Index: head/sys/fs/ext2fs/ext2_vnops.c =================================================================== --- head/sys/fs/ext2fs/ext2_vnops.c (revision 309061) +++ head/sys/fs/ext2fs/ext2_vnops.c (revision 309062) @@ -1,2080 +1,2046 @@ /*- * modified for EXT2FS support in Lites 1.1 * * Aug 1995, Godmar Back (gback@cs.utah.edu) * University of Utah, Department of Computer Science */ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufs_vnops.c 8.7 (Berkeley) 2/3/94 * @(#)ufs_vnops.c 8.27 (Berkeley) 5/27/95 * $FreeBSD$ */ #include "opt_suiddir.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 "opt_directio.h" #include #include #include #include #include #include #include #include static int ext2_makeinode(int mode, struct vnode *, struct vnode **, struct componentname *); static void ext2_itimes_locked(struct vnode *); static int ext4_ext_read(struct vop_read_args *); static int ext2_ind_read(struct vop_read_args *); static vop_access_t ext2_access; static int ext2_chmod(struct vnode *, int, struct ucred *, struct thread *); static int ext2_chown(struct vnode *, uid_t, gid_t, struct ucred *, struct thread *); static vop_close_t ext2_close; static vop_create_t ext2_create; static vop_fsync_t ext2_fsync; static vop_getattr_t ext2_getattr; static vop_ioctl_t ext2_ioctl; static vop_link_t ext2_link; static vop_mkdir_t ext2_mkdir; static vop_mknod_t ext2_mknod; static vop_open_t ext2_open; static vop_pathconf_t ext2_pathconf; static vop_print_t ext2_print; static vop_read_t ext2_read; static vop_readlink_t ext2_readlink; static vop_remove_t ext2_remove; static vop_rename_t ext2_rename; static vop_rmdir_t ext2_rmdir; static vop_setattr_t ext2_setattr; static vop_strategy_t ext2_strategy; static vop_symlink_t ext2_symlink; static vop_write_t ext2_write; static vop_vptofh_t ext2_vptofh; static vop_close_t ext2fifo_close; static vop_kqfilter_t ext2fifo_kqfilter; /* Global vfs data structures for ext2. */ struct vop_vector ext2_vnodeops = { .vop_default = &default_vnodeops, .vop_access = ext2_access, .vop_bmap = ext2_bmap, .vop_cachedlookup = ext2_lookup, .vop_close = ext2_close, .vop_create = ext2_create, .vop_fsync = ext2_fsync, .vop_getpages = vnode_pager_local_getpages, .vop_getpages_async = vnode_pager_local_getpages_async, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_ioctl = ext2_ioctl, .vop_link = ext2_link, .vop_lookup = vfs_cache_lookup, .vop_mkdir = ext2_mkdir, .vop_mknod = ext2_mknod, .vop_open = ext2_open, .vop_pathconf = ext2_pathconf, .vop_poll = vop_stdpoll, .vop_print = ext2_print, .vop_read = ext2_read, .vop_readdir = ext2_readdir, .vop_readlink = ext2_readlink, .vop_reallocblks = ext2_reallocblks, .vop_reclaim = ext2_reclaim, .vop_remove = ext2_remove, .vop_rename = ext2_rename, .vop_rmdir = ext2_rmdir, .vop_setattr = ext2_setattr, .vop_strategy = ext2_strategy, .vop_symlink = ext2_symlink, .vop_write = ext2_write, .vop_vptofh = ext2_vptofh, }; struct vop_vector ext2_fifoops = { .vop_default = &fifo_specops, .vop_access = ext2_access, .vop_close = ext2fifo_close, .vop_fsync = ext2_fsync, .vop_getattr = ext2_getattr, .vop_inactive = ext2_inactive, .vop_kqfilter = ext2fifo_kqfilter, .vop_print = ext2_print, .vop_read = VOP_PANIC, .vop_reclaim = ext2_reclaim, .vop_setattr = ext2_setattr, .vop_write = VOP_PANIC, .vop_vptofh = ext2_vptofh, }; /* * A virgin directory (no blushing please). * Note that the type and namlen fields are reversed relative to ext2. * Also, we don't use `struct odirtemplate', since it would just cause * endianness problems. */ static struct dirtemplate mastertemplate = { 0, 12, 1, EXT2_FT_DIR, ".", 0, DIRBLKSIZ - 12, 2, EXT2_FT_DIR, ".." }; static struct dirtemplate omastertemplate = { 0, 12, 1, EXT2_FT_UNKNOWN, ".", 0, DIRBLKSIZ - 12, 2, EXT2_FT_UNKNOWN, ".." }; static void ext2_itimes_locked(struct vnode *vp) { struct inode *ip; struct timespec ts; ASSERT_VI_LOCKED(vp, __func__); ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE)) == 0) return; if ((vp->v_type == VBLK || vp->v_type == VCHR)) ip->i_flag |= IN_LAZYMOD; else ip->i_flag |= IN_MODIFIED; if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { vfs_timestamp(&ts); if (ip->i_flag & IN_ACCESS) { ip->i_atime = ts.tv_sec; ip->i_atimensec = ts.tv_nsec; } if (ip->i_flag & IN_UPDATE) { ip->i_mtime = ts.tv_sec; ip->i_mtimensec = ts.tv_nsec; ip->i_modrev++; } if (ip->i_flag & IN_CHANGE) { ip->i_ctime = ts.tv_sec; ip->i_ctimensec = ts.tv_nsec; } } ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE); } void ext2_itimes(struct vnode *vp) { VI_LOCK(vp); ext2_itimes_locked(vp); VI_UNLOCK(vp); } /* * Create a regular file */ static int ext2_create(struct vop_create_args *ap) { int error; error = ext2_makeinode(MAKEIMODE(ap->a_vap->va_type, ap->a_vap->va_mode), ap->a_dvp, ap->a_vpp, ap->a_cnp); if (error != 0) return (error); if ((ap->a_cnp->cn_flags & MAKEENTRY) != 0) cache_enter(ap->a_dvp, *ap->a_vpp, ap->a_cnp); return (0); } static int ext2_open(struct vop_open_args *ap) { if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) return (EOPNOTSUPP); /* * Files marked append-only must be opened for appending. */ if ((VTOI(ap->a_vp)->i_flags & APPEND) && (ap->a_mode & (FWRITE | O_APPEND)) == FWRITE) return (EPERM); vnode_create_vobject(ap->a_vp, VTOI(ap->a_vp)->i_size, ap->a_td); return (0); } /* * Close called. * * Update the times on the inode. */ static int ext2_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (0); } static int ext2_access(struct vop_access_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); accmode_t accmode = ap->a_accmode; int error; if (vp->v_type == VBLK || vp->v_type == VCHR) return (EOPNOTSUPP); /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ if (accmode & VWRITE) { switch (vp->v_type) { case VDIR: case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } } /* If immutable bit set, nobody gets to write it. */ if ((accmode & VWRITE) && (ip->i_flags & (SF_IMMUTABLE | SF_SNAPSHOT))) return (EPERM); error = vaccess(vp->v_type, ip->i_mode, ip->i_uid, ip->i_gid, ap->a_accmode, ap->a_cred, NULL); return (error); } static int ext2_getattr(struct vop_getattr_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct vattr *vap = ap->a_vap; ext2_itimes(vp); /* * Copy from inode table */ vap->va_fsid = dev2udev(ip->i_devvp->v_rdev); vap->va_fileid = ip->i_number; vap->va_mode = ip->i_mode & ~IFMT; vap->va_nlink = ip->i_nlink; vap->va_uid = ip->i_uid; vap->va_gid = ip->i_gid; vap->va_rdev = ip->i_rdev; vap->va_size = ip->i_size; vap->va_atime.tv_sec = ip->i_atime; vap->va_atime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_atimensec : 0; vap->va_mtime.tv_sec = ip->i_mtime; vap->va_mtime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_mtimensec : 0; vap->va_ctime.tv_sec = ip->i_ctime; vap->va_ctime.tv_nsec = E2DI_HAS_XTIME(ip) ? ip->i_ctimensec : 0; if E2DI_HAS_XTIME(ip) { vap->va_birthtime.tv_sec = ip->i_birthtime; vap->va_birthtime.tv_nsec = ip->i_birthnsec; } vap->va_flags = ip->i_flags; vap->va_gen = ip->i_gen; vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; vap->va_bytes = dbtob((u_quad_t)ip->i_blocks); vap->va_type = IFTOVT(ip->i_mode); vap->va_filerev = ip->i_modrev; return (0); } /* * Set attribute vnode op. called from several syscalls */ static int ext2_setattr(struct vop_setattr_args *ap) { struct vattr *vap = ap->a_vap; struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); struct ucred *cred = ap->a_cred; struct thread *td = curthread; int error; /* * Check for unsettable attributes. */ if ((vap->va_type != VNON) || (vap->va_nlink != VNOVAL) || (vap->va_fsid != VNOVAL) || (vap->va_fileid != VNOVAL) || (vap->va_blocksize != VNOVAL) || (vap->va_rdev != VNOVAL) || ((int)vap->va_bytes != VNOVAL) || (vap->va_gen != VNOVAL)) { return (EINVAL); } if (vap->va_flags != VNOVAL) { /* Disallow flags not supported by ext2fs. */ if(vap->va_flags & ~(SF_APPEND | SF_IMMUTABLE | UF_NODUMP)) return (EOPNOTSUPP); if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * Callers may only modify the file flags on objects they * have VADMIN rights for. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Unprivileged processes and privileged processes in * jail() are not permitted to unset system flags, or * modify flags if any system flags are set. * Privileged non-jail processes may not modify system flags * if securelevel > 0 and any existing system flags are set. */ if (!priv_check_cred(cred, PRIV_VFS_SYSFLAGS, 0)) { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND)) { error = securelevel_gt(cred, 0); if (error) return (error); } } else { if (ip->i_flags & (SF_IMMUTABLE | SF_APPEND) || ((vap->va_flags ^ ip->i_flags) & SF_SETTABLE)) return (EPERM); } ip->i_flags = vap->va_flags; ip->i_flag |= IN_CHANGE; if (ip->i_flags & (IMMUTABLE | APPEND)) return (0); } if (ip->i_flags & (IMMUTABLE | APPEND)) return (EPERM); /* * Go through the fields and update iff not VNOVAL. */ if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); if ((error = ext2_chown(vp, vap->va_uid, vap->va_gid, cred, td)) != 0) return (error); } if (vap->va_size != VNOVAL) { /* * Disallow write attempts on read-only file systems; * unless the file is a socket, fifo, or a block or * character device resident on the file system. */ switch (vp->v_type) { case VDIR: return (EISDIR); case VLNK: case VREG: if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); break; default: break; } if ((error = ext2_truncate(vp, vap->va_size, 0, cred, td)) != 0) return (error); } if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); /* * From utimes(2): * If times is NULL, ... The caller must be the owner of * the file, have permission to write the file, or be the * super-user. * If times is non-NULL, ... The caller must be the owner of * the file or be the super-user. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td)) && ((vap->va_vaflags & VA_UTIMES_NULL) == 0 || (error = VOP_ACCESS(vp, VWRITE, cred, td)))) return (error); ip->i_flag |= IN_CHANGE | IN_MODIFIED; if (vap->va_atime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_ACCESS; ip->i_atime = vap->va_atime.tv_sec; ip->i_atimensec = vap->va_atime.tv_nsec; } if (vap->va_mtime.tv_sec != VNOVAL) { ip->i_flag &= ~IN_UPDATE; ip->i_mtime = vap->va_mtime.tv_sec; ip->i_mtimensec = vap->va_mtime.tv_nsec; } ip->i_birthtime = vap->va_birthtime.tv_sec; ip->i_birthnsec = vap->va_birthtime.tv_nsec; error = ext2_update(vp, 0); if (error) return (error); } error = 0; if (vap->va_mode != (mode_t)VNOVAL) { if (vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); error = ext2_chmod(vp, (int)vap->va_mode, cred, td); } return (error); } /* * Change the mode on a file. * Inode must be locked before calling. */ static int ext2_chmod(struct vnode *vp, int mode, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); int error; /* * To modify the permissions on a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * Privileged processes may set the sticky bit on non-directories, * as well as set the setgid bit on a file with a group that the * process is not a member of. */ if (vp->v_type != VDIR && (mode & S_ISTXT)) { error = priv_check_cred(cred, PRIV_VFS_STICKYFILE, 0); if (error) return (EFTYPE); } if (!groupmember(ip->i_gid, cred) && (mode & ISGID)) { error = priv_check_cred(cred, PRIV_VFS_SETGID, 0); if (error) return (error); } ip->i_mode &= ~ALLPERMS; ip->i_mode |= (mode & ALLPERMS); ip->i_flag |= IN_CHANGE; return (0); } /* * Perform chown operation on inode ip; * inode must be locked prior to call. */ static int ext2_chown(struct vnode *vp, uid_t uid, gid_t gid, struct ucred *cred, struct thread *td) { struct inode *ip = VTOI(vp); uid_t ouid; gid_t ogid; int error = 0; if (uid == (uid_t)VNOVAL) uid = ip->i_uid; if (gid == (gid_t)VNOVAL) gid = ip->i_gid; /* * To modify the ownership of a file, must possess VADMIN * for that file. */ if ((error = VOP_ACCESS(vp, VADMIN, cred, td))) return (error); /* * To change the owner of a file, or change the group of a file * to a group of which we are not a member, the caller must * have privilege. */ if (uid != ip->i_uid || (gid != ip->i_gid && !groupmember(gid, cred))) { error = priv_check_cred(cred, PRIV_VFS_CHOWN, 0); if (error) return (error); } ogid = ip->i_gid; ouid = ip->i_uid; ip->i_gid = gid; ip->i_uid = uid; ip->i_flag |= IN_CHANGE; if ((ip->i_mode & (ISUID | ISGID)) && (ouid != uid || ogid != gid)) { if (priv_check_cred(cred, PRIV_VFS_RETAINSUGID, 0) != 0) ip->i_mode &= ~(ISUID | ISGID); } return (0); } /* * Synch an open file. */ /* ARGSUSED */ static int ext2_fsync(struct vop_fsync_args *ap) { /* * Flush all dirty buffers associated with a vnode. */ vop_stdfsync(ap); return (ext2_update(ap->a_vp, ap->a_waitfor == MNT_WAIT)); } /* * Mknod vnode call */ /* ARGSUSED */ static int ext2_mknod(struct vop_mknod_args *ap) { struct vattr *vap = ap->a_vap; struct vnode **vpp = ap->a_vpp; struct inode *ip; ino_t ino; int error; error = ext2_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); ip = VTOI(*vpp); ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; if (vap->va_rdev != VNOVAL) { /* * Want to be able to use this to make badblock * inodes, so don't truncate the dev number. */ ip->i_rdev = vap->va_rdev; } /* * Remove inode, then reload it through VFS_VGET so it is * checked to see if it is an alias of an existing entry in * the inode cache. XXX I don't believe this is necessary now. */ (*vpp)->v_type = VNON; ino = ip->i_number; /* Save this before vgone() invalidates ip. */ vgone(*vpp); vput(*vpp); error = VFS_VGET(ap->a_dvp->v_mount, ino, LK_EXCLUSIVE, vpp); if (error) { *vpp = NULL; return (error); } return (0); } static int ext2_remove(struct vop_remove_args *ap) { struct inode *ip; struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; int error; ip = VTOI(vp); if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(dvp)->i_flags & APPEND)) { error = EPERM; goto out; } error = ext2_dirremove(dvp, ap->a_cnp); if (error == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } /* * link vnode call */ static int ext2_link(struct vop_link_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *tdvp = ap->a_tdvp; struct componentname *cnp = ap->a_cnp; struct inode *ip; int error; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_link: no name"); #endif ip = VTOI(vp); if ((nlink_t)ip->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto out; } if (ip->i_flags & (IMMUTABLE | APPEND)) { error = EPERM; goto out; } ip->i_nlink++; ip->i_flag |= IN_CHANGE; error = ext2_update(vp, !DOINGASYNC(vp)); if (!error) error = ext2_direnter(ip, tdvp, cnp); if (error) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; } out: return (error); } /* * Rename system call. * rename("foo", "bar"); * is essentially * unlink("bar"); * link("foo", "bar"); * unlink("foo"); * but ``atomically''. Can't do full commit without saving state in the * inode on disk which isn't feasible at this time. Best we can do is * always guarantee the target exists. * * Basic algorithm is: * * 1) Bump link count on source while we're linking it to the * target. This also ensure the inode won't be deleted out * from underneath us while we work (it may be truncated by * a concurrent `trunc' or `open' for creation). * 2) Link source to destination. If destination already exists, * delete it first. * 3) Unlink source reference to inode if still around. If a * directory was moved and the parent of the destination * is different from the source, patch the ".." entry in the * directory. */ static int ext2_rename(struct vop_rename_args *ap) { struct vnode *tvp = ap->a_tvp; struct vnode *tdvp = ap->a_tdvp; struct vnode *fvp = ap->a_fvp; struct vnode *fdvp = ap->a_fdvp; struct componentname *tcnp = ap->a_tcnp; struct componentname *fcnp = ap->a_fcnp; struct inode *ip, *xp, *dp; struct dirtemplate dirbuf; int doingdirectory = 0, oldparent = 0, newparent = 0; int error = 0; u_char namlen; #ifdef INVARIANTS if ((tcnp->cn_flags & HASBUF) == 0 || (fcnp->cn_flags & HASBUF) == 0) panic("ext2_rename: no name"); #endif /* * Check for cross-device rename. */ if ((fvp->v_mount != tdvp->v_mount) || (tvp && (fvp->v_mount != tvp->v_mount))) { error = EXDEV; abortit: if (tdvp == tvp) vrele(tdvp); else vput(tdvp); if (tvp) vput(tvp); vrele(fdvp); vrele(fvp); return (error); } if (tvp && ((VTOI(tvp)->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (VTOI(tdvp)->i_flags & APPEND))) { error = EPERM; goto abortit; } /* * Renaming a file to itself has no effect. The upper layers should * not call us in that case. Temporarily just warn if they do. */ if (fvp == tvp) { printf("ext2_rename: fvp == tvp (can't happen)\n"); error = 0; goto abortit; } if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0) goto abortit; dp = VTOI(fdvp); ip = VTOI(fvp); if (ip->i_nlink >= EXT2_LINK_MAX) { VOP_UNLOCK(fvp, 0); error = EMLINK; goto abortit; } if ((ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND)) || (dp->i_flags & APPEND)) { VOP_UNLOCK(fvp, 0); error = EPERM; goto abortit; } if ((ip->i_mode & IFMT) == IFDIR) { /* * Avoid ".", "..", and aliases of "." for obvious reasons. */ if ((fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || dp == ip || (fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT || (ip->i_flag & IN_RENAME)) { VOP_UNLOCK(fvp, 0); error = EINVAL; goto abortit; } ip->i_flag |= IN_RENAME; oldparent = dp->i_number; doingdirectory++; } vrele(fdvp); /* * When the target exists, both the directory * and target vnodes are returned locked. */ dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); /* * 1) Bump link count while we're moving stuff * around. If we crash somewhere before * completing our work, the link count * may be wrong, but correctable. */ ip->i_nlink++; ip->i_flag |= IN_CHANGE; if ((error = ext2_update(fvp, !DOINGASYNC(fvp))) != 0) { VOP_UNLOCK(fvp, 0); goto bad; } /* * If ".." must be changed (ie the directory gets a new * parent) then the source directory must not be in the * directory hierarchy above the target, as this would * orphan everything below the source directory. Also * the user must have write permission in the source so * as to be able to change "..". We must repeat the call * to namei, as the parent directory is unlocked by the * call to checkpath(). */ error = VOP_ACCESS(fvp, VWRITE, tcnp->cn_cred, tcnp->cn_thread); VOP_UNLOCK(fvp, 0); if (oldparent != dp->i_number) newparent = dp->i_number; if (doingdirectory && newparent) { if (error) /* write access check above */ goto bad; if (xp != NULL) vput(tvp); error = ext2_checkpath(ip, dp, tcnp->cn_cred); if (error) goto out; VREF(tdvp); error = relookup(tdvp, &tvp, tcnp); if (error) goto out; vrele(tdvp); dp = VTOI(tdvp); xp = NULL; if (tvp) xp = VTOI(tvp); } /* * 2) If target doesn't exist, link the target * to the source and unlink the source. * Otherwise, rewrite the target directory * entry to reference the source inode and * expunge the original entry's existence. */ if (xp == NULL) { if (dp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Account for ".." in new directory. * When source and destination have the same * parent we don't fool with the link count. */ if (doingdirectory && newparent) { if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto bad; } dp->i_nlink++; dp->i_flag |= IN_CHANGE; error = ext2_update(tdvp, !DOINGASYNC(tdvp)); if (error) goto bad; } error = ext2_direnter(ip, tdvp, tcnp); if (error) { if (doingdirectory && newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; (void)ext2_update(tdvp, 1); } goto bad; } vput(tdvp); } else { if (xp->i_devvp != dp->i_devvp || xp->i_devvp != ip->i_devvp) panic("ext2_rename: EXDEV"); /* * Short circuit rename(foo, foo). */ if (xp->i_number == ip->i_number) panic("ext2_rename: same file"); /* * If the parent directory is "sticky", then the user must * own the parent directory, or the destination of the rename, * otherwise the destination may not be changed (except by * root). This implements append-only directories. */ if ((dp->i_mode & S_ISTXT) && tcnp->cn_cred->cr_uid != 0 && tcnp->cn_cred->cr_uid != dp->i_uid && xp->i_uid != tcnp->cn_cred->cr_uid) { error = EPERM; goto bad; } /* * Target must be empty if a directory and have no links * to it. Also, ensure source and target are compatible * (both directories, or both not directories). */ if ((xp->i_mode&IFMT) == IFDIR) { if (! ext2_dirempty(xp, dp->i_number, tcnp->cn_cred) || xp->i_nlink > 2) { error = ENOTEMPTY; goto bad; } if (!doingdirectory) { error = ENOTDIR; goto bad; } cache_purge(tdvp); } else if (doingdirectory) { error = EISDIR; goto bad; } error = ext2_dirrewrite(dp, ip, tcnp); if (error) goto bad; /* * If the target directory is in the same * directory as the source directory, * decrement the link count on the parent * of the target directory. */ if (doingdirectory && !newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; } vput(tdvp); /* * Adjust the link count of the target to * reflect the dirrewrite above. If this is * a directory it is empty and there are * no links to it, so we can squash the inode and * any space associated with it. We disallowed * renaming over top of a directory with links to * it above, as the remaining link would point to * a directory without "." or ".." entries. */ xp->i_nlink--; if (doingdirectory) { if (--xp->i_nlink != 0) panic("ext2_rename: linked directory"); error = ext2_truncate(tvp, (off_t)0, IO_SYNC, tcnp->cn_cred, tcnp->cn_thread); } xp->i_flag |= IN_CHANGE; vput(tvp); xp = NULL; } /* * 3) Unlink the source. */ fcnp->cn_flags &= ~MODMASK; fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; VREF(fdvp); error = relookup(fdvp, &fvp, fcnp); if (error == 0) vrele(fdvp); if (fvp != NULL) { xp = VTOI(fvp); dp = VTOI(fdvp); } else { /* * From name has disappeared. IN_RENAME is not sufficient * to protect against directory races due to timing windows, * so we can't panic here. */ vrele(ap->a_fvp); return (0); } /* * Ensure that the directory entry still exists and has not * changed while the new name has been entered. If the source is * a file then the entry may have been unlinked or renamed. In * either case there is no further work to be done. If the source * is a directory then it cannot have been rmdir'ed; its link * count of three would cause a rmdir to fail with ENOTEMPTY. * The IN_RENAME flag ensures that it cannot be moved by another * rename. */ if (xp != ip) { /* * From name resolves to a different inode. IN_RENAME is * not sufficient protection against timing window races * so we can't panic here. */ } else { /* * If the source is a directory with a * new parent, the link count of the old * parent directory must be decremented * and ".." set to point to the new parent. */ if (doingdirectory && newparent) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; error = vn_rdwr(UIO_READ, fvp, (caddr_t)&dirbuf, sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); if (error == 0) { /* Like ufs little-endian: */ namlen = dirbuf.dotdot_type; if (namlen != 2 || dirbuf.dotdot_name[0] != '.' || dirbuf.dotdot_name[1] != '.') { ext2_dirbad(xp, (doff_t)12, "rename: mangled dir"); } else { dirbuf.dotdot_ino = newparent; (void) vn_rdwr(UIO_WRITE, fvp, (caddr_t)&dirbuf, sizeof(struct dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, tcnp->cn_cred, NOCRED, NULL, NULL); cache_purge(fdvp); } } } error = ext2_dirremove(fdvp, fcnp); if (!error) { xp->i_nlink--; xp->i_flag |= IN_CHANGE; } xp->i_flag &= ~IN_RENAME; } if (dp) vput(fdvp); if (xp) vput(fvp); vrele(ap->a_fvp); return (error); bad: if (xp) vput(ITOV(xp)); vput(ITOV(dp)); out: if (doingdirectory) ip->i_flag &= ~IN_RENAME; if (vn_lock(fvp, LK_EXCLUSIVE) == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; ip->i_flag &= ~IN_RENAME; vput(fvp); } else vrele(fvp); return (error); } /* * Mkdir system call */ static int ext2_mkdir(struct vop_mkdir_args *ap) { struct vnode *dvp = ap->a_dvp; struct vattr *vap = ap->a_vap; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; struct vnode *tvp; struct dirtemplate dirtemplate, *dtp; int error, dmode; #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_mkdir: no name"); #endif dp = VTOI(dvp); if ((nlink_t)dp->i_nlink >= EXT2_LINK_MAX) { error = EMLINK; goto out; } dmode = vap->va_mode & 0777; dmode |= IFDIR; /* * Must simulate part of ext2_makeinode here to acquire the inode, * but not have it entered in the parent directory. The entry is * made later after writing "." and ".." entries. */ error = ext2_valloc(dvp, dmode, cnp->cn_cred, &tvp); if (error) goto out; ip = VTOI(tvp); ip->i_gid = dp->i_gid; #ifdef SUIDDIR { /* * if we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * The new directory also inherits the SUID bit. * If user's UID and dir UID are the same, * 'give it away' so that the SUID is still forced on. */ if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (dp->i_mode & ISUID) && dp->i_uid) { dmode |= ISUID; ip->i_uid = dp->i_uid; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = dmode; tvp->v_type = VDIR; /* Rest init'd in getnewvnode(). */ ip->i_nlink = 2; if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; error = ext2_update(tvp, 1); /* * Bump link count in parent directory * to reflect work done below. Should * be done before reference is created * so reparation is possible if we crash. */ dp->i_nlink++; dp->i_flag |= IN_CHANGE; error = ext2_update(dvp, !DOINGASYNC(dvp)); if (error) goto bad; /* Initialize directory with "." and ".." from static template. */ if (EXT2_HAS_INCOMPAT_FEATURE(ip->i_e2fs, EXT2F_INCOMPAT_FTYPE)) dtp = &mastertemplate; else dtp = &omastertemplate; dirtemplate = *dtp; dirtemplate.dot_ino = ip->i_number; dirtemplate.dotdot_ino = dp->i_number; /* note that in ext2 DIRBLKSIZ == blocksize, not DEV_BSIZE * so let's just redefine it - for this function only */ #undef DIRBLKSIZ #define DIRBLKSIZ VTOI(dvp)->i_e2fs->e2fs_bsize dirtemplate.dotdot_reclen = DIRBLKSIZ - 12; error = vn_rdwr(UIO_WRITE, tvp, (caddr_t)&dirtemplate, sizeof(dirtemplate), (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_SYNC | IO_NOMACCHECK, cnp->cn_cred, NOCRED, NULL, NULL); if (error) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; goto bad; } if (DIRBLKSIZ > VFSTOEXT2(dvp->v_mount)->um_mountp->mnt_stat.f_bsize) /* XXX should grow with balloc() */ panic("ext2_mkdir: blksize"); else { ip->i_size = DIRBLKSIZ; ip->i_flag |= IN_CHANGE; } /* Directory set up, now install its entry in the parent directory. */ error = ext2_direnter(ip, dvp, cnp); if (error) { dp->i_nlink--; dp->i_flag |= IN_CHANGE; } bad: /* * No need to do an explicit VOP_TRUNCATE here, vrele will do this * for us because we set the link count to 0. */ if (error) { ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); } else *ap->a_vpp = tvp; out: return (error); #undef DIRBLKSIZ #define DIRBLKSIZ DEV_BSIZE } /* * Rmdir system call. */ static int ext2_rmdir(struct vop_rmdir_args *ap) { struct vnode *vp = ap->a_vp; struct vnode *dvp = ap->a_dvp; struct componentname *cnp = ap->a_cnp; struct inode *ip, *dp; int error; ip = VTOI(vp); dp = VTOI(dvp); /* * Verify the directory is empty (and valid). * (Rmdir ".." won't be valid since * ".." will contain a reference to * the current directory and thus be * non-empty.) */ if (ip->i_nlink != 2 || !ext2_dirempty(ip, dp->i_number, cnp->cn_cred)) { error = ENOTEMPTY; goto out; } if ((dp->i_flags & APPEND) || (ip->i_flags & (NOUNLINK | IMMUTABLE | APPEND))) { error = EPERM; goto out; } /* * Delete reference to directory before purging * inode. If we crash in between, the directory * will be reattached to lost+found, */ error = ext2_dirremove(dvp, cnp); if (error) goto out; dp->i_nlink--; dp->i_flag |= IN_CHANGE; cache_purge(dvp); VOP_UNLOCK(dvp, 0); /* * Truncate inode. The only stuff left * in the directory is "." and "..". The * "." reference is inconsequential since * we're quashing it. The ".." reference * has already been adjusted above. We've * removed the "." reference and the reference * in the parent directory, but there may be * other hard links so decrement by 2 and * worry about them later. */ ip->i_nlink -= 2; error = ext2_truncate(vp, (off_t)0, IO_SYNC, cnp->cn_cred, cnp->cn_thread); cache_purge(ITOV(ip)); if (vn_lock(dvp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { VOP_UNLOCK(vp, 0); vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } out: return (error); } /* * symlink -- make a symbolic link */ static int ext2_symlink(struct vop_symlink_args *ap) { struct vnode *vp, **vpp = ap->a_vpp; struct inode *ip; int len, error; error = ext2_makeinode(IFLNK | ap->a_vap->va_mode, ap->a_dvp, vpp, ap->a_cnp); if (error) return (error); vp = *vpp; len = strlen(ap->a_target); if (len < vp->v_mount->mnt_maxsymlinklen) { ip = VTOI(vp); bcopy(ap->a_target, (char *)ip->i_shortlink, len); ip->i_size = len; ip->i_flag |= IN_CHANGE | IN_UPDATE; } else error = vn_rdwr(UIO_WRITE, vp, ap->a_target, len, (off_t)0, UIO_SYSSPACE, IO_NODELOCKED | IO_NOMACCHECK, ap->a_cnp->cn_cred, NOCRED, NULL, NULL); if (error) vput(vp); return (error); } /* * Return target name of a symbolic link */ static int ext2_readlink(struct vop_readlink_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); int isize; isize = ip->i_size; if (isize < vp->v_mount->mnt_maxsymlinklen) { uiomove((char *)ip->i_shortlink, isize, ap->a_uio); return (0); } return (VOP_READ(vp, ap->a_uio, 0, ap->a_cred)); } /* * Calculate the logical to physical mapping if not done already, * then call the device strategy routine. * * In order to be able to swap to a file, the ext2_bmaparray() operation may not * deadlock on memory. See ext2_bmap() for details. */ static int ext2_strategy(struct vop_strategy_args *ap) { struct buf *bp = ap->a_bp; struct vnode *vp = ap->a_vp; struct bufobj *bo; daddr_t blkno; int error; if (vp->v_type == VBLK || vp->v_type == VCHR) panic("ext2_strategy: spec"); if (bp->b_blkno == bp->b_lblkno) { error = ext2_bmaparray(vp, bp->b_lblkno, &blkno, NULL, NULL); bp->b_blkno = blkno; if (error) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return (0); } if ((long)bp->b_blkno == -1) vfs_bio_clrbuf(bp); } if ((long)bp->b_blkno == -1) { bufdone(bp); return (0); } bp->b_iooffset = dbtob(bp->b_blkno); bo = VFSTOEXT2(vp->v_mount)->um_bo; BO_STRATEGY(bo, bp); return (0); } /* * Print out the contents of an inode. */ static int ext2_print(struct vop_print_args *ap) { struct vnode *vp = ap->a_vp; struct inode *ip = VTOI(vp); vn_printf(ip->i_devvp, "\tino %ju", (uintmax_t)ip->i_number); if (vp->v_type == VFIFO) fifo_printinfo(vp); printf("\n"); return (0); } /* * Close wrapper for fifos. * * Update the times on the inode then do device close. */ static int ext2fifo_close(struct vop_close_args *ap) { struct vnode *vp = ap->a_vp; VI_LOCK(vp); if (vp->v_usecount > 1) ext2_itimes_locked(vp); VI_UNLOCK(vp); return (fifo_specops.vop_close(ap)); } /* * Kqfilter wrapper for fifos. * * Fall through to ext2 kqfilter routines if needed */ static int ext2fifo_kqfilter(struct vop_kqfilter_args *ap) { int error; error = fifo_specops.vop_kqfilter(ap); if (error) error = vfs_kqfilter(ap); return (error); } /* * Return POSIX pathconf information applicable to ext2 filesystems. */ static int ext2_pathconf(struct vop_pathconf_args *ap) { int error = 0; switch (ap->a_name) { case _PC_LINK_MAX: *ap->a_retval = EXT2_LINK_MAX; break; case _PC_NAME_MAX: *ap->a_retval = NAME_MAX; break; case _PC_PATH_MAX: *ap->a_retval = PATH_MAX; break; case _PC_PIPE_BUF: *ap->a_retval = PIPE_BUF; break; case _PC_CHOWN_RESTRICTED: *ap->a_retval = 1; break; case _PC_NO_TRUNC: *ap->a_retval = 1; break; case _PC_MIN_HOLE_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_ASYNC_IO: /* _PC_ASYNC_IO should have been handled by upper layers. */ KASSERT(0, ("_PC_ASYNC_IO should not get here")); error = EINVAL; break; case _PC_PRIO_IO: *ap->a_retval = 0; break; case _PC_SYNC_IO: *ap->a_retval = 0; break; case _PC_ALLOC_SIZE_MIN: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_bsize; break; case _PC_FILESIZEBITS: *ap->a_retval = 64; break; case _PC_REC_INCR_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_MAX_XFER_SIZE: *ap->a_retval = -1; /* means ``unlimited'' */ break; case _PC_REC_MIN_XFER_SIZE: *ap->a_retval = ap->a_vp->v_mount->mnt_stat.f_iosize; break; case _PC_REC_XFER_ALIGN: *ap->a_retval = PAGE_SIZE; break; case _PC_SYMLINK_MAX: *ap->a_retval = MAXPATHLEN; break; default: error = EINVAL; break; } return (error); } /* * Vnode pointer to File handle */ /* ARGSUSED */ static int ext2_vptofh(struct vop_vptofh_args *ap) { struct inode *ip; struct ufid *ufhp; ip = VTOI(ap->a_vp); ufhp = (struct ufid *)ap->a_fhp; ufhp->ufid_len = sizeof(struct ufid); ufhp->ufid_ino = ip->i_number; ufhp->ufid_gen = ip->i_gen; return (0); } /* * Initialize the vnode associated with a new inode, handle aliased * vnodes. */ int ext2_vinit(struct mount *mntp, struct vop_vector *fifoops, struct vnode **vpp) { struct inode *ip; struct vnode *vp; vp = *vpp; ip = VTOI(vp); vp->v_type = IFTOVT(ip->i_mode); if (vp->v_type == VFIFO) vp->v_op = fifoops; if (ip->i_number == EXT2_ROOTINO) vp->v_vflag |= VV_ROOT; ip->i_modrev = init_va_filerev(); *vpp = vp; return (0); } /* * Allocate a new inode. */ static int ext2_makeinode(int mode, struct vnode *dvp, struct vnode **vpp, struct componentname *cnp) { struct inode *ip, *pdir; struct vnode *tvp; int error; pdir = VTOI(dvp); #ifdef INVARIANTS if ((cnp->cn_flags & HASBUF) == 0) panic("ext2_makeinode: no name"); #endif *vpp = NULL; if ((mode & IFMT) == 0) mode |= IFREG; error = ext2_valloc(dvp, mode, cnp->cn_cred, &tvp); if (error) { return (error); } ip = VTOI(tvp); ip->i_gid = pdir->i_gid; #ifdef SUIDDIR { /* * if we are * not the owner of the directory, * and we are hacking owners here, (only do this where told to) * and we are not giving it TOO root, (would subvert quotas) * then go ahead and give it to the other user. * Note that this drops off the execute bits for security. */ if ( (dvp->v_mount->mnt_flag & MNT_SUIDDIR) && (pdir->i_mode & ISUID) && (pdir->i_uid != cnp->cn_cred->cr_uid) && pdir->i_uid) { ip->i_uid = pdir->i_uid; mode &= ~07111; } else { ip->i_uid = cnp->cn_cred->cr_uid; } } #else ip->i_uid = cnp->cn_cred->cr_uid; #endif ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; ip->i_mode = mode; tvp->v_type = IFTOVT(mode); /* Rest init'd in getnewvnode(). */ ip->i_nlink = 1; if ((ip->i_mode & ISGID) && !groupmember(ip->i_gid, cnp->cn_cred)) { if (priv_check_cred(cnp->cn_cred, PRIV_VFS_RETAINSUGID, 0)) ip->i_mode &= ~ISGID; } if (cnp->cn_flags & ISWHITEOUT) ip->i_flags |= UF_OPAQUE; /* * Make sure inode goes to disk before directory entry. */ error = ext2_update(tvp, !DOINGASYNC(tvp)); if (error) goto bad; error = ext2_direnter(ip, dvp, cnp); if (error) goto bad; *vpp = tvp; return (0); bad: /* * Write error occurred trying to update the inode * or the directory so must deallocate the inode. */ ip->i_nlink = 0; ip->i_flag |= IN_CHANGE; vput(tvp); return (error); } /* * Vnode op for reading. */ static int ext2_read(struct vop_read_args *ap) { struct vnode *vp; struct inode *ip; int error; vp = ap->a_vp; ip = VTOI(vp); /*EXT4_EXT_LOCK(ip);*/ if (ip->i_flag & IN_E4EXTENTS) error = ext4_ext_read(ap); else error = ext2_ind_read(ap); /*EXT4_EXT_UNLOCK(ip);*/ return (error); } /* * Vnode op for reading. */ static int ext2_ind_read(struct vop_read_args *ap) { struct vnode *vp; struct inode *ip; struct uio *uio; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn, nextlbn; off_t bytesinfile; long size, xfersize, blkoffset; int error, orig_resid, seqcount; int ioflag; vp = ap->a_vp; uio = ap->a_uio; ioflag = ap->a_ioflag; seqcount = ap->a_ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_READ) panic("%s: mode", "ext2_read"); if (vp->v_type == VLNK) { if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) panic("%s: short symlink", "ext2_read"); } else if (vp->v_type != VREG && vp->v_type != VDIR) panic("%s: type %d", "ext2_read", vp->v_type); #endif orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("ext2_read: uio->uio_resid < 0")); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("ext2_read: uio->uio_offset < 0")); fs = ip->i_e2fs; if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize) return (EOVERFLOW); for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); nextlbn = lbn + 1; size = blksize(fs, ip, lbn); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (bytesinfile < xfersize) xfersize = bytesinfile; if (lblktosize(fs, nextlbn) >= ip->i_size) error = bread(vp, lbn, size, NOCRED, &bp); else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { error = cluster_read(vp, ip->i_size, lbn, size, NOCRED, blkoffset + uio->uio_resid, seqcount, 0, &bp); } else if (seqcount > 1) { u_int nextsize = blksize(fs, ip, nextlbn); error = breadn(vp, lbn, size, &nextlbn, &nextsize, 1, NOCRED, &bp); } else error = bread(vp, lbn, size, NOCRED, &bp); if (error) { brelse(bp); bp = NULL; break; } /* - * If IO_DIRECT then set B_DIRECT for the buffer. This - * will cause us to attempt to release the buffer later on - * and will cause the buffer cache to attempt to free the - * underlying pages. - */ - if (ioflag & IO_DIRECT) - bp->b_flags |= B_DIRECT; - - /* * We should only get non-zero b_resid when an I/O error * has occurred, which should cause us to break above. * However, if the short read did not cause an error, * then we want to ensure that we do not uiomove bad * or uninitialized data. */ size -= bp->b_resid; if (size < xfersize) { if (size == 0) break; xfersize = size; } error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); if (error) break; - - if (ioflag & (IO_VMIO|IO_DIRECT)) { - /* - * If it's VMIO or direct I/O, then we don't - * need the buf, mark it available for - * freeing. If it's non-direct VMIO, the VM has - * the data. - */ - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - /* - * Otherwise let whoever - * made the request take care of - * freeing it. We just queue - * it onto another list. - */ - bqrelse(bp); - } + vfs_bio_brelse(bp, ioflag); } /* * This can only happen in the case of an error * because the loop above resets bp to NULL on each iteration * and on normal completion has not set a new value into it. * so it must have come from a 'break' statement */ - if (bp != NULL) { - if (ioflag & (IO_VMIO|IO_DIRECT)) { - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - bqrelse(bp); - } - } + if (bp != NULL) + vfs_bio_brelse(bp, ioflag); if ((error == 0 || uio->uio_resid != orig_resid) && (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) ip->i_flag |= IN_ACCESS; return (error); } static int ext2_ioctl(struct vop_ioctl_args *ap) { switch (ap->a_command) { case FIOSEEKDATA: case FIOSEEKHOLE: return (vn_bmap_seekhole(ap->a_vp, ap->a_command, (off_t *)ap->a_data, ap->a_cred)); default: return (ENOTTY); } } /* * this function handles ext4 extents block mapping */ static int ext4_ext_read(struct vop_read_args *ap) { static unsigned char zeroes[EXT2_MAX_BLOCK_SIZE]; struct vnode *vp; struct inode *ip; struct uio *uio; struct m_ext2fs *fs; struct buf *bp; struct ext4_extent nex, *ep; struct ext4_extent_path path; daddr_t lbn, newblk; off_t bytesinfile; int cache_type; ssize_t orig_resid; int error; long size, xfersize, blkoffset; vp = ap->a_vp; ip = VTOI(vp); uio = ap->a_uio; memset(&path, 0, sizeof(path)); orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("%s: uio->uio_resid < 0", __func__)); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("%s: uio->uio_offset < 0", __func__)); fs = ip->i_e2fs; if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->e2fs_maxfilesize) return (EOVERFLOW); while (uio->uio_resid > 0) { if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); size = blksize(fs, ip, lbn); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; xfersize = MIN(xfersize, uio->uio_resid); xfersize = MIN(xfersize, bytesinfile); /* get block from ext4 extent cache */ cache_type = ext4_ext_in_cache(ip, lbn, &nex); switch (cache_type) { case EXT4_EXT_CACHE_NO: ext4_ext_find_extent(fs, ip, lbn, &path); if (path.ep_is_sparse) ep = &path.ep_sparse_ext; else ep = path.ep_ext; if (ep == NULL) return (EIO); ext4_ext_put_cache(ip, ep, path.ep_is_sparse ? EXT4_EXT_CACHE_GAP : EXT4_EXT_CACHE_IN); newblk = lbn - ep->e_blk + (ep->e_start_lo | (daddr_t)ep->e_start_hi << 32); if (path.ep_bp != NULL) { brelse(path.ep_bp); path.ep_bp = NULL; } break; case EXT4_EXT_CACHE_GAP: /* block has not been allocated yet */ break; case EXT4_EXT_CACHE_IN: newblk = lbn - nex.e_blk + (nex.e_start_lo | (daddr_t)nex.e_start_hi << 32); break; default: panic("%s: invalid cache type", __func__); } if (cache_type == EXT4_EXT_CACHE_GAP || (cache_type == EXT4_EXT_CACHE_NO && path.ep_is_sparse)) { if (xfersize > sizeof(zeroes)) xfersize = sizeof(zeroes); error = uiomove(zeroes, xfersize, uio); if (error) return (error); } else { error = bread(ip->i_devvp, fsbtodb(fs, newblk), size, NOCRED, &bp); if (error) { brelse(bp); return (error); } size -= bp->b_resid; if (size < xfersize) { if (size == 0) { bqrelse(bp); break; } xfersize = size; } error = uiomove(bp->b_data + blkoffset, xfersize, uio); bqrelse(bp); if (error) return (error); } } return (0); } /* * Vnode op for writing. */ static int ext2_write(struct vop_write_args *ap) { struct vnode *vp; struct uio *uio; struct inode *ip; struct m_ext2fs *fs; struct buf *bp; daddr_t lbn; off_t osize; int blkoffset, error, flags, ioflag, resid, size, seqcount, xfersize; ioflag = ap->a_ioflag; uio = ap->a_uio; vp = ap->a_vp; seqcount = ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE) panic("%s: mode", "ext2_write"); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = ip->i_size; if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) return (EPERM); /* FALLTHROUGH */ case VLNK: break; case VDIR: /* XXX differs from ffs -- this is called from ext2_mkdir(). */ if ((ioflag & IO_SYNC) == 0) panic("ext2_write: nonsync dir write"); break; default: panic("ext2_write: type %p %d (%jd,%jd)", (void *)vp, vp->v_type, (intmax_t)uio->uio_offset, (intmax_t)uio->uio_resid); } KASSERT(uio->uio_resid >= 0, ("ext2_write: uio->uio_resid < 0")); KASSERT(uio->uio_offset >= 0, ("ext2_write: uio->uio_offset < 0")); fs = ip->i_e2fs; if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->e2fs_maxfilesize) return (EFBIG); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, I don't think it matters. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); resid = uio->uio_resid; osize = ip->i_size; if (seqcount > BA_SEQMAX) flags = BA_SEQMAX << BA_SEQSHIFT; else flags = seqcount << BA_SEQSHIFT; if ((ioflag & IO_SYNC) && !DOINGASYNC(vp)) flags |= IO_SYNC; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->e2fs_fsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (uio->uio_offset + xfersize > ip->i_size) vnode_pager_setsize(vp, uio->uio_offset + xfersize); /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->e2fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; error = ext2_balloc(ip, lbn, blkoffset + xfersize, ap->a_cred, &bp, flags); if (error != 0) break; if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) bp->b_flags |= B_NOCACHE; if (uio->uio_offset + xfersize > ip->i_size) ip->i_size = uio->uio_offset + xfersize; size = blksize(fs, ip, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); /* * If the buffer is not already filled and we encounter an * error while trying to fill it, we have to clear out any * garbage data from the pages instantiated for the buffer. * If we do not, a failed uiomove() during a write can leave * the prior contents of the pages exposed to a userland mmap. * * Note that we need only clear buffers with a transfer size * equal to the block size because buffers with a shorter * transfer size were cleared above by the call to ext2_balloc() * with the BA_CLRBUF flag set. * * If the source region for uiomove identically mmaps the * buffer, uiomove() performed the NOP copy, and the buffer * content remains valid because the page fault handler * validated the pages. */ if (error != 0 && (bp->b_flags & B_CACHE) == 0 && fs->e2fs_bsize == xfersize) vfs_bio_clrbuf(bp); - if (ioflag & (IO_VMIO|IO_DIRECT)) { - bp->b_flags |= B_RELBUF; - } + + vfs_bio_set_flags(bp, ioflag); /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || (ioflag & IO_ASYNC)) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else if (xfersize + blkoffset == fs->e2fs_fsize) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { bp->b_flags |= B_CLUSTEROK; cluster_write(vp, bp, ip->i_size, seqcount, 0); } else { bawrite(bp); } } else if (ioflag & IO_DIRECT) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else { bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } if (error || xfersize == 0) break; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ap->a_cred) { if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) ip->i_mode &= ~(ISUID | ISGID); } if (error) { if (ioflag & IO_UNIT) { (void)ext2_truncate(vp, osize, ioflag & IO_SYNC, ap->a_cred, uio->uio_td); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } if (uio->uio_resid != resid) { ip->i_flag |= IN_CHANGE | IN_UPDATE; if (ioflag & IO_SYNC) error = ext2_update(vp, 1); } return (error); } Index: head/sys/kern/vfs_bio.c =================================================================== --- head/sys/kern/vfs_bio.c (revision 309061) +++ head/sys/kern/vfs_bio.c (revision 309062) @@ -1,4963 +1,5002 @@ /*- * Copyright (c) 2004 Poul-Henning Kamp * Copyright (c) 1994,1997 John S. Dyson * Copyright (c) 2013 The FreeBSD Foundation * All rights reserved. * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * this file contains a new buffer I/O scheme implementing a coherent * VM object and buffer cache scheme. Pains have been taken to make * sure that the performance degradation associated with schemes such * as this is not realized. * * Author: John S. Dyson * Significant help during the development and debugging phases * had been provided by David Greenman, also of the FreeBSD core team. * * see man buf(9) for more info. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_compat.h" #include "opt_swap.h" static MALLOC_DEFINE(M_BIOBUF, "biobuf", "BIO buffer"); struct bio_ops bioops; /* I/O operation notification */ struct buf_ops buf_ops_bio = { .bop_name = "buf_ops_bio", .bop_write = bufwrite, .bop_strategy = bufstrategy, .bop_sync = bufsync, .bop_bdflush = bufbdflush, }; static struct buf *buf; /* buffer header pool */ extern struct buf *swbuf; /* Swap buffer header pool. */ caddr_t unmapped_buf; /* Used below and for softdep flushing threads in ufs/ffs/ffs_softdep.c */ struct proc *bufdaemonproc; struct proc *bufspacedaemonproc; static int inmem(struct vnode *vp, daddr_t blkno); static void vm_hold_free_pages(struct buf *bp, int newbsize); static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to); static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m); static void vfs_clean_pages_dirty_buf(struct buf *bp); static void vfs_setdirty_locked_object(struct buf *bp); static void vfs_vmio_invalidate(struct buf *bp); static void vfs_vmio_truncate(struct buf *bp, int npages); static void vfs_vmio_extend(struct buf *bp, int npages, int size); static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno); static int buf_flush(struct vnode *vp, int); static int buf_recycle(bool); static int buf_scan(bool); static int flushbufqueues(struct vnode *, int, int); static void buf_daemon(void); static void bremfreel(struct buf *bp); static __inline void bd_wakeup(void); static int sysctl_runningspace(SYSCTL_HANDLER_ARGS); static void bufkva_reclaim(vmem_t *, int); static void bufkva_free(struct buf *); static int buf_import(void *, void **, int, int); static void buf_release(void *, void **, int); #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS); #endif int vmiodirenable = TRUE; SYSCTL_INT(_vfs, OID_AUTO, vmiodirenable, CTLFLAG_RW, &vmiodirenable, 0, "Use the VM system for directory writes"); long runningbufspace; SYSCTL_LONG(_vfs, OID_AUTO, runningbufspace, CTLFLAG_RD, &runningbufspace, 0, "Amount of presently outstanding async buffer io"); static long bufspace; #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) SYSCTL_PROC(_vfs, OID_AUTO, bufspace, CTLTYPE_LONG|CTLFLAG_MPSAFE|CTLFLAG_RD, &bufspace, 0, sysctl_bufspace, "L", "Virtual memory used for buffers"); #else SYSCTL_LONG(_vfs, OID_AUTO, bufspace, CTLFLAG_RD, &bufspace, 0, "Physical memory used for buffers"); #endif static long bufkvaspace; SYSCTL_LONG(_vfs, OID_AUTO, bufkvaspace, CTLFLAG_RD, &bufkvaspace, 0, "Kernel virtual memory used for buffers"); static long maxbufspace; SYSCTL_LONG(_vfs, OID_AUTO, maxbufspace, CTLFLAG_RW, &maxbufspace, 0, "Maximum allowed value of bufspace (including metadata)"); static long bufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, bufmallocspace, CTLFLAG_RD, &bufmallocspace, 0, "Amount of malloced memory for buffers"); static long maxbufmallocspace; SYSCTL_LONG(_vfs, OID_AUTO, maxmallocbufspace, CTLFLAG_RW, &maxbufmallocspace, 0, "Maximum amount of malloced memory for buffers"); static long lobufspace; SYSCTL_LONG(_vfs, OID_AUTO, lobufspace, CTLFLAG_RW, &lobufspace, 0, "Minimum amount of buffers we want to have"); long hibufspace; SYSCTL_LONG(_vfs, OID_AUTO, hibufspace, CTLFLAG_RW, &hibufspace, 0, "Maximum allowed value of bufspace (excluding metadata)"); long bufspacethresh; SYSCTL_LONG(_vfs, OID_AUTO, bufspacethresh, CTLFLAG_RW, &bufspacethresh, 0, "Bufspace consumed before waking the daemon to free some"); static int buffreekvacnt; SYSCTL_INT(_vfs, OID_AUTO, buffreekvacnt, CTLFLAG_RW, &buffreekvacnt, 0, "Number of times we have freed the KVA space from some buffer"); static int bufdefragcnt; SYSCTL_INT(_vfs, OID_AUTO, bufdefragcnt, CTLFLAG_RW, &bufdefragcnt, 0, "Number of times we have had to repeat buffer allocation to defragment"); static long lorunningspace; SYSCTL_PROC(_vfs, OID_AUTO, lorunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &lorunningspace, 0, sysctl_runningspace, "L", "Minimum preferred space used for in-progress I/O"); static long hirunningspace; SYSCTL_PROC(_vfs, OID_AUTO, hirunningspace, CTLTYPE_LONG | CTLFLAG_MPSAFE | CTLFLAG_RW, &hirunningspace, 0, sysctl_runningspace, "L", "Maximum amount of space to use for in-progress I/O"); int dirtybufferflushes; SYSCTL_INT(_vfs, OID_AUTO, dirtybufferflushes, CTLFLAG_RW, &dirtybufferflushes, 0, "Number of bdwrite to bawrite conversions to limit dirty buffers"); int bdwriteskip; SYSCTL_INT(_vfs, OID_AUTO, bdwriteskip, CTLFLAG_RW, &bdwriteskip, 0, "Number of buffers supplied to bdwrite with snapshot deadlock risk"); int altbufferflushes; SYSCTL_INT(_vfs, OID_AUTO, altbufferflushes, CTLFLAG_RW, &altbufferflushes, 0, "Number of fsync flushes to limit dirty buffers"); static int recursiveflushes; SYSCTL_INT(_vfs, OID_AUTO, recursiveflushes, CTLFLAG_RW, &recursiveflushes, 0, "Number of flushes skipped due to being recursive"); static int numdirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, numdirtybuffers, CTLFLAG_RD, &numdirtybuffers, 0, "Number of buffers that are dirty (has unwritten changes) at the moment"); static int lodirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, lodirtybuffers, CTLFLAG_RW, &lodirtybuffers, 0, "How many buffers we want to have free before bufdaemon can sleep"); static int hidirtybuffers; SYSCTL_INT(_vfs, OID_AUTO, hidirtybuffers, CTLFLAG_RW, &hidirtybuffers, 0, "When the number of dirty buffers is considered severe"); int dirtybufthresh; SYSCTL_INT(_vfs, OID_AUTO, dirtybufthresh, CTLFLAG_RW, &dirtybufthresh, 0, "Number of bdwrite to bawrite conversions to clear dirty buffers"); static int numfreebuffers; SYSCTL_INT(_vfs, OID_AUTO, numfreebuffers, CTLFLAG_RD, &numfreebuffers, 0, "Number of free buffers"); static int lofreebuffers; SYSCTL_INT(_vfs, OID_AUTO, lofreebuffers, CTLFLAG_RW, &lofreebuffers, 0, "Target number of free buffers"); static int hifreebuffers; SYSCTL_INT(_vfs, OID_AUTO, hifreebuffers, CTLFLAG_RW, &hifreebuffers, 0, "Threshold for clean buffer recycling"); static int getnewbufcalls; SYSCTL_INT(_vfs, OID_AUTO, getnewbufcalls, CTLFLAG_RW, &getnewbufcalls, 0, "Number of calls to getnewbuf"); static int getnewbufrestarts; SYSCTL_INT(_vfs, OID_AUTO, getnewbufrestarts, CTLFLAG_RW, &getnewbufrestarts, 0, "Number of times getnewbuf has had to restart a buffer acquisition"); static int mappingrestarts; SYSCTL_INT(_vfs, OID_AUTO, mappingrestarts, CTLFLAG_RW, &mappingrestarts, 0, "Number of times getblk has had to restart a buffer mapping for " "unmapped buffer"); static int numbufallocfails; SYSCTL_INT(_vfs, OID_AUTO, numbufallocfails, CTLFLAG_RW, &numbufallocfails, 0, "Number of times buffer allocations failed"); static int flushbufqtarget = 100; SYSCTL_INT(_vfs, OID_AUTO, flushbufqtarget, CTLFLAG_RW, &flushbufqtarget, 0, "Amount of work to do in flushbufqueues when helping bufdaemon"); static long notbufdflushes; SYSCTL_LONG(_vfs, OID_AUTO, notbufdflushes, CTLFLAG_RD, ¬bufdflushes, 0, "Number of dirty buffer flushes done by the bufdaemon helpers"); static long barrierwrites; SYSCTL_LONG(_vfs, OID_AUTO, barrierwrites, CTLFLAG_RW, &barrierwrites, 0, "Number of barrier writes"); SYSCTL_INT(_vfs, OID_AUTO, unmapped_buf_allowed, CTLFLAG_RD, &unmapped_buf_allowed, 0, "Permit the use of the unmapped i/o"); /* * This lock synchronizes access to bd_request. */ static struct mtx_padalign bdlock; /* * This lock protects the runningbufreq and synchronizes runningbufwakeup and * waitrunningbufspace(). */ static struct mtx_padalign rbreqlock; /* * Lock that protects needsbuffer and the sleeps/wakeups surrounding it. */ static struct rwlock_padalign nblock; /* * Lock that protects bdirtywait. */ static struct mtx_padalign bdirtylock; /* * Wakeup point for bufdaemon, as well as indicator of whether it is already * active. Set to 1 when the bufdaemon is already "on" the queue, 0 when it * is idling. */ static int bd_request; /* * Request/wakeup point for the bufspace daemon. */ static int bufspace_request; /* * Request for the buf daemon to write more buffers than is indicated by * lodirtybuf. This may be necessary to push out excess dependencies or * defragment the address space where a simple count of the number of dirty * buffers is insufficient to characterize the demand for flushing them. */ static int bd_speedupreq; /* * bogus page -- for I/O to/from partially complete buffers * this is a temporary solution to the problem, but it is not * really that bad. it would be better to split the buffer * for input in the case of buffers partially already in memory, * but the code is intricate enough already. */ vm_page_t bogus_page; /* * Synchronization (sleep/wakeup) variable for active buffer space requests. * Set when wait starts, cleared prior to wakeup(). * Used in runningbufwakeup() and waitrunningbufspace(). */ static int runningbufreq; /* * Synchronization (sleep/wakeup) variable for buffer requests. * Can contain the VFS_BIO_NEED flags defined below; setting/clearing is done * by and/or. * Used in numdirtywakeup(), bufspace_wakeup(), bwillwrite(), * getnewbuf(), and getblk(). */ static volatile int needsbuffer; /* * Synchronization for bwillwrite() waiters. */ static int bdirtywait; /* * Definitions for the buffer free lists. */ #define QUEUE_NONE 0 /* on no queue */ #define QUEUE_EMPTY 1 /* empty buffer headers */ #define QUEUE_DIRTY 2 /* B_DELWRI buffers */ #define QUEUE_CLEAN 3 /* non-B_DELWRI buffers */ #define QUEUE_SENTINEL 1024 /* not an queue index, but mark for sentinel */ /* Maximum number of clean buffer queues. */ #define CLEAN_QUEUES 16 /* Configured number of clean queues. */ static int clean_queues; /* Maximum number of buffer queues. */ #define BUFFER_QUEUES (QUEUE_CLEAN + CLEAN_QUEUES) /* Queues for free buffers with various properties */ static TAILQ_HEAD(bqueues, buf) bufqueues[BUFFER_QUEUES] = { { 0 } }; #ifdef INVARIANTS static int bq_len[BUFFER_QUEUES]; #endif /* * Lock for each bufqueue */ static struct mtx_padalign bqlocks[BUFFER_QUEUES]; /* * per-cpu empty buffer cache. */ uma_zone_t buf_zone; /* * Single global constant for BUF_WMESG, to avoid getting multiple references. * buf_wmesg is referred from macros. */ const char *buf_wmesg = BUF_WMESG; static int sysctl_runningspace(SYSCTL_HANDLER_ARGS) { long value; int error; value = *(long *)arg1; error = sysctl_handle_long(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL) return (error); mtx_lock(&rbreqlock); if (arg1 == &hirunningspace) { if (value < lorunningspace) error = EINVAL; else hirunningspace = value; } else { KASSERT(arg1 == &lorunningspace, ("%s: unknown arg1", __func__)); if (value > hirunningspace) error = EINVAL; else lorunningspace = value; } mtx_unlock(&rbreqlock); return (error); } #if defined(COMPAT_FREEBSD4) || defined(COMPAT_FREEBSD5) || \ defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD7) static int sysctl_bufspace(SYSCTL_HANDLER_ARGS) { long lvalue; int ivalue; if (sizeof(int) == sizeof(long) || req->oldlen >= sizeof(long)) return (sysctl_handle_long(oidp, arg1, arg2, req)); lvalue = *(long *)arg1; if (lvalue > INT_MAX) /* On overflow, still write out a long to trigger ENOMEM. */ return (sysctl_handle_long(oidp, &lvalue, 0, req)); ivalue = lvalue; return (sysctl_handle_int(oidp, &ivalue, 0, req)); } #endif static int bqcleanq(void) { static int nextq; return ((atomic_fetchadd_int(&nextq, 1) % clean_queues) + QUEUE_CLEAN); } static int bqisclean(int qindex) { return (qindex >= QUEUE_CLEAN && qindex < QUEUE_CLEAN + CLEAN_QUEUES); } /* * bqlock: * * Return the appropriate queue lock based on the index. */ static inline struct mtx * bqlock(int qindex) { return (struct mtx *)&bqlocks[qindex]; } /* * bdirtywakeup: * * Wakeup any bwillwrite() waiters. */ static void bdirtywakeup(void) { mtx_lock(&bdirtylock); if (bdirtywait) { bdirtywait = 0; wakeup(&bdirtywait); } mtx_unlock(&bdirtylock); } /* * bdirtysub: * * Decrement the numdirtybuffers count by one and wakeup any * threads blocked in bwillwrite(). */ static void bdirtysub(void) { if (atomic_fetchadd_int(&numdirtybuffers, -1) == (lodirtybuffers + hidirtybuffers) / 2) bdirtywakeup(); } /* * bdirtyadd: * * Increment the numdirtybuffers count by one and wakeup the buf * daemon if needed. */ static void bdirtyadd(void) { /* * Only do the wakeup once as we cross the boundary. The * buf daemon will keep running until the condition clears. */ if (atomic_fetchadd_int(&numdirtybuffers, 1) == (lodirtybuffers + hidirtybuffers) / 2) bd_wakeup(); } /* * bufspace_wakeup: * * Called when buffer space is potentially available for recovery. * getnewbuf() will block on this flag when it is unable to free * sufficient buffer space. Buffer space becomes recoverable when * bp's get placed back in the queues. */ static void bufspace_wakeup(void) { /* * If someone is waiting for bufspace, wake them up. * * Since needsbuffer is set prior to doing an additional queue * scan it is safe to check for the flag prior to acquiring the * lock. The thread that is preparing to scan again before * blocking would discover the buf we released. */ if (needsbuffer) { rw_rlock(&nblock); if (atomic_cmpset_int(&needsbuffer, 1, 0) == 1) wakeup(__DEVOLATILE(void *, &needsbuffer)); rw_runlock(&nblock); } } /* * bufspace_daemonwakeup: * * Wakeup the daemon responsible for freeing clean bufs. */ static void bufspace_daemonwakeup(void) { rw_rlock(&nblock); if (bufspace_request == 0) { bufspace_request = 1; wakeup(&bufspace_request); } rw_runlock(&nblock); } /* * bufspace_adjust: * * Adjust the reported bufspace for a KVA managed buffer, possibly * waking any waiters. */ static void bufspace_adjust(struct buf *bp, int bufsize) { long space; int diff; KASSERT((bp->b_flags & B_MALLOC) == 0, ("bufspace_adjust: malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) { atomic_subtract_long(&bufspace, -diff); bufspace_wakeup(); } else { space = atomic_fetchadd_long(&bufspace, diff); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + diff >= bufspacethresh) bufspace_daemonwakeup(); } bp->b_bufsize = bufsize; } /* * bufspace_reserve: * * Reserve bufspace before calling allocbuf(). metadata has a * different space limit than data. */ static int bufspace_reserve(int size, bool metadata) { long limit; long space; if (metadata) limit = maxbufspace; else limit = hibufspace; do { space = bufspace; if (space + size > limit) return (ENOSPC); } while (atomic_cmpset_long(&bufspace, space, space + size) == 0); /* Wake up the daemon on the transition. */ if (space < bufspacethresh && space + size >= bufspacethresh) bufspace_daemonwakeup(); return (0); } /* * bufspace_release: * * Release reserved bufspace after bufspace_adjust() has consumed it. */ static void bufspace_release(int size) { atomic_subtract_long(&bufspace, size); bufspace_wakeup(); } /* * bufspace_wait: * * Wait for bufspace, acting as the buf daemon if a locked vnode is * supplied. needsbuffer must be set in a safe fashion prior to * polling for space. The operation must be re-tried on return. */ static void bufspace_wait(struct vnode *vp, int gbflags, int slpflag, int slptimeo) { struct thread *td; int error, fl, norunbuf; if ((gbflags & GB_NOWAIT_BD) != 0) return; td = curthread; rw_wlock(&nblock); while (needsbuffer != 0) { if (vp != NULL && vp->v_type != VCHR && (td->td_pflags & TDP_BUFNEED) == 0) { rw_wunlock(&nblock); /* * getblk() is called with a vnode locked, and * some majority of the dirty buffers may as * well belong to the vnode. Flushing the * buffers there would make a progress that * cannot be achieved by the buf_daemon, that * cannot lock the vnode. */ norunbuf = ~(TDP_BUFNEED | TDP_NORUNNINGBUF) | (td->td_pflags & TDP_NORUNNINGBUF); /* * Play bufdaemon. The getnewbuf() function * may be called while the thread owns lock * for another dirty buffer for the same * vnode, which makes it impossible to use * VOP_FSYNC() there, due to the buffer lock * recursion. */ td->td_pflags |= TDP_BUFNEED | TDP_NORUNNINGBUF; fl = buf_flush(vp, flushbufqtarget); td->td_pflags &= norunbuf; rw_wlock(&nblock); if (fl != 0) continue; if (needsbuffer == 0) break; } error = rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, (PRIBIO + 4) | slpflag, "newbuf", slptimeo); if (error != 0) break; } rw_wunlock(&nblock); } /* * bufspace_daemon: * * buffer space management daemon. Tries to maintain some marginal * amount of free buffer space so that requesting processes neither * block nor work to reclaim buffers. */ static void bufspace_daemon(void) { for (;;) { kproc_suspend_check(bufspacedaemonproc); /* * Free buffers from the clean queue until we meet our * targets. * * Theory of operation: The buffer cache is most efficient * when some free buffer headers and space are always * available to getnewbuf(). This daemon attempts to prevent * the excessive blocking and synchronization associated * with shortfall. It goes through three phases according * demand: * * 1) The daemon wakes up voluntarily once per-second * during idle periods when the counters are below * the wakeup thresholds (bufspacethresh, lofreebuffers). * * 2) The daemon wakes up as we cross the thresholds * ahead of any potential blocking. This may bounce * slightly according to the rate of consumption and * release. * * 3) The daemon and consumers are starved for working * clean buffers. This is the 'bufspace' sleep below * which will inefficiently trade bufs with bqrelse * until we return to condition 2. */ while (bufspace > lobufspace || numfreebuffers < hifreebuffers) { if (buf_recycle(false) != 0) { atomic_set_int(&needsbuffer, 1); if (buf_recycle(false) != 0) { rw_wlock(&nblock); if (needsbuffer) rw_sleep(__DEVOLATILE(void *, &needsbuffer), &nblock, PRIBIO|PDROP, "bufspace", hz/10); else rw_wunlock(&nblock); } } maybe_yield(); } /* * Re-check our limits under the exclusive nblock. */ rw_wlock(&nblock); if (bufspace < bufspacethresh && numfreebuffers > lofreebuffers) { bufspace_request = 0; rw_sleep(&bufspace_request, &nblock, PRIBIO|PDROP, "-", hz); } else rw_wunlock(&nblock); } } static struct kproc_desc bufspace_kp = { "bufspacedaemon", bufspace_daemon, &bufspacedaemonproc }; SYSINIT(bufspacedaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &bufspace_kp); /* * bufmallocadjust: * * Adjust the reported bufspace for a malloc managed buffer, possibly * waking any waiters. */ static void bufmallocadjust(struct buf *bp, int bufsize) { int diff; KASSERT((bp->b_flags & B_MALLOC) != 0, ("bufmallocadjust: non-malloc buf %p", bp)); diff = bufsize - bp->b_bufsize; if (diff < 0) atomic_subtract_long(&bufmallocspace, -diff); else atomic_add_long(&bufmallocspace, diff); bp->b_bufsize = bufsize; } /* * runningwakeup: * * Wake up processes that are waiting on asynchronous writes to fall * below lorunningspace. */ static void runningwakeup(void) { mtx_lock(&rbreqlock); if (runningbufreq) { runningbufreq = 0; wakeup(&runningbufreq); } mtx_unlock(&rbreqlock); } /* * runningbufwakeup: * * Decrement the outstanding write count according. */ void runningbufwakeup(struct buf *bp) { long space, bspace; bspace = bp->b_runningbufspace; if (bspace == 0) return; space = atomic_fetchadd_long(&runningbufspace, -bspace); KASSERT(space >= bspace, ("runningbufspace underflow %ld %ld", space, bspace)); bp->b_runningbufspace = 0; /* * Only acquire the lock and wakeup on the transition from exceeding * the threshold to falling below it. */ if (space < lorunningspace) return; if (space - bspace > lorunningspace) return; runningwakeup(); } /* * waitrunningbufspace() * * runningbufspace is a measure of the amount of I/O currently * running. This routine is used in async-write situations to * prevent creating huge backups of pending writes to a device. * Only asynchronous writes are governed by this function. * * This does NOT turn an async write into a sync write. It waits * for earlier writes to complete and generally returns before the * caller's write has reached the device. */ void waitrunningbufspace(void) { mtx_lock(&rbreqlock); while (runningbufspace > hirunningspace) { runningbufreq = 1; msleep(&runningbufreq, &rbreqlock, PVM, "wdrain", 0); } mtx_unlock(&rbreqlock); } /* * vfs_buf_test_cache: * * Called when a buffer is extended. This function clears the B_CACHE * bit if the newly extended portion of the buffer does not contain * valid data. */ static __inline void vfs_buf_test_cache(struct buf *bp, vm_ooffset_t foff, vm_offset_t off, vm_offset_t size, vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (bp->b_flags & B_CACHE) { int base = (foff + off) & PAGE_MASK; if (vm_page_is_valid(m, base, size) == 0) bp->b_flags &= ~B_CACHE; } } /* Wake up the buffer daemon if necessary */ static __inline void bd_wakeup(void) { mtx_lock(&bdlock); if (bd_request == 0) { bd_request = 1; wakeup(&bd_request); } mtx_unlock(&bdlock); } /* * bd_speedup - speedup the buffer cache flushing code */ void bd_speedup(void) { int needwake; mtx_lock(&bdlock); needwake = 0; if (bd_speedupreq == 0 || bd_request == 0) needwake = 1; bd_speedupreq = 1; bd_request = 1; if (needwake) wakeup(&bd_request); mtx_unlock(&bdlock); } #ifndef NSWBUF_MIN #define NSWBUF_MIN 16 #endif #ifdef __i386__ #define TRANSIENT_DENOM 5 #else #define TRANSIENT_DENOM 10 #endif /* * Calculating buffer cache scaling values and reserve space for buffer * headers. This is called during low level kernel initialization and * may be called more then once. We CANNOT write to the memory area * being reserved at this time. */ caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est) { int tuned_nbuf; long maxbuf, maxbuf_sz, buf_sz, biotmap_sz; /* * physmem_est is in pages. Convert it to kilobytes (assumes * PAGE_SIZE is >= 1K) */ physmem_est = physmem_est * (PAGE_SIZE / 1024); /* * The nominal buffer size (and minimum KVA allocation) is BKVASIZE. * For the first 64MB of ram nominally allocate sufficient buffers to * cover 1/4 of our ram. Beyond the first 64MB allocate additional * buffers to cover 1/10 of our ram over 64MB. When auto-sizing * the buffer cache we limit the eventual kva reservation to * maxbcache bytes. * * factor represents the 1/4 x ram conversion. */ if (nbuf == 0) { int factor = 4 * BKVASIZE / 1024; nbuf = 50; if (physmem_est > 4096) nbuf += min((physmem_est - 4096) / factor, 65536 / factor); if (physmem_est > 65536) nbuf += min((physmem_est - 65536) * 2 / (factor * 5), 32 * 1024 * 1024 / (factor * 5)); if (maxbcache && nbuf > maxbcache / BKVASIZE) nbuf = maxbcache / BKVASIZE; tuned_nbuf = 1; } else tuned_nbuf = 0; /* XXX Avoid unsigned long overflows later on with maxbufspace. */ maxbuf = (LONG_MAX / 3) / BKVASIZE; if (nbuf > maxbuf) { if (!tuned_nbuf) printf("Warning: nbufs lowered from %d to %ld\n", nbuf, maxbuf); nbuf = maxbuf; } /* * Ideal allocation size for the transient bio submap is 10% * of the maximal space buffer map. This roughly corresponds * to the amount of the buffer mapped for typical UFS load. * * Clip the buffer map to reserve space for the transient * BIOs, if its extent is bigger than 90% (80% on i386) of the * maximum buffer map extent on the platform. * * The fall-back to the maxbuf in case of maxbcache unset, * allows to not trim the buffer KVA for the architectures * with ample KVA space. */ if (bio_transient_maxcnt == 0 && unmapped_buf_allowed) { maxbuf_sz = maxbcache != 0 ? maxbcache : maxbuf * BKVASIZE; buf_sz = (long)nbuf * BKVASIZE; if (buf_sz < maxbuf_sz / TRANSIENT_DENOM * (TRANSIENT_DENOM - 1)) { /* * There is more KVA than memory. Do not * adjust buffer map size, and assign the rest * of maxbuf to transient map. */ biotmap_sz = maxbuf_sz - buf_sz; } else { /* * Buffer map spans all KVA we could afford on * this platform. Give 10% (20% on i386) of * the buffer map to the transient bio map. */ biotmap_sz = buf_sz / TRANSIENT_DENOM; buf_sz -= biotmap_sz; } if (biotmap_sz / INT_MAX > MAXPHYS) bio_transient_maxcnt = INT_MAX; else bio_transient_maxcnt = biotmap_sz / MAXPHYS; /* * Artificially limit to 1024 simultaneous in-flight I/Os * using the transient mapping. */ if (bio_transient_maxcnt > 1024) bio_transient_maxcnt = 1024; if (tuned_nbuf) nbuf = buf_sz / BKVASIZE; } /* * swbufs are used as temporary holders for I/O, such as paging I/O. * We have no less then 16 and no more then 256. */ nswbuf = min(nbuf / 4, 256); TUNABLE_INT_FETCH("kern.nswbuf", &nswbuf); if (nswbuf < NSWBUF_MIN) nswbuf = NSWBUF_MIN; /* * Reserve space for the buffer cache buffers */ swbuf = (void *)v; v = (caddr_t)(swbuf + nswbuf); buf = (void *)v; v = (caddr_t)(buf + nbuf); return(v); } /* Initialize the buffer subsystem. Called before use of any buffers. */ void bufinit(void) { struct buf *bp; int i; CTASSERT(MAXBCACHEBUF >= MAXBSIZE); mtx_init(&bqlocks[QUEUE_DIRTY], "bufq dirty lock", NULL, MTX_DEF); mtx_init(&bqlocks[QUEUE_EMPTY], "bufq empty lock", NULL, MTX_DEF); for (i = QUEUE_CLEAN; i < QUEUE_CLEAN + CLEAN_QUEUES; i++) mtx_init(&bqlocks[i], "bufq clean lock", NULL, MTX_DEF); mtx_init(&rbreqlock, "runningbufspace lock", NULL, MTX_DEF); rw_init(&nblock, "needsbuffer lock"); mtx_init(&bdlock, "buffer daemon lock", NULL, MTX_DEF); mtx_init(&bdirtylock, "dirty buf lock", NULL, MTX_DEF); /* next, make a null set of free lists */ for (i = 0; i < BUFFER_QUEUES; i++) TAILQ_INIT(&bufqueues[i]); unmapped_buf = (caddr_t)kva_alloc(MAXPHYS); /* finally, initialize each buffer header and stick on empty q */ for (i = 0; i < nbuf; i++) { bp = &buf[i]; bzero(bp, sizeof *bp); bp->b_flags = B_INVAL; bp->b_rcred = NOCRED; bp->b_wcred = NOCRED; bp->b_qindex = QUEUE_EMPTY; bp->b_xflags = 0; bp->b_data = bp->b_kvabase = unmapped_buf; LIST_INIT(&bp->b_dep); BUF_LOCKINIT(bp); TAILQ_INSERT_TAIL(&bufqueues[QUEUE_EMPTY], bp, b_freelist); #ifdef INVARIANTS bq_len[QUEUE_EMPTY]++; #endif } /* * maxbufspace is the absolute maximum amount of buffer space we are * allowed to reserve in KVM and in real terms. The absolute maximum * is nominally used by metadata. hibufspace is the nominal maximum * used by most other requests. The differential is required to * ensure that metadata deadlocks don't occur. * * maxbufspace is based on BKVASIZE. Allocating buffers larger then * this may result in KVM fragmentation which is not handled optimally * by the system. XXX This is less true with vmem. We could use * PAGE_SIZE. */ maxbufspace = (long)nbuf * BKVASIZE; hibufspace = lmax(3 * maxbufspace / 4, maxbufspace - MAXBCACHEBUF * 10); lobufspace = (hibufspace / 20) * 19; /* 95% */ bufspacethresh = lobufspace + (hibufspace - lobufspace) / 2; /* * Note: The 16 MiB upper limit for hirunningspace was chosen * arbitrarily and may need further tuning. It corresponds to * 128 outstanding write IO requests (if IO size is 128 KiB), * which fits with many RAID controllers' tagged queuing limits. * The lower 1 MiB limit is the historical upper limit for * hirunningspace. */ hirunningspace = lmax(lmin(roundup(hibufspace / 64, MAXBCACHEBUF), 16 * 1024 * 1024), 1024 * 1024); lorunningspace = roundup((hirunningspace * 2) / 3, MAXBCACHEBUF); /* * Limit the amount of malloc memory since it is wired permanently into * the kernel space. Even though this is accounted for in the buffer * allocation, we don't want the malloced region to grow uncontrolled. * The malloc scheme improves memory utilization significantly on * average (small) directories. */ maxbufmallocspace = hibufspace / 20; /* * Reduce the chance of a deadlock occurring by limiting the number * of delayed-write dirty buffers we allow to stack up. */ hidirtybuffers = nbuf / 4 + 20; dirtybufthresh = hidirtybuffers * 9 / 10; numdirtybuffers = 0; /* * To support extreme low-memory systems, make sure hidirtybuffers * cannot eat up all available buffer space. This occurs when our * minimum cannot be met. We try to size hidirtybuffers to 3/4 our * buffer space assuming BKVASIZE'd buffers. */ while ((long)hidirtybuffers * BKVASIZE > 3 * hibufspace / 4) { hidirtybuffers >>= 1; } lodirtybuffers = hidirtybuffers / 2; /* * lofreebuffers should be sufficient to avoid stalling waiting on * buf headers under heavy utilization. The bufs in per-cpu caches * are counted as free but will be unavailable to threads executing * on other cpus. * * hifreebuffers is the free target for the bufspace daemon. This * should be set appropriately to limit work per-iteration. */ lofreebuffers = MIN((nbuf / 25) + (20 * mp_ncpus), 128 * mp_ncpus); hifreebuffers = (3 * lofreebuffers) / 2; numfreebuffers = nbuf; bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL | VM_ALLOC_WIRED); /* Setup the kva and free list allocators. */ vmem_set_reclaim(buffer_arena, bufkva_reclaim); buf_zone = uma_zcache_create("buf free cache", sizeof(struct buf), NULL, NULL, NULL, NULL, buf_import, buf_release, NULL, 0); /* * Size the clean queue according to the amount of buffer space. * One queue per-256mb up to the max. More queues gives better * concurrency but less accurate LRU. */ clean_queues = MIN(howmany(maxbufspace, 256*1024*1024), CLEAN_QUEUES); } #ifdef INVARIANTS static inline void vfs_buf_check_mapped(struct buf *bp) { KASSERT(bp->b_kvabase != unmapped_buf, ("mapped buf: b_kvabase was not updated %p", bp)); KASSERT(bp->b_data != unmapped_buf, ("mapped buf: b_data was not updated %p", bp)); KASSERT(bp->b_data < unmapped_buf || bp->b_data >= unmapped_buf + MAXPHYS, ("b_data + b_offset unmapped %p", bp)); } static inline void vfs_buf_check_unmapped(struct buf *bp) { KASSERT(bp->b_data == unmapped_buf, ("unmapped buf: corrupted b_data %p", bp)); } #define BUF_CHECK_MAPPED(bp) vfs_buf_check_mapped(bp) #define BUF_CHECK_UNMAPPED(bp) vfs_buf_check_unmapped(bp) #else #define BUF_CHECK_MAPPED(bp) do {} while (0) #define BUF_CHECK_UNMAPPED(bp) do {} while (0) #endif static int isbufbusy(struct buf *bp) { if (((bp->b_flags & B_INVAL) == 0 && BUF_ISLOCKED(bp)) || ((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI)) return (1); return (0); } /* * Shutdown the system cleanly to prepare for reboot, halt, or power off. */ void bufshutdown(int show_busybufs) { static int first_buf_printf = 1; struct buf *bp; int iter, nbusy, pbusy; #ifndef PREEMPTION int subiter; #endif /* * Sync filesystems for shutdown */ wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); /* * With soft updates, some buffers that are * written will be remarked as dirty until other * buffers are written. */ for (iter = pbusy = 0; iter < 20; iter++) { nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) if (isbufbusy(bp)) nbusy++; if (nbusy == 0) { if (first_buf_printf) printf("All buffers synced."); break; } if (first_buf_printf) { printf("Syncing disks, buffers remaining... "); first_buf_printf = 0; } printf("%d ", nbusy); if (nbusy < pbusy) iter = 0; pbusy = nbusy; wdog_kern_pat(WD_LASTVAL); sys_sync(curthread, NULL); #ifdef PREEMPTION /* * Drop Giant and spin for a while to allow * interrupt threads to run. */ DROP_GIANT(); DELAY(50000 * iter); PICKUP_GIANT(); #else /* * Drop Giant and context switch several times to * allow interrupt threads to run. */ DROP_GIANT(); for (subiter = 0; subiter < 50 * iter; subiter++) { thread_lock(curthread); mi_switch(SW_VOL, NULL); thread_unlock(curthread); DELAY(1000); } PICKUP_GIANT(); #endif } printf("\n"); /* * Count only busy local buffers to prevent forcing * a fsck if we're just a client of a wedged NFS server */ nbusy = 0; for (bp = &buf[nbuf]; --bp >= buf; ) { if (isbufbusy(bp)) { #if 0 /* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */ if (bp->b_dev == NULL) { TAILQ_REMOVE(&mountlist, bp->b_vp->v_mount, mnt_list); continue; } #endif nbusy++; if (show_busybufs > 0) { printf( "%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:", nbusy, bp, bp->b_vp, bp->b_flags, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); BUF_LOCKPRINTINFO(bp); if (show_busybufs > 1) vn_printf(bp->b_vp, "vnode content: "); } } } if (nbusy) { /* * Failed to sync all blocks. Indicate this and don't * unmount filesystems (thus forcing an fsck on reboot). */ printf("Giving up on %d buffers\n", nbusy); DELAY(5000000); /* 5 seconds */ } else { if (!first_buf_printf) printf("Final sync complete\n"); /* * Unmount filesystems */ if (panicstr == NULL) vfs_unmountall(); } swapoff_all(); DELAY(100000); /* wait for console output to finish */ } static void bpmap_qenter(struct buf *bp) { BUF_CHECK_MAPPED(bp); /* * bp->b_data is relative to bp->b_offset, but * bp->b_offset may be offset into the first page. */ bp->b_data = (caddr_t)trunc_page((vm_offset_t)bp->b_data); pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); bp->b_data = (caddr_t)((vm_offset_t)bp->b_data | (vm_offset_t)(bp->b_offset & PAGE_MASK)); } /* * binsfree: * * Insert the buffer into the appropriate free list. */ static void binsfree(struct buf *bp, int qindex) { struct mtx *olock, *nlock; if (qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } /* * Stick to the same clean queue for the lifetime of the buf to * limit locking below. Otherwise pick ont sequentially. */ if (qindex == QUEUE_CLEAN) { if (bqisclean(bp->b_qindex)) qindex = bp->b_qindex; else qindex = bqcleanq(); } /* * Handle delayed bremfree() processing. */ nlock = bqlock(qindex); if (bp->b_flags & B_REMFREE) { olock = bqlock(bp->b_qindex); mtx_lock(olock); bremfreel(bp); if (olock != nlock) { mtx_unlock(olock); mtx_lock(nlock); } } else mtx_lock(nlock); if (bp->b_qindex != QUEUE_NONE) panic("binsfree: free buffer onto another queue???"); bp->b_qindex = qindex; if (bp->b_flags & B_AGE) TAILQ_INSERT_HEAD(&bufqueues[bp->b_qindex], bp, b_freelist); else TAILQ_INSERT_TAIL(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS bq_len[bp->b_qindex]++; #endif mtx_unlock(nlock); } /* * buf_free: * * Free a buffer to the buf zone once it no longer has valid contents. */ static void buf_free(struct buf *bp) { if (bp->b_flags & B_REMFREE) bremfreef(bp); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 1"); if (bp->b_rcred != NOCRED) { crfree(bp->b_rcred); bp->b_rcred = NOCRED; } if (bp->b_wcred != NOCRED) { crfree(bp->b_wcred); bp->b_wcred = NOCRED; } if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); bufkva_free(bp); BUF_UNLOCK(bp); uma_zfree(buf_zone, bp); atomic_add_int(&numfreebuffers, 1); bufspace_wakeup(); } /* * buf_import: * * Import bufs into the uma cache from the buf list. The system still * expects a static array of bufs and much of the synchronization * around bufs assumes type stable storage. As a result, UMA is used * only as a per-cpu cache of bufs still maintained on a global list. */ static int buf_import(void *arg, void **store, int cnt, int flags) { struct buf *bp; int i; mtx_lock(&bqlocks[QUEUE_EMPTY]); for (i = 0; i < cnt; i++) { bp = TAILQ_FIRST(&bufqueues[QUEUE_EMPTY]); if (bp == NULL) break; bremfreel(bp); store[i] = bp; } mtx_unlock(&bqlocks[QUEUE_EMPTY]); return (i); } /* * buf_release: * * Release bufs from the uma cache back to the buffer queues. */ static void buf_release(void *arg, void **store, int cnt) { int i; for (i = 0; i < cnt; i++) binsfree(store[i], QUEUE_EMPTY); } /* * buf_alloc: * * Allocate an empty buffer header. */ static struct buf * buf_alloc(void) { struct buf *bp; bp = uma_zalloc(buf_zone, M_NOWAIT); if (bp == NULL) { bufspace_daemonwakeup(); atomic_add_int(&numbufallocfails, 1); return (NULL); } /* * Wake-up the bufspace daemon on transition. */ if (atomic_fetchadd_int(&numfreebuffers, -1) == lofreebuffers) bufspace_daemonwakeup(); if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) panic("getnewbuf_empty: Locked buf %p on free queue.", bp); KASSERT(bp->b_vp == NULL, ("bp: %p still has vnode %p.", bp, bp->b_vp)); KASSERT((bp->b_flags & (B_DELWRI | B_NOREUSE)) == 0, ("invalid buffer %p flags %#x", bp, bp->b_flags)); KASSERT((bp->b_xflags & (BX_VNCLEAN|BX_VNDIRTY)) == 0, ("bp: %p still on a buffer list. xflags %X", bp, bp->b_xflags)); KASSERT(bp->b_npages == 0, ("bp: %p still has %d vm pages\n", bp, bp->b_npages)); KASSERT(bp->b_kvasize == 0, ("bp: %p still has kva\n", bp)); KASSERT(bp->b_bufsize == 0, ("bp: %p still has bufspace\n", bp)); bp->b_flags = 0; bp->b_ioflags = 0; bp->b_xflags = 0; bp->b_vflags = 0; bp->b_vp = NULL; bp->b_blkno = bp->b_lblkno = 0; bp->b_offset = NOOFFSET; bp->b_iodone = 0; bp->b_error = 0; bp->b_resid = 0; bp->b_bcount = 0; bp->b_npages = 0; bp->b_dirtyoff = bp->b_dirtyend = 0; bp->b_bufobj = NULL; bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_fsprivate1 = NULL; bp->b_fsprivate2 = NULL; bp->b_fsprivate3 = NULL; LIST_INIT(&bp->b_dep); return (bp); } /* * buf_qrecycle: * * Free a buffer from the given bufqueue. kva controls whether the * freed buf must own some kva resources. This is used for * defragmenting. */ static int buf_qrecycle(int qindex, bool kva) { struct buf *bp, *nbp; if (kva) atomic_add_int(&bufdefragcnt, 1); nbp = NULL; mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); /* * Run scan, possibly freeing data and/or kva mappings on the fly * depending. */ while ((bp = nbp) != NULL) { /* * Calculate next bp (we can only use it if we do not * release the bqlock). */ nbp = TAILQ_NEXT(bp, b_freelist); /* * If we are defragging then we need a buffer with * some kva to reclaim. */ if (kva && bp->b_kvasize == 0) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) continue; /* * Skip buffers with background writes in progress. