xref: /dflybsd-src/sys/vfs/mfs/mfs_vfsops.c (revision b370aff7747b2e03ce9b829fbf2877dffdadfb64)
1 /*
2  * Copyright (c) 1989, 1990, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)mfs_vfsops.c	8.11 (Berkeley) 6/19/95
34  * $FreeBSD: src/sys/ufs/mfs/mfs_vfsops.c,v 1.81.2.3 2001/07/04 17:35:21 tegge Exp $
35  * $DragonFly: src/sys/vfs/mfs/mfs_vfsops.c,v 1.41 2008/07/26 22:31:54 mneumann Exp $
36  */
37 
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/conf.h>
42 #include <sys/device.h>
43 #include <sys/kernel.h>
44 #include <sys/proc.h>
45 #include <sys/buf.h>
46 #include <sys/mount.h>
47 #include <sys/signalvar.h>
48 #include <sys/signal2.h>
49 #include <sys/vnode.h>
50 #include <sys/malloc.h>
51 #include <sys/sysproto.h>
52 #include <sys/mman.h>
53 #include <sys/linker.h>
54 #include <sys/fcntl.h>
55 #include <sys/nlookup.h>
56 #include <sys/devfs.h>
57 
58 #include <vm/vm.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_page.h>
61 #include <vm/vm_pager.h>
62 #include <vm/vnode_pager.h>
63 #include <vm/vm_extern.h>
64 
65 #include <sys/buf2.h>
66 #include <sys/thread2.h>
67 
68 #include <vfs/ufs/quota.h>
69 #include <vfs/ufs/inode.h>
70 #include <vfs/ufs/ufsmount.h>
71 #include <vfs/ufs/ufs_extern.h>
72 #include <vfs/ufs/fs.h>
73 #include <vfs/ufs/ffs_extern.h>
74 
75 #include "mfsnode.h"
76 #include "mfs_extern.h"
77 
78 MALLOC_DEFINE(M_MFSNODE, "MFS node", "MFS vnode private part");
79 
80 static int	mfs_mount (struct mount *mp,
81 			char *path, caddr_t data, struct ucred *td);
82 static int	mfs_start (struct mount *mp, int flags);
83 static int	mfs_statfs (struct mount *mp, struct statfs *sbp,
84 			struct ucred *cred);
85 static int	mfs_init (struct vfsconf *);
86 static void	mfs_doio(struct bio *bio, struct mfsnode *mfsp);
87 
88 d_open_t	mfsopen;
89 d_close_t	mfsclose;
90 d_strategy_t	mfsstrategy;
91 
92 static struct dev_ops mfs_ops = {
93 	{ "MFS", -1, D_DISK },
94 	.d_open =	mfsopen,
95 	.d_close =	mfsclose,
96 	.d_read =	physread,
97 	.d_write =	physwrite,
98 	.d_strategy =	mfsstrategy,
99 };
100 
101 /*
102  * mfs vfs operations.
103  */
104 static struct vfsops mfs_vfsops = {
105 	.vfs_mount =     	mfs_mount,
106 	.vfs_start =    	mfs_start,
107 	.vfs_unmount =   	ffs_unmount,
108 	.vfs_root =     	ufs_root,
109 	.vfs_quotactl =  	ufs_quotactl,
110 	.vfs_statfs =   	mfs_statfs,
111 	.vfs_sync =     	ffs_sync,
112 	.vfs_vget =      	ffs_vget,
113 	.vfs_fhtovp =   	ffs_fhtovp,
114 	.vfs_checkexp =  	ufs_check_export,
115 	.vfs_vptofh =   	ffs_vptofh,
116 	.vfs_init =     	mfs_init
117 };
118 
119 VFS_SET(mfs_vfsops, mfs, 0);
120 MODULE_VERSION(mfs, 1);
121 
122 /*
123  * We allow the underlying MFS block device to be opened and read.
