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. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)mfs_vfsops.c 8.11 (Berkeley) 6/19/95
30 * $FreeBSD: src/sys/ufs/mfs/mfs_vfsops.c,v 1.81.2.3 2001/07/04 17:35:21 tegge Exp $
31 */
32
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/conf.h>
37 #include <sys/device.h>
38 #include <sys/kernel.h>
39 #include <sys/proc.h>
40 #include <sys/buf.h>
41 #include <sys/mount.h>
42 #include <sys/signalvar.h>
43 #include <sys/signal2.h>
44 #include <sys/spinlock2.h>
45 #include <sys/vnode.h>
46 #include <sys/malloc.h>
47 #include <sys/sysmsg.h>
48 #include <sys/mman.h>
49 #include <sys/linker.h>
50 #include <sys/fcntl.h>
51 #include <sys/nlookup.h>
52 #include <sys/devfs.h>
53
54 #include <vm/vm.h>
55 #include <vm/vm_object.h>
56 #include <vm/vm_page.h>
57 #include <vm/vm_pager.h>
58 #include <vm/vnode_pager.h>
59 #include <vm/vm_extern.h>
60
61 #include <sys/buf2.h>
62 #include <sys/thread2.h>
63
64 #include <vfs/ufs/quota.h>
65 #include <vfs/ufs/inode.h>
66 #include <vfs/ufs/ufsmount.h>
67 #include <vfs/ufs/ufs_extern.h>
68 #include <vfs/ufs/fs.h>
69 #include <vfs/ufs/ffs_extern.h>
70
71 #include "mfsnode.h"
72 #include "mfs_extern.h"
73
74 MALLOC_DEFINE(M_MFSNODE, "MFS node", "MFS vnode private part");
75
76 static int mfs_mount (struct mount *mp,
77 char *path, caddr_t data, struct ucred *td);
78 static int mfs_start (struct mount *mp, int flags);
79 static int mfs_statfs (struct mount *mp, struct statfs *sbp,
80 struct ucred *cred);
81 static int mfs_init (struct vfsconf *);
82 static void mfs_doio(struct bio *bio, struct mfsnode *mfsp);
83
84 d_open_t mfsopen;
85 d_close_t mfsclose;
86 d_strategy_t mfsstrategy;
87
88 static struct dev_ops mfs_ops = {
89 { "MFS", -1, D_DISK | D_NOEMERGPGR },
90 .d_open = mfsopen,
91 .d_close = mfsclose,
92 .d_read = physread,
93 .d_write = physwrite,
94 .d_strategy = mfsstrategy,
95 };
96
97 /*
98 * mfs vfs operations.
99 */
100 static struct vfsops mfs_vfsops = {
101 .vfs_flags = 0,
102 .vfs_mount = mfs_mount,
103 .vfs_start = mfs_start,
104 .vfs_unmount = ffs_unmount,
105 .vfs_root = ufs_root,
106 .vfs_quotactl = ufs_quotactl,
107 .vfs_statfs = mfs_statfs,
108 .vfs_sync = ffs_sync,
109 .vfs_vget = ffs_vget,
110 .vfs_fhtovp = ffs_fhtovp,
111 .vfs_checkexp = ufs_check_export,
112 .vfs_vptofh = ffs_vptofh,
113 .vfs_init = mfs_init
114 };
115
116 VFS_SET(mfs_vfsops, mfs, 0);
117 MODULE_VERSION(mfs, 1);
118
119 /*
120 * We allow the underlying MFS block device to be opened and read.
121 */
122 int
mfsopen(struct dev_open_args * ap)123 mfsopen(struct dev_open_args *ap)
124 {
125 cdev_t dev = ap->a_head.a_dev;
126
127 #if 0
128 if (ap->a_oflags & FWRITE)
129 return(EROFS);
130 #endif
131 if (dev->si_drv1)
132 return(0);
133 return(ENXIO);
134 }
135
136 int
mfsclose(struct dev_close_args * ap)137 mfsclose(struct dev_close_args *ap)
138 {
139 cdev_t dev = ap->a_head.a_dev;
140 struct mfsnode *mfsp;
141
142 if ((mfsp = dev->si_drv1) == NULL)
143 return(0);
144 mfsp->mfs_active = 0;
145 wakeup((caddr_t)mfsp);
146 return(0);
147 }
148
149 int
mfsstrategy(struct dev_strategy_args * ap)150 mfsstrategy(struct dev_strategy_args *ap)
151 {
152 cdev_t dev = ap->a_head.a_dev;
153 struct bio *bio = ap->a_bio;
154 struct buf *bp = bio->bio_buf;
155 off_t boff = bio->bio_offset;
156 off_t eoff = boff + bp->b_bcount;
157 struct mfsnode *mfsp;
158
159 if ((mfsp = dev->si_drv1) == NULL) {
160 bp->b_error = ENXIO;
161 goto error;
162 }
163 if (boff < 0)
164 goto bad;
165 if (eoff > mfsp->mfs_size) {
166 if (boff > mfsp->mfs_size || (bp->b_flags & B_BNOCLIP))
167 goto bad;
168 /*
169 * Return EOF by completing the I/O with 0 bytes transfered.
