X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..13f56ec4e58bf8687e2a68032c093c0213dd519b:/bsd/vfs/vfs_subr.c?ds=inline diff --git a/bsd/vfs/vfs_subr.c b/bsd/vfs/vfs_subr.c index be1ba3291..43352545d 100644 --- a/bsd/vfs/vfs_subr.c +++ b/bsd/vfs/vfs_subr.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2010 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -105,6 +105,7 @@ #include #include #include +#include #include #include @@ -117,12 +118,17 @@ #include #include +#include #include /* kalloc()/kfree() */ #include /* delay_for_interval() */ #include /* OSAddAtomic() */ +#ifdef JOE_DEBUG +#include +#endif + #include /* vnode_pager_vrele() */ #if CONFIG_MACF @@ -132,6 +138,10 @@ extern lck_grp_t *vnode_lck_grp; extern lck_attr_t *vnode_lck_attr; +#if CONFIG_TRIGGERS +extern lck_grp_t *trigger_vnode_lck_grp; +extern lck_attr_t *trigger_vnode_lck_attr; +#endif extern lck_mtx_t * mnt_list_mtx_lock; @@ -144,6 +154,16 @@ int vttoif_tab[9] = { S_IFSOCK, S_IFIFO, S_IFMT, }; + +/* XXX These should be in a BSD accessible Mach header, but aren't. */ +extern void memory_object_mark_used( + memory_object_control_t control); + +extern void memory_object_mark_unused( + memory_object_control_t control, + boolean_t rage); + + /* XXX next protptype should be from */ extern int nfs_vinvalbuf(vnode_t, int, vfs_context_t, int); @@ -160,8 +180,11 @@ __private_extern__ kern_return_t reset_vmobjectcache(unsigned int val1, unsigned int val2); __private_extern__ int unlink1(vfs_context_t, struct nameidata *, int); +extern int system_inshutdown; + static void vnode_list_add(vnode_t); static void vnode_list_remove(vnode_t); +static void vnode_list_remove_locked(vnode_t); static errno_t vnode_drain(vnode_t); static void vgone(vnode_t, int flags); @@ -169,8 +192,6 @@ static void vclean(vnode_t vp, int flag); static void vnode_reclaim_internal(vnode_t, int, int, int); static void vnode_dropiocount (vnode_t); -static errno_t vnode_getiocount(vnode_t vp, int vid, int vflags); -static int vget_internal(vnode_t, int, int); static vnode_t checkalias(vnode_t vp, dev_t nvp_rdev); static int vnode_reload(vnode_t); @@ -180,13 +201,26 @@ static void insmntque(vnode_t vp, mount_t mp); static int mount_getvfscnt(void); static int mount_fillfsids(fsid_t *, int ); static void vnode_iterate_setup(mount_t); -static int vnode_umount_preflight(mount_t, vnode_t, int); +int vnode_umount_preflight(mount_t, vnode_t, int); static int vnode_iterate_prepare(mount_t); static int vnode_iterate_reloadq(mount_t); static void vnode_iterate_clear(mount_t); +static mount_t vfs_getvfs_locked(fsid_t *); +static int vn_create_reg(vnode_t dvp, vnode_t *vpp, struct nameidata *ndp, + struct vnode_attr *vap, uint32_t flags, int fmode, uint32_t *statusp, vfs_context_t ctx); +static int vnode_authattr_new_internal(vnode_t dvp, struct vnode_attr *vap, int noauth, uint32_t *defaulted_fieldsp, vfs_context_t ctx); errno_t rmdir_remove_orphaned_appleDouble(vnode_t, vfs_context_t, int *); +#ifdef JOE_DEBUG +static void record_vp(vnode_t vp, int count); +#endif + +#if CONFIG_TRIGGERS +static int vnode_resolver_create(mount_t, vnode_t, struct vnode_trigger_param *, boolean_t external); +static void vnode_resolver_detach(vnode_t); +#endif + TAILQ_HEAD(freelst, vnode) vnode_free_list; /* vnode free list */ TAILQ_HEAD(deadlst, vnode) vnode_dead_list; /* vnode dead list */ @@ -259,7 +293,7 @@ static int nummounts = 0; * place for now... it should be deprecated out of the * exports and removed eventually. */ -unsigned long vnodetarget; /* target for vnreclaim() */ +u_int32_t vnodetarget; /* target for vnreclaim() */ #define VNODE_FREE_TARGET 20 /* Default value for vnodetarget */ /* @@ -304,6 +338,10 @@ reset_vmobjectcache(unsigned int val1, unsigned int val2) vm_size_t oval = val1 - VNODE_FREE_MIN; vm_size_t nval; + if (val1 == val2) { + return KERN_SUCCESS; + } + if(val2 < VNODE_FREE_MIN) nval = 0; else @@ -323,9 +361,9 @@ vnode_waitforwrites(vnode_t vp, int output_target, int slpflag, int slptimeout, if (vp->v_numoutput > output_target) { - slpflag &= ~PDROP; + slpflag |= PDROP; - vnode_lock(vp); + vnode_lock_spin(vp); while ((vp->v_numoutput > output_target) && error == 0) { if (output_target) @@ -336,6 +374,8 @@ vnode_waitforwrites(vnode_t vp, int output_target, int slpflag, int slptimeout, ts.tv_sec = (slptimeout/100); ts.tv_nsec = (slptimeout % 1000) * 10 * NSEC_PER_USEC * 1000 ; error = msleep((caddr_t)&vp->v_numoutput, &vp->v_lock, (slpflag | (PRIBIO + 1)), msg, &ts); + + vnode_lock_spin(vp); } vnode_unlock(vp); } @@ -356,29 +396,27 @@ void vnode_writedone(vnode_t vp) { if (vp) { - OSAddAtomic(-1, &vp->v_numoutput); + int need_wakeup = 0; - if (vp->v_numoutput <= 1) { - int need_wakeup = 0; + OSAddAtomic(-1, &vp->v_numoutput); - vnode_lock_spin(vp); + vnode_lock_spin(vp); - if (vp->v_numoutput < 0) - panic("vnode_writedone: numoutput < 0"); + if (vp->v_numoutput < 0) + panic("vnode_writedone: numoutput < 0"); - if ((vp->v_flag & VTHROTTLED) && (vp->v_numoutput <= 1)) { - vp->v_flag &= ~VTHROTTLED; - need_wakeup = 1; - } - if ((vp->v_flag & VBWAIT) && (vp->v_numoutput == 0)) { - vp->v_flag &= ~VBWAIT; - need_wakeup = 1; - } - vnode_unlock(vp); - - if (need_wakeup) - wakeup((caddr_t)&vp->v_numoutput); + if ((vp->v_flag & VTHROTTLED)) { + vp->v_flag &= ~VTHROTTLED; + need_wakeup = 1; + } + if ((vp->v_flag & VBWAIT) && (vp->v_numoutput == 0)) { + vp->v_flag &= ~VBWAIT; + need_wakeup = 1; } + vnode_unlock(vp); + + if (need_wakeup) + wakeup((caddr_t)&vp->v_numoutput); } } @@ -437,7 +475,7 @@ vnode_iterate_setup(mount_t mp) } -static int +int vnode_umount_preflight(mount_t mp, vnode_t skipvp, int flags) { vnode_t vp; @@ -645,6 +683,12 @@ mount_lock(mount_t mp) lck_mtx_lock(&mp->mnt_mlock); } +void +mount_lock_spin(mount_t mp) +{ + lck_mtx_lock_spin(&mp->mnt_mlock); +} + void mount_unlock(mount_t mp) { @@ -656,7 +700,7 @@ void mount_ref(mount_t mp, int locked) { if ( !locked) - mount_lock(mp); + mount_lock_spin(mp); mp->mnt_count++; @@ -669,7 +713,7 @@ void mount_drop(mount_t mp, int locked) { if ( !locked) - mount_lock(mp); + mount_lock_spin(mp); mp->mnt_count--; @@ -761,6 +805,13 @@ mount_refdrain(mount_t mp) return(0); } +/* Tags the mount point as not supportine extended readdir for NFS exports */ +void +mount_set_noreaddirext(mount_t mp) { + mount_lock (mp); + mp->mnt_kern_flag |= MNTK_DENY_READDIREXT; + mount_unlock (mp); +} /* * Mark a mount point as busy. Used to synchronize access and to delay @@ -852,8 +903,8 @@ vfs_rootmountalloc_internal(struct vfstable *vfsp, const char *devname) { mount_t mp; - mp = _MALLOC_ZONE((u_long)sizeof(struct mount), M_MOUNT, M_WAITOK); - bzero((char *)mp, (u_long)sizeof(struct mount)); + mp = _MALLOC_ZONE(sizeof(struct mount), M_MOUNT, M_WAITOK); + bzero((char *)mp, sizeof(struct mount)); /* Initialize the default IO constraints */ mp->mnt_maxreadcnt = mp->mnt_maxwritecnt = MAXPHYS; @@ -862,9 +913,13 @@ vfs_rootmountalloc_internal(struct vfstable *vfsp, const char *devname) mp->mnt_maxsegwritesize = mp->mnt_maxwritecnt; mp->mnt_devblocksize = DEV_BSIZE; mp->mnt_alignmentmask = PAGE_MASK; + mp->mnt_ioqueue_depth = MNT_DEFAULT_IOQUEUE_DEPTH; + mp->mnt_ioscale = 1; mp->mnt_ioflags = 0; mp->mnt_realrootvp = NULLVP; mp->mnt_authcache_ttl = CACHED_LOOKUP_RIGHT_TTL; + mp->mnt_throttle_mask = LOWPRI_MAX_NUM_DEV - 1; + mp->mnt_devbsdunit = 0; mount_lock_init(mp); (void)vfs_busy(mp, LK_NOWAIT); @@ -1045,9 +1100,12 @@ vfs_mountroot(void) dounmount(mp, MNT_FORCE, 0, ctx); goto fail; } - - /* VFS_ROOT provides reference so flags = 0 */ error = vnode_label(mp, NULL, vp, NULL, 0, ctx); + /* + * get rid of reference provided by VFS_ROOT + */ + vnode_put(vp); + if (error) { printf("%s() vnode_label() returned %d\n", __func__, error); @@ -1071,7 +1129,6 @@ fail: /* * Lookup a mount point by filesystem identifier. */ -extern mount_t vfs_getvfs_locked(fsid_t *); struct mount * vfs_getvfs(fsid_t *fsid) @@ -1079,7 +1136,7 @@ vfs_getvfs(fsid_t *fsid) return (mount_list_lookupby_fsid(fsid, 0, 0)); } -struct mount * +static struct mount * vfs_getvfs_locked(fsid_t *fsid) { return(mount_list_lookupby_fsid(fsid, 1, 0)); @@ -1096,6 +1153,8 @@ vfs_getvfs_by_mntonname(char *path) if (!strncmp(mp->mnt_vfsstat.f_mntonname, path, sizeof(mp->mnt_vfsstat.f_mntonname))) { retmp = mp; + if (mount_iterref(retmp, 1)) + retmp = NULL; goto out; } } @@ -1160,7 +1219,7 @@ insmntque(vnode_t vp, mount_t mp) panic("insmntque: vp not in mount vnode list"); vp->v_lflag &= ~VNAMED_MOUNT; - mount_lock(lmp); + mount_lock_spin(lmp); mount_drop(lmp, 1); @@ -1185,7 +1244,7 @@ insmntque(vnode_t vp, mount_t mp) * Insert into list of vnodes for the new mount point, if available. */ if ((vp->v_mount = mp) != NULL) { - mount_lock(mp); + mount_lock_spin(mp); if ((vp->v_mntvnodes.tqe_next != 0) && (vp->v_mntvnodes.tqe_prev != 0)) panic("vp already in mount list"); if (mp->mnt_lflag & MNT_LITER) @@ -1194,8 +1253,6 @@ insmntque(vnode_t vp, mount_t mp) TAILQ_INSERT_HEAD(&mp->mnt_vnodelist, vp, v_mntvnodes); if (vp->v_lflag & VNAMED_MOUNT) panic("insmntque: vp already in mount vnode list"); - if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb)) - panic("insmntque: vp on the free list\n"); vp->v_lflag |= VNAMED_MOUNT; mount_ref(mp, 1); mount_unlock(mp); @@ -1289,6 +1346,7 @@ checkalias(struct vnode *nvp, dev_t nvp_rdev) { struct vnode *vp; struct vnode **vpp; + struct specinfo *sin = NULL; int vid = 0; vpp = &speclisth[SPECHASH(nvp_rdev)]; @@ -1304,6 +1362,7 @@ loop: SPECHASH_UNLOCK(); if (vp) { +found_alias: if (vnode_getwithvid(vp,vid)) { goto loop; } @@ -1316,41 +1375,67 @@ loop: * Alias, but not in use, so flush it out. */ if ((vp->v_iocount == 1) && (vp->v_usecount == 0)) { - vnode_reclaim_internal(vp, 1, 0, 0); + vnode_reclaim_internal(vp, 1, 1, 0); + vnode_put_locked(vp); vnode_unlock(vp); - vnode_put(vp); goto loop; } + } if (vp == NULL || vp->v_tag != VT_NON) { -retnullvp: - MALLOC_ZONE(nvp->v_specinfo, struct specinfo *, sizeof(struct specinfo), - M_SPECINFO, M_WAITOK); + if (sin == NULL) { + MALLOC_ZONE(sin, struct specinfo *, sizeof(struct specinfo), + M_SPECINFO, M_WAITOK); + } + + nvp->v_specinfo = sin; bzero(nvp->v_specinfo, sizeof(struct specinfo)); nvp->v_rdev = nvp_rdev; nvp->v_specflags = 0; nvp->v_speclastr = -1; + nvp->v_specinfo->si_opencount = 0; + nvp->v_specinfo->si_initted = 0; + nvp->v_specinfo->si_throttleable = 0; SPECHASH_LOCK(); + + /* We dropped the lock, someone could have added */ + if (vp == NULLVP) { + for (vp = *vpp; vp; vp = vp->v_specnext) { + if (nvp_rdev == vp->v_rdev && nvp->v_type == vp->v_type) { + vid = vp->v_id; + SPECHASH_UNLOCK(); + goto found_alias; + } + } + } + nvp->v_hashchain = vpp; nvp->v_specnext = *vpp; *vpp = nvp; - SPECHASH_UNLOCK(); if (vp != NULLVP) { - nvp->v_flag |= VALIASED; - vp->v_flag |= VALIASED; + nvp->v_specflags |= SI_ALIASED; + vp->v_specflags |= SI_ALIASED; + SPECHASH_UNLOCK(); + vnode_put_locked(vp); vnode_unlock(vp); - vnode_put(vp); + } else { + SPECHASH_UNLOCK(); } + return (NULLVP); } + + if (sin) { + FREE_ZONE(sin, sizeof(struct specinfo), M_SPECINFO); + } + if ((vp->v_flag & (VBDEVVP | VDEVFLUSH)) != 0) return(vp); - else { - panic("checkalias with VT_NON vp that shouldn't: %x", (unsigned int)vp); - goto retnullvp; - } + + panic("checkalias with VT_NON vp that shouldn't: %p", vp); + return (vp); } @@ -1365,26 +1450,20 @@ retnullvp: * and an error returned to indicate that the vnode is no longer * usable (possibly having been changed to a new file system type). */ -static int +int vget_internal(vnode_t vp, int vid, int vflags) { int error = 0; - int vpid; vnode_lock_spin(vp); - if (vflags & VNODE_WITHID) - vpid = vid; - else - vpid = vp->v_id; // save off the original v_id - if ((vflags & VNODE_WRITEABLE) && (vp->v_writecount == 0)) /* * vnode to be returned only if it has writers opened */ error = EINVAL; else - error = vnode_getiocount(vp, vpid, vflags); + error = vnode_getiocount(vp, vid, vflags); vnode_unlock(vp); @@ -1399,7 +1478,7 @@ int vnode_ref(vnode_t vp) { - return (vnode_ref_ext(vp, 0)); + return (vnode_ref_ext(vp, 0, 0)); } /* @@ -1407,7 +1486,7 @@ vnode_ref(vnode_t vp) * ENOENT No such file or directory [terminating] */ int -vnode_ref_ext(vnode_t vp, int fmode) +vnode_ref_ext(vnode_t vp, int fmode, int flags) { int error = 0; @@ -1424,10 +1503,12 @@ vnode_ref_ext(vnode_t vp, int fmode) /* * if you are the owner of drain/termination, can acquire usecount */ - if ((vp->v_lflag & (VL_DRAIN | VL_TERMINATE | VL_DEAD))) { - if (vp->v_owner != current_thread()) { - error = ENOENT; - goto out; + if ((flags & VNODE_REF_FORCE) == 0) { + if ((vp->v_lflag & (VL_DRAIN | VL_TERMINATE | VL_DEAD))) { + if (vp->v_owner != current_thread()) { + error = ENOENT; + goto out; + } } } vp->v_usecount++; @@ -1460,6 +1541,13 @@ vnode_ref_ext(vnode_t vp, int fmode) vnode_list_remove(vp); } } + if (vp->v_usecount == 1 && vp->v_type == VREG && !(vp->v_flag & VSYSTEM)) { + + if (vp->v_ubcinfo) { + vnode_lock_convert(vp); + memory_object_mark_used(vp->v_ubcinfo->ui_control); + } + } out: vnode_unlock(vp); @@ -1474,11 +1562,15 @@ out: static void vnode_list_add(vnode_t vp) { +#if DIAGNOSTIC + lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED); +#endif /* * if it is already on a list or non zero references return */ - if (VONLIST(vp) || (vp->v_usecount != 0) || (vp->v_iocount != 0)) + if (VONLIST(vp) || (vp->v_usecount != 0) || (vp->v_iocount != 0) || (vp->v_lflag & VL_TERMINATE)) return; + vnode_list_lock(); if ((vp->v_flag & VRAGE) && !(vp->v_lflag & VL_DEAD)) { @@ -1523,12 +1615,40 @@ vnode_list_add(vnode_t vp) vnode_list_unlock(); } + +/* + * remove the vnode from appropriate free list. + * called with vnode LOCKED and + * the list lock held + */ +static void +vnode_list_remove_locked(vnode_t vp) +{ + if (VONLIST(vp)) { + /* + * the v_listflag field is + * protected by the vnode_list_lock + */ + if (vp->v_listflag & VLIST_RAGE) + VREMRAGE("vnode_list_remove", vp); + else if (vp->v_listflag & VLIST_DEAD) + VREMDEAD("vnode_list_remove", vp); + else + VREMFREE("vnode_list_remove", vp); + } +} + + /* * remove the vnode from appropriate free list. + * called with vnode LOCKED */ static void vnode_list_remove(vnode_t vp) { +#if DIAGNOSTIC + lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED); +#endif /* * we want to avoid taking the list lock * in the case where we're not on the free @@ -1548,24 +1668,16 @@ vnode_list_remove(vnode_t vp) /* * however, we're not guaranteed that * we won't go from the on-list state - * to the non-on-list state until we + * to the not-on-list state until we * hold the vnode_list_lock... this - * is due to new_vnode removing vnodes + * is due to "new_vnode" removing vnodes * from the free list uder the list_lock * w/o the vnode lock... so we need to * check again whether we're currently * on the free list */ - if (VONLIST(vp)) { - if (vp->v_listflag & VLIST_RAGE) - VREMRAGE("vnode_list_remove", vp); - else if (vp->v_listflag & VLIST_DEAD) - VREMDEAD("vnode_list_remove", vp); - else - VREMFREE("vnode_list_remove", vp); + vnode_list_remove_locked(vp); - VLISTNONE(vp); - } vnode_list_unlock(); } } @@ -1588,22 +1700,27 @@ vnode_rele_ext(vnode_t vp, int fmode, int dont_reenter) void vnode_rele_internal(vnode_t vp, int fmode, int dont_reenter, int locked) { + if ( !locked) vnode_lock_spin(vp); - +#if DIAGNOSTIC + else + lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED); +#endif if (--vp->v_usecount < 0) - panic("vnode_rele_ext: vp %p usecount -ve : %d", vp, vp->v_usecount); + panic("vnode_rele_ext: vp %p usecount -ve : %d. v_tag = %d, v_type = %d, v_flag = %x.", vp, vp->v_usecount, vp->v_tag, vp->v_type, vp->v_flag); if (fmode & FWRITE) { if (--vp->v_writecount < 0) - panic("vnode_rele_ext: vp %p writecount -ve : %ld", vp, vp->v_writecount); + panic("vnode_rele_ext: vp %p writecount -ve : %d. v_tag = %d, v_type = %d, v_flag = %x.", vp, vp->v_writecount, vp->v_tag, vp->v_type, vp->v_flag); } if (fmode & O_EVTONLY) { if (--vp->v_kusecount < 0) - panic("vnode_rele_ext: vp %p kusecount -ve : %d", vp, vp->v_kusecount); + panic("vnode_rele_ext: vp %p kusecount -ve : %d. v_tag = %d, v_type = %d, v_flag = %x.", vp, vp->v_kusecount, vp->v_tag, vp->v_type, vp->v_flag); } if (vp->v_kusecount > vp->v_usecount) - panic("vnode_rele_ext: vp %p kusecount(%d) out of balance with usecount(%d)\n",vp, vp->v_kusecount, vp->v_usecount); + panic("vnode_rele_ext: vp %p kusecount(%d) out of balance with usecount(%d). v_tag = %d, v_type = %d, v_flag = %x.",vp, vp->v_kusecount, vp->v_usecount, vp->v_tag, vp->v_type, vp->v_flag); + if ((vp->v_iocount > 0) || (vp->v_usecount > 0)) { /* * vnode is still busy... if we're the last @@ -1614,9 +1731,7 @@ vnode_rele_internal(vnode_t vp, int fmode, int dont_reenter, int locked) vp->v_lflag |= VL_NEEDINACTIVE; vp->v_flag &= ~(VNOCACHE_DATA | VRAOFF | VOPENEVT); } - if ( !locked) - vnode_unlock(vp); - return; + goto done; } vp->v_flag &= ~(VNOCACHE_DATA | VRAOFF | VOPENEVT); @@ -1634,9 +1749,8 @@ vnode_rele_internal(vnode_t vp, int fmode, int dont_reenter, int locked) vp->v_flag |= VAGE; } vnode_list_add(vp); - if ( !locked) - vnode_unlock(vp); - return; + + goto done; } /* * at this point both the iocount and usecount @@ -1671,15 +1785,22 @@ vnode_rele_internal(vnode_t vp, int fmode, int dont_reenter, int locked) if (ut->uu_defer_reclaims) { vp->v_defer_reclaimlist = ut->uu_vreclaims; - ut->uu_vreclaims = vp; - goto defer_reclaim; + ut->uu_vreclaims = vp; + goto done; } vnode_lock_convert(vp); - vnode_reclaim_internal(vp, 1, 0, 0); + vnode_reclaim_internal(vp, 1, 1, 0); } vnode_dropiocount(vp); vnode_list_add(vp); -defer_reclaim: +done: + if (vp->v_usecount == 0 && vp->v_type == VREG && !