freebsd-skq/sys/kern/uipc_mqueue.c

2919 lines
64 KiB
C
Raw Normal View History

/*-
* Copyright (c) 2005 David Xu <davidxu@freebsd.org>
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*/
/*
* POSIX message queue implementation.
*
* 1) A mqueue filesystem can be mounted, each message queue appears
* in mounted directory, user can change queue's permission and
* ownership, or remove a queue. Manually creating a file in the
* directory causes a message queue to be created in the kernel with
* default message queue attributes applied and same name used, this
* method is not advocated since mq_open syscall allows user to specify
* different attributes. Also the file system can be mounted multiple
* times at different mount points but shows same contents.
*
* 2) Standard POSIX message queue API. The syscalls do not use vfs layer,
* but directly operate on internal data structure, this allows user to
* use the IPC facility without having to mount mqueue file system.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
#include "opt_capsicum.h"
#include "opt_compat.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/buf.h>
#include <sys/capsicum.h>
#include <sys/dirent.h>
#include <sys/event.h>
#include <sys/eventhandler.h>
#include <sys/fcntl.h>
#include <sys/file.h>
#include <sys/filedesc.h>
#include <sys/jail.h>
#include <sys/lock.h>
#include <sys/module.h>
#include <sys/mount.h>
#include <sys/mqueue.h>
#include <sys/mutex.h>
#include <sys/namei.h>
#include <sys/posix4.h>
#include <sys/poll.h>
#include <sys/priv.h>
#include <sys/proc.h>
#include <sys/queue.h>
#include <sys/sysproto.h>
#include <sys/stat.h>
#include <sys/syscall.h>
#include <sys/syscallsubr.h>
#include <sys/sysent.h>
#include <sys/sx.h>
#include <sys/sysctl.h>
#include <sys/taskqueue.h>
#include <sys/unistd.h>
#include <sys/user.h>
#include <sys/vnode.h>
#include <machine/atomic.h>
FEATURE(p1003_1b_mqueue, "POSIX P1003.1B message queues support");
/*
* Limits and constants
*/
#define MQFS_NAMELEN NAME_MAX
#define MQFS_DELEN (8 + MQFS_NAMELEN)
/* node types */
typedef enum {
mqfstype_none = 0,
mqfstype_root,
mqfstype_dir,
mqfstype_this,
mqfstype_parent,
mqfstype_file,
mqfstype_symlink,
} mqfs_type_t;
struct mqfs_node;
/*
* mqfs_info: describes a mqfs instance
*/
struct mqfs_info {
struct sx mi_lock;
struct mqfs_node *mi_root;
struct unrhdr *mi_unrhdr;
};
struct mqfs_vdata {
LIST_ENTRY(mqfs_vdata) mv_link;
struct mqfs_node *mv_node;
struct vnode *mv_vnode;
struct task mv_task;
};
/*
* mqfs_node: describes a node (file or directory) within a mqfs
*/
struct mqfs_node {
char mn_name[MQFS_NAMELEN+1];
struct mqfs_info *mn_info;
struct mqfs_node *mn_parent;
LIST_HEAD(,mqfs_node) mn_children;
LIST_ENTRY(mqfs_node) mn_sibling;
LIST_HEAD(,mqfs_vdata) mn_vnodes;
const void *mn_pr_root;
int mn_refcount;
mqfs_type_t mn_type;
int mn_deleted;
uint32_t mn_fileno;
void *mn_data;
struct timespec mn_birth;
struct timespec mn_ctime;
struct timespec mn_atime;
struct timespec mn_mtime;
uid_t mn_uid;
gid_t mn_gid;
int mn_mode;
};
#define VTON(vp) (((struct mqfs_vdata *)((vp)->v_data))->mv_node)
#define VTOMQ(vp) ((struct mqueue *)(VTON(vp)->mn_data))
#define VFSTOMQFS(m) ((struct mqfs_info *)((m)->mnt_data))
#define FPTOMQ(fp) ((struct mqueue *)(((struct mqfs_node *) \
(fp)->f_data)->mn_data))
TAILQ_HEAD(msgq, mqueue_msg);
struct mqueue;
struct mqueue_notifier {
LIST_ENTRY(mqueue_notifier) nt_link;
struct sigevent nt_sigev;
ksiginfo_t nt_ksi;
struct proc *nt_proc;
};
struct mqueue {
struct mtx mq_mutex;
int mq_flags;
long mq_maxmsg;
long mq_msgsize;
long mq_curmsgs;
long mq_totalbytes;
struct msgq mq_msgq;
int mq_receivers;
int mq_senders;
struct selinfo mq_rsel;
struct selinfo mq_wsel;
struct mqueue_notifier *mq_notifier;
};
#define MQ_RSEL 0x01
#define MQ_WSEL 0x02
struct mqueue_msg {
TAILQ_ENTRY(mqueue_msg) msg_link;
unsigned int msg_prio;
unsigned int msg_size;
/* following real data... */
};
static SYSCTL_NODE(_kern, OID_AUTO, mqueue, CTLFLAG_RW, 0,
"POSIX real time message queue");
static int default_maxmsg = 10;
static int default_msgsize = 1024;
static int maxmsg = 100;
SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsg, CTLFLAG_RW,
&maxmsg, 0, "Default maximum messages in queue");
static int maxmsgsize = 16384;
SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmsgsize, CTLFLAG_RW,
&maxmsgsize, 0, "Default maximum message size");
static int maxmq = 100;
SYSCTL_INT(_kern_mqueue, OID_AUTO, maxmq, CTLFLAG_RW,
&maxmq, 0, "maximum message queues");
static int curmq = 0;
SYSCTL_INT(_kern_mqueue, OID_AUTO, curmq, CTLFLAG_RW,
&curmq, 0, "current message queue number");
static int unloadable = 0;
static MALLOC_DEFINE(M_MQUEUEDATA, "mqdata", "mqueue data");
static eventhandler_tag exit_tag;
/* Only one instance per-system */
static struct mqfs_info mqfs_data;
static uma_zone_t mqnode_zone;
static uma_zone_t mqueue_zone;
static uma_zone_t mvdata_zone;
static uma_zone_t mqnoti_zone;
static struct vop_vector mqfs_vnodeops;
static struct fileops mqueueops;
static unsigned mqfs_osd_jail_slot;
/*
* Directory structure construction and manipulation
*/
#ifdef notyet
static struct mqfs_node *mqfs_create_dir(struct mqfs_node *parent,
const char *name, int namelen, struct ucred *cred, int mode);
static struct mqfs_node *mqfs_create_link(struct mqfs_node *parent,
const char *name, int namelen, struct ucred *cred, int mode);
#endif
static struct mqfs_node *mqfs_create_file(struct mqfs_node *parent,
const char *name, int namelen, struct ucred *cred, int mode);
static int mqfs_destroy(struct mqfs_node *mn);
static void mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn);
static void mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn);
static int mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn);
static int mqfs_prison_remove(void *obj, void *data);
/*
* Message queue construction and maniplation
*/
static struct mqueue *mqueue_alloc(const struct mq_attr *attr);
static void mqueue_free(struct mqueue *mq);
static int mqueue_send(struct mqueue *mq, const char *msg_ptr,
size_t msg_len, unsigned msg_prio, int waitok,
const struct timespec *abs_timeout);
static int mqueue_receive(struct mqueue *mq, char *msg_ptr,
size_t msg_len, unsigned *msg_prio, int waitok,
const struct timespec *abs_timeout);
static int _mqueue_send(struct mqueue *mq, struct mqueue_msg *msg,
int timo);
static int _mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg,
int timo);
static void mqueue_send_notification(struct mqueue *mq);
static void mqueue_fdclose(struct thread *td, int fd, struct file *fp);
static void mq_proc_exit(void *arg, struct proc *p);
/*
* kqueue filters
*/
static void filt_mqdetach(struct knote *kn);
static int filt_mqread(struct knote *kn, long hint);
static int filt_mqwrite(struct knote *kn, long hint);
struct filterops mq_rfiltops = {
.f_isfd = 1,
.f_detach = filt_mqdetach,
.f_event = filt_mqread,
};
struct filterops mq_wfiltops = {
.f_isfd = 1,
.f_detach = filt_mqdetach,
.f_event = filt_mqwrite,
};
/*
* Initialize fileno bitmap
*/
static void
mqfs_fileno_init(struct mqfs_info *mi)
{
struct unrhdr *up;
up = new_unrhdr(1, INT_MAX, NULL);
mi->mi_unrhdr = up;
}
/*
* Tear down fileno bitmap
*/
static void
mqfs_fileno_uninit(struct mqfs_info *mi)
{
struct unrhdr *up;
up = mi->mi_unrhdr;
mi->mi_unrhdr = NULL;
delete_unrhdr(up);
}
/*
* Allocate a file number
*/
static void
mqfs_fileno_alloc(struct mqfs_info *mi, struct mqfs_node *mn)
{
/* make sure our parent has a file number */
if (mn->mn_parent && !mn->mn_parent->mn_fileno)
mqfs_fileno_alloc(mi, mn->mn_parent);
switch (mn->mn_type) {
case mqfstype_root:
case mqfstype_dir:
case mqfstype_file:
case mqfstype_symlink:
mn->mn_fileno = alloc_unr(mi->mi_unrhdr);
break;
case mqfstype_this:
KASSERT(mn->mn_parent != NULL,
("mqfstype_this node has no parent"));
mn->mn_fileno = mn->mn_parent->mn_fileno;
break;
case mqfstype_parent:
KASSERT(mn->mn_parent != NULL,
("mqfstype_parent node has no parent"));
if (mn->mn_parent == mi->mi_root) {
mn->mn_fileno = mn->mn_parent->mn_fileno;
break;
}
KASSERT(mn->mn_parent->mn_parent != NULL,
("mqfstype_parent node has no grandparent"));
mn->mn_fileno = mn->mn_parent->mn_parent->mn_fileno;
break;
default:
KASSERT(0,
("mqfs_fileno_alloc() called for unknown type node: %d",
mn->mn_type));
break;
}
}
/*
* Release a file number
*/
static void
mqfs_fileno_free(struct mqfs_info *mi, struct mqfs_node *mn)
{
switch (mn->mn_type) {
case mqfstype_root:
case mqfstype_dir:
case mqfstype_file:
case mqfstype_symlink:
free_unr(mi->mi_unrhdr, mn->mn_fileno);
break;
case mqfstype_this:
case mqfstype_parent:
/* ignore these, as they don't "own" their file number */
break;
default:
KASSERT(0,
("mqfs_fileno_free() called for unknown type node: %d",
mn->mn_type));
break;
}
}
static __inline struct mqfs_node *
mqnode_alloc(void)
{
return uma_zalloc(mqnode_zone, M_WAITOK | M_ZERO);
}
static __inline void
mqnode_free(struct mqfs_node *node)
{
uma_zfree(mqnode_zone, node);
}
static __inline void
mqnode_addref(struct mqfs_node *node)
{
atomic_fetchadd_int(&node->mn_refcount, 1);
}
static __inline void
