1 /* $NetBSD: lfs_vfsops.c,v 1.146.2.1 2004/05/29 09:05:04 tron Exp $ */
2
3 /*-
4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
5 * All rights reserved.
6 *
7 * This code is derived from software contributed to The NetBSD Foundation
8 * by Konrad E. Schroder <perseant@hhhh.org>.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. All advertising materials mentioning features or use of this software
19 * must display the following acknowledgement:
20 * This product includes software developed by the NetBSD
21 * Foundation, Inc. and its contributors.
22 * 4. Neither the name of The NetBSD Foundation nor the names of its
23 * contributors may be used to endorse or promote products derived
24 * from this software without specific prior written permission.
25 *
26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36 * POSSIBILITY OF SUCH DAMAGE.
37 */
38 /*-
39 * Copyright (c) 1989, 1991, 1993, 1994
40 * The Regents of the University of California. All rights reserved.
41 *
42 * Redistribution and use in source and binary forms, with or without
43 * modification, are permitted provided that the following conditions
44 * are met:
45 * 1. Redistributions of source code must retain the above copyright
46 * notice, this list of conditions and the following disclaimer.
47 * 2. Redistributions in binary form must reproduce the above copyright
48 * notice, this list of conditions and the following disclaimer in the
49 * documentation and/or other materials provided with the distribution.
50 * 3. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
53 *
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64 * SUCH DAMAGE.
65 *
66 * @(#)lfs_vfsops.c 8.20 (Berkeley) 6/10/95
67 */
68
69 #include <sys/cdefs.h>
70 __KERNEL_RCSID(0, "$NetBSD: lfs_vfsops.c,v 1.146.2.1 2004/05/29 09:05:04 tron Exp $");
71
72 #if defined(_KERNEL_OPT)
73 #include "opt_quota.h"
74 #endif
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/namei.h>
79 #include <sys/proc.h>
80 #include <sys/kernel.h>
81 #include <sys/vnode.h>
82 #include <sys/mount.h>
83 #include <sys/kthread.h>
84 #include <sys/buf.h>
85 #include <sys/device.h>
86 #include <sys/mbuf.h>
87 #include <sys/file.h>
88 #include <sys/disklabel.h>
89 #include <sys/ioctl.h>
90 #include <sys/errno.h>
91 #include <sys/malloc.h>
92 #include <sys/pool.h>
93 #include <sys/socket.h>
94 #include <uvm/uvm_extern.h>
95 #include <sys/sysctl.h>
96 #include <sys/conf.h>
97
98 #include <miscfs/specfs/specdev.h>
99
100 #include <ufs/ufs/quota.h>
101 #include <ufs/ufs/inode.h>
102 #include <ufs/ufs/ufsmount.h>
103 #include <ufs/ufs/ufs_extern.h>
104
105 #include <uvm/uvm.h>
106 #include <uvm/uvm_stat.h>
107 #include <uvm/uvm_pager.h>
108 #include <uvm/uvm_pdaemon.h>
109
110 #include <ufs/lfs/lfs.h>
111 #include <ufs/lfs/lfs_extern.h>
112
113 #include <miscfs/genfs/genfs.h>
114 #include <miscfs/genfs/genfs_node.h>
115
116 static int lfs_gop_write(struct vnode *, struct vm_page **, int, int);
117 static boolean_t lfs_issequential_hole(const struct ufsmount *,
118 daddr_t, daddr_t);
119
120 static int lfs_mountfs(struct vnode *, struct mount *, struct proc *);
121 static daddr_t check_segsum(struct lfs *, daddr_t, u_int64_t,
122 struct ucred *, int, int *, struct proc *);
123
124 extern const struct vnodeopv_desc lfs_vnodeop_opv_desc;
125 extern const struct vnodeopv_desc lfs_specop_opv_desc;
126 extern const struct vnodeopv_desc lfs_fifoop_opv_desc;
127
128 pid_t lfs_writer_daemon = 0;
129 int lfs_do_flush = 0;
130
131 const struct vnodeopv_desc * const lfs_vnodeopv_descs[] = {
132 &lfs_vnodeop_opv_desc,
133 &lfs_specop_opv_desc,
134 &lfs_fifoop_opv_desc,
135 NULL,
136 };
137
138 struct vfsops lfs_vfsops = {
139 MOUNT_LFS,
140 lfs_mount,
141 ufs_start,
142 lfs_unmount,
143 ufs_root,
144 ufs_quotactl,
145 lfs_statfs,
146 lfs_sync,
147 lfs_vget,
148 lfs_fhtovp,
149 lfs_vptofh,
150 lfs_init,
151 lfs_reinit,
152 lfs_done,
153 NULL,
154 lfs_mountroot,
155 ufs_check_export,
156 lfs_vnodeopv_descs,
157 };
158
159 struct genfs_ops lfs_genfsops = {
160 lfs_gop_size,
161 ufs_gop_alloc,
162 lfs_gop_write,
163 };
164
165 struct pool lfs_inode_pool;
166 struct pool lfs_dinode_pool;
167 struct pool lfs_inoext_pool;
168
169 /*
170 * The writer daemon. UVM keeps track of how many dirty pages we are holding
171 * in lfs_subsys_pages; the daemon flushes the filesystem when this value
172 * crosses the (user-defined) threshhold LFS_MAX_PAGES.
