TRAILINGSLASH; * * Which is problematic since it modifies data read * from userspace. Then if fast path lookup was to * abort we would have to either restore it or convey * the flag. Since this is a corner case just ignore * it for simplicity. */ return (cache_fpl_partial(fpl)); } } ndp->ni_next = cp; /* * Check for degenerate name (e.g. / or "") * which is a way of talking about a directory, * e.g. like "/." or ".". * * TODO * Another corner case handled by the regular lookup */ if (__predict_false(cnp->cn_nameptr[0] == '\0')) { return (cache_fpl_partial(fpl)); } return (0); } static void cache_fplookup_parse_advance(struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; ndp = fpl->ndp; cnp = fpl->cnp; cnp->cn_nameptr = ndp->ni_next; while (*cnp->cn_nameptr == '/') { cnp->cn_nameptr++; ndp->ni_pathlen--; } } /* * See the API contract for VOP_FPLOOKUP_VEXEC. */ static int __noinline cache_fplookup_failed_vexec(struct cache_fpl *fpl, int error) { + struct componentname *cnp; struct vnode *dvp; seqc_t dvp_seqc; + cnp = fpl->cnp; dvp = fpl->dvp; dvp_seqc = fpl->dvp_seqc; /* * Hack: they may be looking up foo/bar, where foo is a * regular file. In such a case we need to turn ENOTDIR, * but we may happen to get here with a different error. */ if (dvp->v_type != VDIR) { /* * The check here is predominantly to catch * EOPNOTSUPP from dead_vnodeops. If the vnode * gets doomed past this point it is going to * fail seqc verification. */ if (VN_IS_DOOMED(dvp)) { return (cache_fpl_aborted(fpl)); } error = ENOTDIR; } + /* + * Hack: handle O_SEARCH. + * + * Open Group Base Specifications Issue 7, 2018 edition states: + * If the access mode of the open file description associated with the + * file descriptor is not O_SEARCH, the function shall check whether + * directory searches are permitted using the current permissions of + * the directory underlying the file descriptor. If the access mode is + * O_SEARCH, the function shall not perform the check. + * + * Regular lookup tests for the NOEXECCHECK flag for every path + * component to decide whether to do the permission check. However, + * since most lookups never have the flag (and when they do it is only + * present for the first path component), lockless lookup only acts on + * it if there is a permission problem. Here the flag is represented + * with a boolean so that we don't have to clear it on the way out. + * + * For simplicity this always aborts. + * TODO: check if this is the first lookup and ignore the permission + * problem. Note the flag has to survive fallback (if it happens to be + * performed). + */ + if (fpl->fsearch) { + return (cache_fpl_aborted(fpl)); + } + switch (error) { case EAGAIN: if (!vn_seqc_consistent(dvp, dvp_seqc)) { error = cache_fpl_aborted(fpl); } else { cache_fpl_partial(fpl); } break; default: if (!vn_seqc_consistent(dvp, dvp_seqc)) { error = cache_fpl_aborted(fpl); } else { cache_fpl_smr_exit(fpl); cache_fpl_handled(fpl, error); } break; } return (error); } static int cache_fplookup_impl(struct vnode *dvp, struct cache_fpl *fpl) { struct nameidata *ndp; struct componentname *cnp; struct mount *mp; int error; error = CACHE_FPL_FAILED; ndp = fpl->ndp; cnp = fpl->cnp; cache_fpl_checkpoint(fpl, &fpl->snd); fpl->dvp = dvp; fpl->dvp_seqc = vn_seqc_read_any(fpl->dvp); if (seqc_in_modify(fpl->dvp_seqc)) { cache_fpl_aborted(fpl); goto out; } mp = atomic_load_ptr(&fpl->dvp->v_mount); if (!cache_fplookup_mp_supported(mp)) { cache_fpl_aborted(fpl); goto out; } VNPASS(cache_fplookup_vnode_supported(fpl->dvp), fpl->dvp); for (;;) { error = cache_fplookup_parse(fpl); if (__predict_false(error != 0)) { break; } VNPASS(cache_fplookup_vnode_supported(fpl->dvp), fpl->dvp); error = VOP_FPLOOKUP_VEXEC(fpl->dvp, cnp->cn_cred); if (__predict_false(error != 0)) { error = cache_fplookup_failed_vexec(fpl, error); break; } if (__predict_false(cache_fpl_isdotdot(cnp))) { error = cache_fplookup_dotdot(fpl); if (__predict_false(error != 0)) { break; } } else { error = cache_fplookup_next(fpl); if (__predict_false(error != 0)) { break; } VNPASS(!seqc_in_modify(fpl->tvp_seqc), fpl->tvp); if (cache_fplookup_need_climb_mount(fpl)) { error = cache_fplookup_climb_mount(fpl); if (__predict_false(error != 0)) { break; } } } VNPASS(!seqc_in_modify(fpl->tvp_seqc), fpl->tvp); if (cache_fpl_islastcn(ndp)) { error = cache_fplookup_final(fpl); break; } if (!vn_seqc_consistent(fpl->dvp, fpl->dvp_seqc)) { error = cache_fpl_aborted(fpl); break; } fpl->dvp = fpl->tvp; fpl->dvp_seqc = fpl->tvp_seqc; cache_fplookup_parse_advance(fpl); cache_fpl_checkpoint(fpl, &fpl->snd); } out: switch (fpl->status) { case CACHE_FPL_STATUS_UNSET: __assert_unreachable(); break; case CACHE_FPL_STATUS_PARTIAL: cache_fpl_smr_assert_entered(fpl); return (cache_fplookup_partial_setup(fpl)); case CACHE_FPL_STATUS_ABORTED: if (fpl->in_smr) cache_fpl_smr_exit(fpl); return (CACHE_FPL_FAILED); case CACHE_FPL_STATUS_HANDLED: MPASS(error != CACHE_FPL_FAILED); cache_fpl_smr_assert_not_entered(fpl); if (__predict_false(error != 0)) { ndp->ni_dvp = NULL; ndp->ni_vp = NULL; cache_fpl_cleanup_cnp(cnp); return (error); } ndp->ni_dvp = fpl->dvp; ndp->ni_vp = fpl->tvp; if (cnp->cn_flags & SAVENAME) cnp->cn_flags |= HASBUF; else cache_fpl_cleanup_cnp(cnp); return (error); } } /* * Fast path lookup protected with SMR and sequence counters. * * Note: all VOP_FPLOOKUP_VEXEC routines have a comment referencing this one. * * Filesystems can opt in by setting the MNTK_FPLOOKUP flag and meeting criteria * outlined below. * * Traditional vnode lookup conceptually looks like this: * * vn_lock(current); * for (;;) { * next = find(); * vn_lock(next); * vn_unlock(current); * current = next; * if (last) * break; * } * return (current); * * Each jump to the next vnode is safe memory-wise and atomic with respect to * any modifications thanks to holding respective locks. * * The same guarantee can be provided with a combination of safe memory * reclamation and sequence counters instead. If all operations which affect * the relationship between the current vnode and the one we are looking for * also modify the counter, we can verify whether all the conditions held as * we made the jump. This includes things like permissions, mount points etc. * Counter modification is provided by enclosing relevant places in * vn_seqc_write_begin()/end() calls. * * Thus this translates to: * * vfs_smr_enter(); * dvp_seqc = seqc_read_any(dvp); * if (seqc_in_modify(dvp_seqc)) // someone is altering the vnode * abort(); * for (;;) { * tvp = find(); * tvp_seqc = seqc_read_any(tvp); * if (seqc_in_modify(tvp_seqc)) // someone is altering the target vnode * abort(); * if (!seqc_consistent(dvp, dvp_seqc) // someone is altering the vnode * abort(); * dvp = tvp; // we know nothing of importance has changed * dvp_seqc = tvp_seqc; // store the counter for the tvp iteration * if (last) * break; * } * vget(); // secure the vnode * if (!seqc_consistent(tvp, tvp_seqc) // final check * abort(); * // at this point we know nothing has changed for any parent<->child pair * // as they were crossed during