mqnode_release(struct mqfs_node *node)
{
struct mqfs_info *mqfs;
int old, exp;
mqfs = node->mn_info;
old = atomic_fetchadd_int(&node->mn_refcount, -1);
if (node->mn_type == mqfstype_dir ||
node->mn_type == mqfstype_root)
exp = 3; /* include . and .. */
else
exp = 1;
if (old == exp) {
int locked = sx_xlocked(&mqfs->mi_lock);
if (!locked)
sx_xlock(&mqfs->mi_lock);
mqfs_destroy(node);
if (!locked)
sx_xunlock(&mqfs->mi_lock);
}
}
/*
* Add a node to a directory
*/
static int
mqfs_add_node(struct mqfs_node *parent, struct mqfs_node *node)
{
KASSERT(parent != NULL, ("%s(): parent is NULL", __func__));
KASSERT(parent->mn_info != NULL,
("%s(): parent has no mn_info", __func__));
KASSERT(parent->mn_type == mqfstype_dir ||
parent->mn_type == mqfstype_root,
("%s(): parent is not a directory", __func__));
node->mn_info = parent->mn_info;
node->mn_parent = parent;
LIST_INIT(&node->mn_children);
LIST_INIT(&node->mn_vnodes);
LIST_INSERT_HEAD(&parent->mn_children, node, mn_sibling);
mqnode_addref(parent);
return (0);
}
static struct mqfs_node *
mqfs_create_node(const char *name, int namelen, struct ucred *cred, int mode,
int nodetype)
{
struct mqfs_node *node;
node = mqnode_alloc();
strncpy(node->mn_name, name, namelen);
node->mn_pr_root = cred->cr_prison->pr_root;
node->mn_type = nodetype;
node->mn_refcount = 1;
vfs_timestamp(&node->mn_birth);
node->mn_ctime = node->mn_atime = node->mn_mtime
= node->mn_birth;
node->mn_uid = cred->cr_uid;
node->mn_gid = cred->cr_gid;
node->mn_mode = mode;
return (node);
}
/*
* Create a file
*/
static struct mqfs_node *
mqfs_create_file(struct mqfs_node *parent, const char *name, int namelen,
struct ucred *cred, int mode)
{
struct mqfs_node *node;
node = mqfs_create_node(name, namelen, cred, mode, mqfstype_file);
if (mqfs_add_node(parent, node) != 0) {
mqnode_free(node);
return (NULL);
}
return (node);
}
/*
* Add . and .. to a directory
*/
static int
mqfs_fixup_dir(struct mqfs_node *parent)
{
struct mqfs_node *dir;
dir = mqnode_alloc();
dir->mn_name[0] = '.';
dir->mn_type = mqfstype_this;
dir->mn_refcount = 1;
if (mqfs_add_node(parent, dir) != 0) {
mqnode_free(dir);
return (-1);
}
dir = mqnode_alloc();
dir->mn_name[0] = dir->mn_name[1] = '.';
dir->mn_type = mqfstype_parent;
dir->mn_refcount = 1;
if (mqfs_add_node(parent, dir) != 0) {
mqnode_free(dir);
return (-1);
}
return (0);
}
#ifdef notyet
/*
* Create a directory
*/
static struct mqfs_node *
mqfs_create_dir(struct mqfs_node *parent, const char *name, int namelen,
struct ucred *cred, int mode)
{
struct mqfs_node *node;
node = mqfs_create_node(name, namelen, cred, mode, mqfstype_dir);
if (mqfs_add_node(parent, node) != 0) {
mqnode_free(node);
return (NULL);
}
if (mqfs_fixup_dir(node) != 0) {
mqfs_destroy(node);
return (NULL);
}
return (node);
}
/*
* Create a symlink
*/
static struct mqfs_node *
mqfs_create_link(struct mqfs_node *parent, const char *name, int namelen,
struct ucred *cred, int mode)
{
struct mqfs_node *node;
node = mqfs_create_node(name, namelen, cred, mode, mqfstype_symlink);
if (mqfs_add_node(parent, node) != 0) {
mqnode_free(node);
return (NULL);
}
return (node);
}
#endif
/*
* Destroy a node or a tree of nodes
*/
static int
mqfs_destroy(struct mqfs_node *node)
{
struct mqfs_node *parent;
KASSERT(node != NULL,
("%s(): node is NULL", __func__));
KASSERT(node->mn_info != NULL,
("%s(): node has no mn_info", __func__));
/* destroy children */
if (node->mn_type == mqfstype_dir || node->mn_type == mqfstype_root)
while (! LIST_EMPTY(&node->mn_children))
mqfs_destroy(LIST_FIRST(&node->mn_children));
/* unlink from parent */
if ((parent = node->mn_parent) != NULL) {
KASSERT(parent->mn_info == node->mn_info,
("%s(): parent has different mn_info", __func__));
LIST_REMOVE(node, mn_sibling);
}
if (node->mn_fileno != 0)
mqfs_fileno_free(node->mn_info, node);
if (node->mn_data != NULL)
mqueue_free(node->mn_data);
mqnode_free(node);
return (0);
}
/*
* Mount a mqfs instance
*/
static int
mqfs_mount(struct mount *mp)
{
struct statfs *sbp;
if (mp->mnt_flag & MNT_UPDATE)
return (EOPNOTSUPP);
mp->mnt_data = &mqfs_data;
MNT_ILOCK(mp);
mp->mnt_flag |= MNT_LOCAL;
MNT_IUNLOCK(mp);
vfs_getnewfsid(mp);
sbp = &mp->mnt_stat;
vfs_mountedfrom(mp, "mqueue");
sbp->f_bsize = PAGE_SIZE;
sbp->f_iosize = PAGE_SIZE;
sbp->f_blocks = 1;
sbp->f_bfree = 0;
sbp->f_bavail = 0;
sbp->f_files = 1;
sbp->f_ffree = 0;
return (0);
}
/*
* Unmount a mqfs instance
*/
static int
mqfs_unmount(struct mount *mp, int mntflags)
{
int error;
error = vflush(mp, 0, (mntflags & MNT_FORCE) ? FORCECLOSE : 0,
curthread);
return (error);
}
/*
* Return a root vnode
*/
static int
mqfs_root(struct mount *mp, int flags, struct vnode **vpp)
{
struct mqfs_info *mqfs;
int ret;
mqfs = VFSTOMQFS(mp);
ret = mqfs_allocv(mp, vpp, mqfs->mi_root);
return (ret);
}
/*
* Return filesystem stats
*/
static int
mqfs_statfs(struct mount *mp, struct statfs *sbp)
{
/* XXX update statistics */
return (0);
}
/*
* Initialize a mqfs instance
*/
static int
mqfs_init(struct vfsconf *vfc)
{
struct mqfs_node *root;
struct mqfs_info *mi;
osd_method_t methods[PR_MAXMETHOD] = {
[PR_METHOD_REMOVE] = mqfs_prison_remove,
};
mqnode_zone = uma_zcreate("mqnode", sizeof(struct mqfs_node),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
mqueue_zone = uma_zcreate("mqueue", sizeof(struct mqueue),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
mvdata_zone = uma_zcreate("mvdata",
sizeof(struct mqfs_vdata), NULL, NULL, NULL,
NULL, UMA_ALIGN_PTR, 0);
mqnoti_zone = uma_zcreate("mqnotifier", sizeof(struct mqueue_notifier),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
mi = &mqfs_data;
sx_init(&mi->mi_lock, "mqfs lock");
/* set up the root diretory */
root = mqfs_create_node("/", 1, curthread->td_ucred, 01777,
mqfstype_root);
root->mn_info = mi;
LIST_INIT(&root->mn_children);
LIST_INIT(&root->mn_vnodes);
mi->mi_root = root;
mqfs_fileno_init(mi);
mqfs_fileno_alloc(mi, root);
mqfs_fixup_dir(root);
exit_tag = EVENTHANDLER_REGISTER(process_exit, mq_proc_exit, NULL,
EVENTHANDLER_PRI_ANY);
mq_fdclose = mqueue_fdclose;
p31b_setcfg(CTL_P1003_1B_MESSAGE_PASSING, _POSIX_MESSAGE_PASSING);
mqfs_osd_jail_slot = osd_jail_register(NULL, methods);
return (0);
}
/*
* Destroy a mqfs instance
*/
static int
mqfs_uninit(struct vfsconf *vfc)
{
struct mqfs_info *mi;
if (!unloadable)
return (EOPNOTSUPP);
osd_jail_deregister(mqfs_osd_jail_slot);
EVENTHANDLER_DEREGISTER(process_exit, exit_tag);
mi = &mqfs_data;
mqfs_destroy(mi->mi_root);
mi->mi_root = NULL;
mqfs_fileno_uninit(mi);
sx_destroy(&mi->mi_lock);
uma_zdestroy(mqnode_zone);
uma_zdestroy(mqueue_zone);
uma_zdestroy(mvdata_zone);
uma_zdestroy(mqnoti_zone);
return (0);
}
/*
* task routine
*/
static void
do_recycle(void *context, int pending __unused)
{
struct vnode *vp = (struct vnode *)context;
vrecycle(vp);
vdrop(vp);
}
/*
* Allocate a vnode
*/
static int
mqfs_allocv(struct mount *mp, struct vnode **vpp, struct mqfs_node *pn)
{
struct mqfs_vdata *vd;
struct mqfs_info *mqfs;
struct vnode *newvpp;
int error;
mqfs = pn->mn_info;
*vpp = NULL;
sx_xlock(&mqfs->mi_lock);
LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
if (vd->mv_vnode->v_mount == mp) {
vhold(vd->mv_vnode);
break;
}
}
if (vd != NULL) {
found:
*vpp = vd->mv_vnode;
sx_xunlock(&mqfs->mi_lock);
error = vget(*vpp, LK_RETRY | LK_EXCLUSIVE, curthread);
vdrop(*vpp);
return (error);
}
sx_xunlock(&mqfs->mi_lock);
error = getnewvnode("mqueue", mp, &mqfs_vnodeops, &newvpp);
if (error)
return (error);
vn_lock(newvpp, LK_EXCLUSIVE | LK_RETRY);
error = insmntque(newvpp, mp);
if (error != 0)
return (error);
sx_xlock(&mqfs->mi_lock);
/*
* Check if it has already been allocated
* while we were blocked.
*/
LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
if (vd->mv_vnode->v_mount == mp) {
vhold(vd->mv_vnode);
sx_xunlock(&mqfs->mi_lock);
vgone(newvpp);
vput(newvpp);
goto found;
}
}
*vpp = newvpp;
vd = uma_zalloc(mvdata_zone, M_WAITOK);
(*vpp)->v_data = vd;
vd->mv_vnode = *vpp;
vd->mv_node = pn;
TASK_INIT(&vd->mv_task, 0, do_recycle, *vpp);
LIST_INSERT_HEAD(&pn->mn_vnodes, vd, mv_link);
mqnode_addref(pn);
switch (pn->mn_type) {
case mqfstype_root:
(*vpp)->v_vflag = VV_ROOT;
/* fall through */
case mqfstype_dir:
case mqfstype_this:
case mqfstype_parent:
(*vpp)->v_type = VDIR;
break;
case mqfstype_file:
(*vpp)->v_type = VREG;
break;
case mqfstype_symlink:
(*vpp)->v_type = VLNK;
break;
case mqfstype_none:
KASSERT(0, ("mqfs_allocf called for null node\n"));
default:
panic("%s has unexpected type: %d", pn->mn_name, pn->mn_type);
}
sx_xunlock(&mqfs->mi_lock);
return (0);
}
/*
* Search a directory entry
*/
static struct mqfs_node *
mqfs_search(struct mqfs_node *pd, const char *name, int len, struct ucred *cred)
{
struct mqfs_node *pn;
const void *pr_root;
sx_assert(&pd->mn_info->mi_lock, SX_LOCKED);
pr_root = cred->cr_prison->pr_root;
LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
/* Only match names within the same prison root directory */
if ((pn->mn_pr_root == NULL || pn->mn_pr_root == pr_root) &&
strncmp(pn->mn_name, name, len) == 0 &&
pn->mn_name[len] == '\0')
return (pn);
}
return (NULL);
}
/*
* Look up a file or directory.