173 */
174 static void
175 lfs_writerd(void *arg)
176 {
177 #ifdef LFS_PD
178 struct mount *mp, *nmp;
179 struct lfs *fs;
180 #endif
181
182 lfs_writer_daemon = curproc->p_pid;
183
184 simple_lock(&lfs_subsys_lock);
185 for (;;) {
186 ltsleep(&lfs_writer_daemon, PVM | PNORELOCK, "lfswriter", 0,
187 &lfs_subsys_lock);
188
189 #ifdef LFS_PD
190 /*
191 * Look through the list of LFSs to see if any of them
192 * have requested pageouts.
193 */
194 simple_lock(&mountlist_slock);
195 for (mp = CIRCLEQ_FIRST(&mountlist); mp != (void *)&mountlist;
196 mp = nmp) {
197 if (vfs_busy(mp, LK_NOWAIT, &mountlist_slock)) {
198 nmp = CIRCLEQ_NEXT(mp, mnt_list);
199 continue;
200 }
201 if (strncmp(&mp->mnt_stat.f_fstypename[0], MOUNT_LFS,
202 MFSNAMELEN) == 0) {
203 fs = VFSTOUFS(mp)->um_lfs;
204 if (fs->lfs_pdflush ||
205 !TAILQ_EMPTY(&fs->lfs_pchainhd)) {
206 fs->lfs_pdflush = 0;
207 lfs_flush_fs(fs, 0);
208 }
209 }
210
211 simple_lock(&mountlist_slock);
212 nmp = CIRCLEQ_NEXT(mp, mnt_list);
213 vfs_unbusy(mp);
214 }
215 simple_unlock(&mountlist_slock);
216 #endif /* LFS_PD */
217
218 /*
219 * If global state wants a flush, flush everything.
220 */
221 simple_lock(&lfs_subsys_lock);
222 while (lfs_do_flush || locked_queue_count > LFS_MAX_BUFS ||
223 locked_queue_bytes > LFS_MAX_BYTES ||
224 lfs_subsys_pages > LFS_MAX_PAGES) {
225
226 #ifdef DEBUG_LFS_FLUSH
227 if (lfs_do_flush)
228 printf("daemon: lfs_do_flush\n");
229 if (locked_queue_count > LFS_MAX_BUFS)
230 printf("daemon: lqc = %d, max %d\n",
231 locked_queue_count, LFS_MAX_BUFS);
232 if (locked_queue_bytes > LFS_MAX_BYTES)
233 printf("daemon: lqb = %ld, max %ld\n",
234 locked_queue_bytes, LFS_MAX_BYTES);
235 if (lfs_subsys_pages > LFS_MAX_PAGES)
236 printf("daemon: lssp = %d, max %d\n",
237 lfs_subsys_pages, LFS_MAX_PAGES);
238 #endif /* DEBUG_LFS_FLUSH */
239 lfs_flush(NULL, SEGM_WRITERD);
240 lfs_do_flush = 0;
241 }
242 }
243 /* NOTREACHED */
244 }
245
246 /*
247 * Initialize the filesystem, most work done by ufs_init.
248 */
249 void
250 lfs_init()
251 {
252 #ifdef _LKM
253 malloc_type_attach(M_SEGMENT);
254 #endif
255 ufs_init();
256
257 /*
258 * XXX Same structure as FFS inodes? Should we share a common pool?