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0) { BUF_UNLOCK(bp); continue; } KASSERT(bp->b_qindex == qindex, ("getnewbuf: inconsistent queue %d bp %p", qindex, bp)); /* * NOTE: nbp is now entirely invalid. We can only restart * the scan from this point on. */ bremfreel(bp); mtx_unlock(&bqlocks[qindex]); /* * Requeue the background write buffer with error and * restart the scan. */ if ((bp->b_vflags & BV_BKGRDERR) != 0) { bqrelse(bp); mtx_lock(&bqlocks[qindex]); nbp = TAILQ_FIRST(&bufqueues[qindex]); continue; } bp->b_flags |= B_INVAL; brelse(bp); return (0); } mtx_unlock(&bqlocks[qindex]); return (ENOBUFS); } /* * buf_recycle: * * Iterate through all clean queues until we find a buf to recycle or * exhaust the search. */ static int buf_recycle(bool kva) { int qindex, first_qindex; qindex = first_qindex = bqcleanq(); do { if (buf_qrecycle(qindex, kva) == 0) return (0); if (++qindex == QUEUE_CLEAN + clean_queues) qindex = QUEUE_CLEAN; } while (qindex != first_qindex); return (ENOBUFS); } /* * buf_scan: * * Scan the clean queues looking for a buffer to recycle. needsbuffer * is set on failure so that the caller may optionally bufspace_wait() * in a race-free fashion. */ static int buf_scan(bool defrag) { int error; /* * To avoid heavy synchronization and wakeup races we set * needsbuffer and re-poll before failing. This ensures that * no frees can be missed between an unsuccessful poll and * going to sleep in a synchronized fashion. */ if ((error = buf_recycle(defrag)) != 0) { atomic_set_int(&needsbuffer, 1); bufspace_daemonwakeup(); error = buf_recycle(defrag); } if (error == 0) atomic_add_int(&getnewbufrestarts, 1); return (error); } /* * bremfree: * * Mark the buffer for removal from the appropriate free list. * */ void bremfree(struct buf *bp) { CTR3(KTR_BUF, "bremfree(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT((bp->b_flags & B_REMFREE) == 0, ("bremfree: buffer %p already marked for delayed removal.", bp)); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfree: buffer %p not on a queue.", bp)); BUF_ASSERT_XLOCKED(bp); bp->b_flags |= B_REMFREE; } /* * bremfreef: * * Force an immediate removal from a free list. Used only in nfs when * it abuses the b_freelist pointer. */ void bremfreef(struct buf *bp) { struct mtx *qlock; qlock = bqlock(bp->b_qindex); mtx_lock(qlock); bremfreel(bp); mtx_unlock(qlock); } /* * bremfreel: * * Removes a buffer from the free list, must be called with the * correct qlock held. */ static void bremfreel(struct buf *bp) { CTR3(KTR_BUF, "bremfreel(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_qindex != QUEUE_NONE, ("bremfreel: buffer %p not on a queue.", bp)); if (bp->b_qindex != QUEUE_EMPTY) { BUF_ASSERT_XLOCKED(bp); } mtx_assert(bqlock(bp->b_qindex), MA_OWNED); TAILQ_REMOVE(&bufqueues[bp->b_qindex], bp, b_freelist); #ifdef INVARIANTS KASSERT(bq_len[bp->b_qindex] >= 1, ("queue %d underflow", bp->b_qindex)); bq_len[bp->b_qindex]--; #endif bp->b_qindex = QUEUE_NONE; bp->b_flags &= ~B_REMFREE; } /* * bufkva_free: * * Free the kva allocation for a buffer. * */ static void bufkva_free(struct buf *bp) { #ifdef INVARIANTS if (bp->b_kvasize == 0) { KASSERT(bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf, ("Leaked KVA space on %p", bp)); } else if (buf_mapped(bp)) BUF_CHECK_MAPPED(bp); else BUF_CHECK_UNMAPPED(bp); #endif if (bp->b_kvasize == 0) return; vmem_free(buffer_arena, (vm_offset_t)bp->b_kvabase, bp->b_kvasize); atomic_subtract_long(&bufkvaspace, bp->b_kvasize); atomic_add_int(&buffreekvacnt, 1); bp->b_data = bp->b_kvabase = unmapped_buf; bp->b_kvasize = 0; } /* * bufkva_alloc: * * Allocate the buffer KVA and set b_kvasize and b_kvabase. */ static int bufkva_alloc(struct buf *bp, int maxsize, int gbflags) { vm_offset_t addr; int error; KASSERT((gbflags & GB_UNMAPPED) == 0 || (gbflags & GB_KVAALLOC) != 0, ("Invalid gbflags 0x%x in %s", gbflags, __func__)); bufkva_free(bp); addr = 0; error = vmem_alloc(buffer_arena, maxsize, M_BESTFIT | M_NOWAIT, &addr); if (error != 0) { /* * Buffer map is too fragmented. Request the caller * to defragment the map. */ return (error); } bp->b_kvabase = (caddr_t)addr; bp->b_kvasize = maxsize; atomic_add_long(&bufkvaspace, bp->b_kvasize); if ((gbflags & GB_UNMAPPED) != 0) { bp->b_data = unmapped_buf; BUF_CHECK_UNMAPPED(bp); } else { bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); } return (0); } /* * bufkva_reclaim: * * Reclaim buffer kva by freeing buffers holding kva. This is a vmem * callback that fires to avoid returning failure. */ static void bufkva_reclaim(vmem_t *vmem, int flags) { int i; for (i = 0; i < 5; i++) if (buf_scan(true) != 0) break; return; } /* * Attempt to initiate asynchronous I/O on read-ahead blocks. We must * clear BIO_ERROR and B_INVAL prior to initiating I/O . If B_CACHE is set, * the buffer is valid and we do not have to do anything. */ void breada(struct vnode * vp, daddr_t * rablkno, int * rabsize, int cnt, struct ucred * cred) { struct buf *rabp; int i; for (i = 0; i < cnt; i++, rablkno++, rabsize++) { if (inmem(vp, *rablkno)) continue; rabp = getblk(vp, *rablkno, *rabsize, 0, 0, 0); if ((rabp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, rabp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } rabp->b_flags |= B_ASYNC; rabp->b_flags &= ~B_INVAL; rabp->b_ioflags &= ~BIO_ERROR; rabp->b_iocmd = BIO_READ; if (rabp->b_rcred == NOCRED && cred != NOCRED) rabp->b_rcred = crhold(cred); vfs_busy_pages(rabp, 0); BUF_KERNPROC(rabp); rabp->b_iooffset = dbtob(rabp->b_blkno); bstrategy(rabp); } else { brelse(rabp); } } } /* * Entry point for bread() and breadn() via #defines in sys/buf.h. * * Get a buffer with the specified data. Look in the cache first. We * must clear BIO_ERROR and B_INVAL prior to initiating I/O. If B_CACHE * is set, the buffer is valid and we do not have to do anything, see * getblk(). Also starts asynchronous I/O on read-ahead blocks. * * Always return a NULL buffer pointer (in bpp) when returning an error. */ int breadn_flags(struct vnode *vp, daddr_t blkno, int size, daddr_t *rablkno, int *rabsize, int cnt, struct ucred *cred, int flags, struct buf **bpp) { struct buf *bp; int rv = 0, readwait = 0; CTR3(KTR_BUF, "breadn(%p, %jd, %d)", vp, blkno, size); /* * Can only return NULL if GB_LOCK_NOWAIT flag is specified. */ *bpp = bp = getblk(vp, blkno, size, 0, 0, flags); if (bp == NULL) return (EBUSY); /* if not found in cache, do some I/O */ if ((bp->b_flags & B_CACHE) == 0) { if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 0); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_inblock++; } bp->b_iocmd = BIO_READ; bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; if (bp->b_rcred == NOCRED && cred != NOCRED) bp->b_rcred = crhold(cred); vfs_busy_pages(bp, 0); bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); ++readwait; } breada(vp, rablkno, rabsize, cnt, cred); if (readwait) { rv = bufwait(bp); if (rv != 0) { brelse(bp); *bpp = NULL; } } return (rv); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ int bufwrite(struct buf *bp) { int oldflags; struct vnode *vp; long space; int vp_md; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if ((bp->b_bufobj->bo_flag & BO_DEAD) != 0) { bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_CACHE; brelse(bp); return (ENXIO); } if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (bp->b_flags & B_BARRIER) barrierwrites++; oldflags = bp->b_flags; BUF_ASSERT_HELD(bp); KASSERT(!(bp->b_vflags & BV_BKGRDINPROG), ("FFS background buffer should not get here %p", bp)); vp = bp->b_vp; if (vp) vp_md = vp->v_vflag & VV_MD; else vp_md = 0; /* * Mark the buffer clean. Increment the bufobj write count * before bundirty() call, to prevent other thread from seeing * empty dirty list and zero counter for writes in progress, * falsely indicating that the bufobj is clean. */ bufobj_wref(bp->b_bufobj); bundirty(bp); bp->b_flags &= ~B_DONE; bp->b_ioflags &= ~BIO_ERROR; bp->b_flags |= B_CACHE; bp->b_iocmd = BIO_WRITE; vfs_busy_pages(bp, 1); /* * Normal bwrites pipeline writes */ bp->b_runningbufspace = bp->b_bufsize; space = atomic_fetchadd_long(&runningbufspace, bp->b_runningbufspace); if (!TD_IS_IDLETHREAD(curthread)) { #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_buf(curproc, bp, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ curthread->td_ru.ru_oublock++; } if (oldflags & B_ASYNC) BUF_KERNPROC(bp); bp->b_iooffset = dbtob(bp->b_blkno); buf_track(bp, __func__); bstrategy(bp); if ((oldflags & B_ASYNC) == 0) { int rtval = bufwait(bp); brelse(bp); return (rtval); } else if (space > hirunningspace) { /* * don't allow the async write to saturate the I/O * system. We will not deadlock here because * we are blocking waiting for I/O that is already in-progress * to complete. We do not block here if it is the update * or syncer daemon trying to clean up as that can lead * to deadlock. */ if ((curthread->td_pflags & TDP_NORUNNINGBUF) == 0 && !vp_md) waitrunningbufspace(); } return (0); } void bufbdflush(struct bufobj *bo, struct buf *bp) { struct buf *nbp; if (bo->bo_dirty.bv_cnt > dirtybufthresh + 10) { (void) VOP_FSYNC(bp->b_vp, MNT_NOWAIT, curthread); altbufferflushes++; } else if (bo->bo_dirty.bv_cnt > dirtybufthresh) { BO_LOCK(bo); /* * Try to find a buffer to flush. */ TAILQ_FOREACH(nbp, &bo->bo_dirty.bv_hd, b_bobufs) { if ((nbp->b_vflags & BV_BKGRDINPROG) || BUF_LOCK(nbp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) continue; if (bp == nbp) panic("bdwrite: found ourselves"); BO_UNLOCK(bo); /* Don't countdeps with the bo lock held. */ if (buf_countdeps(nbp, 0)) { BO_LOCK(bo); BUF_UNLOCK(nbp); continue; } if (nbp->b_flags & B_CLUSTEROK) { vfs_bio_awrite(nbp); } else { bremfree(nbp); bawrite(nbp); } dirtybufferflushes++; break; } if (nbp == NULL) BO_UNLOCK(bo); } } /* * Delayed write. (Buffer is marked dirty). Do not bother writing * anything if the buffer is marked invalid. * * Note that since the buffer must be completely valid, we can safely * set B_CACHE. In fact, we have to set B_CACHE here rather then in * biodone() in order to prevent getblk from writing the buffer * out synchronously. */ void bdwrite(struct buf *bp) { struct thread *td = curthread; struct vnode *vp; struct bufobj *bo; CTR3(KTR_BUF, "bdwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT((bp->b_flags & B_BARRIER) == 0, ("Barrier request in delayed write %p", bp)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_INVAL) { brelse(bp); return; } /* * If we have too many dirty buffers, don't create any more. * If we are wildly over our limit, then force a complete * cleanup. Otherwise, just keep the situation from getting * out of control. Note that we have to avoid a recursive * disaster and not try to clean up after our own cleanup! */ vp = bp->b_vp; bo = bp->b_bufobj; if ((td->td_pflags & (TDP_COWINPROGRESS|TDP_INBDFLUSH)) == 0) { td->td_pflags |= TDP_INBDFLUSH; BO_BDFLUSH(bo, bp); td->td_pflags &= ~TDP_INBDFLUSH; } else recursiveflushes++; bdirty(bp); /* * Set B_CACHE, indicating that the buffer is fully valid. This is * true even of NFS now. */ bp->b_flags |= B_CACHE; /* * This bmap keeps the system from needing to do the bmap later, * perhaps when the system is attempting to do a sync. Since it * is likely that the indirect block -- or whatever other datastructure * that the filesystem needs is still in memory now, it is a good * thing to do this. Note also, that if the pageout daemon is * requesting a sync -- there might not be enough memory to do * the bmap then... So, this is important to do. */ if (vp->v_type != VCHR && bp->b_lblkno == bp->b_blkno) { VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL); } buf_track(bp, __func__); /* * Set the *dirty* buffer range based upon the VM system dirty * pages. * * Mark the buffer pages as clean. We need to do this here to * satisfy the vnode_pager and the pageout daemon, so that it * thinks that the pages have been "cleaned". Note that since * the pages are in a delayed write buffer -- the VFS layer * "will" see that the pages get written out on the next sync, * or perhaps the cluster will be completed. */ vfs_clean_pages_dirty_buf(bp); bqrelse(bp); /* * note: we cannot initiate I/O from a bdwrite even if we wanted to, * due to the softdep code. */ } /* * bdirty: * * Turn buffer into delayed write request. We must clear BIO_READ and * B_RELBUF, and we must set B_DELWRI. We reassign the buffer to * itself to properly update it in the dirty/clean lists. We mark it * B_DONE to ensure that any asynchronization of the buffer properly * clears B_DONE ( else a panic will occur later ). * * bdirty() is kinda like bdwrite() - we have to clear B_INVAL which * might have been set pre-getblk(). Unlike bwrite/bdwrite, bdirty() * should only be called if the buffer is known-good. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bdirty(struct buf *bp) { CTR3(KTR_BUF, "bdirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bdirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); bp->b_flags &= ~(B_RELBUF); bp->b_iocmd = BIO_WRITE; if ((bp->b_flags & B_DELWRI) == 0) { bp->b_flags |= /* XXX B_DONE | */ B_DELWRI; reassignbuf(bp); bdirtyadd(); } } /* * bundirty: * * Clear B_DELWRI for buffer. * * Since the buffer is not on a queue, we do not update the numfreebuffers * count. * * The buffer must be on QUEUE_NONE. */ void bundirty(struct buf *bp) { CTR3(KTR_BUF, "bundirty(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); KASSERT(bp->b_flags & B_REMFREE || bp->b_qindex == QUEUE_NONE, ("bundirty: buffer %p still on queue %d", bp, bp->b_qindex)); BUF_ASSERT_HELD(bp); if (bp->b_flags & B_DELWRI) { bp->b_flags &= ~B_DELWRI; reassignbuf(bp); bdirtysub(); } /* * Since it is now being written, we can clear its deferred write flag. */ bp->b_flags &= ~B_DEFERRED; } /* * bawrite: * * Asynchronous write. Start output on a buffer, but do not wait for * it to complete. The buffer is released when the output completes. * * bwrite() ( or the VOP routine anyway ) is responsible for handling * B_INVAL buffers. Not us. */ void bawrite(struct buf *bp) { bp->b_flags |= B_ASYNC; (void) bwrite(bp); } /* * babarrierwrite: * * Asynchronous barrier write. Start output on a buffer, but do not * wait for it to complete. Place a write barrier after this write so * that this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ void babarrierwrite(struct buf *bp) { bp->b_flags |= B_ASYNC | B_BARRIER; (void) bwrite(bp); } /* * bbarrierwrite: * * Synchronous barrier write. Start output on a buffer and wait for * it to complete. Place a write barrier after this write so that * this buffer and all buffers written before it are committed to * the disk before any buffers written after this write are committed * to the disk. The buffer is released when the output completes. */ int bbarrierwrite(struct buf *bp) { bp->b_flags |= B_BARRIER; return (bwrite(bp)); } /* * bwillwrite: * * Called prior to the locking of any vnodes when we are expecting to * write. We do not want to starve the buffer cache with too many * dirty buffers so we block here. By blocking prior to the locking * of any vnodes we attempt to avoid the situation where a locked vnode * prevents the various system daemons from flushing related buffers. */ void bwillwrite(void) { if (numdirtybuffers >= hidirtybuffers) { mtx_lock(&bdirtylock); while (numdirtybuffers >= hidirtybuffers) { bdirtywait = 1; msleep(&bdirtywait, &bdirtylock, (PRIBIO + 4), "flswai", 0); } mtx_unlock(&bdirtylock); } } /* * Return true if we have too many dirty buffers. */ int buf_dirty_count_severe(void) { return(numdirtybuffers >= hidirtybuffers); } /* * brelse: * * Release a busy buffer and, if requested, free its resources. The * buffer will be stashed in the appropriate bufqueue[] allowing it * to be accessed later as a cache entity or reused for other purposes. */ void brelse(struct buf *bp) { int qindex; /* * Many functions erroneously call brelse with a NULL bp under rare * error conditions. Simply return when called with a NULL bp. */ if (bp == NULL) return; CTR3(KTR_BUF, "brelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("brelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); KASSERT((bp->b_flags & B_VMIO) != 0 || (bp->b_flags & B_NOREUSE) == 0, ("brelse: non-VMIO buffer marked NOREUSE")); if (BUF_LOCKRECURSED(bp)) { /* * Do not process, in particular, do not handle the * B_INVAL/B_RELBUF and do not release to free list. */ BUF_UNLOCK(bp); return; } if (bp->b_flags & B_MANAGED) { bqrelse(bp); return; } if ((bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); bdirty(bp); } if (bp->b_iocmd == BIO_WRITE && (bp->b_ioflags & BIO_ERROR) && !(bp->b_flags & B_INVAL)) { /* * Failed write, redirty. Must clear BIO_ERROR to prevent * pages from being scrapped. */ bp->b_ioflags &= ~BIO_ERROR; bdirty(bp); } else if ((bp->b_flags & (B_NOCACHE | B_INVAL)) || (bp->b_ioflags & BIO_ERROR) || (bp->b_bufsize <= 0)) { /* * Either a failed read I/O or we were asked to free or not * cache the buffer. */ bp->b_flags |= B_INVAL; if (!LIST_EMPTY(&bp->b_dep)) buf_deallocate(bp); if (bp->b_flags & B_DELWRI) bdirtysub(); bp->b_flags &= ~(B_DELWRI | B_CACHE); if ((bp->b_flags & B_VMIO) == 0) { allocbuf(bp, 0); if (bp->b_vp) brelvp(bp); } } /* * We must clear B_RELBUF if B_DELWRI is set. If vfs_vmio_truncate() * is called with B_DELWRI set, the underlying pages may wind up * getting freed causing a previous write (bdwrite()) to get 'lost' * because pages associated with a B_DELWRI bp are marked clean. * * We still allow the B_INVAL case to call vfs_vmio_truncate(), even * if B_DELWRI is set. */ if (bp->b_flags & B_DELWRI) bp->b_flags &= ~B_RELBUF; /* * VMIO buffer rundown. It is not very necessary to keep a VMIO buffer * constituted, not even NFS buffers now. Two flags effect this. If * B_INVAL, the struct buf is invalidated but the VM object is kept * around ( i.e. so it is trivial to reconstitute the buffer later ). * * If BIO_ERROR or B_NOCACHE is set, pages in the VM object will be * invalidated. BIO_ERROR cannot be set for a failed write unless the * buffer is also B_INVAL because it hits the re-dirtying code above. * * Normally we can do this whether a buffer is B_DELWRI or not. If * the buffer is an NFS buffer, it is tracking piecemeal writes or * the commit state and we cannot afford to lose the buffer. If the * buffer has a background write in progress, we need to keep it * around to prevent it from being reconstituted and starting a second * background write. */ if ((bp->b_flags & B_VMIO) && (bp->b_flags & B_NOCACHE || (bp->b_ioflags & BIO_ERROR && bp->b_iocmd == BIO_READ)) && !(bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && !vn_isdisk(bp->b_vp, NULL) && (bp->b_flags & B_DELWRI))) { vfs_vmio_invalidate(bp); allocbuf(bp, 0); } if ((bp->b_flags & (B_INVAL | B_RELBUF)) != 0 || (bp->b_flags & (B_DELWRI | B_NOREUSE)) == B_NOREUSE) { allocbuf(bp, 0); bp->b_flags &= ~B_NOREUSE; if (bp->b_vp != NULL) brelvp(bp); } /* * If the buffer has junk contents signal it and eventually * clean up B_DELWRI and diassociate the vnode so that gbincore() * doesn't find it. */ if (bp->b_bufsize == 0 || (bp->b_ioflags & BIO_ERROR) != 0 || (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF)) != 0) bp->b_flags |= B_INVAL; if (bp->b_flags & B_INVAL) { if (bp->b_flags & B_DELWRI) bundirty(bp); if (bp->b_vp) brelvp(bp); } buf_track(bp, __func__); /* buffers with no memory */ if (bp->b_bufsize == 0) { buf_free(bp); return; } /* buffers with junk contents */ if (bp->b_flags & (B_INVAL | B_NOCACHE | B_RELBUF) || (bp->b_ioflags & BIO_ERROR)) { bp->b_xflags &= ~(BX_BKGRDWRITE | BX_ALTDATA); if (bp->b_vflags & BV_BKGRDINPROG) panic("losing buffer 2"); qindex = QUEUE_CLEAN; bp->b_flags |= B_AGE; /* remaining buffers */ } else if (bp->b_flags & B_DELWRI) qindex = QUEUE_DIRTY; else qindex = QUEUE_CLEAN; binsfree(bp, qindex); bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF | B_DIRECT); if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("brelse: not dirty"); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Release a buffer back to the appropriate queue but do not try to free * it. The buffer is expected to be used again soon. * * bqrelse() is used by bdwrite() to requeue a delayed write, and used by * biodone() to requeue an async I/O on completion. It is also used when * known good buffers need to be requeued but we think we may need the data * again soon. * * XXX we should be able to leave the B_RELBUF hint set on completion. */ void bqrelse(struct buf *bp) { int qindex; CTR3(KTR_BUF, "bqrelse(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(!(bp->b_flags & (B_CLUSTER|B_PAGING)), ("bqrelse: inappropriate B_PAGING or B_CLUSTER bp %p", bp)); qindex = QUEUE_NONE; if (BUF_LOCKRECURSED(bp)) { /* do not release to free list */ BUF_UNLOCK(bp); return; } bp->b_flags &= ~(B_ASYNC | B_NOCACHE | B_AGE | B_RELBUF); if (bp->b_flags & B_MANAGED) { if (bp->b_flags & B_REMFREE) bremfreef(bp); goto out; } /* buffers with stale but valid contents */ if ((bp->b_flags & B_DELWRI) != 0 || (bp->b_vflags & (BV_BKGRDINPROG | BV_BKGRDERR)) == BV_BKGRDERR) { BO_LOCK(bp->b_bufobj); bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); qindex = QUEUE_DIRTY; } else { if ((bp->b_flags & B_DELWRI) == 0 && (bp->b_xflags & BX_VNDIRTY)) panic("bqrelse: not dirty"); if ((bp->b_flags & B_NOREUSE) != 0) { brelse(bp); return; } qindex = QUEUE_CLEAN; } binsfree(bp, qindex); out: buf_track(bp, __func__); /* unlock */ BUF_UNLOCK(bp); if (qindex == QUEUE_CLEAN) bufspace_wakeup(); } /* * Complete I/O to a VMIO backed page. Validate the pages as appropriate, * restore bogus pages. */ static void vfs_vmio_iodone(struct buf *bp) { vm_ooffset_t foff; vm_page_t m; vm_object_t obj; struct vnode *vp; int bogus, i, iosize; obj = bp->b_bufobj->bo_object; KASSERT(obj->paging_in_progress >= bp->b_npages, ("vfs_vmio_iodone: paging in progress(%d) < b_npages(%d)", obj->paging_in_progress, bp->b_npages)); vp = bp->b_vp; KASSERT(vp->v_holdcnt > 0, ("vfs_vmio_iodone: vnode %p has zero hold count", vp)); KASSERT(vp->v_object != NULL, ("vfs_vmio_iodone: vnode %p has no vm_object", vp)); foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_vmio_iodone: bp %p has no buffer offset", bp)); bogus = 0; iosize = bp->b_bcount - bp->b_resid; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { int resid; resid = ((foff + PAGE_SIZE) & ~(off_t)PAGE_MASK) - foff; if (resid > iosize) resid = iosize; /* * cleanup bogus pages, restoring the originals */ m = bp->b_pages[i]; if (m == bogus_page) { bogus = 1; m = vm_page_lookup(obj, OFF_TO_IDX(foff)); if (m == NULL) panic("biodone: page disappeared!"); bp->b_pages[i] = m; } else if ((bp->b_iocmd == BIO_READ) && resid > 0) { /* * In the write case, the valid and clean bits are * already changed correctly ( see bdwrite() ), so we * only need to do this here in the read case. */ KASSERT((m->dirty & vm_page_bits(foff & PAGE_MASK, resid)) == 0, ("vfs_vmio_iodone: page %p " "has unexpected dirty bits", m)); vfs_page_set_valid(bp, foff, m); } KASSERT(OFF_TO_IDX(foff) == m->pindex, ("vfs_vmio_iodone: foff(%jd)/pindex(%ju) mismatch", (intmax_t)foff, (uintmax_t)m->pindex)); vm_page_sunbusy(m); foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; iosize -= resid; } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * Unwire a page held by a buf and place it on the appropriate vm queue. */ static void vfs_vmio_unwire(struct buf *bp, vm_page_t m) { bool freed; vm_page_lock(m); if (vm_page_unwire(m, PQ_NONE)) { /* * Determine if the page should be freed before adding * it to the inactive queue. */ if (m->valid == 0) { freed = !vm_page_busied(m); if (freed) vm_page_free(m); } else if ((bp->b_flags & B_DIRECT) != 0) freed = vm_page_try_to_free(m); else freed = false; if (!freed) { /* * If the page is unlikely to be reused, let the * VM know. Otherwise, maintain LRU page * ordering and put the page at the tail of the * inactive queue. */ if ((bp->b_flags & B_NOREUSE) != 0) vm_page_deactivate_noreuse(m); else vm_page_deactivate(m); } } vm_page_unlock(m); } /* * Perform page invalidation when a buffer is released. The fully invalid * pages will be reclaimed later in vfs_vmio_truncate(). */ static void vfs_vmio_invalidate(struct buf *bp) { vm_object_t obj; vm_page_t m; int i, resid, poffset, presid; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove(trunc_page((vm_offset_t)bp->b_data), bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); /* * Get the base offset and length of the buffer. Note that * in the VMIO case if the buffer block size is not * page-aligned then b_data pointer may not be page-aligned. * But our b_pages[] array *IS* page aligned. * * block sizes less then DEV_BSIZE (usually 512) are not * supported due to the page granularity bits (m->valid, * m->dirty, etc...). * * See man buf(9) for more information */ obj = bp->b_bufobj->bo_object; resid = bp->b_bufsize; poffset = bp->b_offset & PAGE_MASK; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) panic("vfs_vmio_invalidate: Unexpected bogus page."); bp->b_pages[i] = NULL; presid = resid > (PAGE_SIZE - poffset) ? (PAGE_SIZE - poffset) : resid; KASSERT(presid >= 0, ("brelse: extra page")); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(obj); vm_page_busy_sleep(m, "mbncsh", true); VM_OBJECT_WLOCK(obj); } if (pmap_page_wired_mappings(m) == 0) vm_page_set_invalid(m, poffset, presid); vfs_vmio_unwire(bp, m); resid -= presid; poffset = 0; } VM_OBJECT_WUNLOCK(obj); bp->b_npages = 0; } /* * Page-granular truncation of an existing VMIO buffer. */ static void vfs_vmio_truncate(struct buf *bp, int desiredpages) { vm_object_t obj; vm_page_t m; int i; if (bp->b_npages == desiredpages) return; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qremove((vm_offset_t)trunc_page((vm_offset_t)bp->b_data) + (desiredpages << PAGE_SHIFT), bp->b_npages - desiredpages); } else BUF_CHECK_UNMAPPED(bp); obj = bp->b_bufobj->bo_object; if (obj != NULL) VM_OBJECT_WLOCK(obj); for (i = desiredpages; i < bp->b_npages; i++) { m = bp->b_pages[i]; KASSERT(m != bogus_page, ("allocbuf: bogus page found")); bp->b_pages[i] = NULL; vfs_vmio_unwire(bp, m); } if (obj != NULL) VM_OBJECT_WUNLOCK(obj); bp->b_npages = desiredpages; } /* * Byte granular extension of VMIO buffers. */ static void vfs_vmio_extend(struct buf *bp, int desiredpages, int size) { /* * We are growing the buffer, possibly in a * byte-granular fashion. */ vm_object_t obj; vm_offset_t toff; vm_offset_t tinc; vm_page_t m; /* * Step 1, bring in the VM pages from the object, allocating * them if necessary. We must clear B_CACHE if these pages * are not valid for the range covered by the buffer. */ obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); while (bp->b_npages < desiredpages) { /* * We must allocate system pages since blocking * here could interfere with paging I/O, no * matter which process we are. * * Only exclusive busy can be tested here. * Blocking on shared busy might lead to * deadlocks once allocbuf() is called after * pages are vfs_busy_pages(). */ m = vm_page_grab(obj, OFF_TO_IDX(bp->b_offset) + bp->b_npages, VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY | VM_ALLOC_COUNT(desiredpages - bp->b_npages)); if (m->valid == 0) bp->b_flags &= ~B_CACHE; bp->b_pages[bp->b_npages] = m; ++bp->b_npages; } /* * Step 2. We've loaded the pages into the buffer, * we have to figure out if we can still have B_CACHE * set. Note that B_CACHE is set according to the * byte-granular range ( bcount and size ), not the * aligned range ( newbsize ). * * The VM test is against m->valid, which is DEV_BSIZE * aligned. Needless to say, the validity of the data * needs to also be DEV_BSIZE aligned. Note that this * fails with NFS if the server or some other client * extends the file's EOF. If our buffer is resized, * B_CACHE may remain set! XXX */ toff = bp->b_bcount; tinc = PAGE_SIZE - ((bp->b_offset + toff) & PAGE_MASK); while ((bp->b_flags & B_CACHE) && toff < size) { vm_pindex_t pi; if (tinc > (size - toff)) tinc = size - toff; pi = ((bp->b_offset & PAGE_MASK) + toff) >> PAGE_SHIFT; m = bp->b_pages[pi]; vfs_buf_test_cache(bp, bp->b_offset, toff, tinc, m); toff += tinc; tinc = PAGE_SIZE; } VM_OBJECT_WUNLOCK(obj); /* * Step 3, fixup the KVA pmap. */ if (buf_mapped(bp)) bpmap_qenter(bp); else BUF_CHECK_UNMAPPED(bp); } /* * Check to see if a block at a particular lbn is available for a clustered * write. */ static int vfs_bio_clcheck(struct vnode *vp, int size, daddr_t lblkno, daddr_t blkno) { struct buf *bpa; int match; match = 0; /* If the buf isn't in core skip it */ if ((bpa = gbincore(&vp->v_bufobj, lblkno)) == NULL) return (0); /* If the buf is busy we don't want to wait for it */ if (BUF_LOCK(bpa, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0) return (0); /* Only cluster with valid clusterable delayed write buffers */ if ((bpa->b_flags & (B_DELWRI | B_CLUSTEROK | B_INVAL)) != (B_DELWRI | B_CLUSTEROK)) goto done; if (bpa->b_bufsize != size) goto done; /* * Check to see if it is in the expected place on disk and that the * block has been mapped. */ if ((bpa->b_blkno != bpa->b_lblkno) && (bpa->b_blkno == blkno)) match = 1; done: BUF_UNLOCK(bpa); return (match); } /* * vfs_bio_awrite: * * Implement clustered async writes for clearing out B_DELWRI buffers. * This is much better then the old way of writing only one buffer at * a time. Note that we may not be presented with the buffers in the * correct order, so we search for the cluster in both directions. */ int vfs_bio_awrite(struct buf *bp) { struct bufobj *bo; int i; int j; daddr_t lblkno = bp->b_lblkno; struct vnode *vp = bp->b_vp; int ncl; int nwritten; int size; int maxcl; int gbflags; bo = &vp->v_bufobj; gbflags = (bp->b_data == unmapped_buf) ? GB_UNMAPPED : 0; /* * right now we support clustered writing only to regular files. If * we find a clusterable block we could be in the middle of a cluster * rather then at the beginning. */ if ((vp->v_type == VREG) && (vp->v_mount != 0) && /* Only on nodes that have the size info */ (bp->b_flags & (B_CLUSTEROK | B_INVAL)) == B_CLUSTEROK) { size = vp->v_mount->mnt_stat.f_iosize; maxcl = MAXPHYS / size; BO_RLOCK(bo); for (i = 1; i < maxcl; i++) if (vfs_bio_clcheck(vp, size, lblkno + i, bp->b_blkno + ((i * size) >> DEV_BSHIFT)) == 0) break; for (j = 1; i + j <= maxcl && j <= lblkno; j++) if (vfs_bio_clcheck(vp, size, lblkno - j, bp->b_blkno - ((j * size) >> DEV_BSHIFT)) == 0) break; BO_RUNLOCK(bo); --j; ncl = i + j; /* * this is a possible cluster write */ if (ncl != 1) { BUF_UNLOCK(bp); nwritten = cluster_wbuild(vp, size, lblkno - j, ncl, gbflags); return (nwritten); } } bremfree(bp); bp->b_flags |= B_ASYNC; /* * default (old) behavior, writing out only one block * * XXX returns b_bufsize instead of b_bcount for nwritten? */ nwritten = bp->b_bufsize; (void) bwrite(bp); return (nwritten); } /* * getnewbuf_kva: * * Allocate KVA for an empty buf header according to gbflags. */ static int getnewbuf_kva(struct buf *bp, int gbflags, int maxsize) { if ((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_UNMAPPED) { /* * In order to keep fragmentation sane we only allocate kva * in BKVASIZE chunks. XXX with vmem we can do page size. */ maxsize = (maxsize + BKVAMASK) & ~BKVAMASK; if (maxsize != bp->b_kvasize && bufkva_alloc(bp, maxsize, gbflags)) return (ENOSPC); } return (0); } /* * getnewbuf: * * Find and initialize a new buffer header, freeing up existing buffers * in the bufqueues as necessary. The new buffer is returned locked. * * We block if: * We have insufficient buffer headers * We have insufficient buffer space * buffer_arena is too fragmented ( space reservation fails ) * If we have to flush dirty buffers ( but we try to avoid this ) * * The caller is responsible for releasing the reserved bufspace after * allocbuf() is called. */ static struct buf * getnewbuf(struct vnode *vp, int slpflag, int slptimeo, int maxsize, int gbflags) { struct buf *bp; bool metadata, reserved; bp = NULL; KASSERT((gbflags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); if (!unmapped_buf_allowed) gbflags &= ~(GB_UNMAPPED | GB_KVAALLOC); if (vp == NULL || (vp->v_vflag & (VV_MD | VV_SYSTEM)) != 0 || vp->v_type == VCHR) metadata = true; else metadata = false; atomic_add_int(&getnewbufcalls, 1); reserved = false; do { if (reserved == false && bufspace_reserve(maxsize, metadata) != 0) continue; reserved = true; if ((bp = buf_alloc()) == NULL) continue; if (getnewbuf_kva(bp, gbflags, maxsize) == 0) return (bp); break; } while(buf_scan(false) == 0); if (reserved) atomic_subtract_long(&bufspace, maxsize); if (bp != NULL) { bp->b_flags |= B_INVAL; brelse(bp); } bufspace_wait(vp, gbflags, slpflag, slptimeo); return (NULL); } /* * buf_daemon: * * buffer flushing daemon. Buffers are normally flushed by the * update daemon but if it cannot keep up this process starts to * take the load in an attempt to prevent getnewbuf() from blocking. */ static struct kproc_desc buf_kp = { "bufdaemon", buf_daemon, &bufdaemonproc }; SYSINIT(bufdaemon, SI_SUB_KTHREAD_BUF, SI_ORDER_FIRST, kproc_start, &buf_kp); static int buf_flush(struct vnode *vp, int target) { int flushed; flushed = flushbufqueues(vp, target, 0); if (flushed == 0) { /* * Could not find any buffers without rollback * dependencies, so just write the first one * in the hopes of eventually making progress. */ if (vp != NULL && target > 2) target /= 2; flushbufqueues(vp, target, 1); } return (flushed); } static void buf_daemon() { int lodirty; /* * This process needs to be suspended prior to shutdown sync. */ EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, bufdaemonproc, SHUTDOWN_PRI_LAST); /* * This process is allowed to take the buffer cache to the limit */ curthread->td_pflags |= TDP_NORUNNINGBUF | TDP_BUFNEED; mtx_lock(&bdlock); for (;;) { bd_request = 0; mtx_unlock(&bdlock); kproc_suspend_check(bufdaemonproc); lodirty = lodirtybuffers; if (bd_speedupreq) { lodirty = numdirtybuffers / 2; bd_speedupreq = 0; } /* * Do the flush. Limit the amount of in-transit I/O we * allow to build up, otherwise we would completely saturate * the I/O system. */ while (numdirtybuffers > lodirty) { if (buf_flush(NULL, numdirtybuffers - lodirty) == 0) break; kern_yield(PRI_USER); } /* * Only clear bd_request if we have reached our low water * mark. The buf_daemon normally waits 1 second and * then incrementally flushes any dirty buffers that have * built up, within reason. * * If we were unable to hit our low water mark and couldn't * find any flushable buffers, we sleep for a short period * to avoid endless loops on unlockable buffers. */ mtx_lock(&bdlock); if (numdirtybuffers <= lodirtybuffers) { /* * We reached our low water mark, reset the * request and sleep until we are needed again. * The sleep is just so the suspend code works. */ bd_request = 0; /* * Do an extra wakeup in case dirty threshold * changed via sysctl and the explicit transition * out of shortfall was missed. */ bdirtywakeup(); if (runningbufspace <= lorunningspace) runningwakeup(); msleep(&bd_request, &bdlock, PVM, "psleep", hz); } else { /* * We couldn't find any flushable dirty buffers but * still have too many dirty buffers, we * have to sleep and try again. (rare) */ msleep(&bd_request, &bdlock, PVM, "qsleep", hz / 10); } } } /* * flushbufqueues: * * Try to flush a buffer in the dirty queue. We must be careful to * free up B_INVAL buffers instead of write them, which NFS is * particularly sensitive to. */ static int flushwithdeps = 0; SYSCTL_INT(_vfs, OID_AUTO, flushwithdeps, CTLFLAG_RW, &flushwithdeps, 0, "Number of buffers flushed with dependecies that require rollbacks"); static int flushbufqueues(struct vnode *lvp, int target, int flushdeps) { struct buf *sentinel; struct vnode *vp; struct mount *mp; struct buf *bp; int hasdeps; int flushed; int queue; int error; bool unlock; flushed = 0; queue = QUEUE_DIRTY; bp = NULL; sentinel = malloc(sizeof(struct buf), M_TEMP, M_WAITOK | M_ZERO); sentinel->b_qindex = QUEUE_SENTINEL; mtx_lock(&bqlocks[queue]); TAILQ_INSERT_HEAD(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); while (flushed != target) { maybe_yield(); mtx_lock(&bqlocks[queue]); bp = TAILQ_NEXT(sentinel, b_freelist); if (bp != NULL) { TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); TAILQ_INSERT_AFTER(&bufqueues[queue], bp, sentinel, b_freelist); } else { mtx_unlock(&bqlocks[queue]); break; } /* * Skip sentinels inserted by other invocations of the * flushbufqueues(), taking care to not reorder them. * * Only flush the buffers that belong to the * vnode locked by the curthread. */ if (bp->b_qindex == QUEUE_SENTINEL || (lvp != NULL && bp->b_vp != lvp)) { mtx_unlock(&bqlocks[queue]); continue; } error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL); mtx_unlock(&bqlocks[queue]); if (error != 0) continue; /* * BKGRDINPROG can only be set with the buf and bufobj * locks both held. We tolerate a race to clear it here. */ if ((bp->b_vflags & BV_BKGRDINPROG) != 0 || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); continue; } if (bp->b_flags & B_INVAL) { bremfreef(bp); brelse(bp); flushed++; continue; } if (!LIST_EMPTY(&bp->b_dep) && buf_countdeps(bp, 0)) { if (flushdeps == 0) { BUF_UNLOCK(bp); continue; } hasdeps = 1; } else hasdeps = 0; /* * We must hold the lock on a vnode before writing * one of its buffers. Otherwise we may confuse, or * in the case of a snapshot vnode, deadlock the * system. * * The lock order here is the reverse of the normal * of vnode followed by buf lock. This is ok because * the NOWAIT will prevent deadlock. */ vp = bp->b_vp; if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { BUF_UNLOCK(bp); continue; } if (lvp == NULL) { unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT); } else { ASSERT_VOP_LOCKED(vp, "getbuf"); unlock = false; error = VOP_ISLOCKED(vp) == LK_EXCLUSIVE ? 0 : vn_lock(vp, LK_TRYUPGRADE); } if (error == 0) { CTR3(KTR_BUF, "flushbufqueue(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (curproc == bufdaemonproc) { vfs_bio_awrite(bp); } else { bremfree(bp); bwrite(bp); notbufdflushes++; } vn_finished_write(mp); if (unlock) VOP_UNLOCK(vp, 0); flushwithdeps += hasdeps; flushed++; /* * Sleeping on runningbufspace while holding * vnode lock leads to deadlock. */ if (curproc == bufdaemonproc && runningbufspace > hirunningspace) waitrunningbufspace(); continue; } vn_finished_write(mp); BUF_UNLOCK(bp); } mtx_lock(&bqlocks[queue]); TAILQ_REMOVE(&bufqueues[queue], sentinel, b_freelist); mtx_unlock(&bqlocks[queue]); free(sentinel, M_TEMP); return (flushed); } /* * Check to see if a block is currently memory resident. */ struct buf * incore(struct bufobj *bo, daddr_t blkno) { struct buf *bp; BO_RLOCK(bo); bp = gbincore(bo, blkno); BO_RUNLOCK(bo); return (bp); } /* * Returns true if no I/O is needed to access the * associated VM object. This is like incore except * it also hunts around in the VM system for the data. */ static int inmem(struct vnode * vp, daddr_t blkno) { vm_object_t obj; vm_offset_t toff, tinc, size; vm_page_t m; vm_ooffset_t off; ASSERT_VOP_LOCKED(vp, "inmem"); if (incore(&vp->v_bufobj, blkno)) return 1; if (vp->v_mount == NULL) return 0; obj = vp->v_object; if (obj == NULL) return (0); size = PAGE_SIZE; if (size > vp->v_mount->mnt_stat.f_iosize) size = vp->v_mount->mnt_stat.f_iosize; off = (vm_ooffset_t)blkno * (vm_ooffset_t)vp->v_mount->mnt_stat.f_iosize; VM_OBJECT_RLOCK(obj); for (toff = 0; toff < vp->v_mount->mnt_stat.f_iosize; toff += tinc) { m = vm_page_lookup(obj, OFF_TO_IDX(off + toff)); if (!m) goto notinmem; tinc = size; if (tinc > PAGE_SIZE - ((toff + off) & PAGE_MASK)) tinc = PAGE_SIZE - ((toff + off) & PAGE_MASK); if (vm_page_is_valid(m, (vm_offset_t) ((toff + off) & PAGE_MASK), tinc) == 0) goto notinmem; } VM_OBJECT_RUNLOCK(obj); return 1; notinmem: VM_OBJECT_RUNLOCK(obj); return (0); } /* * Set the dirty range for a buffer based on the status of the dirty * bits in the pages comprising the buffer. The range is limited * to the size of the buffer. * * Tell the VM system that the pages associated with this buffer * are clean. This is used for delayed writes where the data is * going to go to disk eventually without additional VM intevention. * * Note that while we only really need to clean through to b_bcount, we * just go ahead and clean through to b_bufsize. */ static void vfs_clean_pages_dirty_buf(struct buf *bp) { vm_ooffset_t foff, noff, eoff; vm_page_t m; int i; if ((bp->b_flags & B_VMIO) == 0 || bp->b_bufsize == 0) return; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_clean_pages_dirty_buf: no buffer offset")); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); vfs_drain_busy_pages(bp); vfs_setdirty_locked_object(bp); for (i = 0; i < bp->b_npages; i++) { noff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; eoff = noff; if (eoff > bp->b_offset + bp->b_bufsize) eoff = bp->b_offset + bp->b_bufsize; m = bp->b_pages[i]; vfs_page_set_validclean(bp, foff, m); /* vm_page_clear_dirty(m, foff & PAGE_MASK, eoff - foff); */ foff = noff; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } static void vfs_setdirty_locked_object(struct buf *bp) { vm_object_t object; int i; object = bp->b_bufobj->bo_object; VM_OBJECT_ASSERT_WLOCKED(object); /* * We qualify the scan for modified pages on whether the * object has been flushed yet. */ if ((object->flags & OBJ_MIGHTBEDIRTY) != 0) { vm_offset_t boffset; vm_offset_t eoffset; /* * test the pages to see if they have been modified directly * by users through the VM system. */ for (i = 0; i < bp->b_npages; i++) vm_page_test_dirty(bp->b_pages[i]); /* * Calculate the encompassing dirty range, boffset and eoffset, * (eoffset - boffset) bytes. */ for (i = 0; i < bp->b_npages; i++) { if (bp->b_pages[i]->dirty) break; } boffset = (i << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); for (i = bp->b_npages - 1; i >= 0; --i) { if (bp->b_pages[i]->dirty) { break; } } eoffset = ((i + 1) << PAGE_SHIFT) - (bp->b_offset & PAGE_MASK); /* * Fit it to the buffer. */ if (eoffset > bp->b_bcount) eoffset = bp->b_bcount; /* * If we have a good dirty range, merge with the existing * dirty range. */ if (boffset < eoffset) { if (bp->b_dirtyoff > boffset) bp->b_dirtyoff = boffset; if (bp->b_dirtyend < eoffset) bp->b_dirtyend = eoffset; } } } /* * Allocate the KVA mapping for an existing buffer. * If an unmapped buffer is provided but a mapped buffer is requested, take * also care to properly setup mappings between pages and KVA. */ static void bp_unmapped_get_kva(struct buf *bp, daddr_t blkno, int size, int gbflags) { int bsize, maxsize, need_mapping, need_kva; off_t offset; need_mapping = bp->b_data == unmapped_buf && (gbflags & GB_UNMAPPED) == 0; need_kva = bp->b_kvabase == unmapped_buf && bp->b_data == unmapped_buf && (gbflags & GB_KVAALLOC) != 0; if (!need_mapping && !need_kva) return; BUF_CHECK_UNMAPPED(bp); if (need_mapping && bp->b_kvabase != unmapped_buf) { /* * Buffer is not mapped, but the KVA was already * reserved at the time of the instantiation. Use the * allocated space. */ goto has_addr; } /* * Calculate the amount of the address space we would reserve * if the buffer was mapped. */ bsize = vn_isdisk(bp->b_vp, NULL) ? DEV_BSIZE : bp->b_bufobj->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; maxsize = size + (offset & PAGE_MASK); maxsize = imax(maxsize, bsize); while (bufkva_alloc(bp, maxsize, gbflags) != 0) { if ((gbflags & GB_NOWAIT_BD) != 0) { /* * XXXKIB: defragmentation cannot * succeed, not sure what else to do. */ panic("GB_NOWAIT_BD and GB_UNMAPPED %p", bp); } atomic_add_int(&mappingrestarts, 1); bufspace_wait(bp->b_vp, gbflags, 0, 0); } has_addr: if (need_mapping) { /* b_offset is handled by bpmap_qenter. */ bp->b_data = bp->b_kvabase; BUF_CHECK_MAPPED(bp); bpmap_qenter(bp); } } /* * getblk: * * Get a block given a specified block and offset into a file/device. * The buffers B_DONE bit will be cleared on return, making it almost * ready for an I/O initiation. B_INVAL may or may not be set on * return. The caller should clear B_INVAL prior to initiating a * READ. * * For a non-VMIO buffer, B_CACHE is set to the opposite of B_INVAL for * an existing buffer. * * For a VMIO buffer, B_CACHE is modified according to the backing VM. * If getblk()ing a previously 0-sized invalid buffer, B_CACHE is set * and then cleared based on the backing VM. If the previous buffer is * non-0-sized but invalid, B_CACHE will be cleared. * * If getblk() must create a new buffer, the new buffer is returned with * both B_INVAL and B_CACHE clear unless it is a VMIO buffer, in which * case it is returned with B_INVAL clear and B_CACHE set based on the * backing VM. * * getblk() also forces a bwrite() for any B_DELWRI buffer whos * B_CACHE bit is clear. * * What this means, basically, is that the caller should use B_CACHE to * determine whether the buffer is fully valid or not and should clear * B_INVAL prior to issuing a read. If the caller intends to validate * the buffer by loading its data area with something, the caller needs * to clear B_INVAL. If the caller does this without issuing an I/O, * the caller should set B_CACHE ( as an optimization ), else the caller * should issue the I/O and biodone() will set B_CACHE if the I/O was * a write attempt or if it was a successful read. If the caller * intends to issue a READ, the caller must clear B_INVAL and BIO_ERROR * prior to issuing the READ. biodone() will *not* clear B_INVAL. */ struct buf * getblk(struct vnode *vp, daddr_t blkno, int size, int slpflag, int slptimeo, int flags) { struct buf *bp; struct bufobj *bo; int bsize, error, maxsize, vmio; off_t offset; CTR3(KTR_BUF, "getblk(%p, %ld, %d)", vp, (long)blkno, size); KASSERT((flags & (GB_UNMAPPED | GB_KVAALLOC)) != GB_KVAALLOC, ("GB_KVAALLOC only makes sense with GB_UNMAPPED")); ASSERT_VOP_LOCKED(vp, "getblk"); if (size > MAXBCACHEBUF) panic("getblk: size(%d) > MAXBCACHEBUF(%d)\n", size, MAXBCACHEBUF); if (!unmapped_buf_allowed) flags &= ~(GB_UNMAPPED | GB_KVAALLOC); bo = &vp->v_bufobj; loop: BO_RLOCK(bo); bp = gbincore(bo, blkno); if (bp != NULL) { int lockflags; /* * Buffer is in-core. If the buffer is not busy nor managed, * it must be on a queue. */ lockflags = LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK; if (flags & GB_LOCK_NOWAIT) lockflags |= LK_NOWAIT; error = BUF_TIMELOCK(bp, lockflags, BO_LOCKPTR(bo), "getblk", slpflag, slptimeo); /* * If we slept and got the lock we have to restart in case * the buffer changed identities. */ if (error == ENOLCK) goto loop; /* We timed out or were interrupted. */ else if (error) return (NULL); /* If recursed, assume caller knows the rules. */ else if (BUF_LOCKRECURSED(bp)) goto end; /* * The buffer is locked. B_CACHE is cleared if