uvm_swap.c

来自「基于组件方式开发操作系统的OSKIT源代码」· C语言 代码 · 共 1,884 行 · 第 1/4 页

		for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
		     sdp != (void *)&spp->spi_swapdev;
		     sdp = CIRCLEQ_NEXT(sdp, swd_next))
			if (pgno >= sdp->swd_drumoffset &&
			    pgno < (sdp->swd_drumoffset + sdp->swd_drumsize)) {
				return sdp;
			}
	return NULL;
}

/*
 * sys_swapctl: main entry point for swapctl(2) system call
 * 	[with two helper functions: swap_on and swap_off]
 */
int
sys_swapctl(p, v, retval)
	struct proc *p;
	void *v;
	register_t *retval;
{
	struct sys_swapctl_args /* {
		syscallarg(int) cmd;
		syscallarg(void *) arg;
		syscallarg(int) misc;
	} */ *uap = (struct sys_swapctl_args *)v;
	struct vnode *vp;
#ifndef OSKIT
	struct nameidata nd;
#endif
	struct swappri *spp;
	struct swapdev *sdp;
	struct swapent *sep;
#ifndef OSKIT
	char	userpath[PATH_MAX + 1];
	size_t	len;
#endif
	int	count, error, misc;
	int	priority;
	UVMHIST_FUNC("sys_swapctl"); UVMHIST_CALLED(pdhist);

	misc = SCARG(uap, misc);

	/*
	 * ensure serialized syscall access by grabbing the swap_syscall_lock
	 */
	lockmgr(&swap_syscall_lock, LK_EXCLUSIVE, NULL);

	/*
	 * we handle the non-priv NSWAP and STATS request first.
	 *
	 * SWAP_NSWAP: return number of config'd swap devices 
	 * [can also be obtained with uvmexp sysctl]
	 */
	if (SCARG(uap, cmd) == SWAP_NSWAP) {
		UVMHIST_LOG(pdhist, "<- done SWAP_NSWAP=%d", uvmexp.nswapdev,
		    0, 0, 0);
		*retval = uvmexp.nswapdev;
		error = 0;
		goto out;
	}

	/*
	 * SWAP_STATS: get stats on current # of configured swap devs
	 *
	 * note that the swap_priority list can't change as long 
	 * as we are holding the swap_syscall_lock.  we don't want
	 * to grab the uvm.swap_data_lock because we may fault&sleep during 
	 * copyout() and we don't want to be holding that lock then!
	 */
	if (SCARG(uap, cmd) == SWAP_STATS
#if defined(COMPAT_13)
	    || SCARG(uap, cmd) == SWAP_OSTATS
#endif
	    ) {
		sep = (struct swapent *)SCARG(uap, arg);
		count = 0;

		for (spp = LIST_FIRST(&swap_priority); spp != NULL;
		    spp = LIST_NEXT(spp, spi_swappri)) {
			for (sdp = CIRCLEQ_FIRST(&spp->spi_swapdev);
			     sdp != (void *)&spp->spi_swapdev && misc-- > 0;
			     sdp = CIRCLEQ_NEXT(sdp, swd_next)) {
			  	/*
				 * backwards compatibility for system call.
				 * note that we use 'struct oswapent' as an
				 * overlay into both 'struct swapdev' and
				 * the userland 'struct swapent', as we
				 * want to retain backwards compatibility
				 * with NetBSD 1.3.
				 */
				sdp->swd_ose.ose_inuse = 
				    btodb(sdp->swd_npginuse << PAGE_SHIFT);
				error = copyout(&sdp->swd_ose, sep,
						sizeof(struct oswapent));

				/* now copy out the path if necessary */
#if defined(COMPAT_13)
				if (error == 0 && SCARG(uap, cmd) == SWAP_STATS)
#else
				if (error == 0)
#endif
					error = copyout(sdp->swd_path,
					    &sep->se_path, sdp->swd_pathlen);

				if (error)
					goto out;
				count++;
#if defined(COMPAT_13)
				if (SCARG(uap, cmd) == SWAP_OSTATS)
					((struct oswapent *)sep)++;
				else
#endif
					sep++;
			}
		}