*/
static int
mqfs_lookupx(struct vop_cachedlookup_args *ap)
{
struct componentname *cnp;
struct vnode *dvp, **vpp;
struct mqfs_node *pd;
struct mqfs_node *pn;
struct mqfs_info *mqfs;
int nameiop, flags, error, namelen;
char *pname;
struct thread *td;
cnp = ap->a_cnp;
vpp = ap->a_vpp;
dvp = ap->a_dvp;
pname = cnp->cn_nameptr;
namelen = cnp->cn_namelen;
td = cnp->cn_thread;
flags = cnp->cn_flags;
nameiop = cnp->cn_nameiop;
pd = VTON(dvp);
pn = NULL;
mqfs = pd->mn_info;
*vpp = NULLVP;
if (dvp->v_type != VDIR)
return (ENOTDIR);
error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, cnp->cn_thread);
if (error)
return (error);
/* shortcut: check if the name is too long */
if (cnp->cn_namelen >= MQFS_NAMELEN)
return (ENOENT);
/* self */
if (namelen == 1 && pname[0] == '.') {
if ((flags & ISLASTCN) && nameiop != LOOKUP)
return (EINVAL);
pn = pd;
*vpp = dvp;
VREF(dvp);
return (0);
}
/* parent */
if (cnp->cn_flags & ISDOTDOT) {
if (dvp->v_vflag & VV_ROOT)
return (EIO);
if ((flags & ISLASTCN) && nameiop != LOOKUP)
return (EINVAL);
VOP_UNLOCK(dvp, 0);
KASSERT(pd->mn_parent, ("non-root directory has no parent"));
pn = pd->mn_parent;
error = mqfs_allocv(dvp->v_mount, vpp, pn);
vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
return (error);
}
/* named node */
sx_xlock(&mqfs->mi_lock);
pn = mqfs_search(pd, pname, namelen, cnp->cn_cred);
if (pn != NULL)
mqnode_addref(pn);
sx_xunlock(&mqfs->mi_lock);
/* found */
if (pn != NULL) {
/* DELETE */
if (nameiop == DELETE && (flags & ISLASTCN)) {
error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
if (error) {
mqnode_release(pn);
return (error);
}
if (*vpp == dvp) {
VREF(dvp);
*vpp = dvp;
mqnode_release(pn);
return (0);
}
}
/* allocate vnode */
error = mqfs_allocv(dvp->v_mount, vpp, pn);
mqnode_release(pn);
if (error == 0 && cnp->cn_flags & MAKEENTRY)
cache_enter(dvp, *vpp, cnp);
return (error);
}
/* not found */
/* will create a new entry in the directory ? */
if ((nameiop == CREATE || nameiop == RENAME) && (flags & LOCKPARENT)
&& (flags & ISLASTCN)) {
error = VOP_ACCESS(dvp, VWRITE, cnp->cn_cred, td);
if (error)
return (error);
cnp->cn_flags |= SAVENAME;
return (EJUSTRETURN);
}
return (ENOENT);
}
#if 0
struct vop_lookup_args {
struct vop_generic_args a_gen;
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
};
#endif
/*
* vnode lookup operation
*/
static int
mqfs_lookup(struct vop_cachedlookup_args *ap)
{
int rc;
rc = mqfs_lookupx(ap);
return (rc);
}
#if 0
struct vop_create_args {
struct vnode *a_dvp;
struct vnode **a_vpp;
struct componentname *a_cnp;
struct vattr *a_vap;
};
#endif
/*
* vnode creation operation
*/
static int
mqfs_create(struct vop_create_args *ap)
{
struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
struct componentname *cnp = ap->a_cnp;
struct mqfs_node *pd;
struct mqfs_node *pn;
struct mqueue *mq;
int error;
pd = VTON(ap->a_dvp);
if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
return (ENOTDIR);
mq = mqueue_alloc(NULL);
if (mq == NULL)
return (EAGAIN);
sx_xlock(&mqfs->mi_lock);
if ((cnp->cn_flags & HASBUF) == 0)
panic("%s: no name", __func__);
pn = mqfs_create_file(pd, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_cred, ap->a_vap->va_mode);
if (pn == NULL) {
sx_xunlock(&mqfs->mi_lock);
error = ENOSPC;
} else {
mqnode_addref(pn);
sx_xunlock(&mqfs->mi_lock);
error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
mqnode_release(pn);
if (error)
mqfs_destroy(pn);
else
pn->mn_data = mq;
}
if (error)
mqueue_free(mq);
return (error);
}
/*
* Remove an entry
*/
static
int do_unlink(struct mqfs_node *pn, struct ucred *ucred)
{
struct mqfs_node *parent;
struct mqfs_vdata *vd;
int error = 0;
sx_assert(&pn->mn_info->mi_lock, SX_LOCKED);
if (ucred->cr_uid != pn->mn_uid &&
(error = priv_check_cred(ucred, PRIV_MQ_ADMIN, 0)) != 0)
error = EACCES;
else if (!pn->mn_deleted) {
parent = pn->mn_parent;
pn->mn_parent = NULL;
pn->mn_deleted = 1;
LIST_REMOVE(pn, mn_sibling);
LIST_FOREACH(vd, &pn->mn_vnodes, mv_link) {
cache_purge(vd->mv_vnode);
vhold(vd->mv_vnode);
taskqueue_enqueue(taskqueue_thread, &vd->mv_task);
}
mqnode_release(pn);
mqnode_release(parent);
} else
error = ENOENT;
return (error);
}
#if 0
struct vop_remove_args {
struct vnode *a_dvp;
struct vnode *a_vp;
struct componentname *a_cnp;
};
#endif
/*
* vnode removal operation
*/
static int
mqfs_remove(struct vop_remove_args *ap)
{
struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
struct mqfs_node *pn;
int error;
if (ap->a_vp->v_type == VDIR)
return (EPERM);
pn = VTON(ap->a_vp);
sx_xlock(&mqfs->mi_lock);
error = do_unlink(pn, ap->a_cnp->cn_cred);
sx_xunlock(&mqfs->mi_lock);
return (error);
}
#if 0
struct vop_inactive_args {
struct vnode *a_vp;
struct thread *a_td;
};
#endif
static int
mqfs_inactive(struct vop_inactive_args *ap)
{
struct mqfs_node *pn = VTON(ap->a_vp);
if (pn->mn_deleted)
vrecycle(ap->a_vp);
return (0);
}
#if 0
struct vop_reclaim_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
struct thread *a_td;
};
#endif
static int
mqfs_reclaim(struct vop_reclaim_args *ap)
{
struct mqfs_info *mqfs = VFSTOMQFS(ap->a_vp->v_mount);
struct vnode *vp = ap->a_vp;
struct mqfs_node *pn;
struct mqfs_vdata *vd;
vd = vp->v_data;
pn = vd->mv_node;
sx_xlock(&mqfs->mi_lock);
vp->v_data = NULL;
LIST_REMOVE(vd, mv_link);
uma_zfree(mvdata_zone, vd);
mqnode_release(pn);
sx_xunlock(&mqfs->mi_lock);
return (0);
}
#if 0
struct vop_open_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
int a_mode;
struct ucred *a_cred;
struct thread *a_td;
struct file *a_fp;
};
#endif
static int
mqfs_open(struct vop_open_args *ap)
{
return (0);
}
#if 0
struct vop_close_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
int a_fflag;
struct ucred *a_cred;
struct thread *a_td;
};
#endif
static int
mqfs_close(struct vop_close_args *ap)
{
return (0);
}
#if 0
struct vop_access_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
accmode_t a_accmode;
struct ucred *a_cred;
struct thread *a_td;
};
#endif
/*
* Verify permissions
*/
static int
mqfs_access(struct vop_access_args *ap)
{
struct vnode *vp = ap->a_vp;
struct vattr vattr;
int error;
error = VOP_GETATTR(vp, &vattr, ap->a_cred);
if (error)
return (error);
error = vaccess(vp->v_type, vattr.va_mode, vattr.va_uid,
vattr.va_gid, ap->a_accmode, ap->a_cred, NULL);
return (error);
}
#if 0
struct vop_getattr_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
struct vattr *a_vap;
struct ucred *a_cred;
};
#endif
/*
* Get file attributes
*/
static int
mqfs_getattr(struct vop_getattr_args *ap)
{
struct vnode *vp = ap->a_vp;
struct mqfs_node *pn = VTON(vp);
struct vattr *vap = ap->a_vap;
int error = 0;
vap->va_type = vp->v_type;
vap->va_mode = pn->mn_mode;
vap->va_nlink = 1;
vap->va_uid = pn->mn_uid;
vap->va_gid = pn->mn_gid;
vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
vap->va_fileid = pn->mn_fileno;
vap->va_size = 0;
vap->va_blocksize = PAGE_SIZE;
vap->va_bytes = vap->va_size = 0;
vap->va_atime = pn->mn_atime;
vap->va_mtime = pn->mn_mtime;
vap->va_ctime = pn->mn_ctime;
vap->va_birthtime = pn->mn_birth;
vap->va_gen = 0;
vap->va_flags = 0;
vap->va_rdev = NODEV;
vap->va_bytes = 0;
vap->va_filerev = 0;
return (error);
}
#if 0
struct vop_setattr_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
struct vattr *a_vap;
struct ucred *a_cred;
};
#endif
/*
* Set attributes
*/
static int
mqfs_setattr(struct vop_setattr_args *ap)
{
struct mqfs_node *pn;
struct vattr *vap;
struct vnode *vp;
struct thread *td;
int c, error;
uid_t uid;
gid_t gid;
td = curthread;
vap = ap->a_vap;
vp = ap->a_vp;
if ((vap->va_type != VNON) ||
(vap->va_nlink != VNOVAL) ||
(vap->va_fsid != VNOVAL) ||
(vap->va_fileid != VNOVAL) ||
(vap->va_blocksize != VNOVAL) ||
(vap->va_flags != VNOVAL && vap->va_flags != 0) ||
(vap->va_rdev != VNOVAL) ||
((int)vap->va_bytes != VNOVAL) ||
(vap->va_gen != VNOVAL)) {
return (EINVAL);
}
pn = VTON(vp);
error = c = 0;
if (vap->va_uid == (uid_t)VNOVAL)
uid = pn->mn_uid;
else
uid = vap->va_uid;
if (vap->va_gid == (gid_t)VNOVAL)
gid = pn->mn_gid;
else
gid = vap->va_gid;
if (uid != pn->mn_uid || gid != pn->mn_gid) {
/*
* To modify the ownership of a file, must possess VADMIN
* for that file.
*/
if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)))
return (error);
/*
* XXXRW: Why is there a privilege check here: shouldn't the
* check in VOP_ACCESS() be enough? Also, are the group bits
* below definitely right?