the buffer is * invalid. Otherwise, for a non-VMIO buffer, B_CACHE is set * and for a VMIO buffer B_CACHE is adjusted according to the * backing VM cache. */ if (bp->b_flags & B_INVAL) bp->b_flags &= ~B_CACHE; else if ((bp->b_flags & (B_VMIO | B_INVAL)) == 0) bp->b_flags |= B_CACHE; if (bp->b_flags & B_MANAGED) MPASS(bp->b_qindex == QUEUE_NONE); else bremfree(bp); /* * check for size inconsistencies for non-VMIO case. */ if (bp->b_bcount != size) { if ((bp->b_flags & B_VMIO) == 0 || (size > bp->b_kvasize)) { if (bp->b_flags & B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); } else { if (LIST_EMPTY(&bp->b_dep)) { bp->b_flags |= B_RELBUF; brelse(bp); } else { bp->b_flags |= B_NOCACHE; bwrite(bp); } } goto loop; } } /* * Handle the case of unmapped buffer which should * become mapped, or the buffer for which KVA * reservation is requested. */ bp_unmapped_get_kva(bp, blkno, size, flags); /* * If the size is inconsistent in the VMIO case, we can resize * the buffer. This might lead to B_CACHE getting set or * cleared. If the size has not changed, B_CACHE remains * unchanged from its previous state. */ allocbuf(bp, size); KASSERT(bp->b_offset != NOOFFSET, ("getblk: no buffer offset")); /* * A buffer with B_DELWRI set and B_CACHE clear must * be committed before we can return the buffer in * order to prevent the caller from issuing a read * ( due to B_CACHE not being set ) and overwriting * it. * * Most callers, including NFS and FFS, need this to * operate properly either because they assume they * can issue a read if B_CACHE is not set, or because * ( for example ) an uncached B_DELWRI might loop due * to softupdates re-dirtying the buffer. In the latter * case, B_CACHE is set after the first write completes, * preventing further loops. * NOTE! b*write() sets B_CACHE. If we cleared B_CACHE * above while extending the buffer, we cannot allow the * buffer to remain with B_CACHE set after the write * completes or it will represent a corrupt state. To * deal with this we set B_NOCACHE to scrap the buffer * after the write. * * We might be able to do something fancy, like setting * B_CACHE in bwrite() except if B_DELWRI is already set, * so the below call doesn't set B_CACHE, but that gets real * confusing. This is much easier. */ if ((bp->b_flags & (B_CACHE|B_DELWRI)) == B_DELWRI) { bp->b_flags |= B_NOCACHE; bwrite(bp); goto loop; } bp->b_flags &= ~B_DONE; } else { /* * Buffer is not in-core, create new buffer. The buffer * returned by getnewbuf() is locked. Note that the returned * buffer is also considered valid (not marked B_INVAL). */ BO_RUNLOCK(bo); /* * If the user does not want us to create the buffer, bail out * here. */ if (flags & GB_NOCREAT) return NULL; if (numfreebuffers == 0 && TD_IS_IDLETHREAD(curthread)) return NULL; bsize = vn_isdisk(vp, NULL) ? DEV_BSIZE : bo->bo_bsize; KASSERT(bsize != 0, ("bsize == 0, check bo->bo_bsize")); offset = blkno * bsize; vmio = vp->v_object != NULL; if (vmio) { maxsize = size + (offset & PAGE_MASK); } else { maxsize = size; /* Do not allow non-VMIO notmapped buffers. */ flags &= ~(GB_UNMAPPED | GB_KVAALLOC); } maxsize = imax(maxsize, bsize); bp = getnewbuf(vp, slpflag, slptimeo, maxsize, flags); if (bp == NULL) { if (slpflag || slptimeo) return NULL; /* * XXX This is here until the sleep path is diagnosed * enough to work under very low memory conditions. * * There's an issue on low memory, 4BSD+non-preempt * systems (eg MIPS routers with 32MB RAM) where buffer * exhaustion occurs without sleeping for buffer * reclaimation. This just sticks in a loop and * constantly attempts to allocate a buffer, which * hits exhaustion and tries to wakeup bufdaemon. * This never happens because we never yield. * * The real solution is to identify and fix these cases * so we aren't effectively busy-waiting in a loop * until the reclaimation path has cycles to run. */ kern_yield(PRI_USER); goto loop; } /* * This code is used to make sure that a buffer is not * created while the getnewbuf routine is blocked. * This can be a problem whether the vnode is locked or not. * If the buffer is created out from under us, we have to * throw away the one we just created. * * Note: this must occur before we associate the buffer * with the vp especially considering limitations in * the splay tree implementation when dealing with duplicate * lblkno's. */ BO_LOCK(bo); if (gbincore(bo, blkno)) { BO_UNLOCK(bo); bp->b_flags |= B_INVAL; brelse(bp); bufspace_release(maxsize); goto loop; } /* * Insert the buffer into the hash, so that it can * be found by incore. */ bp->b_blkno = bp->b_lblkno = blkno; bp->b_offset = offset; bgetvp(vp, bp); BO_UNLOCK(bo); /* * set B_VMIO bit. allocbuf() the buffer bigger. Since the * buffer size starts out as 0, B_CACHE will be set by * allocbuf() for the VMIO case prior to it testing the * backing store for validity. */ if (vmio) { bp->b_flags |= B_VMIO; KASSERT(vp->v_object == bp->b_bufobj->bo_object, ("ARGH! different b_bufobj->bo_object %p %p %p\n", bp, vp->v_object, bp->b_bufobj->bo_object)); } else { bp->b_flags &= ~B_VMIO; KASSERT(bp->b_bufobj->bo_object == NULL, ("ARGH! has b_bufobj->bo_object %p %p\n", bp, bp->b_bufobj->bo_object)); BUF_CHECK_MAPPED(bp); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags &= ~B_DONE; } CTR4(KTR_BUF, "getblk(%p, %ld, %d) = %p", vp, (long)blkno, size, bp); BUF_ASSERT_HELD(bp); end: buf_track(bp, __func__); KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); return (bp); } /* * Get an empty, disassociated buffer of given size. The buffer is initially * set to B_INVAL. */ struct buf * geteblk(int size, int flags) { struct buf *bp; int maxsize; maxsize = (size + BKVAMASK) & ~BKVAMASK; while ((bp = getnewbuf(NULL, 0, 0, maxsize, flags)) == NULL) { if ((flags & GB_NOWAIT_BD) && (curthread->td_pflags & TDP_BUFNEED) != 0) return (NULL); } allocbuf(bp, size); bufspace_release(maxsize); bp->b_flags |= B_INVAL; /* b_dep cleared by getnewbuf() */ BUF_ASSERT_HELD(bp); return (bp); } /* * Truncate the backing store for a non-vmio buffer. */ static void vfs_nonvmio_truncate(struct buf *bp, int newbsize) { if (bp->b_flags & B_MALLOC) { /* * malloced buffers are not shrunk */ if (newbsize == 0) { bufmallocadjust(bp, 0); free(bp->b_data, M_BIOBUF); bp->b_data = bp->b_kvabase; bp->b_flags &= ~B_MALLOC; } return; } vm_hold_free_pages(bp, newbsize); bufspace_adjust(bp, newbsize); } /* * Extend the backing for a non-VMIO buffer. */ static void vfs_nonvmio_extend(struct buf *bp, int newbsize) { caddr_t origbuf; int origbufsize; /* * We only use malloced memory on the first allocation. * and revert to page-allocated memory when the buffer * grows. * * There is a potential smp race here that could lead * to bufmallocspace slightly passing the max. It * is probably extremely rare and not worth worrying * over. */ if (bp->b_bufsize == 0 && newbsize <= PAGE_SIZE/2 && bufmallocspace < maxbufmallocspace) { bp->b_data = malloc(newbsize, M_BIOBUF, M_WAITOK); bp->b_flags |= B_MALLOC; bufmallocadjust(bp, newbsize); return; } /* * If the buffer is growing on its other-than-first * allocation then we revert to the page-allocation * scheme. */ origbuf = NULL; origbufsize = 0; if (bp->b_flags & B_MALLOC) { origbuf = bp->b_data; origbufsize = bp->b_bufsize; bp->b_data = bp->b_kvabase; bufmallocadjust(bp, 0); bp->b_flags &= ~B_MALLOC; newbsize = round_page(newbsize); } vm_hold_load_pages(bp, (vm_offset_t) bp->b_data + bp->b_bufsize, (vm_offset_t) bp->b_data + newbsize); if (origbuf != NULL) { bcopy(origbuf, bp->b_data, origbufsize); free(origbuf, M_BIOBUF); } bufspace_adjust(bp, newbsize); } /* * This code constitutes the buffer memory from either anonymous system * memory (in the case of non-VMIO operations) or from an associated * VM object (in the case of VMIO operations). This code is able to * resize a buffer up or down. * * Note that this code is tricky, and has many complications to resolve * deadlock or inconsistent data situations. Tread lightly!!! * There are B_CACHE and B_DELWRI interactions that must be dealt with by * the caller. Calling this code willy nilly can result in the loss of data. * * allocbuf() only adjusts B_CACHE for VMIO buffers. getblk() deals with * B_CACHE for the non-VMIO case. */ int allocbuf(struct buf *bp, int size) { int newbsize; BUF_ASSERT_HELD(bp); if (bp->b_bcount == size) return (1); if (bp->b_kvasize != 0 && bp->b_kvasize < size) panic("allocbuf: buffer too small"); newbsize = roundup2(size, DEV_BSIZE); if ((bp->b_flags & B_VMIO) == 0) { if ((bp->b_flags & B_MALLOC) == 0) newbsize = round_page(newbsize); /* * Just get anonymous memory from the kernel. Don't * mess with B_CACHE. */ if (newbsize < bp->b_bufsize) vfs_nonvmio_truncate(bp, newbsize); else if (newbsize > bp->b_bufsize) vfs_nonvmio_extend(bp, newbsize); } else { int desiredpages; desiredpages = (size == 0) ? 0 : num_pages((bp->b_offset & PAGE_MASK) + newbsize); if (bp->b_flags & B_MALLOC) panic("allocbuf: VMIO buffer can't be malloced"); /* * Set B_CACHE initially if buffer is 0 length or will become * 0-length. */ if (size == 0 || bp->b_bufsize == 0) bp->b_flags |= B_CACHE; if (newbsize < bp->b_bufsize) vfs_vmio_truncate(bp, desiredpages); /* XXX This looks as if it should be newbsize > b_bufsize */ else if (size > bp->b_bcount) vfs_vmio_extend(bp, desiredpages, size); bufspace_adjust(bp, newbsize); } bp->b_bcount = size; /* requested buffer size. */ return (1); } extern int inflight_transient_maps; void biodone(struct bio *bp) { struct mtx *mtxp; void (*done)(struct bio *); vm_offset_t start, end; biotrack(bp, __func__); if ((bp->bio_flags & BIO_TRANSIENT_MAPPING) != 0) { bp->bio_flags &= ~BIO_TRANSIENT_MAPPING; bp->bio_flags |= BIO_UNMAPPED; start = trunc_page((vm_offset_t)bp->bio_data); end = round_page((vm_offset_t)bp->bio_data + bp->bio_length); bp->bio_data = unmapped_buf; pmap_qremove(start, OFF_TO_IDX(end - start)); vmem_free(transient_arena, start, end - start); atomic_add_int(&inflight_transient_maps, -1); } done = bp->bio_done; if (done == NULL) { mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->bio_flags |= BIO_DONE; wakeup(bp); mtx_unlock(mtxp); } else { bp->bio_flags |= BIO_DONE; done(bp); } } /* * Wait for a BIO to finish. */ int biowait(struct bio *bp, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->bio_flags & BIO_DONE) == 0) msleep(bp, mtxp, PRIBIO, wchan, 0); mtx_unlock(mtxp); if (bp->bio_error != 0) return (bp->bio_error); if (!(bp->bio_flags & BIO_ERROR)) return (0); return (EIO); } void biofinish(struct bio *bp, struct devstat *stat, int error) { if (error) { bp->bio_error = error; bp->bio_flags |= BIO_ERROR; } if (stat != NULL) devstat_end_transaction_bio(stat, bp); biodone(bp); } #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) void biotrack_buf(struct bio *bp, const char *location) { buf_track(bp->bio_track_bp, location); } #endif /* * bufwait: * * Wait for buffer I/O completion, returning error status. The buffer * is left locked and B_DONE on return. B_EINTR is converted into an EINTR * error and cleared. */ int bufwait(struct buf *bp) { if (bp->b_iocmd == BIO_READ) bwait(bp, PRIBIO, "biord"); else bwait(bp, PRIBIO, "biowr"); if (bp->b_flags & B_EINTR) { bp->b_flags &= ~B_EINTR; return (EINTR); } if (bp->b_ioflags & BIO_ERROR) { return (bp->b_error ? bp->b_error : EIO); } else { return (0); } } /* * bufdone: * * Finish I/O on a buffer, optionally calling a completion function. * This is usually called from an interrupt so process blocking is * not allowed. * * biodone is also responsible for setting B_CACHE in a B_VMIO bp. * In a non-VMIO bp, B_CACHE will be set on the next getblk() * assuming B_INVAL is clear. * * For the VMIO case, we set B_CACHE if the op was a read and no * read error occurred, or if the op was a write. B_CACHE is never * set if the buffer is invalid or otherwise uncacheable. * * biodone does not mess with B_INVAL, allowing the I/O routine or the * initiator to leave B_INVAL set to brelse the buffer out of existence * in the biodone routine. */ void bufdone(struct buf *bp) { struct bufobj *dropobj; void (*biodone)(struct buf *); buf_track(bp, __func__); CTR3(KTR_BUF, "bufdone(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); dropobj = NULL; KASSERT(!(bp->b_flags & B_DONE), ("biodone: bp %p already done", bp)); BUF_ASSERT_HELD(bp); runningbufwakeup(bp); if (bp->b_iocmd == BIO_WRITE) dropobj = bp->b_bufobj; /* call optional completion function if requested */ if (bp->b_iodone != NULL) { biodone = bp->b_iodone; bp->b_iodone = NULL; (*biodone) (bp); if (dropobj) bufobj_wdrop(dropobj); return; } bufdone_finish(bp); if (dropobj) bufobj_wdrop(dropobj); } void bufdone_finish(struct buf *bp) { BUF_ASSERT_HELD(bp); if (!LIST_EMPTY(&bp->b_dep)) buf_complete(bp); if (bp->b_flags & B_VMIO) { /* * Set B_CACHE if the op was a normal read and no error * occurred. B_CACHE is set for writes in the b*write() * routines. */ if (bp->b_iocmd == BIO_READ && !(bp->b_flags & (B_INVAL|B_NOCACHE)) && !(bp->b_ioflags & BIO_ERROR)) bp->b_flags |= B_CACHE; vfs_vmio_iodone(bp); } /* * For asynchronous completions, release the buffer now. The brelse * will do a wakeup there if necessary - so no need to do a wakeup * here in the async case. The sync case always needs to do a wakeup. */ if (bp->b_flags & B_ASYNC) { if ((bp->b_flags & (B_NOCACHE | B_INVAL | B_RELBUF)) || (bp->b_ioflags & BIO_ERROR)) brelse(bp); else bqrelse(bp); } else bdone(bp); } /* * This routine is called in lieu of iodone in the case of * incomplete I/O. This keeps the busy status for pages * consistent. */ void vfs_unbusy_pages(struct buf *bp) { int i; vm_object_t obj; vm_page_t m; runningbufwakeup(bp); if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; VM_OBJECT_WLOCK(obj); for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (m == bogus_page) { m = vm_page_lookup(obj, OFF_TO_IDX(bp->b_offset) + i); if (!m) panic("vfs_unbusy_pages: page missing\n"); bp->b_pages[i] = m; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } else BUF_CHECK_UNMAPPED(bp); } vm_page_sunbusy(m); } vm_object_pip_wakeupn(obj, bp->b_npages); VM_OBJECT_WUNLOCK(obj); } /* * vfs_page_set_valid: * * Set the valid bits in a page based on the supplied offset. The * range is restricted to the buffer's size. * * This routine is typically called after a read completes. */ static void vfs_page_set_valid(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t eoff; /* * Compute the end offset, eoff, such that [off, eoff) does not span a * page boundary and eoff is not greater than the end of the buffer. * The end of the buffer, in this case, is our file EOF, not the * allocation size of the buffer. */ eoff = (off + PAGE_SIZE) & ~(vm_ooffset_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > off) vm_page_set_valid_range(m, off & PAGE_MASK, eoff - off); } /* * vfs_page_set_validclean: * * Set the valid bits and clear the dirty bits in a page based on the * supplied offset. The range is restricted to the buffer's size. */ static void vfs_page_set_validclean(struct buf *bp, vm_ooffset_t off, vm_page_t m) { vm_ooffset_t soff, eoff; /* * Start and end offsets in buffer. eoff - soff may not cross a * page boundary or cross the end of the buffer. The end of the * buffer, in this case, is our file EOF, not the allocation size * of the buffer. */ soff = off; eoff = (off + PAGE_SIZE) & ~(off_t)PAGE_MASK; if (eoff > bp->b_offset + bp->b_bcount) eoff = bp->b_offset + bp->b_bcount; /* * Set valid range. This is typically the entire buffer and thus the * entire page. */ if (eoff > soff) { vm_page_set_validclean( m, (vm_offset_t) (soff & PAGE_MASK), (vm_offset_t) (eoff - soff) ); } } /* * Ensure that all buffer pages are not exclusive busied. If any page is * exclusive busy, drain it. */ void vfs_drain_busy_pages(struct buf *bp) { vm_page_t m; int i, last_busied; VM_OBJECT_ASSERT_WLOCKED(bp->b_bufobj->bo_object); last_busied = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if (vm_page_xbusied(m)) { for (; last_busied < i; last_busied++) vm_page_sbusy(bp->b_pages[last_busied]); while (vm_page_xbusied(m)) { vm_page_lock(m); VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); vm_page_busy_sleep(m, "vbpage", true); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); } } } for (i = 0; i < last_busied; i++) vm_page_sunbusy(bp->b_pages[i]); } /* * This routine is called before a device strategy routine. * It is used to tell the VM system that paging I/O is in * progress, and treat the pages associated with the buffer * almost as being exclusive busy. Also the object paging_in_progress * flag is handled to make sure that the object doesn't become * inconsistent. * * Since I/O has not been initiated yet, certain buffer flags * such as BIO_ERROR or B_INVAL may be in an inconsistent state * and should be ignored. */ void vfs_busy_pages(struct buf *bp, int clear_modify) { int i, bogus; vm_object_t obj; vm_ooffset_t foff; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; obj = bp->b_bufobj->bo_object; foff = bp->b_offset; KASSERT(bp->b_offset != NOOFFSET, ("vfs_busy_pages: no buffer offset")); VM_OBJECT_WLOCK(obj); vfs_drain_busy_pages(bp); if (bp->b_bufsize != 0) vfs_setdirty_locked_object(bp); bogus = 0; for (i = 0; i < bp->b_npages; i++) { m = bp->b_pages[i]; if ((bp->b_flags & B_CLUSTER) == 0) { vm_object_pip_add(obj, 1); vm_page_sbusy(m); } /* * When readying a buffer for a read ( i.e * clear_modify == 0 ), it is important to do * bogus_page replacement for valid pages in * partially instantiated buffers. Partially * instantiated buffers can, in turn, occur when * reconstituting a buffer from its VM backing store * base. We only have to do this if B_CACHE is * clear ( which causes the I/O to occur in the * first place ). The replacement prevents the read * I/O from overwriting potentially dirty VM-backed * pages. XXX bogus page replacement is, uh, bogus. * It may not work properly with small-block devices. * We need to find a better way. */ if (clear_modify) { pmap_remove_write(m); vfs_page_set_validclean(bp, foff, m); } else if (m->valid == VM_PAGE_BITS_ALL && (bp->b_flags & B_CACHE) == 0) { bp->b_pages[i] = bogus_page; bogus++; } foff = (foff + PAGE_SIZE) & ~(off_t)PAGE_MASK; } VM_OBJECT_WUNLOCK(obj); if (bogus && buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); pmap_qenter(trunc_page((vm_offset_t)bp->b_data), bp->b_pages, bp->b_npages); } } /* * vfs_bio_set_valid: * * Set the range within the buffer to valid. The range is * relative to the beginning of the buffer, b_offset. Note that * b_offset itself may be offset from the beginning of the first * page. */ void vfs_bio_set_valid(struct buf *bp, int base, int size) { int i, n; vm_page_t m; if (!(bp->b_flags & B_VMIO)) return; /* * Fixup base to be relative to beginning of first page. * Set initial n to be the maximum number of bytes in the * first page that can be validated. */ base += (bp->b_offset & PAGE_MASK); n = PAGE_SIZE - (base & PAGE_MASK); VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; vm_page_set_valid_range(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); } /* * vfs_bio_clrbuf: * * If the specified buffer is a non-VMIO buffer, clear the entire * buffer. If the specified buffer is a VMIO buffer, clear and * validate only the previously invalid portions of the buffer. * This routine essentially fakes an I/O, so we need to clear * BIO_ERROR and B_INVAL. * * Note that while we only theoretically need to clear through b_bcount, * we go ahead and clear through b_bufsize. */ void vfs_bio_clrbuf(struct buf *bp) { int i, j, mask, sa, ea, slide; if ((bp->b_flags & (B_VMIO | B_MALLOC)) != B_VMIO) { clrbuf(bp); return; } bp->b_flags &= ~B_INVAL; bp->b_ioflags &= ~BIO_ERROR; VM_OBJECT_WLOCK(bp->b_bufobj->bo_object); if ((bp->b_npages == 1) && (bp->b_bufsize < PAGE_SIZE) && (bp->b_offset & PAGE_MASK) == 0) { if (bp->b_pages[0] == bogus_page) goto unlock; mask = (1 << (bp->b_bufsize / DEV_BSIZE)) - 1; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[0]->object); if ((bp->b_pages[0]->valid & mask) == mask) goto unlock; if ((bp->b_pages[0]->valid & mask) == 0) { pmap_zero_page_area(bp->b_pages[0], 0, bp->b_bufsize); bp->b_pages[0]->valid |= mask; goto unlock; } } sa = bp->b_offset & PAGE_MASK; slide = 0; for (i = 0; i < bp->b_npages; i++, sa = 0) { slide = imin(slide + PAGE_SIZE, bp->b_offset + bp->b_bufsize); ea = slide & PAGE_MASK; if (ea == 0) ea = PAGE_SIZE; if (bp->b_pages[i] == bogus_page) continue; j = sa / DEV_BSIZE; mask = ((1 << ((ea - sa) / DEV_BSIZE)) - 1) << j; VM_OBJECT_ASSERT_WLOCKED(bp->b_pages[i]->object); if ((bp->b_pages[i]->valid & mask) == mask) continue; if ((bp->b_pages[i]->valid & mask) == 0) pmap_zero_page_area(bp->b_pages[i], sa, ea - sa); else { for (; sa < ea; sa += DEV_BSIZE, j++) { if ((bp->b_pages[i]->valid & (1 << j)) == 0) { pmap_zero_page_area(bp->b_pages[i], sa, DEV_BSIZE); } } } bp->b_pages[i]->valid |= mask; } unlock: VM_OBJECT_WUNLOCK(bp->b_bufobj->bo_object); bp->b_resid = 0; } void vfs_bio_bzero_buf(struct buf *bp, int base, int size) { vm_page_t m; int i, n; if (buf_mapped(bp)) { BUF_CHECK_MAPPED(bp); bzero(bp->b_data + base, size); } else { BUF_CHECK_UNMAPPED(bp); n = PAGE_SIZE - (base & PAGE_MASK); for (i = base / PAGE_SIZE; size > 0 && i < bp->b_npages; ++i) { m = bp->b_pages[i]; if (n > size) n = size; pmap_zero_page_area(m, base & PAGE_MASK, n); base += n; size -= n; n = PAGE_SIZE; } } } /* + * Update buffer flags based on I/O request parameters, optionally releasing the + * buffer. If it's VMIO or direct I/O, the buffer pages are released to the VM, + * where they may be placed on a page queue (VMIO) or freed immediately (direct + * I/O). Otherwise the buffer is released to the cache. + */ +static void +b_io_dismiss(struct buf *bp, int ioflag, bool release) +{ + + KASSERT((ioflag & IO_NOREUSE) == 0 || (ioflag & IO_VMIO) != 0, + ("buf %p non-VMIO noreuse", bp)); + + if ((ioflag & IO_DIRECT) != 0) + bp->b_flags |= B_DIRECT; + if ((ioflag & (IO_VMIO | IO_DIRECT)) != 0 && LIST_EMPTY(&bp->b_dep)) { + bp->b_flags |= B_RELBUF; + if ((ioflag & IO_NOREUSE) != 0) + bp->b_flags |= B_NOREUSE; + if (release) + brelse(bp); + } else if (release) + bqrelse(bp); +} + +void +vfs_bio_brelse(struct buf *bp, int ioflag) +{ + + b_io_dismiss(bp, ioflag, true); +} + +void +vfs_bio_set_flags(struct buf *bp, int ioflag) +{ + + b_io_dismiss(bp, ioflag, false); +} + +/* * vm_hold_load_pages and vm_hold_free_pages get pages into * a buffers address space. The pages are anonymous and are * not associated with a file object. */ static void vm_hold_load_pages(struct buf *bp, vm_offset_t from, vm_offset_t to) { vm_offset_t pg; vm_page_t p; int index; BUF_CHECK_MAPPED(bp); to = round_page(to); from = round_page(from); index = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; for (pg = from; pg < to; pg += PAGE_SIZE, index++) { tryagain: /* * note: must allocate system pages since blocking here * could interfere with paging I/O, no matter which * process we are. */ p = vm_page_alloc(NULL, 0, VM_ALLOC_SYSTEM | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_COUNT((to - pg) >> PAGE_SHIFT)); if (p == NULL) { VM_WAIT; goto tryagain; } pmap_qenter(pg, &p, 1); bp->b_pages[index] = p; } bp->b_npages = index; } /* Return pages associated with this buf to the vm system */ static void vm_hold_free_pages(struct buf *bp, int newbsize) { vm_offset_t from; vm_page_t p; int index, newnpages; BUF_CHECK_MAPPED(bp); from = round_page((vm_offset_t)bp->b_data + newbsize); newnpages = (from - trunc_page((vm_offset_t)bp->b_data)) >> PAGE_SHIFT; if (bp->b_npages > newnpages) pmap_qremove(from, bp->b_npages - newnpages); for (index = newnpages; index < bp->b_npages; index++) { p = bp->b_pages[index]; bp->b_pages[index] = NULL; if (vm_page_sbusied(p)) printf("vm_hold_free_pages: blkno: %jd, lblkno: %jd\n", (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno); p->wire_count--; vm_page_free(p); atomic_subtract_int(&vm_cnt.v_wire_count, 1); } bp->b_npages = newnpages; } /* * Map an IO request into kernel virtual address space. * * All requests are (re)mapped into kernel VA space. * Notice that we use b_bufsize for the size of the buffer * to be mapped. b_bcount might be modified by the driver. * * Note that even if the caller determines that the address space should * be valid, a race or a smaller-file mapped into a larger space may * actually cause vmapbuf() to fail, so all callers of vmapbuf() MUST * check the return value. * * This function only works with pager buffers. */ int vmapbuf(struct buf *bp, int mapbuf) { vm_prot_t prot; int pidx; if (bp->b_bufsize < 0) return (-1); prot = VM_PROT_READ; if (bp->b_iocmd == BIO_READ) prot |= VM_PROT_WRITE; /* Less backwards than it looks */ if ((pidx = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map, (vm_offset_t)bp->b_data, bp->b_bufsize, prot, bp->b_pages, btoc(MAXPHYS))) < 0) return (-1); bp->b_npages = pidx; bp->b_offset = ((vm_offset_t)bp->b_data) & PAGE_MASK; if (mapbuf || !unmapped_buf_allowed) { pmap_qenter((vm_offset_t)bp->b_kvabase, bp->b_pages, pidx); bp->b_data = bp->b_kvabase + bp->b_offset; } else bp->b_data = unmapped_buf; return(0); } /* * Free the io map PTEs associated with this IO operation. * We also invalidate the TLB entries and restore the original b_addr. * * This function only works with pager buffers. */ void vunmapbuf(struct buf *bp) { int npages; npages = bp->b_npages; if (buf_mapped(bp)) pmap_qremove(trunc_page((vm_offset_t)bp->b_data), npages); vm_page_unhold_pages(bp->b_pages, npages); bp->b_data = unmapped_buf; } void bdone(struct buf *bp) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); bp->b_flags |= B_DONE; wakeup(bp); mtx_unlock(mtxp); } void bwait(struct buf *bp, u_char pri, const char *wchan) { struct mtx *mtxp; mtxp = mtx_pool_find(mtxpool_sleep, bp); mtx_lock(mtxp); while ((bp->b_flags & B_DONE) == 0) msleep(bp, mtxp, pri, wchan, 0); mtx_unlock(mtxp); } int bufsync(struct bufobj *bo, int waitfor) { return (VOP_FSYNC(bo2vnode(bo), waitfor, curthread)); } void bufstrategy(struct bufobj *bo, struct buf *bp) { int i = 0; struct vnode *vp; vp = bp->b_vp; KASSERT(vp == bo->bo_private, ("Inconsistent vnode bufstrategy")); KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("Wrong vnode in bufstrategy(bp=%p, vp=%p)", bp, vp)); i = VOP_STRATEGY(vp, bp); KASSERT(i == 0, ("VOP_STRATEGY failed bp=%p vp=%p", bp, bp->b_vp)); } void bufobj_wrefl(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); ASSERT_BO_WLOCKED(bo); bo->bo_numoutput++; } void bufobj_wref(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wref")); BO_LOCK(bo); bo->bo_numoutput++; BO_UNLOCK(bo); } void bufobj_wdrop(struct bufobj *bo) { KASSERT(bo != NULL, ("NULL bo in bufobj_wdrop")); BO_LOCK(bo); KASSERT(bo->bo_numoutput > 0, ("bufobj_wdrop non-positive count")); if ((--bo->bo_numoutput == 0) && (bo->bo_flag & BO_WWAIT)) { bo->bo_flag &= ~BO_WWAIT; wakeup(&bo->bo_numoutput); } BO_UNLOCK(bo); } int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo) { int error; KASSERT(bo != NULL, ("NULL bo in bufobj_wwait")); ASSERT_BO_WLOCKED(bo); error = 0; while (bo->bo_numoutput) { bo->bo_flag |= BO_WWAIT; error = msleep(&bo->bo_numoutput, BO_LOCKPTR(bo), slpflag | (PRIBIO + 1), "bo_wwait", timeo); if (error) break; } return (error); } /* * Set bio_data or bio_ma for struct bio from the struct buf. */ void bdata2bio(struct buf *bp, struct bio *bip) { if (!buf_mapped(bp)) { KASSERT(unmapped_buf_allowed, ("unmapped")); bip->bio_ma = bp->b_pages; bip->bio_ma_n = bp->b_npages; bip->bio_data = unmapped_buf; bip->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bip->bio_flags |= BIO_UNMAPPED; KASSERT(round_page(bip->bio_ma_offset + bip->bio_length) / PAGE_SIZE == bp->b_npages, ("Buffer %p too short: %d %lld %d", bp, bip->bio_ma_offset, (long long)bip->bio_length, bip->bio_ma_n)); } else { bip->bio_data = bp->b_data; bip->bio_ma = NULL; } } /* * The MIPS pmap code currently doesn't handle aliased pages. * The VIPT caches may not handle page aliasing themselves, leading * to data corruption. * * As such, this code makes a system extremely unhappy if said * system doesn't support unaliasing the above situation in hardware. * Some "recent" systems (eg some mips24k/mips74k cores) don't enable * this feature at build time, so it has to be handled in software. * * Once the MIPS pmap/cache code grows to support this function on * earlier chips, it should be flipped back off. */ #ifdef __mips__ static int buf_pager_relbuf = 1; #else static int buf_pager_relbuf = 0; #endif SYSCTL_INT(_vfs, OID_AUTO, buf_pager_relbuf, CTLFLAG_RWTUN, &buf_pager_relbuf, 0, "Make buffer pager release buffers after reading"); /* * The buffer pager. It uses buffer reads to validate pages. * * In contrast to the generic local pager from vm/vnode_pager.c, this * pager correctly and easily handles volumes where the underlying * device block size is greater than the machine page size. The * buffer cache transparently extends the requested page run to be * aligned at the block boundary, and does the necessary bogus page * replacements in the addends to avoid obliterating already valid * pages. * * The only non-trivial issue is that the exclusive busy state for * pages, which is assumed by the vm_pager_getpages() interface, is * incompatible with the VMIO buffer cache's desire to share-busy the * pages. This function performs a trivial downgrade of the pages' * state before reading buffers, and a less trivial upgrade from the * shared-busy to excl-busy state after the read. */ int vfs_bio_getpages(struct vnode *vp, vm_page_t *ma, int count, int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno, vbg_get_blksize_t get_blksize) { vm_page_t m; vm_object_t object; struct buf *bp; struct mount *mp; daddr_t lbn, lbnp; vm_ooffset_t la, lb, poff, poffe; long bsize; int bo_bs, br_flags, error, i, pgsin, pgsin_a, pgsin_b; bool redo, lpart; object = vp->v_object; mp = vp->v_mount; la = IDX_TO_OFF(ma[count - 1]->pindex); if (la >= object->un_pager.vnp.vnp_size) return (VM_PAGER_BAD); lpart = la + PAGE_SIZE > object->un_pager.vnp.vnp_size; bo_bs = get_blksize(vp, get_lblkno(vp, IDX_TO_OFF(ma[0]->pindex))); /* * Calculate read-ahead, behind and total pages. */ pgsin = count; lb = IDX_TO_OFF(ma[0]->pindex); pgsin_b = OFF_TO_IDX(lb - rounddown2(lb, bo_bs)); pgsin += pgsin_b; if (rbehind != NULL) *rbehind = pgsin_b; pgsin_a = OFF_TO_IDX(roundup2(la, bo_bs) - la); if (la + IDX_TO_OFF(pgsin_a) >= object->un_pager.vnp.vnp_size) pgsin_a = OFF_TO_IDX(roundup2(object->un_pager.vnp.vnp_size, PAGE_SIZE) - la); pgsin += pgsin_a; if (rahead != NULL) *rahead = pgsin_a; PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, pgsin); br_flags = (mp != NULL && (mp->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0) ? GB_UNMAPPED : 0; VM_OBJECT_WLOCK(object); again: for (i = 0; i < count; i++) vm_page_busy_downgrade(ma[i]); VM_OBJECT_WUNLOCK(object); lbnp = -1; for (i = 0; i < count; i++) { m = ma[i]; /* * Pages are shared busy and the object lock is not * owned, which together allow for the pages' * invalidation. The racy test for validity avoids * useless creation of the buffer for the most typical * case when invalidation is not used in redo or for * parallel read. The shared->excl upgrade loop at * the end of the function catches the race in a * reliable way (protected by the object lock). */ if (m->valid == VM_PAGE_BITS_ALL) continue; poff = IDX_TO_OFF(m->pindex); poffe = MIN(poff + PAGE_SIZE, object->un_pager.vnp.vnp_size); for (; poff < poffe; poff += bsize) { lbn = get_lblkno(vp, poff); if (lbn == lbnp) goto next_page; lbnp = lbn; bsize = get_blksize(vp, lbn); error = bread_gb(vp, lbn, bsize, curthread->td_ucred, br_flags, &bp); if (error != 0) goto end_pages; if (LIST_EMPTY(&bp->b_dep)) { /* * Invalidation clears m->valid, but * may leave B_CACHE flag if the * buffer existed at the invalidation * time. In this case, recycle the * buffer to do real read on next * bread() after redo. * * Otherwise B_RELBUF is not strictly * necessary, enable to reduce buf * cache pressure. */ if (buf_pager_relbuf || m->valid != VM_PAGE_BITS_ALL) bp->b_flags |= B_RELBUF; bp->b_flags &= ~B_NOCACHE; brelse(bp); } else { bqrelse(bp); } } KASSERT(1 /* racy, enable for debugging */ || m->valid == VM_PAGE_BITS_ALL || i == count - 1, ("buf %d %p invalid", i, m)); if (i == count - 1 && lpart) { VM_OBJECT_WLOCK(object); if (m->valid != 0 && m->valid != VM_PAGE_BITS_ALL) vm_page_zero_invalid(m, TRUE); VM_OBJECT_WUNLOCK(object); } next_page:; } end_pages: VM_OBJECT_WLOCK(object); redo = false; for (i = 0; i < count; i++) { vm_page_sunbusy(ma[i]); ma[i] = vm_page_grab(object, ma[i]->pindex, VM_ALLOC_NORMAL); /* * Since the pages were only sbusy while neither the * buffer nor the object lock was held by us, or * reallocated while vm_page_grab() slept for busy * relinguish, they could have been invalidated. * Recheck the valid bits and re-read as needed. * * Note that the last page is made fully valid in the * read loop, and partial validity for the page at * index count - 1 could mean that the page was * invalidated or removed, so we must restart for * safety as well. */ if (ma[i]->valid != VM_PAGE_BITS_ALL) redo = true; } if (redo && error == 0) goto again; VM_OBJECT_WUNLOCK(object); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } #include "opt_ddb.h" #ifdef DDB #include /* DDB command to show buffer data */ DB_SHOW_COMMAND(buffer, db_show_buffer) { /* get args */ struct buf *bp = (struct buf *)addr; #ifdef FULL_BUF_TRACKING uint32_t i, j; #endif if (!have_addr) { db_printf("usage: show buffer \n"); return; } db_printf("buf at %p\n", bp); db_printf("b_flags = 0x%b, b_xflags=0x%b, b_vflags=0x%b\n", (u_int)bp->b_flags, PRINT_BUF_FLAGS, (u_int)bp->b_xflags, PRINT_BUF_XFLAGS, (u_int)bp->b_vflags, PRINT_BUF_VFLAGS); db_printf( "b_error = %d, b_bufsize = %ld, b_bcount = %ld, b_resid = %ld\n" "b_bufobj = (%p), b_data = %p, b_blkno = %jd, b_lblkno = %jd, " "b_dep = %p\n", bp->b_error, bp->b_bufsize, bp->b_bcount, bp->b_resid, bp->b_bufobj, bp->b_data, (intmax_t)bp->b_blkno, (intmax_t)bp->b_lblkno, bp->b_dep.lh_first); db_printf("b_kvabase = %p, b_kvasize = %d\n", bp->b_kvabase, bp->b_kvasize); if (bp->b_npages) { int i; db_printf("b_npages = %d, pages(OBJ, IDX, PA): ", bp->b_npages); for (i = 0; i < bp->b_npages; i++) { vm_page_t m; m = bp->b_pages[i]; if (m != NULL) db_printf("(%p, 0x%lx, 0x%lx)", m->object, (u_long)m->pindex, (u_long)VM_PAGE_TO_PHYS(m)); else db_printf("( ??? )"); if ((i + 1) < bp->b_npages) db_printf(","); } db_printf("\n"); } #if defined(FULL_BUF_TRACKING) db_printf("b_io_tracking: b_io_tcnt = %u\n", bp->b_io_tcnt); i = bp->b_io_tcnt % BUF_TRACKING_SIZE; for (j = 1; j <= BUF_TRACKING_SIZE; j++) db_printf(" %2u: %s\n", j, bp->b_io_tracking[BUF_TRACKING_ENTRY(i - j)]); #elif defined(BUF_TRACKING) db_printf("b_io_tracking: %s\n", bp->b_io_tracking); #endif db_printf(" "); BUF_LOCKPRINTINFO(bp); } DB_SHOW_COMMAND(lockedbufs, lockedbufs) { struct buf *bp; int i; for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (BUF_ISLOCKED(bp)) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } } DB_SHOW_COMMAND(vnodebufs, db_show_vnodebufs) { struct vnode *vp; struct buf *bp; if (!have_addr) { db_printf("usage: show vnodebufs \n"); return; } vp = (struct vnode *)addr; db_printf("Clean buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_clean.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } db_printf("Dirty buffers:\n"); TAILQ_FOREACH(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs) { db_show_buffer((uintptr_t)bp, 1, 0, NULL); db_printf("\n"); } } DB_COMMAND(countfreebufs, db_coundfreebufs) { struct buf *bp; int i, used = 0, nfree = 0; if (have_addr) { db_printf("usage: countfreebufs\n"); return; } for (i = 0; i < nbuf; i++) { bp = &buf[i]; if (bp->b_qindex == QUEUE_EMPTY) nfree++; else used++; } db_printf("Counted %d free, %d used (%d tot)\n", nfree, used, nfree + used); db_printf("numfreebuffers is %d\n", numfreebuffers); } #endif /* DDB */ Index: head/sys/sys/buf.h =================================================================== --- head/sys/sys/buf.h (revision 309061) +++ head/sys/sys/buf.h (revision 309062) @@ -1,569 +1,571 @@ /*- * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)buf.h 8.9 (Berkeley) 3/30/95 * $FreeBSD$ */ #ifndef _SYS_BUF_H_ #define _SYS_BUF_H_ #include #include #include #include struct bio; struct buf; struct bufobj; struct mount; struct vnode; struct uio; /* * To avoid including */ LIST_HEAD(workhead, worklist); /* * These are currently used only by the soft dependency code, hence * are stored once in a global variable. If other subsystems wanted * to use these hooks, a pointer to a set of bio_ops could be added * to each buffer. */ extern struct bio_ops { void (*io_start)(struct buf *); void (*io_complete)(struct buf *); void (*io_deallocate)(struct buf *); int (*io_countdeps)(struct buf *, int); } bioops; struct vm_object; struct vm_page; typedef unsigned char b_xflags_t; /* * The buffer header describes an I/O operation in the kernel. * * NOTES: * b_bufsize, b_bcount. b_bufsize is the allocation size of the * buffer, either DEV_BSIZE or PAGE_SIZE aligned. b_bcount is the * originally requested buffer size and can serve as a bounds check * against EOF. For most, but not all uses, b_bcount == b_bufsize. * * b_dirtyoff, b_dirtyend. Buffers support piecemeal, unaligned * ranges of dirty data that need to be written to backing store. * The range is typically clipped at b_bcount ( not b_bufsize ). * * b_resid. Number of bytes remaining in I/O. After an I/O operation * completes, b_resid is usually 0 indicating 100% success. * * All fields are protected by the buffer lock except those marked: * V - Protected by owning bufobj lock * Q - Protected by the buf queue lock * D - Protected by an dependency implementation specific lock */ struct buf { struct bufobj *b_bufobj; long b_bcount; void *b_caller1; caddr_t b_data; int b_error; uint16_t b_iocmd; /* BIO_* bio_cmd from bio.h */ uint16_t b_ioflags; /* BIO_* bio_flags from bio.h */ off_t b_iooffset; long b_resid; void (*b_iodone)(struct buf *); daddr_t b_blkno; /* Underlying physical block number. */ off_t b_offset; /* Offset into file. */ TAILQ_ENTRY(buf) b_bobufs; /* (V) Buffer's associated vnode. */ uint32_t b_vflags; /* (V) BV_* flags */ unsigned short b_qindex; /* (Q) buffer queue index */ uint32_t b_flags; /* B_* flags. */ b_xflags_t b_xflags; /* extra flags */ struct lock b_lock; /* Buffer lock */ long b_bufsize; /* Allocated buffer size. */ int b_runningbufspace; /* when I/O is running, pipelining */ int b_kvasize; /* size of kva for buffer */ int b_dirtyoff; /* Offset in buffer of dirty region. */ int b_dirtyend; /* Offset of end of dirty region. */ caddr_t b_kvabase; /* base kva for buffer */ daddr_t b_lblkno; /* Logical block number. */ struct vnode *b_vp; /* Device vnode. */ struct ucred *b_rcred; /* Read credentials reference. */ struct ucred *b_wcred; /* Write credentials reference. */ union { TAILQ_ENTRY(buf) b_freelist; /* (Q) */ struct { void (*b_pgiodone)(void *, vm_page_t *, int, int); int b_pgbefore; int b_pgafter; }; }; union cluster_info { TAILQ_HEAD(cluster_list_head, buf) cluster_head; TAILQ_ENTRY(buf) cluster_entry; } b_cluster; struct vm_page *b_pages[btoc(MAXPHYS)]; int b_npages; struct workhead b_dep; /* (D) List of filesystem dependencies. */ void *b_fsprivate1; void *b_fsprivate2; void *b_fsprivate3; #if defined(FULL_BUF_TRACKING) #define BUF_TRACKING_SIZE 32 #define BUF_TRACKING_ENTRY(x) ((x) & (BUF_TRACKING_SIZE - 1)) const char *b_io_tracking[BUF_TRACKING_SIZE]; uint32_t b_io_tcnt; #elif defined(BUF_TRACKING) const char *b_io_tracking; #endif }; #define b_object b_bufobj->bo_object /* * These flags are kept in b_flags. * * Notes: * * B_ASYNC VOP calls on bp's are usually async whether or not * B_ASYNC is set, but some subsystems, such as NFS, like * to know what is best for the caller so they can * optimize the I/O. * * B_PAGING Indicates that bp is being used by the paging system or * some paging system and that the bp is not linked into * the b_vp's clean/dirty linked lists or ref counts. * Buffer vp reassignments are illegal in this case. * * B_CACHE This may only be set if the buffer is entirely valid. * The situation where B_DELWRI is set and B_CACHE is * clear MUST be committed to disk by getblk() so * B_DELWRI can also be cleared. See the comments for * getblk() in kern/vfs_bio.c. If B_CACHE is clear, * the caller is expected to clear BIO_ERROR and B_INVAL, * set BIO_READ, and initiate an I/O. * * The 'entire buffer' is defined to be the range from * 0 through b_bcount. * * B_MALLOC Request that the buffer be allocated from the malloc * pool, DEV_BSIZE aligned instead of PAGE_SIZE aligned. * * B_CLUSTEROK This flag is typically set for B_DELWRI buffers * by filesystems that allow clustering when the buffer * is fully dirty and indicates that it may be clustered * with other adjacent dirty buffers. Note the clustering * may not be used with the stage 1 data write under NFS * but may be used for the commit rpc portion. * * B_VMIO Indicates that the buffer is tied into an VM object. * The buffer's data is always PAGE_SIZE aligned even * if b_bufsize and b_bcount are not. ( b_bufsize is * always at least DEV_BSIZE aligned, though ). * * B_DIRECT Hint that we should attempt to completely free * the pages underlying the buffer. B_DIRECT is * sticky until the buffer is released and typically * only has an effect when B_RELBUF is also set. * */ #define B_AGE 0x00000001 /* Move to age queue when I/O done. */ #define B_NEEDCOMMIT 0x00000002 /* Append-write in progress. */ #define B_ASYNC 0x00000004 /* Start I/O, do not wait. */ #define B_DIRECT 0x00000008 /* direct I/O flag (pls free vmio) */ #define B_DEFERRED 0x00000010 /* Skipped over for cleaning */ #define B_CACHE 0x00000020 /* Bread found us in the cache. */ #define B_VALIDSUSPWRT 0x00000040 /* Valid write during suspension. */ #define B_DELWRI 0x00000080 /* Delay I/O until buffer reused. */ #define B_00000100 0x00000100 /* Available flag. */ #define B_DONE 0x00000200 /* I/O completed. */ #define B_EINTR 0x00000400 /* I/O was interrupted */ #define B_NOREUSE 0x00000800 /* Contents not reused once released. */ #define B_00001000 0x00001000 /* Available flag. */ #define B_INVAL 0x00002000 /* Does not contain valid info. */ #define B_BARRIER 0x00004000 /* Write this and all preceding first. */ #define B_NOCACHE 0x00008000 /* Do not cache block after use. */ #define B_MALLOC 0x00010000 /* malloced b_data */ #define B_CLUSTEROK 0x00020000 /* Pagein op, so swap() can count it. */ #define B_00040000 0x00040000 /* Available flag. */ #define B_00080000 0x00080000 /* Available flag. */ #define B_00100000 0x00100000 /* Available flag. */ #define B_00200000 0x00200000 /* Available flag. */ #define B_RELBUF 0x00400000 /* Release VMIO buffer. */ #define B_FS_FLAG1 0x00800000 /* Available flag for FS use. */ #define B_NOCOPY 0x01000000 /* Don't copy-on-write this buf. */ #define B_INFREECNT 0x02000000 /* buf is counted in numfreebufs */ #define B_PAGING 0x04000000 /* volatile paging I/O -- bypass VMIO */ #define B_MANAGED 0x08000000 /* Managed by FS. */ #define B_RAM 0x10000000 /* Read ahead mark (flag) */ #define B_VMIO 0x20000000 /* VMIO