170 * Set B_INVAL to indicate that any data in the buffer is not
171 * valid.
172 */
173 if (boff == mfsp->mfs_size) {
174 bp->b_resid = bp->b_bcount;
175 bp->b_flags |= B_INVAL;
176 goto done;
177 }
178 bp->b_bcount = mfsp->mfs_size - boff;
179 }
180
181 /*
182 * Initiate I/O
183 */
184 if (mfsp->mfs_td == curthread) {
185 mfs_doio(bio, mfsp);
186 } else {
187 bioq_insert_tail(&mfsp->bio_queue, bio);
188 wakeup((caddr_t)mfsp);
189 }
190 return(0);
191
192 /*
193 * Failure conditions on bio
194 */
195 bad:
196 bp->b_error = EINVAL;
197 error:
198 bp->b_flags |= B_ERROR | B_INVAL;
199 done:
200 biodone(bio);
201 return(0);
202 }
203
204 /*
205 * mfs_mount
206 *
207 * Called when mounting local physical media
208 *
209 * PARAMETERS:
210 * mountroot
211 * mp mount point structure
212 * path NULL (flag for root mount!!!)
213 * data <unused>
214 * ndp <unused>
215 * p process (user credentials check [statfs])
216 *
217 * mount
218 * mp mount point structure
219 * path path to mount point
220 * data pointer to argument struct in user space
221 * ndp mount point namei() return (used for
222 * credentials on reload), reused to look
223 * up block device.
224 * p process (user credentials check)
225 *
226 * RETURNS: 0 Success
227 * !0 error number (errno.h)
228 *
229 * LOCK STATE:
230 *
231 * ENTRY
232 * mount point is locked
233 * EXIT
234 * mount point is locked
235 *
236 * NOTES:
237 * A NULL path can be used for a flag since the mount
238 * system call will fail with EFAULT in copyinstr in
239 * namei() if it is a genuine NULL from the user.
240 */
241 /* ARGSUSED */
242 static int
mfs_mount(struct mount * mp,char * path,caddr_t data,struct ucred * cred)243 mfs_mount(struct mount *mp, char *path, caddr_t data, struct ucred *cred)
244 {
245 struct vnode *devvp;
246 struct mfs_args args;
247 struct ufsmount *ump;
248 struct fs *fs;
249 struct mfsnode *mfsp;
250 struct nlookupdata nd;
251 size_t size;
252 char devname[16];
253 int flags;
254 int minnum;
255 int error;
256 cdev_t dev;
257
258 /*
259 * Use NULL path to flag a root mount
260 */
261 if (path == NULL) {
262 /*
263 ***
264 * Mounting root file system
265 ***
266 */
267
268 /* you lose */
269 panic("mfs_mount: mount MFS as root: not configured!");
270 }
271
272 mfsp = NULL;
273
274 /*
275 ***
276 * Mounting non-root file system or updating a file system
277 ***
278 */
279
280 /* copy in user arguments*/
281 error = copyin(data, (caddr_t)&args, sizeof (struct mfs_args));
282 if (error)
283 goto error_1;
284
285 /*
286 * If updating, check whether changing from read-only to
287 * read/write; if there is no device name, that's all we do.
288 */
289 if (mp->mnt_flag & MNT_UPDATE) {
290 /*
291 ********************
292 * UPDATE
293 ********************
294 */
295 ump = VFSTOUFS(mp);
296 fs = ump->um_fs;
297 if (fs->fs_ronly == 0 && (mp->mnt_flag & MNT_RDONLY)) {
298 flags = WRITECLOSE;
299 if (mp->mnt_flag & MNT_FORCE)
300 flags |= FORCECLOSE;
301 error = ffs_flushfiles(mp, flags);
302 if (error)
303 goto error_1;
304 }
305 if (fs->fs_ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
306 /* XXX reopen the device vnode read-write */
307 fs->fs_ronly = 0;
308 }
309 /* if not updating name...*/
310 if (args.fspec == 0) {
311 /*
312 * Process export requests. Jumping to "success"
313 * will return the vfs_export() error code.
314 */
315 error = vfs_export(mp, &ump->um_export, &args.export);
316 goto success;
317 }
318
319 /* XXX MFS does not support name updating*/
320 goto success;
321 }
322
323 /*
324 * Do the MALLOC before the make_dev since doing so afterward
325 * might cause a bogus v_data pointer to get dereferenced
326 * elsewhere if MALLOC should block.