*/
if (((ap->a_cred->cr_uid != pn->mn_uid) || uid != pn->mn_uid ||
(gid != pn->mn_gid && !groupmember(gid, ap->a_cred))) &&
(error = priv_check(td, PRIV_MQ_ADMIN)) != 0)
return (error);
pn->mn_uid = uid;
pn->mn_gid = gid;
c = 1;
}
if (vap->va_mode != (mode_t)VNOVAL) {
if ((ap->a_cred->cr_uid != pn->mn_uid) &&
(error = priv_check(td, PRIV_MQ_ADMIN)))
return (error);
pn->mn_mode = vap->va_mode;
c = 1;
}
if (vap->va_atime.tv_sec != VNOVAL || vap->va_mtime.tv_sec != VNOVAL) {
/* See the comment in ufs_vnops::ufs_setattr(). */
if ((error = VOP_ACCESS(vp, VADMIN, ap->a_cred, td)) &&
((vap->va_vaflags & VA_UTIMES_NULL) == 0 ||
(error = VOP_ACCESS(vp, VWRITE, ap->a_cred, td))))
return (error);
if (vap->va_atime.tv_sec != VNOVAL) {
pn->mn_atime = vap->va_atime;
}
if (vap->va_mtime.tv_sec != VNOVAL) {
pn->mn_mtime = vap->va_mtime;
}
c = 1;
}
if (c) {
vfs_timestamp(&pn->mn_ctime);
}
return (0);
}
#if 0
struct vop_read_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
struct uio *a_uio;
int a_ioflag;
struct ucred *a_cred;
};
#endif
/*
* Read from a file
*/
static int
mqfs_read(struct vop_read_args *ap)
{
char buf[80];
struct vnode *vp = ap->a_vp;
struct uio *uio = ap->a_uio;
struct mqfs_node *pn;
struct mqueue *mq;
int len, error;
if (vp->v_type != VREG)
return (EINVAL);
pn = VTON(vp);
mq = VTOMQ(vp);
snprintf(buf, sizeof(buf),
"QSIZE:%-10ld MAXMSG:%-10ld CURMSG:%-10ld MSGSIZE:%-10ld\n",
mq->mq_totalbytes,
mq->mq_maxmsg,
mq->mq_curmsgs,
mq->mq_msgsize);
buf[sizeof(buf)-1] = '\0';
len = strlen(buf);
error = uiomove_frombuf(buf, len, uio);
return (error);
}
#if 0
struct vop_readdir_args {
struct vop_generic_args a_gen;
struct vnode *a_vp;
struct uio *a_uio;
struct ucred *a_cred;
int *a_eofflag;
int *a_ncookies;
u_long **a_cookies;
};
#endif
/*
* Return directory entries.
*/
static int
mqfs_readdir(struct vop_readdir_args *ap)
{
struct vnode *vp;
struct mqfs_info *mi;
struct mqfs_node *pd;
struct mqfs_node *pn;
struct dirent entry;
struct uio *uio;
const void *pr_root;
int *tmp_ncookies = NULL;
off_t offset;
int error, i;
vp = ap->a_vp;
mi = VFSTOMQFS(vp->v_mount);
pd = VTON(vp);
uio = ap->a_uio;
if (vp->v_type != VDIR)
return (ENOTDIR);
if (uio->uio_offset < 0)
return (EINVAL);
if (ap->a_ncookies != NULL) {
tmp_ncookies = ap->a_ncookies;
*ap->a_ncookies = 0;
ap->a_ncookies = NULL;
}
error = 0;
offset = 0;
pr_root = ap->a_cred->cr_prison->pr_root;
sx_xlock(&mi->mi_lock);
LIST_FOREACH(pn, &pd->mn_children, mn_sibling) {
entry.d_reclen = sizeof(entry);
2016-04-14 17:07:26 +00:00
/*
* Only show names within the same prison root directory
* (or not associated with a prison, e.g. "." and "..").
*/
if (pn->mn_pr_root != NULL && pn->mn_pr_root != pr_root)
continue;
if (!pn->mn_fileno)
mqfs_fileno_alloc(mi, pn);
entry.d_fileno = pn->mn_fileno;
for (i = 0; i < MQFS_NAMELEN - 1 && pn->mn_name[i] != '\0'; ++i)
entry.d_name[i] = pn->mn_name[i];
entry.d_name[i] = 0;
entry.d_namlen = i;
switch (pn->mn_type) {
case mqfstype_root:
case mqfstype_dir:
case mqfstype_this:
case mqfstype_parent:
entry.d_type = DT_DIR;
break;
case mqfstype_file:
entry.d_type = DT_REG;
break;
case mqfstype_symlink:
entry.d_type = DT_LNK;
break;
default:
panic("%s has unexpected node type: %d", pn->mn_name,
pn->mn_type);
}
if (entry.d_reclen > uio->uio_resid)
break;
if (offset >= uio->uio_offset) {
error = vfs_read_dirent(ap, &entry, offset);
if (error)
break;
}
offset += entry.d_reclen;
}
sx_xunlock(&mi->mi_lock);
uio->uio_offset = offset;
if (tmp_ncookies != NULL)
ap->a_ncookies = tmp_ncookies;
return (error);
}
#ifdef notyet
#if 0
struct vop_mkdir_args {
struct vnode *a_dvp;
struvt vnode **a_vpp;
struvt componentname *a_cnp;
struct vattr *a_vap;
};
#endif
/*
* Create a directory.
*/
static int
mqfs_mkdir(struct vop_mkdir_args *ap)
{
struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
struct componentname *cnp = ap->a_cnp;
struct mqfs_node *pd = VTON(ap->a_dvp);
struct mqfs_node *pn;
int error;
if (pd->mn_type != mqfstype_root && pd->mn_type != mqfstype_dir)
return (ENOTDIR);
sx_xlock(&mqfs->mi_lock);
if ((cnp->cn_flags & HASBUF) == 0)
panic("%s: no name", __func__);
pn = mqfs_create_dir(pd, cnp->cn_nameptr, cnp->cn_namelen,
ap->a_vap->cn_cred, ap->a_vap->va_mode);
if (pn != NULL)
mqnode_addref(pn);
sx_xunlock(&mqfs->mi_lock);
if (pn == NULL) {
error = ENOSPC;
} else {
error = mqfs_allocv(ap->a_dvp->v_mount, ap->a_vpp, pn);
mqnode_release(pn);
}
return (error);
}
#if 0
struct vop_rmdir_args {
struct vnode *a_dvp;
struct vnode *a_vp;
struct componentname *a_cnp;
};
#endif
/*
* Remove a directory.
*/
static int
mqfs_rmdir(struct vop_rmdir_args *ap)
{
struct mqfs_info *mqfs = VFSTOMQFS(ap->a_dvp->v_mount);
struct mqfs_node *pn = VTON(ap->a_vp);
struct mqfs_node *pt;
if (pn->mn_type != mqfstype_dir)
return (ENOTDIR);
sx_xlock(&mqfs->mi_lock);
if (pn->mn_deleted) {
sx_xunlock(&mqfs->mi_lock);
return (ENOENT);
}
pt = LIST_FIRST(&pn->mn_children);
pt = LIST_NEXT(pt, mn_sibling);
pt = LIST_NEXT(pt, mn_sibling);
if (pt != NULL) {
sx_xunlock(&mqfs->mi_lock);
return (ENOTEMPTY);
}
pt = pn->mn_parent;
pn->mn_parent = NULL;
pn->mn_deleted = 1;
LIST_REMOVE(pn, mn_sibling);
mqnode_release(pn);
mqnode_release(pt);
sx_xunlock(&mqfs->mi_lock);
cache_purge(ap->a_vp);
return (0);
}
#endif /* notyet */
/*
* See if this prison root is obsolete, and clean up associated queues if it is.
*/
static int
mqfs_prison_remove(void *obj, void *data __unused)
{
const struct prison *pr = obj;
const struct prison *tpr;
struct mqfs_node *pn, *tpn;
int found;
found = 0;
TAILQ_FOREACH(tpr, &allprison, pr_list) {
if (tpr->pr_root == pr->pr_root && tpr != pr && tpr->pr_ref > 0)
found = 1;
}
if (!found) {
/*
* No jails are rooted in this directory anymore,
* so no queues should be either.
*/
sx_xlock(&mqfs_data.mi_lock);
LIST_FOREACH_SAFE(pn, &mqfs_data.mi_root->mn_children,
mn_sibling, tpn) {
if (pn->mn_pr_root == pr->pr_root)
(void)do_unlink(pn, curthread->td_ucred);
}
sx_xunlock(&mqfs_data.mi_lock);
}
return (0);
}
/*
* Allocate a message queue
*/
static struct mqueue *
mqueue_alloc(const struct mq_attr *attr)
{
struct mqueue *mq;
if (curmq >= maxmq)
return (NULL);
mq = uma_zalloc(mqueue_zone, M_WAITOK | M_ZERO);
TAILQ_INIT(&mq->mq_msgq);
if (attr != NULL) {
mq->mq_maxmsg = attr->mq_maxmsg;
mq->mq_msgsize = attr->mq_msgsize;
} else {
mq->mq_maxmsg = default_maxmsg;
mq->mq_msgsize = default_msgsize;
}
mtx_init(&mq->mq_mutex, "mqueue lock", NULL, MTX_DEF);
knlist_init_mtx(&mq->mq_rsel.si_note, &mq->mq_mutex);
knlist_init_mtx(&mq->mq_wsel.si_note, &mq->mq_mutex);
atomic_add_int(&curmq, 1);
return (mq);
}
/*
* Destroy a message queue
*/
static void
mqueue_free(struct mqueue *mq)
{
struct mqueue_msg *msg;
while ((msg = TAILQ_FIRST(&mq->mq_msgq)) != NULL) {
TAILQ_REMOVE(&mq->mq_msgq, msg, msg_link);
free(msg, M_MQUEUEDATA);
}
mtx_destroy(&mq->mq_mutex);
seldrain(&mq->mq_rsel);
seldrain(&mq->mq_wsel);
knlist_destroy(&mq->mq_rsel.si_note);
knlist_destroy(&mq->mq_wsel.si_note);
uma_zfree(mqueue_zone, mq);
atomic_add_int(&curmq, -1);
}
/*
* Load a message from user space
*/
static struct mqueue_msg *
mqueue_loadmsg(const char *msg_ptr, size_t msg_size, int msg_prio)
{
struct mqueue_msg *msg;
size_t len;
int error;
len = sizeof(struct mqueue_msg) + msg_size;
msg = malloc(len, M_MQUEUEDATA, M_WAITOK);
error = copyin(msg_ptr, ((char *)msg) + sizeof(struct mqueue_msg),
msg_size);
if (error) {
free(msg, M_MQUEUEDATA);
msg = NULL;
} else {
msg->msg_size = msg_size;
msg->msg_prio = msg_prio;
}
return (msg);
}
/*
* Save a message to user space
*/
static int
mqueue_savemsg(struct mqueue_msg *msg, char *msg_ptr, int *msg_prio)
{
int error;
error = copyout(((char *)msg) + sizeof(*msg), msg_ptr,
msg->msg_size);
if (error == 0 && msg_prio != NULL)
error = copyout(&msg->msg_prio, msg_prio, sizeof(int));
return (error);
}
/*
* Free a message's memory
*/
static __inline void
mqueue_freemsg(struct mqueue_msg *msg)
{
free(msg, M_MQUEUEDATA);
}
/*
* Send a message. if waitok is false, thread will not be
* blocked if there is no data in queue, otherwise, absolute
* time will be checked.