		UVMHIST_LOG(pdhist, "<- done SWAP_STATS", 0, 0, 0, 0);

		*retval = count;
		error = 0;
		goto out;
	} 

#ifndef OSKIT
	/*
	 * all other requests require superuser privs.   verify.
	 */
	if ((error = suser(p->p_ucred, &p->p_acflag)))
		goto out;

	if (SCARG(uap, cmd) == SWAP_GETDUMPDEV) {
		dev_t	*devp = (dev_t *)SCARG(uap, arg);

		error = copyout(&dumpdev, devp, sizeof(dumpdev));
		goto out;
	}
#endif

	/*
	 * at this point we expect a path name in arg.   we will
	 * use namei() to gain a vnode reference (vref), and lock
	 * the vnode (VOP_LOCK).
	 *
	 * XXX: a NULL arg means use the root vnode pointer (e.g. for
	 * miniroot)
	 */
#ifdef OSKIT
	vp = (struct vnode*)SCARG(uap, arg); /* kill const */
#else
	if (SCARG(uap, arg) == NULL) {
		vp = rootvp;		/* miniroot */
		if (vget(vp, LK_EXCLUSIVE)) {
			error = EBUSY;
			goto out;
		}
		if (SCARG(uap, cmd) == SWAP_ON &&
		    copystr("miniroot", userpath, sizeof userpath, &len))
			panic("swapctl: miniroot copy failed");
	} else {
		int	space;
		char	*where;

		if (SCARG(uap, cmd) == SWAP_ON) {
			if ((error = copyinstr(SCARG(uap, arg), userpath,
			    sizeof userpath, &len)))
				goto out;
			space = UIO_SYSSPACE;
			where = userpath;
		} else {
			space = UIO_USERSPACE;
			where = (char *)SCARG(uap, arg);
		}
		NDINIT(&nd, LOOKUP, FOLLOW|LOCKLEAF, space, where, p);
		if ((error = namei(&nd)))
			goto out;
		vp = nd.ni_vp;
	}
#endif /*OSKIT*/
	/* note: "vp" is referenced and locked */

	error = 0;		/* assume no error */
	switch(SCARG(uap, cmd)) {

#ifndef OSKIT
	case SWAP_DUMPDEV:
		if (vp->v_type != VBLK) {
			error = ENOTBLK;
			goto out;
		}
		dumpdev = vp->v_rdev;
		
		break;
#endif

	case SWAP_CTL:
		/*
		 * get new priority, remove old entry (if any) and then
		 * reinsert it in the correct place.  finally, prune out
		 * any empty priority structures.
		 */
		priority = SCARG(uap, misc);
		spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK);
		simple_lock(&uvm.swap_data_lock);
		if ((sdp = swaplist_find(vp, 1)) == NULL) {
			error = ENOENT;
		} else {
			swaplist_insert(sdp, spp, priority);
			swaplist_trim();
		}
		simple_unlock(&uvm.swap_data_lock);
		if (error)
			free(spp, M_VMSWAP);
		break;

	case SWAP_ON:

		/*
		 * check for duplicates.   if none found, then insert a
		 * dummy entry on the list to prevent someone else from
		 * trying to enable this device while we are working on
		 * it.
		 */