124  */
125 int
126 mfsopen(struct dev_open_args *ap)
127 {
128 	cdev_t dev = ap->a_head.a_dev;
129 
130 #if 0
131 	if (ap->a_oflags & FWRITE)
132 		return(EROFS);
133 #endif
134 	if (dev->si_drv1)
135 		return(0);
136 	return(ENXIO);
137 }
138 
139 int
140 mfsclose(struct dev_close_args *ap)
141 {
142 	cdev_t dev = ap->a_head.a_dev;
143 	struct mfsnode *mfsp;
144 
145 	if ((mfsp = dev->si_drv1) == NULL)
146 		return(0);
147         mfsp->mfs_active = 0;
148         wakeup((caddr_t)mfsp);
149 	return(0);
150 }
151 
152 int
153 mfsstrategy(struct dev_strategy_args *ap)
154 {
155 	cdev_t dev = ap->a_head.a_dev;
156 	struct bio *bio = ap->a_bio;
157 	struct buf *bp = bio->bio_buf;
158 	off_t boff = bio->bio_offset;
159 	off_t eoff = boff + bp->b_bcount;
160 	struct mfsnode *mfsp;
161 
162 	if ((mfsp = dev->si_drv1) == NULL) {
163 		bp->b_error = ENXIO;
164 		goto error;
165 	}
166 	if (boff < 0)
167 		goto bad;
168 	if (eoff > mfsp->mfs_size) {
169 		if (boff > mfsp->mfs_size || (bp->b_flags & B_BNOCLIP))
170 			goto bad;
171 		/*
172 		 * Return EOF by completing the I/O with 0 bytes transfered.
173 		 * Set B_INVAL to indicate that any data in the buffer is not
174 		 * valid.
175 		 */
176 		if (boff == mfsp->mfs_size) {
177 			bp->b_resid = bp->b_bcount;
178 			bp->b_flags |= B_INVAL;
179 			goto done;
180 		}
181 		bp->b_bcount = mfsp->mfs_size - boff;
182 	}
183 
184 	/*
185 	 * Initiate I/O
186 	 */
187 	if (mfsp->mfs_td == curthread) {
188 		mfs_doio(bio, mfsp);
189 	} else {
190 		bioq_insert_tail(&mfsp->bio_queue, bio);
191 		wakeup((caddr_t)mfsp);
192 	}
193 	return(0);
194 
195 	/*
196 	 * Failure conditions on bio
197 	 */
198 bad:
199 	bp->b_error = EINVAL;
200 error:
201 	bp->b_flags |= B_ERROR | B_INVAL;
202 done:
203 	biodone(bio);
204 	return(0);
205 }
206 
207 /*
208  * mfs_mount
209  *
210  * Called when mounting local physical media
211  *
212  * PARAMETERS:
213  *		mountroot
214  *			mp	mount point structure
215  *			path	NULL (flag for root mount!!!)
216  *			data	<unused>
217  *			ndp	<unused>
218  *			p	process (user credentials check [statfs])
219  *
220  *		mount
221  *			mp	mount point structure
222  *			path	path to mount point
223  *			data	pointer to argument struct in user space
224  *			ndp	mount point namei() return (used for
225  *				credentials on reload), reused to look
226  *				up block device.
227  *			p	process (user credentials check)
228  *
229  * RETURNS:	0	Success
230  *		!0	error number (errno.h)
231  *
232  * LOCK STATE:
233  *
234  *		ENTRY
235  *			mount point is locked
236  *		EXIT
237  *			mount point is locked
238  *
239  * NOTES:
240  *		A NULL path can be used for a flag since the mount
241  *		system call will fail with EFAULT in copyinstr in
242  *		namei() if it is a genuine NULL from the user.