259 */
260 pool_init(&lfs_inode_pool, sizeof(struct inode), 0, 0, 0,
261 "lfsinopl", &pool_allocator_nointr);
262 pool_init(&lfs_dinode_pool, sizeof(struct ufs1_dinode), 0, 0, 0,
263 "lfsdinopl", &pool_allocator_nointr);
264 pool_init(&lfs_inoext_pool, sizeof(struct lfs_inode_ext), 8, 0, 0,
265 "lfsinoextpl", &pool_allocator_nointr);
266 #ifdef DEBUG
267 memset(lfs_log, 0, sizeof(lfs_log));
268 #endif
269 simple_lock_init(&lfs_subsys_lock);
270 }
271
272 void
273 lfs_reinit()
274 {
275 ufs_reinit();
276 }
277
278 void
279 lfs_done()
280 {
281 ufs_done();
282 pool_destroy(&lfs_inode_pool);
283 pool_destroy(&lfs_dinode_pool);
284 pool_destroy(&lfs_inoext_pool);
285 #ifdef _LKM
286 malloc_type_detach(M_SEGMENT);
287 #endif
288 }
289
290 /*
291 * Called by main() when ufs is going to be mounted as root.
292 */
293 int
294 lfs_mountroot()
295 {
296 extern struct vnode *rootvp;
297 struct mount *mp;
298 struct proc *p = curproc; /* XXX */
299 int error;
300
301 if (root_device->dv_class != DV_DISK)
302 return (ENODEV);
303
304 if (rootdev == NODEV)
305 return (ENODEV);
306 /*
307 * Get vnodes for swapdev and rootdev.
308 */
309 if ((error = bdevvp(rootdev, &rootvp))) {
310 printf("lfs_mountroot: can't setup bdevvp's");
311 return (error);
312 }
313 if ((error = vfs_rootmountalloc(MOUNT_LFS, "root_device", &mp))) {
314 vrele(rootvp);
315 return (error);
316 }
317 if ((error = lfs_mountfs(rootvp, mp, p))) {
318 mp->mnt_op->vfs_refcount--;
319 vfs_unbusy(mp);
320 free(mp, M_MOUNT);
321 vrele(rootvp);
322 return (error);
323 }
324 simple_lock(&mountlist_slock);
325 CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
326 simple_unlock(&mountlist_slock);
327 (void)lfs_statfs(mp, &mp->mnt_stat, p);
328 vfs_unbusy(mp);
329 inittodr(VFSTOUFS(mp)->um_lfs->lfs_tstamp);
330 return (0);
331 }
332
333 /*
334 * VFS Operations.
335 *
336 * mount system call
337 */
338 int
339 lfs_mount(struct mount *mp, const char *path, void *data, struct nameidata *ndp, struct proc *p)
340 {
341 struct vnode *devvp;
342 struct ufs_args args;
343 struct ufsmount *ump = NULL;
344 struct lfs *fs = NULL; /* LFS */
345 int error;
346 mode_t accessmode;
347
348 if (mp->mnt_flag & MNT_GETARGS) {
349 ump = VFSTOUFS(mp);
350 if (ump == NULL)
351 return EIO;
352 args.fspec = NULL;
353 vfs_showexport(mp, &args.export, &ump->um_export);
354 return copyout(&args, data, sizeof(args));
355 }
356 error = copyin(data, &args, sizeof (struct ufs_args));
357 if (error)
358 return (error);
359
360 /*
361 * If updating, check whether changing from read-only to
362 * read/write; if there is no device name, that's all we do.
363 */
364 if (mp->mnt_flag & MNT_UPDATE) {
365 ump = VFSTOUFS(mp);
366 fs = ump->um_lfs;
367 if (fs->lfs_ronly && (mp->mnt_iflag & IMNT_WANTRDWR)) {
368 /*
369 * If upgrade to read-write by non-root, then verify
370 * that user has necessary permissions on the device.
371 */
372 if (p->p_ucred->cr_uid != 0) {
373 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
374 error = VOP_ACCESS(ump->um_devvp, VREAD|VWRITE,
375 p->p_ucred, p);
376 VOP_UNLOCK(ump->um_devvp, 0);
377 if (error)
378 return (error);
379 }
380 fs->lfs_ronly = 0;
381 }
382 if (args.fspec == 0) {
383 /*
384 * Process export requests.
385 */
386 return (vfs_export(mp, &ump->um_export, &args.export));
387 }
388 }
389 /*
390 * Not an update, or updating the name: look up the name
391 * and verify that it refers to a sensible block device.
392 */
393 NDINIT(ndp, LOOKUP, FOLLOW, UIO_USERSPACE, args.fspec, p);
394 if ((error = namei(ndp)) != 0)
395 return (error);
396 devvp = ndp->ni_vp;
397 if (devvp->v_type != VBLK) {
398 vrele(devvp);
399 return (ENOTBLK);
400 }
401 if (bdevsw_lookup(devvp->v_rdev) == NULL) {
402 vrele(devvp);
403 return (ENXIO);
404 }
405 /*
406 * If mount by non-root, then verify that user has necessary
407 * permissions on the device.