(vp->v_flag & VSYSTEM)) { + + if (vp->v_ubcinfo) { + vnode_lock_convert(vp); + memory_object_mark_unused(vp->v_ubcinfo->ui_control, (vp->v_flag & VRAGE) == VRAGE); + } + } if ( !locked) vnode_unlock(vp); return; @@ -1707,7 +1828,7 @@ vflush(struct mount *mp, struct vnode *skipvp, int flags) int busy = 0; int reclaimed = 0; int retval; - int vid; + unsigned int vid; mount_lock(mp); vnode_iterate_setup(mp); @@ -1735,17 +1856,20 @@ loop: return(retval); } - /* iterate over all the vnodes */ - while (!TAILQ_EMPTY(&mp->mnt_workerqueue)) { - vp = TAILQ_FIRST(&mp->mnt_workerqueue); - TAILQ_REMOVE(&mp->mnt_workerqueue, vp, v_mntvnodes); - TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); - if ( (vp->v_mount != mp) || (vp == skipvp)) { - continue; - } - vid = vp->v_id; - mount_unlock(mp); - vnode_lock(vp); + /* iterate over all the vnodes */ + while (!TAILQ_EMPTY(&mp->mnt_workerqueue)) { + + vp = TAILQ_FIRST(&mp->mnt_workerqueue); + TAILQ_REMOVE(&mp->mnt_workerqueue, vp, v_mntvnodes); + TAILQ_INSERT_TAIL(&mp->mnt_vnodelist, vp, v_mntvnodes); + + if ( (vp->v_mount != mp) || (vp == skipvp)) { + continue; + } + vid = vp->v_id; + mount_unlock(mp); + + vnode_lock_spin(vp); if ((vp->v_id != vid) || ((vp->v_lflag & (VL_DEAD | VL_TERMINATE)))) { vnode_unlock(vp); @@ -1772,7 +1896,7 @@ loop: continue; } /* - * If requested, skip over vnodes marked VSWAP. + * If requested, skip over vnodes marked VROOT. */ if ((flags & SKIPROOT) && (vp->v_flag & VROOT)) { vnode_unlock(vp); @@ -1795,15 +1919,17 @@ loop: */ if (((vp->v_usecount == 0) || ((vp->v_usecount - vp->v_kusecount) == 0))) { + + vnode_lock_convert(vp); vp->v_iocount++; /* so that drain waits for * other iocounts */ #ifdef JOE_DEBUG record_vp(vp, 1); #endif - vnode_reclaim_internal(vp, 1, 0, 0); + vnode_reclaim_internal(vp, 1, 1, 0); vnode_dropiocount(vp); vnode_list_add(vp); - vnode_unlock(vp); + reclaimed++; mount_lock(mp); continue; @@ -1814,12 +1940,14 @@ loop: * anonymous device. For all other files, just kill them. */ if (flags & FORCECLOSE) { + vnode_lock_convert(vp); + if (vp->v_type != VBLK && vp->v_type != VCHR) { vp->v_iocount++; /* so that drain waits * for other iocounts */ #ifdef JOE_DEBUG record_vp(vp, 1); #endif - vnode_reclaim_internal(vp, 1, 0, 0); + vnode_reclaim_internal(vp, 1, 1, 0); vnode_dropiocount(vp); vnode_list_add(vp); vnode_unlock(vp); @@ -1864,7 +1992,7 @@ loop: return (0); } -long num_recycledvnodes = 0; /* long for OSAddAtomic */ +long num_recycledvnodes = 0; /* * Disassociate the underlying file system from a vnode. * The vnode lock is held on entry. @@ -1877,6 +2005,9 @@ vclean(vnode_t vp, int flags) int need_inactive; int already_terminating; int clflags = 0; +#if NAMEDSTREAMS + int is_namedstream; +#endif /* * Check to see if the vnode is in use. @@ -1908,13 +2039,13 @@ vclean(vnode_t vp, int flags) */ insmntque(vp, (struct mount *)0); +#if NAMEDSTREAMS + is_namedstream = vnode_isnamedstream(vp); +#endif + vnode_unlock(vp); - OSAddAtomic(1, &num_recycledvnodes); - /* - * purge from the name cache as early as possible... - */ - cache_purge(vp); + OSAddAtomicLong(1, &num_recycledvnodes); if (flags & DOCLOSE) clflags |= IO_NDELAY; @@ -1935,17 +2066,40 @@ vclean(vnode_t vp, int flags) #endif { VNOP_FSYNC(vp, MNT_WAIT, ctx); - buf_invalidateblks(vp, BUF_WRITE_DATA, 0, 0); + buf_invalidateblks(vp, BUF_WRITE_DATA | BUF_INVALIDATE_LOCKED, 0, 0); } if (UBCINFOEXISTS(vp)) /* * Clean the pages in VM. */ - (void)ubc_sync_range(vp, (off_t)0, ubc_getsize(vp), UBC_PUSHALL); + (void)ubc_msync(vp, (off_t)0, ubc_getsize(vp), NULL, UBC_PUSHALL | UBC_INVALIDATE | UBC_SYNC); } if (active || need_inactive) VNOP_INACTIVE(vp, ctx); +#if NAMEDSTREAMS + if ((is_namedstream != 0) && (vp->v_parent != NULLVP)) { + vnode_t pvp = vp->v_parent; + + /* Delete the shadow stream file before we reclaim its vnode */ + if (vnode_isshadow(vp)) { + vnode_relenamedstream(pvp, vp, ctx); + } + + /* + * No more streams associated with the parent. We + * have a ref on it, so its identity is stable. + * If the parent is on an opaque volume, then we need to know + * whether it has associated named streams. + */ + if (vfs_authopaque(pvp->v_mount)) { + vnode_lock_spin(pvp); + pvp->v_lflag &= ~VL_HASSTREAMS; + vnode_unlock(pvp); + } + } +#endif + /* * Destroy ubc named reference * cluster_release is done on this path @@ -1953,6 +2107,14 @@ vclean(vnode_t vp, int flags) */ ubc_destroy_named(vp); +#if CONFIG_TRIGGERS + /* + * cleanup trigger info from vnode (if any) + */ + if (vp->v_resolve) + vnode_resolver_detach(vp); +#endif + /* * Reclaim the vnode. */ @@ -1960,7 +2122,7 @@ vclean(vnode_t vp, int flags) panic("vclean: cannot reclaim"); // make sure the name & parent ptrs get cleaned out! - vnode_update_identity(vp, NULLVP, NULL, 0, 0, VNODE_UPDATE_PARENT | VNODE_UPDATE_NAME); + vnode_update_identity(vp, NULLVP, NULL, 0, 0, VNODE_UPDATE_PARENT | VNODE_UPDATE_NAME | VNODE_UPDATE_PURGE); vnode_lock(vp); @@ -2002,23 +2164,20 @@ vn_revoke(vnode_t vp, __unused int flags, __unused vfs_context_t a_context) panic("vnop_revoke"); #endif - if (vp->v_flag & VALIASED) { + if (vnode_isaliased(vp)) { /* * If a vgone (or vclean) is already in progress, - * wait until it is done and return. + * return an immediate error */ - vnode_lock(vp); - if (vp->v_lflag & VL_TERMINATE) { - vnode_unlock(vp); + if (vp->v_lflag & VL_TERMINATE) return(ENOENT); - } - vnode_unlock(vp); + /* * Ensure that vp will not be vgone'd while we * are eliminating its aliases. */ SPECHASH_LOCK(); - while (vp->v_flag & VALIASED) { + while ((vp->v_specflags & SI_ALIASED)) { for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type || vp == vq) @@ -2029,7 +2188,7 @@ vn_revoke(vnode_t vp, __unused int flags, __unused vfs_context_t a_context) SPECHASH_LOCK(); break; } - vnode_reclaim_internal(vq, 0, 0, 0); + vnode_reclaim_internal(vq, 0, 1, 0); vnode_put(vq); SPECHASH_LOCK(); break; @@ -2049,14 +2208,16 @@ vn_revoke(vnode_t vp, __unused int flags, __unused vfs_context_t a_context) int vnode_recycle(struct vnode *vp) { - vnode_lock(vp); + vnode_lock_spin(vp); if (vp->v_iocount || vp->v_usecount) { vp->v_lflag |= VL_MARKTERM; vnode_unlock(vp); return(0); } + vnode_lock_convert(vp); vnode_reclaim_internal(vp, 1, 0, 0); + vnode_unlock(vp); return (1); @@ -2113,7 +2274,7 @@ vgone(vnode_t vp, int flags) if (vq == NULL) panic("missing bdev"); } - if (vp->v_flag & VALIASED) { + if (vp->v_specflags & SI_ALIASED) { vx = NULL; for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { if (vq->v_rdev != vp->v_rdev || @@ -2126,8 +2287,8 @@ vgone(vnode_t vp, int flags) if (vx == NULL) panic("missing alias"); if (vq == NULL) - vx->v_flag &= ~VALIASED; - vp->v_flag &= ~VALIASED; + vx->v_specflags &= ~SI_ALIASED; + vp->v_specflags &= ~SI_ALIASED; } SPECHASH_UNLOCK(); { @@ -2182,8 +2343,8 @@ vcount(vnode_t vp) int vid; loop: - if ((vp->v_flag & VALIASED) == 0) - return (vp->v_usecount - vp->v_kusecount); + if (!vnode_isaliased(vp)) + return (vp->v_specinfo->si_opencount); count = 0; SPECHASH_LOCK(); @@ -2209,12 +2370,12 @@ loop: /* * Alias, but not in use, so flush it out. */ - vnode_reclaim_internal(vq, 1, 0, 0); + vnode_reclaim_internal(vq, 1, 1, 0); + vnode_put_locked(vq); vnode_unlock(vq); - vnode_put(vq); goto loop; } - count += (vq->v_usecount - vq->v_kusecount); + count += vq->v_specinfo->si_opencount; } vnode_unlock(vq); @@ -2242,10 +2403,8 @@ int prtactive = 0; /* 1 => print out reclaim of active vnodes */ /* * Print out a description of a vnode. */ -#if !CONFIG_NO_PRINTF_STRINGS static const char *typename[] = { "VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD" }; -#endif void vprint(const char *label, struct vnode *vp) @@ -2254,7 +2413,7 @@ vprint(const char *label, struct vnode *vp) if (label != NULL) printf("%s: ", label); - printf("type %s, usecount %d, writecount %ld", + printf("type %s, usecount %d, writecount %d", typename[vp->v_type], vp->v_usecount, vp->v_writecount); sbuf[0] = '\0'; if (vp->v_flag & VROOT) @@ -2267,7 +2426,7 @@ vprint(const char *label, struct vnode *vp) strlcat(sbuf, "|VNOFLUSH", sizeof(sbuf)); if (vp->v_flag & VBWAIT) strlcat(sbuf, "|VBWAIT", sizeof(sbuf)); - if (vp->v_flag & VALIASED) + if (vnode_isaliased(vp)) strlcat(sbuf, "|VALIASED", sizeof(sbuf)); if (sbuf[0] != '\0') printf(" flags (%s)", &sbuf[1]); @@ -2280,6 +2439,11 @@ vn_getpath(struct vnode *vp, char *pathbuf, int *len) return build_path(vp, pathbuf, *len, len, BUILDPATH_NO_FS_ENTER, vfs_context_current()); } +int +vn_getpath_fsenter(struct vnode *vp, char *pathbuf, int *len) +{ + return build_path(vp, pathbuf, *len, len, 0, vfs_context_current()); +} int vn_getcdhash(struct vnode *vp, off_t offset, unsigned char *cdhash) @@ -2311,32 +2475,47 @@ extension_cmp(const void *a, const void *b) // them (i.e. a short 8 character name can't have an 8 // character extension). // +extern lck_mtx_t *pkg_extensions_lck; + __private_extern__ int -set_package_extensions_table(void *data, int nentries, int maxwidth) +set_package_extensions_table(user_addr_t data, int nentries, int maxwidth) { - char *new_exts; + char *new_exts, *old_exts; int error; if (nentries <= 0 || nentries > 1024 || maxwidth <= 0 || maxwidth > 255) { return EINVAL; } - MALLOC(new_exts, char *, nentries * maxwidth, M_TEMP, M_WAITOK); + + // allocate one byte extra so we can guarantee null termination + MALLOC(new_exts, char *, (nentries * maxwidth) + 1, M_TEMP, M_WAITOK); + if (new_exts == NULL) { + return ENOMEM; + } - error = copyin(CAST_USER_ADDR_T(data), new_exts, nentries * maxwidth); + error = copyin(data, new_exts, nentries * maxwidth); if (error) { FREE(new_exts, M_TEMP); return error; } - if (extension_table) { - FREE(extension_table, M_TEMP); - } + new_exts[(nentries * maxwidth)] = '\0'; // guarantee null termination of the block + + qsort(new_exts, nentries, maxwidth, extension_cmp); + + lck_mtx_lock(pkg_extensions_lck); + + old_exts = extension_table; extension_table = new_exts; nexts = nentries; max_ext_width = maxwidth; - qsort(extension_table, nexts, maxwidth, extension_cmp); + lck_mtx_unlock(pkg_extensions_lck); + + if (old_exts) { + FREE(old_exts, M_TEMP); + } return 0; } @@ -2367,16 +2546,21 @@ is_package_name(const char *name, int len) // advance over the "." name_ext++; + lck_mtx_lock(pkg_extensions_lck); + // now iterate over all the extensions to see if any match ptr = &extension_table[0]; for(i=0; i < nexts; i++, ptr+=max_ext_width) { extlen = strlen(ptr); if (strncasecmp(name_ext, ptr, extlen) == 0 && name_ext[extlen] == '\0') { // aha, a match! + lck_mtx_unlock(pkg_extensions_lck); return 1; } } + lck_mtx_unlock(pkg_extensions_lck); + // if we get here, no extension matched return 0; } @@ -2422,6 +2606,25 @@ vn_path_package_check(__unused vnode_t vp, char *path, int pathlen, int *compone return 0; } +/* + * Determine if a name is inappropriate for a searchfs query. + * This list consists of /System currently. + */ + +int vn_searchfs_inappropriate_name(const char *name, int len) { + const char *bad_names[] = { "System" }; + int bad_len[] = { 6 }; + int i; + + for(i=0; i < (int) (sizeof(bad_names) / sizeof(bad_names[0])); i++) { + if (len == bad_len[i] && strncmp(name, bad_names[i], strlen(bad_names[i]) + 1) == 0) { + return 1; + } + } + + // if we get here, no name matched + return 0; +} /* * Top level filesystem related information gathering. @@ -2436,7 +2639,7 @@ vfs_sysctl(int *name, u_int namelen, user_addr_t oldp, size_t *oldlenp, int *username; u_int usernamelen; int error; - struct vfsconf *vfsc; + struct vfsconf vfsc; /* All non VFS_GENERIC and in VFS_GENERIC, * VFS_MAXTYPENUM, VFS_CONF, VFS_SET_PACKAGE_EXTS @@ -2468,16 +2671,27 @@ vfs_sysctl(int *name, u_int namelen, user_addr_t oldp, size_t *oldlenp, if (namelen < 2) return (EISDIR); /* overloaded */ if (name[0] != VFS_GENERIC) { - for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) - if (vfsp->vfc_typenum == name[0]) + + mount_list_lock(); + for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) + if (vfsp->vfc_typenum == name[0]) { + vfsp->vfc_refcount++; break; + } + mount_list_unlock(); + if (vfsp == NULL) return (ENOTSUP); /* XXX current context proxy for proc p? */ - return ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, + error = ((*vfsp->vfc_vfsops->vfs_sysctl)(&name[1], namelen - 1, oldp, oldlenp, newp, newlen, vfs_context_current())); + + mount_list_lock(); + vfsp->vfc_refcount--; + mount_list_unlock(); + return error; } switch (name[1]) { case VFS_MAXTYPENUM: @@ -2485,31 +2699,31 @@ vfs_sysctl(int *name, u_int namelen, user_addr_t oldp, size_t *oldlenp, case VFS_CONF: if (namelen < 3) return (ENOTDIR); /* overloaded */ + + mount_list_lock(); for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) if (vfsp->vfc_typenum == name[2]) break; - if (vfsp == NULL) + + if (vfsp == NULL) { + mount_list_unlock(); return (ENOTSUP); - vfsc = (struct vfsconf *)vfsp; - if (proc_is64bit(p)) { - struct user_vfsconf usr_vfsc; - usr_vfsc.vfc_vfsops = CAST_USER_ADDR_T(vfsc->vfc_vfsops); - bcopy(vfsc->vfc_name, usr_vfsc.vfc_name, sizeof(usr_vfsc.vfc_name)); - usr_vfsc.vfc_typenum = vfsc->vfc_typenum; - usr_vfsc.vfc_refcount = vfsc->vfc_refcount; - usr_vfsc.vfc_flags = vfsc->vfc_flags; - usr_vfsc.vfc_mountroot = CAST_USER_ADDR_T(vfsc->vfc_mountroot); - usr_vfsc.vfc_next = CAST_USER_ADDR_T(vfsc->vfc_next); - return (sysctl_rdstruct(oldp, oldlenp, newp, &usr_vfsc, - sizeof(usr_vfsc))); - } - else { - return (sysctl_rdstruct(oldp, oldlenp, newp, vfsc, - sizeof(struct vfsconf))); } + + vfsc.vfc_reserved1 = 0; + bcopy(vfsp->vfc_name, vfsc.vfc_name, sizeof(vfsc.vfc_name)); + vfsc.vfc_typenum = vfsp->vfc_typenum; + vfsc.vfc_refcount = vfsp->vfc_refcount; + vfsc.vfc_flags = vfsp->vfc_flags; + vfsc.vfc_reserved2 = 0; + vfsc.vfc_reserved3 = 0; + + mount_list_unlock(); + return (sysctl_rdstruct(oldp, oldlenp, newp, &vfsc, + sizeof(struct vfsconf))); case VFS_SET_PACKAGE_EXTS: - return set_package_extensions_table((void *)name[1], name[2], name[3]); + return set_package_extensions_table((user_addr_t)((unsigned)name[1]), name[2], name[3]); } /* * We need to get back into the general MIB, so we need to re-prepend @@ -2521,7 +2735,7 @@ vfs_sysctl(int *name, u_int namelen, user_addr_t oldp, size_t *oldlenp, bcopy(name, username + 1, namelen * sizeof(*name)); username[0] = CTL_VFS; error = userland_sysctl(p, username, usernamelen, oldp, - oldlenp, 1, newp, newlen, oldlenp); + oldlenp, newp, newlen, oldlenp); FREE(username, M_TEMP); return (error); } @@ -2539,7 +2753,7 @@ sysctl_vnode } SYSCTL_PROC(_kern, KERN_VNODE, vnode, - CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MASKED, + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_MASKED | CTLFLAG_LOCKED, 0, 0, sysctl_vnode, "S,", ""); @@ -2557,7 +2771,7 @@ vfs_mountedon(struct vnode *vp) error = EBUSY; goto out; } - if (vp->v_flag & VALIASED) { + if (vp->v_specflags & SI_ALIASED) { for (vq = *vp->v_hashchain; vq; vq = vq->v_specnext) { if (vq->v_rdev != vp->v_rdev || vq->v_type != vp->v_type) @@ -2599,7 +2813,7 @@ vfs_unmountall(void) continue; } else if (error == EBUSY) { /* If EBUSY is returned, the unmount was already in progress */ - printf("unmount of %x failed (", (unsigned int)mp); + printf("unmount of %p failed (", mp); printf("BUSY)\n"); } mount_list_lock(); @@ -2618,58 +2832,62 @@ vnode_pager_vrele(vnode_t vp) { struct ubc_info *uip; - vnode_lock(vp); + vnode_lock_spin(vp); vp->v_lflag &= ~VNAMED_UBC; uip = vp->v_ubcinfo; vp->v_ubcinfo = UBC_INFO_NULL; - ubc_info_deallocate(uip); - vnode_unlock(vp); + + ubc_info_deallocate(uip); } #include +u_int32_t rootunit = (u_int32_t)-1; + errno_t vfs_init_io_attributes(vnode_t devvp, mount_t mp) { int error; - off_t readblockcnt; - off_t writeblockcnt; - off_t readmaxcnt; - off_t writemaxcnt; - off_t readsegcnt; - off_t writesegcnt; - off_t readsegsize; - off_t writesegsize; - off_t alignment; - u_long blksize; + off_t readblockcnt = 0; + off_t writeblockcnt = 0; + off_t readmaxcnt = 0; + off_t writemaxcnt = 0; + off_t readsegcnt = 0; + off_t writesegcnt = 0; + off_t readsegsize = 0; + off_t