243  */
244 /* ARGSUSED */
245 static int
246 mfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
247 {
248 	struct vnode *devvp;
249 	struct mfs_args args;
250 	struct ufsmount *ump;
251 	struct fs *fs;
252 	struct mfsnode *mfsp;
253 	struct nlookupdata nd;
254 	size_t size;
255 	char devname[16];
256 	int flags;
257 	int minnum;
258 	int error;
259 	cdev_t dev;
260 
261 	/*
262 	 * Use NULL path to flag a root mount
263 	 */
264 	if (path == NULL) {
265 		/*
266 		 ***
267 		 * Mounting root file system
268 		 ***
269 		 */
270 
271 		/* you lose */
272 		panic("mfs_mount: mount MFS as root: not configured!");
273 	}
274 
275 	mfsp = NULL;
276 
277 	/*
278 	 ***
279 	 * Mounting non-root file system or updating a file system
280 	 ***
281 	 */
282 
283 	/* copy in user arguments*/
284 	error = copyin(data, (caddr_t)&args, sizeof (struct mfs_args));
285 	if (error)
286 		goto error_1;
287 
288 	/*
289 	 * If updating, check whether changing from read-only to
290 	 * read/write; if there is no device name, that's all we do.
291 	 */
292 	if (mp->mnt_flag & MNT_UPDATE) {
293 		/*
294 		 ********************
295 		 * UPDATE
296 		 ********************
297 		 */
298 		ump = VFSTOUFS(mp);
299 		fs = ump->um_fs;
300 		if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
301 			flags = WRITECLOSE;
302 			if (mp->mnt_flag & MNT_FORCE)
303 				flags |= FORCECLOSE;
304 			error = ffs_flushfiles(mp, flags);
305 			if (error)
306 				goto error_1;
307 		}
308 		if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
309 			/* XXX reopen the device vnode read-write */
310 			fs->fs_ronly = 0;
311 		}
312 		/* if not updating name...*/
313 		if (args.fspec == 0) {
314 			/*
315 			 * Process export requests.  Jumping to "success"
316 			 * will return the vfs_export() error code.
317 			 */
318 			error = vfs_export(mp, &ump->um_export, &args.export);
319 			goto success;
320 		}
321 
322 		/* XXX MFS does not support name updating*/
323 		goto success;
324 	}
325 
326 	/*
327 	 * Do the MALLOC before the make_dev since doing so afterward
328 	 * might cause a bogus v_data pointer to get dereferenced
329 	 * elsewhere if MALLOC should block.
330 	 */
331 	MALLOC(mfsp, struct mfsnode *, sizeof *mfsp, M_MFSNODE,
332 	       M_WAITOK|M_ZERO);
333 
334 	minnum = (int)curproc->p_pid;
335 
336 	dev = make_dev(&mfs_ops, minnum, UID_ROOT, GID_WHEEL, 0600,
337 		       "mfs%d", minnum);
338 	/* It is not clear that these will get initialized otherwise */
339 	dev->si_bsize_phys = DEV_BSIZE;
340 	dev->si_iosize_max = DFLTPHYS;
341 	dev->si_drv1 = mfsp;
342 	mfsp->mfs_baseoff = args.base;
343 	mfsp->mfs_size = args.size;
344 	mfsp->mfs_dev = dev;
345 	mfsp->mfs_td = curthread;
346 	mfsp->mfs_active = 1;
347 	bioq_init(&mfsp->bio_queue);
348 
349 	devfs_config();	/* sync devfs work */
350 	ksnprintf(devname, sizeof(devname), "/dev/mfs%d", minnum);
351 	nlookup_init(&nd, devname, UIO_SYSSPACE, 0);
352 	devvp = NULL;
353 	error = nlookup(&nd);
354 	if (error == 0) {
355 		devvp = nd.nl_nch.ncp->nc_vp;
356 		if (devvp == NULL)
357 			error = ENOENT;
358 		error = vget(devvp, LK_SHARED);
359 	}
360 	nlookup_done(&nd);
361 
362 	if (error)
363 		goto error_1;
364 	vn_unlock(devvp);
365 
366 	/*
367 	 * Our 'block' device must be backed by a VM object.  Theoretically
368 	 * we could use the anonymous memory VM object supplied by userland,
369 	 * but it would be somewhat of a complex task to deal with it
370 	 * that way since it would result in I/O requests which supply
371 	 * the VM pages from our own object.