408 */
409 if (p->p_ucred->cr_uid != 0) {
410 accessmode = VREAD;
411 if ((mp->mnt_flag & MNT_RDONLY) == 0)
412 accessmode |= VWRITE;
413 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
414 error = VOP_ACCESS(devvp, accessmode, p->p_ucred, p);
415 if (error) {
416 vput(devvp);
417 return (error);
418 }
419 VOP_UNLOCK(devvp, 0);
420 }
421 if ((mp->mnt_flag & MNT_UPDATE) == 0)
422 error = lfs_mountfs(devvp, mp, p); /* LFS */
423 else {
424 if (devvp != ump->um_devvp)
425 error = EINVAL; /* needs translation */
426 else
427 vrele(devvp);
428 }
429 if (error) {
430 vrele(devvp);
431 return (error);
432 }
433 ump = VFSTOUFS(mp);
434 fs = ump->um_lfs; /* LFS */
435 return set_statfs_info(path, UIO_USERSPACE, args.fspec,
436 UIO_USERSPACE, mp, p);
437 }
438
439 /*
440 * Roll-forward code.
441 */
442
443 /*
444 * Load the appropriate indirect block, and change the appropriate pointer.
445 * Mark the block dirty. Do segment and avail accounting.
446 */
447 static int
448 update_meta(struct lfs *fs, ino_t ino, int version, daddr_t lbn,
449 daddr_t ndaddr, size_t size, struct proc *p)
450 {
451 int error;
452 struct vnode *vp;
453 struct inode *ip;
454 #ifdef DEBUG_LFS_RFW
455 daddr_t odaddr;
456 struct indir a[NIADDR];
457 int num;
458 int i;
459 #endif /* DEBUG_LFS_RFW */
460 struct buf *bp;
461 SEGUSE *sup;
462
463 KASSERT(lbn >= 0); /* no indirect blocks */
464
465 if ((error = lfs_rf_valloc(fs, ino, version, p, &vp)) != 0) {
466 #ifdef DEBUG_LFS_RFW
467 printf("update_meta: ino %d: lfs_rf_valloc returned %d\n", ino,
468 error);
469 #endif /* DEBUG_LFS_RFW */
470 return error;
471 }
472
473 if ((error = VOP_BALLOC(vp, (lbn << fs->lfs_bshift), size,
474 NOCRED, 0, &bp)) != 0) {
475 vput(vp);
476 return (error);
477 }
478 /* No need to write, the block is already on disk */
479 if (bp->b_flags & B_DELWRI) {
480 LFS_UNLOCK_BUF(bp);
481 fs->lfs_avail += btofsb(fs, bp->b_bcount);
482 }
483 bp->b_flags |= B_INVAL;
484 brelse(bp);
485
486 /*
487 * Extend the file, if it is not large enough already.
488 * XXX this is not exactly right, we don't know how much of the
489 * XXX last block is actually used. We hope that an inode will
490 * XXX appear later to give the correct size.
491 */
492 ip = VTOI(vp);
493 if (ip->i_size <= (lbn << fs->lfs_bshift)) {
494 u_int64_t newsize;
495
496 if (lbn < NDADDR)
497 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) +
498 (size - fs->lfs_fsize) + 1;
499 else
500 newsize = ip->i_ffs1_size = (lbn << fs->lfs_bshift) + 1;
501
502 if (ip->i_size < newsize) {
503 ip->i_size = newsize;
504 /*
505 * tell vm our new size for the case the inode won't
506 * appear later.