writesegsize = 0; + off_t alignment = 0; + off_t ioqueue_depth = 0; + u_int32_t blksize; u_int64_t temp; u_int32_t features; vfs_context_t ctx = vfs_context_current(); - + int isssd = 0; int isvirtual = 0; + + + VNOP_IOCTL(devvp, DKIOCGETTHROTTLEMASK, (caddr_t)&mp->mnt_throttle_mask, 0, NULL); /* - * determine if this mount point exists on the same device as the root - * partition... if so, then it comes under the hard throttle control + * as a reasonable approximation, only use the lowest bit of the mask + * to generate a disk unit number */ - int thisunit = -1; - static int rootunit = -1; + mp->mnt_devbsdunit = num_trailing_0(mp->mnt_throttle_mask); - if (rootunit == -1) { - if (VNOP_IOCTL(rootvp, DKIOCGETBSDUNIT, (caddr_t)&rootunit, 0, ctx)) - rootunit = -1; - else if (rootvp == devvp) - mp->mnt_kern_flag |= MNTK_ROOTDEV; - } - if (devvp != rootvp && rootunit != -1) { - if (VNOP_IOCTL(devvp, DKIOCGETBSDUNIT, (caddr_t)&thisunit, 0, ctx) == 0) { - if (thisunit == rootunit) - mp->mnt_kern_flag |= MNTK_ROOTDEV; - } + if (devvp == rootvp) + rootunit = mp->mnt_devbsdunit; + + if (mp->mnt_devbsdunit == rootunit) { + /* + * this mount point exists on the same device as the root + * partition, so it comes under the hard throttle control... + * this is true even for the root mount point itself + */ + mp->mnt_kern_flag |= MNTK_ROOTDEV; } /* * force the spec device to re-cache @@ -2685,11 +2903,24 @@ vfs_init_io_attributes(vnode_t devvp, mount_t mp) mp->mnt_devblocksize = blksize; + /* + * set the maximum possible I/O size + * this may get clipped to a smaller value + * based on which constraints are being advertised + * and if those advertised constraints result in a smaller + * limit for a given I/O + */ + mp->mnt_maxreadcnt = MAX_UPL_SIZE * PAGE_SIZE; + mp->mnt_maxwritecnt = MAX_UPL_SIZE * PAGE_SIZE; + if (VNOP_IOCTL(devvp, DKIOCISVIRTUAL, (caddr_t)&isvirtual, 0, ctx) == 0) { if (isvirtual) mp->mnt_kern_flag |= MNTK_VIRTUALDEV; } - + if (VNOP_IOCTL(devvp, DKIOCISSOLIDSTATE, (caddr_t)&isssd, 0, ctx) == 0) { + if (isssd) + mp->mnt_kern_flag |= MNTK_SSD; + } if ((error = VNOP_IOCTL(devvp, DKIOCGETFEATURES, (caddr_t)&features, 0, ctx))) return (error); @@ -2730,36 +2961,52 @@ vfs_init_io_attributes(vnode_t devvp, mount_t mp) (caddr_t)&alignment, 0, ctx))) return (error); + if ((error = VNOP_IOCTL(devvp, DKIOCGETCOMMANDPOOLSIZE, + (caddr_t)&ioqueue_depth, 0, ctx))) + return (error); + if (readmaxcnt) - temp = (readmaxcnt > UINT32_MAX) ? UINT32_MAX : readmaxcnt; - else { - if (readblockcnt) { - temp = readblockcnt * blksize; - temp = (temp > UINT32_MAX) ? UINT32_MAX : temp; - } else - temp = MAXPHYS; + mp->mnt_maxreadcnt = (readmaxcnt > UINT32_MAX) ? UINT32_MAX : readmaxcnt; + + if (readblockcnt) { + temp = readblockcnt * blksize; + temp = (temp > UINT32_MAX) ? UINT32_MAX : temp; + + if (temp < mp->mnt_maxreadcnt) + mp->mnt_maxreadcnt = (u_int32_t)temp; } - mp->mnt_maxreadcnt = (u_int32_t)temp; if (writemaxcnt) - temp = (writemaxcnt > UINT32_MAX) ? UINT32_MAX : writemaxcnt; - else { - if (writeblockcnt) { - temp = writeblockcnt * blksize; - temp = (temp > UINT32_MAX) ? UINT32_MAX : temp; - } else - temp = MAXPHYS; + mp->mnt_maxwritecnt = (writemaxcnt > UINT32_MAX) ? UINT32_MAX : writemaxcnt; + + if (writeblockcnt) { + temp = writeblockcnt * blksize; + temp = (temp > UINT32_MAX) ? UINT32_MAX : temp; + + if (temp < mp->mnt_maxwritecnt) + mp->mnt_maxwritecnt = (u_int32_t)temp; } - mp->mnt_maxwritecnt = (u_int32_t)temp; if (readsegcnt) { temp = (readsegcnt > UINT16_MAX) ? UINT16_MAX : readsegcnt; - mp->mnt_segreadcnt = (u_int16_t)temp; + } else { + temp = mp->mnt_maxreadcnt / PAGE_SIZE; + + if (temp > UINT16_MAX) + temp = UINT16_MAX; } + mp->mnt_segreadcnt = (u_int16_t)temp; + if (writesegcnt) { temp = (writesegcnt > UINT16_MAX) ? UINT16_MAX : writesegcnt; - mp->mnt_segwritecnt = (u_int16_t)temp; + } else { + temp = mp->mnt_maxwritecnt / PAGE_SIZE; + + if (temp > UINT16_MAX) + temp = UINT16_MAX; } + mp->mnt_segwritecnt = (u_int16_t)temp; + if (readsegsize) temp = (readsegsize > UINT32_MAX) ? UINT32_MAX : readsegsize; else @@ -2778,26 +3025,44 @@ vfs_init_io_attributes(vnode_t devvp, mount_t mp) temp = 0; mp->mnt_alignmentmask = temp; + + if (ioqueue_depth > MNT_DEFAULT_IOQUEUE_DEPTH) + temp = ioqueue_depth; + else + temp = MNT_DEFAULT_IOQUEUE_DEPTH; + + mp->mnt_ioqueue_depth = temp; + mp->mnt_ioscale = (mp->mnt_ioqueue_depth + (MNT_DEFAULT_IOQUEUE_DEPTH - 1)) / MNT_DEFAULT_IOQUEUE_DEPTH; + + if (mp->mnt_ioscale > 1) + printf("ioqueue_depth = %d, ioscale = %d\n", (int)mp->mnt_ioqueue_depth, (int)mp->mnt_ioscale); + if (features & DK_FEATURE_FORCE_UNIT_ACCESS) mp->mnt_ioflags |= MNT_IOFLAGS_FUA_SUPPORTED; - + if (features & DK_FEATURE_UNMAP) + mp->mnt_ioflags |= MNT_IOFLAGS_UNMAP_SUPPORTED; return (error); } static struct klist fs_klist; +lck_grp_t *fs_klist_lck_grp; +lck_mtx_t *fs_klist_lock; void vfs_event_init(void) { klist_init(&fs_klist); + fs_klist_lck_grp = lck_grp_alloc_init("fs_klist", NULL); + fs_klist_lock = lck_mtx_alloc_init(fs_klist_lck_grp, NULL); } void vfs_event_signal(__unused fsid_t *fsid, u_int32_t event, __unused intptr_t data) { - + lck_mtx_lock(fs_klist_lock); KNOTE(&fs_klist, event); + lck_mtx_unlock(fs_klist_lock); } /* @@ -2896,6 +3161,10 @@ again: return (ENOMEM); MALLOC(fsidlst, fsid_t *, req->oldlen, M_TEMP, M_WAITOK); + if (fsidlst == NULL) { + return (ENOMEM); + } + error = sysctl_vfs_getvfslist(fsidlst, req->oldlen / sizeof(fsid_t), &actual); /* @@ -2921,8 +3190,7 @@ static int sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, struct sysctl_req *req) { - struct vfsidctl vc; - struct user_vfsidctl user_vc; + union union_vfsidctl vc; struct mount *mp; struct vfsstatfs *sp; int *name, flags, namelen; @@ -2935,26 +3203,14 @@ sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, namelen = arg2; is_64_bit = proc_is64bit(p); - if (is_64_bit) { - error = SYSCTL_IN(req, &user_vc, sizeof(user_vc)); - if (error) - goto out; - if (user_vc.vc_vers != VFS_CTL_VERS1) { - error = EINVAL; - goto out; - } - mp = mount_list_lookupby_fsid(&user_vc.vc_fsid, 0, 1); - } - else { - error = SYSCTL_IN(req, &vc, sizeof(vc)); - if (error) - goto out; - if (vc.vc_vers != VFS_CTL_VERS1) { - error = EINVAL; - goto out; - } - mp = mount_list_lookupby_fsid(&vc.vc_fsid, 0, 1); + error = SYSCTL_IN(req, &vc, is_64_bit? sizeof(vc.vc64):sizeof(vc.vc32)); + if (error) + goto out; + if (vc.vc32.vc_vers != VFS_CTL_VERS1) { /* works for 32 and 64 */ + error = EINVAL; + goto out; } + mp = mount_list_lookupby_fsid(&vc.vc32.vc_fsid, 0, 1); /* works for 32 and 64 */ if (mp == NULL) { error = ENOENT; goto out; @@ -2993,12 +3249,12 @@ sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, case VFS_CTL_UMOUNT: req->newidx = 0; if (is_64_bit) { - req->newptr = user_vc.vc_ptr; - req->newlen = (size_t)user_vc.vc_len; + req->newptr = vc.vc64.vc_ptr; + req->newlen = (size_t)vc.vc64.vc_len; } else { - req->newptr = CAST_USER_ADDR_T(vc.vc_ptr); - req->newlen = vc.vc_len; + req->newptr = CAST_USER_ADDR_T(vc.vc32.vc_ptr); + req->newlen = vc.vc32.vc_len; } error = SYSCTL_IN(req, &flags, sizeof(flags)); if (error) @@ -3013,44 +3269,48 @@ sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, case VFS_CTL_STATFS: req->newidx = 0; if (is_64_bit) { - req->newptr = user_vc.vc_ptr; - req->newlen = (size_t)user_vc.vc_len; + req->newptr = vc.vc64.vc_ptr; + req->newlen = (size_t)vc.vc64.vc_len; } else { - req->newptr = CAST_USER_ADDR_T(vc.vc_ptr); - req->newlen = vc.vc_len; + req->newptr = CAST_USER_ADDR_T(vc.vc32.vc_ptr); + req->newlen = vc.vc32.vc_len; } error = SYSCTL_IN(req, &flags, sizeof(flags)); if (error) break; sp = &mp->mnt_vfsstat; - if (((flags & MNT_NOWAIT) == 0 || (flags & MNT_WAIT)) && + if (((flags & MNT_NOWAIT) == 0 || (flags & (MNT_WAIT | MNT_DWAIT))) && (error = vfs_update_vfsstat(mp, ctx, VFS_USER_EVENT))) goto out; if (is_64_bit) { - struct user_statfs sfs; + struct user64_statfs sfs; bzero(&sfs, sizeof(sfs)); sfs.f_flags = mp->mnt_flag & MNT_VISFLAGMASK; sfs.f_type = mp->mnt_vtable->vfc_typenum; - sfs.f_bsize = (user_long_t)sp->f_bsize; - sfs.f_iosize = (user_long_t)sp->f_iosize; - sfs.f_blocks = (user_long_t)sp->f_blocks; - sfs.f_bfree = (user_long_t)sp->f_bfree; - sfs.f_bavail = (user_long_t)sp->f_bavail; - sfs.f_files = (user_long_t)sp->f_files; - sfs.f_ffree = (user_long_t)sp->f_ffree; + sfs.f_bsize = (user64_long_t)sp->f_bsize; + sfs.f_iosize = (user64_long_t)sp->f_iosize; + sfs.f_blocks = (user64_long_t)sp->f_blocks; + sfs.f_bfree = (user64_long_t)sp->f_bfree; + sfs.f_bavail = (user64_long_t)sp->f_bavail; + sfs.f_files = (user64_long_t)sp->f_files; + sfs.f_ffree = (user64_long_t)sp->f_ffree; sfs.f_fsid = sp->f_fsid; sfs.f_owner = sp->f_owner; - strlcpy(sfs.f_fstypename, sp->f_fstypename, MFSNAMELEN); + if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) { + strlcpy(&sfs.f_fstypename[0], &mp->fstypename_override[0], MFSTYPENAMELEN); + } else { + strlcpy(sfs.f_fstypename, sp->f_fstypename, MFSNAMELEN); + } strlcpy(sfs.f_mntonname, sp->f_mntonname, MNAMELEN); strlcpy(sfs.f_mntfromname, sp->f_mntfromname, MNAMELEN); error = SYSCTL_OUT(req, &sfs, sizeof(sfs)); } else { - struct statfs sfs; - bzero(&sfs, sizeof(struct statfs)); + struct user32_statfs sfs; + bzero(&sfs, sizeof(sfs)); sfs.f_flags = mp->mnt_flag & MNT_VISFLAGMASK; sfs.f_type = mp->mnt_vtable->vfc_typenum; @@ -3059,7 +3319,7 @@ sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, * have to fudge the numbers here in that case. We inflate the blocksize in order * to reflect the filesystem size as best we can. */ - if (sp->f_blocks > LONG_MAX) { + if (sp->f_blocks > INT_MAX) { int shift; /* @@ -3072,31 +3332,35 @@ sysctl_vfs_ctlbyfsid(__unused struct sysctl_oid *oidp, void *arg1, int arg2, * being smaller than f_bsize. */ for (shift = 0; shift < 32; shift++) { - if ((sp->f_blocks >> shift) <= LONG_MAX) + if ((sp->f_blocks >> shift) <= INT_MAX) break; - if ((sp->f_bsize << (shift + 1)) > LONG_MAX) + if ((((long long)sp->f_bsize) << (shift + 1)) > INT_MAX) break; } -#define __SHIFT_OR_CLIP(x, s) ((((x) >> (s)) > LONG_MAX) ? LONG_MAX : ((x) >> (s))) - sfs.f_blocks = (long)__SHIFT_OR_CLIP(sp->f_blocks, shift); - sfs.f_bfree = (long)__SHIFT_OR_CLIP(sp->f_bfree, shift); - sfs.f_bavail = (long)__SHIFT_OR_CLIP(sp->f_bavail, shift); +#define __SHIFT_OR_CLIP(x, s) ((((x) >> (s)) > INT_MAX) ? INT_MAX : ((x) >> (s))) + sfs.f_blocks = (user32_long_t)__SHIFT_OR_CLIP(sp->f_blocks, shift); + sfs.f_bfree = (user32_long_t)__SHIFT_OR_CLIP(sp->f_bfree, shift); + sfs.f_bavail = (user32_long_t)__SHIFT_OR_CLIP(sp->f_bavail, shift); #undef __SHIFT_OR_CLIP - sfs.f_bsize = (long)(sp->f_bsize << shift); + sfs.f_bsize = (user32_long_t)(sp->f_bsize << shift); sfs.f_iosize = lmax(sp->f_iosize, sp->f_bsize); } else { - sfs.f_bsize = (long)sp->f_bsize; - sfs.f_iosize = (long)sp->f_iosize; - sfs.f_blocks = (long)sp->f_blocks; - sfs.f_bfree = (long)sp->f_bfree; - sfs.f_bavail = (long)sp->f_bavail; + sfs.f_bsize = (user32_long_t)sp->f_bsize; + sfs.f_iosize = (user32_long_t)sp->f_iosize; + sfs.f_blocks = (user32_long_t)sp->f_blocks; + sfs.f_bfree = (user32_long_t)sp->f_bfree; + sfs.f_bavail = (user32_long_t)sp->f_bavail; } - sfs.f_files = (long)sp->f_files; - sfs.f_ffree = (long)sp->f_ffree; + sfs.f_files = (user32_long_t)sp->f_files; + sfs.f_ffree = (user32_long_t)sp->f_ffree; sfs.f_fsid = sp->f_fsid; sfs.f_owner = sp->f_owner; - strlcpy(sfs.f_fstypename, sp->f_fstypename, MFSNAMELEN); + if (mp->mnt_kern_flag & MNTK_TYPENAME_OVERRIDE) { + strlcpy(&sfs.f_fstypename[0], &mp->fstypename_override[0], MFSTYPENAMELEN); + } else { + strlcpy(sfs.f_fstypename, sp->f_fstypename, MFSNAMELEN); + } strlcpy(sfs.f_mntonname, sp->f_mntonname, MNAMELEN); strlcpy(sfs.f_mntfromname, sp->f_mntfromname, MNAMELEN); @@ -3116,24 +3380,29 @@ out: static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); - -struct filterops fs_filtops = - { 0, filt_fsattach, filt_fsdetach, filt_fsevent }; +struct filterops fs_filtops = { + .f_attach = filt_fsattach, + .f_detach = filt_fsdetach, + .f_event = filt_fsevent, +}; static int filt_fsattach(struct knote *kn) { + lck_mtx_lock(fs_klist_lock); kn->kn_flags |= EV_CLEAR; KNOTE_ATTACH(&fs_klist, kn); + lck_mtx_unlock(fs_klist_lock); return (0); } static void filt_fsdetach(struct knote *kn) { - + lck_mtx_lock(fs_klist_lock); KNOTE_DETACH(&fs_klist, kn); + lck_mtx_unlock(fs_klist_lock); } static int @@ -3189,27 +3458,27 @@ sysctl_vfs_noremotehang(__unused struct sysctl_oid *oidp, } if (pid < 0) - OSBitAndAtomic(~((uint32_t)P_NOREMOTEHANG), (UInt32 *)&p->p_flag); + OSBitAndAtomic(~((uint32_t)P_NOREMOTEHANG), &p->p_flag); else - OSBitOrAtomic(P_NOREMOTEHANG, (UInt32 *)&p->p_flag); + OSBitOrAtomic(P_NOREMOTEHANG, &p->p_flag); proc_rele(p); return (0); } /* the vfs.generic. branch. */ -SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RW|CTLFLAG_LOCKED, NULL, "vfs generic hinge"); +SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RW | CTLFLAG_LOCKED, NULL, "vfs generic hinge"); /* retreive a list of mounted filesystem fsid_t */ -SYSCTL_PROC(_vfs_generic, OID_AUTO, vfsidlist, CTLFLAG_RD, +SYSCTL_PROC(_vfs_generic, OID_AUTO, vfsidlist, CTLFLAG_RD | CTLFLAG_LOCKED, NULL, 0, sysctl_vfs_vfslist, "S,fsid", "List of mounted filesystem ids"); /* perform operations on filesystem via fsid_t */ -SYSCTL_NODE(_vfs_generic, OID_AUTO, ctlbyfsid, CTLFLAG_RW|CTLFLAG_LOCKED, +SYSCTL_NODE(_vfs_generic, OID_AUTO, ctlbyfsid, CTLFLAG_RW | CTLFLAG_LOCKED, sysctl_vfs_ctlbyfsid, "ctlbyfsid"); -SYSCTL_PROC(_vfs_generic, OID_AUTO, noremotehang, CTLFLAG_RW|CTLFLAG_ANYBODY, +SYSCTL_PROC(_vfs_generic, OID_AUTO, noremotehang, CTLFLAG_RW | CTLFLAG_ANYBODY, NULL, 0, sysctl_vfs_noremotehang, "I", "noremotehang"); -long num_reusedvnodes = 0; /* long for OSAddAtomic */ +long num_reusedvnodes = 0; static int new_vnode(vnode_t *vpp) @@ -3217,12 +3486,13 @@ new_vnode(vnode_t *vpp) vnode_t vp; int retries = 0; /* retry incase of tablefull */ int force_alloc = 0, walk_count = 0; - int vpid; + unsigned int vpid; struct timespec ts; struct timeval current_tv; +#ifndef __LP64__ struct unsafe_fsnode *l_unsafefs = 0; +#endif /* __LP64__ */ proc_t curproc = current_proc(); - pid_t current_pid = proc_pid(curproc); retry: microuptime(¤t_tv); @@ -3231,33 +3501,37 @@ retry: vnode_list_lock(); - if ( !TAILQ_EMPTY(&vnode_dead_list)) { - /* - * Can always reuse a dead one + if ((numvnodes - deadvnodes) < desiredvnodes || force_alloc) { + if ( !TAILQ_EMPTY(&vnode_dead_list)) { + /* + * Can always reuse a dead one + */ + vp = TAILQ_FIRST(&vnode_dead_list); + goto steal_this_vp; + } + /* + * no dead vnodes available... if we're under + * the limit, we'll create a new vnode */ - vp = TAILQ_FIRST(&vnode_dead_list); - goto steal_this_vp; - } - /* - * no dead vnodes available... if we're under - * the limit, we'll create a new vnode - */ - if (numvnodes < desiredvnodes || force_alloc) { numvnodes++; vnode_list_unlock(); + MALLOC_ZONE(vp, struct vnode *, sizeof(*vp), M_VNODE, M_WAITOK); bzero((char *)vp, sizeof(*vp)); VLISTNONE(vp); /* avoid double queue removal */ lck_mtx_init(&vp->v_lock, vnode_lck_grp, vnode_lck_attr); + klist_init(&vp->v_knotes); nanouptime(&ts); vp->v_id = ts.tv_nsec; vp->v_flag = VSTANDARD; #if CONFIG_MACF - mac_vnode_label_init(vp); + if (mac_vnode_label_init_needed(vp)) + mac_vnode_label_init(vp); #endif /* MAC */ + vp->v_iocount = 1; goto done; } @@ -3268,21 +3542,26 @@ retry: (current_tv.tv_sec - rage_tv.tv_sec) >= RAGE_TIME_LIMIT)) { TAILQ_FOREACH(vp, &vnode_rage_list, v_freelist) { - if ( !(vp->v_listflag & VLIST_RAGE) || !(vp->v_flag & VRAGE)) - panic("new_vnode: vp on RAGE list not marked both VLIST_RAGE and VRAGE"); - - // skip vnodes which have a dependency on this process - // (i.e. they're vnodes in a disk image and this process - // is diskimages-helper) - // - if (vp->v_mount && vp->v_mount->mnt_dependent_pid != current_pid && vp->v_mount->mnt_dependent_process != curproc) { - break; + if ( !