372 	 *
373 	 * vnode_pager_alloc() is typically called when a VM object is
374 	 * being referenced externally.  We have to undo the refs for
375 	 * the self reference between vnode and object.
376 	 */
377 	vnode_pager_setsize(devvp, args.size);
378 
379 	/* Save "mounted from" info for mount point (NULL pad)*/
380 	copyinstr(args.fspec,			/* device name*/
381 		  mp->mnt_stat.f_mntfromname,	/* save area*/
382 		  MNAMELEN - 1,			/* max size*/
383 		  &size);			/* real size*/
384 	bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
385 	/* vref is eaten by mount? */
386 
387 	error = ffs_mountfs(devvp, mp, M_MFSNODE);
388 	if (error) {
389 		mfsp->mfs_active = 0;
390 		goto error_2;
391 	}
392 
393 	/*
394 	 * Initialize FS stat information in mount struct; uses
395 	 * mp->mnt_stat.f_mntfromname.
396 	 *
397 	 * This code is common to root and non-root mounts
398 	 */
399 	VFS_STATFS(mp, &mp->mnt_stat, cred);
400 
401 	goto success;
402 
403 error_2:	/* error with devvp held*/
404 	vrele(devvp);
405 
406 error_1:	/* no state to back out*/
407 	if (mfsp) {
408 		if (mfsp->mfs_dev) {
409 			destroy_dev(mfsp->mfs_dev);
410 			mfsp->mfs_dev = NULL;
411 		}
412 		FREE(mfsp, M_MFSNODE);
413 	}
414 
415 success:
416 	return(error);
417 }
418 
419 /*
420  * Used to grab the process and keep it in the kernel to service
421  * memory filesystem I/O requests.
422  *
423  * Loop servicing I/O requests.
424  * Copy the requested data into or out of the memory filesystem
425  * address space.
426  */
427 /* ARGSUSED */
428 static int
429 mfs_start(struct mount *mp, int flags)
430 {
431 	struct vnode *vp = VFSTOUFS(mp)->um_devvp;
432 	struct mfsnode *mfsp = vp->v_rdev->si_drv1;
433 	struct bio *bio;
434 	struct buf *bp;
435 	int gotsig = 0, sig;
436 	thread_t td = curthread;
437 
438 	/*
439 	 * We must prevent the system from trying to swap
440 	 * out or kill ( when swap space is low, see vm/pageout.c ) the
441 	 * process.  A deadlock can occur if the process is swapped out,
442 	 * and the system can loop trying to kill the unkillable ( while
443 	 * references exist ) MFS process when swap space is low.
444 	 */
445 	KKASSERT(curproc);
446 	PHOLD(curproc);
447 
448 	mfsp->mfs_td = td;
449 
450 	while (mfsp->mfs_active) {
451 		crit_enter();
452 
453 		while ((bio = bioq_first(&mfsp->bio_queue)) != NULL) {
454 			bioq_remove(&mfsp->bio_queue, bio);
455 			crit_exit();
456 			bp = bio->bio_buf;
457 			mfs_doio(bio, mfsp);
458 			wakeup(bp);
459 			crit_enter();
460 		}
461 
462 		crit_exit();
463 
464 		/*
465 		 * If a non-ignored signal is received, try to unmount.
466 		 * If that fails, clear the signal (it has been "processed"),
467 		 * otherwise we will loop here, as tsleep will always return
468 		 * EINTR/ERESTART.
