/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ /* Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T */ /* All Rights Reserved */ /* * University Copyright- Copyright (c) 1982, 1986, 1988 * The Regents of the University of California * All Rights Reserved * * University Acknowledgment- Portions of this document are derived from * software developed by the University of California, Berkeley, and its * contributors. */ #pragma ident "%Z%%M% %I% %E% SMI" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Private interfaces to create vopstats-related data structures */ extern void initialize_vopstats(vopstats_t *); extern vopstats_t *get_fstype_vopstats(struct vfs *, struct vfssw *); extern vsk_anchor_t *get_vskstat_anchor(struct vfs *); static void vfs_clearmntopt_nolock(mntopts_t *, const char *, int); static void vfs_setmntopt_nolock(mntopts_t *, const char *, const char *, int, int); static int vfs_optionisset_nolock(const mntopts_t *, const char *, char **); static void vfs_freemnttab(struct vfs *); static void vfs_freeopt(mntopt_t *); static void vfs_swapopttbl_nolock(mntopts_t *, mntopts_t *); static void vfs_swapopttbl(mntopts_t *, mntopts_t *); static void vfs_copyopttbl_extend(const mntopts_t *, mntopts_t *, int); static void vfs_createopttbl_extend(mntopts_t *, const char *, const mntopts_t *); static char **vfs_copycancelopt_extend(char **const, int); static void vfs_freecancelopt(char **); static char *getrootfs(void); static int getmacpath(dev_info_t *, void *); struct ipmnt { struct ipmnt *mip_next; dev_t mip_dev; struct vfs *mip_vfsp; }; static kmutex_t vfs_miplist_mutex; static struct ipmnt *vfs_miplist = NULL; static struct ipmnt *vfs_miplist_end = NULL; /* * VFS global data. */ vnode_t *rootdir; /* pointer to root inode vnode. */ vnode_t *devicesdir; /* pointer to inode of devices root */ vnode_t *devdir; /* pointer to inode of dev root */ char *server_rootpath; /* root path for diskless clients */ char *server_hostname; /* hostname of diskless server */ static struct vfs root; static struct vfs devices; static struct vfs dev; struct vfs *rootvfs = &root; /* pointer to root vfs; head of VFS list. */ rvfs_t *rvfs_list; /* array of vfs ptrs for vfs hash list */ int vfshsz = 512; /* # of heads/locks in vfs hash arrays */ /* must be power of 2! */ timespec_t vfs_mnttab_ctime; /* mnttab created time */ timespec_t vfs_mnttab_mtime; /* mnttab last modified time */ char *vfs_dummyfstype = "\0"; struct pollhead vfs_pollhd; /* for mnttab pollers */ /* * Table for generic options recognized in the VFS layer and acted * on at this level before parsing file system specific options. * The nosuid option is stronger than any of the devices and setuid * options, so those are canceled when nosuid is seen. * * All options which are added here need to be added to the * list of standard options in usr/src/cmd/fs.d/fslib.c as well. */ /* * VFS Mount options table */ static char *ro_cancel[] = { MNTOPT_RW, NULL }; static char *rw_cancel[] = { MNTOPT_RO, NULL }; static char *suid_cancel[] = { MNTOPT_NOSUID, NULL }; static char *nosuid_cancel[] = { MNTOPT_SUID, MNTOPT_DEVICES, MNTOPT_NODEVICES, MNTOPT_NOSETUID, MNTOPT_SETUID, NULL }; static char *devices_cancel[] = { MNTOPT_NODEVICES, NULL }; static char *nodevices_cancel[] = { MNTOPT_DEVICES, NULL }; static char *setuid_cancel[] = { MNTOPT_NOSETUID, NULL }; static char *nosetuid_cancel[] = { MNTOPT_SETUID, NULL }; static char *nbmand_cancel[] = { MNTOPT_NONBMAND, NULL }; static char *nonbmand_cancel[] = { MNTOPT_NBMAND, NULL }; static char *exec_cancel[] = { MNTOPT_NOEXEC, NULL }; static char *noexec_cancel[] = { MNTOPT_EXEC, NULL }; static const mntopt_t mntopts[] = { /* * option name cancel options default arg flags */ { MNTOPT_REMOUNT, NULL, NULL, MO_NODISPLAY, (void *)0 }, { MNTOPT_RO, ro_cancel, NULL, 0, (void *)0 }, { MNTOPT_RW, rw_cancel, NULL, 0, (void *)0 }, { MNTOPT_SUID, suid_cancel, NULL, 0, (void *)0 }, { MNTOPT_NOSUID, nosuid_cancel, NULL, 0, (void *)0 }, { MNTOPT_DEVICES, devices_cancel, NULL, 0, (void *)0 }, { MNTOPT_NODEVICES, nodevices_cancel, NULL, 0, (void *)0 }, { MNTOPT_SETUID, setuid_cancel, NULL, 0, (void *)0 }, { MNTOPT_NOSETUID, nosetuid_cancel, NULL, 0, (void *)0 }, { MNTOPT_NBMAND, nbmand_cancel, NULL, 0, (void *)0 }, { MNTOPT_NONBMAND, nonbmand_cancel, NULL, 0, (void *)0 }, { MNTOPT_EXEC, exec_cancel, NULL, 0, (void *)0 }, { MNTOPT_NOEXEC, noexec_cancel, NULL, 0, (void *)0 }, }; const mntopts_t vfs_mntopts = { sizeof (mntopts) / sizeof (mntopt_t), (mntopt_t *)&mntopts[0] }; /* * File system operation dispatch functions. */ int fsop_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr) { return (*(vfsp)->vfs_op->vfs_mount)(vfsp, mvp, uap, cr); } int fsop_unmount(vfs_t *vfsp, int flag, cred_t *cr) { return (*(vfsp)->vfs_op->vfs_unmount)(vfsp, flag, cr); } int fsop_root(vfs_t *vfsp, vnode_t **vpp) { refstr_t *mntpt; int ret = (*(vfsp)->vfs_op->vfs_root)(vfsp, vpp); /* * Make sure this root has a path. With lofs, it is possible to have * a NULL mountpoint. */ if (ret == 0 && vfsp->vfs_mntpt != NULL && (*vpp)->v_path == NULL) { mntpt = vfs_getmntpoint(vfsp); vn_setpath_str(*vpp, refstr_value(mntpt), strlen(refstr_value(mntpt))); refstr_rele(mntpt); } return (ret); } int fsop_statfs(vfs_t *vfsp, statvfs64_t *sp) { return (*(vfsp)->vfs_op->vfs_statvfs)(vfsp, sp); } int fsop_sync(vfs_t *vfsp, short flag, cred_t *cr) { return (*(vfsp)->vfs_op->vfs_sync)(vfsp, flag, cr); } int fsop_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp) { return (*(vfsp)->vfs_op->vfs_vget)(vfsp, vpp, fidp); } int fsop_mountroot(vfs_t *vfsp, enum whymountroot reason) { return (*(vfsp)->vfs_op->vfs_mountroot)(vfsp, reason); } void fsop_freefs(vfs_t *vfsp) { (*(vfsp)->vfs_op->vfs_freevfs)(vfsp); } int fsop_vnstate(vfs_t *vfsp, vnode_t *vp, vntrans_t nstate) { return ((*(vfsp)->vfs_op->vfs_vnstate)(vfsp, vp, nstate)); } int fsop_sync_by_kind(int fstype, short flag, cred_t *cr) { ASSERT((fstype >= 0) && (fstype < nfstype)); if (ALLOCATED_VFSSW(&vfssw[fstype]) && VFS_INSTALLED(&vfssw[fstype])) return (*vfssw[fstype].vsw_vfsops.vfs_sync) (NULL, flag, cr); else return (ENOTSUP); } /* * File system initialization. vfs_setfsops() must be called from a file * system's init routine. */ static int fs_copyfsops(const fs_operation_def_t *template, vfsops_t *actual, int *unused_ops) { static const fs_operation_trans_def_t vfs_ops_table[] = { VFSNAME_MOUNT, offsetof(vfsops_t, vfs_mount), fs_nosys, fs_nosys, VFSNAME_UNMOUNT, offsetof(vfsops_t, vfs_unmount), fs_nosys, fs_nosys, VFSNAME_ROOT, offsetof(vfsops_t, vfs_root), fs_nosys, fs_nosys, VFSNAME_STATVFS, offsetof(vfsops_t, vfs_statvfs), fs_nosys, fs_nosys, VFSNAME_SYNC, offsetof(vfsops_t, vfs_sync), (fs_generic_func_p) fs_sync, (fs_generic_func_p) fs_sync, /* No errors allowed */ VFSNAME_VGET, offsetof(vfsops_t, vfs_vget), fs_nosys, fs_nosys, VFSNAME_MOUNTROOT, offsetof(vfsops_t, vfs_mountroot), fs_nosys, fs_nosys, VFSNAME_FREEVFS, offsetof(vfsops_t, vfs_freevfs), (fs_generic_func_p)fs_freevfs, (fs_generic_func_p)fs_freevfs, /* Shouldn't fail */ VFSNAME_VNSTATE, offsetof(vfsops_t, vfs_vnstate), (fs_generic_func_p)fs_nosys, (fs_generic_func_p)fs_nosys, NULL, 0, NULL, NULL }; return (fs_build_vector(actual, unused_ops, vfs_ops_table, template)); } int vfs_setfsops(int fstype, const fs_operation_def_t *template, vfsops_t **actual) { int error; int unused_ops; /* * Verify that fstype refers to a valid fs. Note that * 0 is valid since it's used to set "stray" ops. */ if ((fstype < 0) || (fstype >= nfstype)) return (EINVAL); if (!ALLOCATED_VFSSW(&vfssw[fstype])) return (EINVAL); /* Set up the operations vector. */ error = fs_copyfsops(template, &vfssw[fstype].vsw_vfsops, &unused_ops); if (error != 0) return (error); vfssw[fstype].vsw_flag |= VSW_INSTALLED; if (actual != NULL) *actual = &vfssw[fstype].vsw_vfsops; #if DEBUG if (unused_ops != 0) cmn_err(CE_WARN, "vfs_setfsops: %s: %d operations supplied " "but not used", vfssw[fstype].vsw_name, unused_ops); #endif return (0); } int vfs_makefsops(const fs_operation_def_t *template, vfsops_t **actual) { int error; int unused_ops; *actual = (vfsops_t *)kmem_alloc(sizeof (vfsops_t), KM_SLEEP); error = fs_copyfsops(template, *actual, &unused_ops); if (error != 0) { kmem_free(*actual, sizeof (vfsops_t)); *actual = NULL; return (error); } return (0); } /* * Free a vfsops structure created as a result of vfs_makefsops(). * NOTE: For a vfsops structure initialized by vfs_setfsops(), use * vfs_freevfsops_by_type(). */ void vfs_freevfsops(vfsops_t *vfsops) { kmem_free(vfsops, sizeof (vfsops_t)); } /* * Since the vfsops structure is part of the vfssw table and wasn't * really allocated, we're not really freeing anything. We keep * the name for consistency with vfs_freevfsops(). We do, however, * need to take care of a little bookkeeping. * NOTE: For a vfsops structure created by vfs_setfsops(), use * vfs_freevfsops_by_type(). */ int vfs_freevfsops_by_type(int fstype) { /* Verify that fstype refers to a loaded fs (and not fsid 0). */ if ((fstype <= 0) || (fstype >= nfstype)) return (EINVAL); WLOCK_VFSSW(); if ((vfssw[fstype].vsw_flag & VSW_INSTALLED) == 0) { WUNLOCK_VFSSW(); return (EINVAL); } vfssw[fstype].vsw_flag &= ~VSW_INSTALLED; WUNLOCK_VFSSW(); return (0); } /* Support routines used to reference vfs_op */ /* Set the operations vector for a vfs */ void vfs_setops(vfs_t *vfsp, vfsops_t *vfsops) { vfsops_t *op; ASSERT(vfsp != NULL); ASSERT(vfsops != NULL); op = vfsp->vfs_op; membar_consumer(); if ((vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) && casptr(&vfsp->vfs_op, op, vfsops) == op) { return; } fsem_setvfsops(vfsp, vfsops); } /* Retrieve the operations vector for a vfs */ vfsops_t * vfs_getops(vfs_t *vfsp) { vfsops_t *op; ASSERT(vfsp != NULL); op = vfsp->vfs_op; membar_consumer(); if ((vfsp->vfs_implp == NULL || vfsp->vfs_femhead == NULL) && op == vfsp->vfs_op) { return (op); } else { return (fsem_getvfsops(vfsp)); } } /* * Returns non-zero (1) if the vfsops matches that of the vfs. * Returns zero (0) if not. */ int vfs_matchops(vfs_t *vfsp, vfsops_t *vfsops) { return (vfs_getops(vfsp) == vfsops); } /* * Returns non-zero (1) if the file system has installed a non-default, * non-error vfs_sync routine. Returns zero (0) otherwise. */ int vfs_can_sync(vfs_t *vfsp) { /* vfs_sync() routine is not the default/error function */ return (vfs_getops(vfsp)->vfs_sync != fs_sync); } /* * Initialize a vfs structure. */ void vfs_init(vfs_t *vfsp, vfsops_t *op, void *data) { vfsp->vfs_count = 0; vfsp->vfs_next = vfsp; vfsp->vfs_prev = vfsp; vfsp->vfs_zone_next = vfsp; vfsp->vfs_zone_prev = vfsp; vfsp->vfs_flag = 0; vfsp->vfs_data = (data); vfsp->vfs_resource = NULL; vfsp->vfs_mntpt = NULL; vfsp->vfs_mntopts.mo_count = 0; vfsp->vfs_mntopts.mo_list = NULL; vfsp->vfs_implp = NULL; vfsp->vfs_zone = NULL; /* * Note: Don't initialize any member of the vfs_impl_t structure * here as it could be a problem for unbundled file systems. */ vfs_setops((vfsp), (op)); sema_init(&vfsp->vfs_reflock, 1, NULL, SEMA_DEFAULT, NULL); } /* * Allocate and initialize the vfs implementation private data * structure, vfs_impl_t. */ void vfsimpl_setup(vfs_t *vfsp) { vfsp->vfs_implp = kmem_alloc(sizeof (vfs_impl_t), KM_SLEEP); /* Note that this are #define'd in vfs.h */ vfsp->vfs_femhead = NULL; vfsp->vfs_vskap = NULL; vfsp->vfs_fstypevsp = NULL; } /* * Release the vfs_impl_t structure, if it exists. Some unbundled * filesystems may not use the newer version of vfs and thus * would not contain this implementation private data structure. */ void vfsimpl_teardown(vfs_t *vfsp) { vfs_impl_t *vip = vfsp->vfs_implp; if (vip == NULL) return; if (vip->vi_femhead) { ASSERT(vip->vi_femhead->femh_list == NULL); mutex_destroy(&vip->vi_femhead->femh_lock); kmem_free(vip->vi_femhead, sizeof (*(vip->vi_femhead))); vip->vi_femhead = NULL; } kmem_free(vfsp->vfs_implp, sizeof (vfs_impl_t)); vfsp->vfs_implp = NULL; } /* * VFS system calls: mount, umount, syssync, statfs, fstatfs, statvfs, * fstatvfs, and sysfs moved to common/syscall. */ /* * Update every mounted file system. We call the vfs_sync operation of * each file system type, passing it a NULL vfsp to indicate that all * mounted file systems of that type should be updated. */ void vfs_sync(int flag) { struct vfssw *vswp; RLOCK_VFSSW(); for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) { if (ALLOCATED_VFSSW(vswp) && VFS_INSTALLED(vswp)) { vfs_refvfssw(vswp); RUNLOCK_VFSSW(); (void) (*vswp->vsw_vfsops.vfs_sync)(NULL, flag, CRED()); vfs_unrefvfssw(vswp); RLOCK_VFSSW(); } } RUNLOCK_VFSSW(); } void sync(void) { vfs_sync(0); } /* * External routines. */ krwlock_t vfssw_lock; /* lock accesses to vfssw */ /* * Lock for accessing the vfs linked list. Initialized in vfs_mountroot(), * but otherwise should be accessed only via vfs_list_lock() and * vfs_list_unlock(). Also used to protect the timestamp for mods to the list. */ static krwlock_t vfslist; /* * Mount devfs on /devices. This is done right after root is mounted * to provide device access support for the system */ static void vfs_mountdevices(void) { struct vfssw *vsw; struct vnode *mvp; struct mounta mounta = { /* fake mounta for devfs_mount() */ NULL, NULL, MS_SYSSPACE, NULL, NULL, 0, NULL, 0 }; /* * _init devfs module to fill in the vfssw */ if (modload("fs", "devfs") == -1) panic("Cannot _init devfs module"); /* * Hold vfs */ RLOCK_VFSSW(); vsw = vfs_getvfsswbyname("devfs"); VFS_INIT(&devices, &vsw->vsw_vfsops, NULL); VFS_HOLD(&devices); /* * Locate mount point */ if (lookupname("/devices", UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp)) panic("Cannot find /devices"); /* * Perform the mount of /devices */ if (VFS_MOUNT(&devices, mvp, &mounta, CRED())) panic("Cannot mount /devices"); RUNLOCK_VFSSW(); /* * Set appropriate members and add to vfs list for mnttab display */ vfs_setresource(&devices, "/devices"); vfs_setmntpoint(&devices, "/devices"); /* * Hold the root of /devices so it won't go away */ if (VFS_ROOT(&devices, &devicesdir)) panic("vfs_mountdevices: not devices root"); if (vfs_lock(&devices) != 0) { VN_RELE(devicesdir); cmn_err(CE_NOTE, "Cannot acquire vfs_lock of /devices"); return; } if (vn_vfswlock(mvp) != 0) { vfs_unlock(&devices); VN_RELE(devicesdir); cmn_err(CE_NOTE, "Cannot acquire vfswlock of /devices"); return; } vfs_add(mvp, &devices, 0); vn_vfsunlock(mvp); vfs_unlock(&devices); VN_RELE(devicesdir); } /* * mount the first instance of /dev to root and remain mounted */ static void vfs_mountdev1(void) { struct vfssw *vsw; struct vnode *mvp; struct mounta mounta = { /* fake mounta for sdev_mount() */ NULL, NULL, MS_SYSSPACE | MS_OVERLAY, NULL, NULL, 0, NULL, 0 }; /* * _init dev module to fill in the vfssw */ if (modload("fs", "dev") == -1) cmn_err(CE_PANIC, "Cannot _init dev module\n"); /* * Hold vfs */ RLOCK_VFSSW(); vsw = vfs_getvfsswbyname("dev"); VFS_INIT(&dev, &vsw->vsw_vfsops, NULL); VFS_HOLD(&dev); /* * Locate mount point */ if (lookupname("/dev", UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp)) cmn_err(CE_PANIC, "Cannot find /dev\n"); /* * Perform the mount of /dev */ if (VFS_MOUNT(&dev, mvp, &mounta, CRED())) cmn_err(CE_PANIC, "Cannot mount /dev 1\n"); RUNLOCK_VFSSW(); /* * Set appropriate members and add to vfs list for mnttab display */ vfs_setresource(&dev, "/dev"); vfs_setmntpoint(&dev, "/dev"); /* * Hold the root of /dev so it won't go away */ if (VFS_ROOT(&dev, &devdir)) cmn_err(CE_PANIC, "vfs_mountdev1: not dev root"); if (vfs_lock(&dev) != 0) { VN_RELE(devdir); cmn_err(CE_NOTE, "Cannot acquire vfs_lock of /dev"); return; } if (vn_vfswlock(mvp) != 0) { vfs_unlock(&dev); VN_RELE(devdir); cmn_err(CE_NOTE, "Cannot acquire vfswlock of /dev"); return; } vfs_add(mvp, &dev, 0); vn_vfsunlock(mvp); vfs_unlock(&dev); VN_RELE(devdir); } /* * Mount required filesystem. This is done right after root is mounted. */ static void vfs_mountfs(char *module, char *spec, char *path) { struct vnode *mvp; struct mounta mounta; vfs_t *vfsp; mounta.flags = MS_SYSSPACE | MS_DATA; mounta.fstype = module; mounta.spec = spec; mounta.dir = path; if (lookupname(path, UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp)) { cmn_err(CE_WARN, "Cannot find %s", path); return; } if (domount(NULL, &mounta, mvp, CRED(), &vfsp)) cmn_err(CE_WARN, "Cannot mount %s", path); else VFS_RELE(vfsp); VN_RELE(mvp); } /* * vfs_mountroot is called by main() to mount the root filesystem. */ void vfs_mountroot(void) { struct vnode *rvp = NULL; char *path; size_t plen; struct vfssw *vswp; rw_init(&vfssw_lock, NULL, RW_DEFAULT, NULL); rw_init(&vfslist, NULL, RW_DEFAULT, NULL); /* * Alloc the vfs hash bucket array and locks */ rvfs_list = kmem_zalloc(vfshsz * sizeof (rvfs_t), KM_SLEEP); /* * Call machine-dependent routine "rootconf" to choose a root * file system type. */ if (rootconf()) panic("vfs_mountroot: cannot mount root"); /* * Get vnode for '/'. Set up rootdir, u.u_rdir and u.u_cdir * to point to it. These are used by lookuppn() so that it * knows where to start from ('/' or '.'). */ vfs_setmntpoint(rootvfs, "/"); if (VFS_ROOT(rootvfs, &rootdir)) panic("vfs_mountroot: no root vnode"); PTOU(curproc)->u_cdir = rootdir; VN_HOLD(PTOU(curproc)->u_cdir); PTOU(curproc)->u_rdir = NULL; /* * Setup the global zone's rootvp, now that it exists. */ global_zone->zone_rootvp = rootdir; VN_HOLD(global_zone->zone_rootvp); /* * Notify the module code that it can begin using the * root filesystem instead of the boot program's services. */ modrootloaded = 1; /* * Set up mnttab information for root */ vfs_setresource(rootvfs, rootfs.bo_name); /* * Notify cluster software that the root filesystem is available. */ clboot_mountroot(); /* Now that we're all done with the root FS, set up its vopstats */ if ((vswp = vfs_getvfsswbyvfsops(vfs_getops(rootvfs))) != NULL) { /* Set flag for statistics collection */ if (vswp->vsw_flag & VSW_STATS) { initialize_vopstats(&rootvfs->vfs_vopstats); rootvfs->vfs_flag |= VFS_STATS; rootvfs->vfs_fstypevsp = get_fstype_vopstats(rootvfs, vswp); rootvfs->vfs_vskap = get_vskstat_anchor(rootvfs); } vfs_unrefvfssw(vswp); } /* * Mount /devices, /dev instance 1, /system/contract, /etc/mnttab, * /etc/svc/volatile, /etc/dfs/sharetab, /system/object, and /proc. */ vfs_mountdevices(); vfs_mountdev1(); vfs_mountfs("ctfs", "ctfs", CTFS_ROOT); vfs_mountfs("proc", "/proc", "/proc"); vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab"); vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile"); vfs_mountfs("objfs", "objfs", OBJFS_ROOT); if (getzoneid() == GLOBAL_ZONEID) { vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab"); } #ifdef __sparc /* * This bit of magic can go away when we convert sparc to * the new boot architecture based on ramdisk. * * Booting off a mirrored root volume: * At this point, we have booted and mounted root on a * single component of the mirror. Complete the boot * by configuring SVM and converting the root to the * dev_t of the mirrored root device. This dev_t conversion * only works because the underlying device doesn't change. */ if (root_is_svm) { if (svm_rootconf()) { panic("vfs_mountroot: cannot remount root"); } /* * mnttab should reflect the new root device */ vfs_lock_wait(rootvfs); vfs_setresource(rootvfs, rootfs.bo_name); vfs_unlock(rootvfs); } #endif /* __sparc */ /* * Look up the root device via devfs so that a dv_node is * created for it. The vnode is never VN_RELE()ed. * We allocate more than MAXPATHLEN so that the * buffer passed to i_ddi_prompath_to_devfspath() is * exactly MAXPATHLEN (the function expects a buffer * of that length). */ plen = strlen("/devices"); path = kmem_alloc(plen + MAXPATHLEN, KM_SLEEP); (void) strcpy(path, "/devices"); if (i_ddi_prompath_to_devfspath(rootfs.bo_name, path + plen) != DDI_SUCCESS || lookupname(path, UIO_SYSSPACE, FOLLOW, NULLVPP, &rvp)) { /* NUL terminate in case "path" has garbage */ path[plen + MAXPATHLEN - 1] = '\0'; #ifdef DEBUG cmn_err(CE_WARN, "!Cannot lookup root device: %s", path); #endif } kmem_free(path, plen + MAXPATHLEN); } /* * If remount failed and we're in a zone we need to check for the zone * root path and strip it before the call to vfs_setpath(). * * If strpath doesn't begin with the zone_rootpath the original * strpath is returned unchanged. */ static const char * stripzonepath(const char *strpath) { char *str1, *str2; int i; zone_t *zonep = curproc->p_zone; if (zonep->zone_rootpath == NULL || strpath == NULL) { return (NULL); } /* * we check for the end of the string at one past the * current position because the zone_rootpath always * ends with "/" but we don't want to strip that off. */ str1 = zonep->zone_rootpath; str2 = (char *)strpath; ASSERT(str1[0] != '\0'); for (i = 0; str1[i + 1] != '\0'; i++) { if (str1[i] != str2[i]) return ((char *)strpath); } return (&str2[i]); } /* * Common mount code. Called from the system call entry point, from autofs, * and from pxfs. * * Takes the effective file system type, mount arguments, the mount point * vnode, flags specifying whether the mount is a remount and whether it * should be entered into the vfs list, and credentials. Fills in its vfspp * parameter with the mounted file system instance's vfs. * * Note that the effective file system type is specified as a string. It may * be null, in which case it's determined from the mount arguments, and may * differ from the type specified in the mount arguments; this is a hook to * allow interposition when instantiating file system instances. * * The caller is responsible for releasing its own hold on the mount point * vp (this routine does its own hold when necessary). * Also note that for remounts, the mount point vp should be the vnode for * the root of the file system rather than the vnode that the file system * is mounted on top of. */ int domount(char *fsname, struct mounta *uap, vnode_t *vp, struct cred *credp, struct vfs **vfspp) { struct vfssw *vswp; vfsops_t *vfsops; struct vfs *vfsp; struct vnode *bvp; dev_t bdev = 0; mntopts_t mnt_mntopts; int error = 0; int copyout_error = 0; int ovflags; char *opts = uap->optptr; char *inargs = opts; int optlen = uap->optlen; int remount; int rdonly; int nbmand = 0; int delmip = 0; int addmip = 0; int splice = ((uap->flags & MS_NOSPLICE) == 0); int fromspace = (uap->flags & MS_SYSSPACE) ? UIO_SYSSPACE : UIO_USERSPACE; char *resource = NULL, *mountpt = NULL; refstr_t *oldresource, *oldmntpt; struct pathname pn, rpn; vsk_anchor_t *vskap; /* * The v_flag value for the mount point vp is permanently set * to VVFSLOCK so that no one bypasses the vn_vfs*locks routine * for mount point locking. */ mutex_enter(&vp->v_lock); vp->v_flag |= VVFSLOCK; mutex_exit(&vp->v_lock); mnt_mntopts.mo_count = 0; /* * Find the ops vector to use to invoke the file system-specific mount * method. If the fsname argument is non-NULL, use it directly. * Otherwise, dig the file system type information out of the mount * arguments. * * A side effect is to hold the vfssw entry. * * Mount arguments can be specified in several ways, which are * distinguished by flag bit settings. The preferred way is to set * MS_OPTIONSTR, indicating an 8 argument mount with the file system * type supplied as a character string and the last two arguments * being a pointer to a character buffer and the size of the buffer. * On entry, the buffer holds a null terminated list of options; on * return, the string is the list of options the file system * recognized. If MS_DATA is set arguments five and six point to a * block of binary data which the file system interprets. * A further wrinkle is that some callers don't set MS_FSS and MS_DATA * consistently with these conventions. To handle them, we check to * see whether the pointer to the file system name has a numeric value * less than 256. If so, we treat it as an index. */ if (fsname != NULL) { if ((vswp = vfs_getvfssw(fsname)) == NULL) { return (EINVAL); } } else if (uap->flags & (MS_OPTIONSTR | MS_DATA | MS_FSS)) { size_t n; uint_t fstype; char name[FSTYPSZ]; if ((fstype = (uintptr_t)uap->fstype) < 256) { RLOCK_VFSSW(); if (fstype == 0 || fstype >= nfstype || !ALLOCATED_VFSSW(&vfssw[fstype])) { RUNLOCK_VFSSW(); return (EINVAL); } (void) strcpy(name, vfssw[fstype].vsw_name); RUNLOCK_VFSSW(); if ((vswp = vfs_getvfssw(name)) == NULL) return (EINVAL); } else { /* * Handle either kernel or user address space. */ if (uap->flags & MS_SYSSPACE) { error = copystr(uap->fstype, name, FSTYPSZ, &n); } else { error = copyinstr(uap->fstype, name, FSTYPSZ, &n); } if (error) { if (error == ENAMETOOLONG) return (EINVAL); return (error); } if ((vswp = vfs_getvfssw(name)) == NULL) return (EINVAL); } } else { if ((vswp = vfs_getvfsswbyvfsops(vfs_getops(rootvfs))) == NULL) return (EINVAL); } if (!VFS_INSTALLED(vswp)) return (EINVAL); vfsops = &vswp->vsw_vfsops; vfs_copyopttbl(&vswp->vsw_optproto, &mnt_mntopts); /* * Fetch mount options and parse them for generic vfs options */ if (uap->flags & MS_OPTIONSTR) { /* * Limit the buffer size */ if (optlen < 0 || optlen > MAX_MNTOPT_STR) { error = EINVAL; goto errout; } if ((uap->flags & MS_SYSSPACE) == 0) { inargs = kmem_alloc(MAX_MNTOPT_STR, KM_SLEEP); inargs[0] = '\0'; if (optlen) { error = copyinstr(opts, inargs, (size_t)optlen, NULL); if (error) { goto errout; } } } vfs_parsemntopts(&mnt_mntopts, inargs, 0); } /* * Flag bits override the options string. */ if (uap->flags & MS_REMOUNT) vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_REMOUNT, NULL, 0, 0); if (uap->flags & MS_RDONLY) vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_RO, NULL, 0, 0); if (uap->flags & MS_NOSUID) vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_NOSUID, NULL, 0, 0); /* * Check if this is a remount; must be set in the option string and * the file system must support a remount option. */ if (remount = vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_REMOUNT, NULL)) { if (!(vswp->vsw_flag & VSW_CANREMOUNT)) { error = ENOTSUP; goto errout; } uap->flags |= MS_REMOUNT; } /* * uap->flags and vfs_optionisset() should agree. */ if (rdonly = vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_RO, NULL)) { uap->flags |= MS_RDONLY; } if (vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_NOSUID, NULL)) { uap->flags |= MS_NOSUID; } nbmand = vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_NBMAND, NULL); ASSERT(splice || !remount); /* * If we are splicing the fs into the namespace, * perform mount point checks. * * We want to resolve the path for the mount point to eliminate * '.' and ".." and symlinks in mount points; we can't do the * same for the resource string, since it would turn * "/dev/dsk/c0t0d0s0" into "/devices/pci@...". We need to do * this before grabbing vn_vfswlock(), because otherwise we * would deadlock with lookuppn(). */ if (splice) { ASSERT(vp->v_count > 0); /* * Pick up mount point and device from appropriate space. */ if (pn_get(uap->spec, fromspace, &pn) == 0) { resource = kmem_alloc(pn.pn_pathlen + 1, KM_SLEEP); (void) strcpy(resource, pn.pn_path); pn_free(&pn); } /* * Do a lookupname prior to taking the * writelock. Mark this as completed if * successful for later cleanup and addition to * the mount in progress table. */ if ((uap->flags & MS_GLOBAL) == 0 && lookupname(uap->spec, fromspace, FOLLOW, NULL, &bvp) == 0) { addmip = 1; } if ((error = pn_get(uap->dir, fromspace, &pn)) == 0) { pathname_t *pnp; if (*pn.pn_path != '/') { error = EINVAL; pn_free(&pn); goto errout; } pn_alloc(&rpn); /* * Kludge to prevent autofs from deadlocking with * itself when it calls domount(). * * If autofs is calling, it is because it is doing * (autofs) mounts in the process of an NFS mount. A * lookuppn() here would cause us to block waiting for * said NFS mount to complete, which can't since this * is the thread that was supposed to doing it. */ if (fromspace == UIO_USERSPACE) { if ((error = lookuppn(&pn, &rpn, FOLLOW, NULL, NULL)) == 0) { pnp = &rpn; } else { /* * The file disappeared or otherwise * became inaccessible since we opened * it; might as well fail the mount * since the mount point is no longer * accessible. */ pn_free(&rpn); pn_free(&pn); goto errout; } } else { pnp = &pn; } mountpt = kmem_alloc(pnp->pn_pathlen + 1, KM_SLEEP); (void) strcpy(mountpt, pnp->pn_path); /* * If the addition of the zone's rootpath * would push us over a total path length * of MAXPATHLEN, we fail the mount with * ENAMETOOLONG, which is what we would have * gotten if we were trying to perform the same * mount in the global zone. * * strlen() doesn't count the trailing * '\0', but zone_rootpathlen counts both a * trailing '/' and the terminating '\0'. */ if ((curproc->p_zone->zone_rootpathlen - 1 + strlen(mountpt)) > MAXPATHLEN || (resource != NULL && (curproc->p_zone->zone_rootpathlen - 1 + strlen(resource)) > MAXPATHLEN)) { error = ENAMETOOLONG; } pn_free(&rpn); pn_free(&pn); } if (error) goto errout; /* * Prevent path name resolution from proceeding past * the mount point. */ if (vn_vfswlock(vp) != 0) { error = EBUSY; goto errout; } /* * Verify that it's legitimate to establish a mount on * the prospective mount point. */ if (vn_mountedvfs(vp) != NULL) { /* * The mount point lock was obtained after some * other thread raced through and established a mount. */ vn_vfsunlock(vp); error = EBUSY; goto errout; } if (vp->v_flag & VNOMOUNT) { vn_vfsunlock(vp); error = EINVAL; goto errout; } } if ((uap->flags & (MS_DATA | MS_OPTIONSTR)) == 0) { uap->dataptr = NULL; uap->datalen = 0; } /* * If this is a remount, we don't want to create a new VFS. * Instead, we pass the existing one with a remount flag. */ if (remount) { /* * Confirm that the mount point is the root vnode of the * file system that is being remounted. * This can happen if the user specifies a different * mount point directory pathname in the (re)mount command. * * Code below can only be reached if splice is true, so it's * safe to do vn_vfsunlock() here. */ if ((vp->v_flag & VROOT) == 0) { vn_vfsunlock(vp); error = ENOENT; goto errout; } /* * Disallow making file systems read-only unless file system * explicitly allows it in its vfssw. Ignore other flags. */ if (rdonly && vn_is_readonly(vp) == 0 && (vswp->vsw_flag & VSW_CANRWRO) == 0) { vn_vfsunlock(vp); error = EINVAL; goto errout; } /* * Changing the NBMAND setting on remounts is permitted * but logged since it can lead to unexpected behavior. * We also counsel against using it for / and /usr. */ if ((nbmand && ((vp->v_vfsp->vfs_flag & VFS_NBMAND) == 0)) || (!nbmand && (vp->v_vfsp->vfs_flag & VFS_NBMAND))) { cmn_err(CE_WARN, "domount: nbmand turned %s via " "remounting %s", nbmand ? "on" : "off", refstr_value(vp->v_vfsp->vfs_mntpt)); } vfsp = vp->v_vfsp; ovflags = vfsp->vfs_flag; vfsp->vfs_flag |= VFS_REMOUNT; vfsp->vfs_flag &= ~VFS_RDONLY; } else { vfsp = kmem_alloc(sizeof (vfs_t), KM_SLEEP); VFS_INIT(vfsp, vfsops, NULL); } VFS_HOLD(vfsp); /* * The vfs_reflock is not used anymore the code below explicitly * holds it preventing others accesing it directly. */ if ((sema_tryp(&vfsp->vfs_reflock) == 0) && !(vfsp->vfs_flag & VFS_REMOUNT)) cmn_err(CE_WARN, "mount type %s couldn't get vfs_reflock", vswp->vsw_name); /* * Lock the vfs. If this is a remount we want to avoid spurious umount * failures that happen as a side-effect of fsflush() and other mount * and unmount operations that might be going on simultaneously and * may have locked the vfs currently. To not return EBUSY immediately * here we use vfs_lock_wait() instead vfs_lock() for the remount case. */ if (!remount) { if (error = vfs_lock(vfsp)) { vfsp->vfs_flag = ovflags; if (splice) vn_vfsunlock(vp); if (vfsp->vfs_implp) vfsimpl_teardown(vfsp); kmem_free(vfsp, sizeof (struct vfs)); goto errout; } } else { vfs_lock_wait(vfsp); } /* * Add device to mount in progress table, global mounts require special * handling. It is possible that we have already done the lookupname * on a spliced, non-global fs. If so, we don't want to do it again * since we cannot do a lookupname after taking the * wlock above. This case is for a non-spliced, non-global filesystem. */ if (!addmip) { if ((uap->flags & MS_GLOBAL) == 0 && lookupname(uap->spec, fromspace, FOLLOW, NULL, &bvp) == 0) { addmip = 1; } } if (addmip) { bdev = bvp->v_rdev; VN_RELE(bvp); vfs_addmip(bdev, vfsp); addmip = 0; delmip = 1; } /* * Invalidate cached entry for the mount point. */ if (splice) dnlc_purge_vp(vp); /* * If have an option string but the filesystem doesn't supply a * prototype options table, create a table with the global * options and sufficient room to accept all the options in the * string. Then parse the passed in option string * accepting all the options in the string. This gives us an * option table with all the proper cancel properties for the * global options. * * Filesystems that supply a prototype options table are handled * earlier in this function. */ if (uap->flags & MS_OPTIONSTR) { if (!(vswp->vsw_flag & VSW_HASPROTO)) { mntopts_t tmp_mntopts; tmp_mntopts.mo_count = 0; vfs_createopttbl_extend(&tmp_mntopts, inargs, &mnt_mntopts); vfs_parsemntopts(&tmp_mntopts, inargs, 1); vfs_swapopttbl_nolock(&mnt_mntopts, &tmp_mntopts); vfs_freeopttbl(&tmp_mntopts); } } /* * Serialize with zone creations. */ mount_in_progress(); /* * Instantiate (or reinstantiate) the file system. If appropriate, * splice it into the file system name space. * * We want VFS_MOUNT() to be able to override the vfs_resource * string if necessary (ie, mntfs), and also for a remount to * change the same (necessary when remounting '/' during boot). * So we set up vfs_mntpt and vfs_resource to what we think they * should be, then hand off control to VFS_MOUNT() which can * override this. * * For safety's sake, when changing vfs_resource or vfs_mntpt of * a vfs which is on the vfs list (i.e. during a remount), we must * never set those fields to NULL. Several bits of code make * assumptions that the fields are always valid. */ vfs_swapopttbl(&mnt_mntopts, &vfsp->vfs_mntopts); if (remount) { if ((oldresource = vfsp->vfs_resource) != NULL) refstr_hold(oldresource); if ((oldmntpt = vfsp->vfs_mntpt) != NULL) refstr_hold(oldmntpt); } vfs_setresource(vfsp, resource); vfs_setmntpoint(vfsp, mountpt); error = VFS_MOUNT(vfsp, vp, uap, credp); if (uap->flags & MS_RDONLY) vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0); if (uap->flags & MS_NOSUID) vfs_setmntopt(vfsp, MNTOPT_NOSUID, NULL, 0); if (uap->flags & MS_GLOBAL) vfs_setmntopt(vfsp, MNTOPT_GLOBAL, NULL, 0); if (error) { if (remount) { /* put back pre-remount options */ vfs_swapopttbl(&mnt_mntopts, &vfsp->vfs_mntopts); vfs_setmntpoint(vfsp, (stripzonepath( refstr_value(oldmntpt)))); if (oldmntpt) refstr_rele(oldmntpt); vfs_setresource(vfsp, (stripzonepath( refstr_value(oldresource)))); if (oldresource) refstr_rele(oldresource); vfsp->vfs_flag = ovflags; vfs_unlock(vfsp); VFS_RELE(vfsp); } else { vfs_unlock(vfsp); vfs_freemnttab(vfsp); if (vfsp->vfs_implp) vfsimpl_teardown(vfsp); kmem_free(vfsp, sizeof (struct vfs)); } } else { /* * Set the mount time to now */ vfsp->vfs_mtime = ddi_get_time(); if (remount) { vfsp->vfs_flag &= ~VFS_REMOUNT; if (oldresource) refstr_rele(oldresource); if (oldmntpt) refstr_rele(oldmntpt); } else if (splice) { /* * Link vfsp into the name space at the mount * point. Vfs_add() is responsible for * holding the mount point which will be * released when vfs_remove() is called. */ vfs_add(vp, vfsp, uap->flags); } else { /* * Hold the reference to file system which is * not linked into the name space. */ vfsp->vfs_zone = NULL; VFS_HOLD(vfsp); vfsp->vfs_vnodecovered = NULL; } /* * Set flags for global options encountered */ if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) vfsp->vfs_flag |= VFS_RDONLY; else vfsp->vfs_flag &= ~VFS_RDONLY; if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) { vfsp->vfs_flag |= (VFS_NOSETUID|VFS_NODEVICES); } else { if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) vfsp->vfs_flag |= VFS_NODEVICES; else vfsp->vfs_flag &= ~VFS_NODEVICES; if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) vfsp->vfs_flag |= VFS_NOSETUID; else vfsp->vfs_flag &= ~VFS_NOSETUID; } if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) vfsp->vfs_flag |= VFS_NBMAND; else vfsp->vfs_flag &= ~VFS_NBMAND; if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) vfsp->vfs_flag |= VFS_XATTR; else vfsp->vfs_flag &= ~VFS_XATTR; if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) vfsp->vfs_flag |= VFS_NOEXEC; else vfsp->vfs_flag &= ~VFS_NOEXEC; /* * Now construct the output option string of options * we recognized. */ if (uap->flags & MS_OPTIONSTR) { vfs_list_read_lock(); copyout_error = vfs_buildoptionstr( &vfsp->vfs_mntopts, inargs, optlen); vfs_list_unlock(); if (copyout_error == 0 && (uap->flags & MS_SYSSPACE) == 0) { copyout_error = copyoutstr(inargs, opts, optlen, NULL); } } /* * If this isn't a remount, set up the vopstats before * anyone can touch this. We only allow spliced file * systems (file systems which are in the namespace) to * have the VFS_STATS flag set. * NOTE: PxFS mounts the underlying file system with * MS_NOSPLICE set and copies those vfs_flags to its private * vfs structure. As a result, PxFS should never have * the VFS_STATS flag or else we might access the vfs * statistics-related fields prior to them being * properly initialized. */ if (!remount && (vswp->vsw_flag & VSW_STATS) && splice) { initialize_vopstats(&vfsp->vfs_vopstats); /* * We need to set vfs_vskap to NULL because there's * a chance it won't be set below. This is checked * in teardown_vopstats() so we can't have garbage. */ vfsp->vfs_vskap = NULL; vfsp->vfs_flag |= VFS_STATS; vfsp->vfs_fstypevsp = get_fstype_vopstats(vfsp, vswp); } if (vswp->vsw_flag & VSW_XID) vfsp->vfs_flag |= VFS_XID; vfs_unlock(vfsp); } mount_completed(); if (splice) vn_vfsunlock(vp); if ((error == 0) && (copyout_error == 0)) { if (!remount) { /* * Don't call get_vskstat_anchor() while holding * locks since it allocates memory and calls * VFS_STATVFS(). For NFS, the latter can generate * an over-the-wire call. */ vskap = get_vskstat_anchor(vfsp); /* Only take the lock if we have something to do */ if (vskap != NULL) { vfs_lock_wait(vfsp); if (vfsp->vfs_flag & VFS_STATS) { vfsp->vfs_vskap = vskap; } vfs_unlock(vfsp); } } /* Return vfsp to caller. */ *vfspp = vfsp; } errout: vfs_freeopttbl(&mnt_mntopts); if (resource != NULL) kmem_free(resource, strlen(resource) + 1); if (mountpt != NULL) kmem_free(mountpt, strlen(mountpt) + 1); /* * It is possible we errored prior to adding to mount in progress * table. Must free vnode we acquired with successful lookupname. */ if (addmip) VN_RELE(bvp); if (delmip) vfs_delmip(vfsp); ASSERT(vswp != NULL); vfs_unrefvfssw(vswp); if (inargs != opts) kmem_free(inargs, MAX_MNTOPT_STR); if (copyout_error) { VFS_RELE(vfsp); error = copyout_error; } return (error); } static void vfs_setpath(struct vfs *vfsp, refstr_t **refp, const char *newpath) { size_t len; refstr_t *ref; zone_t *zone = curproc->p_zone; char *sp; int have_list_lock = 0; ASSERT(!VFS_ON_LIST(vfsp) || vfs_lock_held(vfsp)); /* * New path must be less than MAXPATHLEN because mntfs * will only display up to MAXPATHLEN bytes. This is currently * safe, because domount() uses pn_get(), and other callers * similarly cap the size to fewer than MAXPATHLEN bytes. */ ASSERT(strlen(newpath) < MAXPATHLEN); /* mntfs requires consistency while vfs list lock is held */ if (VFS_ON_LIST(vfsp)) { have_list_lock = 1; vfs_list_lock(); } if (*refp != NULL) refstr_rele(*refp); /* Do we need to modify the path? */ if (zone == global_zone || *newpath != '/') { ref = refstr_alloc(newpath); goto out; } /* * Truncate the trailing '/' in the zoneroot, and merge * in the zone's rootpath with the "newpath" (resource * or mountpoint) passed in. * * The size of the required buffer is thus the size of * the buffer required for the passed-in newpath * (strlen(newpath) + 1), plus the size of the buffer * required to hold zone_rootpath (zone_rootpathlen) * minus one for one of the now-superfluous NUL * terminations, minus one for the trailing '/'. * * That gives us: * * (strlen(newpath) + 1) + zone_rootpathlen - 1 - 1 * * Which is what we have below. */ len = strlen(newpath) + zone->zone_rootpathlen - 1; sp = kmem_alloc(len, KM_SLEEP); /* * Copy everything including the trailing slash, which * we then overwrite with the NUL character. */ (void) strcpy(sp, zone->zone_rootpath); sp[zone->zone_rootpathlen - 2] = '\0'; (void) strcat(sp, newpath); ref = refstr_alloc(sp); kmem_free(sp, len); out: *refp = ref; if (have_list_lock) { vfs_mnttab_modtimeupd(); vfs_list_unlock(); } } /* * Record a mounted resource name in a vfs structure. * If vfsp is already mounted, caller must hold the vfs lock. */ void vfs_setresource(struct vfs *vfsp, const char *resource) { if (resource == NULL || resource[0] == '\0') resource = VFS_NORESOURCE; vfs_setpath(vfsp, &vfsp->vfs_resource, resource); } /* * Record a mount point name in a vfs structure. * If vfsp is already mounted, caller must hold the vfs lock. */ void vfs_setmntpoint(struct vfs *vfsp, const char *mntpt) { if (mntpt == NULL || mntpt[0] == '\0') mntpt = VFS_NOMNTPT; vfs_setpath(vfsp, &vfsp->vfs_mntpt, mntpt); } /* Returns the vfs_resource. Caller must call refstr_rele() when finished. */ refstr_t * vfs_getresource(const struct vfs *vfsp) { refstr_t *resource; vfs_list_read_lock(); resource = vfsp->vfs_resource; refstr_hold(resource); vfs_list_unlock(); return (resource); } /* Returns the vfs_mntpt. Caller must call refstr_rele() when finished. */ refstr_t * vfs_getmntpoint(const struct vfs *vfsp) { refstr_t *mntpt; vfs_list_read_lock(); mntpt = vfsp->vfs_mntpt; refstr_hold(mntpt); vfs_list_unlock(); return (mntpt); } /* * Create an empty options table with enough empty slots to hold all * The options in the options string passed as an argument. * Potentially prepend another options table. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mops. */ static void vfs_createopttbl_extend(mntopts_t *mops, const char *opts, const mntopts_t *mtmpl) { const char *s = opts; uint_t count; if (opts == NULL || *opts == '\0') { count = 0; } else { count = 1; /* * Count number of options in the string */ for (s = strchr(s, ','); s != NULL; s = strchr(s, ',')) { count++; s++; } } vfs_copyopttbl_extend(mtmpl, mops, count); } /* * Create an empty options table with enough empty slots to hold all * The options in the options string passed as an argument. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mops. */ void vfs_createopttbl(mntopts_t *mops, const char *opts) { vfs_createopttbl_extend(mops, opts, NULL); } /* * Swap two mount options tables */ static void vfs_swapopttbl_nolock(mntopts_t *optbl1, mntopts_t *optbl2) { uint_t tmpcnt; mntopt_t *tmplist; tmpcnt = optbl2->mo_count; tmplist = optbl2->mo_list; optbl2->mo_count = optbl1->mo_count; optbl2->mo_list = optbl1->mo_list; optbl1->mo_count = tmpcnt; optbl1->mo_list = tmplist; } static void vfs_swapopttbl(mntopts_t *optbl1, mntopts_t *optbl2) { vfs_list_lock(); vfs_swapopttbl_nolock(optbl1, optbl2); vfs_mnttab_modtimeupd(); vfs_list_unlock(); } static char ** vfs_copycancelopt_extend(char **const moc, int extend) { int i = 0; int j; char **result; if (moc != NULL) { for (; moc[i] != NULL; i++) /* count number of options to cancel */; } if (i + extend == 0) return (NULL); result = kmem_alloc((i + extend + 1) * sizeof (char *), KM_SLEEP); for (j = 0; j < i; j++) { result[j] = kmem_alloc(strlen(moc[j]) + 1, KM_SLEEP); (void) strcpy(result[j], moc[j]); } for (; j <= i + extend; j++) result[j] = NULL; return (result); } static void vfs_copyopt(const mntopt_t *s, mntopt_t *d) { char *sp, *dp; d->mo_flags = s->mo_flags; d->mo_data = s->mo_data; sp = s->mo_name; if (sp != NULL) { dp = kmem_alloc(strlen(sp) + 1, KM_SLEEP); (void) strcpy(dp, sp); d->mo_name = dp; } else { d->mo_name = NULL; /* should never happen */ } d->mo_cancel = vfs_copycancelopt_extend(s->mo_cancel, 0); sp = s->mo_arg; if (sp != NULL) { dp = kmem_alloc(strlen(sp) + 1, KM_SLEEP); (void) strcpy(dp, sp); d->mo_arg = dp; } else { d->mo_arg = NULL; } } /* * Copy a mount options table, possibly allocating some spare * slots at the end. It is permissible to copy_extend the NULL table. */ static void vfs_copyopttbl_extend(const mntopts_t *smo, mntopts_t *dmo, int extra) { uint_t i, count; mntopt_t *motbl; /* * Clear out any existing stuff in the options table being initialized */ vfs_freeopttbl(dmo); count = (smo == NULL) ? 0 : smo->mo_count; if ((count + extra) == 0) /* nothing to do */ return; dmo->mo_count = count + extra; motbl = kmem_zalloc((count + extra) * sizeof (mntopt_t), KM_SLEEP); dmo->mo_list = motbl; for (i = 0; i < count; i++) { vfs_copyopt(&smo->mo_list[i], &motbl[i]); } for (i = count; i < count + extra; i++) { motbl[i].mo_flags = MO_EMPTY; } } /* * Copy a mount options table. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect smo and dmo. */ void vfs_copyopttbl(const mntopts_t *smo, mntopts_t *dmo) { vfs_copyopttbl_extend(smo, dmo, 0); } static char ** vfs_mergecancelopts(const mntopt_t *mop1, const mntopt_t *mop2) { int c1 = 0; int c2 = 0; char **result; char **sp1, **sp2, **dp; /* * First we count both lists of cancel options. * If either is NULL or has no elements, we return a copy of * the other. */ if (mop1->mo_cancel != NULL) { for (; mop1->mo_cancel[c1] != NULL; c1++) /* count cancel options in mop1 */; } if (c1 == 0) return (vfs_copycancelopt_extend(mop2->mo_cancel, 0)); if (mop2->mo_cancel != NULL) { for (; mop2->mo_cancel[c2] != NULL; c2++) /* count cancel options in mop2 */; } result = vfs_copycancelopt_extend(mop1->mo_cancel, c2); if (c2 == 0) return (result); /* * When we get here, we've got two sets of cancel options; * we need to merge the two sets. We know that the result * array has "c1+c2+1" entries and in the end we might shrink * it. * Result now has a copy of the c1 entries from mop1; we'll * now lookup all the entries of mop2 in mop1 and copy it if * it is unique. * This operation is O(n^2) but it's only called once per * filesystem per duplicate option. This is a situation * which doesn't arise with the filesystems in ON and * n is generally 1. */ dp = &result[c1]; for (sp2 = mop2->mo_cancel; *sp2 != NULL; sp2++) { for (sp1 = mop1->mo_cancel; *sp1 != NULL; sp1++) { if (strcmp(*sp1, *sp2) == 0) break; } if (*sp1 == NULL) { /* * Option *sp2 not found in mop1, so copy it. * The calls to vfs_copycancelopt_extend() * guarantee that there's enough room. */ *dp = kmem_alloc(strlen(*sp2) + 1, KM_SLEEP); (void) strcpy(*dp++, *sp2); } } if (dp != &result[c1+c2]) { size_t bytes = (dp - result + 1) * sizeof (char *); char **nres = kmem_alloc(bytes, KM_SLEEP); bcopy(result, nres, bytes); kmem_free(result, (c1 + c2 + 1) * sizeof (char *)); result = nres; } return (result); } /* * Merge two mount option tables (outer and inner) into one. This is very * similar to "merging" global variables and automatic variables in C. * * This isn't (and doesn't have to be) fast. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect omo, imo & dmo. */ void vfs_mergeopttbl(const mntopts_t *omo, const mntopts_t *imo, mntopts_t *dmo) { uint_t i, count; mntopt_t *mop, *motbl; uint_t freeidx; /* * First determine how much space we need to allocate. */ count = omo->mo_count; for (i = 0; i < imo->mo_count; i++) { if (imo->mo_list[i].mo_flags & MO_EMPTY) continue; if (vfs_hasopt(omo, imo->mo_list[i].mo_name) == NULL) count++; } ASSERT(count >= omo->mo_count && count <= omo->mo_count + imo->mo_count); motbl = kmem_alloc(count * sizeof (mntopt_t), KM_SLEEP); for (i = 0; i < omo->mo_count; i++) vfs_copyopt(&omo->mo_list[i], &motbl[i]); freeidx = omo->mo_count; for (i = 0; i < imo->mo_count; i++) { if (imo->mo_list[i].mo_flags & MO_EMPTY) continue; if ((mop = vfs_hasopt(omo, imo->mo_list[i].mo_name)) != NULL) { char **newcanp; uint_t index = mop - omo->mo_list; newcanp = vfs_mergecancelopts(mop, &motbl[index]); vfs_freeopt(&motbl[index]); vfs_copyopt(&imo->mo_list[i], &motbl[index]); vfs_freecancelopt(motbl[index].mo_cancel); motbl[index].mo_cancel = newcanp; } else { /* * If it's a new option, just copy it over to the first * free location. */ vfs_copyopt(&imo->mo_list[i], &motbl[freeidx++]); } } dmo->mo_count = count; dmo->mo_list = motbl; } /* * Functions to set and clear mount options in a mount options table. */ /* * Clear a mount option, if it exists. * * The update_mnttab arg indicates whether mops is part of a vfs that is on * the vfs list. */ static void vfs_clearmntopt_nolock(mntopts_t *mops, const char *opt, int update_mnttab) { struct mntopt *mop; uint_t i, count; ASSERT(!update_mnttab || RW_WRITE_HELD(&vfslist)); count = mops->mo_count; for (i = 0; i < count; i++) { mop = &mops->mo_list[i]; if (mop->mo_flags & MO_EMPTY) continue; if (strcmp(opt, mop->mo_name)) continue; mop->mo_flags &= ~MO_SET; if (mop->mo_arg != NULL) { kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1); } mop->mo_arg = NULL; if (update_mnttab) vfs_mnttab_modtimeupd(); break; } } void vfs_clearmntopt(struct vfs *vfsp, const char *opt) { int gotlock = 0; if (VFS_ON_LIST(vfsp)) { gotlock = 1; vfs_list_lock(); } vfs_clearmntopt_nolock(&vfsp->vfs_mntopts, opt, gotlock); if (gotlock) vfs_list_unlock(); } /* * Set a mount option on. If it's not found in the table, it's silently * ignored. If the option has MO_IGNORE set, it is still set unless the * VFS_NOFORCEOPT bit is set in the flags. Also, VFS_DISPLAY/VFS_NODISPLAY flag * bits can be used to toggle the MO_NODISPLAY bit for the option. * If the VFS_CREATEOPT flag bit is set then the first option slot with * MO_EMPTY set is created as the option passed in. * * The update_mnttab arg indicates whether mops is part of a vfs that is on * the vfs list. */ static void vfs_setmntopt_nolock(mntopts_t *mops, const char *opt, const char *arg, int flags, int update_mnttab) { mntopt_t *mop; uint_t i, count; char *sp; ASSERT(!update_mnttab || RW_WRITE_HELD(&vfslist)); if (flags & VFS_CREATEOPT) { if (vfs_hasopt(mops, opt) != NULL) { flags &= ~VFS_CREATEOPT; } } count = mops->mo_count; for (i = 0; i < count; i++) { mop = &mops->mo_list[i]; if (mop->mo_flags & MO_EMPTY) { if ((flags & VFS_CREATEOPT) == 0) continue; sp = kmem_alloc(strlen(opt) + 1, KM_SLEEP); (void) strcpy(sp, opt); mop->mo_name = sp; if (arg != NULL) mop->mo_flags = MO_HASVALUE; else mop->mo_flags = 0; } else if (strcmp(opt, mop->mo_name)) { continue; } if ((mop->mo_flags & MO_IGNORE) && (flags & VFS_NOFORCEOPT)) break; if (arg != NULL && (mop->mo_flags & MO_HASVALUE) != 0) { sp = kmem_alloc(strlen(arg) + 1, KM_SLEEP); (void) strcpy(sp, arg); } else { sp = NULL; } if (mop->mo_arg != NULL) kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1); mop->mo_arg = sp; if (flags & VFS_DISPLAY) mop->mo_flags &= ~MO_NODISPLAY; if (flags & VFS_NODISPLAY) mop->mo_flags |= MO_NODISPLAY; mop->mo_flags |= MO_SET; if (mop->mo_cancel != NULL) { char **cp; for (cp = mop->mo_cancel; *cp != NULL; cp++) vfs_clearmntopt_nolock(mops, *cp, 0); } if (update_mnttab) vfs_mnttab_modtimeupd(); break; } } void vfs_setmntopt(struct vfs *vfsp, const char *opt, const char *arg, int flags) { int gotlock = 0; if (VFS_ON_LIST(vfsp)) { gotlock = 1; vfs_list_lock(); } vfs_setmntopt_nolock(&vfsp->vfs_mntopts, opt, arg, flags, gotlock); if (gotlock) vfs_list_unlock(); } /* * Add a "tag" option to a mounted file system's options list. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mops. */ static mntopt_t * vfs_addtag(mntopts_t *mops, const char *tag) { uint_t count; mntopt_t *mop, *motbl; count = mops->mo_count + 1; motbl = kmem_zalloc(count * sizeof (mntopt_t), KM_SLEEP); if (mops->mo_count) { size_t len = (count - 1) * sizeof (mntopt_t); bcopy(mops->mo_list, motbl, len); kmem_free(mops->mo_list, len); } mops->mo_count = count; mops->mo_list = motbl; mop = &motbl[count - 1]; mop->mo_flags = MO_TAG; mop->mo_name = kmem_alloc(strlen(tag) + 1, KM_SLEEP); (void) strcpy(mop->mo_name, tag); return (mop); } /* * Allow users to set arbitrary "tags" in a vfs's mount options. * Broader use within the kernel is discouraged. */ int vfs_settag(uint_t major, uint_t minor, const char *mntpt, const char *tag, cred_t *cr) { vfs_t *vfsp; mntopts_t *mops; mntopt_t *mop; int found = 0; dev_t dev = makedevice(major, minor); int err = 0; char *buf = kmem_alloc(MAX_MNTOPT_STR, KM_SLEEP); /* * Find the desired mounted file system */ vfs_list_lock(); vfsp = rootvfs; do { if (vfsp->vfs_dev == dev && strcmp(mntpt, refstr_value(vfsp->vfs_mntpt)) == 0) { found = 1; break; } vfsp = vfsp->vfs_next; } while (vfsp != rootvfs); if (!found) { err = EINVAL; goto out; } err = secpolicy_fs_config(cr, vfsp); if (err != 0) goto out; mops = &vfsp->vfs_mntopts; /* * Add tag if it doesn't already exist */ if ((mop = vfs_hasopt(mops, tag)) == NULL) { int len; (void) vfs_buildoptionstr(mops, buf, MAX_MNTOPT_STR); len = strlen(buf); if (len + strlen(tag) + 2 > MAX_MNTOPT_STR) { err = ENAMETOOLONG; goto out; } mop = vfs_addtag(mops, tag); } if ((mop->mo_flags & MO_TAG) == 0) { err = EINVAL; goto out; } vfs_setmntopt_nolock(mops, tag, NULL, 0, 1); out: vfs_list_unlock(); kmem_free(buf, MAX_MNTOPT_STR); return (err); } /* * Allow users to remove arbitrary "tags" in a vfs's mount options. * Broader use within the kernel is discouraged. */ int vfs_clrtag(uint_t major, uint_t minor, const char *mntpt, const char *tag, cred_t *cr) { vfs_t *vfsp; mntopt_t *mop; int found = 0; dev_t dev = makedevice(major, minor); int err = 0; /* * Find the desired mounted file system */ vfs_list_lock(); vfsp = rootvfs; do { if (vfsp->vfs_dev == dev && strcmp(mntpt, refstr_value(vfsp->vfs_mntpt)) == 0) { found = 1; break; } vfsp = vfsp->vfs_next; } while (vfsp != rootvfs); if (!found) { err = EINVAL; goto out; } err = secpolicy_fs_config(cr, vfsp); if (err != 0) goto out; if ((mop = vfs_hasopt(&vfsp->vfs_mntopts, tag)) == NULL) { err = EINVAL; goto out; } if ((mop->mo_flags & MO_TAG) == 0) { err = EINVAL; goto out; } vfs_clearmntopt_nolock(&vfsp->vfs_mntopts, tag, 1); out: vfs_list_unlock(); return (err); } /* * Function to parse an option string and fill in a mount options table. * Unknown options are silently ignored. The input option string is modified * by replacing separators with nulls. If the create flag is set, options * not found in the table are just added on the fly. The table must have * an option slot marked MO_EMPTY to add an option on the fly. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mops.. */ void vfs_parsemntopts(mntopts_t *mops, char *osp, int create) { char *s = osp, *p, *nextop, *valp, *cp, *ep; int setflg = VFS_NOFORCEOPT; if (osp == NULL) return; while (*s != '\0') { p = strchr(s, ','); /* find next option */ if (p == NULL) { cp = NULL; p = s + strlen(s); } else { cp = p; /* save location of comma */ *p++ = '\0'; /* mark end and point to next option */ } nextop = p; p = strchr(s, '='); /* look for value */ if (p == NULL) { valp = NULL; /* no value supplied */ } else { ep = p; /* save location of equals */ *p++ = '\0'; /* end option and point to value */ valp = p; } /* * set option into options table */ if (create) setflg |= VFS_CREATEOPT; vfs_setmntopt_nolock(mops, s, valp, setflg, 0); if (cp != NULL) *cp = ','; /* restore the comma */ if (valp != NULL) *ep = '='; /* restore the equals */ s = nextop; } } /* * Function to inquire if an option exists in a mount options table. * Returns a pointer to the option if it exists, else NULL. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mops. */ struct mntopt * vfs_hasopt(const mntopts_t *mops, const char *opt) { struct mntopt *mop; uint_t i, count; count = mops->mo_count; for (i = 0; i < count; i++) { mop = &mops->mo_list[i]; if (mop->mo_flags & MO_EMPTY) continue; if (strcmp(opt, mop->mo_name) == 0) return (mop); } return (NULL); } /* * Function to inquire if an option is set in a mount options table. * Returns non-zero if set and fills in the arg pointer with a pointer to * the argument string or NULL if there is no argument string. */ static int vfs_optionisset_nolock(const mntopts_t *mops, const char *opt, char **argp) { struct mntopt *mop; uint_t i, count; count = mops->mo_count; for (i = 0; i < count; i++) { mop = &mops->mo_list[i]; if (mop->mo_flags & MO_EMPTY) continue; if (strcmp(opt, mop->mo_name)) continue; if ((mop->mo_flags & MO_SET) == 0) return (0); if (argp != NULL && (mop->mo_flags & MO_HASVALUE) != 0) *argp = mop->mo_arg; return (1); } return (0); } int vfs_optionisset(const struct vfs *vfsp, const char *opt, char **argp) { int ret; vfs_list_read_lock(); ret = vfs_optionisset_nolock(&vfsp->vfs_mntopts, opt, argp); vfs_list_unlock(); return (ret); } /* * Construct a comma separated string of the options set in the given * mount table, return the string in the given buffer. Return non-zero if * the buffer would overflow. * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mp. */ int vfs_buildoptionstr(const mntopts_t *mp, char *buf, int len) { char *cp; uint_t i; buf[0] = '\0'; cp = buf; for (i = 0; i < mp->mo_count; i++) { struct mntopt *mop; mop = &mp->mo_list[i]; if (mop->mo_flags & MO_SET) { int optlen, comma = 0; if (buf[0] != '\0') comma = 1; optlen = strlen(mop->mo_name); if (strlen(buf) + comma + optlen + 1 > len) goto err; if (comma) *cp++ = ','; (void) strcpy(cp, mop->mo_name); cp += optlen; /* * Append option value if there is one */ if (mop->mo_arg != NULL) { int arglen; arglen = strlen(mop->mo_arg); if (strlen(buf) + arglen + 2 > len) goto err; *cp++ = '='; (void) strcpy(cp, mop->mo_arg); cp += arglen; } } } return (0); err: return (EOVERFLOW); } static void vfs_freecancelopt(char **moc) { if (moc != NULL) { int ccnt = 0; char **cp; for (cp = moc; *cp != NULL; cp++) { kmem_free(*cp, strlen(*cp) + 1); ccnt++; } kmem_free(moc, (ccnt + 1) * sizeof (char *)); } } static void vfs_freeopt(mntopt_t *mop) { if (mop->mo_name != NULL) kmem_free(mop->mo_name, strlen(mop->mo_name) + 1); vfs_freecancelopt(mop->mo_cancel); if (mop->mo_arg != NULL) kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1); } /* * Free a mount options table * * This function is *not* for general use by filesystems. * * Note: caller is responsible for locking the vfs list, if needed, * to protect mp. */ void vfs_freeopttbl(mntopts_t *mp) { uint_t i, count; count = mp->mo_count; for (i = 0; i < count; i++) { vfs_freeopt(&mp->mo_list[i]); } if (count) { kmem_free(mp->mo_list, sizeof (mntopt_t) * count); mp->mo_count = 0; mp->mo_list = NULL; } } /* * Free any mnttab information recorded in the vfs struct. * The vfs must not be on the vfs list. */ static void vfs_freemnttab(struct vfs *vfsp) { ASSERT(!VFS_ON_LIST(vfsp)); /* * Free device and mount point information */ if (vfsp->vfs_mntpt != NULL) { refstr_rele(vfsp->vfs_mntpt); vfsp->vfs_mntpt = NULL; } if (vfsp->vfs_resource != NULL) { refstr_rele(vfsp->vfs_resource); vfsp->vfs_resource = NULL; } /* * Now free mount options information */ vfs_freeopttbl(&vfsp->vfs_mntopts); } /* * Return the last mnttab modification time */ void vfs_mnttab_modtime(timespec_t *ts) { ASSERT(RW_LOCK_HELD(&vfslist)); *ts = vfs_mnttab_mtime; } /* * See if mnttab is changed */ void vfs_mnttab_poll(timespec_t *old, struct pollhead **phpp) { int changed; *phpp = (struct pollhead *)NULL; /* * Note: don't grab vfs list lock before accessing vfs_mnttab_mtime. * Can lead to deadlock against vfs_mnttab_modtimeupd(). It is safe * to not grab the vfs list lock because tv_sec is monotonically * increasing. */ changed = (old->tv_nsec != vfs_mnttab_mtime.tv_nsec) || (old->tv_sec != vfs_mnttab_mtime.tv_sec); if (!changed) { *phpp = &vfs_pollhd; } } /* * Update the mnttab