(vp->v_listflag & VLIST_RAGE)) + panic("new_vnode: vp (%p) on RAGE list not marked VLIST_RAGE", vp); + + // if we're a dependency-capable process, skip vnodes that can + // cause recycling deadlocks. (i.e. this process is diskimages + // helper and the vnode is in a disk image). Querying the + // mnt_kern_flag for the mount's virtual device status + // is safer than checking the mnt_dependent_process, which + // may not be updated if there are multiple devnode layers + // in between the disk image and the final consumer. + + if ((curproc->p_flag & P_DEPENDENCY_CAPABLE) == 0 || vp->v_mount == NULL || + (vp->v_mount->mnt_kern_flag & MNTK_VIRTUALDEV) == 0) { + break; } // don't iterate more than MAX_WALK_COUNT vnodes to // avoid keeping the vnode list lock held for too long. if (walk_count++ > MAX_WALK_COUNT) { - vp = NULL; + vp = NULL; break; } } @@ -3295,12 +3574,18 @@ retry: */ walk_count = 0; TAILQ_FOREACH(vp, &vnode_free_list, v_freelist) { - // skip vnodes which have a dependency on this process - // (i.e. they're vnodes in a disk image and this process - // is diskimages-helper) - // - if (vp->v_mount && vp->v_mount->mnt_dependent_pid != current_pid && vp->v_mount->mnt_dependent_process != curproc) { - break; + + // if we're a dependency-capable process, skip vnodes that can + // cause recycling deadlocks. (i.e. this process is diskimages + // helper and the vnode is in a disk image). Querying the + // mnt_kern_flag for the mount's virtual device status + // is safer than checking the mnt_dependent_process, which + // may not be updated if there are multiple devnode layers + // in between the disk image and the final consumer. + + if ((curproc->p_flag & P_DEPENDENCY_CAPABLE) == 0 || vp->v_mount == NULL || + (vp->v_mount->mnt_kern_flag & MNTK_VIRTUALDEV) == 0) { + break; } // don't iterate more than MAX_WALK_COUNT vnodes to @@ -3329,7 +3614,7 @@ retry: } if (vp == NULL) { - /* + /* * we've reached the system imposed maximum number of vnodes * but there isn't a single one available * wait a bit and then retry... if we can't get a vnode @@ -3346,25 +3631,34 @@ retry: log(LOG_EMERG, "%d desired, %d numvnodes, " "%d free, %d dead, %d rage\n", desiredvnodes, numvnodes, freevnodes, deadvnodes, ragevnodes); +#if CONFIG_EMBEDDED + /* + * Running out of vnodes tends to make a system unusable. Start killing + * processes that jetsam knows are killable. + */ + if (jetsam_kill_top_proc(TRUE, kJetsamFlagsKilledVnodes) < 0) { + /* + * If jetsam can't find any more processes to kill and there + * still aren't any free vnodes, panic. Hopefully we'll get a + * panic log to tell us why we ran out. + */ + panic("vnode table is full\n"); + } + + delay_for_interval(1, 1000 * 1000); + goto retry; +#endif + *vpp = NULL; return (ENFILE); } steal_this_vp: vpid = vp->v_id; - /* - * the v_listflag field is - * protected by the vnode_list_lock - */ - if (vp->v_listflag & VLIST_DEAD) - VREMDEAD("new_vnode", vp); - else if (vp->v_listflag & VLIST_RAGE) - VREMRAGE("new_vnode", vp); - else - VREMFREE("new_vnode", vp); - VLISTNONE(vp); + vnode_list_remove_locked(vp); vnode_list_unlock(); + vnode_lock_spin(vp); /* @@ -3407,30 +3701,32 @@ steal_this_vp: vnode_unlock(vp); goto retry; } - OSAddAtomic(1, &num_reusedvnodes); + OSAddAtomicLong(1, &num_reusedvnodes); /* Checks for anyone racing us for recycle */ if (vp->v_type != VBAD) { if (vp->v_lflag & VL_DEAD) - panic("new_vnode: the vnode is VL_DEAD but not VBAD"); + panic("new_vnode(%p): the vnode is VL_DEAD but not VBAD", vp); vnode_lock_convert(vp); - (void)vnode_reclaim_internal(vp, 1, 1, 0); if ((VONLIST(vp))) - panic("new_vnode: vp on list "); + panic("new_vnode(%p): vp on list", vp); if (vp->v_usecount || vp->v_iocount || vp->v_kusecount || (vp->v_lflag & (VNAMED_UBC | VNAMED_MOUNT | VNAMED_FSHASH))) - panic("new_vnode: free vnode still referenced\n"); + panic("new_vnode(%p): free vnode still referenced", vp); if ((vp->v_mntvnodes.tqe_prev != 0) && (vp->v_mntvnodes.tqe_next != 0)) - panic("new_vnode: vnode seems to be on mount list "); + panic("new_vnode(%p): vnode seems to be on mount list", vp); if ( !LIST_EMPTY(&vp->v_nclinks) || !LIST_EMPTY(&vp->v_ncchildren)) - panic("new_vnode: vnode still hooked into the name cache"); + panic("new_vnode(%p): vnode still hooked into the name cache", vp); } + +#ifndef __LP64__ if (vp->v_unsafefs) { l_unsafefs = vp->v_unsafefs; vp->v_unsafefs = (struct unsafe_fsnode *)NULL; } +#endif /* __LP64__ */ #if CONFIG_MACF /* @@ -3442,9 +3738,14 @@ steal_this_vp: if (vp->v_lflag & VL_LABELED) { vnode_lock_convert(vp); mac_vnode_label_recycle(vp); + } else if (mac_vnode_label_init_needed(vp)) { + vnode_lock_convert(vp); + mac_vnode_label_init(vp); } + #endif /* MAC */ + vp->v_iocount = 1; vp->v_lflag = 0; vp->v_writecount = 0; vp->v_references = 0; @@ -3456,10 +3757,13 @@ steal_this_vp: vnode_unlock(vp); +#ifndef __LP64__ if (l_unsafefs) { lck_mtx_destroy(&l_unsafefs->fsnodelock, vnode_lck_grp); FREE_ZONE((void *)l_unsafefs, sizeof(struct unsafe_fsnode), M_UNSAFEFS); } +#endif /* __LP64__ */ + done: *vpp = vp; @@ -3501,7 +3805,9 @@ vnode_get(struct vnode *vp) int vnode_get_locked(struct vnode *vp) { - +#if DIAGNOSTIC + lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED); +#endif if ((vp->v_iocount == 0) && (vp->v_lflag & (VL_TERMINATE | VL_DEAD))) { return(ENOENT); } @@ -3512,10 +3818,27 @@ vnode_get_locked(struct vnode *vp) return (0); } +/* + * vnode_getwithvid() cuts in line in front of a vnode drain (that is, + * while the vnode is draining, but at no point after that) to prevent + * deadlocks when getting vnodes from filesystem hashes while holding + * resources that may prevent other iocounts from being released. + */ +int +vnode_getwithvid(vnode_t vp, uint32_t vid) +{ + return(vget_internal(vp, vid, ( VNODE_NODEAD | VNODE_WITHID | VNODE_DRAINO ))); +} + +/* + * vnode_getwithvid_drainok() is like vnode_getwithvid(), but *does* block behind a vnode + * drain; it exists for use in the VFS name cache, where we really do want to block behind + * vnode drain to prevent holding off an unmount. + */ int -vnode_getwithvid(vnode_t vp, int vid) +vnode_getwithvid_drainok(vnode_t vp, uint32_t vid) { - return(vget_internal(vp, vid, ( VNODE_NODEAD| VNODE_WITHID))); + return(vget_internal(vp, vid, ( VNODE_NODEAD | VNODE_WITHID ))); } int @@ -3525,6 +3848,12 @@ vnode_getwithref(vnode_t vp) } +__private_extern__ int +vnode_getalways(vnode_t vp) +{ + return(vget_internal(vp, 0, VNODE_ALWAYS)); +} + int vnode_put(vnode_t vp) { @@ -3542,6 +3871,9 @@ vnode_put_locked(vnode_t vp) { vfs_context_t ctx = vfs_context_current(); /* hoist outside loop */ +#if DIAGNOSTIC + lck_mtx_assert(&vp->v_lock, LCK_MTX_ASSERT_OWNED); +#endif retry: if (vp->v_iocount < 1) panic("vnode_put(%p): iocount < 1", vp); @@ -3550,7 +3882,7 @@ retry: vnode_dropiocount(vp); return(0); } - if ((vp->v_lflag & (VL_MARKTERM | VL_TERMINATE | VL_DEAD | VL_NEEDINACTIVE)) == VL_NEEDINACTIVE) { + if ((vp->v_lflag & (VL_DEAD | VL_NEEDINACTIVE)) == VL_NEEDINACTIVE) { vp->v_lflag &= ~VL_NEEDINACTIVE; vnode_unlock(vp); @@ -3573,7 +3905,7 @@ retry: if ((vp->v_lflag & (VL_MARKTERM | VL_TERMINATE | VL_DEAD)) == VL_MARKTERM) { vnode_lock_convert(vp); - vnode_reclaim_internal(vp, 1, 0, 0); + vnode_reclaim_internal(vp, 1, 1, 0); } vnode_dropiocount(vp); vnode_list_add(vp); @@ -3615,17 +3947,15 @@ out: errno_t vnode_resume(vnode_t vp) { + if ((vp->v_lflag & VL_SUSPENDED) && vp->v_owner == current_thread()) { - vnode_lock_spin(vp); - - if (vp->v_owner == current_thread()) { + vnode_lock_spin(vp); vp->v_lflag &= ~VL_SUSPENDED; vp->v_owner = NULL; vnode_unlock(vp); - wakeup(&vp->v_iocount); - } else - vnode_unlock(vp); + wakeup(&vp->v_iocount); + } return(0); } @@ -3665,7 +3995,7 @@ vnode_drain(vnode_t vp) { if (vp->v_lflag & VL_DRAIN) { - panic("vnode_drain: recursuve drain"); + panic("vnode_drain: recursive drain"); return(ENOENT); } vp->v_lflag |= VL_DRAIN; @@ -3673,13 +4003,16 @@ vnode_drain(vnode_t vp) while (vp->v_iocount > 1) msleep(&vp->v_iocount, &vp->v_lock, PVFS, "vnode_drain", NULL); + + vp->v_lflag &= ~VL_DRAIN; + return(0); } /* * if the number of recent references via vnode_getwithvid or vnode_getwithref - * exceeds this threshhold, than 'UN-AGE' the vnode by removing it from + * exceeds this threshold, than 'UN-AGE' the vnode by removing it from * the LRU list if it's currently on it... once the iocount and usecount both drop * to 0, it will get put back on the end of the list, effectively making it younger * this allows us to keep actively referenced vnodes in the list without having @@ -3688,11 +4021,13 @@ vnode_drain(vnode_t vp) */ #define UNAGE_THRESHHOLD 25 -static errno_t -vnode_getiocount(vnode_t vp, int vid, int vflags) +errno_t +vnode_getiocount(vnode_t vp, unsigned int vid, int vflags) { int nodead = vflags & VNODE_NODEAD; int nosusp = vflags & VNODE_NOSUSPEND; + int always = vflags & VNODE_ALWAYS; + int beatdrain = vflags & VNODE_DRAINO; for (;;) { /* @@ -3721,6 +4056,21 @@ vnode_getiocount(vnode_t vp, int vid, int vflags) (vp->v_owner == current_thread())) { break; } + + if (always != 0) + break; + + /* + * In some situations, we want to get an iocount + * even if the vnode is draining to prevent deadlock, + * e.g. if we're in the filesystem, potentially holding + * resources that could prevent other iocounts from + * being released. + */ + if (beatdrain && (vp->v_lflag & VL_DRAIN)) { + break; + } + vnode_lock_convert(vp); if (vp->v_lflag & VL_TERMINATE) { @@ -3730,7 +4080,7 @@ vnode_getiocount(vnode_t vp, int vid, int vflags) } else msleep(&vp->v_iocount, &vp->v_lock, PVFS, "vnode_getiocount", NULL); } - if (vid != vp->v_id) { + if (((vflags & VNODE_WITHID) != 0) && vid != vp->v_id) { return(ENOENT); } if (++vp->v_references >= UNAGE_THRESHHOLD) { @@ -3754,10 +4104,8 @@ vnode_dropiocount (vnode_t vp) #ifdef JOE_DEBUG record_vp(vp, -1); #endif - if ((vp->v_lflag & (VL_DRAIN | VL_SUSPENDED)) && (vp->v_iocount <= 1)) { - vnode_lock_convert(vp); + if ((vp->v_lflag & (VL_DRAIN | VL_SUSPENDED)) && (vp->v_iocount <= 1)) wakeup(&vp->v_iocount); - } } @@ -3783,22 +4131,36 @@ vnode_reclaim_internal(struct vnode * vp, int locked, int reuse, int flags) vn_clearunionwait(vp, 1); - if (vnode_drain(vp)) { - panic("vnode drain failed"); - vnode_unlock(vp); - return; - } + vnode_drain(vp); + isfifo = (vp->v_type == VFIFO); if (vp->v_type != VBAD) vgone(vp, flags); /* clean and reclaim the vnode */ /* - * give the vnode a new identity so - * that vnode_getwithvid will fail - * on any stale cache accesses + * give the vnode a new identity so that vnode_getwithvid will fail + * on any stale cache accesses... + * grab the list_lock so that if we're in "new_vnode" + * behind the list_lock trying to steal this vnode, the v_id is stable... + * once new_vnode drops the list_lock, it will block trying to take + * the vnode lock until we release it... at that point it will evaluate + * whether the v_vid has changed + * also need to make sure that the vnode isn't on a list where "new_vnode" + * can find it after the v_id has been bumped until we are completely done + * with the vnode (i.e. putting it back on a list has to be the very last + * thing we do to this vnode... many of the callers of vnode_reclaim_internal + * are holding an io_count on the vnode... they need to drop the io_count + * BEFORE doing a vnode_list_add or make sure to hold the vnode lock until + * they are completely done with the vnode */ + vnode_list_lock(); + + vnode_list_remove_locked(vp); vp->v_id++; + + vnode_list_unlock(); + if (isfifo) { struct fifoinfo * fip; @@ -3806,13 +4168,12 @@ vnode_reclaim_internal(struct vnode * vp, int locked, int reuse, int flags) vp->v_fifoinfo = NULL; FREE(fip, M_TEMP); } - vp->v_type = VBAD; if (vp->v_data) panic("vnode_reclaim_internal: cleaned vnode isn't"); if (vp->v_numoutput) - panic("vnode_reclaim_internal: Clean vnode has pending I/O's"); + panic("vnode_reclaim_internal: clean vnode has pending I/O's"); if (UBCINFOEXISTS(vp)) panic("vnode_reclaim_internal: ubcinfo not cleaned"); if (vp->v_parent) @@ -3823,19 +4184,22 @@ vnode_reclaim_internal(struct vnode * vp, int locked, int reuse, int flags) vp->v_socket = NULL; vp->v_lflag &= ~VL_TERMINATE; - vp->v_lflag &= ~VL_DRAIN; vp->v_owner = NULL; + KNOTE(&vp->v_knotes, NOTE_REVOKE); + + /* Make sure that when we reuse the vnode, no knotes left over */ + klist_init(&vp->v_knotes); + if (vp->v_lflag & VL_TERMWANT) { vp->v_lflag &= ~VL_TERMWANT; wakeup(&vp->v_lflag); } - if (!reuse && vp->v_usecount == 0) { + if (!reuse) { /* * make sure we get on the - * dead list + * dead list if appropriate */ - vnode_list_remove(vp); vnode_list_add(vp); } if (!locked) @@ -3846,10 +4210,9 @@ vnode_reclaim_internal(struct vnode * vp, int locked, int reuse, int flags) * The following api creates a vnode and associates all the parameter specified in vnode_fsparam * structure and returns a vnode handle with a reference. device aliasing is handled here so checkalias * is obsoleted by this. - * vnode_create(int flavor, size_t size, void * param, vnode_t *vp) */ int -vnode_create(int flavor, size_t size, void *data, vnode_t *vpp) +vnode_create(uint32_t flavor, uint32_t size, void *data, vnode_t *vpp) { int error; int insert = 1; @@ -3859,155 +4222,210 @@ vnode_create(int flavor, size_t size, void *data, vnode_t *vpp) struct uthread *ut; struct componentname *cnp; struct vnode_fsparam *param = (struct vnode_fsparam *)data; - - if (flavor == VNCREATE_FLAVOR && (size == VCREATESIZE) && param) { - if ( (error = new_vnode(&vp)) ) { +#if CONFIG_TRIGGERS + struct vnode_trigger_param *tinfo = NULL; +#endif + if (param == NULL) + return (EINVAL); + +#if CONFIG_TRIGGERS + if ((flavor == VNCREATE_TRIGGER) && (size == VNCREATE_TRIGGER_SIZE)) { + tinfo = (struct vnode_trigger_param *)data; + + /* Validate trigger vnode input */ + if ((param->vnfs_vtype != VDIR) || + (tinfo->vnt_resolve_func == NULL) || + (tinfo->vnt_flags & ~VNT_VALID_MASK)) { + return (EINVAL); + } + /* Fall through a normal create (params will be the same) */ + flavor = VNCREATE_FLAVOR; + size = VCREATESIZE; + } +#endif + if ((flavor != VNCREATE_FLAVOR) || (size != VCREATESIZE)) + return (EINVAL); + + if ( (error = new_vnode(&vp)) ) + return(error); + + dvp = param->vnfs_dvp; + cnp = param->vnfs_cnp; + + vp->v_op = param->vnfs_vops; + vp->v_type = param->vnfs_vtype; + vp->v_data = param->vnfs_fsnode; + + if (param->vnfs_markroot) + vp->v_flag |= VROOT; + if (param->vnfs_marksystem) + vp->v_flag |= VSYSTEM; + if (vp->v_type == VREG) { + error = ubc_info_init_withsize(vp, param->vnfs_filesize); + if (error) { +#ifdef JOE_DEBUG + record_vp(vp, 1); +#endif + vp->v_mount = NULL; + vp->v_op = dead_vnodeop_p; + vp->v_tag = VT_NON; + vp->v_data = NULL; + vp->v_type = VBAD; + vp->v_lflag |= VL_DEAD; + + vnode_put(vp); return(error); - } else { - dvp = param->vnfs_dvp; - cnp = param->vnfs_cnp; + } + } +#ifdef JOE_DEBUG + record_vp(vp, 1); +#endif - vp->v_op = param->vnfs_vops; - vp->v_type = param->vnfs_vtype; - vp->v_data = param->vnfs_fsnode; - vp->v_iocount = 1; - - if (param->vnfs_markroot) - vp->v_flag |= VROOT; - if (param->vnfs_marksystem) - vp->v_flag |= VSYSTEM; - if (vp->v_type == VREG) { - error = ubc_info_init_withsize(vp, param->vnfs_filesize); - if (error) { +#if CONFIG_TRIGGERS + /* + * For trigger vnodes, attach trigger info to vnode + */ + if ((vp->v_type == VDIR) && (tinfo != NULL)) { + /* + * Note: has a side effect of incrementing trigger count on the + * mount if successful, which we would need to undo on a + * subsequent failure. + */ #ifdef JOE_DEBUG - record_vp(vp, 1); + record_vp(vp, -1); #endif - vp->v_mount = NULL; - vp->v_op = dead_vnodeop_p; - vp->v_tag = VT_NON; - vp->v_data = NULL; - vp->v_type = VBAD; - vp->v_lflag |= VL_DEAD; - - vnode_put(vp); - return(error); - } - } + error = vnode_resolver_create(param->vnfs_mp, vp, tinfo, FALSE); + if (error) { + printf("vnode_create: vnode_resolver_create() err %d\n", error); + vp->v_mount = NULL; + vp->v_op = dead_vnodeop_p; + vp->v_tag = VT_NON; + vp->v_data = NULL; + vp->v_type = VBAD; + vp->v_lflag |= VL_DEAD; #ifdef JOE_DEBUG record_vp(vp, 1); #endif - if (vp->v_type == VCHR || vp->v_type == VBLK) { - - vp->v_tag = VT_DEVFS; /* callers will reset if needed (bdevvp) */ - - if ( (nvp = checkalias(vp, param->vnfs_rdev)) ) { - /* - * if checkalias returns a vnode, it will be locked - * - * first get rid of the unneeded vnode we acquired - */ - vp->v_data = NULL; - vp->v_op = spec_vnodeop_p; - vp->v_type = VBAD; - vp->v_lflag = VL_DEAD; - vp->v_data = NULL; - vp->v_tag = VT_NON; - vnode_put(vp); + vnode_put(vp); + return (error); + } + } +#endif + if (vp->v_type == VCHR || vp->v_type == VBLK) { - /* - * switch to aliased vnode and finish - * preparing it - */ - vp = nvp; - - vclean(vp, 0); - vp->v_op = param->vnfs_vops; - vp->v_type = param->vnfs_vtype; - vp->v_data = param->vnfs_fsnode; - vp->v_lflag = 0; - vp->v_mount = NULL; - insmntque(vp, param->vnfs_mp); - insert = 0; - vnode_unlock(vp); - } - } + vp->v_tag = VT_DEVFS; /* callers will reset if needed (bdevvp) */ - if (vp->v_type == VFIFO) { - struct fifoinfo *fip; + if ( (nvp = checkalias(vp, param->vnfs_rdev)) ) { + /* + * if checkalias returns a vnode, it will be locked + * + * first get rid of the unneeded vnode we acquired + */ + vp->v_data = NULL; + vp->v_op = spec_vnodeop_p; + vp->v_type = VBAD; + vp->v_lflag = VL_DEAD; + vp->v_data = NULL; + vp->v_tag = VT_NON; + vnode_put(vp); - MALLOC(fip, struct fifoinfo *, - sizeof(*fip), M_TEMP, M_WAITOK); - bzero(fip, sizeof(struct fifoinfo )); - vp->v_fifoinfo = fip; - } - /* The file systems usually pass the address of the location where - * where there store the vnode pointer. When we add the vnode in mount - * point and name cache they are discoverable. So the file system