507 */
508 uvm_vnp_setsize(vp, newsize);
509 }
510 }
511
512 lfs_update_single(fs, NULL, vp, lbn, ndaddr, size);
513
514 LFS_SEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
515 sup->su_nbytes += size;
516 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, ndaddr), bp);
517
518 /* differences here should be due to UNWRITTEN indirect blocks. */
519 KASSERT((lblkno(fs, ip->i_size) > NDADDR &&
520 ip->i_lfs_effnblks == ip->i_ffs1_blocks) ||
521 ip->i_lfs_effnblks >= ip->i_ffs1_blocks);
522
523 #ifdef DEBUG_LFS_RFW
524 /* Now look again to make sure it worked */
525 ufs_bmaparray(vp, lbn, &odaddr, &a[0], &num, NULL, NULL);
526 for (i = num; i > 0; i--) {
527 if (!a[i].in_exists)
528 panic("update_meta: absent %d lv indirect block", i);
529 }
530 if (dbtofsb(fs, odaddr) != ndaddr)
531 printf("update_meta: failed setting ino %d lbn %" PRId64
532 " to %" PRId64 "\n", ino, lbn, ndaddr);
533 #endif /* DEBUG_LFS_RFW */
534 vput(vp);
535 return 0;
536 }
537
538 static int
539 update_inoblk(struct lfs *fs, daddr_t offset, struct ucred *cred,
540 struct proc *p)
541 {
542 struct vnode *devvp, *vp;
543 struct inode *ip;
544 struct ufs1_dinode *dip;
545 struct buf *dbp, *ibp;
546 int error;
547 daddr_t daddr;
548 IFILE *ifp;
549 SEGUSE *sup;
550
551 devvp = VTOI(fs->lfs_ivnode)->i_devvp;
552
553 /*
554 * Get the inode, update times and perms.
555 * DO NOT update disk blocks, we do that separately.
556 */
557 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_ibsize, cred, &dbp);
558 if (error) {
559 #ifdef DEBUG_LFS_RFW
560 printf("update_inoblk: bread returned %d\n", error);
561 #endif
562 return error;
563 }
564 dip = ((struct ufs1_dinode *)(dbp->b_data)) + INOPB(fs);
565 while (--dip >= (struct ufs1_dinode *)dbp->b_data) {
566 if (dip->di_inumber > LFS_IFILE_INUM) {
567 /* printf("ino %d version %d\n", dip->di_inumber,
568 dip->di_gen); */
569 error = lfs_rf_valloc(fs, dip->di_inumber, dip->di_gen,
570 p, &vp);
571 if (error) {
572 #ifdef DEBUG_LFS_RFW
573 printf("update_inoblk: lfs_rf_valloc returned %d\n", error);
574 #endif
575 continue;
576 }
577 ip = VTOI(vp);
578 if (dip->di_size != ip->i_size)
579 VOP_TRUNCATE(vp, dip->di_size, 0, NOCRED, p);
580 /* Get mode, link count, size, and times */
581 memcpy(ip->i_din.ffs1_din, dip,
582 offsetof(struct ufs1_dinode, di_db[0]));
583
584 /* Then the rest, except di_blocks */
585 ip->i_flags = ip->i_ffs1_flags = dip->di_flags;
586 ip->i_gen = ip->i_ffs1_gen = dip->di_gen;
587 ip->i_uid = ip->i_ffs1_uid = dip->di_uid;
588 ip->i_gid = ip->i_ffs1_gid = dip->di_gid;
589
590 ip->i_mode = ip->i_ffs1_mode;
591 ip->i_nlink = ip->i_ffs_effnlink = ip->i_ffs1_nlink;
592 ip->i_size = ip->i_ffs1_size;
593
594 LFS_SET_UINO(ip, IN_CHANGE | IN_MODIFIED | IN_UPDATE);
595
596 /* Re-initialize to get type right */
597 ufs_vinit(vp->v_mount, lfs_specop_p, lfs_fifoop_p,
598 &vp);
599 vput(vp);
600
601 /* Record change in location */
602 LFS_IENTRY(ifp, fs, dip->di_inumber, ibp);
603 daddr = ifp->if_daddr;
604 ifp->if_daddr = dbtofsb(fs, dbp->b_blkno);
605 error = LFS_BWRITE_LOG(ibp); /* Ifile */
606 /* And do segment accounting */
607 if (dtosn(fs, daddr) != dtosn(fs, dbtofsb(fs, dbp->b_blkno))) {
608 if (daddr > 0) {
609 LFS_SEGENTRY(sup, fs, dtosn(fs, daddr),
610 ibp);
611 sup->su_nbytes -= sizeof (struct ufs1_dinode);
612 LFS_WRITESEGENTRY(sup, fs,
613 dtosn(fs, daddr),
614 ibp);
615 }
616 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
617 ibp);
618 sup->su_nbytes += sizeof (struct ufs1_dinode);
619 LFS_WRITESEGENTRY(sup, fs,
620 dtosn(fs, dbtofsb(fs, dbp->b_blkno)),
621 ibp);
622 }
623 }
624 }
625 dbp->b_flags |= B_AGE;
626 brelse(dbp);
627
628 return 0;
629 }
630
631 #define CHECK_CKSUM 0x0001 /* Check the checksum to make sure it's valid */
632 #define CHECK_UPDATE 0x0002 /* Update Ifile for new data blocks / inodes */
633
634 static daddr_t
635 check_segsum(struct lfs *fs, daddr_t offset, u_int64_t nextserial,
636 struct ucred *cred, int flags, int *pseg_flags, struct proc *p)
637 {
638 struct vnode *devvp;
639 struct buf *bp, *dbp;
640 int error, nblocks = 0, ninos, i, j; /* XXX: gcc */
641 SEGSUM *ssp;
642 u_long *dp = NULL, *datap = NULL; /* XXX u_int32_t */
643 daddr_t oldoffset;
644 int32_t *iaddr; /* XXX ondisk32 */
645 FINFO *fip;
646 SEGUSE *sup;
647 size_t size;
648
649 devvp = VTOI(fs->lfs_ivnode)->i_devvp;
650 /*
651 * If the segment has a superblock and we're at the top
652 * of the segment, skip the superblock.