		priority = SCARG(uap, misc);
		simple_lock(&uvm.swap_data_lock);
		if ((sdp = swaplist_find(vp, 0)) != NULL) {
			error = EBUSY;
			simple_unlock(&uvm.swap_data_lock);
			break;
		}
		sdp = malloc(sizeof *sdp, M_VMSWAP, M_WAITOK);
		spp = malloc(sizeof *spp, M_VMSWAP, M_WAITOK);
		memset(sdp, 0, sizeof(*sdp));
		sdp->swd_flags = SWF_FAKE;	/* placeholder only */
		sdp->swd_vp = vp;
		sdp->swd_dev = (vp->v_type == VBLK) ? vp->v_rdev : NODEV;
		BUFQ_INIT(&sdp->swd_tab);
#ifdef OSKIT
		pthread_mutex_init(&sdp->swd_lock, NULL);
#endif
		/*
		 * XXX Is NFS elaboration necessary?
		 */
		if (vp->v_type == VREG) {
			sdp->swd_cred = crdup(p->p_ucred);
		}

		swaplist_insert(sdp, spp, priority);
		simple_unlock(&uvm.swap_data_lock);

#ifndef OSKIT
		sdp->swd_pathlen = len;
		sdp->swd_path = malloc(sdp->swd_pathlen, M_VMSWAP, M_WAITOK);
		if (copystr(userpath, sdp->swd_path, sdp->swd_pathlen, 0) != 0)
			panic("swapctl: copystr");
#endif

		/*
		 * we've now got a FAKE placeholder in the swap list.
		 * now attempt to enable swap on it.  if we fail, undo
		 * what we've done and kill the fake entry we just inserted.
		 * if swap_on is a success, it will clear the SWF_FAKE flag
		 */

		if ((error = swap_on(p, sdp)) != 0) {
			simple_lock(&uvm.swap_data_lock);
			(void) swaplist_find(vp, 1);  /* kill fake entry */
			swaplist_trim();
			simple_unlock(&uvm.swap_data_lock);
			if (vp->v_type == VREG) {
				crfree(sdp->swd_cred);
			}
#ifdef OSKIT
			pthread_mutex_destroy(&sdp->swd_lock);
#endif
			free(sdp->swd_path, M_VMSWAP);
			free(sdp, M_VMSWAP);
			break;
		}
		break;

	case SWAP_OFF:
		simple_lock(&uvm.swap_data_lock);
		if ((sdp = swaplist_find(vp, 0)) == NULL) {
			simple_unlock(&uvm.swap_data_lock);
			error = ENXIO;
			break;
		}

		/*
		 * If a device isn't in use or enabled, we
		 * can't stop swapping from it (again).
		 */
		if ((sdp->swd_flags & (SWF_INUSE|SWF_ENABLE)) == 0) {
			simple_unlock(&uvm.swap_data_lock);
			error = EBUSY;
			break;
		}

		/*
		 * do the real work.
		 */
		if ((error = swap_off(p, sdp)) != 0)
			goto out;

		break;

	default:
		error = EINVAL;
	}

	/*
	 * done!  release the ref gained by namei() and unlock.
	 */
	vput(vp);

out:
	lockmgr(&swap_syscall_lock, LK_RELEASE, NULL);

	UVMHIST_LOG(pdhist, "<- done!  error=%d", error, 0, 0, 0);
	return (error);
}

/*
 * swap_on: attempt to enable a swapdev for swapping.   note that the
 *	swapdev is already on the global list, but disabled (marked
 *	SWF_FAKE).
 *
 * => we avoid the start of the disk (to protect disk labels)
 * => we also avoid the miniroot, if we are swapping to root.
 * => caller should leave uvm.swap_data_lock unlocked, we may lock it
 *	if needed.
 */
static int
swap_on(p, sdp)
	struct proc *p;
	struct swapdev *sdp;
{
	static int count = 0;	/* static */
	struct vnode *vp;
	int error, npages, nblocks, size;
	long addr;
	struct vattr va;
#ifdef NFS
	extern int (**nfsv2_vnodeop_p) __P((void *));
#endif /* NFS */
	dev_t dev;
	char *name;
	UVMHIST_FUNC("swap_on"); UVMHIST_CALLED(pdhist);

	/*
	 * we want to enable swapping on sdp.   the swd_vp contains
	 * the vnode we want (locked and ref'd), and the swd_dev
	 * contains the dev_t of the file, if it a block device.
	 */

	vp = sdp->swd_vp;
	dev = sdp->swd_dev;