modification time and wake up any waiters for * mnttab changes */ void vfs_mnttab_modtimeupd() { hrtime_t oldhrt, newhrt; ASSERT(RW_WRITE_HELD(&vfslist)); oldhrt = ts2hrt(&vfs_mnttab_mtime); gethrestime(&vfs_mnttab_mtime); newhrt = ts2hrt(&vfs_mnttab_mtime); if (oldhrt == (hrtime_t)0) vfs_mnttab_ctime = vfs_mnttab_mtime; /* * Attempt to provide unique mtime (like uniqtime but not). */ if (newhrt == oldhrt) { newhrt++; hrt2ts(newhrt, &vfs_mnttab_mtime); } pollwakeup(&vfs_pollhd, (short)POLLRDBAND); } int dounmount(struct vfs *vfsp, int flag, cred_t *cr) { vnode_t *coveredvp; int error; extern void teardown_vopstats(vfs_t *); /* * Get covered vnode. This will be NULL if the vfs is not linked * into the file system name space (i.e., domount() with MNT_NOSPICE). */ coveredvp = vfsp->vfs_vnodecovered; ASSERT(coveredvp == NULL || vn_vfswlock_held(coveredvp)); /* * Purge all dnlc entries for this vfs. */ (void) dnlc_purge_vfsp(vfsp, 0); /* For forcible umount, skip VFS_SYNC() since it may hang */ if ((flag & MS_FORCE) == 0) (void) VFS_SYNC(vfsp, 0, cr); /* * Lock the vfs to maintain fs status quo during unmount. This * has to be done after the sync because ufs_update tries to acquire * the vfs_reflock. */ vfs_lock_wait(vfsp); if (error = VFS_UNMOUNT(vfsp, flag, cr)) { vfs_unlock(vfsp); if (coveredvp != NULL) vn_vfsunlock(coveredvp); } else if (coveredvp != NULL) { teardown_vopstats(vfsp); /* * vfs_remove() will do a VN_RELE(vfsp->vfs_vnodecovered) * when it frees vfsp so we do a VN_HOLD() so we can * continue to use coveredvp afterwards. */ VN_HOLD(coveredvp); vfs_remove(vfsp); vn_vfsunlock(coveredvp); VN_RELE(coveredvp); } else { teardown_vopstats(vfsp); /* * Release the reference to vfs that is not linked * into the name space. */ vfs_unlock(vfsp); VFS_RELE(vfsp); } return (error); } /* * Vfs_unmountall() is called by uadmin() to unmount all * mounted file systems (except the root file system) during shutdown. * It follows the existing locking protocol when traversing the vfs list * to sync and unmount vfses. Even though there should be no * other thread running while the system is shutting down, it is prudent * to still follow the locking protocol. */ void vfs_unmountall(void) { struct vfs *vfsp; struct vfs *prev_vfsp = NULL; int error; /* * Toss all dnlc entries now so that the per-vfs sync * and unmount operations don't have to slog through * a bunch of uninteresting vnodes over and over again. */ dnlc_purge(); vfs_list_lock(); for (vfsp = rootvfs->vfs_prev; vfsp != rootvfs; vfsp = prev_vfsp) { prev_vfsp = vfsp->vfs_prev; if (vfs_lock(vfsp) != 0) continue; error = vn_vfswlock(vfsp->vfs_vnodecovered); vfs_unlock(vfsp); if (error) continue; vfs_list_unlock(); (void) VFS_SYNC(vfsp, SYNC_CLOSE, CRED()); (void) dounmount(vfsp, 0, CRED()); /* * Since we dropped the vfslist lock above we must * verify that next_vfsp still exists, else start over. */ vfs_list_lock(); for (vfsp = rootvfs->vfs_prev; vfsp != rootvfs; vfsp = vfsp->vfs_prev) if (vfsp == prev_vfsp) break; if (vfsp == rootvfs && prev_vfsp != rootvfs) prev_vfsp = rootvfs->vfs_prev; } vfs_list_unlock(); } /* * Called to add an entry to the end of the vfs mount in progress list */ void vfs_addmip(dev_t dev, struct vfs *vfsp) { struct ipmnt *mipp; mipp = (struct ipmnt *)kmem_alloc(sizeof (struct ipmnt), KM_SLEEP); mipp->mip_next = NULL; mipp->mip_dev = dev; mipp->mip_vfsp = vfsp; mutex_enter(&vfs_miplist_mutex); if (vfs_miplist_end != NULL) vfs_miplist_end->mip_next = mipp; else vfs_miplist = mipp; vfs_miplist_end = mipp; mutex_exit(&vfs_miplist_mutex); } /* * Called to remove an entry from the mount in progress list * Either because the mount completed or it failed. */ void vfs_delmip(struct vfs *vfsp) { struct ipmnt *mipp, *mipprev; mutex_enter(&vfs_miplist_mutex); mipprev = NULL; for (mipp = vfs_miplist; mipp && mipp->mip_vfsp != vfsp; mipp = mipp->mip_next) { mipprev = mipp; } if (mipp == NULL) return; /* shouldn't happen */ if (mipp == vfs_miplist_end) vfs_miplist_end = mipprev; if (mipprev == NULL) vfs_miplist = mipp->mip_next; else mipprev->mip_next = mipp->mip_next; mutex_exit(&vfs_miplist_mutex); kmem_free(mipp, sizeof (struct ipmnt)); } /* * vfs_add is called by a specific filesystem's mount routine to add * the new vfs into the vfs list/hash and to cover the mounted-on vnode. * The vfs should already have been locked by the caller. * * coveredvp is NULL if this is the root. */ void vfs_add(vnode_t *coveredvp, struct vfs *vfsp, int mflag) { int newflag; ASSERT(vfs_lock_held(vfsp)); VFS_HOLD(vfsp); newflag = vfsp->vfs_flag; if (mflag & MS_RDONLY) newflag |= VFS_RDONLY; else newflag &= ~VFS_RDONLY; if (mflag & MS_NOSUID) newflag |= (VFS_NOSETUID|VFS_NODEVICES); else newflag &= ~(VFS_NOSETUID|VFS_NODEVICES); if (mflag & MS_NOMNTTAB) newflag |= VFS_NOMNTTAB; else newflag &= ~VFS_NOMNTTAB; if (coveredvp != NULL) { ASSERT(vn_vfswlock_held(coveredvp)); coveredvp->v_vfsmountedhere = vfsp; VN_HOLD(coveredvp); } vfsp->vfs_vnodecovered = coveredvp; vfsp->vfs_flag = newflag; vfs_list_add(vfsp); } /* * Remove a vfs from the vfs list, null out the pointer from the * covered vnode to the vfs (v_vfsmountedhere), and null out the pointer * from the vfs to the covered vnode (vfs_vnodecovered). Release the * reference to the vfs and to the covered vnode. * * Called from dounmount after it's confirmed with the file system * that the unmount is legal. */ void vfs_remove(struct vfs *vfsp) { vnode_t *vp; ASSERT(vfs_lock_held(vfsp)); /* * Can't unmount root. Should never happen because fs will * be busy. */ if (vfsp == rootvfs) panic("vfs_remove: unmounting root"); vfs_list_remove(vfsp); /* * Unhook from the file system name space. */ vp = vfsp->vfs_vnodecovered; ASSERT(vn_vfswlock_held(vp)); vp->v_vfsmountedhere = NULL; vfsp->vfs_vnodecovered = NULL; VN_RELE(vp); /* * Release lock and wakeup anybody waiting. */ vfs_unlock(vfsp); VFS_RELE(vfsp); } /* * Lock a filesystem to prevent access to it while mounting, * unmounting and syncing. Return EBUSY immediately if lock * can't be acquired. */ int vfs_lock(vfs_t *vfsp) { vn_vfslocks_entry_t *vpvfsentry; vpvfsentry = vn_vfslocks_getlock(vfsp); if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER)) return (0); vn_vfslocks_rele(vpvfsentry); return (EBUSY); } int vfs_rlock(vfs_t *vfsp) { vn_vfslocks_entry_t *vpvfsentry; vpvfsentry = vn_vfslocks_getlock(vfsp); if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER)) return (0); vn_vfslocks_rele(vpvfsentry); return (EBUSY); } void vfs_lock_wait(vfs_t *vfsp) { vn_vfslocks_entry_t *vpvfsentry; vpvfsentry = vn_vfslocks_getlock(vfsp); rwst_enter(&vpvfsentry->ve_lock, RW_WRITER); } void vfs_rlock_wait(vfs_t *vfsp) { vn_vfslocks_entry_t *vpvfsentry; vpvfsentry = vn_vfslocks_getlock(vfsp); rwst_enter(&vpvfsentry->ve_lock, RW_READER); } /* * Unlock a locked filesystem. */ void vfs_unlock(vfs_t *vfsp) { vn_vfslocks_entry_t *vpvfsentry; /* * vfs_unlock will mimic sema_v behaviour to fix 4748018. * And these changes should remain for the patch changes as it is. */ if (panicstr) return; /* * ve_refcount needs to be dropped twice here. * 1. To release refernce after a call to vfs_locks_getlock() * 2. To release the reference from the locking routines like * vfs_rlock_wait/vfs_wlock_wait/vfs_wlock etc,. */ vpvfsentry = vn_vfslocks_getlock(vfsp); vn_vfslocks_rele(vpvfsentry); rwst_exit(&vpvfsentry->ve_lock); vn_vfslocks_rele(vpvfsentry); } /* * Utility routine that allows a filesystem to construct its * fsid in "the usual way" - by munging some underlying dev_t and * the filesystem type number into the 64-bit fsid. Note that * this implicitly relies on dev_t persistence to make filesystem * id's persistent. * * There's nothing to prevent an individual fs from constructing its * fsid in a different way, and indeed they should. * * Since we want fsids to be 32-bit quantities (so that they can be * exported identically by either 32-bit or 64-bit APIs, as well as * the fact that fsid's are "known" to NFS), we compress the device * number given down to 32-bits, and panic if that isn't possible. */ void vfs_make_fsid(fsid_t *fsi, dev_t dev, int val) { if (!cmpldev((dev32_t *)&fsi->val[0], dev)) panic("device number too big for fsid!"); fsi->val[1] = val; } int vfs_lock_held(vfs_t *vfsp) { int held; vn_vfslocks_entry_t *vpvfsentry; /* * vfs_lock_held will mimic sema_held behaviour * if panicstr is set. And these changes should remain * for the patch changes as it is. */ if (panicstr) return (1); vpvfsentry = vn_vfslocks_getlock(vfsp); held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER); vn_vfslocks_rele(vpvfsentry); return (held); } struct _kthread * vfs_lock_owner(vfs_t *vfsp) { struct _kthread *owner; vn_vfslocks_entry_t *vpvfsentry; /* * vfs_wlock_held will mimic sema_held behaviour * if panicstr is set. And these changes should remain * for the patch changes as it is. */ if (panicstr) return (NULL); vpvfsentry = vn_vfslocks_getlock(vfsp); owner = rwst_owner(&vpvfsentry->ve_lock); vn_vfslocks_rele(vpvfsentry); return (owner); } /* * vfs list locking. * * Rather than manipulate the vfslist lock directly, we abstract into lock * and unlock routines to allow the locking implementation to be changed for * clustering. * * Whenever the vfs list is modified through its hash links, the overall list * lock must be obtained before locking the relevant hash bucket. But to see * whether a given vfs is on the list, it suffices to obtain the lock for the * hash bucket without getting the overall list lock. (See getvfs() below.) */ void vfs_list_lock() { rw_enter(&vfslist, RW_WRITER); } void vfs_list_read_lock() { rw_enter(&vfslist, RW_READER); } void vfs_list_unlock() { rw_exit(&vfslist); } /* * Low level worker routines for adding entries to and removing entries from * the vfs list. */ static void vfs_hash_add(struct vfs *vfsp, int insert_at_head) { int vhno; struct vfs **hp; dev_t dev; ASSERT(RW_WRITE_HELD(&vfslist)); dev = expldev(vfsp->vfs_fsid.val[0]); vhno = VFSHASH(getmajor(dev), getminor(dev)); mutex_enter(&rvfs_list[vhno].rvfs_lock); /* * Link into the hash table, inserting it at the end, so that LOFS * with the same fsid as UFS (or other) file systems will not hide the * UFS. */ if (insert_at_head) { vfsp->vfs_hash = rvfs_list[vhno].rvfs_head; rvfs_list[vhno].rvfs_head = vfsp; } else { for (hp = &rvfs_list[vhno].rvfs_head; *hp != NULL; hp = &(*hp)->vfs_hash) continue; /* * hp now contains the address of the pointer to update * to effect the insertion. */ vfsp->vfs_hash = NULL; *hp = vfsp; } rvfs_list[vhno].rvfs_len++; mutex_exit(&rvfs_list[vhno].rvfs_lock); } static void vfs_hash_remove(struct vfs *vfsp) { int vhno; struct vfs *tvfsp; dev_t dev; ASSERT(RW_WRITE_HELD(&vfslist)); dev = expldev(vfsp->vfs_fsid.val[0]); vhno = VFSHASH(getmajor(dev), getminor(dev)); mutex_enter(&rvfs_list[vhno].rvfs_lock); /* * Remove from hash. */ if (rvfs_list[vhno].rvfs_head == vfsp) { rvfs_list[vhno].rvfs_head = vfsp->vfs_hash; rvfs_list[vhno].rvfs_len--; goto foundit; } for (tvfsp = rvfs_list[vhno].rvfs_head; tvfsp != NULL; tvfsp = tvfsp->vfs_hash) { if (tvfsp->vfs_hash == vfsp) { tvfsp->vfs_hash = vfsp->vfs_hash; rvfs_list[vhno].rvfs_len--; goto foundit; } } cmn_err(CE_WARN, "vfs_list_remove: vfs not found in hash"); foundit: mutex_exit(&rvfs_list[vhno].rvfs_lock); } void vfs_list_add(struct vfs *vfsp) { zone_t *zone; /* * The zone that owns the mount is the one that performed the mount. * Note that this isn't necessarily the same as the zone mounted into. * The corresponding zone_rele() will be done when the vfs_t is * being free'd. */ vfsp->vfs_zone = curproc->p_zone; zone_hold(vfsp->vfs_zone); /* * Find the zone mounted into, and put this mount on its vfs list. */ zone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt)); ASSERT(zone != NULL); /* * Special casing for the root vfs. This structure is allocated * statically and hooked onto rootvfs at link time. During