327 */
328 mfsp = kmalloc(sizeof *mfsp, M_MFSNODE, M_WAITOK | M_ZERO);
329
330 minnum = (int)curproc->p_pid;
331
332 dev = make_dev(&mfs_ops, minnum, UID_ROOT, GID_WHEEL, 0600,
333 "mfs%d", minnum);
334 /* It is not clear that these will get initialized otherwise */
335 dev->si_bsize_phys = DEV_BSIZE;
336 dev->si_iosize_max = MAXPHYS;
337 dev->si_drv1 = mfsp;
338 mfsp->mfs_baseoff = args.base;
339 mfsp->mfs_size = args.size;
340 mfsp->mfs_dev = dev;
341 mfsp->mfs_td = curthread;
342 mfsp->mfs_active = 1;
343 bioq_init(&mfsp->bio_queue);
344
345 devfs_config(); /* sync devfs work */
346 ksnprintf(devname, sizeof(devname), "/dev/mfs%d", minnum);
347 nlookup_init(&nd, devname, UIO_SYSSPACE, 0);
348 devvp = NULL;
349 error = nlookup(&nd);
350 if (error == 0) {
351 devvp = nd.nl_nch.ncp->nc_vp;
352 if (devvp == NULL)
353 error = ENOENT;
354 error = vget(devvp, LK_SHARED);
355 }
356 nlookup_done(&nd);
357
358 if (error)
359 goto error_1;
360 vn_unlock(devvp);
361
362 /*
363 * Our 'block' device must be backed by a VM object. Theoretically
364 * we could use the anonymous memory VM object supplied by userland,
365 * but it would be somewhat of a complex task to deal with it
366 * that way since it would result in I/O requests which supply
367 * the VM pages from our own object.
368 *
369 * vnode_pager_alloc() is typically called when a VM object is
370 * being referenced externally. We have to undo the refs for
371 * the self reference between vnode and object.
372 */
373 vnode_pager_setsize(devvp, args.size);
374
375 /* Save "mounted from" info for mount point (NULL pad)*/
376 copyinstr(args.fspec, /* device name*/
377 mp->mnt_stat.f_mntfromname, /* save area*/
378 MNAMELEN - 1, /* max size*/
379 &size); /* real size*/
380 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
381 /* vref is eaten by mount? */
382
383 error = ffs_mountfs(devvp, mp, M_MFSNODE);
384 if (error) {
385 mfsp->mfs_active = 0;
386 goto error_2;
387 }
388
389 /*
390 * Initialize FS stat information in mount struct; uses
391 * mp->mnt_stat.f_mntfromname.
392 *
393 * This code is common to root and non-root mounts
394 */
395 VFS_STATFS(mp, &mp->mnt_stat, cred);
396
397 /*
398 * Mark VFS_START MPSAFE; this is to avoid accessing
399 * per-mount token after VFS_START exits
400 */
401 mp->mnt_kern_flag |= MNTK_ST_MPSAFE;
402
403 goto success;
404
405 error_2: /* error with devvp held*/
406 vrele(devvp);
407
408 error_1: /* no state to back out*/
409 if (mfsp) {
410 if (mfsp->mfs_dev) {
411 destroy_dev(mfsp->mfs_dev);
412 mfsp->mfs_dev = NULL;
413 }
414 kfree(mfsp, M_MFSNODE);
415 }
416
417 success:
418 return(error);
419 }
420
421 /*
422 * Used to grab the process and keep it in the kernel to service
423 * memory filesystem I/O requests.
424 *
425 * Loop servicing I/O requests.
426 * Copy the requested data into or out of the memory filesystem
427 * address space.
428 */
429 /* ARGSUSED */
430 static int
mfs_start(struct mount * mp,int flags)431 mfs_start(struct mount *mp, int flags)
432 {
433 struct vnode *vp = VFSTOUFS(mp)->um_devvp;
434 struct mfsnode *mfsp = vp->v_rdev->si_drv1;
435 struct bio *bio;
436 struct buf *bp;
437 int gotsig = 0, sig;
438 thread_t td = curthread;
439
440 /*
441 * We must prevent the system from trying to swap
442 * out or kill ( when swap space is low, see vm/pageout.c ) the
443 * process. A deadlock can occur if the process is swapped out,
444 * and the system can loop trying to kill the unkillable ( while
445 * references exist ) MFS process when swap space is low.
446 */
447 KKASSERT(curproc);
448 PHOLD(curproc);
449
450 mfsp->mfs_td = td;
451
452 while (mfsp->mfs_active) {
453 crit_enter();
454
455 while ((bio = bioq_takefirst(&mfsp->bio_queue)) != NULL) {
456 crit_exit();
457 bp = bio->bio_buf;
458 mfs_doio(bio, mfsp);
459 wakeup(bp);
460 crit_enter();
461 }
462
463 crit_exit();
464
465 /*
466 * If a non-ignored signal is received, try to unmount.