469 		 */
470 		/*
471 		 * Note that dounmount() may fail if work was queued after
472 		 * we slept. We have to jump hoops here to make sure that we
473 		 * process any buffers after the sleep, before we dounmount()
474 		 */
475 		if (gotsig) {
476 			gotsig = 0;
477 			if (dounmount(mp, 0) != 0) {
478 				KKASSERT(td->td_proc);
479 				sig = CURSIG(td->td_lwp);
480 				if (sig) {
481 					spin_lock(&td->td_lwp->lwp_spin);
482 					lwp_delsig(td->td_lwp, sig);
483 					spin_unlock(&td->td_lwp->lwp_spin);
484 				}
485 			}
486 		}
487 		else if (tsleep((caddr_t)mfsp, PCATCH, "mfsidl", 0))
488 			gotsig++;	/* try to unmount in next pass */
489 	}
490 	PRELE(curproc);
491         if (mfsp->mfs_dev) {
492                 destroy_dev(mfsp->mfs_dev);
493                 mfsp->mfs_dev = NULL;
494         }
495 	FREE(mfsp, M_MFSNODE);
496 	return (0);
497 }
498 
499 /*
500  * Get file system statistics.
501  */
502 static int
503 mfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
504 {
505 	int error;
506 
507 	error = ffs_statfs(mp, sbp, cred);
508 	sbp->f_type = mp->mnt_vfc->vfc_typenum;
509 	return (error);
510 }
511 
512 /*
513  * Memory based filesystem initialization.
514  */
515 static int
516 mfs_init(struct vfsconf *vfsp)
517 {
518 	return (0);
519 }
520 
521 /*
522  * Memory file system I/O.
523  *
524  * Trivial on the HP since buffer has already been mapping into KVA space.
525  *
526  * Read and Write are handled with a simple copyin and copyout.
527  *
528  * We also partially support VOP_FREEBLKS().  We can't implement
529  * completely -- for example, on fragments or inode metadata, but we can
530  * implement it for page-aligned requests.
531  */
532 static void
533 mfs_doio(struct bio *bio, struct mfsnode *mfsp)
534 {
535 	struct buf *bp = bio->bio_buf;
536 	caddr_t base = mfsp->mfs_baseoff + bio->bio_offset;
537 	int bytes;
538 
539 	switch(bp->b_cmd) {
540 	case BUF_CMD_FREEBLKS:
541 		/*
542 		 * Implement FREEBLKS, which allows the filesystem to tell
543 		 * a block device when blocks are no longer needed (like when
544 		 * a file is deleted).  We use the hook to MADV_FREE the VM.
545 		 * This makes an MFS filesystem work as well or better then
546 		 * a sun-style swap-mounted filesystem.
547 		 */
548 		bytes = bp->b_bcount;
549 
550 		if ((vm_offset_t)base & PAGE_MASK) {
551 			int n = PAGE_SIZE - ((vm_offset_t)base & PAGE_MASK);
552 			bytes -= n;
553 			base += n;
554 		}
555                 if (bytes > 0) {
556                         struct madvise_args uap;
557 
558 			bytes &= ~PAGE_MASK;
559 			if (bytes != 0) {
560 				bzero(&uap, sizeof(uap));
561 				uap.addr  = base;
562 				uap.len   = bytes;
563 				uap.behav = MADV_FREE;
564 				sys_madvise(&uap);
565 			}
566                 }
567 		bp->b_error = 0;
568 		break;
569 	case BUF_CMD_READ:
570 		/*
571 		 * Read data from our 'memory' disk
572 		 */
573 		bp->b_error = copyin(base, bp->b_data, bp->b_bcount);
574 		break;
575 	case BUF_CMD_WRITE:
576 		/*
577 		 * Write data to our 'memory' disk
578 		 */
579 		bp->b_error = copyout(bp->b_data, base, bp->b_bcount);
580 		break;
581 	default:
582 		panic("mfs: bad b_cmd %d\n", bp->b_cmd);
583 	}
584 	if (bp->b_error)
585 		bp->b_flags |= B_ERROR;
586 	biodone(bio);
587 }
588