653 */
654 if (sntod(fs, dtosn(fs, offset)) == offset) {
655 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
656 if (sup->su_flags & SEGUSE_SUPERBLOCK)
657 offset += btofsb(fs, LFS_SBPAD);
658 brelse(bp);
659 }
660
661 /* Read in the segment summary */
662 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_sumsize, cred, &bp);
663 if (error)
664 return -1;
665
666 /* Check summary checksum */
667 ssp = (SEGSUM *)bp->b_data;
668 if (flags & CHECK_CKSUM) {
669 if (ssp->ss_sumsum != cksum(&ssp->ss_datasum,
670 fs->lfs_sumsize -
671 sizeof(ssp->ss_sumsum))) {
672 #ifdef DEBUG_LFS_RFW
673 printf("Sumsum error at 0x%" PRIx64 "\n", offset);
674 #endif
675 offset = -1;
676 goto err1;
677 }
678 if (ssp->ss_nfinfo == 0 && ssp->ss_ninos == 0) {
679 #ifdef DEBUG_LFS_RFW
680 printf("Empty pseg at 0x%" PRIx64 "\n", offset);
681 #endif
682 offset = -1;
683 goto err1;
684 }
685 if (ssp->ss_create < fs->lfs_tstamp) {
686 #ifdef DEBUG_LFS_RFW
687 printf("Old data at 0x%" PRIx64 "\n", offset);
688 #endif
689 offset = -1;
690 goto err1;
691 }
692 }
693 if (fs->lfs_version > 1) {
694 if (ssp->ss_serial != nextserial) {
695 #ifdef DEBUG_LFS_RFW
696 printf("Unexpected serial number at 0x%" PRIx64
697 "\n", offset);
698 #endif
699 offset = -1;
700 goto err1;
701 }
702 if (ssp->ss_ident != fs->lfs_ident) {
703 #ifdef DEBUG_LFS_RFW
704 printf("Incorrect fsid (0x%x vs 0x%x) at 0x%"
705 PRIx64 "\n", ssp->ss_ident, fs->lfs_ident, offset);
706 #endif
707 offset = -1;
708 goto err1;
709 }
710 }
711 if (pseg_flags)
712 *pseg_flags = ssp->ss_flags;
713 oldoffset = offset;
714 offset += btofsb(fs, fs->lfs_sumsize);
715
716 ninos = howmany(ssp->ss_ninos, INOPB(fs));
717 /* XXX ondisk32 */
718 iaddr = (int32_t *)(bp->b_data + fs->lfs_sumsize - sizeof(int32_t));
719 if (flags & CHECK_CKSUM) {
720 /* Count blocks */
721 nblocks = 0;
722 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
723 for (i = 0; i < ssp->ss_nfinfo; ++i) {
724 nblocks += fip->fi_nblocks;
725 if (fip->fi_nblocks <= 0)
726 break;
727 /* XXX ondisk32 */
728 fip = (FINFO *)(((char *)fip) + FINFOSIZE +
729 (fip->fi_nblocks * sizeof(int32_t)));
730 }
731 nblocks += ninos;
732 /* Create the sum array */
733 datap = dp = (u_long *)malloc(nblocks * sizeof(u_long),
734 M_SEGMENT, M_WAITOK);
735 }
736
737 /* Handle individual blocks */
738 fip = (FINFO *)(bp->b_data + SEGSUM_SIZE(fs));
739 for (i = 0; i < ssp->ss_nfinfo || ninos; ++i) {
740 /* Inode block? */
741 if (ninos && *iaddr == offset) {
742 if (flags & CHECK_CKSUM) {
743 /* Read in the head and add to the buffer */
744 error = bread(devvp, fsbtodb(fs, offset), fs->lfs_bsize,
745 cred, &dbp);
746 if (error) {
747 offset = -1;
748 goto err2;
749 }
750 (*dp++) = ((u_long *)(dbp->b_data))[0];
751 dbp->b_flags |= B_AGE;
752 brelse(dbp);
753 }
754 if (flags & CHECK_UPDATE) {
755 if ((error = update_inoblk(fs, offset, cred, p))
756 != 0) {
757 offset = -1;
758 goto err2;
759 }
760 }
761 offset += btofsb(fs, fs->lfs_ibsize);
762 --iaddr;
763 --ninos;
764 --i; /* compensate */