flag */ #define B_CLUSTER 0x40000000 /* pagein op, so swap() can count it */ #define B_REMFREE 0x80000000 /* Delayed bremfree */ #define PRINT_BUF_FLAGS "\20\40remfree\37cluster\36vmio\35ram\34managed" \ "\33paging\32infreecnt\31nocopy\30b23\27relbuf\26b21\25b20" \ "\24b19\23b18\22clusterok\21malloc\20nocache\17b14\16inval" \ "\15b12\14noreuse\13eintr\12done\11b8\10delwri" \ "\7validsuspwrt\6cache\5deferred\4direct\3async\2needcommit\1age" /* * These flags are kept in b_xflags. */ #define BX_VNDIRTY 0x00000001 /* On vnode dirty list */ #define BX_VNCLEAN 0x00000002 /* On vnode clean list */ #define BX_BKGRDWRITE 0x00000010 /* Do writes in background */ #define BX_BKGRDMARKER 0x00000020 /* Mark buffer for splay tree */ #define BX_ALTDATA 0x00000040 /* Holds extended data */ #define PRINT_BUF_XFLAGS "\20\7altdata\6bkgrdmarker\5bkgrdwrite\2clean\1dirty" #define NOOFFSET (-1LL) /* No buffer offset calculated yet */ /* * These flags are kept in b_vflags. */ #define BV_SCANNED 0x00000001 /* VOP_FSYNC funcs mark written bufs */ #define BV_BKGRDINPROG 0x00000002 /* Background write in progress */ #define BV_BKGRDWAIT 0x00000004 /* Background write waiting */ #define BV_BKGRDERR 0x00000008 /* Error from background write */ #define PRINT_BUF_VFLAGS "\20\4bkgrderr\3bkgrdwait\2bkgrdinprog\1scanned" #ifdef _KERNEL /* * Buffer locking */ extern const char *buf_wmesg; /* Default buffer lock message */ #define BUF_WMESG "bufwait" #include /* XXX for curthread */ #include /* * Initialize a lock. */ #define BUF_LOCKINIT(bp) \ lockinit(&(bp)->b_lock, PRIBIO + 4, buf_wmesg, 0, 0) /* * * Get a lock sleeping non-interruptably until it becomes available. */ #define BUF_LOCK(bp, locktype, interlock) \ _lockmgr_args_rw(&(bp)->b_lock, (locktype), (interlock), \ LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, \ LOCK_FILE, LOCK_LINE) /* * Get a lock sleeping with specified interruptably and timeout. */ #define BUF_TIMELOCK(bp, locktype, interlock, wmesg, catch, timo) \ _lockmgr_args_rw(&(bp)->b_lock, (locktype) | LK_TIMELOCK, \ (interlock), (wmesg), (PRIBIO + 4) | (catch), (timo), \ LOCK_FILE, LOCK_LINE) /* * Release a lock. Only the acquiring process may free the lock unless * it has been handed off to biodone. */ #define BUF_UNLOCK(bp) do { \ KASSERT(((bp)->b_flags & B_REMFREE) == 0, \ ("BUF_UNLOCK %p while B_REMFREE is still set.", (bp))); \ \ (void)_lockmgr_args(&(bp)->b_lock, LK_RELEASE, NULL, \ LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, \ LOCK_FILE, LOCK_LINE); \ } while (0) /* * Check if a buffer lock is recursed. */ #define BUF_LOCKRECURSED(bp) \ lockmgr_recursed(&(bp)->b_lock) /* * Check if a buffer lock is currently held. */ #define BUF_ISLOCKED(bp) \ lockstatus(&(bp)->b_lock) /* * Free a buffer lock. */ #define BUF_LOCKFREE(bp) \ lockdestroy(&(bp)->b_lock) /* * Print informations on a buffer lock. */ #define BUF_LOCKPRINTINFO(bp) \ lockmgr_printinfo(&(bp)->b_lock) /* * Buffer lock assertions. */ #if defined(INVARIANTS) && defined(INVARIANT_SUPPORT) #define BUF_ASSERT_LOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_LOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_SLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_SLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_XLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_XLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_UNLOCKED(bp) \ _lockmgr_assert(&(bp)->b_lock, KA_UNLOCKED, LOCK_FILE, LOCK_LINE) #define BUF_ASSERT_HELD(bp) #define BUF_ASSERT_UNHELD(bp) #else #define BUF_ASSERT_LOCKED(bp) #define BUF_ASSERT_SLOCKED(bp) #define BUF_ASSERT_XLOCKED(bp) #define BUF_ASSERT_UNLOCKED(bp) #define BUF_ASSERT_HELD(bp) #define BUF_ASSERT_UNHELD(bp) #endif #ifdef _SYS_PROC_H_ /* Avoid #include pollution */ /* * When initiating asynchronous I/O, change ownership of the lock to the * kernel. Once done, the lock may legally released by biodone. The * original owning process can no longer acquire it recursively, but must * wait until the I/O is completed and the lock has been freed by biodone. */ #define BUF_KERNPROC(bp) \ _lockmgr_disown(&(bp)->b_lock, LOCK_FILE, LOCK_LINE) #endif #endif /* _KERNEL */ struct buf_queue_head { TAILQ_HEAD(buf_queue, buf) queue; daddr_t last_pblkno; struct buf *insert_point; struct buf *switch_point; }; /* * This structure describes a clustered I/O. */ struct cluster_save { long bs_bcount; /* Saved b_bcount. */ long bs_bufsize; /* Saved b_bufsize. */ int bs_nchildren; /* Number of associated buffers. */ struct buf **bs_children; /* List of associated buffers. */ }; #ifdef _KERNEL static __inline int bwrite(struct buf *bp) { KASSERT(bp->b_bufobj != NULL, ("bwrite: no bufobj bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops != NULL, ("bwrite: no bo_ops bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops->bop_write != NULL, ("bwrite: no bop_write bp=%p", bp)); return (BO_WRITE(bp->b_bufobj, bp)); } static __inline void bstrategy(struct buf *bp) { KASSERT(bp->b_bufobj != NULL, ("bstrategy: no bufobj bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops != NULL, ("bstrategy: no bo_ops bp=%p", bp)); KASSERT(bp->b_bufobj->bo_ops->bop_strategy != NULL, ("bstrategy: no bop_strategy bp=%p", bp)); BO_STRATEGY(bp->b_bufobj, bp); } static __inline void buf_start(struct buf *bp) { if (bioops.io_start) (*bioops.io_start)(bp); } static __inline void buf_complete(struct buf *bp) { if (bioops.io_complete) (*bioops.io_complete)(bp); } static __inline void buf_deallocate(struct buf *bp) { if (bioops.io_deallocate) (*bioops.io_deallocate)(bp); } static __inline int buf_countdeps(struct buf *bp, int i) { if (bioops.io_countdeps) return ((*bioops.io_countdeps)(bp, i)); else return (0); } static __inline void buf_track(struct buf *bp, const char *location) { #if defined(FULL_BUF_TRACKING) bp->b_io_tracking[BUF_TRACKING_ENTRY(bp->b_io_tcnt++)] = location; #elif defined(BUF_TRACKING) bp->b_io_tracking = location; #endif } #endif /* _KERNEL */ /* * Zero out the buffer's data area. */ #define clrbuf(bp) { \ bzero((bp)->b_data, (u_int)(bp)->b_bcount); \ (bp)->b_resid = 0; \ } /* * Flags for getblk's last parameter. */ #define GB_LOCK_NOWAIT 0x0001 /* Fail if we block on a buf lock. */ #define GB_NOCREAT 0x0002 /* Don't create a buf if not found. */ #define GB_NOWAIT_BD 0x0004 /* Do not wait for bufdaemon. */ #define GB_UNMAPPED 0x0008 /* Do not mmap buffer pages. */ #define GB_KVAALLOC 0x0010 /* But allocate KVA. */ #ifdef _KERNEL extern int nbuf; /* The number of buffer headers */ extern long maxswzone; /* Max KVA for swap structures */ extern long maxbcache; /* Max KVA for buffer cache */ extern long runningbufspace; extern long hibufspace; extern int dirtybufthresh; extern int bdwriteskip; extern int dirtybufferflushes; extern int altbufferflushes; extern int nswbuf; /* Number of swap I/O buffer headers. */ extern int cluster_pbuf_freecnt; /* Number of pbufs for clusters */ extern int vnode_pbuf_freecnt; /* Number of pbufs for vnode pager */ extern int vnode_async_pbuf_freecnt; /* Number of pbufs for vnode pager, asynchronous reads */ extern caddr_t unmapped_buf; /* Data address for unmapped buffers. */ static inline int buf_mapped(struct buf *bp) { return (bp->b_data != unmapped_buf); } void runningbufwakeup(struct buf *); void waitrunningbufspace(void); caddr_t kern_vfs_bio_buffer_alloc(caddr_t v, long physmem_est); void bufinit(void); void bufshutdown(int); void bdata2bio(struct buf *bp, struct bio *bip); void bwillwrite(void); int buf_dirty_count_severe(void); void bremfree(struct buf *); void bremfreef(struct buf *); /* XXX Force bremfree, only for nfs. */ #define bread(vp, blkno, size, cred, bpp) \ breadn_flags(vp, blkno, size, NULL, NULL, 0, cred, 0, bpp) #define bread_gb(vp, blkno, size, cred, gbflags, bpp) \ breadn_flags(vp, blkno, size, NULL, NULL, 0, cred, \ gbflags, bpp) #define breadn(vp, blkno, size, rablkno, rabsize, cnt, cred, bpp) \ breadn_flags(vp, blkno, size, rablkno, rabsize, cnt, cred, 0, bpp) int breadn_flags(struct vnode *, daddr_t, int, daddr_t *, int *, int, struct ucred *, int, struct buf **); void breada(struct vnode *, daddr_t *, int *, int, struct ucred *); void bdwrite(struct buf *); void bawrite(struct buf *); void babarrierwrite(struct buf *); int bbarrierwrite(struct buf *); void bdirty(struct buf *); void bundirty(struct buf *); void bufstrategy(struct bufobj *, struct buf *); void brelse(struct buf *); void bqrelse(struct buf *); int vfs_bio_awrite(struct buf *); void vfs_drain_busy_pages(struct buf *bp); struct buf * getpbuf(int *); struct buf *incore(struct bufobj *, daddr_t); struct buf *gbincore(struct bufobj *, daddr_t); struct buf *getblk(struct vnode *, daddr_t, int, int, int, int); struct buf *geteblk(int, int); int bufwait(struct buf *); int bufwrite(struct buf *); void bufdone(struct buf *); void bufdone_finish(struct buf *); void bd_speedup(void); int cluster_read(struct vnode *, u_quad_t, daddr_t, long, struct ucred *, long, int, int, struct buf **); int cluster_wbuild(struct vnode *, long, daddr_t, int, int); void cluster_write(struct vnode *, struct buf *, u_quad_t, int, int); +void vfs_bio_brelse(struct buf *bp, int ioflags); void vfs_bio_bzero_buf(struct buf *bp, int base, int size); -void vfs_bio_set_valid(struct buf *, int base, int size); void vfs_bio_clrbuf(struct buf *); +void vfs_bio_set_flags(struct buf *bp, int ioflags); +void vfs_bio_set_valid(struct buf *, int base, int size); void vfs_busy_pages(struct buf *, int clear_modify); void vfs_unbusy_pages(struct buf *); int vmapbuf(struct buf *, int); void vunmapbuf(struct buf *); void relpbuf(struct buf *, int *); void brelvp(struct buf *); void bgetvp(struct vnode *, struct buf *); void pbgetbo(struct bufobj *bo, struct buf *bp); void pbgetvp(struct vnode *, struct buf *); void pbrelbo(struct buf *); void pbrelvp(struct buf *); int allocbuf(struct buf *bp, int size); void reassignbuf(struct buf *); struct buf *trypbuf(int *); void bwait(struct buf *, u_char, const char *); void bdone(struct buf *); typedef daddr_t (vbg_get_lblkno_t)(struct vnode *, vm_ooffset_t); typedef int (vbg_get_blksize_t)(struct vnode *, daddr_t); int vfs_bio_getpages(struct vnode *vp, struct vm_page **ma, int count, int *rbehind, int *rahead, vbg_get_lblkno_t get_lblkno, vbg_get_blksize_t get_blksize); #endif /* _KERNEL */ #endif /* !_SYS_BUF_H_ */ Index: head/sys/sys/vnode.h =================================================================== --- head/sys/sys/vnode.h (revision 309061) +++ head/sys/sys/vnode.h (revision 309062) @@ -1,884 +1,885 @@ /*- * Copyright (c) 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. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vnode.h 8.7 (Berkeley) 2/4/94 * $FreeBSD$ */ #ifndef _SYS_VNODE_H_ #define _SYS_VNODE_H_ #include #include #include #include #include #include #include #include #include #include /* * The vnode is the focus of all file activity in UNIX. There is a * unique vnode allocated for each active file, each current directory, * each mounted-on file, text file, and the root. */ /* * Vnode types. VNON means no type. */ enum vtype { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD, VMARKER }; /* * Each underlying filesystem allocates its own private area and hangs * it from v_data. If non-null, this area is freed in getnewvnode(). */ struct namecache; struct vpollinfo { struct mtx vpi_lock; /* lock to protect below */ struct selinfo vpi_selinfo; /* identity of poller(s) */ short vpi_events; /* what they are looking for */ short vpi_revents; /* what has happened */ }; /* * Reading or writing any of these items requires holding the appropriate lock. * * Lock reference: * c - namecache mutex * i - interlock * l - mp mnt_listmtx or freelist mutex * I - updated with atomics, 0->1 and 1->0 transitions with interlock held * m - mount point interlock * p - pollinfo lock * u - Only a reference to the vnode is needed to read. * v - vnode lock * * Vnodes may be found on many lists. The general way to deal with operating * on a vnode that is on a list is: * 1) Lock the list and find the vnode. * 2) Lock interlock so that the vnode does not go away. * 3) Unlock the list to avoid lock order reversals. * 4) vget with LK_INTERLOCK and check for ENOENT, or * 5) Check for DOOMED if the vnode lock is not required. * 6) Perform your operation, then vput(). */ #if defined(_KERNEL) || defined(_KVM_VNODE) struct vnode { /* * Fields which define the identity of the vnode. These fields are * owned by the filesystem (XXX: and vgone() ?) */ const char *v_tag; /* u type of underlying data */ struct vop_vector *v_op; /* u vnode operations vector */ void *v_data; /* u private data for fs */ /* * Filesystem instance stuff */ struct mount *v_mount; /* u ptr to vfs we are in */ TAILQ_ENTRY(vnode) v_nmntvnodes; /* m vnodes for mount point */ /* * Type specific fields, only one applies to any given vnode. * See #defines below for renaming to v_* namespace. */ union { struct mount *vu_mount; /* v ptr to mountpoint (VDIR) */ struct socket *vu_socket; /* v unix domain net (VSOCK) */ struct cdev *vu_cdev; /* v device (VCHR, VBLK) */ struct fifoinfo *vu_fifoinfo; /* v fifo (VFIFO) */ } v_un; /* * vfs_hash: (mount + inode) -> vnode hash. The hash value * itself is grouped with other int fields, to avoid padding. */ LIST_ENTRY(vnode) v_hashlist; /* * VFS_namecache stuff */ LIST_HEAD(, namecache) v_cache_src; /* c Cache entries from us */ TAILQ_HEAD(, namecache) v_cache_dst; /* c Cache entries to us */ struct namecache *v_cache_dd; /* c Cache entry for .. vnode */ /* * Locking */ struct lock v_lock; /* u (if fs don't have one) */ struct mtx v_interlock; /* lock for "i" things */ struct lock *v_vnlock; /* u pointer to vnode lock */ /* * The machinery of being a vnode */ TAILQ_ENTRY(vnode) v_actfreelist; /* l vnode active/free lists */ struct bufobj v_bufobj; /* * Buffer cache object */ /* * Hooks for various subsystems and features. */ struct vpollinfo *v_pollinfo; /* i Poll events, p for *v_pi */ struct label *v_label; /* MAC label for vnode */ struct lockf *v_lockf; /* Byte-level advisory lock list */ struct rangelock v_rl; /* Byte-range lock */ /* * clustering stuff */ daddr_t v_cstart; /* v start block of cluster */ daddr_t v_lasta; /* v last allocation */ daddr_t v_lastw; /* v last write */ int v_clen; /* v length of cur. cluster */ u_int v_holdcnt; /* I prevents recycling. */ u_int v_usecount; /* I ref count of users */ u_int v_iflag; /* i vnode flags (see below) */ u_int v_vflag; /* v vnode flags */ u_int v_mflag; /* l mnt-specific vnode flags */ int v_writecount; /* v ref count of writers */ u_int v_hash; enum vtype v_type; /* u vnode type */ }; #endif /* defined(_KERNEL) || defined(_KVM_VNODE) */ #define v_mountedhere v_un.vu_mount #define v_socket v_un.vu_socket #define v_rdev v_un.vu_cdev #define v_fifoinfo v_un.vu_fifoinfo #define bo2vnode(bo) __containerof((bo), struct vnode, v_bufobj) /* XXX: These are temporary to avoid a source sweep at this time */ #define v_object v_bufobj.bo_object /* * Userland version of struct vnode, for sysctl. */ struct xvnode { size_t xv_size; /* sizeof(struct xvnode) */ void *xv_vnode; /* address of real vnode */ u_long xv_flag; /* vnode vflags */ int xv_usecount; /* reference count of users */ int xv_writecount; /* reference count of writers */ int xv_holdcnt; /* page & buffer references */ u_long xv_id; /* capability identifier */ void *xv_mount; /* address of parent mount */ long xv_numoutput; /* num of writes in progress */ enum vtype xv_type; /* vnode type */ union { void *xvu_socket; /* socket, if VSOCK */ void *xvu_fifo; /* fifo, if VFIFO */ dev_t xvu_rdev; /* maj/min, if VBLK/VCHR */ struct { dev_t xvu_dev; /* device, if VDIR/VREG/VLNK */ ino_t xvu_ino; /* id, if VDIR/VREG/VLNK */ } xv_uns; } xv_un; }; #define xv_socket xv_un.xvu_socket #define xv_fifo xv_un.xvu_fifo #define xv_rdev xv_un.xvu_rdev #define xv_dev xv_un.xv_uns.xvu_dev #define xv_ino xv_un.xv_uns.xvu_ino /* We don't need to lock the knlist */ #define VN_KNLIST_EMPTY(vp) ((vp)->v_pollinfo == NULL || \ KNLIST_EMPTY(&(vp)->v_pollinfo->vpi_selinfo.si_note)) #define VN_KNOTE(vp, b, a) \ do { \ if (!VN_KNLIST_EMPTY(vp)) \ KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b), \ (a) | KNF_NOKQLOCK); \ } while (0) #define VN_KNOTE_LOCKED(vp, b) VN_KNOTE(vp, b, KNF_LISTLOCKED) #define VN_KNOTE_UNLOCKED(vp, b) VN_KNOTE(vp, b, 0) /* * Vnode flags. * VI flags are protected by interlock and live in v_iflag * VV flags are protected by the vnode lock and live in v_vflag * * VI_DOOMED is doubly protected by the interlock and vnode lock. Both * are required for writing but the status may be checked with either. */ #define VI_MOUNT 0x0020 /* Mount in progress */ #define VI_DOOMED 0x0080 /* This vnode is being recycled */ #define VI_FREE 0x0100 /* This vnode is on the freelist */ #define VI_ACTIVE 0x0200 /* This vnode is on the active list */ #define VI_DOINGINACT 0x0800 /* VOP_INACTIVE is in progress */ #define VI_OWEINACT 0x1000 /* Need to call inactive */ #define VV_ROOT 0x0001 /* root of its filesystem */ #define VV_ISTTY 0x0002 /* vnode represents a tty */ #define VV_NOSYNC 0x0004 /* unlinked, stop syncing */ #define VV_ETERNALDEV 0x0008 /* device that is never destroyed */ #define VV_CACHEDLABEL 0x0010 /* Vnode has valid cached MAC label */ #define VV_TEXT 0x0020 /* vnode is a pure text prototype */ #define VV_COPYONWRITE 0x0040 /* vnode is doing copy-on-write */ #define VV_SYSTEM 0x0080 /* vnode being used by kernel */ #define VV_PROCDEP 0x0100 /* vnode is process dependent */ #define VV_NOKNOTE 0x0200 /* don't activate knotes on this vnode */ #define VV_DELETED 0x0400 /* should be removed */ #define VV_MD 0x0800 /* vnode backs the md device */ #define VV_FORCEINSMQ 0x1000 /* force the insmntque to succeed */ #define VMP_TMPMNTFREELIST 0x0001 /* Vnode is on mnt's tmp free list */ /* * Vnode attributes. A field value of VNOVAL represents a field whose value * is unavailable (getattr) or which is not to be changed (setattr). */ struct vattr { enum vtype va_type; /* vnode type (for create) */ u_short va_mode; /* files access mode and type */ short va_nlink; /* number of references to file */ uid_t va_uid; /* owner user id */ gid_t va_gid; /* owner group id */ dev_t va_fsid; /* filesystem id */ long va_fileid; /* file id */ u_quad_t va_size; /* file size in bytes */ long va_blocksize; /* blocksize preferred for i/o */ struct timespec va_atime; /* time of last access */ struct timespec va_mtime; /* time of last modification */ struct timespec va_ctime; /* time file changed */ struct timespec va_birthtime; /* time file created */ u_long va_gen; /* generation number of file */ u_long va_flags; /* flags defined for file */ dev_t va_rdev; /* device the special file represents */ u_quad_t va_bytes; /* bytes of disk space held by file */ u_quad_t va_filerev; /* file modification number */ u_int va_vaflags; /* operations flags, see below */ long va_spare; /* remain quad aligned */ }; /* * Flags for va_vaflags. */ #define VA_UTIMES_NULL 0x01 /* utimes argument was NULL */ #define VA_EXCLUSIVE 0x02 /* exclusive create request */ #define VA_SYNC 0x04 /* O_SYNC truncation */ /* * Flags for ioflag. (high 16 bits used to ask for read-ahead and * help with write clustering) * NB: IO_NDELAY and IO_DIRECT are linked to fcntl.h */ #define IO_UNIT 0x0001 /* do I/O as atomic unit */ #define IO_APPEND 0x0002 /* append write to end */ #define IO_NDELAY 0x0004 /* FNDELAY flag set in file table */ #define IO_NODELOCKED 0x0008 /* underlying node already locked */ #define IO_ASYNC 0x0010 /* bawrite rather then bdwrite */ #define IO_VMIO 0x0020 /* data already in VMIO space */ #define IO_INVAL 0x0040 /* invalidate after I/O */ #define IO_SYNC 0x0080 /* do I/O synchronously */ #define IO_DIRECT 0x0100 /* attempt to bypass buffer cache */ +#define IO_NOREUSE 0x0200 /* VMIO data won't be reused */ #define IO_EXT 0x0400 /* operate on external attributes */ #define IO_NORMAL 0x0800 /* operate on regular data */ #define IO_NOMACCHECK 0x1000 /* MAC checks unnecessary */ #define IO_BUFLOCKED 0x2000 /* ffs flag; indir buf is locked */ #define IO_RANGELOCKED 0x4000 /* range locked */ #define IO_SEQMAX 0x7F /* seq heuristic max value */ #define IO_SEQSHIFT 16 /* seq heuristic in upper 16 bits */ /* * Flags for accmode_t. */ #define VEXEC 000000000100 /* execute/search permission */ #define VWRITE 000000000200 /* write permission */ #define VREAD 000000000400 /* read permission */ #define VADMIN 000000010000 /* being the file owner */ #define VAPPEND 000000040000 /* permission to write/append */ /* * VEXPLICIT_DENY makes VOP_ACCESSX(9) return EPERM or EACCES only * if permission was denied explicitly, by a "deny" rule in NFSv4 ACL, * and 0 otherwise. This never happens with ordinary unix access rights * or POSIX.1e ACLs. Obviously, VEXPLICIT_DENY must be OR-ed with * some other V* constant. */ #define VEXPLICIT_DENY 000000100000 #define VREAD_NAMED_ATTRS 000000200000 /* not used */ #define VWRITE_NAMED_ATTRS 000000400000 /* not used */ #define VDELETE_CHILD 000001000000 #define VREAD_ATTRIBUTES 000002000000 /* permission to stat(2) */ #define VWRITE_ATTRIBUTES 000004000000 /* change {m,c,a}time */ #define VDELETE 000010000000 #define VREAD_ACL 000020000000 /* read ACL and file mode */ #define VWRITE_ACL 000040000000 /* change ACL and/or file mode */ #define VWRITE_OWNER 000100000000 /* change file owner */ #define VSYNCHRONIZE 000200000000 /* not used */ #define VCREAT 000400000000 /* creating new file */ #define VVERIFY 001000000000 /* verification required */ /* * Permissions that were traditionally granted only to the file owner. */ #define VADMIN_PERMS (VADMIN | VWRITE_ATTRIBUTES | VWRITE_ACL | \ VWRITE_OWNER) /* * Permissions that were traditionally granted to everyone. */ #define VSTAT_PERMS (VREAD_ATTRIBUTES | VREAD_ACL) /* * Permissions that allow to change the state of the file in any way. */ #define VMODIFY_PERMS (VWRITE | VAPPEND | VADMIN_PERMS | VDELETE_CHILD | \ VDELETE) /* * Token indicating no attribute value yet assigned. */ #define VNOVAL (-1) /* * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon) */ #define VLKTIMEOUT (hz / 20 + 1) #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_VNODE); #endif extern u_int ncsizefactor; /* * Convert between vnode types and inode formats (since POSIX.1 * defines mode word of stat structure in terms of inode formats). */ extern enum vtype iftovt_tab[]; extern int vttoif_tab[]; #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12]) #define VTTOIF(indx) (vttoif_tab[(int)(indx)]) #define MAKEIMODE(indx, mode) (int)(VTTOIF(indx) | (mode)) /* * Flags to various vnode functions. */ #define SKIPSYSTEM 0x0001 /* vflush: skip vnodes marked VSYSTEM */ #define FORCECLOSE 0x0002 /* vflush: force file closure */ #define WRITECLOSE 0x0004 /* vflush: only close writable files */ #define EARLYFLUSH 0x0008 /* vflush: early call for ffs_flushfiles */ #define V_SAVE 0x0001 /* vinvalbuf: sync file first */ #define V_ALT 0x0002 /* vinvalbuf: invalidate only alternate bufs */ #define V_NORMAL 0x0004 /* vinvalbuf: invalidate only regular bufs */ #define V_CLEANONLY 0x0008 /* vinvalbuf: invalidate only clean bufs */ #define REVOKEALL 0x0001 /* vop_revoke: revoke all aliases */ #define V_WAIT 0x0001 /* vn_start_write: sleep for suspend */ #define V_NOWAIT 0x0002 /* vn_start_write: don't sleep for suspend */ #define V_XSLEEP 0x0004 /* vn_start_write: just return after sleep */ #define V_MNTREF 0x0010 /* vn_start_write: mp is already ref-ed */ #define VR_START_WRITE 0x0001 /* vfs_write_resume: start write atomically */ #define VR_NO_SUSPCLR 0x0002 /* vfs_write_resume: do not clear suspension */ #define VS_SKIP_UNMOUNT 0x0001 /* vfs_write_suspend: fail if the filesystem is being unmounted */ #define VREF(vp) vref(vp) #ifdef DIAGNOSTIC #define VATTR_NULL(vap) vattr_null(vap) #else #define VATTR_NULL(vap) (*(vap) = va_null) /* initialize a vattr */ #endif /* DIAGNOSTIC */ #define NULLVP ((struct vnode *)NULL) /* * Global vnode data. */ extern struct vnode *rootvnode; /* root (i.e. "/") vnode */ extern struct mount *rootdevmp; /* "/dev" mount */ extern int desiredvnodes; /* number of vnodes desired */ extern struct uma_zone *namei_zone; extern struct vattr va_null; /* predefined null vattr structure */ #define VI_LOCK(vp) mtx_lock(&(vp)->v_interlock) #define VI_LOCK_FLAGS(vp, flags) mtx_lock_flags(&(vp)->v_interlock, (flags)) #define VI_TRYLOCK(vp) mtx_trylock(&(vp)->v_interlock) #define VI_UNLOCK(vp) mtx_unlock(&(vp)->v_interlock) #define VI_MTX(vp) (&(vp)->v_interlock) #define VN_LOCK_AREC(vp) lockallowrecurse((vp)->v_vnlock) #define VN_LOCK_ASHARE(vp) lockallowshare((vp)->v_vnlock) #define VN_LOCK_DSHARE(vp) lockdisableshare((vp)->v_vnlock) #endif /* _KERNEL */ /* * Mods for extensibility. */ /* * Flags for vdesc_flags: */ #define VDESC_MAX_VPS 16 /* Low order 16 flag bits are reserved for willrele flags for vp arguments. */ #define VDESC_VP0_WILLRELE 0x0001 #define VDESC_VP1_WILLRELE 0x0002 #define VDESC_VP2_WILLRELE 0x0004 #define VDESC_VP3_WILLRELE 0x0008 #define VDESC_NOMAP_VPP 0x0100 #define VDESC_VPP_WILLRELE 0x0200 /* * A generic structure. * This can be used by bypass routines to identify generic arguments. */ struct vop_generic_args { struct vnodeop_desc *a_desc; /* other random data follows, presumably */ }; typedef int vop_bypass_t(struct vop_generic_args *); /* * VDESC_NO_OFFSET is used to identify the end of the offset list * and in places where no such field exists. */ #define VDESC_NO_OFFSET -1 /* * This structure describes the vnode operation taking place. */ struct vnodeop_desc { char *vdesc_name; /* a readable name for debugging */ int vdesc_flags; /* VDESC_* flags */ vop_bypass_t *vdesc_call; /* Function to call */ /* * These ops are used by bypass routines to map and locate arguments. * Creds and procs are not needed in bypass routines, but sometimes * they are useful to (for example) transport layers. * Nameidata is useful because it has a cred in it. */ int *vdesc_vp_offsets; /* list ended by VDESC_NO_OFFSET */ int vdesc_vpp_offset; /* return vpp location */ int vdesc_cred_offset; /* cred location, if any */ int vdesc_thread_offset; /* thread location, if any */ int vdesc_componentname_offset; /* if any */ }; #ifdef _KERNEL /* * A list of all the operation descs. */ extern struct vnodeop_desc *vnodeop_descs[]; #define VOPARG_OFFSETOF(s_type, field) __offsetof(s_type, field) #define VOPARG_OFFSETTO(s_type, s_offset, struct_p) \ ((s_type)(((char*)(struct_p)) + (s_offset))) #ifdef DEBUG_VFS_LOCKS /* * Support code to aid in debugging VFS locking problems. Not totally * reliable since if the thread sleeps between changing the lock * state and checking it with the assert, some other thread could * change the state. They are good enough for debugging a single * filesystem using a single-threaded test. Note that the unreliability is * limited to false negatives; efforts were made to ensure that false * positives cannot occur. */ void assert_vi_locked(struct vnode *vp, const char *str); void assert_vi_unlocked(struct vnode *vp, const char *str); void assert_vop_elocked(struct vnode *vp, const char *str); void assert_vop_locked(struct vnode *vp, const char *str); void assert_vop_unlocked(struct vnode *vp, const char *str); #define ASSERT_VI_LOCKED(vp, str) assert_vi_locked((vp), (str)) #define ASSERT_VI_UNLOCKED(vp, str) assert_vi_unlocked((vp), (str)) #define ASSERT_VOP_ELOCKED(vp, str) assert_vop_elocked((vp), (str)) #define ASSERT_VOP_LOCKED(vp, str) assert_vop_locked((vp), (str)) #define ASSERT_VOP_UNLOCKED(vp, str) assert_vop_unlocked((vp), (str)) #else /* !DEBUG_VFS_LOCKS */ #define ASSERT_VI_LOCKED(vp, str) ((void)0) #define ASSERT_VI_UNLOCKED(vp, str) ((void)0) #define ASSERT_VOP_ELOCKED(vp, str) ((void)0) #define ASSERT_VOP_LOCKED(vp, str) ((void)0) #define ASSERT_VOP_UNLOCKED(vp, str) ((void)0) #endif /* DEBUG_VFS_LOCKS */ /* * This call works for vnodes in the kernel. */ #define VCALL(c) ((c)->a_desc->vdesc_call(c)) #define DOINGASYNC(vp) \ (((vp)->v_mount->mnt_kern_flag & MNTK_ASYNC) != 0 && \ ((curthread->td_pflags & TDP_SYNCIO) == 0)) /* * VMIO support inline */ extern int vmiodirenable; static __inline int vn_canvmio(struct vnode *vp) { if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR))) return(TRUE); return(FALSE); } /* * Finally, include the default set of vnode operations. */ typedef void vop_getpages_iodone_t(void *, vm_page_t *, int, int); #include "vnode_if.h" /* vn_open_flags */ #define VN_OPEN_NOAUDIT 0x00000001 #define VN_OPEN_NOCAPCHECK 0x00000002 #define VN_OPEN_NAMECACHE 0x00000004 /* * Public vnode manipulation functions. */ struct componentname; struct file; struct mount; struct nameidata; struct ostat; struct thread; struct proc; struct stat; struct nstat; struct ucred; struct uio; struct vattr; struct vfsops; struct vnode; typedef int (*vn_get_ino_t)(struct mount *, void *, int, struct vnode **); int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn); /* cache_* may belong in namei.h. */ void cache_changesize(int newhashsize); #define cache_enter(dvp, vp, cnp) \ cache_enter_time(dvp, vp, cnp, NULL, NULL) void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp); int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp); void cache_purge(struct vnode *vp); void cache_purge_negative(struct vnode *vp); void cache_purgevfs(struct mount *mp, bool force); int change_dir(struct vnode *vp, struct thread *td); void cvtstat(struct stat *st, struct ostat *ost); void cvtnstat(struct stat *sb, struct nstat *nsb); int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp); void getnewvnode_reserve(u_int count); void getnewvnode_drop_reserve(void); int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg); int insmntque(struct vnode *vp, struct mount *mp); u_quad_t init_va_filerev(void); int speedup_syncer(void); int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen); int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); struct vnode * vn_dir_dd_ino(struct vnode *vp); int vn_commname(struct vnode *vn, char *buf, u_int buflen); int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen); int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused); int vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *aclp, accmode_t accmode, struct ucred *cred, int *privused); int vaccess_acl_posix1e(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *acl, accmode_t accmode, struct ucred *cred, int *privused); void vattr_null(struct vattr *vap); int vcount(struct vnode *vp); #define vdrop(vp) _vdrop((vp), 0) #define vdropl(vp) _vdrop((vp), 1) void _vdrop(struct vnode *, bool); int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td); int vget(struct vnode *vp, int lockflag, struct thread *td); void vgone(struct vnode *vp); #define vhold(vp) _vhold((vp), 0) #define vholdl(vp) _vhold((vp), 1) void _vhold(struct vnode *, bool); void vinactive(struct vnode *, struct thread *); int vinvalbuf(struct vnode *vp, int save, int slpflag, int slptimeo); int vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize); void vunref(struct vnode *); void vn_printf(struct vnode *vp, const char *fmt, ...) __printflike(2,3); int vrecycle(struct vnode *vp); int vrecyclel(struct vnode *vp); int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred); int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td); void vn_finished_write(struct mount *mp); void vn_finished_secondary_write(struct mount *mp); int vn_isdisk(struct vnode *vp, int *errp); int _vn_lock(struct vnode *vp, int flags, char *file, int line); #define vn_lock(vp, flags) _vn_lock(vp, flags, __FILE__, __LINE__) int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp); int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp); int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp); void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end); int vn_pollrecord(struct vnode *vp, struct thread *p, int events); int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td); int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td); int vn_rlimit_fsize(const struct vnode *vn, const struct uio *uio, struct thread *td); int vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, struct ucred *file_cred, struct thread *td); int vn_start_write(struct vnode *vp, struct mount **mpp, int flags); int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags); int vn_writechk(struct vnode *vp); int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td); int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td); int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td); int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp); int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp); int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td); int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio); int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio); #define vn_rangelock_unlock(vp, cookie) \ rangelock_unlock(&(vp)->v_rl, (cookie), VI_MTX(vp)) #define vn_rangelock_unlock_range(vp, cookie, start, end) \ rangelock_unlock_range(&(vp)->v_rl, (cookie), (start), (end), \ VI_MTX(vp)) #define vn_rangelock_rlock(vp, start, end) \ rangelock_rlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_wlock(vp, start, end) \ rangelock_wlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) int vfs_cache_lookup(struct vop_lookup_args *ap); void vfs_timestamp(struct timespec *); void vfs_write_resume(struct mount *mp, int flags); int vfs_write_suspend(struct mount *mp, int flags); int vfs_write_suspend_umnt(struct mount *mp); void vnlru_free(int, struct vfsops *); int vop_stdbmap(struct vop_bmap_args *); int vop_stdfdatasync_buf(struct vop_fdatasync_args *); int vop_stdfsync(struct vop_fsync_args *); int vop_stdgetwritemount(struct vop_getwritemount_args *); int vop_stdgetpages(struct vop_getpages_args *); int vop_stdinactive(struct vop_inactive_args *); int vop_stdislocked(struct vop_islocked_args *); int vop_stdkqfilter(struct vop_kqfilter_args *); int vop_stdlock(struct vop_lock1_args *); int vop_stdputpages(struct vop_putpages_args *); int vop_stdunlock(struct vop_unlock_args *); int vop_nopoll(struct vop_poll_args *); int vop_stdaccess(struct vop_access_args *ap); int vop_stdaccessx(struct vop_accessx_args *ap); int vop_stdadvise(struct vop_advise_args *ap); int vop_stdadvlock(struct vop_advlock_args *ap); int vop_stdadvlockasync(struct vop_advlockasync_args *ap); int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap); int vop_stdallocate(struct vop_allocate_args *ap); int vop_stdpathconf(struct vop_pathconf_args *); int vop_stdpoll(struct vop_poll_args *); int vop_stdvptocnp(struct vop_vptocnp_args *ap); int vop_stdvptofh(struct vop_vptofh_args *ap); int vop_stdunp_bind(struct vop_unp_bind_args *ap); int vop_stdunp_connect(struct vop_unp_connect_args *ap); int vop_stdunp_detach(struct vop_unp_detach_args *ap); int vop_eopnotsupp(struct vop_generic_args *ap); int vop_ebadf(struct vop_generic_args *ap); int vop_einval(struct vop_generic_args *ap); int vop_enoent(struct vop_generic_args *ap); int vop_enotty(struct vop_generic_args *ap); int vop_null(struct vop_generic_args *ap); int vop_panic(struct vop_generic_args *ap); int dead_poll(struct vop_poll_args *ap); int dead_read(struct vop_read_args *ap); int dead_write(struct vop_write_args *ap); /* These are called from within the actual VOPS. */ void vop_close_post(void *a, int rc); void vop_create_post(void *a, int rc); void vop_deleteextattr_post(void *a, int rc); void vop_link_post(void *a, int rc); void vop_lookup_post(void *a, int rc); void vop_lookup_pre(void *a); void vop_mkdir_post(void *a, int rc); void vop_mknod_post(void *a, int rc); void vop_open_post(void *a, int rc); void vop_read_post(void *a, int rc); void vop_readdir_post(void *a, int rc); void vop_reclaim_post(void *a, int rc); void vop_remove_post(void *a, int rc); void vop_rename_post(void *a, int rc); void vop_rename_pre(void *a); void vop_rmdir_post(void *a, int rc); void vop_setattr_post(void *a, int rc); void vop_setextattr_post(void *a, int rc); void vop_symlink_post(void *a, int rc); #ifdef DEBUG_VFS_LOCKS void vop_strategy_pre(void *a); void vop_lock_pre(void *a); void vop_lock_post(void *a, int rc); void vop_unlock_post(void *a, int rc); void vop_unlock_pre(void *a); #else #define vop_strategy_pre(x) do { } while (0) #define vop_lock_pre(x) do { } while (0) #define vop_lock_post(x, y) do { } while (0) #define vop_unlock_post(x, y) do { } while (0) #define vop_unlock_pre(x) do { } while (0) #endif void vop_rename_fail(struct vop_rename_args *ap); #define VOP_WRITE_PRE(ap) \ struct vattr va; \ int error; \ off_t osize, ooffset, noffset; \ \ osize = ooffset = noffset = 0; \ if (!VN_KNLIST_EMPTY((ap)->a_vp)) { \ error = VOP_GETATTR((ap)->a_vp, &va, (ap)->a_cred); \ if (error) \ return (error); \ ooffset = (ap)->a_uio->uio_offset; \ osize = (off_t)va.va_size; \ } #define VOP_WRITE_POST(ap, ret) \ noffset = (ap)->a_uio->uio_offset; \ if (noffset > ooffset && !VN_KNLIST_EMPTY((ap)->a_vp)) { \ VFS_KNOTE_LOCKED((ap)->a_vp, NOTE_WRITE \ | (noffset > osize ? NOTE_EXTEND : 0)); \ } #define VOP_LOCK(vp, flags) VOP_LOCK1(vp, flags, __FILE__, __LINE__) void vput(struct vnode *vp); void vrele(struct vnode *vp); void vref(struct vnode *vp); void vrefl(struct vnode *vp); int vrefcnt(struct vnode *vp); void v_addpollinfo(struct vnode *vp); int vnode_create_vobject(struct vnode *vp, off_t size, struct thread *td); void vnode_destroy_vobject(struct vnode *vp); extern struct vop_vector fifo_specops; extern struct vop_vector dead_vnodeops; extern struct vop_vector default_vnodeops; #define VOP_PANIC ((void*)(uintptr_t)vop_panic) #define VOP_NULL ((void*)(uintptr_t)vop_null) #define VOP_EBADF ((void*)(uintptr_t)vop_ebadf) #define VOP_ENOTTY ((void*)(uintptr_t)vop_enotty) #define VOP_EINVAL ((void*)(uintptr_t)vop_einval) #define VOP_ENOENT ((void*)(uintptr_t)vop_enoent) #define VOP_EOPNOTSUPP ((void*)(uintptr_t)vop_eopnotsupp) /* fifo_vnops.c */ int fifo_printinfo(struct vnode *); /* vfs_hash.c */ typedef int vfs_hash_cmp_t(struct vnode *vp, void *arg); void vfs_hash_changesize(int newhashsize); int vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); u_int vfs_hash_index(struct vnode *vp); int vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_ref(const struct mount *mp, u_int hash, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_rehash(struct vnode *vp, u_int hash); void vfs_hash_remove(struct vnode *vp); int vfs_kqfilter(struct vop_kqfilter_args *); void vfs_mark_atime(struct vnode *vp, struct ucred *cred); struct dirent; int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off); int vfs_unixify_accmode(accmode_t *accmode); void vfs_unp_reclaim(struct vnode *vp); int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode); int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid); int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); #endif /* _KERNEL */ #endif /* !_SYS_VNODE_H_ */ Index: head/sys/ufs/ffs/ffs_vnops.c =================================================================== --- head/sys/ufs/ffs/ffs_vnops.c (revision 309061) +++ head/sys/ufs/ffs/ffs_vnops.c (revision 309062) @@ -1,1822 +1,1745 @@ /*- * Copyright (c) 2002, 2003 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Marshall * Kirk McKusick and Network Associates Laboratories, the Security * Research Division of Network Associates, Inc. under DARPA/SPAWAR * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS * research program * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * 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. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95 * from: $FreeBSD: .../ufs/ufs_readwrite.c,v 1.96 2002/08/12 09:22:11 phk ... * @(#)ffs_vnops.c 8.15 (Berkeley) 5/14/95 */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "opt_directio.h" #include "opt_ffs.h" #ifdef DIRECTIO extern int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone); #endif static vop_fdatasync_t ffs_fdatasync; static vop_fsync_t ffs_fsync; static vop_getpages_t ffs_getpages; static vop_lock1_t ffs_lock; static vop_read_t ffs_read; static vop_write_t ffs_write; static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag); static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *cred); static vop_strategy_t ffsext_strategy; static vop_closeextattr_t ffs_closeextattr; static vop_deleteextattr_t ffs_deleteextattr; static vop_getextattr_t ffs_getextattr; static vop_listextattr_t ffs_listextattr; static vop_openextattr_t ffs_openextattr; static vop_setextattr_t ffs_setextattr; static vop_vptofh_t ffs_vptofh; /* Global vfs data structures for ufs. */ struct vop_vector ffs_vnodeops1 = { .vop_default = &ufs_vnodeops, .vop_fsync = ffs_fsync, .vop_fdatasync = ffs_fdatasync, .vop_getpages = ffs_getpages, .vop_getpages_async = vnode_pager_local_getpages_async, .vop_lock1 = ffs_lock, .vop_read = ffs_read, .vop_reallocblks = ffs_reallocblks, .vop_write = ffs_write, .vop_vptofh = ffs_vptofh, }; struct vop_vector ffs_fifoops1 = { .vop_default = &ufs_fifoops, .vop_fsync = ffs_fsync, .vop_fdatasync = ffs_fdatasync, .vop_reallocblks = ffs_reallocblks, /* XXX: really ??? */ .vop_vptofh = ffs_vptofh, }; /* Global vfs data structures for ufs. */ struct vop_vector ffs_vnodeops2 = { .vop_default = &ufs_vnodeops, .vop_fsync = ffs_fsync, .vop_fdatasync = ffs_fdatasync, .vop_getpages = ffs_getpages, .vop_getpages_async = vnode_pager_local_getpages_async, .vop_lock1 = ffs_lock, .vop_read = ffs_read, .vop_reallocblks = ffs_reallocblks, .vop_write = ffs_write, .vop_closeextattr = ffs_closeextattr, .vop_deleteextattr = ffs_deleteextattr, .vop_getextattr = ffs_getextattr, .vop_listextattr = ffs_listextattr, .vop_openextattr = ffs_openextattr, .vop_setextattr = ffs_setextattr, .vop_vptofh = ffs_vptofh, }; struct vop_vector ffs_fifoops2 = { .vop_default = &ufs_fifoops, .vop_fsync = ffs_fsync, .vop_fdatasync = ffs_fdatasync, .vop_lock1 = ffs_lock, .vop_reallocblks = ffs_reallocblks, .vop_strategy = ffsext_strategy, .vop_closeextattr = ffs_closeextattr, .vop_deleteextattr = ffs_deleteextattr, .vop_getextattr = ffs_getextattr, .vop_listextattr = ffs_listextattr, .vop_openextattr = ffs_openextattr, .vop_setextattr = ffs_setextattr, .vop_vptofh = ffs_vptofh, }; /* * Synch an open file. */ /* ARGSUSED */ static int ffs_fsync(struct vop_fsync_args *ap) { struct vnode *vp; struct bufobj *bo; int error; vp = ap->a_vp; bo = &vp->v_bufobj; retry: error = ffs_syncvnode(vp, ap->a_waitfor, 0); if (error) return (error); if (ap->a_waitfor == MNT_WAIT && DOINGSOFTDEP(vp)) { error = softdep_fsync(vp); if (error) return (error); /* * The softdep_fsync() function may drop vp lock, * allowing for dirty buffers to reappear on the * bo_dirty list. Recheck and resync as needed. */ BO_LOCK(bo); if ((vp->v_type == VREG || vp->v_type == VDIR) && (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) { BO_UNLOCK(bo); goto retry; } BO_UNLOCK(bo); } return (0); } int ffs_syncvnode(struct vnode *vp, int waitfor, int flags) { struct inode *ip; struct bufobj *bo; struct buf *bp, *nbp; ufs_lbn_t lbn; int error, passes; bool still_dirty, wait; ip = VTOI(vp); ip->i_flag &= ~IN_NEEDSYNC; bo = &vp->v_bufobj; /* * When doing MNT_WAIT we must first flush all dependencies * on the inode. */ if (DOINGSOFTDEP(vp) && waitfor == MNT_WAIT && (error = softdep_sync_metadata(vp)) != 0) return (error); /* * Flush all dirty buffers associated with a vnode. */ error = 0; passes = 0; wait = false; /* Always do an async pass first. */ lbn = lblkno(ITOFS(ip), (ip->i_size + ITOFS(ip)->fs_bsize - 1)); BO_LOCK(bo); loop: TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) bp->b_vflags &= ~BV_SCANNED; TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { /* * Reasons to skip this buffer: it has already been considered * on this pass, the buffer has dependencies that will cause * it to be redirtied and it has not already been deferred, * or it is already being written. */ if ((bp->b_vflags & BV_SCANNED) != 0) continue; bp->b_vflags |= BV_SCANNED; /* * Flush indirects in order, if requested. * * Note that if only datasync is requested, we can * skip indirect blocks when softupdates are not * active. Otherwise we must flush them with data, * since dependencies prevent data block writes. */ if (waitfor == MNT_WAIT && bp->b_lblkno <= -NDADDR && (lbn_level(bp->b_lblkno) >= passes || ((flags & DATA_ONLY) != 0 && !DOINGSOFTDEP(vp)))) continue; if (bp->b_lblkno > lbn) panic("ffs_syncvnode: syncing truncated data."); if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL) == 0) { BO_UNLOCK(bo); } else if (wait) { if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) != 0) { bp->b_vflags &= ~BV_SCANNED; goto next; } } else continue; if ((bp->b_flags & B_DELWRI) == 0) panic("ffs_fsync: not dirty"); /* * Check for dependencies and potentially complete them. */ if (!LIST_EMPTY(&bp->b_dep) && (error = softdep_sync_buf(vp, bp, wait ? MNT_WAIT : MNT_NOWAIT)) != 0) { /* I/O error. */ if (error != EBUSY) { BUF_UNLOCK(bp); return (error); } /* If we deferred once, don't defer again. */ if ((bp->b_flags & B_DEFERRED) == 0) { bp->b_flags |= B_DEFERRED; BUF_UNLOCK(bp); goto next; } } if (wait) { bremfree(bp); if ((error = bwrite(bp)) != 0) return (error); } else if ((bp->b_flags & B_CLUSTEROK)) { (void) vfs_bio_awrite(bp); } else { bremfree(bp); (void) bawrite(bp); } next: /* * Since we may have slept during the I/O, we need * to start from a known point. */ BO_LOCK(bo); nbp = TAILQ_FIRST(&bo->bo_dirty.bv_hd); } if (waitfor != MNT_WAIT) { BO_UNLOCK(bo); if ((flags & NO_INO_UPDT) != 0) return (0); else return (ffs_update(vp, 0)); } /* Drain IO to see if we're done. */ bufobj_wwait(bo, 0, 0); /* * Block devices associated with filesystems may have new I/O * requests posted for them even if the vnode is locked, so no * amount of trying will get them clean. We make several passes * as a best effort. * * Regular files may need multiple passes to flush all dependency * work as it is possible that we must write once per indirect * level, once for the leaf, and once for the inode and each of * these will be done with one sync and one async pass. */ if (bo->bo_dirty.bv_cnt > 0) { if ((flags & DATA_ONLY) == 0) { still_dirty = true; } else { /* * For data-only sync, dirty indirect buffers * are ignored. */ still_dirty = false; TAILQ_FOREACH(bp, &bo->bo_dirty.bv_hd, b_bobufs) { if (bp->b_lblkno > -NDADDR) { still_dirty = true; break; } } } if (still_dirty) { /* Write the inode after sync passes to flush deps. */ if (wait && DOINGSOFTDEP(vp) && (flags & NO_INO_UPDT) == 0) { BO_UNLOCK(bo); ffs_update(vp, 1); BO_LOCK(bo); } /* switch between sync/async. */ wait = !wait; if (wait || ++passes < NIADDR + 2) goto loop; #ifdef INVARIANTS if (!vn_isdisk(vp, NULL)) vn_printf(vp, "ffs_fsync: dirty "); #endif } } BO_UNLOCK(bo); error = 0; if ((flags & DATA_ONLY) == 0) { if ((flags & NO_INO_UPDT) == 0) error = ffs_update(vp, 1); if (DOINGSUJ(vp)) softdep_journal_fsync(VTOI(vp)); } return (error); } static int ffs_fdatasync(struct vop_fdatasync_args *ap) { return (ffs_syncvnode(ap->a_vp, MNT_WAIT, DATA_ONLY)); } static int ffs_lock(ap) struct vop_lock1_args /* { struct vnode *a_vp; int a_flags; struct thread *a_td; char *file; int line; } */ *ap; { #ifndef NO_FFS_SNAPSHOT struct vnode *vp; int flags; struct lock *lkp; int result; switch (ap->a_flags & LK_TYPE_MASK) { case LK_SHARED: case LK_UPGRADE: case LK_EXCLUSIVE: vp = ap->a_vp; flags = ap->a_flags; for (;;) { #ifdef DEBUG_VFS_LOCKS KASSERT(vp->v_holdcnt != 0, ("ffs_lock %p: zero hold count", vp)); #endif lkp = vp->v_vnlock; result = _lockmgr_args(lkp, flags, VI_MTX(vp), LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, ap->a_line); if (lkp == vp->v_vnlock || result != 0) break; /* * Apparent success, except that the vnode * mutated between snapshot file vnode and * regular file vnode while this process * slept. The lock currently held is not the * right lock. Release it, and try to get the * new lock. */ (void) _lockmgr_args(lkp, LK_RELEASE, NULL, LK_WMESG_DEFAULT, LK_PRIO_DEFAULT, LK_TIMO_DEFAULT, ap->a_file, ap->a_line); if ((flags & (LK_INTERLOCK | LK_NOWAIT)) == (LK_INTERLOCK | LK_NOWAIT)) return (EBUSY); if ((flags & LK_TYPE_MASK) == LK_UPGRADE) flags = (flags & ~LK_TYPE_MASK) | LK_EXCLUSIVE; flags &= ~LK_INTERLOCK; } break; default: result = VOP_LOCK1_APV(&ufs_vnodeops, ap); } return (result); #else return (VOP_LOCK1_APV(&ufs_vnodeops, ap)); #endif } /* * Vnode op for reading. */ static int ffs_read(ap) struct vop_read_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { struct vnode *vp; struct inode *ip; struct uio *uio; struct fs *fs; struct buf *bp; ufs_lbn_t lbn, nextlbn; off_t bytesinfile; long size, xfersize, blkoffset; ssize_t orig_resid; int error; int seqcount; int ioflag; vp = ap->a_vp; uio = ap->a_uio; ioflag = ap->a_ioflag; if (ap->a_ioflag & IO_EXT) #ifdef notyet return (ffs_extread(vp, uio, ioflag)); #else panic("ffs_read+IO_EXT"); #endif #ifdef DIRECTIO if ((ioflag & IO_DIRECT) != 0) { int workdone; error = ffs_rawread(vp, uio, &workdone); if (error != 0 || workdone != 0) return error; } #endif seqcount = ap->a_ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_READ) panic("ffs_read: mode"); if (vp->v_type == VLNK) { if ((int)ip->i_size < vp->v_mount->mnt_maxsymlinklen) panic("ffs_read: short symlink"); } else if (vp->v_type != VREG && vp->v_type != VDIR) panic("ffs_read: type %d", vp->v_type); #endif orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("ffs_read: uio->uio_resid < 0")); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("ffs_read: uio->uio_offset < 0")); fs = ITOFS(ip); if (uio->uio_offset < ip->i_size && uio->uio_offset >= fs->fs_maxfilesize) return (EOVERFLOW); for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { if ((bytesinfile = ip->i_size - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); nextlbn = lbn + 1; /* * size of buffer. The buffer representing the * end of the file is rounded up to the size of * the block type ( fragment or full block, * depending ). */ size = blksize(fs, ip, lbn); blkoffset = blkoff(fs, uio->uio_offset); /* * The amount we want to transfer in this iteration is * one FS block less the amount of the data before * our startpoint (duh!) */ xfersize = fs->fs_bsize - blkoffset; /* * But if we actually want less than the block, * or the file doesn't have a whole block more of data, * then use the lesser number. */ if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (bytesinfile < xfersize) xfersize = bytesinfile; if (lblktosize(fs, nextlbn) >= ip->i_size) { /* * Don't do readahead if this is the end of the file. */ error = bread_gb(vp, lbn, size, NOCRED, GB_UNMAPPED, &bp); } else if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERR) == 0) { /* * Otherwise if we are allowed to cluster, * grab as much as we can. * * XXX This may not be a win if we are not * doing sequential access. */ error = cluster_read(vp, ip->i_size, lbn, size, NOCRED, blkoffset + uio->uio_resid, seqcount, GB_UNMAPPED, &bp); } else if (seqcount > 1) { /* * If we are NOT allowed to cluster, then * if we appear to be acting sequentially, * fire off a request for a readahead * as well as a read. Note that the 4th and 5th * arguments point to arrays of the size specified in * the 6th argument. */ u_int nextsize = blksize(fs, ip, nextlbn); error = breadn_flags(vp, lbn, size, &nextlbn, &nextsize, 1, NOCRED, GB_UNMAPPED, &bp); } else { /* * Failing all of the above, just read what the * user asked for. Interestingly, the same as * the first option above. */ error = bread_gb(vp, lbn, size, NOCRED, GB_UNMAPPED, &bp); } if (error) { brelse(bp); bp = NULL; break; } /* - * If IO_DIRECT then set B_DIRECT for the buffer. This - * will cause us to attempt to release the buffer later on - * and will cause the buffer cache to attempt to free the - * underlying pages. - */ - if (ioflag & IO_DIRECT) - bp->b_flags |= B_DIRECT; - - /* * We should only get non-zero b_resid when an I/O error * has occurred, which should cause us to break above. * However, if the short read did not cause an error, * then we want to ensure that we do not uiomove bad * or uninitialized data. */ size -= bp->b_resid; if (size < xfersize) { if (size == 0) break; xfersize = size; } if (buf_mapped(bp)) { error = vn_io_fault_uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); } else { error = vn_io_fault_pgmove(bp->b_pages, blkoffset, (int)xfersize, uio); } if (error) break; - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - /* - * If there are no dependencies, and it's VMIO, - * then we don't need the buf, mark it available - * for freeing. For non-direct VMIO reads, the VM - * has the data. - */ - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - /* - * Otherwise let whoever - * made the request take care of - * freeing it. We just queue - * it onto another list. - */ - bqrelse(bp); - } + vfs_bio_brelse(bp, ioflag); } /* * This can only happen in the case of an error * because the loop above resets bp to NULL on each iteration * and on normal completion has not set a new value into it. * so it must have come from a 'break' statement */ - if (bp != NULL) { - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - bqrelse(bp); - } - } + if (bp != NULL) + vfs_bio_brelse(bp, ioflag); if ((error == 0 || uio->uio_resid != orig_resid) && (vp->v_mount->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0 && (ip->i_flag & IN_ACCESS) == 0) { VI_LOCK(vp); ip->i_flag |= IN_ACCESS; VI_UNLOCK(vp); } return (error); } /* * Vnode op for writing. */ static int ffs_write(ap) struct vop_write_args /* { struct vnode *a_vp; struct uio *a_uio; int a_ioflag; struct ucred *a_cred; } */ *ap; { struct vnode *vp; struct uio *uio; struct inode *ip; struct fs *fs; struct buf *bp; ufs_lbn_t lbn; off_t osize; ssize_t resid; int seqcount; int blkoffset, error, flags, ioflag, size, xfersize; vp = ap->a_vp; uio = ap->a_uio; ioflag = ap->a_ioflag; if (ap->a_ioflag & IO_EXT) #ifdef notyet return (ffs_extwrite(vp, uio, ioflag, ap->a_cred)); #else panic("ffs_write+IO_EXT"); #endif seqcount = ap->a_ioflag >> IO_SEQSHIFT; ip = VTOI(vp); #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE) panic("ffs_write: mode"); #endif switch (vp->v_type) { case VREG: if (ioflag & IO_APPEND) uio->uio_offset = ip->i_size; if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size) return (EPERM); /* FALLTHROUGH */ case VLNK: break; case VDIR: panic("ffs_write: dir write"); break; default: panic("ffs_write: type %p %d (%d,%d)", vp, (int)vp->v_type, (int)uio->uio_offset, (int)uio->uio_resid ); } KASSERT(uio->uio_resid >= 0, ("ffs_write: uio->uio_resid < 0")); KASSERT(uio->uio_offset >= 0, ("ffs_write: uio->uio_offset < 0")); fs = ITOFS(ip); if ((uoff_t)uio->uio_offset + uio->uio_resid > fs->fs_maxfilesize) return (EFBIG); /* * Maybe this should be above the vnode op call, but so long as * file servers have no limits, I don't think it matters. */ if (vn_rlimit_fsize(vp, uio, uio->uio_td)) return (EFBIG); resid = uio->uio_resid; osize = ip->i_size; if (seqcount > BA_SEQMAX) flags = BA_SEQMAX << BA_SEQSHIFT; else flags = seqcount << BA_SEQSHIFT; if (ioflag & IO_SYNC) flags |= IO_SYNC; flags |= BA_UNMAPPED; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->fs_bsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (uio->uio_offset + xfersize > ip->i_size) vnode_pager_setsize(vp, uio->uio_offset + xfersize); /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; /* XXX is uio->uio_offset the right thing here? */ error = UFS_BALLOC(vp, uio->uio_offset, xfersize, ap->a_cred, flags, &bp); if (error != 0) { vnode_pager_setsize(vp, ip->i_size); break; } - if (ioflag & IO_DIRECT) - bp->b_flags |= B_DIRECT; if ((ioflag & (IO_SYNC|IO_INVAL)) == (IO_SYNC|IO_INVAL)) bp->b_flags |= B_NOCACHE; if (uio->uio_offset + xfersize > ip->i_size) { ip->i_size = uio->uio_offset + xfersize; DIP_SET(ip, i_size, ip->i_size); } size = blksize(fs, ip, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; if (buf_mapped(bp)) { error = vn_io_fault_uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); } else { error = vn_io_fault_pgmove(bp->b_pages, blkoffset, (int)xfersize, uio); } /* * If the buffer is not already filled and we encounter an * error while trying to fill it, we have to clear out any * garbage data from the pages instantiated for the buffer. * If we do not, a failed uiomove() during a write can leave * the prior contents of the pages exposed to a userland mmap. * * Note that we need only clear buffers with a transfer size * equal to the block size because buffers with a shorter * transfer size were cleared above by the call to UFS_BALLOC() * with the BA_CLRBUF flag set. * * If the source region for uiomove identically mmaps the * buffer, uiomove() performed the NOP copy, and the buffer * content remains valid because the page fault handler * validated the pages. */ if (error != 0 && (bp->b_flags & B_CACHE) == 0 && fs->fs_bsize == xfersize) vfs_bio_clrbuf(bp); - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - bp->b_flags |= B_RELBUF; - } + vfs_bio_set_flags(bp, ioflag); + /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || (ioflag & IO_ASYNC)) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else if (xfersize + blkoffset == fs->fs_bsize) { if ((vp->v_mount->mnt_flag & MNT_NOCLUSTERW) == 0) { bp->b_flags |= B_CLUSTEROK; cluster_write(vp, bp, ip->i_size, seqcount, GB_UNMAPPED); } else { bawrite(bp); } } else if (ioflag & IO_DIRECT) { bp->b_flags |= B_CLUSTEROK; bawrite(bp); } else { bp->b_flags |= B_CLUSTEROK; bdwrite(bp); } if (error || xfersize == 0) break; ip->i_flag |= IN_CHANGE | IN_UPDATE; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ap->a_cred) { if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) { ip->i_mode &= ~(ISUID | ISGID); DIP_SET(ip, i_mode, ip->i_mode); } } if (error) { if (ioflag & IO_UNIT) { (void)ffs_truncate(vp, osize, IO_NORMAL | (ioflag & IO_SYNC), ap->a_cred); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) error = ffs_update(vp, 1); return (error); } /* * Extended attribute area reading. */ static int ffs_extread(struct vnode *vp, struct uio *uio, int ioflag) { struct inode *ip; struct ufs2_dinode *dp; struct fs *fs; struct buf *bp; ufs_lbn_t lbn, nextlbn; off_t bytesinfile; long size, xfersize, blkoffset; ssize_t orig_resid; int error; ip = VTOI(vp); fs = ITOFS(ip); dp = ip->i_din2; #ifdef INVARIANTS if (uio->uio_rw != UIO_READ || fs->fs_magic != FS_UFS2_MAGIC) panic("ffs_extread: mode"); #endif orig_resid = uio->uio_resid; KASSERT(orig_resid >= 0, ("ffs_extread: uio->uio_resid < 0")); if (orig_resid == 0) return (0); KASSERT(uio->uio_offset >= 0, ("ffs_extread: uio->uio_offset < 0")); for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) { if ((bytesinfile = dp->di_extsize - uio->uio_offset) <= 0) break; lbn = lblkno(fs, uio->uio_offset); nextlbn = lbn + 1; /* * size of buffer. The buffer representing the * end of the file is rounded up to the size of * the block type ( fragment or full block, * depending ). */ size = sblksize(fs, dp->di_extsize, lbn); blkoffset = blkoff(fs, uio->uio_offset); /* * The amount we want to transfer in this iteration is * one FS block less the amount of the data before * our startpoint (duh!) */ xfersize = fs->fs_bsize - blkoffset; /* * But if we actually want less than the block, * or the file doesn't have a whole block more of data, * then use the lesser number. */ if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; if (bytesinfile < xfersize) xfersize = bytesinfile; if (lblktosize(fs, nextlbn) >= dp->di_extsize) { /* * Don't do readahead if this is the end of the info. */ error = bread(vp, -1 - lbn, size, NOCRED, &bp); } else { /* * If we have a second block, then * fire off a request for a readahead * as well as a read. Note that the 4th and 5th * arguments point to arrays of the size specified in * the 6th argument. */ u_int nextsize = sblksize(fs, dp->di_extsize, nextlbn); nextlbn = -1 - nextlbn; error = breadn(vp, -1 - lbn, size, &nextlbn, &nextsize, 1, NOCRED, &bp); } if (error) { brelse(bp); bp = NULL; break; } /* - * If IO_DIRECT then set B_DIRECT for the buffer. This - * will cause us to attempt to release the buffer later on - * and will cause the buffer cache to attempt to free the - * underlying pages. - */ - if (ioflag & IO_DIRECT) - bp->b_flags |= B_DIRECT; - - /* * We should only get non-zero b_resid when an I/O error * has occurred, which should cause us to break above. * However, if the short read did not cause an error, * then we want to ensure that we do not uiomove bad * or uninitialized data. */ size -= bp->b_resid; if (size < xfersize) { if (size == 0) break; xfersize = size; } error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); if (error) break; - - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - /* - * If there are no dependencies, and it's VMIO, - * then we don't need the buf, mark it available - * for freeing. For non-direct VMIO reads, the VM - * has the data. - */ - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - /* - * Otherwise let whoever - * made the request take care of - * freeing it. We just queue - * it onto another list. - */ - bqrelse(bp); - } + vfs_bio_brelse(bp, ioflag); } /* * This can only happen in the case of an error * because the loop above resets bp to NULL on each iteration * and on normal completion has not set a new value into it. * so it must have come from a 'break' statement */ - if (bp != NULL) { - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - bp->b_flags |= B_RELBUF; - brelse(bp); - } else { - bqrelse(bp); - } - } + if (bp != NULL) + vfs_bio_brelse(bp, ioflag); return (error); } /* * Extended attribute area writing. */ static int ffs_extwrite(struct vnode *vp, struct uio *uio, int ioflag, struct ucred *ucred) { struct inode *ip; struct ufs2_dinode *dp; struct fs *fs; struct buf *bp; ufs_lbn_t lbn; off_t osize; ssize_t resid; int blkoffset, error, flags, size, xfersize; ip = VTOI(vp); fs = ITOFS(ip); dp = ip->i_din2; #ifdef INVARIANTS if (uio->uio_rw != UIO_WRITE || fs->fs_magic != FS_UFS2_MAGIC) panic("ffs_extwrite: mode"); #endif if (ioflag & IO_APPEND) uio->uio_offset = dp->di_extsize; KASSERT(uio->uio_offset >= 0, ("ffs_extwrite: uio->uio_offset < 0")); KASSERT(uio->uio_resid >= 0, ("ffs_extwrite: uio->uio_resid < 0")); if ((uoff_t)uio->uio_offset + uio->uio_resid > NXADDR * fs->fs_bsize) return (EFBIG); resid = uio->uio_resid; osize = dp->di_extsize; flags = IO_EXT; if (ioflag & IO_SYNC) flags |= IO_SYNC; for (error = 0; uio->uio_resid > 0;) { lbn = lblkno(fs, uio->uio_offset); blkoffset = blkoff(fs, uio->uio_offset); xfersize = fs->fs_bsize - blkoffset; if (uio->uio_resid < xfersize) xfersize = uio->uio_resid; /* * We must perform a read-before-write if the transfer size * does not cover the entire buffer. */ if (fs->fs_bsize > xfersize) flags |= BA_CLRBUF; else flags &= ~BA_CLRBUF; error = UFS_BALLOC(vp, uio->uio_offset, xfersize, ucred, flags, &bp); if (error != 0) break; /* * If the buffer is not valid we have to clear out any * garbage data from the pages instantiated for the buffer. * If we do not, a failed uiomove() during a write can leave * the prior contents of the pages exposed to a userland * mmap(). XXX deal with uiomove() errors a better way. */ if ((bp->b_flags & B_CACHE) == 0 && fs->fs_bsize <= xfersize) vfs_bio_clrbuf(bp); - if (ioflag & IO_DIRECT) - bp->b_flags |= B_DIRECT; if (uio->uio_offset + xfersize > dp->di_extsize) dp->di_extsize = uio->uio_offset + xfersize; size = sblksize(fs, dp->di_extsize, lbn) - bp->b_resid; if (size < xfersize) xfersize = size; error = uiomove((char *)bp->b_data + blkoffset, (int)xfersize, uio); - if ((ioflag & (IO_VMIO|IO_DIRECT)) && - (LIST_EMPTY(&bp->b_dep))) { - bp->b_flags |= B_RELBUF; - } + + vfs_bio_set_flags(bp, ioflag); /* * If IO_SYNC each buffer is written synchronously. Otherwise * if we have a severe page deficiency write the buffer * asynchronously. Otherwise try to cluster, and if that * doesn't do it then either do an async write (if O_DIRECT), * or a delayed write (if not). */ if (ioflag & IO_SYNC) { (void)bwrite(bp); } else if (vm_page_count_severe() || buf_dirty_count_severe() || xfersize + blkoffset == fs->fs_bsize || (ioflag & (IO_ASYNC | IO_DIRECT))) bawrite(bp); else bdwrite(bp); if (error || xfersize == 0) break; ip->i_flag |= IN_CHANGE; } /* * If we successfully wrote any data, and we are not the superuser * we clear the setuid and setgid bits as a precaution against * tampering. */ if ((ip->i_mode & (ISUID | ISGID)) && resid > uio->uio_resid && ucred) { if (priv_check_cred(ucred, PRIV_VFS_RETAINSUGID, 0)) { ip->i_mode &= ~(ISUID | ISGID); dp->di_mode = ip->i_mode; } } if (error) { if (ioflag & IO_UNIT) { (void)ffs_truncate(vp, osize, IO_EXT | (ioflag&IO_SYNC), ucred); uio->uio_offset -= resid - uio->uio_resid; uio->uio_resid = resid; } } else if (resid > uio->uio_resid && (ioflag & IO_SYNC)) error = ffs_update(vp, 1); return (error); } /* * Vnode operating to retrieve a named extended attribute. * * Locate a particular EA (nspace:name) in the area (ptr:length), and return * the length of the EA, and possibly the pointer to the entry and to the data. */ static int ffs_findextattr(u_char *ptr, u_int length, int nspace, const char *name, u_char **eap, u_char **eac) { u_char *p, *pe, *pn, *p0; int eapad1, eapad2, ealength, ealen, nlen; uint32_t ul; pe = ptr + length; nlen = strlen(name); for (p = ptr; p < pe; p = pn) { p0 = p; bcopy(p, &ul, sizeof(ul)); pn = p + ul; /* make sure this entry is complete */ if (pn > pe) break; p += sizeof(uint32_t); if (*p != nspace) continue; p++; eapad2 = *p++; if (*p != nlen) continue; p++; if (bcmp(p, name, nlen)) continue; ealength = sizeof(uint32_t) + 3 + nlen; eapad1 = 8 - (ealength % 8); if (eapad1 == 8) eapad1 = 0; ealength += eapad1; ealen = ul - ealength - eapad2; p += nlen + eapad1; if (eap != NULL) *eap = p0; if (eac != NULL) *eac = p; return (ealen); } return(-1); } static int ffs_rdextattr(u_char **p, struct vnode *vp, struct thread *td, int extra) { struct inode *ip; struct ufs2_dinode *dp; struct fs *fs; struct uio luio; struct iovec liovec; u_int easize; int error; u_char *eae; ip = VTOI(vp); fs = ITOFS(ip); dp = ip->i_din2; easize = dp->di_extsize; if ((uoff_t)easize + extra > NXADDR * fs->fs_bsize) return (EFBIG); eae = malloc(easize + extra, M_TEMP, M_WAITOK); liovec.iov_base = eae; liovec.iov_len = easize; luio.uio_iov = &liovec; luio.uio_iovcnt = 1; luio.uio_offset = 0; luio.uio_resid = easize; luio.uio_segflg = UIO_SYSSPACE; luio.uio_rw = UIO_READ; luio.uio_td = td; error = ffs_extread(vp, &luio, IO_EXT | IO_SYNC); if (error) { free(eae, M_TEMP); return(error); } *p = eae; return (0); } static void ffs_lock_ea(struct vnode *vp) { struct inode *ip; ip = VTOI(vp); VI_LOCK(vp); while (ip->i_flag & IN_EA_LOCKED) { ip->i_flag |= IN_EA_LOCKWAIT; msleep(&ip->i_ea_refs, &vp->v_interlock, PINOD + 2, "ufs_ea", 0); } ip->i_flag |= IN_EA_LOCKED; VI_UNLOCK(vp); } static void ffs_unlock_ea(struct vnode *vp) { struct inode *ip; ip = VTOI(vp); VI_LOCK(vp); if (ip->i_flag & IN_EA_LOCKWAIT) wakeup(&ip->i_ea_refs); ip->i_flag &= ~(IN_EA_LOCKED | IN_EA_LOCKWAIT); VI_UNLOCK(vp); } static int ffs_open_ea(struct vnode *vp, struct ucred *cred, struct thread *td) { struct inode *ip; struct ufs2_dinode *dp; int error; ip = VTOI(vp); ffs_lock_ea(vp); if (ip->i_ea_area != NULL) { ip->i_ea_refs++; ffs_unlock_ea(vp); return (0); } dp = ip->i_din2; error = ffs_rdextattr(&ip->i_ea_area, vp, td, 0); if (error) { ffs_unlock_ea(vp); return (error); } ip->i_ea_len = dp->di_extsize; ip->i_ea_error = 0; ip->i_ea_refs++; ffs_unlock_ea(vp); return (0); } /* * Vnode extattr transaction commit/abort */ static int ffs_close_ea(struct vnode *vp, int commit, struct ucred *cred, struct thread *td) { struct inode *ip; struct uio luio; struct iovec liovec; int error; struct ufs2_dinode *dp; ip = VTOI(vp); ffs_lock_ea(vp); if (ip->i_ea_area == NULL) { ffs_unlock_ea(vp); return (EINVAL); } dp = ip->i_din2; error = ip->i_ea_error; if (commit && error == 0) { ASSERT_VOP_ELOCKED(vp, "ffs_close_ea commit"); if (cred == NOCRED) cred = vp->v_mount->mnt_cred; liovec.iov_base = ip->i_ea_area; liovec.iov_len = ip->i_ea_len; luio.uio_iov = &liovec; luio.uio_iovcnt = 1; luio.uio_offset = 0; luio.uio_resid = ip->i_ea_len; luio.uio_segflg = UIO_SYSSPACE; luio.uio_rw = UIO_WRITE; luio.uio_td = td; /* XXX: I'm not happy about truncating to zero size */ if (ip->i_ea_len < dp->di_extsize) error = ffs_truncate(vp, 0, IO_EXT, cred); error = ffs_extwrite(vp, &luio, IO_EXT | IO_SYNC, cred); } if (--ip->i_ea_refs == 0) { free(ip->i_ea_area, M_TEMP); ip->i_ea_area = NULL; ip->i_ea_len = 0; ip->i_ea_error = 0; } ffs_unlock_ea(vp); return (error); } /* * Vnode extattr strategy routine for fifos. * * We need to check for a read or write of the external attributes. * Otherwise we just fall through and do the usual thing. */ static int ffsext_strategy(struct vop_strategy_args *ap) /* struct vop_strategy_args { struct vnodeop_desc *a_desc; struct vnode *a_vp; struct buf *a_bp; }; */ { struct vnode *vp; daddr_t lbn; vp = ap->a_vp; lbn = ap->a_bp->b_lblkno; if (I_IS_UFS2(VTOI(vp)) && lbn < 0 && lbn >= -NXADDR) return (VOP_STRATEGY_APV(&ufs_vnodeops, ap)); if (vp->v_type == VFIFO) return (VOP_STRATEGY_APV(&ufs_fifoops, ap)); panic("spec nodes went here"); } /* * Vnode extattr transaction commit/abort */ static int ffs_openextattr(struct vop_openextattr_args *ap) /* struct vop_openextattr_args { struct vnodeop_desc *a_desc; struct vnode *a_vp; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); return (ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td)); } /* * Vnode extattr transaction commit/abort */ static int ffs_closeextattr(struct vop_closeextattr_args *ap) /* struct vop_closeextattr_args { struct vnodeop_desc *a_desc; struct vnode *a_vp; int a_commit; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); if (ap->a_commit && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) return (EROFS); return (ffs_close_ea(ap->a_vp, ap->a_commit, ap->a_cred, ap->a_td)); } /* * Vnode operation to remove a named attribute. */ static int ffs_deleteextattr(struct vop_deleteextattr_args *ap) /* vop_deleteextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct inode *ip; struct fs *fs; uint32_t ealength, ul; int ealen, olen, eapad1, eapad2, error, i, easize; u_char *eae, *p; ip = VTOI(ap->a_vp); fs = ITOFS(ip); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); if (strlen(ap->a_name) == 0) return (EINVAL); if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) { /* * ffs_lock_ea is not needed there, because the vnode * must be exclusively locked. */ if (ip->i_ea_area != NULL && ip->i_ea_error == 0) ip->i_ea_error = error; return (error); } error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); if (error) return (error); ealength = eapad1 = ealen = eapad2 = 0; eae = malloc(ip->i_ea_len, M_TEMP, M_WAITOK); bcopy(ip->i_ea_area, eae, ip->i_ea_len); easize = ip->i_ea_len; olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, &p, NULL); if (olen == -1) { /* delete but nonexistent */ free(eae, M_TEMP); ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); return(ENOATTR); } bcopy(p, &ul, sizeof ul); i = p - eae + ul; if (ul != ealength) { bcopy(p + ul, p + ealength, easize - i); easize += (ealength - ul); } if (easize > NXADDR * fs->fs_bsize) { free(eae, M_TEMP); ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); if (ip->i_ea_area != NULL && ip->i_ea_error == 0) ip->i_ea_error = ENOSPC; return(ENOSPC); } p = ip->i_ea_area; ip->i_ea_area = eae; ip->i_ea_len = easize; free(p, M_TEMP); error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); return(error); } /* * Vnode operation to retrieve a named extended attribute. */ static int ffs_getextattr(struct vop_getextattr_args *ap) /* vop_getextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct inode *ip; u_char *eae, *p; unsigned easize; int error, ealen; ip = VTOI(ap->a_vp); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); if (error) return (error); eae = ip->i_ea_area; easize = ip->i_ea_len; ealen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, NULL, &p); if (ealen >= 0) { error = 0; if (ap->a_size != NULL) *ap->a_size = ealen; else if (ap->a_uio != NULL) error = uiomove(p, ealen, ap->a_uio); } else error = ENOATTR; ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); return(error); } /* * Vnode operation to retrieve extended attributes on a vnode. */ static int ffs_listextattr(struct vop_listextattr_args *ap) /* vop_listextattr { IN struct vnode *a_vp; IN int a_attrnamespace; INOUT struct uio *a_uio; OUT size_t *a_size; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct inode *ip; u_char *eae, *p, *pe, *pn; unsigned easize; uint32_t ul; int error, ealen; ip = VTOI(ap->a_vp); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VREAD); if (error) return (error); error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); if (error) return (error); eae = ip->i_ea_area; easize = ip->i_ea_len; error = 0; if (ap->a_size != NULL) *ap->a_size = 0; pe = eae + easize; for(p = eae; error == 0 && p < pe; p = pn) { bcopy(p, &ul, sizeof(ul)); pn = p + ul; if (pn > pe) break; p += sizeof(ul); if (*p++ != ap->a_attrnamespace) continue; p++; /* pad2 */ ealen = *p; if (ap->a_size != NULL) { *ap->a_size += ealen + 1; } else if (ap->a_uio != NULL) { error = uiomove(p, ealen + 1, ap->a_uio); } } ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); return(error); } /* * Vnode operation to set a named attribute. */ static int ffs_setextattr(struct vop_setextattr_args *ap) /* vop_setextattr { IN struct vnode *a_vp; IN int a_attrnamespace; IN const char *a_name; INOUT struct uio *a_uio; IN struct ucred *a_cred; IN struct thread *a_td; }; */ { struct inode *ip; struct fs *fs; uint32_t ealength, ul; ssize_t ealen; int olen, eapad1, eapad2, error, i, easize; u_char *eae, *p; ip = VTOI(ap->a_vp); fs = ITOFS(ip); if (ap->a_vp->v_type == VCHR || ap->a_vp->v_type == VBLK) return (EOPNOTSUPP); if (strlen(ap->a_name) == 0) return (EINVAL); /* XXX Now unsupported API to delete EAs using NULL uio. */ if (ap->a_uio == NULL) return (EOPNOTSUPP); if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) return (EROFS); ealen = ap->a_uio->uio_resid; if (ealen < 0 || ealen > lblktosize(fs, NXADDR)) return (EINVAL); error = extattr_check_cred(ap->a_vp, ap->a_attrnamespace, ap->a_cred, ap->a_td, VWRITE); if (error) { /* * ffs_lock_ea is not needed there, because the vnode * must be exclusively locked. */ if (ip->i_ea_area != NULL && ip->i_ea_error == 0) ip->i_ea_error = error; return (error); } error = ffs_open_ea(ap->a_vp, ap->a_cred, ap->a_td); if (error) return (error); ealength = sizeof(uint32_t) + 3 + strlen(ap->a_name); eapad1 = 8 - (ealength % 8); if (eapad1 == 8) eapad1 = 0; eapad2 = 8 - (ealen % 8); if (eapad2 == 8) eapad2 = 0; ealength += eapad1 + ealen + eapad2; eae = malloc(ip->i_ea_len + ealength, M_TEMP, M_WAITOK); bcopy(ip->i_ea_area, eae, ip->i_ea_len); easize = ip->i_ea_len; olen = ffs_findextattr(eae, easize, ap->a_attrnamespace, ap->a_name, &p, NULL); if (olen == -1) { /* new, append at end */ p = eae + easize; easize += ealength; } else { bcopy(p, &ul, sizeof ul); i = p - eae + ul; if (ul != ealength) { bcopy(p + ul, p + ealength, easize - i); easize += (ealength - ul); } } if (easize > lblktosize(fs, NXADDR)) { free(eae, M_TEMP); ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); if (ip->i_ea_area != NULL && ip->i_ea_error == 0) ip->i_ea_error = ENOSPC; return(ENOSPC); } bcopy(&ealength, p, sizeof(ealength)); p += sizeof(ealength); *p++ = ap->a_attrnamespace; *p++ = eapad2; *p++ = strlen(ap->a_name); strcpy(p, ap->a_name); p += strlen(ap->a_name); bzero(p, eapad1); p += eapad1; error = uiomove(p, ealen, ap->a_uio); if (error) { free(eae, M_TEMP); ffs_close_ea(ap->a_vp, 0, ap->a_cred, ap->a_td); if (ip->i_ea_area != NULL && ip->i_ea_error == 0) ip->i_ea_error = error; return(error); } p += ealen; bzero(p, eapad2); p = ip->i_ea_area; ip->i_ea_area = eae; ip->i_ea_len = easize; free(p, M_TEMP); error = ffs_close_ea(ap->a_vp, 1, ap->a_cred, ap->a_td); return(error); } /* * Vnode pointer to File handle */ static int ffs_vptofh(struct vop_vptofh_args *ap) /* vop_vptofh { IN struct vnode *a_vp; IN struct fid *a_fhp; }; */ { struct inode *ip; struct ufid *ufhp; ip = VTOI(ap->a_vp); ufhp = (struct ufid *)ap->a_fhp; ufhp->ufid_len = sizeof(struct ufid); ufhp->ufid_ino = ip->i_number; ufhp->ufid_gen = ip->i_gen; return (0); } SYSCTL_DECL(_vfs_ffs); static int use_buf_pager = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, use_buf_pager, CTLFLAG_RWTUN, &use_buf_pager, 0, "Always use buffer pager instead of bmap"); static daddr_t ffs_gbp_getblkno(struct vnode *vp, vm_ooffset_t off) { return (lblkno(VFSTOUFS(vp->v_mount)->um_fs, off)); } static int ffs_gbp_getblksz(struct vnode *vp, daddr_t lbn) { return (blksize(VFSTOUFS(vp->v_mount)->um_fs, VTOI(vp), lbn)); } static int ffs_getpages(struct vop_getpages_args *ap) { struct vnode *vp; struct ufsmount *um; vp = ap->a_vp; um = VFSTOUFS(vp->v_mount); if (!use_buf_pager && um->um_devvp->v_bufobj.bo_bsize <= PAGE_SIZE) return (vnode_pager_generic_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL)); return (vfs_bio_getpages(vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, ffs_gbp_getblkno, ffs_gbp_getblksz)); } Index: head/sys/vm/vm_pageout.c =================================================================== --- head/sys/vm/vm_pageout.c (revision 309061) +++ head/sys/vm/vm_pageout.c (revision 309062) @@ -1,2266 +1,2266 @@ /*- * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1994 John S. Dyson * All rights reserved. * Copyright (c) 1994 David Greenman * All rights reserved. * Copyright (c) 2005 Yahoo! Technologies Norway AS * All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_pageout.c 7.4 (Berkeley) 5/7/91 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * The proverbial page-out daemon. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.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 /* * System initialization */ /* the kernel process "vm_pageout"*/ static void vm_pageout(void); static void vm_pageout_init(void); static int vm_pageout_clean(vm_page_t m, int *numpagedout); static int vm_pageout_cluster(vm_page_t m); static bool vm_pageout_scan(struct vm_domain *vmd, int pass); static void vm_pageout_mightbe_oom(struct vm_domain *vmd, int page_shortage, int starting_page_shortage); SYSINIT(pagedaemon_init, SI_SUB_KTHREAD_PAGE, SI_ORDER_FIRST, vm_pageout_init, NULL); struct proc *pageproc; static struct kproc_desc page_kp = { "pagedaemon", vm_pageout, &pageproc }; SYSINIT(pagedaemon, SI_SUB_KTHREAD_PAGE, SI_ORDER_SECOND, kproc_start, &page_kp); SDT_PROVIDER_DEFINE(vm); SDT_PROBE_DEFINE(vm, , , vm__lowmem_scan); #if !defined(NO_SWAPPING) /* the kernel process "vm_daemon"*/ static void vm_daemon(void); static struct proc *vmproc; static struct kproc_desc vm_kp = { "vmdaemon", vm_daemon, &vmproc }; SYSINIT(vmdaemon, SI_SUB_KTHREAD_VM, SI_ORDER_FIRST, kproc_start, &vm_kp); #endif /* Pagedaemon activity rates, in subdivisions of one second. */ #define VM_LAUNDER_RATE 10 #define VM_INACT_SCAN_RATE 2 int vm_pageout_deficit; /* Estimated number of pages deficit */ u_int vm_pageout_wakeup_thresh; static int vm_pageout_oom_seq = 12; bool vm_pageout_wanted; /* Event on which pageout daemon sleeps */ bool vm_pages_needed; /* Are threads waiting for free pages? */ /* Pending request for dirty page laundering. */ static enum { VM_LAUNDRY_IDLE, VM_LAUNDRY_BACKGROUND, VM_LAUNDRY_SHORTFALL } vm_laundry_request = VM_LAUNDRY_IDLE; #if !defined(NO_SWAPPING) static int vm_pageout_req_swapout; /* XXX */ static int vm_daemon_needed; static struct mtx vm_daemon_mtx; /* Allow for use by vm_pageout before vm_daemon is initialized. */ MTX_SYSINIT(vm_daemon, &vm_daemon_mtx, "vm daemon", MTX_DEF); #endif static int vm_pageout_update_period; static int disable_swap_pageouts; static int lowmem_period = 10; static time_t lowmem_uptime; #if defined(NO_SWAPPING) static int vm_swap_enabled = 0; static int vm_swap_idle_enabled = 0; #else static int vm_swap_enabled = 1; static int vm_swap_idle_enabled = 0; #endif static int vm_panic_on_oom = 0; SYSCTL_INT(_vm, OID_AUTO, panic_on_oom, CTLFLAG_RWTUN, &vm_panic_on_oom, 0, "panic on out of memory instead of killing the largest process"); SYSCTL_INT(_vm, OID_AUTO, pageout_wakeup_thresh, CTLFLAG_RW, &vm_pageout_wakeup_thresh, 0, "free page threshold for waking up the pageout daemon"); SYSCTL_INT(_vm, OID_AUTO, pageout_update_period, CTLFLAG_RW, &vm_pageout_update_period, 0, "Maximum active LRU update period"); SYSCTL_INT(_vm, OID_AUTO, lowmem_period, CTLFLAG_RW, &lowmem_period, 0, "Low memory callback period"); #if defined(NO_SWAPPING) SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RD, &vm_swap_enabled, 0, "Enable entire process swapout"); SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RD, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); #else SYSCTL_INT(_vm, VM_SWAPPING_ENABLED, swap_enabled, CTLFLAG_RW, &vm_swap_enabled, 0, "Enable entire process swapout"); SYSCTL_INT(_vm, OID_AUTO, swap_idle_enabled, CTLFLAG_RW, &vm_swap_idle_enabled, 0, "Allow swapout on idle criteria"); #endif SYSCTL_INT(_vm, OID_AUTO, disable_swapspace_pageouts, CTLFLAG_RW, &disable_swap_pageouts, 0, "Disallow swapout of dirty pages"); static int pageout_lock_miss; SYSCTL_INT(_vm, OID_AUTO, pageout_lock_miss, CTLFLAG_RD, &pageout_lock_miss, 0, "vget() lock misses during pageout"); SYSCTL_INT(_vm, OID_AUTO, pageout_oom_seq, CTLFLAG_RW, &vm_pageout_oom_seq, 0, "back-to-back calls to oom detector to start OOM"); static int act_scan_laundry_weight = 3; SYSCTL_INT(_vm, OID_AUTO, act_scan_laundry_weight, CTLFLAG_RW, &act_scan_laundry_weight, 0, "weight given to clean vs. dirty pages in active queue scans"); static u_int vm_background_launder_target; SYSCTL_UINT(_vm, OID_AUTO, background_launder_target, CTLFLAG_RW, &vm_background_launder_target, 0, "background laundering target, in pages"); static u_int vm_background_launder_rate = 4096; SYSCTL_UINT(_vm, OID_AUTO, background_launder_rate, CTLFLAG_RW, &vm_background_launder_rate, 0, "background laundering rate, in kilobytes per second"); static u_int vm_background_launder_max = 20 * 1024; SYSCTL_UINT(_vm, OID_AUTO, background_launder_max, CTLFLAG_RW, &vm_background_launder_max, 0, "background laundering cap, in kilobytes"); #define VM_PAGEOUT_PAGE_COUNT 16 int vm_pageout_page_count = VM_PAGEOUT_PAGE_COUNT; int vm_page_max_wired; /* XXX max # of wired pages system-wide */ SYSCTL_INT(_vm, OID_AUTO, max_wired, CTLFLAG_RW, &vm_page_max_wired, 0, "System-wide limit to wired page count"); static u_int isqrt(u_int num); static boolean_t vm_pageout_fallback_object_lock(vm_page_t, vm_page_t *); static int vm_pageout_launder(struct vm_domain *vmd, int launder, bool in_shortfall); static void vm_pageout_laundry_worker(void *arg); #if !defined(NO_SWAPPING) static void vm_pageout_map_deactivate_pages(vm_map_t, long); static void vm_pageout_object_deactivate_pages(pmap_t, vm_object_t, long); static void vm_req_vmdaemon(int req); #endif static boolean_t vm_pageout_page_lock(vm_page_t, vm_page_t *); /* * Initialize a dummy page for marking the caller's place in the specified * paging queue. In principle, this function only needs to set the flag * PG_MARKER. Nonetheless, it write busies and initializes the hold count * to one as safety precautions. */ static void vm_pageout_init_marker(vm_page_t marker, u_short queue) { bzero(marker, sizeof(*marker)); marker->flags = PG_MARKER; marker->busy_lock = VPB_SINGLE_EXCLUSIVER; marker->queue = queue; marker->hold_count = 1; } /* * vm_pageout_fallback_object_lock: * * Lock vm object currently associated with `m'. VM_OBJECT_TRYWLOCK is * known to have failed and page queue must be either PQ_ACTIVE or * PQ_INACTIVE. To avoid lock order violation, unlock the page queue * while locking the vm object. Use marker page to detect page queue * changes and maintain notion of next page on page queue. Return * TRUE if no changes were detected, FALSE otherwise. vm object is * locked on return. * * This function depends on both the lock portion of struct vm_object * and normal struct vm_page being type stable. */ static boolean_t vm_pageout_fallback_object_lock(vm_page_t m, vm_page_t *next) { struct vm_page marker; struct vm_pagequeue *pq; boolean_t unchanged; u_short queue; vm_object_t object; queue = m->queue; vm_pageout_init_marker(&marker, queue); pq = vm_page_pagequeue(m); object = m->object; TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); vm_pagequeue_unlock(pq); vm_page_unlock(m); VM_OBJECT_WLOCK(object); vm_page_lock(m); vm_pagequeue_lock(pq); /* * The page's object might have changed, and/or the page might * have moved from its original position in the queue. If the * page's object has changed, then the caller should abandon * processing the page because the wrong object lock was * acquired. Use the marker's plinks.q, not the page's, to * determine if the page has been moved. The state of the * page's plinks.q can be indeterminate; whereas, the marker's * plinks.q must be valid. */ *next = TAILQ_NEXT(&marker, plinks.q); unchanged = m->object == object && m == TAILQ_PREV(&marker, pglist, plinks.q); KASSERT(!unchanged || m->queue == queue, ("page %p queue %d %d", m, queue, m->queue)); TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); return (unchanged); } /* * Lock the page while holding the page queue lock. Use marker page * to detect page queue changes and maintain notion of next page on * page queue. Return TRUE if no changes were detected, FALSE * otherwise. The page is locked on return. The page queue lock might * be dropped and reacquired. * * This function depends on normal struct vm_page being type stable. */ static boolean_t vm_pageout_page_lock(vm_page_t m, vm_page_t *next) { struct vm_page marker; struct vm_pagequeue *pq; boolean_t unchanged; u_short queue; vm_page_lock_assert(m, MA_NOTOWNED); if (vm_page_trylock(m)) return (TRUE); queue = m->queue; vm_pageout_init_marker(&marker, queue); pq = vm_page_pagequeue(m); TAILQ_INSERT_AFTER(&pq->pq_pl, m, &marker, plinks.q); vm_pagequeue_unlock(pq); vm_page_lock(m); vm_pagequeue_lock(pq); /* Page queue might have changed. */ *next = TAILQ_NEXT(&marker, plinks.q); unchanged = m == TAILQ_PREV(&marker, pglist, plinks.q); KASSERT(!unchanged || m->queue == queue, ("page %p queue %d %d", m, queue, m->queue)); TAILQ_REMOVE(&pq->pq_pl, &marker, plinks.q); return (unchanged); } /* * Scan for pages at adjacent offsets within the given page's object that are * eligible for laundering, form a cluster of these pages and the given page, * and launder that cluster. */ static int vm_pageout_cluster(vm_page_t m) { vm_object_t object; vm_page_t mc[2 * vm_pageout_page_count], p, pb, ps; vm_pindex_t pindex; int ib, is, page_base, pageout_count; vm_page_assert_locked(m); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); pindex = m->pindex; /* * We can't clean the page if it is busy or held. */ vm_page_assert_unbusied(m); KASSERT(m->hold_count == 0, ("page %p is held", m)); vm_page_unlock(m); mc[vm_pageout_page_count] = pb = ps = m; pageout_count = 1; page_base = vm_pageout_page_count; ib = 1; is = 1; /* * We can cluster only if the page is not clean, busy, or held, and * the page is in the laundry queue. * * During heavy mmap/modification loads the pageout * daemon can really fragment the underlying file * due to flushing pages out of order and not trying to * align the clusters (which leaves sporadic out-of-order * holes). To solve this problem we do the reverse scan * first and attempt to align our cluster, then do a * forward scan if room remains. */ more: while (ib != 0 && pageout_count < vm_pageout_page_count) { if (ib > pindex) { ib = 0; break; } if ((p = vm_page_prev(pb)) == NULL || vm_page_busied(p)) { ib = 0; break; } vm_page_test_dirty(p); if (p->dirty == 0) { ib = 0; break; } vm_page_lock(p); if (!vm_page_in_laundry(p) || p->hold_count != 0) { /* may be undergoing I/O */ vm_page_unlock(p); ib = 0; break; } vm_page_unlock(p); mc[--page_base] = pb = p; ++pageout_count; ++ib; /* * We are at an alignment boundary. Stop here, and switch * directions. Do not clear ib. */ if ((pindex - (ib - 1)) % vm_pageout_page_count == 0) break; } while (pageout_count < vm_pageout_page_count && pindex + is < object->size) { if ((p = vm_page_next(ps)) == NULL || vm_page_busied(p)) break; vm_page_test_dirty(p); if (p->dirty == 0) break; vm_page_lock(p); if (!vm_page_in_laundry(p) || p->hold_count != 0) { /* may be undergoing I/O */ vm_page_unlock(p); break; } vm_page_unlock(p); mc[page_base + pageout_count] = ps = p; ++pageout_count; ++is; } /* * If we exhausted our forward scan, continue with the reverse scan * when possible, even past an alignment boundary. This catches * boundary conditions. */ if (ib != 0 && pageout_count < vm_pageout_page_count) goto more; - return (vm_pageout_flush(&mc[page_base], pageout_count, 0, 0, NULL, - NULL)); + return (vm_pageout_flush(&mc[page_base], pageout_count, + VM_PAGER_PUT_NOREUSE, 0, NULL, NULL)); } /* * vm_pageout_flush() - launder the given pages * * The given pages are laundered. Note that we setup for the start of * I/O ( i.e. busy the page ), mark it read-only, and bump the object * reference count all in here rather then in the parent. If we want * the parent to do more sophisticated things we may have to change * the ordering. * * Returned runlen is the count of pages between mreq and first * page after mreq with status VM_PAGER_AGAIN. * *eio is set to TRUE if pager returned VM_PAGER_ERROR or VM_PAGER_FAIL * for any page in runlen set. */ int vm_pageout_flush(vm_page_t *mc, int count, int flags, int mreq, int *prunlen, boolean_t *eio) { vm_object_t object = mc[0]->object; int pageout_status[count]; int numpagedout = 0; int i, runlen; VM_OBJECT_ASSERT_WLOCKED(object); /* * Initiate I/O. Bump the vm_page_t->busy counter and * mark the pages read-only. * * We do not have to fixup the clean/dirty bits here... we can * allow the pager to do it after the I/O completes. * * NOTE! mc[i]->dirty may be partial or fragmented due to an * edge case with file fragments. */ for (i = 0; i < count; i++) { KASSERT(mc[i]->valid == VM_PAGE_BITS_ALL, ("vm_pageout_flush: partially invalid page %p index %d/%d", mc[i], i, count)); vm_page_sbusy(mc[i]); pmap_remove_write(mc[i]); } vm_object_pip_add(object, count); vm_pager_put_pages(object, mc, count, flags, pageout_status); runlen = count - mreq; if (eio != NULL) *eio = FALSE; for (i = 0; i < count; i++) { vm_page_t mt = mc[i]; KASSERT(pageout_status[i] == VM_PAGER_PEND || !pmap_page_is_write_mapped(mt), ("vm_pageout_flush: page %p is not write protected", mt)); switch (pageout_status[i]) { case VM_PAGER_OK: vm_page_lock(mt); if (vm_page_in_laundry(mt)) vm_page_deactivate_noreuse(mt); vm_page_unlock(mt); /* FALLTHROUGH */ case VM_PAGER_PEND: numpagedout++; break; case VM_PAGER_BAD: /* * The page is outside the object's range. We pretend * that the page out worked and clean the page, so the * changes will be lost if the page is reclaimed by * the page daemon. */ vm_page_undirty(mt); vm_page_lock(mt); if (vm_page_in_laundry(mt)) vm_page_deactivate_noreuse(mt); vm_page_unlock(mt); break; case VM_PAGER_ERROR: case VM_PAGER_FAIL: /* * If the page couldn't be paged out, then reactivate * it so that it doesn't clog the laundry and inactive * queues. (We will try paging it out again later). */ vm_page_lock(mt); vm_page_activate(mt); vm_page_unlock(mt); if (eio != NULL && i >= mreq && i - mreq < runlen) *eio = TRUE; break; case VM_PAGER_AGAIN: if (i >= mreq && i - mreq < runlen) runlen = i - mreq; break; } /* * If the operation is still going, leave the page busy to * block all other accesses. Also, leave the paging in * progress indicator set so that we don't attempt an object * collapse. */ if (pageout_status[i] != VM_PAGER_PEND) { vm_object_pip_wakeup(object); vm_page_sunbusy(mt); } } if (prunlen != NULL) *prunlen = runlen; return (numpagedout); } #if !defined(NO_SWAPPING) /* * vm_pageout_object_deactivate_pages * * Deactivate enough pages to satisfy the inactive target * requirements. * * The object and map must be locked. */ static void vm_pageout_object_deactivate_pages(pmap_t pmap, vm_object_t first_object, long desired) { vm_object_t backing_object, object; vm_page_t p; int act_delta, remove_mode; VM_OBJECT_ASSERT_LOCKED(first_object); if ((first_object->flags & OBJ_FICTITIOUS) != 0) return; for (object = first_object;; object = backing_object) { if (pmap_resident_count(pmap) <= desired) goto unlock_return; VM_OBJECT_ASSERT_LOCKED(object); if ((object->flags & OBJ_UNMANAGED) != 0 || object->paging_in_progress != 0) goto unlock_return; remove_mode = 0; if (object->shadow_count > 1) remove_mode = 1; /* * Scan the object's entire memory queue. */ TAILQ_FOREACH(p, &object->memq, listq) { if (pmap_resident_count(pmap) <= desired) goto unlock_return; if (vm_page_busied(p)) continue; PCPU_INC(cnt.v_pdpages); vm_page_lock(p); if (p->wire_count != 0 || p->hold_count != 0 || !pmap_page_exists_quick(pmap, p)) { vm_page_unlock(p); continue; } act_delta = pmap_ts_referenced(p); if ((p->aflags & PGA_REFERENCED) != 0) { if (act_delta == 0) act_delta = 1; vm_page_aflag_clear(p, PGA_REFERENCED); } if (!vm_page_active(p) && act_delta != 0) { vm_page_activate(p); p->act_count += act_delta; } else if (vm_page_active(p)) { if (act_delta == 0) { p->act_count -= min(p->act_count, ACT_DECLINE); if (!remove_mode && p->act_count == 0) { pmap_remove_all(p); vm_page_deactivate(p); } else vm_page_requeue(p); } else { vm_page_activate(p); if (p->act_count < ACT_MAX - ACT_ADVANCE) p->act_count += ACT_ADVANCE; vm_page_requeue(p); } } else if (vm_page_inactive(p)) pmap_remove_all(p); vm_page_unlock(p); } if ((backing_object = object->backing_object) == NULL) goto unlock_return; VM_OBJECT_RLOCK(backing_object); if (object != first_object) VM_OBJECT_RUNLOCK(object); } unlock_return: if (object != first_object) VM_OBJECT_RUNLOCK(object); } /* * deactivate some number of pages in a map, try to do it fairly, but * that is really hard to do. */ static void vm_pageout_map_deactivate_pages(map, desired) vm_map_t map; long desired; { vm_map_entry_t tmpe; vm_object_t obj, bigobj; int nothingwired; if (!vm_map_trylock(map)) return; bigobj = NULL; nothingwired = TRUE; /* * first, search out the biggest object, and try to free pages from * that. */ tmpe = map->header.next; while (tmpe != &map->header) { if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { obj = tmpe->object.vm_object; if (obj != NULL && VM_OBJECT_TRYRLOCK(obj)) { if (obj->shadow_count <= 1 && (bigobj == NULL || bigobj->resident_page_count < obj->resident_page_count)) { if (bigobj != NULL) VM_OBJECT_RUNLOCK(bigobj); bigobj = obj; } else VM_OBJECT_RUNLOCK(obj); } } if (tmpe->wired_count > 0) nothingwired = FALSE; tmpe = tmpe->next; } if (bigobj != NULL) { vm_pageout_object_deactivate_pages(map->pmap, bigobj, desired); VM_OBJECT_RUNLOCK(bigobj); } /* * Next, hunt around for other pages to deactivate. We actually * do this search sort of wrong -- .text first is not the best idea. */ tmpe = map->header.next; while (tmpe != &map->header) { if (pmap_resident_count(vm_map_pmap(map)) <= desired) break; if ((tmpe->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) { obj = tmpe->object.vm_object; if (obj != NULL) { VM_OBJECT_RLOCK(obj); vm_pageout_object_deactivate_pages(map->pmap, obj, desired); VM_OBJECT_RUNLOCK(obj); } } tmpe = tmpe->next; } /* * Remove all mappings if a process is swapped out, this will free page * table pages. */ if (desired == 0 && nothingwired) { pmap_remove(vm_map_pmap(map), vm_map_min(map), vm_map_max(map)); } vm_map_unlock(map); } #endif /* !defined(NO_SWAPPING) */ /* * Attempt to acquire all of the necessary locks to launder a page and * then call through the clustering layer to PUTPAGES. Wait a short * time for a vnode lock. * * Requires the page and object lock on entry, releases both before return. * Returns 0 on success and an errno otherwise. */ static int vm_pageout_clean(vm_page_t m, int *numpagedout) { struct vnode *vp; struct mount *mp; vm_object_t object; vm_pindex_t pindex; int error, lockmode; vm_page_assert_locked(m); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); error = 0; vp = NULL; mp = NULL; /* * The object is already known NOT to be dead. It * is possible for the vget() to block the whole * pageout daemon, but the new low-memory handling * code should prevent it. * * We can't wait forever for the vnode lock, we might * deadlock due to a vn_read() getting stuck in * vm_wait while holding this vnode. We skip the * vnode if we can't get it in a reasonable amount * of time. */ if (object->type == OBJT_VNODE) { vm_page_unlock(m); vp = object->handle; if (vp->v_type == VREG && vn_start_write(vp, &mp, V_NOWAIT) != 0) { mp = NULL; error = EDEADLK; goto unlock_all; } KASSERT(mp != NULL, ("vp %p with NULL v_mount", vp)); vm_object_reference_locked(object); pindex = m->pindex; VM_OBJECT_WUNLOCK(object); lockmode = MNT_SHARED_WRITES(vp->v_mount) ? LK_SHARED : LK_EXCLUSIVE; if (vget(vp, lockmode | LK_TIMELOCK, curthread)) { vp = NULL; error = EDEADLK; goto unlock_mp; } VM_OBJECT_WLOCK(object); vm_page_lock(m); /* * While the object and page were unlocked, the page * may have been: * (1) moved to a different queue, * (2) reallocated to a different object, * (3) reallocated to a different offset, or * (4) cleaned. */ if (!vm_page_in_laundry(m) || m->object != object || m->pindex != pindex || m->dirty == 0) { vm_page_unlock(m); error = ENXIO; goto unlock_all; } /* * The page may have been busied or held while the object * and page locks were released. */ if (vm_page_busied(m) || m->hold_count != 0) { vm_page_unlock(m); error = EBUSY; goto unlock_all; } } /* * If a page is dirty, then it is either being washed * (but not yet cleaned) or it is still in the * laundry. If it is still in the laundry, then we * start the cleaning operation. */ if ((*numpagedout = vm_pageout_cluster(m)) == 0) error = EIO; unlock_all: VM_OBJECT_WUNLOCK(object); unlock_mp: vm_page_lock_assert(m, MA_NOTOWNED); if (mp != NULL) { if (vp != NULL) vput(vp); vm_object_deallocate(object); vn_finished_write(mp); } return (error); } /* * Attempt to launder the specified number of pages. * * Returns the number of pages successfully laundered. */ static int vm_pageout_launder(struct vm_domain *vmd, int launder, bool in_shortfall) { struct vm_pagequeue *pq; vm_object_t object; vm_page_t m, next; int act_delta, error, maxscan, numpagedout, starting_target; int vnodes_skipped; bool pageout_ok, queue_locked; starting_target = launder; vnodes_skipped = 0; /* * Scan the laundry queue for pages eligible to be laundered. We stop * once the target number of dirty pages have been laundered, or once * we've reached the end of the queue. A single iteration of this loop * may cause more than one page to be laundered because of clustering. * * maxscan ensures that we don't re-examine requeued pages. Any * additional pages written as part of a cluster are subtracted from * maxscan since they must be taken from the laundry queue. */ pq = &vmd->vmd_pagequeues[PQ_LAUNDRY]; maxscan = pq->pq_cnt; vm_pagequeue_lock(pq); queue_locked = true; for (m = TAILQ_FIRST(&pq->pq_pl); m != NULL && maxscan-- > 0 && launder > 0; m = next) { vm_pagequeue_assert_locked(pq); KASSERT(queue_locked, ("unlocked laundry queue")); KASSERT(vm_page_in_laundry(m), ("page %p has an inconsistent queue", m)); next = TAILQ_NEXT(m, plinks.q); if ((m->flags & PG_MARKER) != 0) continue; KASSERT((m->flags & PG_FICTITIOUS) == 0, ("PG_FICTITIOUS page %p cannot be in laundry queue", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("VPO_UNMANAGED page %p cannot be in laundry queue", m)); if (!vm_pageout_page_lock(m, &next) || m->hold_count != 0) { vm_page_unlock(m); continue; } object = m->object; if ((!VM_OBJECT_TRYWLOCK(object) && (!vm_pageout_fallback_object_lock(m, &next) || m->hold_count != 0)) || vm_page_busied(m)) { VM_OBJECT_WUNLOCK(object); vm_page_unlock(m); continue; } /* * Unlock the laundry queue, invalidating the 'next' pointer. * Use a marker to remember our place in the laundry queue. */ TAILQ_INSERT_AFTER(&pq->pq_pl, m, &vmd->vmd_laundry_marker, plinks.q); vm_pagequeue_unlock(pq); queue_locked = false; /* * Invalid pages can be easily freed. They cannot be * mapped; vm_page_free() asserts this. */ if (m->valid == 0) goto free_page; /* * If the page has been referenced and the object is not dead, * reactivate or requeue the page depending on whether the * object is mapped. */ if ((m->aflags & PGA_REFERENCED) != 0) { vm_page_aflag_clear(m, PGA_REFERENCED); act_delta = 1; } else act_delta = 0; if (object->ref_count != 0) act_delta += pmap_ts_referenced(m); else { KASSERT(!pmap_page_is_mapped(m), ("page %p is mapped", m)); } if (act_delta != 0) { if (object->ref_count != 0) { PCPU_INC(cnt.v_reactivated); vm_page_activate(m); /* * Increase the activation count if the page * was referenced while in the laundry queue. * This makes it less likely that the page will * be returned prematurely to the inactive * queue. */ m->act_count += act_delta + ACT_ADVANCE; /* * If this was a background laundering, count * activated pages towards our target. The * purpose of background laundering is to ensure * that pages are eventually cycled through the * laundry queue, and an activation is a valid * way out. */ if (!in_shortfall) launder--; goto drop_page; } else if ((object->flags & OBJ_DEAD) == 0) goto requeue_page; } /* * If the page appears to be clean at the machine-independent * layer, then remove all of its mappings from the pmap in * anticipation of freeing it. If, however, any of the page's * mappings allow write access, then the page may still be * modified until the last of those mappings are removed. */ if (object->ref_count != 0) { vm_page_test_dirty(m); if (m->dirty == 0) pmap_remove_all(m); } /* * Clean pages are freed, and dirty pages are paged out unless * they belong to a dead object. Requeueing dirty pages from * dead objects is pointless, as they are being paged out and * freed by the thread that destroyed the object. */ if (m->dirty == 0) { free_page: vm_page_free(m); PCPU_INC(cnt.v_dfree); } else if ((object->flags & OBJ_DEAD) == 0) { if (object->type != OBJT_SWAP && object->type != OBJT_DEFAULT) pageout_ok = true; else if (disable_swap_pageouts) pageout_ok = false; else pageout_ok = true; if (!pageout_ok) { requeue_page: vm_pagequeue_lock(pq); queue_locked = true; vm_page_requeue_locked(m); goto drop_page; } /* * Form a cluster with adjacent, dirty pages from the * same object, and page out that entire cluster. * * The adjacent, dirty pages must also be in the * laundry. However, their mappings are not checked * for new references. Consequently, a recently * referenced page may be paged out. However, that * page will not be prematurely reclaimed. After page * out, the page will be placed in the inactive queue, * where any new references will be detected and the * page reactivated. */ error = vm_pageout_clean(m, &numpagedout); if (error == 0) { launder -= numpagedout; maxscan -= numpagedout - 1; } else if (error == EDEADLK) { pageout_lock_miss++; vnodes_skipped++; } goto relock_queue; } drop_page: vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); relock_queue: if (!queue_locked) { vm_pagequeue_lock(pq); queue_locked = true; } next = TAILQ_NEXT(&vmd->vmd_laundry_marker, plinks.q); TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_laundry_marker, plinks.q); } vm_pagequeue_unlock(pq); /* * Wakeup the sync daemon if we skipped a vnode in a writeable object * and we didn't launder enough pages. */ if (vnodes_skipped > 0 && launder > 0) (void)speedup_syncer(); return (starting_target - launder); } /* * Compute the integer square root. */ static u_int isqrt(u_int num) { u_int bit, root, tmp; bit = 1u << ((NBBY * sizeof(u_int)) - 2); while (bit > num) bit >>= 2; root = 0; while (bit != 0) { tmp = root + bit; root >>= 1; if (num >= tmp) { num -= tmp; root += bit; } bit >>= 2; } return (root); } /* * Perform the work of the laundry thread: periodically wake up and determine * whether any pages need to be laundered. If so, determine the number of pages * that need to be laundered, and launder them. */ static void vm_pageout_laundry_worker(void *arg) { struct vm_domain *domain; struct vm_pagequeue *pq; uint64_t nclean, ndirty; u_int last_launder, wakeups; int domidx, last_target, launder, shortfall, shortfall_cycle, target; bool in_shortfall; domidx = (uintptr_t)arg; domain = &vm_dom[domidx]; pq = &domain->vmd_pagequeues[PQ_LAUNDRY]; KASSERT(domain->vmd_segs != 0, ("domain without segments")); vm_pageout_init_marker(&domain->vmd_laundry_marker, PQ_LAUNDRY); shortfall = 0; in_shortfall = false; shortfall_cycle = 0; target = 0; last_launder = 0; /* * The pageout laundry worker is never done, so loop forever. */ for (;;) { KASSERT(target >= 0, ("negative target %d", target)); KASSERT(shortfall_cycle >= 0, ("negative cycle %d", shortfall_cycle)); launder = 0; wakeups = VM_METER_PCPU_CNT(v_pdwakeups); /* * First determine whether we need to launder pages to meet a * shortage of free pages. */ if (shortfall > 0) { in_shortfall = true; shortfall_cycle = VM_LAUNDER_RATE / VM_INACT_SCAN_RATE; target = shortfall; } else if (!in_shortfall) goto trybackground; else if (shortfall_cycle == 0 || vm_laundry_target() <= 0) { /* * We recently entered shortfall and began laundering * pages. If we have completed that laundering run * (and we are no longer in shortfall) or we have met * our laundry target through other activity, then we * can stop laundering pages. */ in_shortfall = false; target = 0; goto trybackground; } last_launder = wakeups; launder = target / shortfall_cycle--; goto dolaundry; /* * There's no immediate need to launder any pages; see if we * meet the conditions to perform background laundering: * * 1. The ratio of dirty to clean inactive pages exceeds the * background laundering threshold and the pagedaemon has * been woken up to reclaim pages since our last * laundering, or * 2. we haven't yet reached the target of the current * background laundering run. * * The background laundering threshold is not a constant. * Instead, it is a slowly growing function of the number of * page daemon wakeups since the last laundering. Thus, as the * ratio of dirty to clean inactive pages grows, the amount of * memory pressure required to trigger laundering decreases. */ trybackground: nclean = vm_cnt.v_inactive_count + vm_cnt.v_free_count; ndirty = vm_cnt.v_laundry_count; if (target == 0 && wakeups != last_launder && ndirty * isqrt(wakeups - last_launder) >= nclean) { target = vm_background_launder_target; } /* * We have a non-zero background laundering target. If we've * laundered up to our maximum without observing a page daemon * wakeup, just stop. This is a safety belt that ensures we * don't launder an excessive amount if memory pressure is low * and the ratio of dirty to clean pages is large. Otherwise, * proceed at the background laundering rate. */ if (target > 0) { if (wakeups != last_launder) { last_launder = wakeups; last_target = target; } else if (last_target - target >= vm_background_launder_max * PAGE_SIZE / 1024) { target = 0; } launder = vm_background_launder_rate * PAGE_SIZE / 1024; launder /= VM_LAUNDER_RATE; if (launder > target) launder = target; } dolaundry: if (launder > 0) { /* * Because of I/O clustering, the number of laundered * pages could exceed "target" by the maximum size of * a cluster minus one. */ target -= min(vm_pageout_launder(domain, launder, in_shortfall), target); pause("laundp", hz / VM_LAUNDER_RATE); } /* * If we're not currently laundering pages and the page daemon * hasn't posted a new request, sleep until the page daemon * kicks us. */ vm_pagequeue_lock(pq); if (target == 0 && vm_laundry_request == VM_LAUNDRY_IDLE) (void)mtx_sleep(&vm_laundry_request, vm_pagequeue_lockptr(pq), PVM, "launds", 0); /* * If the pagedaemon has indicated that it's in shortfall, start * a shortfall laundering unless we're already in the middle of * one. This may preempt a background laundering. */ if (vm_laundry_request == VM_LAUNDRY_SHORTFALL && (!in_shortfall || shortfall_cycle == 0)) { shortfall = vm_laundry_target() + vm_pageout_deficit; target = 0; } else shortfall = 0; if (target == 0) vm_laundry_request = VM_LAUNDRY_IDLE; vm_pagequeue_unlock(pq); } } /* * vm_pageout_scan does the dirty work for the pageout daemon. * * pass == 0: Update active LRU/deactivate pages * pass >= 1: Free inactive pages * * Returns true if pass was zero or enough pages were freed by the inactive * queue scan to meet the target. */ static bool vm_pageout_scan(struct vm_domain *vmd, int pass) { vm_page_t m, next; struct vm_pagequeue *pq; vm_object_t object; long min_scan; int act_delta, addl_page_shortage, deficit, inactq_shortage, maxscan; int page_shortage, scan_tick, scanned, starting_page_shortage; boolean_t queue_locked; /* * If we need to reclaim memory ask kernel caches to return * some. We rate limit to avoid thrashing. */ if (vmd == &vm_dom[0] && pass > 0 && (time_uptime - lowmem_uptime) >= lowmem_period) { /* * Decrease registered cache sizes. */ SDT_PROBE0(vm, , , vm__lowmem_scan); EVENTHANDLER_INVOKE(vm_lowmem, 0); /* * We do this explicitly after the caches have been * drained above. */ uma_reclaim(); lowmem_uptime = time_uptime; } /* * The addl_page_shortage is the number of temporarily * stuck pages in the inactive queue. In other words, the * number of pages from the inactive count that should be * discounted in setting the target for the active queue scan. */ addl_page_shortage = 0; /* * Calculate the number of pages that we want to free. This number * can be negative if many pages are freed between the wakeup call to * the page daemon and this calculation. */ if (pass > 0) { deficit = atomic_readandclear_int(&vm_pageout_deficit); page_shortage = vm_paging_target() + deficit; } else page_shortage = deficit = 0; starting_page_shortage = page_shortage; /* * Start scanning the inactive queue for pages that we can free. The * scan will stop when we reach the target or we have scanned the * entire queue. (Note that m->act_count is not used to make * decisions for the inactive queue, only for the active queue.) */ pq = &vmd->vmd_pagequeues[PQ_INACTIVE]; maxscan = pq->pq_cnt; vm_pagequeue_lock(pq); queue_locked = TRUE; for (m = TAILQ_FIRST(&pq->pq_pl); m != NULL && maxscan-- > 0 && page_shortage > 0; m = next) { vm_pagequeue_assert_locked(pq); KASSERT(queue_locked, ("unlocked inactive queue")); KASSERT(vm_page_inactive(m), ("Inactive queue %p", m)); PCPU_INC(cnt.v_pdpages); next = TAILQ_NEXT(m, plinks.q); /* * skip marker pages */ if (m->flags & PG_MARKER) continue; KASSERT((m->flags & PG_FICTITIOUS) == 0, ("Fictitious page %p cannot be in inactive queue", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("Unmanaged page %p cannot be in inactive queue", m)); /* * The page or object lock acquisitions fail if the * page was removed from the queue or moved to a * different position within the queue. In either * case, addl_page_shortage should not be incremented. */ if (!vm_pageout_page_lock(m, &next)) goto unlock_page; else if (m->hold_count != 0) { /* * Held pages are essentially stuck in the * queue. So, they ought to be discounted * from the inactive count. See the * calculation of inactq_shortage before the * loop over the active queue below. */ addl_page_shortage++; goto unlock_page; } object = m->object; if (!VM_OBJECT_TRYWLOCK(object)) { if (!vm_pageout_fallback_object_lock(m, &next)) goto unlock_object; else if (m->hold_count != 0) { addl_page_shortage++; goto unlock_object; } } if (vm_page_busied(m)) { /* * Don't mess with busy pages. Leave them at * the front of the queue. Most likely, they * are being paged out and will leave the * queue shortly after the scan finishes. So, * they ought to be discounted from the * inactive count. */ addl_page_shortage++; unlock_object: VM_OBJECT_WUNLOCK(object); unlock_page: vm_page_unlock(m); continue; } KASSERT(m->hold_count == 0, ("Held page %p", m)); /* * Dequeue the inactive page and unlock the inactive page * queue, invalidating the 'next' pointer. Dequeueing the * page here avoids a later reacquisition (and release) of * the inactive page queue lock when vm_page_activate(), * vm_page_free(), or vm_page_launder() is called. Use a * marker to remember our place in the inactive queue. */ TAILQ_INSERT_AFTER(&pq->pq_pl, m, &vmd->vmd_marker, plinks.q); vm_page_dequeue_locked(m); vm_pagequeue_unlock(pq); queue_locked = FALSE; /* * Invalid pages can be easily freed. They cannot be * mapped, vm_page_free() asserts this. */ if (m->valid == 0) goto free_page; /* * If the page has been referenced and the object is not dead, * reactivate or requeue the page depending on whether the * object is mapped. */ if ((m->aflags & PGA_REFERENCED) != 0) { vm_page_aflag_clear(m, PGA_REFERENCED); act_delta = 1; } else act_delta = 0; if (object->ref_count != 0) { act_delta += pmap_ts_referenced(m); } else { KASSERT(!pmap_page_is_mapped(m), ("vm_pageout_scan: page %p is mapped", m)); } if (act_delta != 0) { if (object->ref_count != 0) { PCPU_INC(cnt.v_reactivated); vm_page_activate(m); /* * Increase the activation count if the page * was referenced while in the inactive queue. * This makes it less likely that the page will * be returned prematurely to the inactive * queue. */ m->act_count += act_delta + ACT_ADVANCE; goto drop_page; } else if ((object->flags & OBJ_DEAD) == 0) { vm_pagequeue_lock(pq); queue_locked = TRUE; m->queue = PQ_INACTIVE; TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_inc(pq); goto drop_page; } } /* * If the page appears to be clean at the machine-independent * layer, then remove all of its mappings from the pmap in * anticipation of freeing it. If, however, any of the page's * mappings allow write access, then the page may still be * modified until the last of those mappings are removed. */ if (object->ref_count != 0) { vm_page_test_dirty(m); if (m->dirty == 0) pmap_remove_all(m); } /* * Clean pages can be freed, but dirty pages must be sent back * to the laundry, unless they belong to a dead object. * Requeueing dirty pages from dead objects is pointless, as * they are being paged out and freed by the thread that * destroyed the object. */ if (m->dirty == 0) { free_page: vm_page_free(m); PCPU_INC(cnt.v_dfree); --page_shortage; } else if ((object->flags & OBJ_DEAD) == 0) vm_page_launder(m); drop_page: vm_page_unlock(m); VM_OBJECT_WUNLOCK(object); if (!queue_locked) { vm_pagequeue_lock(pq); queue_locked = TRUE; } next = TAILQ_NEXT(&vmd->vmd_marker, plinks.q); TAILQ_REMOVE(&pq->pq_pl, &vmd->vmd_marker, plinks.q); } vm_pagequeue_unlock(pq); /* * Wake up the laundry thread so that it can perform any needed * laundering. If we didn't meet our target, we're in shortfall and * need to launder more aggressively. */ if (vm_laundry_request == VM_LAUNDRY_IDLE && starting_page_shortage > 0) { pq = &vm_dom[0].vmd_pagequeues[PQ_LAUNDRY]; vm_pagequeue_lock(pq); if (page_shortage > 0) { vm_laundry_request = VM_LAUNDRY_SHORTFALL; PCPU_INC(cnt.v_pdshortfalls); } else if (vm_laundry_request != VM_LAUNDRY_SHORTFALL) vm_laundry_request = VM_LAUNDRY_BACKGROUND; wakeup(&vm_laundry_request); vm_pagequeue_unlock(pq); } #if !defined(NO_SWAPPING) /* * Wakeup the swapout daemon if we didn't free the targeted number of * pages. */ if (vm_swap_enabled && page_shortage > 0) vm_req_vmdaemon(VM_SWAP_NORMAL); #endif /* * If the inactive queue scan fails repeatedly to meet its * target, kill the largest process. */ vm_pageout_mightbe_oom(vmd, page_shortage, starting_page_shortage); /* * Compute the number of pages we want to try to move from the * active queue to either the inactive or laundry queue. * * When scanning active pages, we make clean pages count more heavily * towards the page shortage than dirty pages. This is because dirty * pages must be laundered before they can be reused and thus have less * utility when attempting to quickly alleviate a shortage. However, * this weighting also causes the scan to deactivate dirty pages more * more aggressively, improving the effectiveness of clustering and * ensuring that they can eventually be reused. */ inactq_shortage = vm_cnt.v_inactive_target - (vm_cnt.v_inactive_count + vm_cnt.v_laundry_count / act_scan_laundry_weight) + vm_paging_target() + deficit + addl_page_shortage; page_shortage *= act_scan_laundry_weight; pq = &vmd->vmd_pagequeues[PQ_ACTIVE]; vm_pagequeue_lock(pq); maxscan = pq->pq_cnt; /* * If we're just idle polling attempt to visit every * active page within 'update_period' seconds. */ scan_tick = ticks; if (vm_pageout_update_period != 0) { min_scan = pq->pq_cnt; min_scan *= scan_tick - vmd->vmd_last_active_scan; min_scan /= hz * vm_pageout_update_period; } else min_scan = 0; if (min_scan > 0 || (inactq_shortage > 0 && maxscan > 0)) vmd->vmd_last_active_scan = scan_tick; /* * Scan the active queue for pages that can be deactivated. Update * the per-page activity counter and use it to identify deactivation * candidates. Held pages may be deactivated. */ for (m = TAILQ_FIRST(&pq->pq_pl), scanned = 0; m != NULL && (scanned < min_scan || (inactq_shortage > 0 && scanned < maxscan)); m = next, scanned++) { KASSERT(m->queue == PQ_ACTIVE, ("vm_pageout_scan: page %p isn't active", m)); next = TAILQ_NEXT(m, plinks.q); if ((m->flags & PG_MARKER) != 0) continue; KASSERT((m->flags & PG_FICTITIOUS) == 0, ("Fictitious page %p cannot be in active queue", m)); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("Unmanaged page %p cannot be in active queue", m)); if (!vm_pageout_page_lock(m, &next)) { vm_page_unlock(m); continue; } /* * The count for page daemon pages is updated after checking * the page for eligibility. */ PCPU_INC(cnt.v_pdpages); /* * Check to see "how much" the page has been used. */ if ((m->aflags & PGA_REFERENCED) != 0) { vm_page_aflag_clear(m, PGA_REFERENCED); act_delta = 1; } else act_delta = 0; /* * Perform an unsynchronized object ref count check. While * the page lock ensures that the page is not reallocated to * another object, in particular, one with unmanaged mappings * that cannot support pmap_ts_referenced(), two races are, * nonetheless, possible: * 1) The count was transitioning to zero, but we saw a non- * zero value. pmap_ts_referenced() will return zero * because the page is not mapped. * 2) The count was transitioning to one, but we saw zero. * This race delays the detection of a new reference. At * worst, we will deactivate and reactivate the page. */ if (m->object->ref_count != 0) act_delta += pmap_ts_referenced(m); /* * Advance or decay the act_count based on recent usage. */ if (act_delta != 0) { m->act_count += ACT_ADVANCE + act_delta; if (m->act_count > ACT_MAX) m->act_count = ACT_MAX; } else m->act_count -= min(m->act_count, ACT_DECLINE); /* * Move this page to the tail of the active, inactive or laundry * queue depending on usage. */ if (m->act_count == 0) { /* Dequeue to avoid later lock recursion. */ vm_page_dequeue_locked(m); /* * When not short for inactive pages, let dirty pages go * through the inactive queue before moving to the * laundry queues. This gives them some extra time to * be reactivated, potentially avoiding an expensive * pageout. During a page shortage, the inactive queue * is necessarily small, so we may move dirty pages * directly to the laundry queue. */ if (inactq_shortage <= 0) vm_page_deactivate(m); else { /* * Calling vm_page_test_dirty() here would * require acquisition of the object's write * lock. However, during a page shortage, * directing dirty pages into the laundry * queue is only an optimization and not a * requirement. Therefore, we simply rely on * the opportunistic updates to the page's * dirty field by the pmap. */ if (m->dirty == 0) { vm_page_deactivate(m); inactq_shortage -= act_scan_laundry_weight; } else { vm_page_launder(m); inactq_shortage--; } } } else vm_page_requeue_locked(m); vm_page_unlock(m); } vm_pagequeue_unlock(pq); #if !defined(NO_SWAPPING) /* * Idle process swapout -- run once per second when we are reclaiming * pages. */ if (vm_swap_idle_enabled && pass > 0) { static long lsec; if (time_second != lsec) { vm_req_vmdaemon(VM_SWAP_IDLE); lsec = time_second; } } #endif return (page_shortage <= 0); } static int vm_pageout_oom_vote; /* * The pagedaemon threads randlomly select one to perform the * OOM. Trying to kill processes before all pagedaemons * failed to reach free target is premature. */ static void vm_pageout_mightbe_oom(struct vm_domain *vmd, int page_shortage, int starting_page_shortage) { int old_vote; if (starting_page_shortage <= 0 || starting_page_shortage != page_shortage) vmd->vmd_oom_seq = 0; else vmd->vmd_oom_seq++; if (vmd->vmd_oom_seq < vm_pageout_oom_seq) { if (vmd->vmd_oom) { vmd->vmd_oom = FALSE; atomic_subtract_int(&vm_pageout_oom_vote, 1); } return; } /* * Do not follow the call sequence until OOM condition is * cleared. */ vmd->vmd_oom_seq = 0; if (vmd->vmd_oom) return; vmd->vmd_oom = TRUE; old_vote = atomic_fetchadd_int(&vm_pageout_oom_vote, 1); if (old_vote != vm_ndomains - 1) return; /* * The current pagedaemon thread is the last in the quorum to * start OOM. Initiate the selection and signaling of the * victim. */ vm_pageout_oom(VM_OOM_MEM); /* * After one round of OOM terror, recall our vote. On the * next pass, current pagedaemon would vote again if the low * memory condition is still there, due to vmd_oom being * false. */ vmd->vmd_oom = FALSE; atomic_subtract_int(&vm_pageout_oom_vote, 1); } /* * The OOM killer is the page daemon's action of last resort when * memory allocation requests have been stalled for a prolonged period * of time because it cannot reclaim memory. This function computes * the approximate number of physical pages that could be reclaimed if * the specified address space is destroyed. * * Private, anonymous memory owned by the address space is the * principal resource that we expect to recover after an OOM kill. * Since the physical pages mapped by the address space's COW entries * are typically shared pages, they are unlikely to be released and so * they are not counted. * * To get to the point where the page daemon runs the OOM killer, its * efforts to write-back vnode-backed pages may have stalled. This * could be caused by a memory allocation deadlock in the write path * that might be resolved by an OOM kill. Therefore, physical pages * belonging to vnode-backed objects are counted, because they might * be freed without being written out first if the address space holds * the last reference to an unlinked vnode. * * Similarly, physical pages belonging to OBJT_PHYS objects are * counted because the address space might hold the last reference to * the object. */ static long vm_pageout_oom_pagecount(struct vmspace *vmspace) { vm_map_t map; vm_map_entry_t entry; vm_object_t obj; long res; map = &vmspace->vm_map; KASSERT(!map->system_map, ("system map")); sx_assert(&map->lock, SA_LOCKED); res = 0; for (entry = map->header.next; entry != &map->header; entry = entry->next) { if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) continue; obj = entry->object.vm_object; if (obj == NULL) continue; if ((entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0 && obj->ref_count != 1) continue; switch (obj->type) { case OBJT_DEFAULT: case OBJT_SWAP: case OBJT_PHYS: case OBJT_VNODE: res += obj->resident_page_count; break; } } return (res); } void vm_pageout_oom(int shortage) { struct proc *p, *bigproc; vm_offset_t size, bigsize; struct thread *td; struct vmspace *vm; /* * We keep the process bigproc locked once we find it to keep anyone * from messing with it; however, there is a possibility of * deadlock if process B is bigproc and one of its child processes * attempts to propagate a signal to B while we are waiting for A's * lock while walking this list. To avoid this, we don't block on * the process lock but just skip a process if it is already locked. */ bigproc = NULL; bigsize = 0; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { int breakout; PROC_LOCK(p); /* * If this is a system, protected or killed process, skip it. */ if (p->p_state != PRS_NORMAL || (p->p_flag & (P_INEXEC | P_PROTECTED | P_SYSTEM | P_WEXIT)) != 0 || p->p_pid == 1 || P_KILLED(p) || (p->p_pid < 48 && swap_pager_avail != 0)) { PROC_UNLOCK(p); continue; } /* * If the process is in a non-running type state, * don't touch it. Check all the threads individually. */ breakout = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (!TD_ON_RUNQ(td) && !TD_IS_RUNNING(td) && !TD_IS_SLEEPING(td) && !TD_IS_SUSPENDED(td) && !TD_IS_SWAPPED(td)) { thread_unlock(td); breakout = 1; break; } thread_unlock(td); } if (breakout) { PROC_UNLOCK(p); continue; } /* * get the process size */ vm = vmspace_acquire_ref(p); if (vm == NULL) { PROC_UNLOCK(p); continue; } _PHOLD_LITE(p); PROC_UNLOCK(p); sx_sunlock(&allproc_lock); if (!vm_map_trylock_read(&vm->vm_map)) { vmspace_free(vm); sx_slock(&allproc_lock); PRELE(p); continue; } size = vmspace_swap_count(vm); if (shortage == VM_OOM_MEM) size += vm_pageout_oom_pagecount(vm); vm_map_unlock_read(&vm->vm_map); vmspace_free(vm); sx_slock(&allproc_lock); /* * If this process is bigger than the biggest one, * remember it. */ if (size > bigsize) { if (bigproc != NULL) PRELE(bigproc); bigproc = p; bigsize = size; } else { PRELE(p); } } sx_sunlock(&allproc_lock); if (bigproc != NULL) { if (vm_panic_on_oom != 0) panic("out of swap space"); PROC_LOCK(bigproc); killproc(bigproc, "out of swap space"); sched_nice(bigproc, PRIO_MIN); _PRELE(bigproc); PROC_UNLOCK(bigproc); wakeup(&vm_cnt.v_free_count); } } static void vm_pageout_worker(void *arg) { struct vm_domain *domain; int domidx, pass; bool target_met; domidx = (uintptr_t)arg; domain = &vm_dom[domidx]; pass = 0; target_met = true; /* * XXXKIB It could be useful to bind pageout daemon threads to * the cores belonging to the domain, from which vm_page_array * is allocated. */ KASSERT(domain->vmd_segs != 0, ("domain without segments")); domain->vmd_last_active_scan = ticks; vm_pageout_init_marker(&domain->vmd_marker, PQ_INACTIVE); vm_pageout_init_marker(&domain->vmd_inacthead, PQ_INACTIVE); TAILQ_INSERT_HEAD(&domain->vmd_pagequeues[PQ_INACTIVE].pq_pl, &domain->vmd_inacthead, plinks.q); /* * The pageout daemon worker is never done, so loop forever. */ while (TRUE) { mtx_lock(&vm_page_queue_free_mtx); /* * Generally, after a level >= 1 scan, if there are enough * free pages to wakeup the waiters, then they are already * awake. A call to vm_page_free() during the scan awakened * them. However, in the following case, this wakeup serves * to bound the amount of time that a thread might wait. * Suppose a thread's call to vm_page_alloc() fails, but * before that thread calls VM_WAIT, enough pages are freed by * other threads to alleviate the free page shortage. The * thread will, nonetheless, wait until another page is freed * or this wakeup is performed. */ if (vm_pages_needed && !vm_page_count_min()) { vm_pages_needed = false; wakeup(&vm_cnt.v_free_count); } /* * Do not clear vm_pageout_wanted until we reach our free page * target. Otherwise, we may be awakened over and over again, * wasting CPU time. */ if (vm_pageout_wanted && target_met) vm_pageout_wanted = false; /* * Might the page daemon receive a wakeup call? */ if (vm_pageout_wanted) { /* * No. Either vm_pageout_wanted was set by another * thread during the previous scan, which must have * been a level 0 scan, or vm_pageout_wanted was * already set and the scan failed to free enough * pages. If we haven't yet performed a level >= 1 * (page reclamation) scan, then increase the level * and scan again now. Otherwise, sleep a bit and * try again later. */ mtx_unlock(&vm_page_queue_free_mtx); if (pass >= 1) pause("psleep", hz / VM_INACT_SCAN_RATE); pass++; } else { /* * Yes. Sleep until pages need to be reclaimed or * have their reference stats updated. */ if (mtx_sleep(&vm_pageout_wanted, &vm_page_queue_free_mtx, PDROP | PVM, "psleep", hz) == 0) { PCPU_INC(cnt.v_pdwakeups); pass = 1; } else pass = 0; } target_met = vm_pageout_scan(domain, pass); } } /* * vm_pageout_init initialises basic pageout daemon settings. */ static void vm_pageout_init(void) { /* * Initialize some paging parameters. */ vm_cnt.v_interrupt_free_min = 2; if (vm_cnt.v_page_count < 2000) vm_pageout_page_count = 8; /* * v_free_reserved needs to include enough for the largest * swap pager structures plus enough for any pv_entry structs * when paging. */ if (vm_cnt.v_page_count > 1024) vm_cnt.v_free_min = 4 + (vm_cnt.v_page_count - 1024) / 200; else vm_cnt.v_free_min = 4; vm_cnt.v_pageout_free_min = (2*MAXBSIZE)/PAGE_SIZE + vm_cnt.v_interrupt_free_min; vm_cnt.v_free_reserved = vm_pageout_page_count + vm_cnt.v_pageout_free_min + (vm_cnt.v_page_count / 768); vm_cnt.v_free_severe = vm_cnt.v_free_min / 2; vm_cnt.v_free_target = 4 * vm_cnt.v_free_min + vm_cnt.v_free_reserved; vm_cnt.v_free_min += vm_cnt.v_free_reserved; vm_cnt.v_free_severe += vm_cnt.v_free_reserved; vm_cnt.v_inactive_target = (3 * vm_cnt.v_free_target) / 2; if (vm_cnt.v_inactive_target > vm_cnt.v_free_count / 3) vm_cnt.v_inactive_target = vm_cnt.v_free_count / 3; /* * Set the default wakeup threshold to be 10% above the minimum * page limit. This keeps the steady state out of shortfall. */ vm_pageout_wakeup_thresh = (vm_cnt.v_free_min / 10) * 11; /* * Set interval in seconds for active scan. We want to visit each * page at least once every ten minutes. This is to prevent worst * case paging behaviors with stale active LRU. */ if (vm_pageout_update_period == 0) vm_pageout_update_period = 600; /* XXX does not really belong here */ if (vm_page_max_wired == 0) vm_page_max_wired = vm_cnt.v_free_count / 3; /* * Target amount of memory to move out of the laundry queue during a * background laundering. This is proportional to the amount of system * memory. */ vm_background_launder_target = (vm_cnt.v_free_target - vm_cnt.v_free_min) / 10; } /* * vm_pageout is the high level pageout daemon. */ static void vm_pageout(void) { int error; #ifdef VM_NUMA_ALLOC int i; #endif swap_pager_swap_init(); error = kthread_add(vm_pageout_laundry_worker, NULL, curproc, NULL, 0, 0, "laundry: dom0"); if (error != 0) panic("starting laundry for domain 0, error %d", error); #ifdef VM_NUMA_ALLOC for (i = 1; i < vm_ndomains; i++) { error = kthread_add(vm_pageout_worker, (void *)(uintptr_t)i, curproc, NULL, 0, 0, "dom%d", i); if (error != 0) { panic("starting pageout for domain %d, error %d\n", i, error); } } #endif error = kthread_add(uma_reclaim_worker, NULL, curproc, NULL, 0, 0, "uma"); if (error != 0) panic("starting uma_reclaim helper, error %d\n", error); vm_pageout_worker((void *)(uintptr_t)0); } /* * Unless the free page queue lock is held by the caller, this function * should be regarded as advisory. Specifically, the caller should * not msleep() on &vm_cnt.v_free_count following this function unless * the free page queue lock is held until the msleep() is performed. */ void pagedaemon_wakeup(void) { if (!vm_pageout_wanted && curthread->td_proc != pageproc) { vm_pageout_wanted = true; wakeup(&vm_pageout_wanted); } } #if !defined(NO_SWAPPING) static void vm_req_vmdaemon(int req) { static int lastrun = 0; mtx_lock(&vm_daemon_mtx); vm_pageout_req_swapout |= req; if ((ticks > (lastrun + hz)) || (ticks < lastrun)) { wakeup(&vm_daemon_needed); lastrun = ticks; } mtx_unlock(&vm_daemon_mtx); } static void vm_daemon(void) { struct rlimit rsslim; struct proc *p; struct thread *td; struct vmspace *vm; int breakout, swapout_flags, tryagain, attempts; #ifdef RACCT uint64_t rsize, ravailable; #endif while (TRUE) { mtx_lock(&vm_daemon_mtx); msleep(&vm_daemon_needed, &vm_daemon_mtx, PPAUSE, "psleep", #ifdef RACCT racct_enable ? hz : 0 #else 0 #endif ); swapout_flags = vm_pageout_req_swapout; vm_pageout_req_swapout = 0; mtx_unlock(&vm_daemon_mtx); if (swapout_flags) swapout_procs(swapout_flags); /* * scan the processes for exceeding their rlimits or if * process is swapped out -- deactivate pages */ tryagain = 0; attempts = 0; again: attempts++; sx_slock(&allproc_lock); FOREACH_PROC_IN_SYSTEM(p) { vm_pindex_t limit, size; /* * if this is a system process or if we have already * looked at this process, skip it. */ PROC_LOCK(p); if (p->p_state != PRS_NORMAL || p->p_flag & (P_INEXEC | P_SYSTEM | P_WEXIT)) { PROC_UNLOCK(p); continue; } /* * if the process is in a non-running type state, * don't touch it. */ breakout = 0; FOREACH_THREAD_IN_PROC(p, td) { thread_lock(td); if (!TD_ON_RUNQ(td) && !TD_IS_RUNNING(td) && !TD_IS_SLEEPING(td) && !TD_IS_SUSPENDED(td)) { thread_unlock(td); breakout = 1; break; } thread_unlock(td); } if (breakout) { PROC_UNLOCK(p); continue; } /* * get a limit */ lim_rlimit_proc(p, RLIMIT_RSS, &rsslim); limit = OFF_TO_IDX( qmin(rsslim.rlim_cur, rsslim.rlim_max)); /* * let processes that are swapped out really be * swapped out set the limit to nothing (will force a * swap-out.) */ if ((p->p_flag & P_INMEM) == 0) limit = 0; /* XXX */ vm = vmspace_acquire_ref(p); _PHOLD_LITE(p); PROC_UNLOCK(p); if (vm == NULL) { PRELE(p); continue; } sx_sunlock(&allproc_lock); size = vmspace_resident_count(vm); if (size >= limit) { vm_pageout_map_deactivate_pages( &vm->vm_map, limit); } #ifdef RACCT if (racct_enable) { rsize = IDX_TO_OFF(size); PROC_LOCK(p); racct_set(p, RACCT_RSS, rsize); ravailable = racct_get_available(p, RACCT_RSS); PROC_UNLOCK(p); if (rsize > ravailable) { /* * Don't be overly aggressive; this * might be an innocent process, * and the limit could've been exceeded * by some memory hog. Don't try * to deactivate more than 1/4th * of process' resident set size. */ if (attempts <= 8) { if (ravailable < rsize - (rsize / 4)) { ravailable = rsize - (rsize / 4); } } vm_pageout_map_deactivate_pages( &vm->vm_map, OFF_TO_IDX(ravailable)); /* Update RSS usage after paging out. */ size = vmspace_resident_count(vm); rsize = IDX_TO_OFF(size); PROC_LOCK(p); racct_set(p, RACCT_RSS, rsize); PROC_UNLOCK(p); if (rsize > ravailable) tryagain = 1; } } #endif vmspace_free(vm); sx_slock(&allproc_lock); PRELE(p); } sx_sunlock(&allproc_lock); if (tryagain != 0 && attempts <= 10) goto again; } } #endif /* !defined(NO_SWAPPING) */ Index: head/sys/vm/vm_pager.h =================================================================== --- head/sys/vm/vm_pager.h (revision 309061) +++ head/sys/vm/vm_pager.h (revision 309062) @@ -1,191 +1,192 @@ /*- * Copyright (c) 1990 University of Utah. * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * 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. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vm_pager.h 8.4 (Berkeley) 1/12/94 * $FreeBSD$ */ /* * Pager routine interface definition. */ #ifndef _VM_PAGER_ #define _VM_PAGER_ #include TAILQ_HEAD(pagerlst, vm_object); typedef void pgo_init_t(void); typedef vm_object_t pgo_alloc_t(void *, vm_ooffset_t, vm_prot_t, vm_ooffset_t, struct ucred *); typedef void pgo_dealloc_t(vm_object_t); typedef int pgo_getpages_t(vm_object_t, vm_page_t *, int, int *, int *); typedef void pgo_getpages_iodone_t(void *, vm_page_t *, int, int); typedef int pgo_getpages_async_t(vm_object_t, vm_page_t *, int, int *, int *, pgo_getpages_iodone_t, void *); typedef void pgo_putpages_t(vm_object_t, vm_page_t *, int, int, int *); typedef boolean_t pgo_haspage_t(vm_object_t, vm_pindex_t, int *, int *); typedef void pgo_pageunswapped_t(vm_page_t); struct pagerops { pgo_init_t *pgo_init; /* Initialize pager. */ pgo_alloc_t *pgo_alloc; /* Allocate pager. */ pgo_dealloc_t *pgo_dealloc; /* Disassociate. */ pgo_getpages_t *pgo_getpages; /* Get (read) page. */ pgo_getpages_async_t *pgo_getpages_async; /* Get page asyncly. */ pgo_putpages_t *pgo_putpages; /* Put (write) page. */ pgo_haspage_t *pgo_haspage; /* Query page. */ pgo_pageunswapped_t *pgo_pageunswapped; }; extern struct pagerops defaultpagerops; extern struct pagerops swappagerops; extern struct pagerops vnodepagerops; extern struct pagerops devicepagerops; extern struct pagerops physpagerops; extern struct pagerops sgpagerops; extern struct pagerops mgtdevicepagerops; /* * get/put return values * OK operation was successful * BAD specified data was out of the accepted range * FAIL specified data was in range, but doesn't exist * PEND operations was initiated but not completed * ERROR error while accessing data that is in range and exists * AGAIN temporary resource shortage prevented operation from happening */ #define VM_PAGER_OK 0 #define VM_PAGER_BAD 1 #define VM_PAGER_FAIL 2 #define VM_PAGER_PEND 3 #define VM_PAGER_ERROR 4 #define VM_PAGER_AGAIN 5 #define VM_PAGER_PUT_SYNC 0x0001 #define VM_PAGER_PUT_INVAL 0x0002 +#define VM_PAGER_PUT_NOREUSE 0x0004 #define VM_PAGER_CLUSTER_OK 0x0008 #ifdef _KERNEL extern struct pagerops *pagertab[]; extern struct mtx_padalign pbuf_mtx; vm_object_t vm_pager_allocate(objtype_t, void *, vm_ooffset_t, vm_prot_t, vm_ooffset_t, struct ucred *); void vm_pager_bufferinit(void); void vm_pager_deallocate(vm_object_t); int vm_pager_get_pages(vm_object_t, vm_page_t *, int, int *, int *); int vm_pager_get_pages_async(vm_object_t, vm_page_t *, int, int *, int *, pgo_getpages_iodone_t, void *); void vm_pager_init(void); vm_object_t vm_pager_object_lookup(struct pagerlst *, void *); static __inline void vm_pager_put_pages( vm_object_t object, vm_page_t *m, int count, int flags, int *rtvals ) { VM_OBJECT_ASSERT_WLOCKED(object); (*pagertab[object->type]->pgo_putpages) (object, m, count, flags, rtvals); } /* * vm_pager_haspage * * Check to see if an object's pager has the requested page. The * object's pager will also set before and after to give the caller * some idea of the number of pages before and after the requested * page can be I/O'd efficiently. * * The object must be locked. */ static __inline boolean_t vm_pager_has_page( vm_object_t object, vm_pindex_t offset, int *before, int *after ) { boolean_t ret; VM_OBJECT_ASSERT_WLOCKED(object); ret = (*pagertab[object->type]->pgo_haspage) (object, offset, before, after); return (ret); } /* * vm_pager_page_unswapped * * Destroy swap associated with the page. * * The object containing the page must be locked. * This function may not block. * * XXX: A much better name would be "vm_pager_page_dirtied()" * XXX: It is not obvious if this could be profitably used by any * XXX: pagers besides the swap_pager or if it should even be a * XXX: generic pager_op in the first place. */ static __inline void vm_pager_page_unswapped(vm_page_t m) { VM_OBJECT_ASSERT_LOCKED(m->object); if (pagertab[m->object->type]->pgo_pageunswapped) (*pagertab[m->object->type]->pgo_pageunswapped)(m); } struct cdev_pager_ops { int (*cdev_pg_fault)(vm_object_t vm_obj, vm_ooffset_t offset, int prot, vm_page_t *mres); int (*cdev_pg_ctor)(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred, u_short *color); void (*cdev_pg_dtor)(void *handle); }; vm_object_t cdev_pager_allocate(void *handle, enum obj_type tp, struct cdev_pager_ops *ops, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t foff, struct ucred *cred); vm_object_t cdev_pager_lookup(void *handle); void cdev_pager_free_page(vm_object_t object, vm_page_t m); #endif /* _KERNEL */ #endif /* _VM_PAGER_ */ Index: head/sys/vm/vnode_pager.c =================================================================== --- head/sys/vm/vnode_pager.c (revision 309061) +++ head/sys/vm/vnode_pager.c (revision 309062) @@ -1,1413 +1,1414 @@ /*- * Copyright (c) 1990 University of Utah. * Copyright (c) 1991 The Regents of the University of California. * All rights reserved. * Copyright (c) 1993, 1994 John S. Dyson * Copyright (c) 1995, David Greenman * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vnode_pager.c 7.5 (Berkeley) 4/20/91 */ /* * Page to/from files (vnodes). */ /* * TODO: * Implement VOP_GETPAGES/PUTPAGES interface for filesystems. Will * greatly re-simplify the vnode_pager. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress, int *run); static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m); static int vnode_pager_input_old(vm_object_t object, vm_page_t m); static void vnode_pager_dealloc(vm_object_t); static int vnode_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); static int vnode_pager_getpages_async(vm_object_t, vm_page_t *, int, int *, int *, vop_getpages_iodone_t, void *); static void vnode_pager_putpages(vm_object_t, vm_page_t *, int, int, int *); static boolean_t vnode_pager_haspage(vm_object_t, vm_pindex_t, int *, int *); static vm_object_t vnode_pager_alloc(void *, vm_ooffset_t, vm_prot_t, vm_ooffset_t, struct ucred *cred); static int vnode_pager_generic_getpages_done(struct buf *); static void vnode_pager_generic_getpages_done_async(struct buf *); struct pagerops vnodepagerops = { .pgo_alloc = vnode_pager_alloc, .pgo_dealloc = vnode_pager_dealloc, .pgo_getpages = vnode_pager_getpages, .pgo_getpages_async = vnode_pager_getpages_async, .pgo_putpages = vnode_pager_putpages, .pgo_haspage = vnode_pager_haspage, }; int vnode_pbuf_freecnt; int vnode_async_pbuf_freecnt; /* Create the VM system backing object for this vnode */ int vnode_create_vobject(struct vnode *vp, off_t isize, struct thread *td) { vm_object_t object; vm_ooffset_t size = isize; struct vattr va; if (!vn_isdisk(vp, NULL) && vn_canvmio(vp) == FALSE) return (0); while ((object = vp->v_object) != NULL) { VM_OBJECT_WLOCK(object); if (!(object->flags & OBJ_DEAD)) { VM_OBJECT_WUNLOCK(object); return (0); } VOP_UNLOCK(vp, 0); vm_object_set_flag(object, OBJ_DISCONNECTWNT); VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vodead", 0); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } if (size == 0) { if (vn_isdisk(vp, NULL)) { size = IDX_TO_OFF(INT_MAX); } else { if (VOP_GETATTR(vp, &va, td->td_ucred)) return (0); size = va.va_size; } } object = vnode_pager_alloc(vp, size, 0, 0, td->td_ucred); /* * Dereference the reference we just created. This assumes * that the object is associated with the vp. */ VM_OBJECT_WLOCK(object); object->ref_count--; VM_OBJECT_WUNLOCK(object); vrele(vp); KASSERT(vp->v_object != NULL, ("vnode_create_vobject: NULL object")); return (0); } void vnode_destroy_vobject(struct vnode *vp) { struct vm_object *obj; obj = vp->v_object; if (obj == NULL) return; ASSERT_VOP_ELOCKED(vp, "vnode_destroy_vobject"); VM_OBJECT_WLOCK(obj); umtx_shm_object_terminated(obj); if (obj->ref_count == 0) { /* * don't double-terminate the object */ if ((obj->flags & OBJ_DEAD) == 0) { vm_object_terminate(obj); } else { /* * Waiters were already handled during object * termination. The exclusive vnode lock hopefully * prevented new waiters from referencing the dying * object. */ KASSERT((obj->flags & OBJ_DISCONNECTWNT) == 0, ("OBJ_DISCONNECTWNT set obj %p flags %x", obj, obj->flags)); vp->v_object = NULL; VM_OBJECT_WUNLOCK(obj); } } else { /* * Woe to the process that tries to page now :-). */ vm_pager_deallocate(obj); VM_OBJECT_WUNLOCK(obj); } KASSERT(vp->v_object == NULL, ("vp %p obj %p", vp, vp->v_object)); } /* * Allocate (or lookup) pager for a vnode. * Handle is a vnode pointer. * * MPSAFE */ vm_object_t vnode_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *cred) { vm_object_t object; struct vnode *vp; /* * Pageout to vnode, no can do yet. */ if (handle == NULL) return (NULL); vp = (struct vnode *) handle; /* * If the object is being terminated, wait for it to * go away. */ retry: while ((object = vp->v_object) != NULL) { VM_OBJECT_WLOCK(object); if ((object->flags & OBJ_DEAD) == 0) break; vm_object_set_flag(object, OBJ_DISCONNECTWNT); VM_OBJECT_SLEEP(object, object, PDROP | PVM, "vadead", 0); } KASSERT(vp->v_usecount != 0, ("vnode_pager_alloc: no vnode reference")); if (object == NULL) { /* * Add an object of the appropriate size */ object = vm_object_allocate(OBJT_VNODE, OFF_TO_IDX(round_page(size))); object->un_pager.vnp.vnp_size = size; object->un_pager.vnp.writemappings = 0; object->handle = handle; VI_LOCK(vp); if (vp->v_object != NULL) { /* * Object has been created while we were sleeping */ VI_UNLOCK(vp); VM_OBJECT_WLOCK(object); KASSERT(object->ref_count == 1, ("leaked ref %p %d", object, object->ref_count)); object->type = OBJT_DEAD; object->ref_count = 0; VM_OBJECT_WUNLOCK(object); vm_object_destroy(object); goto retry; } vp->v_object = object; VI_UNLOCK(vp); } else { object->ref_count++; #if VM_NRESERVLEVEL > 0 vm_object_color(object, 0); #endif VM_OBJECT_WUNLOCK(object); } vref(vp); return (object); } /* * The object must be locked. */ static void vnode_pager_dealloc(vm_object_t object) { struct vnode *vp; int refs; vp = object->handle; if (vp == NULL) panic("vnode_pager_dealloc: pager already dealloced"); VM_OBJECT_ASSERT_WLOCKED(object); vm_object_pip_wait(object, "vnpdea"); refs = object->ref_count; object->handle = NULL; object->type = OBJT_DEAD; if (object->flags & OBJ_DISCONNECTWNT) { vm_object_clear_flag(object, OBJ_DISCONNECTWNT); wakeup(object); } ASSERT_VOP_ELOCKED(vp, "vnode_pager_dealloc"); if (object->un_pager.vnp.writemappings > 0) { object->un_pager.vnp.writemappings = 0; VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } vp->v_object = NULL; VOP_UNSET_TEXT(vp); VM_OBJECT_WUNLOCK(object); while (refs-- > 0) vunref(vp); VM_OBJECT_WLOCK(object); } static boolean_t vnode_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after) { struct vnode *vp = object->handle; daddr_t bn; int err; daddr_t reqblock; int poff; int bsize; int pagesperblock, blocksperpage; VM_OBJECT_ASSERT_WLOCKED(object); /* * If no vp or vp is doomed or marked transparent to VM, we do not * have the page. */ if (vp == NULL || vp->v_iflag & VI_DOOMED) return FALSE; /* * If the offset is beyond end of file we do * not have the page. */ if (IDX_TO_OFF(pindex) >= object->un_pager.vnp.vnp_size) return FALSE; bsize = vp->v_mount->mnt_stat.f_iosize; pagesperblock = bsize / PAGE_SIZE; blocksperpage = 0; if (pagesperblock > 0) { reqblock = pindex / pagesperblock; } else { blocksperpage = (PAGE_SIZE / bsize); reqblock = pindex * blocksperpage; } VM_OBJECT_WUNLOCK(object); err = VOP_BMAP(vp, reqblock, NULL, &bn, after, before); VM_OBJECT_WLOCK(object); if (err) return TRUE; if (bn == -1) return FALSE; if (pagesperblock > 0) { poff = pindex - (reqblock * pagesperblock); if (before) { *before *= pagesperblock; *before += poff; } if (after) { /* * The BMAP vop can report a partial block in the * 'after', but must not report blocks after EOF. * Assert the latter, and truncate 'after' in case * of the former. */ KASSERT((reqblock + *after) * pagesperblock < roundup2(object->size, pagesperblock), ("%s: reqblock %jd after %d size %ju", __func__, (intmax_t )reqblock, *after, (uintmax_t )object->size)); *after *= pagesperblock; *after += pagesperblock - (poff + 1); if (pindex + *after >= object->size) *after = object->size - 1 - pindex; } } else { if (before) { *before /= blocksperpage; } if (after) { *after /= blocksperpage; } } return TRUE; } /* * Lets the VM system know about a change in size for a file. * We adjust our own internal size and flush any cached pages in * the associated object that are affected by the size change. * * Note: this routine may be invoked as a result of a pager put * operation (possibly at object termination time), so we must be careful. */ void vnode_pager_setsize(struct vnode *vp, vm_ooffset_t nsize) { vm_object_t object; vm_page_t m; vm_pindex_t nobjsize; if ((object = vp->v_object) == NULL) return; /* ASSERT_VOP_ELOCKED(vp, "vnode_pager_setsize and not locked vnode"); */ VM_OBJECT_WLOCK(object); if (object->type == OBJT_DEAD) { VM_OBJECT_WUNLOCK(object); return; } KASSERT(object->type == OBJT_VNODE, ("not vnode-backed object %p", object)); if (nsize == object->un_pager.vnp.vnp_size) { /* * Hasn't changed size */ VM_OBJECT_WUNLOCK(object); return; } nobjsize = OFF_TO_IDX(nsize + PAGE_MASK); if (nsize < object->un_pager.vnp.vnp_size) { /* * File has shrunk. Toss any cached pages beyond the new EOF. */ if (nobjsize < object->size) vm_object_page_remove(object, nobjsize, object->size, 0); /* * this gets rid of garbage at the end of a page that is now * only partially backed by the vnode. * * XXX for some reason (I don't know yet), if we take a * completely invalid page and mark it partially valid * it can screw up NFS reads, so we don't allow the case. */ if ((nsize & PAGE_MASK) && (m = vm_page_lookup(object, OFF_TO_IDX(nsize))) != NULL && m->valid != 0) { int base = (int)nsize & PAGE_MASK; int size = PAGE_SIZE - base; /* * Clear out partial-page garbage in case * the page has been mapped. */ pmap_zero_page_area(m, base, size); /* * Update the valid bits to reflect the blocks that * have been zeroed. Some of these valid bits may * have already been set. */ vm_page_set_valid_range(m, base, size); /* * Round "base" to the next block boundary so that the * dirty bit for a partially zeroed block is not * cleared. */ base = roundup2(base, DEV_BSIZE); /* * Clear out partial-page dirty bits. * * note that we do not clear out the valid * bits. This would prevent bogus_page * replacement from working properly. */ vm_page_clear_dirty(m, base, PAGE_SIZE - base); } } object->un_pager.vnp.vnp_size = nsize; object->size = nobjsize; VM_OBJECT_WUNLOCK(object); } /* * calculate the linear (byte) disk address of specified virtual * file address */ static int vnode_pager_addr(struct vnode *vp, vm_ooffset_t address, daddr_t *rtaddress, int *run) { int bsize; int err; daddr_t vblock; daddr_t voffset; if (address < 0) return -1; if (vp->v_iflag & VI_DOOMED) return -1; bsize = vp->v_mount->mnt_stat.f_iosize; vblock = address / bsize; voffset = address % bsize; err = VOP_BMAP(vp, vblock, NULL, rtaddress, run, NULL); if (err == 0) { if (*rtaddress != -1) *rtaddress += voffset / DEV_BSIZE; if (run) { *run += 1; *run *= bsize/PAGE_SIZE; *run -= voffset/PAGE_SIZE; } } return (err); } /* * small block filesystem vnode pager input */ static int vnode_pager_input_smlfs(vm_object_t object, vm_page_t m) { struct vnode *vp; struct bufobj *bo; struct buf *bp; struct sf_buf *sf; daddr_t fileaddr; vm_offset_t bsize; vm_page_bits_t bits; int error, i; error = 0; vp = object->handle; if (vp->v_iflag & VI_DOOMED) return VM_PAGER_BAD; bsize = vp->v_mount->mnt_stat.f_iosize; VOP_BMAP(vp, 0, &bo, 0, NULL, NULL); sf = sf_buf_alloc(m, 0); for (i = 0; i < PAGE_SIZE / bsize; i++) { vm_ooffset_t address; bits = vm_page_bits(i * bsize, bsize); if (m->valid & bits) continue; address = IDX_TO_OFF(m->pindex) + i * bsize; if (address >= object->un_pager.vnp.vnp_size) { fileaddr = -1; } else { error = vnode_pager_addr(vp, address, &fileaddr, NULL); if (error) break; } if (fileaddr != -1) { bp = getpbuf(&vnode_pbuf_freecnt); /* build a minimal buffer header */ bp->b_iocmd = BIO_READ; bp->b_iodone = bdone; KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); bp->b_rcred = crhold(curthread->td_ucred); bp->b_wcred = crhold(curthread->td_ucred); bp->b_data = (caddr_t)sf_buf_kva(sf) + i * bsize; bp->b_blkno = fileaddr; pbgetbo(bo, bp); bp->b_vp = vp; bp->b_bcount = bsize; bp->b_bufsize = bsize; bp->b_runningbufspace = bp->b_bufsize; atomic_add_long(&runningbufspace, bp->b_runningbufspace); /* do the input */ bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); bwait(bp, PVM, "vnsrd"); if ((bp->b_ioflags & BIO_ERROR) != 0) error = EIO; /* * free the buffer header back to the swap buffer pool */ bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_pbuf_freecnt); if (error) break; } else bzero((caddr_t)sf_buf_kva(sf) + i * bsize, bsize); KASSERT((m->dirty & bits) == 0, ("vnode_pager_input_smlfs: page %p is dirty", m)); VM_OBJECT_WLOCK(object); m->valid |= bits; VM_OBJECT_WUNLOCK(object); } sf_buf_free(sf); if (error) { return VM_PAGER_ERROR; } return VM_PAGER_OK; } /* * old style vnode pager input routine */ static int vnode_pager_input_old(vm_object_t object, vm_page_t m) { struct uio auio; struct iovec aiov; int error; int size; struct sf_buf *sf; struct vnode *vp; VM_OBJECT_ASSERT_WLOCKED(object); error = 0; /* * Return failure if beyond current EOF */ if (IDX_TO_OFF(m->pindex) >= object->un_pager.vnp.vnp_size) { return VM_PAGER_BAD; } else { size = PAGE_SIZE; if (IDX_TO_OFF(m->pindex) + size > object->un_pager.vnp.vnp_size) size = object->un_pager.vnp.vnp_size - IDX_TO_OFF(m->pindex); vp = object->handle; VM_OBJECT_WUNLOCK(object); /* * Allocate a kernel virtual address and initialize so that * we can use VOP_READ/WRITE routines. */ sf = sf_buf_alloc(m, 0); aiov.iov_base = (caddr_t)sf_buf_kva(sf); aiov.iov_len = size; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = IDX_TO_OFF(m->pindex); auio.uio_segflg = UIO_SYSSPACE; auio.uio_rw = UIO_READ; auio.uio_resid = size; auio.uio_td = curthread; error = VOP_READ(vp, &auio, 0, curthread->td_ucred); if (!error) { int count = size - auio.uio_resid; if (count == 0) error = EINVAL; else if (count != PAGE_SIZE) bzero((caddr_t)sf_buf_kva(sf) + count, PAGE_SIZE - count); } sf_buf_free(sf); VM_OBJECT_WLOCK(object); } KASSERT(m->dirty == 0, ("vnode_pager_input_old: page %p is dirty", m)); if (!error) m->valid = VM_PAGE_BITS_ALL; return error ? VM_PAGER_ERROR : VM_PAGER_OK; } /* * generic vnode pager input routine */ /* * Local media VFS's that do not implement their own VOP_GETPAGES * should have their VOP_GETPAGES call to vnode_pager_generic_getpages() * to implement the previous behaviour. * * All other FS's should use the bypass to get to the local media * backing vp's VOP_GETPAGES. */ static int vnode_pager_getpages(vm_object_t object, vm_page_t *m, int count, int *rbehind, int *rahead) { struct vnode *vp; int rtval; vp = object->handle; VM_OBJECT_WUNLOCK(object); rtval = VOP_GETPAGES(vp, m, count, rbehind, rahead); KASSERT(rtval != EOPNOTSUPP, ("vnode_pager: FS getpages not implemented\n")); VM_OBJECT_WLOCK(object); return rtval; } static int vnode_pager_getpages_async(vm_object_t object, vm_page_t *m, int count, int *rbehind, int *rahead, vop_getpages_iodone_t iodone, void *arg) { struct vnode *vp; int rtval; vp = object->handle; VM_OBJECT_WUNLOCK(object); rtval = VOP_GETPAGES_ASYNC(vp, m, count, rbehind, rahead, iodone, arg); KASSERT(rtval != EOPNOTSUPP, ("vnode_pager: FS getpages_async not implemented\n")); VM_OBJECT_WLOCK(object); return (rtval); } /* * The implementation of VOP_GETPAGES() and VOP_GETPAGES_ASYNC() for * local filesystems, where partially valid pages can only occur at * the end of file. */ int vnode_pager_local_getpages(struct vop_getpages_args *ap) { return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, NULL, NULL)); } int vnode_pager_local_getpages_async(struct vop_getpages_async_args *ap) { return (vnode_pager_generic_getpages(ap->a_vp, ap->a_m, ap->a_count, ap->a_rbehind, ap->a_rahead, ap->a_iodone, ap->a_arg)); } /* * This is now called from local media FS's to operate against their * own vnodes if they fail to implement VOP_GETPAGES. */ int vnode_pager_generic_getpages(struct vnode *vp, vm_page_t *m, int count, int *a_rbehind, int *a_rahead, vop_getpages_iodone_t iodone, void *arg) { vm_object_t object; struct bufobj *bo; struct buf *bp; off_t foff; #ifdef INVARIANTS off_t blkno0; #endif int bsize, pagesperblock, *freecnt; int error, before, after, rbehind, rahead, poff, i; int bytecount, secmask; KASSERT(vp->v_type != VCHR && vp->v_type != VBLK, ("%s does not support devices", __func__)); if (vp->v_iflag & VI_DOOMED) return (VM_PAGER_BAD); object = vp->v_object; foff = IDX_TO_OFF(m[0]->pindex); bsize = vp->v_mount->mnt_stat.f_iosize; pagesperblock = bsize / PAGE_SIZE; KASSERT(foff < object->un_pager.vnp.vnp_size, ("%s: page %p offset beyond vp %p size", __func__, m[0], vp)); KASSERT(count <= sizeof(bp->b_pages), ("%s: requested %d pages", __func__, count)); /* * The last page has valid blocks. Invalid part can only * exist at the end of file, and the page is made fully valid * by zeroing in vm_pager_get_pages(). */ if (m[count - 1]->valid != 0 && --count == 0) { if (iodone != NULL) iodone(arg, m, 1, 0); return (VM_PAGER_OK); } /* * Synchronous and asynchronous paging operations use different * free pbuf counters. This is done to avoid asynchronous requests * to consume all pbufs. * Allocate the pbuf at the very beginning of the function, so that * if we are low on certain kind of pbufs don't even proceed to BMAP, * but sleep. */ freecnt = iodone != NULL ? &vnode_async_pbuf_freecnt : &vnode_pbuf_freecnt; bp = getpbuf(freecnt); /* * Get the underlying device blocks for the file with VOP_BMAP(). * If the file system doesn't support VOP_BMAP, use old way of * getting pages via VOP_READ. */ error = VOP_BMAP(vp, foff / bsize, &bo, &bp->b_blkno, &after, &before); if (error == EOPNOTSUPP) { relpbuf(bp, freecnt); VM_OBJECT_WLOCK(object); for (i = 0; i < count; i++) { PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); error = vnode_pager_input_old(object, m[i]); if (error) break; } VM_OBJECT_WUNLOCK(object); return (error); } else if (error != 0) { relpbuf(bp, freecnt); return (VM_PAGER_ERROR); } /* * If the file system supports BMAP, but blocksize is smaller * than a page size, then use special small filesystem code. */ if (pagesperblock == 0) { relpbuf(bp, freecnt); for (i = 0; i < count; i++) { PCPU_INC(cnt.v_vnodein); PCPU_INC(cnt.v_vnodepgsin); error = vnode_pager_input_smlfs(object, m[i]); if (error) break; } return (error); } /* * A sparse file can be encountered only for a single page request, * which may not be preceded by call to vm_pager_haspage(). */ if (bp->b_blkno == -1) { KASSERT(count == 1, ("%s: array[%d] request to a sparse file %p", __func__, count, vp)); relpbuf(bp, freecnt); pmap_zero_page(m[0]); KASSERT(m[0]->dirty == 0, ("%s: page %p is dirty", __func__, m[0])); VM_OBJECT_WLOCK(object); m[0]->valid = VM_PAGE_BITS_ALL; VM_OBJECT_WUNLOCK(object); return (VM_PAGER_OK); } #ifdef INVARIANTS blkno0 = bp->b_blkno; #endif bp->b_blkno += (foff % bsize) / DEV_BSIZE; /* Recalculate blocks available after/before to pages. */ poff = (foff % bsize) / PAGE_SIZE; before *= pagesperblock; before += poff; after *= pagesperblock; after += pagesperblock - (poff + 1); if (m[0]->pindex + after >= object->size) after = object->size - 1 - m[0]->pindex; KASSERT(count <= after + 1, ("%s: %d pages asked, can do only %d", __func__, count, after + 1)); after -= count - 1; /* Trim requested rbehind/rahead to possible values. */ rbehind = a_rbehind ? *a_rbehind : 0; rahead = a_rahead ? *a_rahead : 0; rbehind = min(rbehind, before); rbehind = min(rbehind, m[0]->pindex); rahead = min(rahead, after); rahead = min(rahead, object->size - m[count - 1]->pindex); /* * Check that total amount of pages fit into buf. Trim rbehind and * rahead evenly if not. */ if (rbehind + rahead + count > nitems(bp->b_pages)) { int trim, sum; trim = rbehind + rahead + count - nitems(bp->b_pages) + 1; sum = rbehind + rahead; if (rbehind == before) { /* Roundup rbehind trim to block size. */ rbehind -= roundup(trim * rbehind / sum, pagesperblock); if (rbehind < 0) rbehind = 0; } else rbehind -= trim * rbehind / sum; rahead -= trim * rahead / sum; } KASSERT(rbehind + rahead + count <= nitems(bp->b_pages), ("%s: behind %d ahead %d count %d", __func__, rbehind, rahead, count)); /* * Fill in the bp->b_pages[] array with requested and optional * read behind or read ahead pages. Read behind pages are looked * up in a backward direction, down to a first cached page. Same * for read ahead pages, but there is no need to shift the array * in case of encountering a cached page. */ i = bp->b_npages = 0; if (rbehind) { vm_pindex_t startpindex, tpindex; vm_page_t p; VM_OBJECT_WLOCK(object); startpindex = m[0]->pindex - rbehind; if ((p = TAILQ_PREV(m[0], pglist, listq)) != NULL && p->pindex >= startpindex) startpindex = p->pindex + 1; /* tpindex is unsigned; beware of numeric underflow. */ for (tpindex = m[0]->pindex - 1; tpindex >= startpindex && tpindex < m[0]->pindex; tpindex--, i++) { p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL); if (p == NULL) { /* Shift the array. */ for (int j = 0; j < i; j++) bp->b_pages[j] = bp->b_pages[j + tpindex + 1 - startpindex]; break; } bp->b_pages[tpindex - startpindex] = p; } bp->b_pgbefore = i; bp->b_npages += i; bp->b_blkno -= IDX_TO_OFF(i) / DEV_BSIZE; } else bp->b_pgbefore = 0; /* Requested pages. */ for (int j = 0; j < count; j++, i++) bp->b_pages[i] = m[j]; bp->b_npages += count; if (rahead) { vm_pindex_t endpindex, tpindex; vm_page_t p; if (!VM_OBJECT_WOWNED(object)) VM_OBJECT_WLOCK(object); endpindex = m[count - 1]->pindex + rahead + 1; if ((p = TAILQ_NEXT(m[count - 1], listq)) != NULL && p->pindex < endpindex) endpindex = p->pindex; if (endpindex > object->size) endpindex = object->size; for (tpindex = m[count - 1]->pindex + 1; tpindex < endpindex; i++, tpindex++) { p = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL); if (p == NULL) break; bp->b_pages[i] = p; } bp->b_pgafter = i - bp->b_npages; bp->b_npages = i; } else bp->b_pgafter = 0; if (VM_OBJECT_WOWNED(object)) VM_OBJECT_WUNLOCK(object); /* Report back actual behind/ahead read. */ if (a_rbehind) *a_rbehind = bp->b_pgbefore; if (a_rahead) *a_rahead = bp->b_pgafter; #ifdef INVARIANTS KASSERT(bp->b_npages <= nitems(bp->b_pages), ("%s: buf %p overflowed", __func__, bp)); for (int j = 1; j < bp->b_npages; j++) KASSERT(bp->b_pages[j]->pindex - 1 == bp->b_pages[j - 1]->pindex, ("%s: pages array not consecutive, bp %p", __func__, bp)); #endif /* * Recalculate first offset and bytecount with regards to read behind. * Truncate bytecount to vnode real size and round up physical size * for real devices. */ foff = IDX_TO_OFF(bp->b_pages[0]->pindex); bytecount = bp->b_npages << PAGE_SHIFT; if ((foff + bytecount) > object->un_pager.vnp.vnp_size) bytecount = object->un_pager.vnp.vnp_size - foff; secmask = bo->bo_bsize - 1; KASSERT(secmask < PAGE_SIZE && secmask > 0, ("%s: sector size %d too large", __func__, secmask + 1)); bytecount = (bytecount + secmask) & ~secmask; /* * And map the pages to be read into the kva, if the filesystem * requires mapped buffers. */ if ((vp->v_mount->mnt_kern_flag & MNTK_UNMAPPED_BUFS) != 0 && unmapped_buf_allowed) { bp->b_data = unmapped_buf; bp->b_offset = 0; } else { bp->b_data = bp->b_kvabase; pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); } /* Build a minimal buffer header. */ bp->b_iocmd = BIO_READ; KASSERT(bp->b_rcred == NOCRED, ("leaking read ucred")); KASSERT(bp->b_wcred == NOCRED, ("leaking write ucred")); bp->b_rcred = crhold(curthread->td_ucred); bp->b_wcred = crhold(curthread->td_ucred); pbgetbo(bo, bp); bp->b_vp = vp; bp->b_bcount = bp->b_bufsize = bp->b_runningbufspace = bytecount; bp->b_iooffset = dbtob(bp->b_blkno); KASSERT(IDX_TO_OFF(m[0]->pindex - bp->b_pages[0]->pindex) == (blkno0 - bp->b_blkno) * DEV_BSIZE + IDX_TO_OFF(m[0]->pindex) % bsize, ("wrong offsets bsize %d m[0] %ju b_pages[0] %ju " "blkno0 %ju b_blkno %ju", bsize, (uintmax_t)m[0]->pindex, (uintmax_t)bp->b_pages[0]->pindex, (uintmax_t)blkno0, (uintmax_t)bp->b_blkno)); atomic_add_long(&runningbufspace, bp->b_runningbufspace); PCPU_INC(cnt.v_vnodein); PCPU_ADD(cnt.v_vnodepgsin, bp->b_npages); if (iodone != NULL) { /* async */ bp->b_pgiodone = iodone; bp->b_caller1 = arg; bp->b_iodone = vnode_pager_generic_getpages_done_async; bp->b_flags |= B_ASYNC; BUF_KERNPROC(bp); bstrategy(bp); return (VM_PAGER_OK); } else { bp->b_iodone = bdone; bstrategy(bp); bwait(bp, PVM, "vnread"); error = vnode_pager_generic_getpages_done(bp); for (i = 0; i < bp->b_npages; i++) bp->b_pages[i] = NULL; bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_pbuf_freecnt); return (error != 0 ? VM_PAGER_ERROR : VM_PAGER_OK); } } static void vnode_pager_generic_getpages_done_async(struct buf *bp) { int error; error = vnode_pager_generic_getpages_done(bp); /* Run the iodone upon the requested range. */ bp->b_pgiodone(bp->b_caller1, bp->b_pages + bp->b_pgbefore, bp->b_npages - bp->b_pgbefore - bp->b_pgafter, error); for (int i = 0; i < bp->b_npages; i++) bp->b_pages[i] = NULL; bp->b_vp = NULL; pbrelbo(bp); relpbuf(bp, &vnode_async_pbuf_freecnt); } static int vnode_pager_generic_getpages_done(struct buf *bp) { vm_object_t object; off_t tfoff, nextoff; int i, error; error = (bp->b_ioflags & BIO_ERROR) != 0 ? EIO : 0; object = bp->b_vp->v_object; if (error == 0 && bp->b_bcount != bp->b_npages * PAGE_SIZE) { if (!buf_mapped(bp)) { bp->b_data = bp->b_kvabase; pmap_qenter((vm_offset_t)bp->b_data, bp->b_pages, bp->b_npages); } bzero(bp->b_data + bp->b_bcount, PAGE_SIZE * bp->b_npages - bp->b_bcount); } if (buf_mapped(bp)) { pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages); bp->b_data = unmapped_buf; } VM_OBJECT_WLOCK(object); for (i = 0, tfoff = IDX_TO_OFF(bp->b_pages[0]->pindex); i < bp->b_npages; i++, tfoff = nextoff) { vm_page_t mt; nextoff = tfoff + PAGE_SIZE; mt = bp->b_pages[i]; if (nextoff <= object->un_pager.vnp.vnp_size) { /* * Read filled up entire page. */ mt->valid = VM_PAGE_BITS_ALL; KASSERT(mt->dirty == 0, ("%s: page %p is dirty", __func__, mt)); KASSERT(!pmap_page_is_mapped(mt), ("%s: page %p is mapped", __func__, mt)); } else { /* * Read did not fill up entire page. * * Currently we do not set the entire page valid, * we just try to clear the piece that we couldn't * read. */ vm_page_set_valid_range(mt, 0, object->un_pager.vnp.vnp_size - tfoff); KASSERT((mt->dirty & vm_page_bits(0, object->un_pager.vnp.vnp_size - tfoff)) == 0, ("%s: page %p is dirty", __func__, mt)); } if (i < bp->b_pgbefore || i >= bp->b_npages - bp->b_pgafter) vm_page_readahead_finish(mt); } VM_OBJECT_WUNLOCK(object); if (error != 0) printf("%s: I/O read error %d\n", __func__, error); return (error); } /* * EOPNOTSUPP is no longer legal. For local media VFS's that do not * implement their own VOP_PUTPAGES, their VOP_PUTPAGES should call to * vnode_pager_generic_putpages() to implement the previous behaviour. * * All other FS's should use the bypass to get to the local media * backing vp's VOP_PUTPAGES. */ static void vnode_pager_putpages(vm_object_t object, vm_page_t *m, int count, int flags, int *rtvals) { int rtval; struct vnode *vp; int bytes = count * PAGE_SIZE; /* * Force synchronous operation if we are extremely low on memory * to prevent a low-memory deadlock. VOP operations often need to * allocate more memory to initiate the I/O ( i.e. do a BMAP * operation ). The swapper handles the case by limiting the amount * of asynchronous I/O, but that sort of solution doesn't scale well * for the vnode pager without a lot of work. * * Also, the backing vnode's iodone routine may not wake the pageout * daemon up. This should be probably be addressed XXX. */ if (vm_cnt.v_free_count < vm_cnt.v_pageout_free_min) flags |= VM_PAGER_PUT_SYNC; /* * Call device-specific putpages function */ vp = object->handle; VM_OBJECT_WUNLOCK(object); rtval = VOP_PUTPAGES(vp, m, bytes, flags, rtvals); KASSERT(rtval != EOPNOTSUPP, ("vnode_pager: stale FS putpages\n")); VM_OBJECT_WLOCK(object); } /* * This is now called from local media FS's to operate against their * own vnodes if they fail to implement VOP_PUTPAGES. * * This is typically called indirectly via the pageout daemon and * clustering has already typically occurred, so in general we ask the * underlying filesystem to write the data out asynchronously rather * then delayed. */ int vnode_pager_generic_putpages(struct vnode *vp, vm_page_t *ma, int bytecount, int flags, int *rtvals) { int i; vm_object_t object; vm_page_t m; int count; int maxsize, ncount; vm_ooffset_t poffset; struct uio auio; struct iovec aiov; int error; int ioflags; int ppscheck = 0; static struct timeval lastfail; static int curfail; object = vp->v_object; count = bytecount / PAGE_SIZE; for (i = 0; i < count; i++) rtvals[i] = VM_PAGER_ERROR; if ((int64_t)ma[0]->pindex < 0) { printf("vnode_pager_putpages: attempt to write meta-data!!! -- 0x%lx(%lx)\n", (long)ma[0]->pindex, (u_long)ma[0]->dirty); rtvals[0] = VM_PAGER_BAD; return VM_PAGER_BAD; } maxsize = count * PAGE_SIZE; ncount = count; poffset = IDX_TO_OFF(ma[0]->pindex); /* * If the page-aligned write is larger then the actual file we * have to invalidate pages occurring beyond the file EOF. However, * there is an edge case where a file may not be page-aligned where * the last page is partially invalid. In this case the filesystem * may not properly clear the dirty bits for the entire page (which * could be VM_PAGE_BITS_ALL due to the page having been mmap()d). * With the page locked we are free to fix-up the dirty bits here. * * We do not under any circumstances truncate the valid bits, as * this will screw up bogus page replacement. */ VM_OBJECT_WLOCK(object); if (maxsize + poffset > object->un_pager.vnp.vnp_size) { if (object->un_pager.vnp.vnp_size > poffset) { int pgoff; maxsize = object->un_pager.vnp.vnp_size - poffset; ncount = btoc(maxsize); if ((pgoff = (int)maxsize & PAGE_MASK) != 0) { /* * If the object is locked and the following * conditions hold, then the page's dirty * field cannot be concurrently changed by a * pmap operation. */ m = ma[ncount - 1]; vm_page_assert_sbusied(m); KASSERT(!pmap_page_is_write_mapped(m), ("vnode_pager_generic_putpages: page %p is not read-only", m)); vm_page_clear_dirty(m, pgoff, PAGE_SIZE - pgoff); } } else { maxsize = 0; ncount = 0; } if (ncount < count) { for (i = ncount; i < count; i++) { rtvals[i] = VM_PAGER_BAD; } } } VM_OBJECT_WUNLOCK(object); /* * pageouts are already clustered, use IO_ASYNC to force a bawrite() * rather then a bdwrite() to prevent paging I/O from saturating * the buffer cache. Dummy-up the sequential heuristic to cause * large ranges to cluster. If neither IO_SYNC or IO_ASYNC is set, * the system decides how to cluster. */ ioflags = IO_VMIO; if (flags & (VM_PAGER_PUT_SYNC | VM_PAGER_PUT_INVAL)) ioflags |= IO_SYNC; else if ((flags & VM_PAGER_CLUSTER_OK) == 0) ioflags |= IO_ASYNC; ioflags |= (flags & VM_PAGER_PUT_INVAL) ? IO_INVAL: 0; + ioflags |= (flags & VM_PAGER_PUT_NOREUSE) ? IO_NOREUSE : 0; ioflags |= IO_SEQMAX << IO_SEQSHIFT; aiov.iov_base = (caddr_t) 0; aiov.iov_len = maxsize; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; auio.uio_offset = poffset; auio.uio_segflg = UIO_NOCOPY; auio.uio_rw = UIO_WRITE; auio.uio_resid = maxsize; auio.uio_td = (struct thread *) 0; error = VOP_WRITE(vp, &auio, ioflags, curthread->td_ucred); PCPU_INC(cnt.v_vnodeout); PCPU_ADD(cnt.v_vnodepgsout, ncount); if (error) { if ((ppscheck = ppsratecheck(&lastfail, &curfail, 1))) printf("vnode_pager_putpages: I/O error %d\n", error); } if (auio.uio_resid) { if (ppscheck || ppsratecheck(&lastfail, &curfail, 1)) printf("vnode_pager_putpages: residual I/O %zd at %lu\n", auio.uio_resid, (u_long)ma[0]->pindex); } for (i = 0; i < ncount; i++) { rtvals[i] = VM_PAGER_OK; } return rtvals[0]; } void vnode_pager_undirty_pages(vm_page_t *ma, int *rtvals, int written) { vm_object_t obj; int i, pos; if (written == 0) return; obj = ma[0]->object; VM_OBJECT_WLOCK(obj); for (i = 0, pos = 0; pos < written; i++, pos += PAGE_SIZE) { if (pos < trunc_page(written)) { rtvals[i] = VM_PAGER_OK; vm_page_undirty(ma[i]); } else { /* Partially written page. */ rtvals[i] = VM_PAGER_AGAIN; vm_page_clear_dirty(ma[i], 0, written & PAGE_MASK); } } VM_OBJECT_WUNLOCK(obj); } void vnode_pager_update_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { struct vnode *vp; vm_ooffset_t old_wm; VM_OBJECT_WLOCK(object); if (object->type != OBJT_VNODE) { VM_OBJECT_WUNLOCK(object); return; } old_wm = object->un_pager.vnp.writemappings; object->un_pager.vnp.writemappings += (vm_ooffset_t)end - start; vp = object->handle; if (old_wm == 0 && object->un_pager.vnp.writemappings != 0) { ASSERT_VOP_ELOCKED(vp, "v_writecount inc"); VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } else if (old_wm != 0 && object->un_pager.vnp.writemappings == 0) { ASSERT_VOP_ELOCKED(vp, "v_writecount dec"); VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } VM_OBJECT_WUNLOCK(object); } void vnode_pager_release_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { struct vnode *vp; struct mount *mp; vm_offset_t inc; VM_OBJECT_WLOCK(object); /* * First, recheck the object type to account for the race when * the vnode is reclaimed. */ if (object->type != OBJT_VNODE) { VM_OBJECT_WUNLOCK(object); return; } /* * Optimize for the case when writemappings is not going to * zero. */ inc = end - start; if (object->un_pager.vnp.writemappings != inc) { object->un_pager.vnp.writemappings -= inc; VM_OBJECT_WUNLOCK(object); return; } vp = object->handle; vhold(vp); VM_OBJECT_WUNLOCK(object); mp = NULL; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); /* * Decrement the object's writemappings, by swapping the start * and end arguments for vnode_pager_update_writecount(). If * there was not a race with vnode reclaimation, then the * vnode's v_writecount is decremented. */ vnode_pager_update_writecount(object, end, start); VOP_UNLOCK(vp, 0); vdrop(vp); if (mp != NULL) vn_finished_write(mp); }