the lookup, meaning we matched the guarantee * // of the locked variant * return (tvp); * * The API contract for VOP_FPLOOKUP_VEXEC routines is as follows: * - they are called while within vfs_smr protection which they must never exit * - EAGAIN can be returned to denote checking could not be performed, it is * always valid to return it * - if the sequence counter has not changed the result must be valid * - if the sequence counter has changed both false positives and false negatives * are permitted (since the result will be rejected later) * - for simple cases of unix permission checks vaccess_vexec_smr can be used * * Caveats to watch out for: * - vnodes are passed unlocked and unreferenced with nothing stopping * VOP_RECLAIM, in turn meaning that ->v_data can become NULL. It is advised * to use atomic_load_ptr to fetch it. * - the aforementioned object can also get freed, meaning absent other means it * should be protected with vfs_smr * - either safely checking permissions as they are modified or guaranteeing * their stability is left to the routine */ int cache_fplookup(struct nameidata *ndp, enum cache_fpl_status *status, struct pwd **pwdp) { struct cache_fpl fpl; struct pwd *pwd; struct vnode *dvp; struct componentname *cnp; struct nameidata_saved orig; int error; MPASS(ndp->ni_lcf == 0); fpl.status = CACHE_FPL_STATUS_UNSET; fpl.ndp = ndp; fpl.cnp = &ndp->ni_cnd; MPASS(curthread == fpl.cnp->cn_thread); if ((fpl.cnp->cn_flags & SAVESTART) != 0) MPASS(fpl.cnp->cn_nameiop != LOOKUP); if (!cache_can_fplookup(&fpl)) { SDT_PROBE3(vfs, fplookup, lookup, done, ndp, fpl.line, fpl.status); *status = fpl.status; return (EOPNOTSUPP); } cache_fpl_checkpoint(&fpl, &orig); cache_fpl_smr_enter_initial(&fpl); + fpl.fsearch = false; pwd = pwd_get_smr(); fpl.pwd = pwd; ndp->ni_rootdir = pwd->pwd_rdir; ndp->ni_topdir = pwd->pwd_jdir; cnp = fpl.cnp; cnp->cn_nameptr = cnp->cn_pnbuf; if (cnp->cn_pnbuf[0] == '/') { cache_fpl_handle_root(ndp, &dvp); } else { - MPASS(ndp->ni_dirfd == AT_FDCWD); - dvp = pwd->pwd_cdir; + if (ndp->ni_dirfd == AT_FDCWD) { + dvp = pwd->pwd_cdir; + } else { + error = cache_fplookup_dirfd(&fpl, &dvp); + if (__predict_false(error != 0)) { + goto out; + } + } } SDT_PROBE4(vfs, namei, lookup, entry, dvp, cnp->cn_pnbuf, cnp->cn_flags, true); error = cache_fplookup_impl(dvp, &fpl); +out: cache_fpl_smr_assert_not_entered(&fpl); SDT_PROBE3(vfs, fplookup, lookup, done, ndp, fpl.line, fpl.status); *status = fpl.status; switch (fpl.status) { case CACHE_FPL_STATUS_UNSET: __assert_unreachable(); break; case CACHE_FPL_STATUS_HANDLED: SDT_PROBE3(vfs, namei, lookup, return, error, (error == 0 ? ndp->ni_vp : NULL), true); break; case CACHE_FPL_STATUS_PARTIAL: *pwdp = fpl.pwd; /* * Status restored by cache_fplookup_partial_setup. */ break; case CACHE_FPL_STATUS_ABORTED: cache_fpl_restore(&fpl, &orig); break; } return (error); } Index: head/sys/sys/file.h =================================================================== --- head/sys/sys/file.h (revision 366596) +++ head/sys/sys/file.h (revision 366597) @@ -1,462 +1,464 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 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. * * @(#)file.h 8.3 (Berkeley) 1/9/95 * $FreeBSD$ */ #ifndef _SYS_FILE_H_ #define _SYS_FILE_H_ #ifndef _KERNEL #include /* XXX */ #include #include #else #include #include #include #include #include struct filedesc; struct stat; struct thread; struct uio; struct knote; struct vnode; +struct nameidata; #endif /* _KERNEL */ #define DTYPE_NONE 0 /* not yet