#ifndef OSKIT
	/*
	 * open the swap file (mostly useful for block device files to
	 * let device driver know what is up).
	 *
	 * we skip the open/close for root on swap because the root
	 * has already been opened when root was mounted (mountroot).
	 */
	if (vp != rootvp) {
		if ((error = VOP_OPEN(vp, FREAD|FWRITE, p->p_ucred, p)))
			return (error);
	}
#endif

	/* XXX this only works for block devices */
	UVMHIST_LOG(pdhist, "  dev=%d, major(dev)=%d", dev, major(dev), 0,0);

	/*
	 * we now need to determine the size of the swap area.   for
	 * block specials we can call the d_psize function.
	 * for normal files, we must stat [get attrs].
	 *
	 * we put the result in nblks.
	 * for normal files, we also want the filesystem block size
	 * (which we get with statfs).
	 */
	switch (vp->v_type) {
	case VBLK:
		if (bdevsw[major(dev)].d_psize == 0 ||
		    (nblocks = (*bdevsw[major(dev)].d_psize)(dev)) == -1) {
			error = ENXIO;
			goto bad;
		}
		break;

	case VREG:
#ifdef OSKIT
#undef VOP_GETATTR
#undef VFS_STATFS
#define VOP_GETATTR(vp, va, cred, p)	oskit_uvm_vop_getattr(vp, va)
#define VFS_STATFS(mount, stat, p)	oskit_uvm_vfs_statfs(vp, stat)
#endif
		if ((error = VOP_GETATTR(vp, &va, p->p_ucred, p)))
			goto bad;
		nblocks = (int)btodb(va.va_size);
		if ((error =
		     VFS_STATFS(vp->v_mount, &vp->v_mount->mnt_stat, p)) != 0)
			goto bad;

		sdp->swd_bsize = vp->v_mount->mnt_stat.f_iosize;
		/*
		 * limit the max # of outstanding I/O requests we issue
		 * at any one time.   take it easy on NFS servers.
		 */
#ifdef NFS
		if (vp->v_op == nfsv2_vnodeop_p)
			sdp->swd_maxactive = 2; /* XXX */
		else
#endif /* NFS */
			sdp->swd_maxactive = 8; /* XXX */
		break;

	default:
		error = ENXIO;
		goto bad;
	}

	/*
	 * save nblocks in a safe place and convert to pages.
	 */

	sdp->swd_ose.ose_nblks = nblocks;
	npages = dbtob((u_int64_t)nblocks) >> PAGE_SHIFT;

	/*
	 * for block special files, we want to make sure that leave
	 * the disklabel and bootblocks alone, so we arrange to skip
	 * over them (arbitrarily choosing to skip PAGE_SIZE bytes).
	 * note that because of this the "size" can be less than the
	 * actual number of blocks on the device.
	 */
	if (vp->v_type == VBLK) {
		/* we use pages 1 to (size - 1) [inclusive] */
		size = npages - 1;
		addr = 1;
	} else {
		/* we use pages 0 to (size - 1) [inclusive] */
		size = npages;
		addr = 0;
	}

	/*
	 * make sure we have enough blocks for a reasonable sized swap
	 * area.   we want at least one page.
	 */

	if (size < 1) {
		UVMHIST_LOG(pdhist, "  size <= 1!!", 0, 0, 0, 0);
		error = EINVAL;
		goto bad;
	}

	UVMHIST_LOG(pdhist, "  dev=%x: size=%d addr=%ld\n", dev, size, addr, 0);

	/*
	 * now we need to allocate an extent to manage this swap device
	 */
	name = malloc(12, M_VMSWAP, M_WAITOK);
	sprintf(name, "swap0x%04x", count++);

	/* note that extent_create's 3rd arg is inclusive, thus "- 1" */
	sdp->swd_ex = extent_create(name, 0, npages - 1, M_VMSWAP,