node - * will have the connection to vnode setup by then + /* + * switch to aliased vnode and finish + * preparing it */ - *vpp = vp; + vp = nvp; - /* Add fs named reference. */ - if (param->vnfs_flags & VNFS_ADDFSREF) { - vp->v_lflag |= VNAMED_FSHASH; - } - if (param->vnfs_mp) { - if (param->vnfs_mp->mnt_kern_flag & MNTK_LOCK_LOCAL) - vp->v_flag |= VLOCKLOCAL; - if (insert) { - /* - * enter in mount vnode list - */ - insmntque(vp, param->vnfs_mp); - } -#ifdef INTERIM_FSNODE_LOCK - if (param->vnfs_mp->mnt_vtable->vfc_threadsafe == 0) { - MALLOC_ZONE(vp->v_unsafefs, struct unsafe_fsnode *, - sizeof(struct unsafe_fsnode), M_UNSAFEFS, M_WAITOK); - vp->v_unsafefs->fsnode_count = 0; - vp->v_unsafefs->fsnodeowner = (void *)NULL; - lck_mtx_init(&vp->v_unsafefs->fsnodelock, vnode_lck_grp, vnode_lck_attr); - } -#endif /* INTERIM_FSNODE_LOCK */ - } - if (dvp && vnode_ref(dvp) == 0) { - vp->v_parent = dvp; - } - if (cnp) { - if (dvp && ((param->vnfs_flags & (VNFS_NOCACHE | VNFS_CANTCACHE)) == 0)) { - /* - * enter into name cache - * we've got the info to enter it into the name cache now - */ - cache_enter(dvp, vp, cnp); - } - vp->v_name = vfs_addname(cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_hash, 0); - if ((cnp->cn_flags & UNIONCREATED) == UNIONCREATED) - vp->v_flag |= VISUNION; - } - if ((param->vnfs_flags & VNFS_CANTCACHE) == 0) { - /* - * this vnode is being created as cacheable in the name cache - * this allows us to re-enter it in the cache - */ - vp->v_flag |= VNCACHEABLE; - } - ut = get_bsdthread_info(current_thread()); - - if ((current_proc()->p_lflag & P_LRAGE_VNODES) || - (ut->uu_flag & UT_RAGE_VNODES)) { - /* - * process has indicated that it wants any - * vnodes created on its behalf to be rapidly - * aged to reduce the impact on the cached set - * of vnodes - */ - vp->v_flag |= VRAGE; - } - return(0); + vclean(vp, 0); + vp->v_op = param->vnfs_vops; + vp->v_type = param->vnfs_vtype; + vp->v_data = param->vnfs_fsnode; + vp->v_lflag = 0; + vp->v_mount = NULL; + insmntque(vp, param->vnfs_mp); + insert = 0; + vnode_unlock(vp); + } + } + + if (vp->v_type == VFIFO) { + struct fifoinfo *fip; + + MALLOC(fip, struct fifoinfo *, + sizeof(*fip), M_TEMP, M_WAITOK); + bzero(fip, sizeof(struct fifoinfo )); + vp->v_fifoinfo = fip; + } + /* The file systems must pass the address of the location where + * they store the vnode pointer. When we add the vnode into the mount + * list and name cache they become discoverable. So the file system node + * must have the connection to vnode setup by then + */ + *vpp = vp; + + /* Add fs named reference. */ + if (param->vnfs_flags & VNFS_ADDFSREF) { + vp->v_lflag |= VNAMED_FSHASH; + } + if (param->vnfs_mp) { + if (param->vnfs_mp->mnt_kern_flag & MNTK_LOCK_LOCAL) + vp->v_flag |= VLOCKLOCAL; + if (insert) { + if ((vp->v_freelist.tqe_prev != (struct vnode **)0xdeadb)) + panic("insmntque: vp on the free list\n"); + + /* + * enter in mount vnode list + */ + insmntque(vp, param->vnfs_mp); } +#ifndef __LP64__ + if ((param->vnfs_mp->mnt_vtable->vfc_vfsflags & VFC_VFSTHREADSAFE) == 0) { + MALLOC_ZONE(vp->v_unsafefs, struct unsafe_fsnode *, + sizeof(struct unsafe_fsnode), M_UNSAFEFS, M_WAITOK); + vp->v_unsafefs->fsnode_count = 0; + vp->v_unsafefs->fsnodeowner = (void *)NULL; + lck_mtx_init(&vp->v_unsafefs->fsnodelock, vnode_lck_grp, vnode_lck_attr); + } +#endif /* __LP64__ */ + } + if (dvp && vnode_ref(dvp) == 0) { + vp->v_parent = dvp; + } + if (cnp) { + if (dvp && ((param->vnfs_flags & (VNFS_NOCACHE | VNFS_CANTCACHE)) == 0)) { + /* + * enter into name cache + * we've got the info to enter it into the name cache now + * cache_enter_create will pick up an extra reference on + * the name entered into the string cache + */ + vp->v_name = cache_enter_create(dvp, vp, cnp); + } else + vp->v_name = vfs_addname(cnp->cn_nameptr, cnp->cn_namelen, cnp->cn_hash, 0); + + if ((cnp->cn_flags & UNIONCREATED) == UNIONCREATED) + vp->v_flag |= VISUNION; + } + if ((param->vnfs_flags & VNFS_CANTCACHE) == 0) { + /* + * this vnode is being created as cacheable in the name cache + * this allows us to re-enter it in the cache + */ + vp->v_flag |= VNCACHEABLE; } - return (EINVAL); + ut = get_bsdthread_info(current_thread()); + + if ((current_proc()->p_lflag & P_LRAGE_VNODES) || + (ut->uu_flag & UT_RAGE_VNODES)) { + /* + * process has indicated that it wants any + * vnodes created on its behalf to be rapidly + * aged to reduce the impact on the cached set + * of vnodes + */ + vp->v_flag |= VRAGE; + } + return (0); } int @@ -4037,13 +4455,14 @@ vnode_removefsref(vnode_t vp) int -vfs_iterate(__unused int flags, int (*callout)(mount_t, void *), void *arg) +vfs_iterate(int flags, int (*callout)(mount_t, void *), void *arg) { mount_t mp; int ret = 0; fsid_t * fsid_list; int count, actualcount, i; void * allocmem; + int indx_start, indx_stop, indx_incr; count = mount_getvfscnt(); count += 10; @@ -4053,7 +4472,21 @@ vfs_iterate(__unused int flags, int (*callout)(mount_t, void *), void *arg) actualcount = mount_fillfsids(fsid_list, count); - for (i=0; i< actualcount; i++) { + /* + * Establish the iteration direction + * VFS_ITERATE_TAIL_FIRST overrides default head first order (oldest first) + */ + if (flags & VFS_ITERATE_TAIL_FIRST) { + indx_start = actualcount - 1; + indx_stop = -1; + indx_incr = -1; + } else /* Head first by default */ { + indx_start = 0; + indx_stop = actualcount; + indx_incr = 1; + } + + for (i=indx_start; i != indx_stop; i += indx_incr) { /* obtain the mount point with iteration reference */ mp = mount_list_lookupby_fsid(&fsid_list[i], 0, 1); @@ -4186,13 +4619,22 @@ vfs_update_vfsstat(mount_t mp, vfs_context_t ctx, __unused int eventtype) return(0); } -void +int mount_list_add(mount_t mp) { + int res; + mount_list_lock(); - TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); - nummounts++; + if (system_inshutdown != 0) { + res = -1; + } else { + TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); + nummounts++; + res = 0; + } mount_list_unlock(); + + return res; } void @@ -4206,7 +4648,6 @@ mount_list_remove(mount_t mp) mount_list_unlock(); } -#if CONFIG_VOLFS mount_t mount_lookupby_volfsid(int volfs_id, int withref) { @@ -4240,8 +4681,6 @@ mount_lookupby_volfsid(int volfs_id, int withref) out: return(cur_mount); } -#endif - mount_t mount_list_lookupby_fsid(fsid_t *fsid, int locked, int withref) @@ -4272,7 +4711,7 @@ vnode_lookup(const char *path, int flags, vnode_t *vpp, vfs_context_t ctx) { struct nameidata nd; int error; - u_long ndflags = 0; + u_int32_t ndflags = 0; if (ctx == NULL) { /* XXX technically an error */ ctx = vfs_context_current(); @@ -4289,7 +4728,8 @@ vnode_lookup(const char *path, int flags, vnode_t *vpp, vfs_context_t ctx) ndflags |= DOWHITEOUT; /* XXX AUDITVNPATH1 needed ? */ - NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx); + NDINIT(&nd, LOOKUP, OP_LOOKUP, ndflags, UIO_SYSSPACE, + CAST_USER_ADDR_T(path), ctx); if ((error = namei(&nd))) return (error); @@ -4304,7 +4744,7 @@ vnode_open(const char *path, int fmode, int cmode, int flags, vnode_t *vpp, vfs_ { struct nameidata nd; int error; - u_long ndflags = 0; + u_int32_t ndflags = 0; int lflags = flags; if (ctx == NULL) { /* XXX technically an error */ @@ -4325,7 +4765,8 @@ vnode_open(const char *path, int fmode, int cmode, int flags, vnode_t *vpp, vfs_ ndflags |= DOWHITEOUT; /* XXX AUDITVNPATH1 needed ? */ - NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, CAST_USER_ADDR_T(path), ctx); + NDINIT(&nd, LOOKUP, OP_OPEN, ndflags, UIO_SYSSPACE, + CAST_USER_ADDR_T(path), ctx); if ((error = vn_open(&nd, fmode, cmode))) *vpp = NULL; @@ -4378,6 +4819,18 @@ vnode_setsize(vnode_t vp, off_t size, int ioflag, vfs_context_t ctx) return(vnode_setattr(vp, &va, ctx)); } +static int +vn_create_reg(vnode_t dvp, vnode_t *vpp, struct nameidata *ndp, struct vnode_attr *vap, uint32_t flags, int fmode, uint32_t *statusp, vfs_context_t ctx) +{ + /* Only use compound VNOP for compound operation */ + if (vnode_compound_open_available(dvp) && ((flags & VN_CREATE_DOOPEN) != 0)) { + *vpp = NULLVP; + return VNOP_COMPOUND_OPEN(dvp, vpp, ndp, VNOP_COMPOUND_OPEN_DO_CREATE, fmode, statusp, vap, ctx); + } else { + return VNOP_CREATE(dvp, vpp, &ndp->ni_cnd, vap, ctx); + } +} + /* * Create a filesystem object of arbitrary type with arbitrary attributes in * the spevied directory with the specified name. @@ -4420,70 +4873,48 @@ vnode_setsize(vnode_t vp, off_t size, int ioflag, vfs_context_t ctx) * in the code they originated. */ errno_t -vn_create(vnode_t dvp, vnode_t *vpp, struct componentname *cnp, struct vnode_attr *vap, int flags, vfs_context_t ctx) +vn_create(vnode_t dvp, vnode_t *vpp, struct nameidata *ndp, struct vnode_attr *vap, uint32_t flags, int fmode, uint32_t *statusp, vfs_context_t ctx) { - kauth_acl_t oacl, nacl; - int initial_acl; - errno_t error; + errno_t error, old_error; vnode_t vp = (vnode_t)0; + boolean_t batched; + struct componentname *cnp; + uint32_t defaulted; + cnp = &ndp->ni_cnd; error = 0; - oacl = nacl = NULL; - initial_acl = 0; + batched = namei_compound_available(dvp, ndp) ? TRUE : FALSE; KAUTH_DEBUG("%p CREATE - '%s'", dvp, cnp->cn_nameptr); + if (flags & VN_CREATE_NOINHERIT) + vap->va_vaflags |= VA_NOINHERIT; + if (flags & VN_CREATE_NOAUTH) + vap->va_vaflags |= VA_NOAUTH; /* - * Handle ACL inheritance. + * Handle ACL inheritance, initialize vap. */ - if (!(flags & VN_CREATE_NOINHERIT) && vfs_extendedsecurity(dvp->v_mount)) { - /* save the original filesec */ - if (VATTR_IS_ACTIVE(vap, va_acl)) { - initial_acl = 1; - oacl = vap->va_acl; - } - - vap->va_acl = NULL; - if ((error = kauth_acl_inherit(dvp, - oacl, - &nacl, - vap->va_type == VDIR, - ctx)) != 0) { - KAUTH_DEBUG("%p CREATE - error %d processing inheritance", dvp, error); - return(error); - } + error = vn_attribute_prepare(dvp, vap, &defaulted, ctx); + if (error) { + return error; + } - /* - * If the generated ACL is NULL, then we can save ourselves some effort - * by clearing the active bit. - */ - if (nacl == NULL) { - VATTR_CLEAR_ACTIVE(vap, va_acl); - } else { - VATTR_SET(vap, va_acl, nacl); - } + if (vap->va_type != VREG && (fmode != 0 || (flags & VN_CREATE_DOOPEN) || statusp)) { + panic("Open parameters, but not a regular file."); } - - /* - * Check and default new attributes. - * This will set va_uid, va_gid, va_mode and va_create_time at least, if the caller - * hasn't supplied them. - */ - if ((error = vnode_authattr_new(dvp, vap, flags & VN_CREATE_NOAUTH, ctx)) != 0) { - KAUTH_DEBUG("%p CREATE - error %d handing/defaulting attributes", dvp, error); - goto out; + if ((fmode != 0) && ((flags & VN_CREATE_DOOPEN) == 0)) { + panic("Mode for open, but not trying to open..."); } - /* * Create the requested node. */ switch(vap->va_type) { case VREG: - error = VNOP_CREATE(dvp, vpp, cnp, vap, ctx); + error = vn_create_reg(dvp, vpp, ndp, vap, flags, fmode, statusp, ctx); break; case VDIR: - error = VNOP_MKDIR(dvp, vpp, cnp, vap, ctx); + error = vn_mkdir(dvp, vpp, ndp, vap, ctx); break; case VSOCK: case VFIFO: @@ -4500,10 +4931,11 @@ vn_create(vnode_t dvp, vnode_t *vpp, struct componentname *cnp, struct vnode_att } vp = *vpp; + old_error = error; + #if CONFIG_MACF if (!(flags & VN_CREATE_NOLABEL)) { - error = vnode_label(vnode_mount(vp), dvp, vp, cnp, - VNODE_LABEL_CREATE|VNODE_LABEL_NEEDREF, ctx); + error = vnode_label(vnode_mount(vp), dvp, vp, cnp, VNODE_LABEL_CREATE, ctx); if (error) goto error; } @@ -4520,24 +4952,22 @@ vn_create(vnode_t dvp, vnode_t *vpp, struct componentname *cnp, struct vnode_att #if CONFIG_MACF error: #endif - if ((error != 0 ) && (vp != (vnode_t)0)) { - *vpp = (vnode_t) 0; - vnode_put(vp); + if ((error != 0) && (vp != (vnode_t)0)) { + + /* If we've done a compound open, close */ + if (batched && (old_error == 0) && (vap->va_type == VREG)) { + VNOP_CLOSE(vp, fmode, ctx); + } + + /* Need to provide notifications if a create succeeded */ + if (!batched) { + *vpp = (vnode_t) 0; + vnode_put(vp); + } } out: - /* - * If the caller supplied a filesec in vap, it has been replaced - * now by the post-inheritance copy. We need to put the original back - * and free the inherited product. - */ - if (initial_acl) { - VATTR_SET(vap, va_acl, oacl); - } else { - VATTR_CLEAR_ACTIVE(vap, va_acl); - } - if (nacl != NULL) - kauth_acl_free(nacl); + vn_attribute_cleanup(vap, defaulted); return(error); } @@ -4568,6 +4998,433 @@ vnode_authorize_init(void) vnode_scope = kauth_register_scope(KAUTH_SCOPE_VNODE, vnode_authorize_callback, NULL); } +#define VATTR_PREPARE_DEFAULTED_UID 0x1 +#define VATTR_PREPARE_DEFAULTED_GID 0x2 +#define VATTR_PREPARE_DEFAULTED_MODE 0x4 + +int +vn_attribute_prepare(vnode_t dvp, struct vnode_attr *vap, uint32_t *defaulted_fieldsp, vfs_context_t ctx) +{ + kauth_acl_t nacl = NULL, oacl = NULL; + int error; + + /* + * Handle ACL inheritance. + */ + if (!(vap->va_vaflags & VA_NOINHERIT) && vfs_extendedsecurity(dvp->v_mount)) { + /* save the original filesec */ + if (VATTR_IS_ACTIVE(vap, va_acl)) { + oacl = vap->va_acl; + } + + vap->va_acl = NULL; + if ((error = kauth_acl_inherit(dvp, + oacl, + &nacl, + vap->va_type == VDIR, + ctx)) != 0) { + KAUTH_DEBUG("%p CREATE - error %d processing inheritance", dvp, error); + return(error); + } + + /* + * If the generated ACL is NULL, then we can save ourselves some effort + * by clearing the active bit. + */ + if (nacl == NULL) { + VATTR_CLEAR_ACTIVE(vap, va_acl); + } else { + vap->va_base_acl = oacl; + VATTR_SET(vap, va_acl, nacl); + } + } + + error = vnode_authattr_new_internal(dvp, vap, (vap->va_vaflags & VA_NOAUTH), defaulted_fieldsp, ctx); + if (error) { + vn_attribute_cleanup(vap, *defaulted_fieldsp); + } + + return error; +} + +void +vn_attribute_cleanup(struct vnode_attr *vap, uint32_t defaulted_fields) +{ + /* + * If the caller supplied a filesec in vap, it has been replaced + * now by the post-inheritance copy. We need to put the original back + * and free the inherited product. + */ + kauth_acl_t nacl, oacl; + + if (VATTR_IS_ACTIVE(vap, va_acl)) { + nacl = vap->va_acl; + oacl = vap->va_base_acl; + + if (oacl) { + VATTR_SET(vap, va_acl, oacl); + vap->va_base_acl = NULL; + } else { + VATTR_CLEAR_ACTIVE(vap, va_acl); + } + + if (nacl != NULL) { + kauth_acl_free(nacl); + } + } + + if ((defaulted_fields & VATTR_PREPARE_DEFAULTED_MODE) != 0) { + VATTR_CLEAR_ACTIVE(vap, va_mode); + } + if ((defaulted_fields & VATTR_PREPARE_DEFAULTED_GID) != 0) { + VATTR_CLEAR_ACTIVE(vap, va_gid); + } + if ((defaulted_fields & VATTR_PREPARE_DEFAULTED_UID) != 0) { + VATTR_CLEAR_ACTIVE(vap, va_uid); + } + + return; +} + +int +vn_authorize_unlink(vnode_t dvp, vnode_t vp, struct componentname *cnp, vfs_context_t ctx, __unused void *reserved) +{ + int error = 0; + + /* + * Normally, unlinking of directories is not supported. + * However, some file systems may have limited support. + */ + if ((vp->v_type == VDIR) && + !(vp->v_mount->mnt_vtable->vfc_vfsflags & VFC_VFSDIRLINKS)) { + return (EPERM); /* POSIX */ + } + + /* authorize the delete operation */ +#if CONFIG_MACF + if (!error) + error = mac_vnode_check_unlink(ctx, dvp, vp, cnp); +#endif /* MAC */ + if (!error) + error = vnode_authorize(vp, dvp, KAUTH_VNODE_DELETE, ctx); + + return error; +} + +int +vn_authorize_open_existing(vnode_t vp, struct componentname *cnp, int fmode, vfs_context_t ctx, void *reserved) +{ + /* Open of existing case */ + kauth_action_t action; + int error = 0; + + if (cnp->cn_ndp == NULL) { + panic("NULL ndp"); + } + if (reserved != NULL) { + panic("reserved not NULL."); + } + +#if CONFIG_MACF + /* XXX may do duplicate work here, but ignore that for now (idempotent) */ + if (vfs_flags(vnode_mount(vp)) & MNT_MULTILABEL) { + error = vnode_label(vnode_mount(vp), NULL, vp, NULL, 0, ctx); + if (error) + return (error); + } +#endif + + if ( (fmode & O_DIRECTORY) && vp->v_type != VDIR ) { + return (ENOTDIR); + } + + if (vp->v_type == VSOCK && vp->v_tag != VT_FDESC) { + return (EOPNOTSUPP); /* Operation not supported on socket */ + } + + if (vp->v_type == VLNK && (fmode & O_NOFOLLOW) != 0) { + return (ELOOP); /* O_NOFOLLOW was specified and the target is a symbolic link */ + } + + /* disallow write operations on directories */ + if (vnode_isdir(vp) && (fmode & (FWRITE | O_TRUNC))) { + return (EISDIR); + } + + if ((cnp->cn_ndp->ni_flag & NAMEI_TRAILINGSLASH)) { + if (vp->v_type != VDIR) { + return (ENOTDIR); + } + } + +#if CONFIG_MACF + /* If a file being opened is a shadow file containing + * namedstream data, ignore the macf checks because it + * is a kernel internal file and access should always + * be allowed. + */ + if (!(vnode_isshadow(vp) && vnode_isnamedstream(vp))) { + error = mac_vnode_check_open(ctx, vp, fmode); + if (error) { + return (error); + } + } +#endif + + /* compute action to be authorized */ + action = 0; + if (fmode & FREAD) { + action |= KAUTH_VNODE_READ_DATA; + } + if (fmode & (FWRITE | O_TRUNC)) { + /* + * If we are writing, appending, and not truncating, + * indicate that we are appending so that if the + * UF_APPEND or SF_APPEND bits are set, we do not deny + * the open. + */ + if ((fmode & O_APPEND) && !