765 continue;
766 }
767 /* printf("check: blocks from ino %d version %d\n",
768 fip->fi_ino, fip->fi_version); */
769 size = fs->lfs_bsize;
770 for (j = 0; j < fip->fi_nblocks; ++j) {
771 if (j == fip->fi_nblocks - 1)
772 size = fip->fi_lastlength;
773 if (flags & CHECK_CKSUM) {
774 error = bread(devvp, fsbtodb(fs, offset), size, cred, &dbp);
775 if (error) {
776 offset = -1;
777 goto err2;
778 }
779 (*dp++) = ((u_long *)(dbp->b_data))[0];
780 dbp->b_flags |= B_AGE;
781 brelse(dbp);
782 }
783 /* Account for and update any direct blocks */
784 if ((flags & CHECK_UPDATE) &&
785 fip->fi_ino > LFS_IFILE_INUM &&
786 fip->fi_blocks[j] >= 0) {
787 update_meta(fs, fip->fi_ino, fip->fi_version,
788 fip->fi_blocks[j], offset, size, p);
789 }
790 offset += btofsb(fs, size);
791 }
792 /* XXX ondisk32 */
793 fip = (FINFO *)(((char *)fip) + FINFOSIZE
794 + fip->fi_nblocks * sizeof(int32_t));
795 }
796 /* Checksum the array, compare */
797 if ((flags & CHECK_CKSUM) &&
798 ssp->ss_datasum != cksum(datap, nblocks * sizeof(u_long)))
799 {
800 #ifdef DEBUG_LFS_RFW
801 printf("Datasum error at 0x%" PRIx64 " (wanted %x got %x)\n",
802 offset, ssp->ss_datasum, cksum(datap, nblocks *
803 sizeof(u_long)));
804 #endif
805 offset = -1;
806 goto err2;
807 }
808
809 /* If we're at the end of the segment, move to the next */
810 if (dtosn(fs, offset + btofsb(fs, fs->lfs_sumsize + fs->lfs_bsize)) !=
811 dtosn(fs, offset)) {
812 if (dtosn(fs, offset) == dtosn(fs, ssp->ss_next)) {
813 offset = -1;
814 goto err2;
815 }
816 offset = ssp->ss_next;
817 #ifdef DEBUG_LFS_RFW
818 printf("LFS roll forward: moving on to offset 0x%" PRIx64
819 " -> segment %d\n", offset, dtosn(fs,offset));
820 #endif
821 }
822
823 if (flags & CHECK_UPDATE) {
824 fs->lfs_avail -= (offset - oldoffset);
825 /* Don't clog the buffer queue */
826 simple_lock(&lfs_subsys_lock);
827 if (locked_queue_count > LFS_MAX_BUFS ||
828 locked_queue_bytes > LFS_MAX_BYTES) {
829 lfs_flush(fs, SEGM_CKP);
830 }
831 simple_unlock(&lfs_subsys_lock);
832 }
833
834 err2:
835 if (flags & CHECK_CKSUM)
836 free(datap, M_SEGMENT);
837 err1:
838 bp->b_flags |= B_AGE;
839 brelse(bp);
840
841 /* XXX should we update the serial number even for bad psegs? */
842 if ((flags & CHECK_UPDATE) && offset > 0 && fs->lfs_version > 1)
843 fs->lfs_serial = nextserial;
844 return offset;
845 }
846
847 /*
848 * Common code for mount and mountroot
849 * LFS specific
850 */
851 int
852 lfs_mountfs(struct vnode *devvp, struct mount *mp, struct proc *p)
853 {
854 extern struct vnode *rootvp;
855 struct dlfs *tdfs, *dfs, *adfs;
856 struct lfs *fs;
857 struct ufsmount *ump;
858 struct vnode *vp;
859 struct buf *bp, *abp;
860 struct partinfo dpart;
861 dev_t dev;
862 int error, i, ronly, secsize, fsbsize;
863 struct ucred *cred;
864 CLEANERINFO *cip;
865 SEGUSE *sup;
866 int flags, dirty, do_rollforward;
867 daddr_t offset, oldoffset, lastgoodpseg, sb_addr;
868 int sn, curseg;
869
870 cred = p ? p->p_ucred : NOCRED;
871 /*
872 * Disallow multiple mounts of the same device.