*/
int
mqueue_send(struct mqueue *mq, const char *msg_ptr,
size_t msg_len, unsigned msg_prio, int waitok,
const struct timespec *abs_timeout)
{
struct mqueue_msg *msg;
struct timespec ts, ts2;
struct timeval tv;
int error;
if (msg_prio >= MQ_PRIO_MAX)
return (EINVAL);
if (msg_len > mq->mq_msgsize)
return (EMSGSIZE);
msg = mqueue_loadmsg(msg_ptr, msg_len, msg_prio);
if (msg == NULL)
return (EFAULT);
/* O_NONBLOCK case */
if (!waitok) {
error = _mqueue_send(mq, msg, -1);
if (error)
goto bad;
return (0);
}
/* we allow a null timeout (wait forever) */
if (abs_timeout == NULL) {
error = _mqueue_send(mq, msg, 0);
if (error)
goto bad;
return (0);
}
/* send it before checking time */
error = _mqueue_send(mq, msg, -1);
if (error == 0)
return (0);
if (error != EAGAIN)
goto bad;
if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
error = EINVAL;
goto bad;
}
for (;;) {
ts2 = *abs_timeout;
getnanotime(&ts);
timespecsub(&ts2, &ts);
if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
error = ETIMEDOUT;
break;
}
TIMESPEC_TO_TIMEVAL(&tv, &ts2);
error = _mqueue_send(mq, msg, tvtohz(&tv));
if (error != ETIMEDOUT)
break;
}
if (error == 0)
return (0);
bad:
mqueue_freemsg(msg);
return (error);
}
/*
* Common routine to send a message
*/
static int
_mqueue_send(struct mqueue *mq, struct mqueue_msg *msg, int timo)
{
struct mqueue_msg *msg2;
int error = 0;
mtx_lock(&mq->mq_mutex);
while (mq->mq_curmsgs >= mq->mq_maxmsg && error == 0) {
if (timo < 0) {
mtx_unlock(&mq->mq_mutex);
return (EAGAIN);
}
mq->mq_senders++;
error = msleep(&mq->mq_senders, &mq->mq_mutex,
PCATCH, "mqsend", timo);
mq->mq_senders--;
if (error == EAGAIN)
error = ETIMEDOUT;
}
if (mq->mq_curmsgs >= mq->mq_maxmsg) {
mtx_unlock(&mq->mq_mutex);
return (error);
}
error = 0;
if (TAILQ_EMPTY(&mq->mq_msgq)) {
TAILQ_INSERT_HEAD(&mq->mq_msgq, msg, msg_link);
} else {
if (msg->msg_prio <= TAILQ_LAST(&mq->mq_msgq, msgq)->msg_prio) {
TAILQ_INSERT_TAIL(&mq->mq_msgq, msg, msg_link);
} else {
TAILQ_FOREACH(msg2, &mq->mq_msgq, msg_link) {
if (msg2->msg_prio < msg->msg_prio)
break;
}
TAILQ_INSERT_BEFORE(msg2, msg, msg_link);
}
}
mq->mq_curmsgs++;
mq->mq_totalbytes += msg->msg_size;
if (mq->mq_receivers)
wakeup_one(&mq->mq_receivers);
else if (mq->mq_notifier != NULL)
mqueue_send_notification(mq);
if (mq->mq_flags & MQ_RSEL) {
mq->mq_flags &= ~MQ_RSEL;
selwakeup(&mq->mq_rsel);
}
KNOTE_LOCKED(&mq->mq_rsel.si_note, 0);
mtx_unlock(&mq->mq_mutex);
return (0);
}
/*
* Send realtime a signal to process which registered itself
* successfully by mq_notify.
*/
static void
mqueue_send_notification(struct mqueue *mq)
{
struct mqueue_notifier *nt;
struct thread *td;
struct proc *p;
int error;
mtx_assert(&mq->mq_mutex, MA_OWNED);
nt = mq->mq_notifier;
if (nt->nt_sigev.sigev_notify != SIGEV_NONE) {
p = nt->nt_proc;
error = sigev_findtd(p, &nt->nt_sigev, &td);
if (error) {
mq->mq_notifier = NULL;
return;
}
if (!KSI_ONQ(&nt->nt_ksi)) {
ksiginfo_set_sigev(&nt->nt_ksi, &nt->nt_sigev);
tdsendsignal(p, td, nt->nt_ksi.ksi_signo, &nt->nt_ksi);
}
PROC_UNLOCK(p);
}
mq->mq_notifier = NULL;
}
/*
* Get a message. if waitok is false, thread will not be
* blocked if there is no data in queue, otherwise, absolute
* time will be checked.
*/
int
mqueue_receive(struct mqueue *mq, char *msg_ptr,
size_t msg_len, unsigned *msg_prio, int waitok,
const struct timespec *abs_timeout)
{
struct mqueue_msg *msg;
struct timespec ts, ts2;
struct timeval tv;
int error;
if (msg_len < mq->mq_msgsize)
return (EMSGSIZE);
/* O_NONBLOCK case */
if (!waitok) {
error = _mqueue_recv(mq, &msg, -1);
if (error)
return (error);
goto received;
}
/* we allow a null timeout (wait forever). */
if (abs_timeout == NULL) {
error = _mqueue_recv(mq, &msg, 0);
if (error)
return (error);
goto received;
}
/* try to get a message before checking time */
error = _mqueue_recv(mq, &msg, -1);
if (error == 0)
goto received;
if (error != EAGAIN)
return (error);
if (abs_timeout->tv_nsec >= 1000000000 || abs_timeout->tv_nsec < 0) {
error = EINVAL;
return (error);
}
for (;;) {
ts2 = *abs_timeout;
getnanotime(&ts);
timespecsub(&ts2, &ts);
if (ts2.tv_sec < 0 || (ts2.tv_sec == 0 && ts2.tv_nsec <= 0)) {
error = ETIMEDOUT;
return (error);
}
TIMESPEC_TO_TIMEVAL(&tv, &ts2);
error = _mqueue_recv(mq, &msg, tvtohz(&tv));
if (error == 0)
break;
if (error != ETIMEDOUT)
return (error);
}
received:
error = mqueue_savemsg(msg, msg_ptr, msg_prio);
if (error == 0) {
curthread->td_retval[0] = msg->msg_size;
curthread->td_retval[1] = 0;
}
mqueue_freemsg(msg);
return (error);
}
/*
* Common routine to receive a message
*/
static int
_mqueue_recv(struct mqueue *mq, struct mqueue_msg **msg, int timo)
{
int error = 0;
mtx_lock(&mq->mq_mutex);
while ((*msg = TAILQ_FIRST(&mq->mq_msgq)) == NULL && error == 0) {
if (timo < 0) {
mtx_unlock(&mq->mq_mutex);
return (EAGAIN);
}
mq->mq_receivers++;
error = msleep(&mq->mq_receivers, &mq->mq_mutex,
PCATCH, "mqrecv", timo);
mq->mq_receivers--;
if (error == EAGAIN)
error = ETIMEDOUT;
}
if (*msg != NULL) {
error = 0;
TAILQ_REMOVE(&mq->mq_msgq, *msg, msg_link);
mq->mq_curmsgs--;
mq->mq_totalbytes -= (*msg)->msg_size;
if (mq->mq_senders)
wakeup_one(&mq->mq_senders);
if (mq->mq_flags & MQ_WSEL) {
mq->mq_flags &= ~MQ_WSEL;
selwakeup(&mq->mq_wsel);
}
KNOTE_LOCKED(&mq->mq_wsel.si_note, 0);
}
if (mq->mq_notifier != NULL && mq->mq_receivers == 0 &&
!TAILQ_EMPTY(&mq->mq_msgq)) {
mqueue_send_notification(mq);
}
mtx_unlock(&mq->mq_mutex);
return (error);
}
static __inline struct mqueue_notifier *
notifier_alloc(void)
{
return (uma_zalloc(mqnoti_zone, M_WAITOK | M_ZERO));
}
static __inline void
notifier_free(struct mqueue_notifier *p)
{
uma_zfree(mqnoti_zone, p);
}
static struct mqueue_notifier *
notifier_search(struct proc *p, int fd)
{
struct mqueue_notifier *nt;
LIST_FOREACH(nt, &p->p_mqnotifier, nt_link) {
if (nt->nt_ksi.ksi_mqd == fd)
break;
}
return (nt);
}
static __inline void
notifier_insert(struct proc *p, struct mqueue_notifier *nt)
{
LIST_INSERT_HEAD(&p->p_mqnotifier, nt, nt_link);
}
static __inline void
notifier_delete(struct proc *p, struct mqueue_notifier *nt)
{
LIST_REMOVE(nt, nt_link);
notifier_free(nt);
}
static void
notifier_remove(struct proc *p, struct mqueue *mq, int fd)
{
struct mqueue_notifier *nt;
mtx_assert(&mq->mq_mutex, MA_OWNED);
PROC_LOCK(p);
nt = notifier_search(p, fd);
if (nt != NULL) {
if (mq->mq_notifier == nt)
mq->mq_notifier = NULL;
sigqueue_take(&nt->nt_ksi);
notifier_delete(p, nt);
}
PROC_UNLOCK(p);
}
static int
kern_kmq_open(struct thread *td, const char *upath, int flags, mode_t mode,
const struct mq_attr *attr)
{
char path[MQFS_NAMELEN + 1];
struct mqfs_node *pn;
struct filedesc *fdp;
struct file *fp;
struct mqueue *mq;
int fd, error, len, cmode;
fdp = td->td_proc->p_fd;
cmode = (((mode & ~fdp->fd_cmask) & ALLPERMS) & ~S_ISTXT);
mq = NULL;
if ((flags & O_CREAT) != 0 && attr != NULL) {
if (attr->mq_maxmsg <= 0 || attr->mq_maxmsg > maxmsg)
return (EINVAL);
if (attr->mq_msgsize <= 0 || attr->mq_msgsize > maxmsgsize)
return (EINVAL);
}
error = copyinstr(upath, path, MQFS_NAMELEN + 1, NULL);
if (error)
return (error);
/*
* The first character of name must be a slash (/) character
* and the remaining characters of name cannot include any slash
* characters.
*/
len = strlen(path);
if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
return (EINVAL);
error = falloc(td, &fp, &fd, O_CLOEXEC);
if (error)
return (error);
sx_xlock(&mqfs_data.mi_lock);
pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
if (pn == NULL) {
if (!(flags & O_CREAT)) {
error = ENOENT;
} else {
mq = mqueue_alloc(attr);
if (mq == NULL) {
error = ENFILE;
} else {
pn = mqfs_create_file(mqfs_data.mi_root,
path + 1, len - 1, td->td_ucred,
cmode);
if (pn == NULL) {
error = ENOSPC;
mqueue_free(mq);
}
}
}
if (error == 0) {
pn->mn_data = mq;
}
} else {
if ((flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL)) {
error = EEXIST;
} else {
accmode_t accmode = 0;
if (flags & FREAD)
accmode |= VREAD;
if (flags & FWRITE)
accmode |= VWRITE;
error = vaccess(VREG, pn->mn_mode, pn->mn_uid,
pn->mn_gid, accmode, td->td_ucred, NULL);
}
}
if (error) {
sx_xunlock(&mqfs_data.mi_lock);
fdclose(td, fp, fd);
fdrop(fp, td);
return (error);
}
mqnode_addref(pn);
sx_xunlock(&mqfs_data.mi_lock);
finit(fp, flags & (FREAD | FWRITE | O_NONBLOCK), DTYPE_MQUEUE, pn,
&mqueueops);
td->td_retval[0] = fd;
fdrop(fp, td);
return (0);
}
/*
* Syscall to open a message queue.
*/
int
sys_kmq_open(struct thread *td, struct kmq_open_args *uap)
{
struct mq_attr attr;
int flags, error;
if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
return (EINVAL);
flags = FFLAGS(uap->flags);
if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
error = copyin(uap->attr, &attr, sizeof(attr));
if (error)
return (error);
}
return (kern_kmq_open(td, uap->path, flags, uap->mode,
uap->attr != NULL ? &attr : NULL));
}
/*
* Syscall to unlink a message queue.