(fmode & O_TRUNC)) { + action |= KAUTH_VNODE_APPEND_DATA; + } else { + action |= KAUTH_VNODE_WRITE_DATA; + } + } + return (vnode_authorize(vp, NULL, action, ctx)); +} + +int +vn_authorize_create(vnode_t dvp, struct componentname *cnp, struct vnode_attr *vap, vfs_context_t ctx, void *reserved) +{ + /* Creation case */ + int error; + + if (cnp->cn_ndp == NULL) { + panic("NULL cn_ndp"); + } + if (reserved != NULL) { + panic("reserved not NULL."); + } + + /* Only validate path for creation if we didn't do a complete lookup */ + if (cnp->cn_ndp->ni_flag & NAMEI_UNFINISHED) { + error = lookup_validate_creation_path(cnp->cn_ndp); + if (error) + return (error); + } + +#if CONFIG_MACF + error = mac_vnode_check_create(ctx, dvp, cnp, vap); + if (error) + return (error); +#endif /* CONFIG_MACF */ + + return (vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_FILE, ctx)); +} + +int +vn_authorize_rename(struct vnode *fdvp, struct vnode *fvp, struct componentname *fcnp, + struct vnode *tdvp, struct vnode *tvp, struct componentname *tcnp, + vfs_context_t ctx, void *reserved) +{ + int error = 0; + int moving = 0; + + if (reserved != NULL) { + panic("Passed something other than NULL as reserved field!"); + } + + /* + * Avoid renaming "." and "..". + * + * XXX No need to check for this in the FS. We should always have the leaves + * in VFS in this case. + */ + if (fvp->v_type == VDIR && + ((fdvp == fvp) || + (fcnp->cn_namelen == 1 && fcnp->cn_nameptr[0] == '.') || + ((fcnp->cn_flags | tcnp->cn_flags) & ISDOTDOT)) ) { + error = EINVAL; + goto out; + } + + if (tvp == NULLVP && vnode_compound_rename_available(tdvp)) { + error = lookup_validate_creation_path(tcnp->cn_ndp); + if (error) + goto out; + } + + /***** *****/ +#if CONFIG_MACF + error = mac_vnode_check_rename_from(ctx, fdvp, fvp, fcnp); + if (error) + goto out; +#endif + +#if CONFIG_MACF + error = mac_vnode_check_rename_to(ctx, + tdvp, tvp, fdvp == tdvp, tcnp); + if (error) + goto out; +#endif + /***** *****/ + + /***** *****/ + if (tvp != NULL) { + if (fvp->v_type == VDIR && tvp->v_type != VDIR) { + error = ENOTDIR; + goto out; + } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) { + error = EISDIR; + goto out; + } + } + + if (fvp == tdvp) { + error = EINVAL; + goto out; + } + + /* + * The following edge case is caught here: + * (to cannot be a descendent of from) + * + * o fdvp + * / + * / + * o fvp + * \ + * \ + * o tdvp + * / + * / + * o tvp + */ + if (tdvp->v_parent == fvp) { + error = EINVAL; + goto out; + } + /***** *****/ + + /***** *****/ + + error = 0; + if ((tvp != NULL) && vnode_isdir(tvp)) { + if (tvp != fdvp) + moving = 1; + } else if (tdvp != fdvp) { + moving = 1; + } + + + /* + * must have delete rights to remove the old name even in + * the simple case of fdvp == tdvp. + * + * If fvp is a directory, and we are changing it's parent, + * then we also need rights to rewrite its ".." entry as well. + */ + if (vnode_isdir(fvp)) { + if ((error = vnode_authorize(fvp, fdvp, KAUTH_VNODE_DELETE | KAUTH_VNODE_ADD_SUBDIRECTORY, ctx)) != 0) + goto out; + } else { + if ((error = vnode_authorize(fvp, fdvp, KAUTH_VNODE_DELETE, ctx)) != 0) + goto out; + } + if (moving) { + /* moving into tdvp or tvp, must have rights to add */ + if ((error = vnode_authorize(((tvp != NULL) && vnode_isdir(tvp)) ? tvp : tdvp, + NULL, + vnode_isdir(fvp) ? KAUTH_VNODE_ADD_SUBDIRECTORY : KAUTH_VNODE_ADD_FILE, + ctx)) != 0) { + goto out; + } + } else { + /* node staying in same directory, must be allowed to add new name */ + if ((error = vnode_authorize(fdvp, NULL, + vnode_isdir(fvp) ? KAUTH_VNODE_ADD_SUBDIRECTORY : KAUTH_VNODE_ADD_FILE, ctx)) != 0) + goto out; + } + /* overwriting tvp */ + if ((tvp != NULL) && !vnode_isdir(tvp) && + ((error = vnode_authorize(tvp, tdvp, KAUTH_VNODE_DELETE, ctx)) != 0)) { + goto out; + } + + /***** *****/ + + /* XXX more checks? */ +out: + return error; +} + +int +vn_authorize_mkdir(vnode_t dvp, struct componentname *cnp, struct vnode_attr *vap, vfs_context_t ctx, void *reserved) +{ + int error; + + if (reserved != NULL) { + panic("reserved not NULL in vn_authorize_mkdir()"); + } + + /* XXX A hack for now, to make shadow files work */ + if (cnp->cn_ndp == NULL) { + return 0; + } + + if (vnode_compound_mkdir_available(dvp)) { + error = lookup_validate_creation_path(cnp->cn_ndp); + if (error) + goto out; + } + +#if CONFIG_MACF + error = mac_vnode_check_create(ctx, + dvp, cnp, vap); + if (error) + goto out; +#endif + + /* authorize addition of a directory to the parent */ + if ((error = vnode_authorize(dvp, NULL, KAUTH_VNODE_ADD_SUBDIRECTORY, ctx)) != 0) + goto out; + +out: + return error; +} + +int +vn_authorize_rmdir(vnode_t dvp, vnode_t vp, struct componentname *cnp, vfs_context_t ctx, void *reserved) +{ + int error; + + if (reserved != NULL) { + panic("Non-NULL reserved argument to vn_authorize_rmdir()"); + } + + if (vp->v_type != VDIR) { + /* + * rmdir only deals with directories + */ + return ENOTDIR; + } + + if (dvp == vp) { + /* + * No rmdir "." please. + */ + return EINVAL; + } + +#if CONFIG_MACF + error = mac_vnode_check_unlink(ctx, dvp, + vp, cnp); + if (error) + return error; +#endif + + return vnode_authorize(vp, dvp, KAUTH_VNODE_DELETE, ctx); +} + /* * Authorize an operation on a vnode. * @@ -4666,8 +5523,24 @@ vauth_node_owner(struct vnode_attr *vap, kauth_cred_t cred) return(result); } +/* + * vauth_node_group + * + * Description: Ask if a cred is a member of the group owning the vnode object + * + * Parameters: vap vnode attribute + * vap->va_gid group owner of vnode object + * cred credential to check + * ismember pointer to where to put the answer + * idontknow Return this if we can't get an answer + * + * Returns: 0 Success + * idontknow Can't get information + * kauth_cred_ismember_gid:? Error from kauth subsystem + * kauth_cred_ismember_gid:? Error from kauth subsystem + */ static int -vauth_node_group(struct vnode_attr *vap, kauth_cred_t cred, int *ismember) +vauth_node_group(struct vnode_attr *vap, kauth_cred_t cred, int *ismember, int idontknow) { int error; int result; @@ -4675,11 +5548,43 @@ vauth_node_group(struct vnode_attr *vap, kauth_cred_t cred, int *ismember) error = 0; result = 0; - /* the caller is expected to have asked the filesystem for a group at some point */ + /* + * The caller is expected to have asked the filesystem for a group + * at some point prior to calling this function. The answer may + * have been that there is no group ownership supported for the + * vnode object, in which case we return + */ if (vap && VATTR_IS_SUPPORTED(vap, va_gid)) { error = kauth_cred_ismember_gid(cred, vap->va_gid, &result); + /* + * Credentials which are opted into external group membership + * resolution which are not known to the external resolver + * will result in an ENOENT error. We translate this into + * the appropriate 'idontknow' response for our caller. + * + * XXX We do not make a distinction here between an ENOENT + * XXX arising from a response from the external resolver, + * XXX and an ENOENT which is internally generated. This is + * XXX a deficiency of the published kauth_cred_ismember_gid() + * XXX KPI which can not be overcome without new KPI. For + * XXX all currently known cases, however, this wil result + * XXX in correct behaviour. + */ + if (error == ENOENT) + error = idontknow; } - /* we could test the group UUID here if we had a policy for it */ + /* + * XXX We could test the group UUID here if we had a policy for it, + * XXX but this is problematic from the perspective of synchronizing + * XXX group UUID and POSIX GID ownership of a file and keeping the + * XXX values coherent over time. The problem is that the local + * XXX system will vend transient group UUIDs for unknown POSIX GID + * XXX values, and these are not persistent, whereas storage of values + * XXX is persistent. One potential solution to this is a local + * XXX (persistent) replica of remote directory entries and vended + * XXX local ids in a local directory server (think in terms of a + * XXX caching DNS server). + */ if (!error) *ismember = result; @@ -4707,16 +5612,39 @@ vauth_file_owner(vauth_ctx vcp) return(result); } + +/* + * vauth_file_ingroup + * + * Description: Ask if a user is a member of the group owning the directory + * + * Parameters: vcp The vnode authorization context that + * contains the user and directory info + * vcp->flags_valid Valid flags + * vcp->flags Flags values + * vcp->vap File vnode attributes + * vcp->ctx VFS Context (for user) + * ismember pointer to where to put the answer + * idontknow Return this if we can't get an answer + * + * Returns: 0 Success + * vauth_node_group:? Error from vauth_node_group() + * + * Implicit returns: *ismember 0 The user is not a group member + * 1 The user is a group member + */ static int -vauth_file_ingroup(vauth_ctx vcp, int *ismember) +vauth_file_ingroup(vauth_ctx vcp, int *ismember, int idontknow) { int error; + /* Check for a cached answer first, to avoid the check if possible */ if (vcp->flags_valid & _VAC_IN_GROUP) { *ismember = (vcp->flags & _VAC_IN_GROUP) ? 1 : 0; error = 0; } else { - error = vauth_node_group(vcp->vap, vcp->ctx->vc_ucred, ismember); + /* Otherwise, go look for it */ + error = vauth_node_group(vcp->vap, vcp->ctx->vc_ucred, ismember, idontknow); if (!error) { /* cache our result */ @@ -4753,16 +5681,38 @@ vauth_dir_owner(vauth_ctx vcp) return(result); } +/* + * vauth_dir_ingroup + * + * Description: Ask if a user is a member of the group owning the directory + * + * Parameters: vcp The vnode authorization context that + * contains the user and directory info + * vcp->flags_valid Valid flags + * vcp->flags Flags values + * vcp->dvap Dir vnode attributes + * vcp->ctx VFS Context (for user) + * ismember pointer to where to put the answer + * idontknow Return this if we can't get an answer + * + * Returns: 0 Success + * vauth_node_group:? Error from vauth_node_group() + * + * Implicit returns: *ismember 0 The user is not a group member + * 1 The user is a group member + */ static int -vauth_dir_ingroup(vauth_ctx vcp, int *ismember) +vauth_dir_ingroup(vauth_ctx vcp, int *ismember, int idontknow) { int error; + /* Check for a cached answer first, to avoid the check if possible */ if (vcp->flags_valid & _VAC_IN_DIR_GROUP) { *ismember = (vcp->flags & _VAC_IN_DIR_GROUP) ? 1 : 0; error = 0; } else { - error = vauth_node_group(vcp->dvap, vcp->ctx->vc_ucred, ismember); + /* Otherwise, go look for it */ + error = vauth_node_group(vcp->dvap, vcp->ctx->vc_ucred, ismember, idontknow); if (!error) { /* cache our result */ @@ -4870,14 +5820,25 @@ vnode_authorize_posix(vauth_ctx vcp, int action, int on_dir) } /* Check group membership (most expensive) */ - ismember = 0; + ismember = 0; /* Default to allow, if the target has no group owner */ + + /* + * In the case we can't get an answer about the user from the call to + * vauth_dir_ingroup() or vauth_file_ingroup(), we want to fail on + * the side of caution, rather than simply granting access, or we will + * fail to correctly implement exclusion groups, so we set the third + * parameter on the basis of the state of 'group_ok'. + */ if (on_dir) { - error = vauth_dir_ingroup(vcp, &ismember); + error = vauth_dir_ingroup(vcp, &ismember, (!group_ok ? EACCES : 0)); } else { - error = vauth_file_ingroup(vcp, &ismember); + error = vauth_file_ingroup(vcp, &ismember, (!group_ok ? EACCES : 0)); + } + if (error) { + if (!group_ok) + ismember = 1; + error = 0; } - if (error) - goto out; if (ismember) { _SETWHERE("group"); if (!group_ok) @@ -4921,15 +5882,21 @@ out: * - Neither the node nor the directory are immutable. * - The user is not the superuser. * - * Deletion is not permitted if the directory is sticky and the caller is not owner of the - * node or directory. + * Deletion is not permitted if the directory is sticky and the caller is + * not owner of the node or directory. + * + * If either the node grants DELETE, or the directory grants DELETE_CHILD, + * the node may be deleted. If neither denies the permission, and the + * caller has Posix write access to the directory, then the node may be + * deleted. * - * If either the node grants DELETE, or the directory grants DELETE_CHILD, the node may be - * deleted. If neither denies the permission, and the caller has Posix write access to the - * directory, then the node may be deleted. + * As an optimization, we cache whether or not delete child is permitted + * on directories without the sticky bit set. */ -static int -vnode_authorize_delete(vauth_ctx vcp) +int +vnode_authorize_delete(vauth_ctx vcp, boolean_t cached_delete_child); +/*static*/ int +vnode_authorize_delete(vauth_ctx vcp, boolean_t cached_delete_child) { struct vnode_attr *vap = vcp->vap; struct vnode_attr *dvap = vcp->dvap; @@ -4939,33 +5906,53 @@ vnode_authorize_delete(vauth_ctx vcp) /* check the ACL on the directory */ delete_child_denied = 0; - if (VATTR_IS_NOT(dvap, va_acl, NULL)) { + if (!cached_delete_child && VATTR_IS_NOT(dvap, va_acl, NULL)) { eval.ae_requested = KAUTH_VNODE_DELETE_CHILD; eval.ae_acl = &dvap->va_acl->acl_ace[0]; eval.ae_count = dvap->va_acl->acl_entrycount; eval.ae_options = 0; if (vauth_dir_owner(vcp)) eval.ae_options |= KAUTH_AEVAL_IS_OWNER; - if ((error = vauth_dir_ingroup(vcp, &ismember)) != 0) + /* + * We use ENOENT as a marker to indicate we could not get + * information in order to delay evaluation until after we + * have the ACL evaluation answer. Previously, we would + * always deny the operation at this point. + */ + if ((error = vauth_dir_ingroup(vcp, &ismember, ENOENT)) != 0 && error != ENOENT) return(error); - if (ismember) + if (error == ENOENT) + eval.ae_options |= KAUTH_AEVAL_IN_GROUP_UNKNOWN; + else if (ismember) eval.ae_options |= KAUTH_AEVAL_IN_GROUP; eval.ae_exp_gall = KAUTH_VNODE_GENERIC_ALL_BITS; eval.ae_exp_gread = KAUTH_VNODE_GENERIC_READ_BITS; eval.ae_exp_gwrite = KAUTH_VNODE_GENERIC_WRITE_BITS; eval.ae_exp_gexec = KAUTH_VNODE_GENERIC_EXECUTE_BITS; + /* + * If there is no entry, we are going to defer to other + * authorization mechanisms. + */ error = kauth_acl_evaluate(cred, &eval); if (error != 0) { KAUTH_DEBUG("%p ERROR during ACL processing - %d", vcp->vp, error); return(error); } - if (eval.ae_result == KAUTH_RESULT_DENY) + switch(eval.ae_result) { + case KAUTH_RESULT_DENY: delete_child_denied = 1; - if (eval.ae_result == KAUTH_RESULT_ALLOW) { - KAUTH_DEBUG("%p ALLOWED - granted by directory ACL", vcp->vp); - return(0); + break; + /* FALLSTHROUGH */ + case KAUTH_RESULT_ALLOW: + KAUTH_DEBUG("%p ALLOWED - granted by directory ACL", vcp->vp); + return(0); + case KAUTH_RESULT_DEFER: + default: + /* Effectively the same as !delete_child_denied */ + KAUTH_DEBUG("%p DEFERRED - directory ACL", vcp->vp); + break; } } @@ -4978,9 +5965,17 @@ vnode_authorize_delete(vauth_ctx vcp) eval.ae_options = 0; if (vauth_file_owner(vcp)) eval.ae_options |= KAUTH_AEVAL_IS_OWNER; - if ((error = vauth_file_ingroup(vcp, &ismember)) != 0) + /* + * We use ENOENT as a marker to indicate we could not get + * information in order to delay evaluation until after we + * have the ACL evaluation answer. Previously, we would + * always deny the operation at this point. + */ + if ((error = vauth_file_ingroup(vcp, &ismember, ENOENT)) != 0 && error != ENOENT) return(error); - if (ismember) + if (error == ENOENT) + eval.ae_options |= KAUTH_AEVAL_IN_GROUP_UNKNOWN; + else if (ismember) eval.ae_options |= KAUTH_AEVAL_IN_GROUP; eval.ae_exp_gall = KAUTH_VNODE_GENERIC_ALL_BITS; eval.ae_exp_gread = KAUTH_VNODE_GENERIC_READ_BITS; @@ -4991,30 +5986,38 @@ vnode_authorize_delete(vauth_ctx vcp) KAUTH_DEBUG("%p ERROR during ACL processing - %d", vcp->vp, error); return(error); } - if (eval.ae_result == KAUTH_RESULT_DENY) + + switch(eval.ae_result) { + case KAUTH_RESULT_DENY: delete_denied = 1; - if (eval.ae_result == KAUTH_RESULT_ALLOW) { + break; + case KAUTH_RESULT_ALLOW: KAUTH_DEBUG("%p ALLOWED - granted by file ACL", vcp->vp); return(0); + case KAUTH_RESULT_DEFER: + default: + /* Effectively the same as !delete_child_denied */ + KAUTH_DEBUG("%p DEFERRED%s - by file ACL", vcp->vp, delete_denied ? "(DENY)" : ""); + break; } } /* if denied by ACL on directory or node, return denial */ if (delete_denied || delete_child_denied) { - KAUTH_DEBUG("%p ALLOWED - denied by ACL", vcp->vp); + KAUTH_DEBUG("%p DENIED - denied by ACL", vcp->vp); return(EACCES); } /* enforce sticky bit behaviour */ if ((dvap->va_mode & S_ISTXT) && !vauth_file_owner(vcp) && !vauth_dir_owner(vcp)) { KAUTH_DEBUG("%p DENIED - sticky bit rules (user %d file %d dir %d)", - vcp->vp, cred->cr_uid, vap->va_uid, dvap->va_uid); + vcp->vp, cred->cr_posix.cr_uid, vap->va_uid, dvap->va_uid); return(EACCES); } /* check