the * vfs_mountroot call at system startup time, the root file system's * VFS_MOUNTROOT routine will call vfs_add with this root vfs struct * as argument. The code below must detect and handle this special * case. The only apparent justification for this special casing is * to ensure that the root file system appears at the head of the * list. * * XXX: I'm assuming that it's ok to do normal list locking when * adding the entry for the root file system (this used to be * done with no locks held). */ vfs_list_lock(); /* * Link into the vfs list proper. */ if (vfsp == &root) { /* * Assert: This vfs is already on the list as its first entry. * Thus, there's nothing to do. */ ASSERT(rootvfs == vfsp); /* * Add it to the head of the global zone's vfslist. */ ASSERT(zone == global_zone); ASSERT(zone->zone_vfslist == NULL); zone->zone_vfslist = vfsp; } else { /* * Link to end of list using vfs_prev (as rootvfs is now a * doubly linked circular list) so list is in mount order for * mnttab use. */ rootvfs->vfs_prev->vfs_next = vfsp; vfsp->vfs_prev = rootvfs->vfs_prev; rootvfs->vfs_prev = vfsp; vfsp->vfs_next = rootvfs; /* * Do it again for the zone-private list (which may be NULL). */ if (zone->zone_vfslist == NULL) { ASSERT(zone != global_zone); zone->zone_vfslist = vfsp; } else { zone->zone_vfslist->vfs_zone_prev->vfs_zone_next = vfsp; vfsp->vfs_zone_prev = zone->zone_vfslist->vfs_zone_prev; zone->zone_vfslist->vfs_zone_prev = vfsp; vfsp->vfs_zone_next = zone->zone_vfslist; } } /* * Link into the hash table, inserting it at the end, so that LOFS * with the same fsid as UFS (or other) file systems will not hide * the UFS. */ vfs_hash_add(vfsp, 0); /* * update the mnttab modification time */ vfs_mnttab_modtimeupd(); vfs_list_unlock(); zone_rele(zone); } void vfs_list_remove(struct vfs *vfsp) { zone_t *zone; zone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt)); ASSERT(zone != NULL); /* * Callers are responsible for preventing attempts to unmount the * root. */ ASSERT(vfsp != rootvfs); vfs_list_lock(); /* * Remove from hash. */ vfs_hash_remove(vfsp); /* * Remove from vfs list. */ vfsp->vfs_prev->vfs_next = vfsp->vfs_next; vfsp->vfs_next->vfs_prev = vfsp->vfs_prev; vfsp->vfs_next = vfsp->vfs_prev = NULL; /* * Remove from zone-specific vfs list. */ if (zone->zone_vfslist == vfsp) zone->zone_vfslist = vfsp->vfs_zone_next; if (vfsp->vfs_zone_next == vfsp) { ASSERT(vfsp->vfs_zone_prev == vfsp); ASSERT(zone->zone_vfslist == vfsp); zone->zone_vfslist = NULL; } vfsp->vfs_zone_prev->vfs_zone_next = vfsp->vfs_zone_next; vfsp->vfs_zone_next->vfs_zone_prev = vfsp->vfs_zone_prev; vfsp->vfs_zone_next = vfsp->vfs_zone_prev = NULL; /* * update the mnttab modification time */ vfs_mnttab_modtimeupd(); vfs_list_unlock(); zone_rele(zone); } struct vfs * getvfs(fsid_t *fsid) { struct vfs *vfsp; int val0 = fsid->val[0]; int val1 = fsid->val[1]; dev_t dev = expldev(val0); int vhno = VFSHASH(getmajor(dev), getminor(dev)); kmutex_t *hmp = &rvfs_list[vhno].rvfs_lock; mutex_enter(hmp); for (vfsp = rvfs_list[vhno].rvfs_head; vfsp; vfsp = vfsp->vfs_hash) { if (vfsp->vfs_fsid.val[0] == val0 && vfsp->vfs_fsid.val[1] == val1) { VFS_HOLD(vfsp); mutex_exit(hmp); return (vfsp); } } mutex_exit(hmp); return (NULL); } /* * Search the vfs mount in progress list for a specified device/vfs entry. * Returns 0 if the first entry in the list that the device matches has the * given vfs pointer as well. If the device matches but a different vfs * pointer is encountered in the list before the given vfs pointer then * a 1 is returned. */ int vfs_devmounting(dev_t dev, struct vfs *vfsp) { int retval = 0; struct ipmnt *mipp; mutex_enter(&vfs_miplist_mutex); for (mipp = vfs_miplist; mipp != NULL; mipp = mipp->mip_next) { if (mipp->mip_dev == dev) { if (mipp->mip_vfsp != vfsp) retval = 1; break; } } mutex_exit(&vfs_miplist_mutex); return (retval); } /* * Search the vfs list for a specified device. Returns 1, if entry is found * or 0 if no suitable entry is found. */ int vfs_devismounted(dev_t dev) { struct vfs *vfsp; int found; vfs_list_read_lock(); vfsp = rootvfs; found = 0; do { if (vfsp->vfs_dev == dev) { found = 1; break; } vfsp = vfsp->vfs_next; } while (vfsp != rootvfs); vfs_list_unlock(); return (found); } /* * Search the vfs list for a specified device. Returns a pointer to it * or NULL if no suitable entry is found. The caller of this routine * is responsible for releasing the returned vfs pointer. */ struct vfs * vfs_dev2vfsp(dev_t dev) { struct vfs *vfsp; int found; vfs_list_read_lock(); vfsp = rootvfs; found = 0; do { /* * The following could be made more efficient by making * the entire loop use vfs_zone_next if the call is from * a zone. The only callers, however, ustat(2) and * umount2(2), don't seem to justify the added * complexity at present. */ if (vfsp->vfs_dev == dev && ZONE_PATH_VISIBLE(refstr_value(vfsp->vfs_mntpt), curproc->p_zone)) { VFS_HOLD(vfsp); found = 1; break; } vfsp = vfsp->vfs_next; } while (vfsp != rootvfs); vfs_list_unlock(); return (found ? vfsp: NULL); } /* * Search the vfs list for a specified mntpoint. Returns a pointer to it * or NULL if no suitable entry is found. The caller of this routine * is responsible for releasing the returned vfs pointer. * * Note that if multiple mntpoints match, the last one matching is * returned in an attempt to return the "top" mount when overlay * mounts are covering the same mount point. This is accomplished by starting * at the end of the list and working our way backwards, stopping at the first * matching mount. */ struct vfs * vfs_mntpoint2vfsp(const char *mp) { struct vfs *vfsp; struct vfs *retvfsp = NULL; zone_t *zone = curproc->p_zone; struct vfs *list; vfs_list_read_lock(); if (getzoneid() == GLOBAL_ZONEID) { /* * The global zone may see filesystems in any zone. */ vfsp = rootvfs->vfs_prev; do { if (strcmp(refstr_value(vfsp->vfs_mntpt), mp) == 0) { retvfsp = vfsp; break; } vfsp = vfsp->vfs_prev; } while (vfsp != rootvfs->vfs_prev); } else if ((list = zone->zone_vfslist) != NULL) { const char *mntpt; vfsp = list->vfs_zone_prev; do { mntpt = refstr_value(vfsp->vfs_mntpt); mntpt = ZONE_PATH_TRANSLATE(mntpt, zone); if (strcmp(mntpt, mp) == 0) { retvfsp = vfsp; break; } vfsp = vfsp->vfs_zone_prev; } while (vfsp != list->vfs_zone_prev); } if (retvfsp) VFS_HOLD(retvfsp); vfs_list_unlock(); return (retvfsp); } /* * Search the vfs list for a specified vfsops. * if vfs entry is found then return 1, else 0. */ int vfs_opsinuse(vfsops_t *ops) { struct vfs *vfsp; int found; vfs_list_read_lock(); vfsp = rootvfs; found = 0; do { if (vfs_getops(vfsp) == ops) { found = 1; break; } vfsp = vfsp->vfs_next; } while (vfsp != rootvfs); vfs_list_unlock(); return (found); } /* * Allocate an entry in vfssw for a file system type */ struct vfssw * allocate_vfssw(char *type) { struct vfssw *vswp; if (type[0] == '\0' || strlen(type) + 1 > _ST_FSTYPSZ) { /* * The vfssw table uses the empty string to identify an * available entry; we cannot add any type which has * a leading NUL. The string length is limited to * the size of the st_fstype array in struct stat. */ return (NULL); } ASSERT(VFSSW_WRITE_LOCKED()); for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) if (!ALLOCATED_VFSSW(vswp)) { vswp->vsw_name = kmem_alloc(strlen(type) + 1, KM_SLEEP); (void) strcpy(vswp->vsw_name, type); ASSERT(vswp->vsw_count == 0); vswp->vsw_count = 1; mutex_init(&vswp->vsw_lock, NULL, MUTEX_DEFAULT, NULL); return (vswp); } return (NULL); } /* * Impose additional layer of translation between vfstype names * and module names in the filesystem. */ static char * vfs_to_modname(char *vfstype) { if (strcmp(vfstype, "proc") == 0) { vfstype = "procfs"; } else if (strcmp(vfstype, "fd") == 0) { vfstype = "fdfs"; } else if (strncmp(vfstype, "nfs", 3) == 0) { vfstype = "nfs"; } return (vfstype); } /* * Find a vfssw entry given a file system type name. * Try to autoload the filesystem if it's not found. * If it's installed, return the vfssw locked to prevent unloading. */ struct vfssw * vfs_getvfssw(char *type) { struct vfssw *vswp; char *modname; RLOCK_VFSSW(); vswp = vfs_getvfsswbyname(type); modname = vfs_to_modname(type); if (rootdir == NULL) { /* * If we haven't yet loaded the root file system, then our * _init won't be called until later. Allocate vfssw entry, * because mod_installfs won't be called. */ if (vswp == NULL) { RUNLOCK_VFSSW(); WLOCK_VFSSW(); if ((vswp = vfs_getvfsswbyname(type)) == NULL) { if ((vswp = allocate_vfssw(type)) == NULL) { WUNLOCK_VFSSW(); return (NULL); } } WUNLOCK_VFSSW(); RLOCK_VFSSW(); } if (!VFS_INSTALLED(vswp)) { RUNLOCK_VFSSW(); (void) modloadonly("fs", modname); } else RUNLOCK_VFSSW(); return (vswp); } /* * Try to load the filesystem. Before calling modload(), we drop * our lock on the VFS switch table, and pick it up after the * module is loaded. However, there is a potential race: the * module could be unloaded after the call to modload() completes * but before we pick up the lock and drive on. Therefore, * we keep reloading the module until we've loaded the module * _and_ we have the lock on the VFS switch table. */ while (vswp == NULL || !VFS_INSTALLED(vswp)) { RUNLOCK_VFSSW(); if (modload("fs", modname) == -1) return (NULL); RLOCK_VFSSW(); if (vswp == NULL) if ((vswp = vfs_getvfsswbyname(type)) == NULL) break; } RUNLOCK_VFSSW(); return (vswp); } /* * Find a vfssw entry given a file system type name. */ struct vfssw * vfs_getvfsswbyname(char *type) { struct vfssw *vswp; ASSERT(VFSSW_LOCKED()); if (type == NULL || *type == '\0') return (NULL); for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) { if (strcmp(type, vswp->vsw_name) == 0) { vfs_refvfssw(vswp); return (vswp); } } return (NULL); } /* * Find a vfssw entry given a set of vfsops. */ struct vfssw * vfs_getvfsswbyvfsops(vfsops_t *vfsops) { struct vfssw *vswp; RLOCK_VFSSW(); for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) { if (ALLOCATED_VFSSW(vswp) && &vswp->vsw_vfsops == vfsops) { vfs_refvfssw(vswp); RUNLOCK_VFSSW(); return (vswp); } } RUNLOCK_VFSSW(); return (NULL); } /* * Reference a vfssw entry. */ void vfs_refvfssw(struct vfssw *vswp) { mutex_enter(&vswp->vsw_lock); vswp->vsw_count++; mutex_exit(&vswp->vsw_lock); } /* * Unreference a vfssw entry. */ void vfs_unrefvfssw(struct vfssw *vswp) { mutex_enter(&vswp->vsw_lock); vswp->vsw_count--; mutex_exit(&vswp->vsw_lock); } int sync_timeout = 30; /* timeout for syncing a page during panic */ int sync_timeleft; /* portion of sync_timeout remaining */ static int sync_retries = 20; /* number of retries when not making progress */ static int sync_triesleft; /* portion of sync_retries remaining */ static pgcnt_t old_pgcnt, new_pgcnt; static int new_bufcnt, old_bufcnt; /* * Sync all of the mounted filesystems, and then wait for the actual i/o to * complete. We wait by counting the number of dirty pages and buffers, * pushing them out using bio_busy() and page_busy(), and then counting again. * This routine is used during both the uadmin A_SHUTDOWN code as well as * the SYNC phase of the panic code (see comments in panic.c). It should only * be used after some higher-level mechanism has quiesced the system so that * new writes are not being initiated while we are waiting for completion. * * To ensure finite running time, our algorithm uses two timeout mechanisms: * sync_timeleft (a timer implemented by the omnipresent deadman() cyclic), and * sync_triesleft (a progress counter used by the vfs_syncall() loop below). * Together these ensure that syncing completes if our i/o paths are stuck. * The counters are declared above so they can be found easily in the debugger. * * The sync_timeleft counter is reset by bio_busy() and page_busy() using the * vfs_syncprogress() subroutine whenever we make progress through the lists of * pages and buffers. It is decremented and expired by the deadman() cyclic. * When vfs_syncall() decides it is done, we disable