*/
int
sys_kmq_unlink(struct thread *td, struct kmq_unlink_args *uap)
{
char path[MQFS_NAMELEN+1];
struct mqfs_node *pn;
int error, len;
error = copyinstr(uap->path, path, MQFS_NAMELEN + 1, NULL);
if (error)
return (error);
len = strlen(path);
if (len < 2 || path[0] != '/' || strchr(path + 1, '/') != NULL)
return (EINVAL);
sx_xlock(&mqfs_data.mi_lock);
pn = mqfs_search(mqfs_data.mi_root, path + 1, len - 1, td->td_ucred);
if (pn != NULL)
error = do_unlink(pn, td->td_ucred);
else
error = ENOENT;
sx_xunlock(&mqfs_data.mi_lock);
return (error);
}
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
typedef int (*_fgetf)(struct thread *, int, cap_rights_t *, struct file **);
/*
* Get message queue by giving file slot
*/
static int
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
_getmq(struct thread *td, int fd, cap_rights_t *rightsp, _fgetf func,
struct file **fpp, struct mqfs_node **ppn, struct mqueue **pmq)
{
struct mqfs_node *pn;
int error;
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
error = func(td, fd, rightsp, fpp);
if (error)
return (error);
if (&mqueueops != (*fpp)->f_ops) {
fdrop(*fpp, td);
return (EBADF);
}
pn = (*fpp)->f_data;
if (ppn)
*ppn = pn;
if (pmq)
*pmq = pn->mn_data;
return (0);
}
static __inline int
getmq(struct thread *td, int fd, struct file **fpp, struct mqfs_node **ppn,
struct mqueue **pmq)
{
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
return _getmq(td, fd, cap_rights_init(&rights, CAP_EVENT), fget,
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
fpp, ppn, pmq);
}
static __inline int
getmq_read(struct thread *td, int fd, struct file **fpp,
struct mqfs_node **ppn, struct mqueue **pmq)
{
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
return _getmq(td, fd, cap_rights_init(&rights, CAP_READ), fget_read,
fpp, ppn, pmq);
}
static __inline int
getmq_write(struct thread *td, int fd, struct file **fpp,
struct mqfs_node **ppn, struct mqueue **pmq)
{
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
return _getmq(td, fd, cap_rights_init(&rights, CAP_WRITE), fget_write,
fpp, ppn, pmq);
}
static int
kern_kmq_setattr(struct thread *td, int mqd, const struct mq_attr *attr,
struct mq_attr *oattr)
{
struct mqueue *mq;
struct file *fp;
u_int oflag, flag;
int error;
if (attr != NULL && (attr->mq_flags & ~O_NONBLOCK) != 0)
return (EINVAL);
error = getmq(td, mqd, &fp, NULL, &mq);
if (error)
return (error);
oattr->mq_maxmsg = mq->mq_maxmsg;
oattr->mq_msgsize = mq->mq_msgsize;
oattr->mq_curmsgs = mq->mq_curmsgs;
if (attr != NULL) {
do {
oflag = flag = fp->f_flag;
flag &= ~O_NONBLOCK;
flag |= (attr->mq_flags & O_NONBLOCK);
} while (atomic_cmpset_int(&fp->f_flag, oflag, flag) == 0);
} else
oflag = fp->f_flag;
oattr->mq_flags = (O_NONBLOCK & oflag);
fdrop(fp, td);
return (error);
}
int
sys_kmq_setattr(struct thread *td, struct kmq_setattr_args *uap)
{
struct mq_attr attr, oattr;
int error;
if (uap->attr != NULL) {
error = copyin(uap->attr, &attr, sizeof(attr));
if (error != 0)
return (error);
}
error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
&oattr);
if (error == 0 && uap->oattr != NULL) {
bzero(oattr.__reserved, sizeof(oattr.__reserved));
error = copyout(&oattr, uap->oattr, sizeof(oattr));
}
return (error);
}
int
sys_kmq_timedreceive(struct thread *td, struct kmq_timedreceive_args *uap)
{
struct mqueue *mq;
struct file *fp;
struct timespec *abs_timeout, ets;
int error;
int waitok;
error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
if (error)
return (error);
if (uap->abs_timeout != NULL) {
error = copyin(uap->abs_timeout, &ets, sizeof(ets));
if (error != 0)
return (error);
abs_timeout = &ets;
} else
abs_timeout = NULL;
waitok = !(fp->f_flag & O_NONBLOCK);
error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
uap->msg_prio, waitok, abs_timeout);
fdrop(fp, td);
return (error);
}
int
sys_kmq_timedsend(struct thread *td, struct kmq_timedsend_args *uap)
{
struct mqueue *mq;
struct file *fp;
struct timespec *abs_timeout, ets;
int error, waitok;
error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
if (error)
return (error);
if (uap->abs_timeout != NULL) {
error = copyin(uap->abs_timeout, &ets, sizeof(ets));
if (error != 0)
return (error);
abs_timeout = &ets;
} else
abs_timeout = NULL;
waitok = !(fp->f_flag & O_NONBLOCK);
error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
uap->msg_prio, waitok, abs_timeout);
fdrop(fp, td);
return (error);
}
static int
kern_kmq_notify(struct thread *td, int mqd, struct sigevent *sigev)
{
2013-09-05 10:24:09 +00:00
#ifdef CAPABILITIES
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
cap_rights_t rights;
2013-09-05 10:24:09 +00:00
#endif
struct filedesc *fdp;
struct proc *p;
struct mqueue *mq;
struct file *fp, *fp2;
struct mqueue_notifier *nt, *newnt = NULL;
int error;
if (sigev != NULL) {
if (sigev->sigev_notify != SIGEV_SIGNAL &&
sigev->sigev_notify != SIGEV_THREAD_ID &&
sigev->sigev_notify != SIGEV_NONE)
return (EINVAL);
if ((sigev->sigev_notify == SIGEV_SIGNAL ||
sigev->sigev_notify == SIGEV_THREAD_ID) &&
!_SIG_VALID(sigev->sigev_signo))
return (EINVAL);
}
p = td->td_proc;
fdp = td->td_proc->p_fd;
error = getmq(td, mqd, &fp, NULL, &mq);
if (error)
return (error);
again:
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SLOCK(fdp);
fp2 = fget_locked(fdp, mqd);
if (fp2 == NULL) {
FILEDESC_SUNLOCK(fdp);
error = EBADF;
goto out;
}
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
#ifdef CAPABILITIES
Change the cap_rights_t type from uint64_t to a structure that we can extend in the future in a backward compatible (API and ABI) way. The cap_rights_t represents capability rights. We used to use one bit to represent one right, but we are running out of spare bits. Currently the new structure provides place for 114 rights (so 50 more than the previous cap_rights_t), but it is possible to grow the structure to hold at least 285 rights, although we can make it even larger if 285 rights won't be enough. The structure definition looks like this: struct cap_rights { uint64_t cr_rights[CAP_RIGHTS_VERSION + 2]; }; The initial CAP_RIGHTS_VERSION is 0. The top two bits in the first element of the cr_rights[] array contain total number of elements in the array - 2. This means if those two bits are equal to 0, we have 2 array elements. The top two bits in all remaining array elements should be 0. The next five bits in all array elements contain array index. Only one bit is used and bit position in this five-bits range defines array index. This means there can be at most five array elements in the future. To define new right the CAPRIGHT() macro must be used. The macro takes two arguments - an array index and a bit to set, eg. #define CAP_PDKILL CAPRIGHT(1, 0x0000000000000800ULL) We still support aliases that combine few rights, but the rights have to belong to the same array element, eg: #define CAP_LOOKUP CAPRIGHT(0, 0x0000000000000400ULL) #define CAP_FCHMOD CAPRIGHT(0, 0x0000000000002000ULL) #define CAP_FCHMODAT (CAP_FCHMOD | CAP_LOOKUP) There is new API to manage the new cap_rights_t structure: cap_rights_t *cap_rights_init(cap_rights_t *rights, ...); void cap_rights_set(cap_rights_t *rights, ...); void cap_rights_clear(cap_rights_t *rights, ...); bool cap_rights_is_set(const cap_rights_t *rights, ...); bool cap_rights_is_valid(const cap_rights_t *rights); void cap_rights_merge(cap_rights_t *dst, const cap_rights_t *src); void cap_rights_remove(cap_rights_t *dst, const cap_rights_t *src); bool cap_rights_contains(const cap_rights_t *big, const cap_rights_t *little); Capability rights to the cap_rights_init(), cap_rights_set(), cap_rights_clear() and cap_rights_is_set() functions are provided by separating them with commas, eg: cap_rights_t rights; cap_rights_init(&rights, CAP_READ, CAP_WRITE, CAP_FSTAT); There is no need to terminate the list of rights, as those functions are actually macros that take care of the termination, eg: #define cap_rights_set(rights, ...) \ __cap_rights_set((rights), __VA_ARGS__, 0ULL) void __cap_rights_set(cap_rights_t *rights, ...); Thanks to using one bit as an array index we can assert in those functions that there are no two rights belonging to different array elements provided together. For example this is illegal and will be detected, because CAP_LOOKUP belongs to element 0 and CAP_PDKILL to element 1: cap_rights_init(&rights, CAP_LOOKUP | CAP_PDKILL); Providing several rights that belongs to the same array's element this way is correct, but is not advised. It should only be used for aliases definition. This commit also breaks compatibility with some existing Capsicum system calls, but I see no other way to do that. This should be fine as Capsicum is still experimental and this change is not going to 9.x. Sponsored by: The FreeBSD Foundation
2013-09-05 00:09:56 +00:00
error = cap_check(cap_rights(fdp, mqd),
cap_rights_init(&rights, CAP_EVENT));
if (error) {
FILEDESC_SUNLOCK(fdp);
goto out;
}