the directory */ - if ((error = vnode_authorize_posix(vcp, VWRITE, 1 /* on_dir */)) != 0) { - KAUTH_DEBUG("%p ALLOWED - granted by posix permisssions", vcp->vp); + if (!cached_delete_child && (error = vnode_authorize_posix(vcp, VWRITE, 1 /* on_dir */)) != 0) { + KAUTH_DEBUG("%p DENIED - denied by posix permisssions", vcp->vp); return(error); } @@ -5068,9 +6071,17 @@ vnode_authorize_simple(vauth_ctx vcp, kauth_ace_rights_t acl_rights, kauth_ace_r eval.ae_options = 0; if (vauth_file_owner(vcp)) eval.ae_options |= KAUTH_AEVAL_IS_OWNER; - if ((error = vauth_file_ingroup(vcp, &ismember)) != 0) + /* + * We use ENOENT as a marker to indicate we could not get + * information in order to delay evaluation until after we + * have the ACL evaluation answer. Previously, we would + * always deny the operation at this point. + */ + if ((error = vauth_file_ingroup(vcp, &ismember, ENOENT)) != 0 && error != ENOENT) return(error); - if (ismember) + if (error == ENOENT) + eval.ae_options |= KAUTH_AEVAL_IN_GROUP_UNKNOWN; + else if (ismember) eval.ae_options |= KAUTH_AEVAL_IN_GROUP; eval.ae_exp_gall = KAUTH_VNODE_GENERIC_ALL_BITS; eval.ae_exp_gread = KAUTH_VNODE_GENERIC_READ_BITS; @@ -5082,14 +6093,20 @@ vnode_authorize_simple(vauth_ctx vcp, kauth_ace_rights_t acl_rights, kauth_ace_r return(error); } - if (eval.ae_result == KAUTH_RESULT_DENY) { + switch(eval.ae_result) { + case KAUTH_RESULT_DENY: KAUTH_DEBUG("%p DENIED - by ACL", vcp->vp); - return(EACCES); /* deny, deny, counter-allege */ - } - if (eval.ae_result == KAUTH_RESULT_ALLOW) { + return(EACCES); /* deny, deny, counter-allege */ + case KAUTH_RESULT_ALLOW: KAUTH_DEBUG("%p ALLOWED - all rights granted by ACL", vcp->vp); return(0); + case KAUTH_RESULT_DEFER: + default: + /* Effectively the same as !delete_child_denied */ + KAUTH_DEBUG("%p DEFERRED - directory ACL", vcp->vp); + break; } + *found_deny = eval.ae_found_deny; /* fall through and evaluate residual rights */ @@ -5241,13 +6258,16 @@ vnode_authorize_checkimmutable(vnode_t vp, struct vnode_attr *vap, int rights, i } } - /* check for file immutability */ + /* + * check for file immutability. first, check if the requested rights are + * allowable for a UF_APPEND file. + */ append = 0; if (vp->v_type == VDIR) { - if ((rights & (KAUTH_VNODE_ADD_FILE | KAUTH_VNODE_ADD_SUBDIRECTORY)) == rights) + if ((rights & (KAUTH_VNODE_ADD_FILE | KAUTH_VNODE_ADD_SUBDIRECTORY | KAUTH_VNODE_WRITE_EXTATTRIBUTES)) == rights) append = 1; } else { - if ((rights & KAUTH_VNODE_APPEND_DATA) == rights) + if ((rights & (KAUTH_VNODE_APPEND_DATA | KAUTH_VNODE_WRITE_EXTATTRIBUTES)) == rights) append = 1; } if ((error = vnode_immutable(vap, append, ignore)) != 0) { @@ -5360,7 +6380,9 @@ vnode_authorize_callback(kauth_cred_t cred, void *idata, kauth_action_t action, vfs_context_t ctx; vnode_t cvp = NULLVP; vnode_t vp, dvp; - int result; + int result = KAUTH_RESULT_DENY; + int parent_iocount = 0; + int parent_action; /* In case we need to use namedstream's data fork for cached rights*/ ctx = (vfs_context_t)arg0; vp = (vnode_t)arg1; @@ -5380,18 +6402,66 @@ vnode_authorize_callback(kauth_cred_t cred, void *idata, kauth_action_t action, */ if (dvp && vp) goto defer; - if (dvp) + if (dvp) { cvp = dvp; - else - cvp = vp; + } else { + /* + * For named streams on local-authorization volumes, rights are cached on the parent; + * authorization is determined by looking at the parent's properties anyway, so storing + * on the parent means that we don't recompute for the named stream and that if + * we need to flush rights (e.g. on VNOP_SETATTR()) we don't need to track down the + * stream to flush its cache separately. If we miss in the cache, then we authorize + * as if there were no cached rights (passing the named stream vnode and desired rights to + * vnode_authorize_callback_int()). + * + * On an opaquely authorized volume, we don't know the relationship between the + * data fork's properties and the rights granted on a stream. Thus, named stream vnodes + * on such a volume are authorized directly (rather than using the parent) and have their + * own caches. When a named stream vnode is created, we mark the parent as having a named + * stream. On a VNOP_SETATTR() for the parent that may invalidate cached authorization, we + * find the stream and flush its cache. + */ + if (vnode_isnamedstream(vp) && (!vfs_authopaque(vp->v_mount))) { + cvp = vnode_getparent(vp); + if (cvp != NULLVP) { + parent_iocount = 1; + } else { + cvp = NULL; + goto defer; /* If we can't use the parent, take the slow path */ + } + + /* Have to translate some actions */ + parent_action = action; + if (parent_action & KAUTH_VNODE_READ_DATA) { + parent_action &= ~KAUTH_VNODE_READ_DATA; + parent_action |= KAUTH_VNODE_READ_EXTATTRIBUTES; + } + if (parent_action & KAUTH_VNODE_WRITE_DATA) { + parent_action &= ~KAUTH_VNODE_WRITE_DATA; + parent_action |= KAUTH_VNODE_WRITE_EXTATTRIBUTES; + } + + } else { + cvp = vp; + } + } - if (vnode_cache_is_authorized(cvp, ctx, action) == TRUE) - return KAUTH_RESULT_ALLOW; + if (vnode_cache_is_authorized(cvp, ctx, parent_iocount ? parent_action : action) == TRUE) { + result = KAUTH_RESULT_ALLOW; + goto out; + } defer: result = vnode_authorize_callback_int(cred, idata, action, arg0, arg1, arg2, arg3); - if (result == KAUTH_RESULT_ALLOW && cvp != NULLVP) + if (result == KAUTH_RESULT_ALLOW && cvp != NULLVP) { + KAUTH_DEBUG("%p - caching action = %x", cvp, action); vnode_cache_authorized_action(cvp, ctx, action); + } + +out: + if (parent_iocount) { + vnode_put(cvp); + } return result; } @@ -5411,7 +6481,7 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i int result; int *errorp; int noimmutable; - boolean_t parent_authorized_for_delete = FALSE; + boolean_t parent_authorized_for_delete_child = FALSE; boolean_t found_deny = FALSE; boolean_t parent_ref= FALSE; @@ -5476,8 +6546,8 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i * can skip a whole bunch of work... we will still have to * authorize that this specific child can be removed */ - if (vnode_cache_is_authorized(dvp, ctx, KAUTH_VNODE_DELETE) == TRUE) - parent_authorized_for_delete = TRUE; + if (vnode_cache_is_authorized(dvp, ctx, KAUTH_VNODE_DELETE_CHILD) == TRUE) + parent_authorized_for_delete_child = TRUE; } else { dvp = NULL; } @@ -5524,7 +6594,7 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i KAUTH_DEBUG("%p ERROR - failed to get vnode attributes - %d", vp, result); goto out; } - if (dvp && parent_authorized_for_delete == FALSE) { + if (dvp) { VATTR_WANTED(&dva, va_mode); VATTR_WANTED(&dva, va_uid); VATTR_WANTED(&dva, va_gid); @@ -5540,7 +6610,7 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i * If the vnode is an extended attribute data vnode (eg. a resource fork), *_DATA becomes * *_EXTATTRIBUTES. */ - if (S_ISXATTR(va.va_mode) || vnode_isnamedstream(vp)) { + if (vnode_isnamedstream(vp)) { if (rights & KAUTH_VNODE_READ_DATA) { rights &= ~KAUTH_VNODE_READ_DATA; rights |= KAUTH_VNODE_READ_EXTATTRIBUTES; @@ -5556,7 +6626,7 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i */ if (vnode_isnamedstream(vp) && (vp->v_parent != NULL) && - (vget_internal(vp->v_parent, 0, VNODE_NODEAD) == 0)) { + (vget_internal(vp->v_parent, 0, VNODE_NODEAD | VNODE_DRAINO) == 0)) { parent_ref = TRUE; vcp->vp = vp = vp->v_parent; if (VATTR_IS_SUPPORTED(&va, va_acl) && (va.va_acl != NULL)) @@ -5580,7 +6650,7 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i if ((result = vnode_authorize_checkimmutable(vp, &va, rights, noimmutable)) != 0) goto out; if ((rights & KAUTH_VNODE_DELETE) && - parent_authorized_for_delete == FALSE && + parent_authorized_for_delete_child == FALSE && ((result = vnode_authorize_checkimmutable(dvp, &dva, KAUTH_VNODE_DELETE_CHILD, 0)) != 0)) goto out; @@ -5593,13 +6663,14 @@ vnode_authorize_callback_int(__unused kauth_cred_t unused_cred, __unused void *i goto out; /* - * If we're not the superuser, authorize based on file properties. + * If we're not the superuser, authorize based on file properties; + * note that even if parent_authorized_for_delete_child is TRUE, we + * need to check on the node itself. */ if (!vfs_context_issuser(ctx)) { /* process delete rights */ if ((rights & KAUTH_VNODE_DELETE) && - parent_authorized_for_delete == FALSE && - ((result = vnode_authorize_delete(vcp)) != 0)) + ((result = vnode_authorize_delete(vcp, parent_authorized_for_delete_child)) != 0)) goto out; /* process remaining rights */ @@ -5650,12 +6721,21 @@ out: vnode_cache_authorized_action(vp, ctx, KAUTH_VNODE_SEARCHBYANYONE); } } - if ((rights & KAUTH_VNODE_DELETE) && parent_authorized_for_delete == FALSE) { + if ((rights & KAUTH_VNODE_DELETE) && parent_authorized_for_delete_child == FALSE) { /* - * parent was successfully and newly authorized for deletions - * add it to the cache + * parent was successfully and newly authorized for content deletions + * add it to the cache, but only if it doesn't have the sticky + * bit set on it. This same check is done earlier guarding + * fetching of dva, and if we jumped to out without having done + * this, we will have returned already because of a non-zero + * 'result' value. */ - vnode_cache_authorized_action(dvp, ctx, KAUTH_VNODE_DELETE); + if (VATTR_IS_SUPPORTED(&dva, va_mode) && + !(dva.va_mode & (S_ISVTX))) { + /* OK to cache delete rights */ + KAUTH_DEBUG("%p - caching DELETE_CHILD rights", dvp); + vnode_cache_authorized_action(dvp, ctx, KAUTH_VNODE_DELETE_CHILD); + } } if (parent_ref) vnode_put(vp); @@ -5667,12 +6747,18 @@ out: return(KAUTH_RESULT_ALLOW); } +int +vnode_authattr_new(vnode_t dvp, struct vnode_attr *vap, int noauth, vfs_context_t ctx) +{ + return vnode_authattr_new_internal(dvp, vap, noauth, NULL, ctx); +} + /* * Check that the attribute information in vattr can be legally applied to * a new file by the context. */ -int -vnode_authattr_new(vnode_t dvp, struct vnode_attr *vap, int noauth, vfs_context_t ctx) +static int +vnode_authattr_new_internal(vnode_t dvp, struct vnode_attr *vap, int noauth, uint32_t *defaulted_fieldsp, vfs_context_t ctx) { int error; int has_priv_suser, ismember, defaulted_owner, defaulted_group, defaulted_mode; @@ -5681,6 +6767,11 @@ vnode_authattr_new(vnode_t dvp, struct vnode_attr *vap, int noauth, vfs_context_ mount_t dmp; error = 0; + + if (defaulted_fieldsp) { + *defaulted_fieldsp = 0; + } + defaulted_owner = defaulted_group = defaulted_mode = 0; /* @@ -5863,6 +6954,17 @@ vnode_authattr_new(vnode_t dvp, struct vnode_attr *vap, int noauth, vfs_context_ } } out: + if (defaulted_fieldsp) { + if (defaulted_mode) { + *defaulted_fieldsp |= VATTR_PREPARE_DEFAULTED_MODE; + } + if (defaulted_group) { + *defaulted_fieldsp |= VATTR_PREPARE_DEFAULTED_GID; + } + if (defaulted_owner) { + *defaulted_fieldsp |= VATTR_PREPARE_DEFAULTED_UID; + } + } return(error); } @@ -5960,6 +7062,14 @@ vnode_authattr(vnode_t vp, struct vnode_attr *vap, kauth_action_t *actionp, vfs_ VATTR_WANTED(&ova, va_flags); } + /* + * If ACLs are being changed, we need the old ACLs. + */ + if (VATTR_IS_ACTIVE(vap, va_acl)) { + KAUTH_DEBUG("ATTR - acl changing, fetching old flags"); + VATTR_WANTED(&ova, va_acl); + } + /* * If the size is being set, make sure it's not a directory. */ @@ -6215,12 +7325,30 @@ vnode_authattr(vnode_t vp, struct vnode_attr *vap, kauth_action_t *actionp, vfs_ */ if (VATTR_IS_ACTIVE(vap, va_uuuid)) { /* if the owner UUID is not actually changing ... */ - if (VATTR_IS_SUPPORTED(&ova, va_uuuid) && kauth_guid_equal(&vap->va_uuuid, &ova.va_uuuid)) - goto no_uuuid_change; + if (VATTR_IS_SUPPORTED(&ova, va_uuuid)) { + if (kauth_guid_equal(&vap->va_uuuid, &ova.va_uuuid)) + goto no_uuuid_change; + + /* + * If the current owner UUID is a null GUID, check + * it against the UUID corresponding to the owner UID. + */ + if (kauth_guid_equal(&ova.va_uuuid, &kauth_null_guid) && + VATTR_IS_SUPPORTED(&ova, va_uid)) { + guid_t uid_guid; + + if (kauth_cred_uid2guid(ova.va_uid, &uid_guid) == 0 && + kauth_guid_equal(&vap->va_uuuid, &uid_guid)) + goto no_uuuid_change; + } + } /* * The owner UUID cannot be set by a non-superuser to anything other than - * their own. + * their own or a null GUID (to "unset" the owner UUID). + * Note that file systems must be prepared to handle the + * null UUID case in a manner appropriate for that file + * system. */ if (!has_priv_suser) { if ((error = kauth_cred_getguid(cred, &changer)) != 0) { @@ -6228,8 +7356,9 @@ vnode_authattr(vnode_t vp, struct vnode_attr *vap, kauth_action_t *actionp, vfs_ /* XXX ENOENT here - no UUID - should perhaps become EPERM */ goto out; } - if (!kauth_guid_equal(&vap->va_uuuid, &changer)) { - KAUTH_DEBUG(" ERROR - cannot set supplied owner UUID - not us"); + if (!kauth_guid_equal(&vap->va_uuuid, &changer) && + !kauth_guid_equal(&vap->va_uuuid, &kauth_null_guid)) { + KAUTH_DEBUG(" ERROR - cannot set supplied owner UUID - not us / null"); error = EPERM; goto out; } @@ -6243,20 +7372,41 @@ no_uuuid_change: */ if (VATTR_IS_ACTIVE(vap, va_guuid)) { /* if the group UUID is not actually changing ... */ - if (VATTR_IS_SUPPORTED(&ova, va_guuid) && kauth_guid_equal(&vap->va_guuid, &ova.va_guuid)) - goto no_guuid_change; + if (VATTR_IS_SUPPORTED(&ova, va_guuid)) { + if (kauth_guid_equal(&vap->va_guuid, &ova.va_guuid)) + goto no_guuid_change; + + /* + * If the current group UUID is a null UUID, check + * it against the UUID corresponding to the group GID. + */ + if (kauth_guid_equal(&ova.va_guuid, &kauth_null_guid) && + VATTR_IS_SUPPORTED(&ova, va_gid)) { + guid_t gid_guid; + + if (kauth_cred_gid2guid(ova.va_gid, &gid_guid) == 0 && + kauth_guid_equal(&vap->va_guuid, &gid_guid)) + goto no_guuid_change; + } + } /* * The group UUID cannot be set by a non-superuser to anything other than - * one of which they are a member. + * one of which they are a member or a null GUID (to "unset" + * the group UUID). + * Note that file systems must be prepared to handle the + * null UUID case in a manner appropriate for that file + * system. */ if (!has_priv_suser) { - if ((error = kauth_cred_ismember_guid(cred, &vap->va_guuid, &ismember)) != 0) { + if (kauth_guid_equal(&vap->va_guuid, &kauth_null_guid)) + ismember = 1; + else if ((error = kauth_cred_ismember_guid(cred, &vap->va_guuid, &ismember)) != 0) { KAUTH_DEBUG(" ERROR - got %d trying to check group membership", error); goto out; } if (!ismember) { - KAUTH_DEBUG(" ERROR - cannot create item with supplied group UUID - not a member"); + KAUTH_DEBUG(" ERROR - cannot set supplied group UUID - not a member / null"); error = EPERM; goto out; } @@ -6325,7 +7475,7 @@ no_guuid_change: KAUTH_DEBUG("CHMOD - adding/removing ACL entries"); } else if (vap->va_acl->acl_entrycount > 0) { /* both ACLs have the same ACE count, said count is 1 or more, bitwise compare ACLs */ - if (!memcmp(&vap->va_acl->acl_ace[0], &ova.va_acl->acl_ace[0], + if (memcmp(&vap->va_acl->acl_ace[0], &ova.va_acl->acl_ace[0], sizeof(struct kauth_ace) * vap->va_acl->acl_entrycount)) { required_action |= KAUTH_VNODE_WRITE_SECURITY; KAUTH_DEBUG("CHMOD - changing ACL entries"); @@ -6348,28 +7498,43 @@ out: return(error); } +static int +setlocklocal_callback(struct vnode *vp, __unused void *cargs) +{ + vnode_lock_spin(vp); + vp->v_flag |= VLOCKLOCAL; + vnode_unlock(vp); + + return (VNODE_RETURNED); +} void vfs_setlocklocal(mount_t mp) { - vnode_t vp; - - mount_lock(mp); + mount_lock_spin(mp); mp->mnt_kern_flag |= MNTK_LOCK_LOCAL; + mount_unlock(mp); /* - * We do not expect anyone to be using any vnodes at the - * time this routine is called. So no need for vnode locking + * The number of active vnodes is expected to be + * very small when vfs_setlocklocal is invoked. */ - TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_mntvnodes) { - vp->v_flag |= VLOCKLOCAL; - } - TAILQ_FOREACH(vp, &mp->mnt_workerqueue, v_mntvnodes) { - vp->v_flag |= VLOCKLOCAL; - } - TAILQ_FOREACH(vp, &mp->mnt_newvnodes, v_mntvnodes) { - vp->v_flag |= VLOCKLOCAL; - } + vnode_iterate(mp, 0, setlocklocal_callback, NULL); +} + +void +vfs_setunmountpreflight(mount_t mp) +{ + mount_lock_spin(mp); + mp->mnt_kern_flag |= MNTK_UNMOUNT_PREFLIGHT; + mount_unlock(mp); +} + +void +vfs_setcompoundopen(mount_t mp) +{ + mount_lock_spin(mp); + mp->mnt_compound_ops |= COMPOUND_VNOP_OPEN; mount_unlock(mp); } @@ -6385,7 +7550,7 @@ vn_setunionwait(vnode_t vp) void vn_checkunionwait(vnode_t vp) { - vnode_lock(vp); + vnode_lock_spin(vp); while ((vp->v_flag & VISUNION) == VISUNION) msleep((caddr_t)&vp->v_flag, &vp->v_lock, 0, 0, 0); vnode_unlock(vp); @@ -6395,7 +7560,7 @@ void vn_clearunionwait(vnode_t vp, int locked) { if (!locked) - vnode_lock(vp); + vnode_lock_spin(vp); if((vp->v_flag & VISUNION) == VISUNION) { vp->v_flag &= ~VISUNION; wakeup((caddr_t)&vp->v_flag); @@ -6518,16 +7683,33 @@ errno_t rmdir_remove_orphaned_appleDouble(vnode_t vp , vfs_context_t ctx, int * cpos += dp->d_reclen; dp = (struct dirent*)cpos; } + + /* + * workaround for HFS/NFS setting eofflag before end of file + */ + if (vp->v_tag == VT_HFS && nentries > 2) + eofflag=0; + + if (vp->v_tag == VT_NFS) { + if (eofflag && !full_erase_flag) { + full_erase_flag = 1; + eofflag = 0; + uio_reset(auio, 0, UIO_SYSSPACE, UIO_READ); + } + else if (!eofflag && full_erase_flag) + full_erase_flag = 0; + } } while (!eofflag); /* - * If we've made it here all the files in the dir are AppleDouble + * If we've made it here all the files in the dir are ._ files. * We can delete the files even though the node is suspended * because we are the owner of the file. */ uio_reset(auio, 0, UIO_SYSSPACE, UIO_READ); eofflag = 0; + full_erase_flag = 0; do { siz = UIO_BUFF_SIZE; @@ -6560,11 +7742,17 @@ errno_t rmdir_remove_orphaned_appleDouble(vnode_t vp , vfs_context_t ctx, int * !