initialized */ #define DTYPE_VNODE 1 /* file */ #define DTYPE_SOCKET 2 /* communications endpoint */ #define DTYPE_PIPE 3 /* pipe */ #define DTYPE_FIFO 4 /* fifo (named pipe) */ #define DTYPE_KQUEUE 5 /* event queue */ #define DTYPE_CRYPTO 6 /* crypto */ #define DTYPE_MQUEUE 7 /* posix message queue */ #define DTYPE_SHM 8 /* swap-backed shared memory */ #define DTYPE_SEM 9 /* posix semaphore */ #define DTYPE_PTS 10 /* pseudo teletype master device */ #define DTYPE_DEV 11 /* Device specific fd type */ #define DTYPE_PROCDESC 12 /* process descriptor */ #define DTYPE_LINUXEFD 13 /* emulation eventfd type */ #define DTYPE_LINUXTFD 14 /* emulation timerfd type */ #ifdef _KERNEL struct file; struct filecaps; struct kaiocb; struct kinfo_file; struct ucred; #define FOF_OFFSET 0x01 /* Use the offset in uio argument */ #define FOF_NOLOCK 0x02 /* Do not take FOFFSET_LOCK */ #define FOF_NEXTOFF_R 0x04 /* Also update f_nextoff[UIO_READ] */ #define FOF_NEXTOFF_W 0x08 /* Also update f_nextoff[UIO_WRITE] */ #define FOF_NOUPDATE 0x10 /* Do not update f_offset */ off_t foffset_lock(struct file *fp, int flags); void foffset_lock_uio(struct file *fp, struct uio *uio, int flags); void foffset_unlock(struct file *fp, off_t val, int flags); void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags); static inline off_t foffset_get(struct file *fp) { return (foffset_lock(fp, FOF_NOLOCK)); } typedef int fo_rdwr_t(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td); typedef int fo_truncate_t(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td); typedef int fo_ioctl_t(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td); typedef int fo_poll_t(struct file *fp, int events, struct ucred *active_cred, struct thread *td); typedef int fo_kqfilter_t(struct file *fp, struct knote *kn); typedef int fo_stat_t(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td); typedef int fo_close_t(struct file *fp, struct thread *td); typedef int fo_chmod_t(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); typedef int fo_chown_t(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); typedef int fo_sendfile_t(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td); typedef int fo_seek_t(struct file *fp, off_t offset, int whence, struct thread *td); typedef int fo_fill_kinfo_t(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp); typedef int fo_mmap_t(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td); typedef int fo_aio_queue_t(struct file *fp, struct kaiocb *job); typedef int fo_add_seals_t(struct file *fp, int flags); typedef int fo_get_seals_t(struct file *fp, int *flags); typedef int fo_fallocate_t(struct file *fp, off_t offset, off_t len, struct thread *td); typedef int fo_flags_t; struct fileops { fo_rdwr_t *fo_read; fo_rdwr_t *fo_write; fo_truncate_t *fo_truncate; fo_ioctl_t *fo_ioctl; fo_poll_t *fo_poll; fo_kqfilter_t *fo_kqfilter; fo_stat_t *fo_stat; fo_close_t *fo_close; fo_chmod_t *fo_chmod; fo_chown_t *fo_chown; fo_sendfile_t *fo_sendfile; fo_seek_t *fo_seek; fo_fill_kinfo_t *fo_fill_kinfo; fo_mmap_t *fo_mmap; fo_aio_queue_t *fo_aio_queue; fo_add_seals_t *fo_add_seals; fo_get_seals_t *fo_get_seals; fo_fallocate_t *fo_fallocate; fo_flags_t fo_flags; /* DFLAG_* below */ }; #define DFLAG_PASSABLE 0x01 /* may be passed via unix sockets. */ #define DFLAG_SEEKABLE 0x02 /* seekable / nonsequential */ #endif /* _KERNEL */ #if defined(_KERNEL) || defined(_WANT_FILE) /* * Kernel descriptor table. * One entry for each open kernel vnode and socket. * * Below is the list of locks that