Merge Capsicum overhaul: - Capability is no longer separate descriptor type. Now every descriptor has set of its own capability rights. - The cap_new(2) system call is left, but it is no longer documented and should not be used in new code. - The new syscall cap_rights_limit(2) should be used instead of cap_new(2), which limits capability rights of the given descriptor without creating a new one. - The cap_getrights(2) syscall is renamed to cap_rights_get(2). - If CAP_IOCTL capability right is present we can further reduce allowed ioctls list with the new cap_ioctls_limit(2) syscall. List of allowed ioctls can be retrived with cap_ioctls_get(2) syscall. - If CAP_FCNTL capability right is present we can further reduce fcntls that can be used with the new cap_fcntls_limit(2) syscall and retrive them with cap_fcntls_get(2). - To support ioctl and fcntl white-listing the filedesc structure was heavly modified. - The audit subsystem, kdump and procstat tools were updated to recognize new syscalls. - Capability rights were revised and eventhough I tried hard to provide backward API and ABI compatibility there are some incompatible changes that are described in detail below: CAP_CREATE old behaviour: - Allow for openat(2)+O_CREAT. - Allow for linkat(2). - Allow for symlinkat(2). CAP_CREATE new behaviour: - Allow for openat(2)+O_CREAT. Added CAP_LINKAT: - Allow for linkat(2). ABI: Reuses CAP_RMDIR bit. - Allow to be target for renameat(2). Added CAP_SYMLINKAT: - Allow for symlinkat(2). Removed CAP_DELETE. Old behaviour: - Allow for unlinkat(2) when removing non-directory object. - Allow to be source for renameat(2). Removed CAP_RMDIR. Old behaviour: - Allow for unlinkat(2) when removing directory. Added CAP_RENAMEAT: - Required for source directory for the renameat(2) syscall. Added CAP_UNLINKAT (effectively it replaces CAP_DELETE and CAP_RMDIR): - Allow for unlinkat(2) on any object. - Required if target of renameat(2) exists and will be removed by this call. Removed CAP_MAPEXEC. CAP_MMAP old behaviour: - Allow for mmap(2) with any combination of PROT_NONE, PROT_READ and PROT_WRITE. CAP_MMAP new behaviour: - Allow for mmap(2)+PROT_NONE. Added CAP_MMAP_R: - Allow for mmap(PROT_READ). Added CAP_MMAP_W: - Allow for mmap(PROT_WRITE). Added CAP_MMAP_X: - Allow for mmap(PROT_EXEC). Added CAP_MMAP_RW: - Allow for mmap(PROT_READ | PROT_WRITE). Added CAP_MMAP_RX: - Allow for mmap(PROT_READ | PROT_EXEC). Added CAP_MMAP_WX: - Allow for mmap(PROT_WRITE | PROT_EXEC). Added CAP_MMAP_RWX: - Allow for mmap(PROT_READ | PROT_WRITE | PROT_EXEC). Renamed CAP_MKDIR to CAP_MKDIRAT. Renamed CAP_MKFIFO to CAP_MKFIFOAT. Renamed CAP_MKNODE to CAP_MKNODEAT. CAP_READ old behaviour: - Allow pread(2). - Disallow read(2), readv(2) (if there is no CAP_SEEK). CAP_READ new behaviour: - Allow read(2), readv(2). - Disallow pread(2) (CAP_SEEK was also required). CAP_WRITE old behaviour: - Allow pwrite(2). - Disallow write(2), writev(2) (if there is no CAP_SEEK). CAP_WRITE new behaviour: - Allow write(2), writev(2). - Disallow pwrite(2) (CAP_SEEK was also required). Added convinient defines: #define CAP_PREAD (CAP_SEEK | CAP_READ) #define CAP_PWRITE (CAP_SEEK | CAP_WRITE) #define CAP_MMAP_R (CAP_MMAP | CAP_SEEK | CAP_READ) #define CAP_MMAP_W (CAP_MMAP | CAP_SEEK | CAP_WRITE) #define CAP_MMAP_X (CAP_MMAP | CAP_SEEK | 0x0000000000000008ULL) #define CAP_MMAP_RW (CAP_MMAP_R | CAP_MMAP_W) #define CAP_MMAP_RX (CAP_MMAP_R | CAP_MMAP_X) #define CAP_MMAP_WX (CAP_MMAP_W | CAP_MMAP_X) #define CAP_MMAP_RWX (CAP_MMAP_R | CAP_MMAP_W | CAP_MMAP_X) #define CAP_RECV CAP_READ #define CAP_SEND CAP_WRITE #define CAP_SOCK_CLIENT \ (CAP_CONNECT | CAP_GETPEERNAME | CAP_GETSOCKNAME | CAP_GETSOCKOPT | \ CAP_PEELOFF | CAP_RECV | CAP_SEND | CAP_SETSOCKOPT | CAP_SHUTDOWN) #define CAP_SOCK_SERVER \ (CAP_ACCEPT | CAP_BIND | CAP_GETPEERNAME | CAP_GETSOCKNAME | \ CAP_GETSOCKOPT | CAP_LISTEN | CAP_PEELOFF | CAP_RECV | CAP_SEND | \ CAP_SETSOCKOPT | CAP_SHUTDOWN) Added defines for backward API compatibility: #define CAP_MAPEXEC CAP_MMAP_X #define CAP_DELETE CAP_UNLINKAT #define CAP_MKDIR CAP_MKDIRAT #define CAP_RMDIR CAP_UNLINKAT #define CAP_MKFIFO CAP_MKFIFOAT #define CAP_MKNOD CAP_MKNODAT #define CAP_SOCK_ALL (CAP_SOCK_CLIENT | CAP_SOCK_SERVER) Sponsored by: The FreeBSD Foundation Reviewed by: Christoph Mallon <christoph.mallon@gmx.de> Many aspects discussed with: rwatson, benl, jonathan ABI compatibility discussed with: kib
2013-03-02 00:53:12 +00:00
#endif
if (fp2 != fp) {
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SUNLOCK(fdp);
error = EBADF;
goto out;
}
mtx_lock(&mq->mq_mutex);
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SUNLOCK(fdp);
if (sigev != NULL) {
if (mq->mq_notifier != NULL) {
error = EBUSY;
} else {
PROC_LOCK(p);
nt = notifier_search(p, mqd);
if (nt == NULL) {
if (newnt == NULL) {
PROC_UNLOCK(p);
mtx_unlock(&mq->mq_mutex);
newnt = notifier_alloc();
goto again;
}
}
if (nt != NULL) {
sigqueue_take(&nt->nt_ksi);
if (newnt != NULL) {
notifier_free(newnt);
newnt = NULL;
}
} else {
nt = newnt;
newnt = NULL;
ksiginfo_init(&nt->nt_ksi);
nt->nt_ksi.ksi_flags |= KSI_INS | KSI_EXT;
nt->nt_ksi.ksi_code = SI_MESGQ;
nt->nt_proc = p;
nt->nt_ksi.ksi_mqd = mqd;
notifier_insert(p, nt);
}
nt->nt_sigev = *sigev;
mq->mq_notifier = nt;
PROC_UNLOCK(p);
/*
* if there is no receivers and message queue
* is not empty, we should send notification
* as soon as possible.
*/
if (mq->mq_receivers == 0 &&
!TAILQ_EMPTY(&mq->mq_msgq))
mqueue_send_notification(mq);
}
} else {
notifier_remove(p, mq, mqd);
}
mtx_unlock(&mq->mq_mutex);
out:
fdrop(fp, td);
if (newnt != NULL)
notifier_free(newnt);
return (error);
}
int
sys_kmq_notify(struct thread *td, struct kmq_notify_args *uap)
{
struct sigevent ev, *evp;
int error;
if (uap->sigev == NULL) {
evp = NULL;
} else {
error = copyin(uap->sigev, &ev, sizeof(ev));
if (error != 0)
return (error);
evp = &ev;
}
return (kern_kmq_notify(td, uap->mqd, evp));
}
static void
mqueue_fdclose(struct thread *td, int fd, struct file *fp)
{
struct filedesc *fdp;
struct mqueue *mq;
fdp = td->td_proc->p_fd;
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_LOCK_ASSERT(fdp);
if (fp->f_ops == &mqueueops) {
mq = FPTOMQ(fp);
mtx_lock(&mq->mq_mutex);
notifier_remove(td->td_proc, mq, fd);
/* have to wakeup thread in same process */
if (mq->mq_flags & MQ_RSEL) {
mq->mq_flags &= ~MQ_RSEL;
selwakeup(&mq->mq_rsel);
}
if (mq->mq_flags & MQ_WSEL) {
mq->mq_flags &= ~MQ_WSEL;
selwakeup(&mq->mq_wsel);
}
mtx_unlock(&mq->mq_mutex);
}
}
static void
mq_proc_exit(void *arg __unused, struct proc *p)
{
struct filedesc *fdp;
struct file *fp;
struct mqueue *mq;
int i;
fdp = p->p_fd;
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SLOCK(fdp);
for (i = 0; i < fdp->fd_nfiles; ++i) {
fp = fget_locked(fdp, i);
if (fp != NULL && fp->f_ops == &mqueueops) {
mq = FPTOMQ(fp);
mtx_lock(&mq->mq_mutex);
notifier_remove(p, FPTOMQ(fp), i);
mtx_unlock(&mq->mq_mutex);
}
}
Replace custom file descriptor array sleep lock constructed using a mutex and flags with an sxlock. This leads to a significant and measurable performance improvement as a result of access to shared locking for frequent lookup operations, reduced general overhead, and reduced overhead in the event of contention. All of these are imported for threaded applications where simultaneous access to a shared file descriptor array occurs frequently. Kris has reported 2x-4x transaction rate improvements on 8-core MySQL benchmarks; smaller improvements can be expected for many workloads as a result of reduced overhead. - Generally eliminate the distinction between "fast" and regular acquisisition of the filedesc lock; the plan is that they will now all be fast. Change all locking instances to either shared or exclusive locks. - Correct a bug (pointed out by kib) in fdfree() where previously msleep() was called without the mutex held; sx_sleep() is now always called with the sxlock held exclusively. - Universally hold the struct file lock over changes to struct file, rather than the filedesc lock or no lock. Always update the f_ops field last. A further memory barrier is required here in the future (discussed with jhb). - Improve locking and reference management in linux_at(), which fails to properly acquire vnode references before using vnode pointers. Annotate improper use of vn_fullpath(), which will be replaced at a future date. In fcntl(), we conservatively acquire an exclusive lock, even though in some cases a shared lock may be sufficient, which should be revisited. The dropping of the filedesc lock in fdgrowtable() is no longer required as the sxlock can be held over the sleep operation; we should consider removing that (pointed out by attilio). Tested by: kris Discussed with: jhb, kris, attilio, jeff
2007-04-04 09:11:34 +00:00
FILEDESC_SUNLOCK(fdp);
KASSERT(LIST_EMPTY(&p->p_mqnotifier), ("mq notifiers left"));
}
static int
mqf_poll(struct file *fp, int events, struct ucred *active_cred,
struct thread *td)
{
struct mqueue *mq = FPTOMQ(fp);
int revents = 0;
mtx_lock(&mq->mq_mutex);
if (events & (POLLIN | POLLRDNORM)) {
if (mq->mq_curmsgs) {
revents |= events & (POLLIN | POLLRDNORM);
} else {
mq->mq_flags |= MQ_RSEL;
selrecord(td, &mq->mq_rsel);
}
}
if (events & POLLOUT) {
if (mq->mq_curmsgs < mq->mq_maxmsg)
revents |= POLLOUT;
else {
mq->mq_flags |= MQ_WSEL;