the deadman() counter by * setting sync_timeleft to zero. This timer guards against vfs_syncall() * deadlocking or hanging inside of a broken filesystem or driver routine. * * The sync_triesleft counter is updated by vfs_syncall() itself. If we make * sync_retries consecutive calls to bio_busy() and page_busy() without * decreasing either the number of dirty buffers or dirty pages below the * lowest count we have seen so far, we give up and return from vfs_syncall(). * * Each loop iteration ends with a call to delay() one second to allow time for * i/o completion and to permit the user time to read our progress messages. */ void vfs_syncall(void) { if (rootdir == NULL && !modrootloaded) return; /* panic during boot - no filesystems yet */ printf("syncing file systems..."); vfs_syncprogress(); sync(); vfs_syncprogress(); sync_triesleft = sync_retries; old_bufcnt = new_bufcnt = INT_MAX; old_pgcnt = new_pgcnt = ULONG_MAX; while (sync_triesleft > 0) { old_bufcnt = MIN(old_bufcnt, new_bufcnt); old_pgcnt = MIN(old_pgcnt, new_pgcnt); new_bufcnt = bio_busy(B_TRUE); new_pgcnt = page_busy(B_TRUE); vfs_syncprogress(); if (new_bufcnt == 0 && new_pgcnt == 0) break; if (new_bufcnt < old_bufcnt || new_pgcnt < old_pgcnt) sync_triesleft = sync_retries; else sync_triesleft--; if (new_bufcnt) printf(" [%d]", new_bufcnt); if (new_pgcnt) printf(" %lu", new_pgcnt); delay(hz); } if (new_bufcnt != 0 || new_pgcnt != 0) printf(" done (not all i/o completed)\n"); else printf(" done\n"); sync_timeleft = 0; delay(hz); } /* * If we are in the middle of the sync phase of panic, reset sync_timeleft to * sync_timeout to indicate that we are making progress and the deadman() * omnipresent cyclic should not yet time us out. Note that it is safe to * store to sync_timeleft here since the deadman() is firing at high-level * on top of us. If we are racing with the deadman(), either the deadman() * will decrement the old value and then we will reset it, or we will * reset it and then the deadman() will immediately decrement it. In either * case, correct behavior results. */ void vfs_syncprogress(void) { if (panicstr) sync_timeleft = sync_timeout; } /* * Map VFS flags to statvfs flags. These shouldn't really be separate * flags at all. */ uint_t vf_to_stf(uint_t vf) { uint_t stf = 0; if (vf & VFS_RDONLY) stf |= ST_RDONLY; if (vf & VFS_NOSETUID) stf |= ST_NOSUID; if (vf & VFS_NOTRUNC) stf |= ST_NOTRUNC; return (stf); } /* * Entries for (illegal) fstype 0. */ /* ARGSUSED */ int vfsstray_sync(struct vfs *vfsp, short arg, struct cred *cr) { cmn_err(CE_PANIC, "stray vfs operation"); return (0); } /* * Entries for (illegal) fstype 0. */ int vfsstray(void) { cmn_err(CE_PANIC, "stray vfs operation"); return (0); } /* * Support for dealing with forced UFS unmount and its interaction with * LOFS. Could be used by any filesystem. * See bug 1203132. */ int vfs_EIO(void) { return (EIO); } /* * We've gotta define the op for sync separately, since the compiler gets * confused if we mix and match ANSI and normal style prototypes when * a "short" argument is present and spits out a warning. */ /*ARGSUSED*/ int vfs_EIO_sync(struct vfs *vfsp, short arg, struct cred *cr) { return (EIO); } vfs_t EIO_vfs; vfsops_t *EIO_vfsops; /* * Called from startup() to initialize all loaded vfs's */ void vfsinit(void) { struct vfssw *vswp; int error; extern int vopstats_enabled; extern void vopstats_startup(); static const fs_operation_def_t EIO_vfsops_template[] = { VFSNAME_MOUNT, { .error = vfs_EIO }, VFSNAME_UNMOUNT, { .error = vfs_EIO }, VFSNAME_ROOT, { .error = vfs_EIO }, VFSNAME_STATVFS, { .error = vfs_EIO }, VFSNAME_SYNC, { .vfs_sync = vfs_EIO_sync }, VFSNAME_VGET, { .error = vfs_EIO }, VFSNAME_MOUNTROOT, { .error = vfs_EIO }, VFSNAME_FREEVFS, { .error = vfs_EIO }, VFSNAME_VNSTATE, { .error = vfs_EIO }, NULL, NULL }; static const fs_operation_def_t stray_vfsops_template[] = { VFSNAME_MOUNT, { .error = vfsstray }, VFSNAME_UNMOUNT, { .error = vfsstray }, VFSNAME_ROOT, { .error = vfsstray }, VFSNAME_STATVFS, { .error = vfsstray }, VFSNAME_SYNC, { .vfs_sync = vfsstray_sync }, VFSNAME_VGET, { .error = vfsstray }, VFSNAME_MOUNTROOT, { .error = vfsstray }, VFSNAME_FREEVFS, { .error = vfsstray }, VFSNAME_VNSTATE, { .error = vfsstray }, NULL, NULL }; /* Initialize the vnode cache (file systems may use it during init). */ vn_create_cache(); /* Setup event monitor framework */ fem_init(); /* Initialize the dummy stray file system type. */ error = vfs_setfsops(0, stray_vfsops_template, NULL); /* Initialize the dummy EIO file system. */ error = vfs_makefsops(EIO_vfsops_template, &EIO_vfsops); if (error != 0) { cmn_err(CE_WARN, "vfsinit: bad EIO vfs ops template"); /* Shouldn't happen, but not bad enough to panic */ } VFS_INIT(&EIO_vfs, EIO_vfsops, (caddr_t)NULL); /* * Default EIO_vfs.vfs_flag to VFS_UNMOUNTED so a lookup * on this vfs can immediately notice it's invalid. */ EIO_vfs.vfs_flag |= VFS_UNMOUNTED; /* * Call the init routines of non-loadable filesystems only. * Filesystems which are loaded as separate modules will be * initialized by the module loading code instead. */ for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) { RLOCK_VFSSW(); if (vswp->vsw_init != NULL) (*vswp->vsw_init)(vswp - vfssw, vswp->vsw_name); RUNLOCK_VFSSW(); } vopstats_startup(); if (vopstats_enabled) { /* EIO_vfs can collect stats, but we don't retrieve them */ initialize_vopstats(&EIO_vfs.vfs_vopstats); EIO_vfs.vfs_fstypevsp = NULL; EIO_vfs.vfs_vskap = NULL; EIO_vfs.vfs_flag |= VFS_STATS; } } /* * Increments the vfs reference count by one atomically. */ void vfs_hold(vfs_t *vfsp) { atomic_add_32(&vfsp->vfs_count, 1); ASSERT(vfsp->vfs_count != 0); } /* * Decrements the vfs reference count by one atomically. When * vfs reference count becomes zero, it calls the file system * specific vfs_freevfs() to free up the resources. */ void vfs_rele(vfs_t *vfsp) { ASSERT(vfsp->vfs_count != 0); if (atomic_add_32_nv(&vfsp->vfs_count, -1) == 0) { VFS_FREEVFS(vfsp); if (vfsp->vfs_zone) zone_rele(vfsp->vfs_zone); vfs_freemnttab(vfsp); if (vfsp->vfs_implp) vfsimpl_teardown(vfsp); sema_destroy(&vfsp->vfs_reflock); kmem_free(vfsp, sizeof (*vfsp)); } } /* * Generic operations vector support. * * This is used to build operations vectors for both the vfs and vnode. * It's normally called only when a file system is loaded. * * There are many possible algorithms for this, including the following: * * (1) scan the list of known operations; for each, see if the file system * includes an entry for it, and fill it in as appropriate. * * (2) set up defaults for all known operations. scan the list of ops * supplied by the file system; for each which is both supplied and * known, fill it in. * * (3) sort the lists of known ops & supplied ops; scan the list, filling * in entries as we go. * * we choose (1) for simplicity, and because performance isn't critical here. * note that (2) could be sped up using a precomputed hash table on known ops. * (3) could be faster than either, but only if the lists were very large or * supplied in sorted order. * */ int fs_build_vector(void *vector, int *unused_ops, const fs_operation_trans_def_t *translation, const fs_operation_def_t *operations) { int i, num_trans, num_ops, used; /* * Count the number of translations and the number of supplied * operations. */ { const fs_operation_trans_def_t *p; for (num_trans = 0, p = translation; p->name != NULL; num_trans++, p++) ; } { const fs_operation_def_t *p; for (num_ops = 0, p = operations; p->name != NULL; num_ops++, p++) ; } /* Walk through each operation known to our caller. There will be */ /* one entry in the supplied "translation table" for each. */ used = 0; for (i = 0; i < num_trans; i++) { int j, found; char *curname; fs_generic_func_p result; fs_generic_func_p *location; curname = translation[i].name; /* Look for a matching operation in the list supplied by the */ /* file system. */ found = 0; for (j = 0; j < num_ops; j++) { if (strcmp(operations[j].name, curname) == 0) { used++; found = 1; break; } } /* * If the file system is using a "placeholder" for default * or error functions, grab the appropriate function out of * the translation table. If the file system didn't supply * this operation at all, use the default function. */ if (found) { result = operations[j].func.fs_generic; if (result == fs_default) { result = translation[i].defaultFunc; } else if (result == fs_error) { result = translation[i].errorFunc; } else if (result == NULL) { /* Null values are PROHIBITED */ return (EINVAL); } } else { result = translation[i].defaultFunc; } /* Now store the function into the operations vector. */ location = (fs_generic_func_p *) (((char *)vector) + translation[i].offset); *location = result; } *unused_ops = num_ops - used; return (0); } /* Placeholder functions, should never be called. */ int fs_error(void) { cmn_err(CE_PANIC, "fs_error called"); return (0); } int fs_default(void) { cmn_err(CE_PANIC, "fs_default called"); return (0); } #ifdef __sparc /* * Part of the implementation of booting off a mirrored root * involves a change of dev_t for the root device. To * accomplish this, first remove the existing hash table * entry for the root device, convert to the new dev_t, * then re-insert in the hash table at the head of the list. */ void vfs_root_redev(vfs_t *vfsp, dev_t ndev, int fstype) { vfs_list_lock(); vfs_hash_remove(vfsp); vfsp->vfs_dev = ndev; vfs_make_fsid(&vfsp->vfs_fsid, ndev, fstype); vfs_hash_add(vfsp, 1); vfs_list_unlock(); } #else /* x86 NEWBOOT */ int rootconf() { int error; struct vfssw *vsw; extern void pm_init(); char *fstyp; fstyp = getrootfs(); if (error = clboot_rootconf()) return (error); if (modload("fs", fstyp) == -1) panic("Cannot _init %s module", fstyp); RLOCK_VFSSW(); vsw = vfs_getvfsswbyname(fstyp); RUNLOCK_VFSSW(); VFS_INIT(rootvfs, &vsw->vsw_vfsops, 0); VFS_HOLD(rootvfs); /* always mount readonly first */ rootvfs->vfs_flag |= VFS_RDONLY; pm_init(); if (netboot) (void) strplumb(); error = VFS_MOUNTROOT(rootvfs, ROOT_INIT); vfs_unrefvfssw(vsw); rootdev = rootvfs->vfs_dev; if (error) panic("cannot mount root path %s", rootfs.bo_name); return (error); } /* * XXX this is called by nfs only and should probably be removed * If booted with ASKNAME, prompt on the console for a filesystem * name and return it. */ void getfsname(char *askfor, char *name, size_t namelen) { if (boothowto & RB_ASKNAME) { printf("%s name: ", askfor); console_gets(name, namelen); } } /* * If server_path exists, then we are booting a diskless * client. Otherwise, we default to ufs. Zfs should perhaps be * another property. */ static char * getrootfs(void) { extern char *strplumb_get_netdev_path(void); char *propstr = NULL; /* check fstype property; it should be nfsdyn for diskless */ if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(), DDI_PROP_DONTPASS, "fstype", &propstr) == DDI_SUCCESS) { (void) strncpy(rootfs.bo_fstype, propstr, BO_MAXFSNAME); ddi_prop_free(propstr); /* * if the boot property 'fstype' is not set, but 'zfs-bootfs' is set, * assume the type of this root filesystem is 'zfs'. */ } else if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(), DDI_PROP_DONTPASS, "zfs-bootfs", &propstr) == DDI_SUCCESS) { (void) strncpy(rootfs.bo_fstype, "zfs", BO_MAXFSNAME); ddi_prop_free(propstr); } if (strncmp(rootfs.bo_fstype, "nfs", 3) != 0) return (rootfs.bo_fstype); ++netboot; /* check if path to network interface is specified in bootpath */ if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(), DDI_PROP_DONTPASS, "bootpath", &propstr) == DDI_SUCCESS) { (void) strncpy(rootfs.bo_name, propstr, BO_MAXOBJNAME); ddi_prop_free(propstr); } else { /* attempt to determine netdev_path via boot_mac address */ netdev_path = strplumb_get_netdev_path(); if (netdev_path == NULL) panic("cannot find boot network interface"); (void) strncpy(rootfs.bo_name, netdev_path, BO_MAXOBJNAME); } return ("nfs"); } #endif