protects members in struct file. * * (a) f_vnode lock required (shared allows both reads and writes) * (f) updated with atomics and blocking on sleepq * (d) cdevpriv_mtx * none not locked */ struct fadvise_info { int fa_advice; /* (f) FADV_* type. */ off_t fa_start; /* (f) Region start. */ off_t fa_end; /* (f) Region end. */ }; struct file { void *f_data; /* file descriptor specific data */ struct fileops *f_ops; /* File operations */ struct ucred *f_cred; /* associated credentials. */ struct vnode *f_vnode; /* NULL or applicable vnode */ short f_type; /* descriptor type */ short f_vnread_flags; /* (f) Sleep lock for f_offset */ volatile u_int f_flag; /* see fcntl.h */ volatile u_int f_count; /* reference count */ /* * DTYPE_VNODE specific fields. */ union { int16_t f_seqcount[2]; /* (a) Count of seq. reads and writes. */ int f_pipegen; }; off_t f_nextoff[2]; /* next expected read/write offset. */ union { struct cdev_privdata *fvn_cdevpriv; /* (d) Private data for the cdev. */ struct fadvise_info *fvn_advice; } f_vnun; /* * DFLAG_SEEKABLE specific fields */ off_t f_offset; }; #define f_cdevpriv f_vnun.fvn_cdevpriv #define f_advice f_vnun.fvn_advice #define FOFFSET_LOCKED 0x1 #define FOFFSET_LOCK_WAITING 0x2 #endif /* _KERNEL || _WANT_FILE */ /* * Userland version of struct file, for sysctl */ struct xfile { ksize_t xf_size; /* size of struct xfile */ pid_t xf_pid; /* owning process */ uid_t xf_uid; /* effective uid of owning process */ int xf_fd; /* descriptor number */ int _xf_int_pad1; kvaddr_t xf_file; /* address of struct file */ short xf_type; /* descriptor type */ short _xf_short_pad1; int xf_count; /* reference count */ int xf_msgcount; /* references from message queue */ int _xf_int_pad2; off_t xf_offset; /* file offset */ kvaddr_t xf_data; /* file descriptor specific data */ kvaddr_t xf_vnode; /* vnode pointer */ u_int xf_flag; /* flags (see fcntl.h) */ int _xf_int_pad3; int64_t _xf_int64_pad[6]; }; #ifdef _KERNEL extern struct fileops vnops; extern struct fileops badfileops; extern struct fileops socketops; extern int maxfiles; /* kernel limit on number of open files */ extern int maxfilesperproc; /* per process limit on number of open files */ int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_mmap(struct thread *td, int fd, cap_rights_t *rightsp, vm_prot_t *maxprotp, struct file **fpp); int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp); int fget_fcntl(struct thread *td, int fd, cap_rights_t *rightsp, int needfcntl, struct file **fpp); int _fdrop(struct file *fp, struct thread *td); fo_rdwr_t invfo_rdwr; fo_truncate_t invfo_truncate; fo_ioctl_t invfo_ioctl; fo_poll_t invfo_poll; fo_kqfilter_t invfo_kqfilter; fo_chmod_t invfo_chmod; fo_chown_t invfo_chown; fo_sendfile_t invfo_sendfile; fo_sendfile_t vn_sendfile; fo_seek_t vn_seek; fo_fill_kinfo_t vn_fill_kinfo; int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif); void finit(struct file *, u_int, short, void *, struct fileops *); void finit_vnode(struct file *, u_int, void *, struct fileops *); int fgetvp(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_exec(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_rights(struct thread *td, int fd, cap_rights_t *needrightsp, struct filecaps *havecaps, struct vnode **vpp); int fgetvp_read(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); int fgetvp_write(struct thread *td, int fd, cap_rights_t *rightsp, struct vnode **vpp); +int fgetvp_lookup_smr(int fd, struct nameidata *ndp, struct vnode **vpp, bool *fsearch); static __inline __result_use_check