((dp->d_namlen == 1 && dp->d_name[0] == '.') || (dp->d_namlen == 2 && dp->d_name[0] == '.' && dp->d_name[1] == '.')) ) { - NDINIT(&nd_temp, DELETE, USEDVP, UIO_SYSSPACE, CAST_USER_ADDR_T(dp->d_name), ctx); + + NDINIT(&nd_temp, DELETE, OP_UNLINK, USEDVP, + UIO_SYSSPACE, CAST_USER_ADDR_T(dp->d_name), + ctx); nd_temp.ni_dvp = vp; error = unlink1(ctx, &nd_temp, 0); - if(error && error != ENOENT) + + if (error && error != ENOENT) { goto outsc; + } + } cpos += dp->d_reclen; dp = (struct dirent*)cpos; @@ -6606,25 +7794,659 @@ outsc: } -#ifdef JOE_DEBUG +void +lock_vnode_and_post(vnode_t vp, int kevent_num) +{ + /* Only take the lock if there's something there! */ + if (vp->v_knotes.slh_first != NULL) { + vnode_lock(vp); + KNOTE(&vp->v_knotes, kevent_num); + vnode_unlock(vp); + } +} -record_vp(vnode_t vp, int count) { +#ifdef JOE_DEBUG +static void record_vp(vnode_t vp, int count) { struct uthread *ut; - int i; +#if CONFIG_TRIGGERS + if (vp->v_resolve) + return; +#endif if ((vp->v_flag & VSYSTEM)) return; ut = get_bsdthread_info(current_thread()); ut->uu_iocount += count; - if (ut->uu_vpindex < 32) { - for (i = 0; i < ut->uu_vpindex; i++) { - if (ut->uu_vps[i] == vp) - return; + if (count == 1) { + if (ut->uu_vpindex < 32) { + OSBacktrace((void **)&ut->uu_pcs[ut->uu_vpindex][0], 10); + + ut->uu_vps[ut->uu_vpindex] = vp; + ut->uu_vpindex++; } - ut->uu_vps[ut->uu_vpindex] = vp; - ut->uu_vpindex++; } } #endif + + +#if CONFIG_TRIGGERS + +#define TRIG_DEBUG 0 + +#if TRIG_DEBUG +#define TRIG_LOG(...) do { printf("%s: ", __FUNCTION__); printf(__VA_ARGS__); } while (0) +#else +#define TRIG_LOG(...) +#endif + +/* + * Resolver result functions + */ + +resolver_result_t +vfs_resolver_result(uint32_t seq, enum resolver_status stat, int aux) +{ + /* + * |<--- 32 --->|<--- 28 --->|<- 4 ->| + * sequence auxiliary status + */ + return (((uint64_t)seq) << 32) | + (((uint64_t)(aux & 0x0fffffff)) << 4) | + (uint64_t)(stat & 0x0000000F); +} + +enum resolver_status +vfs_resolver_status(resolver_result_t result) +{ + /* lower 4 bits is status */ + return (result & 0x0000000F); +} + +uint32_t +vfs_resolver_sequence(resolver_result_t result) +{ + /* upper 32 bits is sequence */ + return (uint32_t)(result >> 32); +} + +int +vfs_resolver_auxiliary(resolver_result_t result) +{ + /* 28 bits of auxiliary */ + return (int)(((uint32_t)(result & 0xFFFFFFF0)) >> 4); +} + +/* + * SPI + * Call in for resolvers to update vnode trigger state + */ +int +vnode_trigger_update(vnode_t vp, resolver_result_t result) +{ + vnode_resolve_t rp; + uint32_t seq; + enum resolver_status stat; + + if (vp->v_resolve == NULL) { + return (EINVAL); + } + + stat = vfs_resolver_status(result); + seq = vfs_resolver_sequence(result); + + if ((stat != RESOLVER_RESOLVED) && (stat != RESOLVER_UNRESOLVED)) { + return (EINVAL); + } + + rp = vp->v_resolve; + lck_mtx_lock(&rp->vr_lock); + + if (seq > rp->vr_lastseq) { + if (stat == RESOLVER_RESOLVED) + rp->vr_flags |= VNT_RESOLVED; + else + rp->vr_flags &= ~VNT_RESOLVED; + + rp->vr_lastseq = seq; + } + + lck_mtx_unlock(&rp->vr_lock); + + return (0); +} + +static int +vnode_resolver_attach(vnode_t vp, vnode_resolve_t rp, boolean_t ref) +{ + int error; + + vnode_lock_spin(vp); + if (vp->v_resolve != NULL) { + vnode_unlock(vp); + return EINVAL; + } else { + vp->v_resolve = rp; + } + vnode_unlock(vp); + + if (ref) { + error = vnode_ref_ext(vp, O_EVTONLY, VNODE_REF_FORCE); + if (error != 0) { + panic("VNODE_REF_FORCE didn't help..."); + } + } + + return 0; +} + +/* + * VFS internal interfaces for vnode triggers + * + * vnode must already have an io count on entry + * v_resolve is stable when io count is non-zero + */ +static int +vnode_resolver_create(mount_t mp, vnode_t vp, struct vnode_trigger_param *tinfo, boolean_t external) +{ + vnode_resolve_t rp; + int result; + char byte; + +#if 1 + /* minimum pointer test (debugging) */ + if (tinfo->vnt_data) + byte = *((char *)tinfo->vnt_data); +#endif + MALLOC(rp, vnode_resolve_t, sizeof(*rp), M_TEMP, M_WAITOK); + if (rp == NULL) + return (ENOMEM); + + lck_mtx_init(&rp->vr_lock, trigger_vnode_lck_grp, trigger_vnode_lck_attr); + + rp->vr_resolve_func = tinfo->vnt_resolve_func; + rp->vr_unresolve_func = tinfo->vnt_unresolve_func; + rp->vr_rearm_func = tinfo->vnt_rearm_func; + rp->vr_reclaim_func = tinfo->vnt_reclaim_func; + rp->vr_data = tinfo->vnt_data; + rp->vr_lastseq = 0; + rp->vr_flags = tinfo->vnt_flags & VNT_VALID_MASK; + if (external) { + rp->vr_flags |= VNT_EXTERNAL; + } + + result = vnode_resolver_attach(vp, rp, external); + if (result != 0) { + goto out; + } + + if (mp) { + OSAddAtomic(1, &mp->mnt_numtriggers); + } + + return (result); + +out: + FREE(rp, M_TEMP); + return result; +} + +static void +vnode_resolver_release(vnode_resolve_t rp) +{ + /* + * Give them a chance to free any private data + */ + if (rp->vr_data && rp->vr_reclaim_func) { + rp->vr_reclaim_func(NULLVP, rp->vr_data); + } + + lck_mtx_destroy(&rp->vr_lock, trigger_vnode_lck_grp); + FREE(rp, M_TEMP); + +} + +/* Called after the vnode has been drained */ +static void +vnode_resolver_detach(vnode_t vp) +{ + vnode_resolve_t rp; + mount_t mp; + + mp = vnode_mount(vp); + + vnode_lock(vp); + rp = vp->v_resolve; + vp->v_resolve = NULL; + vnode_unlock(vp); + + if ((rp->vr_flags & VNT_EXTERNAL) != 0) { + vnode_rele_ext(vp, O_EVTONLY, 1); + } + + vnode_resolver_release(rp); + + /* Keep count of active trigger vnodes per mount */ + OSAddAtomic(-1, &mp->mnt_numtriggers); +} + +/* + * Pathname operations that don't trigger a mount for trigger vnodes + */ +static const u_int64_t ignorable_pathops_mask = + 1LL << OP_MOUNT | + 1LL << OP_UNMOUNT | + 1LL << OP_STATFS | + 1LL << OP_ACCESS | + 1LL << OP_GETATTR | + 1LL << OP_LISTXATTR; + +int +vfs_istraditionaltrigger(enum path_operation op, const struct componentname *cnp) +{ + if (cnp->cn_flags & ISLASTCN) + return ((1LL << op) & ignorable_pathops_mask) == 0; + else + return (1); +} + +__private_extern__ +void +vnode_trigger_rearm(vnode_t vp, vfs_context_t ctx) +{ + vnode_resolve_t rp; + resolver_result_t result; + enum resolver_status status; + uint32_t seq; + + if ((vp->v_resolve == NULL) || + (vp->v_resolve->vr_rearm_func == NULL) || + (vp->v_resolve->vr_flags & VNT_AUTO_REARM) == 0) { + return; + } + + rp = vp->v_resolve; + lck_mtx_lock(&rp->vr_lock); + + /* + * Check if VFS initiated this unmount. If so, we'll catch it after the unresolve completes. + */ + if (rp->vr_flags & VNT_VFS_UNMOUNTED) { + lck_mtx_unlock(&rp->vr_lock); + return; + } + + /* Check if this vnode is already armed */ + if ((rp->vr_flags & VNT_RESOLVED) == 0) { + lck_mtx_unlock(&rp->vr_lock); + return; + } + + lck_mtx_unlock(&rp->vr_lock); + + result = rp->vr_rearm_func(vp, 0, rp->vr_data, ctx); + status = vfs_resolver_status(result); + seq = vfs_resolver_sequence(result); + + lck_mtx_lock(&rp->vr_lock); + if (seq > rp->vr_lastseq) { + if (status == RESOLVER_UNRESOLVED) + rp->vr_flags &= ~VNT_RESOLVED; + rp->vr_lastseq = seq; + } + lck_mtx_unlock(&rp->vr_lock); +} + +__private_extern__ +int +vnode_trigger_resolve(vnode_t vp, struct nameidata *ndp, vfs_context_t ctx) +{ + vnode_resolve_t rp; + enum path_operation op; + resolver_result_t result; + enum resolver_status status; + uint32_t seq; + + /* Only trigger on topmost vnodes */ + if ((vp->v_resolve == NULL) || + (vp->v_resolve->vr_resolve_func == NULL) || + (vp->v_mountedhere != NULL)) { + return (0); + } + + rp = vp->v_resolve; + lck_mtx_lock(&rp->vr_lock); + + /* Check if this vnode is already resolved */ + if (rp->vr_flags & VNT_RESOLVED) { + lck_mtx_unlock(&rp->vr_lock); + return (0); + } + + lck_mtx_unlock(&rp->vr_lock); + + /* + * XXX + * assumes that resolver will not access this trigger vnode (otherwise the kernel will deadlock) + * is there anyway to know this??? + * there can also be other legitimate lookups in parallel + * + * XXX - should we call this on a separate thread with a timeout? + * + * XXX - should we use ISLASTCN to pick the op value??? Perhaps only leafs should + * get the richer set and non-leafs should get generic OP_LOOKUP? TBD + */ + op = (ndp->ni_op < OP_MAXOP) ? ndp->ni_op: OP_LOOKUP; + + result = rp->vr_resolve_func(vp, &ndp->ni_cnd, op, 0, rp->vr_data, ctx); + status = vfs_resolver_status(result); + seq = vfs_resolver_sequence(result); + + lck_mtx_lock(&rp->vr_lock); + if (seq > rp->vr_lastseq) { + if (status == RESOLVER_RESOLVED) + rp->vr_flags |= VNT_RESOLVED; + rp->vr_lastseq = seq; + } + lck_mtx_unlock(&rp->vr_lock); + + /* On resolver errors, propagate the error back up */ + return (status == RESOLVER_ERROR ? vfs_resolver_auxiliary(result) : 0); +} + +static int +vnode_trigger_unresolve(vnode_t vp, int flags, vfs_context_t ctx) +{ + vnode_resolve_t rp; + resolver_result_t result; + enum resolver_status status; + uint32_t seq; + + if ((vp->v_resolve == NULL) || (vp->v_resolve->vr_unresolve_func == NULL)) { + return (0); + } + + rp = vp->v_resolve; + lck_mtx_lock(&rp->vr_lock); + + /* Check if this vnode is already resolved */ + if ((rp->vr_flags & VNT_RESOLVED) == 0) { + printf("vnode_trigger_unresolve: not currently resolved\n"); + lck_mtx_unlock(&rp->vr_lock); + return (0); + } + + rp->vr_flags |= VNT_VFS_UNMOUNTED; + + lck_mtx_unlock(&rp->vr_lock); + + /* + * XXX + * assumes that resolver will not access this trigger vnode (otherwise the kernel will deadlock) + * there can also be other legitimate lookups in parallel + * + * XXX - should we call this on a separate thread with a timeout? + */ + + result = rp->vr_unresolve_func(vp, flags, rp->vr_data, ctx); + status = vfs_resolver_status(result); + seq = vfs_resolver_sequence(result); + + lck_mtx_lock(&rp->vr_lock); + if (seq > rp->vr_lastseq) { + if (status == RESOLVER_UNRESOLVED) + rp->vr_flags &= ~VNT_RESOLVED; + rp->vr_lastseq = seq; + } + rp->vr_flags &= ~VNT_VFS_UNMOUNTED; + lck_mtx_unlock(&rp->vr_lock); + + /* On resolver errors, propagate the error back up */ + return (status == RESOLVER_ERROR ? vfs_resolver_auxiliary(result) : 0); +} + +static int +triggerisdescendant(mount_t mp, mount_t rmp) +{ + int match = FALSE; + + /* + * walk up vnode covered chain looking for a match + */ + name_cache_lock_shared(); + + while (1) { + vnode_t vp; + + /* did we encounter "/" ? */ + if (mp->mnt_flag & MNT_ROOTFS) + break; + + vp = mp->mnt_vnodecovered; + if (vp == NULLVP) + break; + + mp = vp->v_mount; + if (mp == rmp) { + match = TRUE; + break; + } + } + + name_cache_unlock(); + + return (match); +} + +struct trigger_unmount_info { + vfs_context_t ctx; + mount_t top_mp; + vnode_t trigger_vp; + mount_t trigger_mp; + uint32_t trigger_vid; + int flags; +}; + +static int +trigger_unmount_callback(mount_t mp, void * arg) +{ + struct trigger_unmount_info * infop = (struct trigger_unmount_info *)arg; + boolean_t mountedtrigger = FALSE; + + /* + * When we encounter the top level mount we're done + */ + if (mp == infop->top_mp) + return (VFS_RETURNED_DONE); + + if ((mp->mnt_vnodecovered == NULL) || + (vnode_getwithref(mp->mnt_vnodecovered) != 0)) { + return (VFS_RETURNED); + } + + if ((mp->mnt_vnodecovered->v_mountedhere == mp) && + (mp->mnt_vnodecovered->v_resolve != NULL) && + (mp->mnt_vnodecovered->v_resolve->vr_flags & VNT_RESOLVED)) { + mountedtrigger = TRUE; + } + vnode_put(mp->mnt_vnodecovered); + + /* + * When we encounter a mounted trigger, check if its under the top level mount + */ + if ( !mountedtrigger || !triggerisdescendant(mp, infop->top_mp) ) + return (VFS_RETURNED); + + /* + * Process any pending nested mount (now that its not referenced) + */ + if ((infop->trigger_vp != NULLVP) && + (vnode_getwithvid(infop->trigger_vp, infop->trigger_vid) == 0)) { + vnode_t vp = infop->trigger_vp; + int error; + + infop->trigger_vp = NULLVP; + + if (mp == vp->v_mountedhere) { + vnode_put(vp); + printf("trigger_unmount_callback: unexpected match '%s'\n", + mp->mnt_vfsstat.f_mntonname); + return (VFS_RETURNED); + } + if (infop->trigger_mp != vp->v_mountedhere) { + vnode_put(vp); + printf("trigger_unmount_callback: trigger mnt changed! (%p != %p)\n", + infop->trigger_mp, vp->v_mountedhere); + goto savenext; + } + + error = vnode_trigger_unresolve(vp, infop->flags, infop->ctx); + vnode_put(vp); + if (error) { + printf("unresolving: '%s', err %d\n", + vp->v_mountedhere ? vp->v_mountedhere->mnt_vfsstat.f_mntonname : + "???", error); + return (VFS_RETURNED_DONE); /* stop iteration on errors */ + } + } +savenext: + /* + * We can't call resolver here since we hold a mount iter + * ref on mp so save its covered vp for later processing + */ + infop->trigger_vp = mp->mnt_vnodecovered; + if ((infop->trigger_vp != NULLVP) && + (vnode_getwithref(infop->trigger_vp) == 0)) { + if (infop->trigger_vp->v_mountedhere == mp) { + infop->trigger_vid = infop->trigger_vp->v_id; + infop->trigger_mp = mp; + } + vnode_put(infop->trigger_vp); + } + + return (VFS_RETURNED); +} + +/* + * Attempt to unmount any trigger mounts nested underneath a mount. + * This is a best effort attempt and no retries are performed here. + * + * Note: mp->mnt_rwlock is held exclusively on entry (so be carefull) + */ +__private_extern__ +void +vfs_nested_trigger_unmounts(mount_t mp, int flags, vfs_context_t ctx) +{ + struct trigger_unmount_info info; + + /* Must have trigger vnodes */ + if (mp->mnt_numtriggers == 0) { + return; + } + /* Avoid recursive requests (by checking covered vnode) */ + if ((mp->mnt_vnodecovered != NULL) && + (vnode_getwithref(mp->mnt_vnodecovered) == 0)) { + boolean_t recursive = FALSE; + + if ((mp->mnt_vnodecovered->v_mountedhere == mp) && + (mp->mnt_vnodecovered->v_resolve != NULL) && + (mp->mnt_vnodecovered->v_resolve->vr_flags & VNT_VFS_UNMOUNTED)) { + recursive = TRUE; + } + vnode_put(mp->mnt_vnodecovered); + if (recursive) + return; + } + + /* + * Attempt to unmount any nested trigger mounts (best effort) + */ + info.ctx = ctx; + info.top_mp = mp; + info.trigger_vp = NULLVP; + info.trigger_vid = 0; + info.trigger_mp = NULL; + info.flags = flags; + + (void) vfs_iterate(VFS_ITERATE_TAIL_FIRST, trigger_unmount_callback, &info); + + /* + * Process remaining nested mount (now that its not referenced) + */ + if ((info.trigger_vp != NULLVP) && + (vnode_getwithvid(info.trigger_vp, info.trigger_vid) == 0)) { + vnode_t vp = info.trigger_vp; + + if (info.trigger_mp == vp->v_mountedhere) { + (void) vnode_trigger_unresolve(vp, flags, ctx); + } + vnode_put(vp); + } +} + +int +vfs_addtrigger(mount_t mp, const char *relpath, struct vnode_trigger_info *vtip, vfs_context_t ctx) +{ + struct nameidata nd; + int res; + vnode_t rvp, vp; + struct vnode_trigger_param vtp; + + /* + * Must be called for trigger callback, wherein rwlock is held + */ + lck_rw_assert(&mp->mnt_rwlock, LCK_RW_ASSERT_HELD); + + TRIG_LOG("Adding trigger at %s\n", relpath); + TRIG_LOG("Trying VFS_ROOT\n"); + + /* + * We do a lookup starting at the root of the mountpoint, unwilling + * to cross into other mountpoints. + */ + res = VFS_ROOT(mp, &rvp, ctx); + if (res != 0) { + goto out; + } + + TRIG_LOG("Trying namei\n"); + + NDINIT(&nd, LOOKUP, OP_LOOKUP, USEDVP | NOCROSSMOUNT | FOLLOW, UIO_SYSSPACE, + CAST_USER_ADDR_T(relpath), ctx); + nd.ni_dvp = rvp; + res = namei(&nd); + if (res != 0) { + vnode_put(rvp); + goto out; + } + + vp = nd.ni_vp; + nameidone(&nd); + vnode_put(rvp); + + TRIG_LOG("Trying vnode_resolver_create()\n"); + + /* + * Set up blob. vnode_create() takes a larger structure + * with creation info, and we needed something different + * for this case. One needs to win, or we need to munge both; + * vnode_create() wins. + */ + bzero(&vtp, sizeof(vtp)); + vtp.vnt_resolve_func = vtip->vti_resolve_func; + vtp.vnt_unresolve_func = vtip->vti_unresolve_func; + vtp.vnt_rearm_func = vtip->vti_rearm_func; + vtp.vnt_reclaim_func = vtip->vti_reclaim_func; + vtp.vnt_reclaim_func = vtip->vti_reclaim_func; + vtp.vnt_data = vtip->vti_data; + vtp.vnt_flags = vtip->vti_flags; + + res = vnode_resolver_create(mp, vp, &vtp, TRUE); + vnode_put(vp); +out: + TRIG_LOG("Returning %d\n", res); + return res; +} + +#endif /* CONFIG_TRIGGERS */