467 * If that fails, clear the signal (it has been "processed"),
468 * otherwise we will loop here, as tsleep will always return
469 * EINTR/ERESTART.
470 */
471 /*
472 * Note that dounmount() may fail if work was queued after
473 * we slept. We have to jump hoops here to make sure that we
474 * process any buffers after the sleep, before we dounmount()
475 */
476 if (gotsig) {
477 gotsig = 0;
478 if (dounmount(mp, 0, 0) != 0) {
479 KKASSERT(td->td_proc);
480 lwkt_gettoken(&td->td_proc->p_token);
481 sig = CURSIG(td->td_lwp);
482 if (sig) {
483 spin_lock(&td->td_lwp->lwp_spin);
484 lwp_delsig(td->td_lwp, sig, 1);
485 spin_unlock(&td->td_lwp->lwp_spin);
486 }
487 lwkt_reltoken(&td->td_proc->p_token);
488 }
489 }
490 else if (tsleep((caddr_t)mfsp, PCATCH, "mfsidl", 0))
491 gotsig++; /* try to unmount in next pass */
492 }
493 PRELE(curproc);
494 if (mfsp->mfs_dev) {
495 destroy_dev(mfsp->mfs_dev);
496 mfsp->mfs_dev = NULL;
497 }
498 kfree(mfsp, M_MFSNODE);
499 return (EMOUNTEXIT);
500 }
501
502 /*
503 * Get file system statistics.
504 */
505 static int
mfs_statfs(struct mount * mp,struct statfs * sbp,struct ucred * cred)506 mfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
507 {
508 int error;
509
510 error = ffs_statfs(mp, sbp, cred);
511 sbp->f_type = mp->mnt_vfc->vfc_typenum;
512 return (error);
513 }
514
515 /*
516 * Memory based filesystem initialization.
517 */
518 static int
mfs_init(struct vfsconf * vfsp)519 mfs_init(struct vfsconf *vfsp)
520 {
521 return (0);
522 }
523
524 /*
525 * Memory file system I/O.
526 *
527 * Trivial on the HP since buffer has already been mapping into KVA space.
528 *
529 * Read and Write are handled with a simple copyin and copyout.
530 *
531 * We also partially support VOP_FREEBLKS(). We can't implement
532 * completely -- for example, on fragments or inode metadata, but we can
533 * implement it for page-aligned requests.
534 */
535 static void
mfs_doio(struct bio * bio,struct mfsnode * mfsp)536 mfs_doio(struct bio *bio, struct mfsnode *mfsp)
537 {
538 struct buf *bp = bio->bio_buf;
539 caddr_t base = mfsp->mfs_baseoff + bio->bio_offset;
540 int bytes;
541
542 switch(bp->b_cmd) {
543 case BUF_CMD_FREEBLKS:
544 /*
545 * Implement FREEBLKS, which allows the filesystem to tell
546 * a block device when blocks are no longer needed (like when
547 * a file is deleted). We use the hook to MADV_FREE the VM.
548 * This makes an MFS filesystem work as well or better then
549 * a sun-style swap-mounted filesystem.
550 */
551 bytes = bp->b_bcount;
552
553 if ((vm_offset_t)base & PAGE_MASK) {
554 int n = PAGE_SIZE - ((vm_offset_t)base & PAGE_MASK);
555 bytes -= n;
556 base += n;
557 }
558 if (bytes > 0) {
559 struct madvise_args uap;
560
561 bytes &= ~PAGE_MASK;
562 if (bytes != 0) {
563 struct sysmsg sysmsg;
564
565 bzero(&sysmsg, sizeof(sysmsg));
566 bzero(&uap, sizeof(uap));
567 uap.addr = base;
568 uap.len = bytes;
569 uap.behav = MADV_FREE;
570 sys_madvise(&sysmsg, &uap);
571 }
572 }
573 bp->b_error = 0;
574 break;
575 case BUF_CMD_READ:
576 /*
577 * Read data from our 'memory' disk
578 */
579 bp->b_error = copyin(base, bp->b_data, bp->b_bcount);
580 break;
581 case BUF_CMD_WRITE:
582 /*
583 * Write data to our 'memory' disk
584 */
585 bp->b_error = copyout(bp->b_data, base, bp->b_bcount);
586 break;
587 default:
588 panic("mfs: bad b_cmd %d", bp->b_cmd);
589 }
590 if (bp->b_error)
591 bp->b_flags |= B_ERROR;
592 biodone(bio);
593 }
594