bool fhold(struct file *fp) { return (refcount_acquire_checked(&fp->f_count)); } #define fdrop(fp, td) ({ \ struct file *_fp; \ int _error; \ \ _error = 0; \ _fp = (fp); \ if (__predict_false(refcount_release(&_fp->f_count))) \ _error = _fdrop(_fp, td); \ _error; \ }) static __inline fo_rdwr_t fo_read; static __inline fo_rdwr_t fo_write; static __inline fo_truncate_t fo_truncate; static __inline fo_ioctl_t fo_ioctl; static __inline fo_poll_t fo_poll; static __inline fo_kqfilter_t fo_kqfilter; static __inline fo_stat_t fo_stat; static __inline fo_close_t fo_close; static __inline fo_chmod_t fo_chmod; static __inline fo_chown_t fo_chown; static __inline fo_sendfile_t fo_sendfile; static __inline int fo_read(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return ((*fp->f_ops->fo_read)(fp, uio, active_cred, flags, td)); } static __inline int fo_write(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { return ((*fp->f_ops->fo_write)(fp, uio, active_cred, flags, td)); } static __inline int fo_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_truncate)(fp, length, active_cred, td)); } static __inline int fo_ioctl(struct file *fp, u_long com, void *data, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_ioctl)(fp, com, data, active_cred, td)); } static __inline int fo_poll(struct file *fp, int events, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_poll)(fp, events, active_cred, td)); } static __inline int fo_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_stat)(fp, sb, active_cred, td)); } static __inline int fo_close(struct file *fp, struct thread *td) { return ((*fp->f_ops->fo_close)(fp, td)); } static __inline int fo_kqfilter(struct file *fp, struct knote *kn) { return ((*fp->f_ops->fo_kqfilter)(fp, kn)); } static __inline int fo_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_chmod)(fp, mode, active_cred, td)); } static __inline int fo_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { return ((*fp->f_ops->fo_chown)(fp, uid, gid, active_cred, td)); } static __inline int fo_sendfile(struct file *fp, int sockfd, struct uio *hdr_uio, struct uio *trl_uio, off_t offset, size_t nbytes, off_t *sent, int flags, struct thread *td) { return ((*fp->f_ops->fo_sendfile)(fp, sockfd, hdr_uio, trl_uio, offset, nbytes, sent, flags, td)); } static __inline int fo_seek(struct file *fp, off_t offset, int whence, struct thread *td) { return ((*fp->f_ops->fo_seek)(fp, offset, whence, td)); } static __inline int fo_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { return ((*fp->f_ops->fo_fill_kinfo)(fp, kif, fdp)); } static __inline int fo_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, struct thread *td) { if (fp->f_ops->fo_mmap == NULL) return (ENODEV); return ((*fp->f_ops->fo_mmap)(fp, map, addr, size, prot, cap_maxprot, flags, foff, td)); } static __inline int fo_aio_queue(struct file *fp, struct kaiocb *job) { return ((*fp->f_ops->fo_aio_queue)(fp, job)); } static __inline int fo_add_seals(struct file *fp, int seals) { if (fp->f_ops->fo_add_seals == NULL) return (EINVAL); return ((*fp->f_ops->fo_add_seals)(fp, seals)); } static __inline int fo_get_seals(struct file *fp, int *seals) { if (fp->f_ops->fo_get_seals == NULL) return (EINVAL); return ((*fp->f_ops->fo_get_seals)(fp, seals)); } static __inline int fo_fallocate(struct file *fp, off_t offset, off_t len, struct thread *td) { if (fp->f_ops->fo_fallocate == NULL) return (ENODEV); return ((*fp->f_ops->fo_fallocate)(fp, offset, len, td)); } #endif /* _KERNEL */ #endif /* !SYS_FILE_H */