selrecord(td, &mq->mq_wsel);
}
}
mtx_unlock(&mq->mq_mutex);
return (revents);
}
static int
mqf_close(struct file *fp, struct thread *td)
{
struct mqfs_node *pn;
fp->f_ops = &badfileops;
pn = fp->f_data;
fp->f_data = NULL;
sx_xlock(&mqfs_data.mi_lock);
mqnode_release(pn);
sx_xunlock(&mqfs_data.mi_lock);
return (0);
}
static int
mqf_stat(struct file *fp, struct stat *st, struct ucred *active_cred,
struct thread *td)
{
struct mqfs_node *pn = fp->f_data;
bzero(st, sizeof *st);
sx_xlock(&mqfs_data.mi_lock);
st->st_atim = pn->mn_atime;
st->st_mtim = pn->mn_mtime;
st->st_ctim = pn->mn_ctime;
st->st_birthtim = pn->mn_birth;
st->st_uid = pn->mn_uid;
st->st_gid = pn->mn_gid;
st->st_mode = S_IFIFO | pn->mn_mode;
sx_xunlock(&mqfs_data.mi_lock);
return (0);
}
static int
mqf_chmod(struct file *fp, mode_t mode, struct ucred *active_cred,
struct thread *td)
{
struct mqfs_node *pn;
int error;
error = 0;
pn = fp->f_data;
sx_xlock(&mqfs_data.mi_lock);
error = vaccess(VREG, pn->mn_mode, pn->mn_uid, pn->mn_gid, VADMIN,
active_cred, NULL);
if (error != 0)
goto out;
pn->mn_mode = mode & ACCESSPERMS;
out:
sx_xunlock(&mqfs_data.mi_lock);
return (error);
}
static int
mqf_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred,
struct thread *td)
{
struct mqfs_node *pn;
int error;
error = 0;
pn = fp->f_data;
sx_xlock(&mqfs_data.mi_lock);
if (uid == (uid_t)-1)
uid = pn->mn_uid;
if (gid == (gid_t)-1)
gid = pn->mn_gid;
if (((uid != pn->mn_uid && uid != active_cred->cr_uid) ||
(gid != pn->mn_gid && !groupmember(gid, active_cred))) &&
(error = priv_check_cred(active_cred, PRIV_VFS_CHOWN, 0)))
goto out;
pn->mn_uid = uid;
pn->mn_gid = gid;
out:
sx_xunlock(&mqfs_data.mi_lock);
return (error);
}
static int
mqf_kqfilter(struct file *fp, struct knote *kn)
{
struct mqueue *mq = FPTOMQ(fp);
int error = 0;
if (kn->kn_filter == EVFILT_READ) {
kn->kn_fop = &mq_rfiltops;
knlist_add(&mq->mq_rsel.si_note, kn, 0);
} else if (kn->kn_filter == EVFILT_WRITE) {
kn->kn_fop = &mq_wfiltops;
knlist_add(&mq->mq_wsel.si_note, kn, 0);
} else
error = EINVAL;
return (error);
}
static void
filt_mqdetach(struct knote *kn)
{
struct mqueue *mq = FPTOMQ(kn->kn_fp);
if (kn->kn_filter == EVFILT_READ)
knlist_remove(&mq->mq_rsel.si_note, kn, 0);
else if (kn->kn_filter == EVFILT_WRITE)
knlist_remove(&mq->mq_wsel.si_note, kn, 0);
else
panic("filt_mqdetach");
}
static int
filt_mqread(struct knote *kn, long hint)
{
struct mqueue *mq = FPTOMQ(kn->kn_fp);
mtx_assert(&mq->mq_mutex, MA_OWNED);
return (mq->mq_curmsgs != 0);
}
static int
filt_mqwrite(struct knote *kn, long hint)
{
struct mqueue *mq = FPTOMQ(kn->kn_fp);
mtx_assert(&mq->mq_mutex, MA_OWNED);
return (mq->mq_curmsgs < mq->mq_maxmsg);
}
static int
mqf_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
{
kif->kf_type = KF_TYPE_MQUEUE;
return (0);
}
static struct fileops mqueueops = {
.fo_read = invfo_rdwr,
.fo_write = invfo_rdwr,
.fo_truncate = invfo_truncate,
.fo_ioctl = invfo_ioctl,
.fo_poll = mqf_poll,
.fo_kqfilter = mqf_kqfilter,
.fo_stat = mqf_stat,
.fo_close = mqf_close,
.fo_chmod = mqf_chmod,
.fo_chown = mqf_chown,
.fo_sendfile = invfo_sendfile,
.fo_fill_kinfo = mqf_fill_kinfo,
};
static struct vop_vector mqfs_vnodeops = {
.vop_default = &default_vnodeops,
.vop_access = mqfs_access,
.vop_cachedlookup = mqfs_lookup,
.vop_lookup = vfs_cache_lookup,
.vop_reclaim = mqfs_reclaim,
.vop_create = mqfs_create,
.vop_remove = mqfs_remove,
.vop_inactive = mqfs_inactive,
.vop_open = mqfs_open,
.vop_close = mqfs_close,
.vop_getattr = mqfs_getattr,
.vop_setattr = mqfs_setattr,
.vop_read = mqfs_read,
.vop_write = VOP_EOPNOTSUPP,
.vop_readdir = mqfs_readdir,
.vop_mkdir = VOP_EOPNOTSUPP,
.vop_rmdir = VOP_EOPNOTSUPP
};
static struct vfsops mqfs_vfsops = {
.vfs_init = mqfs_init,
.vfs_uninit = mqfs_uninit,
.vfs_mount = mqfs_mount,
.vfs_unmount = mqfs_unmount,
.vfs_root = mqfs_root,
.vfs_statfs = mqfs_statfs,
};
static struct vfsconf mqueuefs_vfsconf = {
.vfc_version = VFS_VERSION,
.vfc_name = "mqueuefs",
.vfc_vfsops = &mqfs_vfsops,
.vfc_typenum = -1,
.vfc_flags = VFCF_SYNTHETIC
};
static struct syscall_helper_data mq_syscalls[] = {
SYSCALL_INIT_HELPER(kmq_open),
SYSCALL_INIT_HELPER(kmq_setattr),
SYSCALL_INIT_HELPER(kmq_timedsend),
SYSCALL_INIT_HELPER(kmq_timedreceive),
SYSCALL_INIT_HELPER(kmq_notify),
SYSCALL_INIT_HELPER(kmq_unlink),
SYSCALL_INIT_LAST
};
#ifdef COMPAT_FREEBSD32
#include <compat/freebsd32/freebsd32.h>
#include <compat/freebsd32/freebsd32_proto.h>
#include <compat/freebsd32/freebsd32_signal.h>
#include <compat/freebsd32/freebsd32_syscall.h>
#include <compat/freebsd32/freebsd32_util.h>
static void
mq_attr_from32(const struct mq_attr32 *from, struct mq_attr *to)
{
to->mq_flags = from->mq_flags;
to->mq_maxmsg = from->mq_maxmsg;
to->mq_msgsize = from->mq_msgsize;
to->mq_curmsgs = from->mq_curmsgs;
}
static void
mq_attr_to32(const struct mq_attr *from, struct mq_attr32 *to)
{
to->mq_flags = from->mq_flags;
to->mq_maxmsg = from->mq_maxmsg;
to->mq_msgsize = from->mq_msgsize;
to->mq_curmsgs = from->mq_curmsgs;
}
int
freebsd32_kmq_open(struct thread *td, struct freebsd32_kmq_open_args *uap)
{
struct mq_attr attr;
struct mq_attr32 attr32;
int flags, error;
if ((uap->flags & O_ACCMODE) == O_ACCMODE || uap->flags & O_EXEC)
return (EINVAL);
flags = FFLAGS(uap->flags);
if ((flags & O_CREAT) != 0 && uap->attr != NULL) {
error = copyin(uap->attr, &attr32, sizeof(attr32));
if (error)
return (error);
mq_attr_from32(&attr32, &attr);
}
return (kern_kmq_open(td, uap->path, flags, uap->mode,
uap->attr != NULL ? &attr : NULL));
}
int
freebsd32_kmq_setattr(struct thread *td, struct freebsd32_kmq_setattr_args *uap)
{
struct mq_attr attr, oattr;
struct mq_attr32 attr32, oattr32;
int error;
if (uap->attr != NULL) {
error = copyin(uap->attr, &attr32, sizeof(attr32));
if (error != 0)
return (error);
mq_attr_from32(&attr32, &attr);
}
error = kern_kmq_setattr(td, uap->mqd, uap->attr != NULL ? &attr : NULL,
&oattr);
if (error == 0 && uap->oattr != NULL) {
mq_attr_to32(&oattr, &oattr32);
bzero(oattr32.__reserved, sizeof(oattr32.__reserved));
error = copyout(&oattr32, uap->oattr, sizeof(oattr32));
}
return (error);
}
int
freebsd32_kmq_timedsend(struct thread *td,
struct freebsd32_kmq_timedsend_args *uap)
{
struct mqueue *mq;
struct file *fp;
struct timespec32 ets32;
struct timespec *abs_timeout, ets;
int error;
int waitok;
error = getmq_write(td, uap->mqd, &fp, NULL, &mq);
if (error)
return (error);
if (uap->abs_timeout != NULL) {
error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
if (error != 0)
return (error);
CP(ets32, ets, tv_sec);
CP(ets32, ets, tv_nsec);
abs_timeout = &ets;
} else
abs_timeout = NULL;
waitok = !(fp->f_flag & O_NONBLOCK);
error = mqueue_send(mq, uap->msg_ptr, uap->msg_len,
uap->msg_prio, waitok, abs_timeout);
fdrop(fp, td);
return (error);
}
int
freebsd32_kmq_timedreceive(struct thread *td,
struct freebsd32_kmq_timedreceive_args *uap)
{
struct mqueue *mq;
struct file *fp;
struct timespec32 ets32;
struct timespec *abs_timeout, ets;
int error, waitok;
error = getmq_read(td, uap->mqd, &fp, NULL, &mq);
if (error)
return (error);
if (uap->abs_timeout != NULL) {
error = copyin(uap->abs_timeout, &ets32, sizeof(ets32));
if (error != 0)
return (error);
CP(ets32, ets, tv_sec);
CP(ets32, ets, tv_nsec);
abs_timeout = &ets;
} else
abs_timeout = NULL;
waitok = !(fp->f_flag & O_NONBLOCK);
error = mqueue_receive(mq, uap->msg_ptr, uap->msg_len,
uap->msg_prio, waitok, abs_timeout);
fdrop(fp, td);
return (error);
}
int
freebsd32_kmq_notify(struct thread *td, struct freebsd32_kmq_notify_args *uap)
{
struct sigevent ev, *evp;
struct sigevent32 ev32;
int error;
if (uap->sigev == NULL) {
evp = NULL;
} else {
error = copyin(uap->sigev, &ev32, sizeof(ev32));
if (error != 0)
return (error);
error = convert_sigevent32(&ev32, &ev);
if (error != 0)
return (error);
evp = &ev;
}
return (kern_kmq_notify(td, uap->mqd, evp));
}
static struct syscall_helper_data mq32_syscalls[] = {
SYSCALL32_INIT_HELPER(freebsd32_kmq_open),
SYSCALL32_INIT_HELPER(freebsd32_kmq_setattr),
SYSCALL32_INIT_HELPER(freebsd32_kmq_timedsend),
SYSCALL32_INIT_HELPER(freebsd32_kmq_timedreceive),
SYSCALL32_INIT_HELPER(freebsd32_kmq_notify),
SYSCALL32_INIT_HELPER_COMPAT(kmq_unlink),
SYSCALL_INIT_LAST
};
#endif
static int
mqinit(void)
{
int error;
error = syscall_helper_register(mq_syscalls, SY_THR_STATIC_KLD);
if (error != 0)
return (error);
#ifdef COMPAT_FREEBSD32
error = syscall32_helper_register(mq32_syscalls, SY_THR_STATIC_KLD);
if (error != 0)
return (error);
#endif
return (0);
}
static int
mqunload(void)
{
#ifdef COMPAT_FREEBSD32
syscall32_helper_unregister(mq32_syscalls);
#endif
syscall_helper_unregister(mq_syscalls);
return (0);
}
static int
mq_modload(struct module *module, int cmd, void *arg)
{
int error = 0;
error = vfs_modevent(module, cmd, arg);
if (error != 0)
return (error);
switch (cmd) {
case MOD_LOAD:
error = mqinit();
if (error != 0)
mqunload();
break;
case MOD_UNLOAD:
error = mqunload();
break;
default:
break;
}
return (error);
}
static moduledata_t mqueuefs_mod = {
"mqueuefs",
mq_modload,
&mqueuefs_vfsconf
};
DECLARE_MODULE(mqueuefs, mqueuefs_mod, SI_SUB_VFS, SI_ORDER_MIDDLE);
MODULE_VERSION(mqueuefs, 1);