873 * Disallow mounting of a device that is currently in use
874 * (except for root, which might share swap device for miniroot).
875 * Flush out any old buffers remaining from a previous use.
876 */
877 if ((error = vfs_mountedon(devvp)) != 0)
878 return (error);
879 if (vcount(devvp) > 1 && devvp != rootvp)
880 return (EBUSY);
881 if ((error = vinvalbuf(devvp, V_SAVE, cred, p, 0, 0)) != 0)
882 return (error);
883
884 ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
885 error = VOP_OPEN(devvp, ronly ? FREAD : FREAD|FWRITE, FSCRED, p);
886 if (error)
887 return (error);
888 if (VOP_IOCTL(devvp, DIOCGPART, &dpart, FREAD, cred, p) != 0)
889 secsize = DEV_BSIZE;
890 else
891 secsize = dpart.disklab->d_secsize;
892
893 /* Don't free random space on error. */
894 bp = NULL;
895 abp = NULL;
896 ump = NULL;
897
898 sb_addr = LFS_LABELPAD / secsize;
899 while (1) {
900 /* Read in the superblock. */
901 error = bread(devvp, sb_addr, LFS_SBPAD, cred, &bp);
902 if (error)
903 goto out;
904 dfs = (struct dlfs *)bp->b_data;
905
906 /* Check the basics. */
907 if (dfs->dlfs_magic != LFS_MAGIC || dfs->dlfs_bsize >= MAXBSIZE ||
908 dfs->dlfs_version > LFS_VERSION ||
909 dfs->dlfs_bsize < sizeof(struct dlfs)) {
910 #ifdef DEBUG_LFS
911 printf("lfs_mountfs: primary superblock sanity failed\n");
912 #endif
913 error = EINVAL; /* XXX needs translation */
914 goto out;
915 }
916 if (dfs->dlfs_inodefmt > LFS_MAXINODEFMT)
917 printf("lfs_mountfs: warning: unknown inode format %d\n",
918 dfs->dlfs_inodefmt);
919
920 if (dfs->dlfs_version == 1)
921 fsbsize = secsize;
922 else {
923 fsbsize = 1 << (dfs->dlfs_bshift - dfs->dlfs_blktodb +
924 dfs->dlfs_fsbtodb);
925 /*
926 * Could be, if the frag size is large enough, that we
927 * don't have the "real" primary superblock. If that's
928 * the case, get the real one, and try again.
929 */
930 if (sb_addr != dfs->dlfs_sboffs[0] <<
931 dfs->dlfs_fsbtodb) {
932 /* #ifdef DEBUG_LFS */
933 printf("lfs_mountfs: sb daddr 0x%llx is not right, trying 0x%llx\n",
934 (long long)sb_addr, (long long)(dfs->dlfs_sboffs[0] <<
935 dfs->dlfs_fsbtodb));
936 /* #endif */
937 sb_addr = dfs->dlfs_sboffs[0] <<
938 dfs->dlfs_fsbtodb;
939 brelse(bp);
940 continue;
941 }
942 }
943 break;
944 }
945
946 /*
947 * Check the second superblock to see which is newer; then mount
948 * using the older of the two. This is necessary to ensure that
949 * the filesystem is valid if it was not unmounted cleanly.
950 */
951
952 if (dfs->dlfs_sboffs[1] &&
953 dfs->dlfs_sboffs[1] - LFS_LABELPAD / fsbsize > LFS_SBPAD / fsbsize)
954 {
955 error = bread(devvp, dfs->dlfs_sboffs[1] * (fsbsize / secsize),
956 LFS_SBPAD, cred, &abp);
957 if (error)
958 goto out;
959 adfs = (struct dlfs *)abp->b_data;
960
961 if (dfs->dlfs_version == 1) {
962 /* 1s resolution comparison */
963 if (adfs->dlfs_tstamp < dfs->dlfs_tstamp)
964 tdfs = adfs;
965 else
966 tdfs = dfs;
967 } else {
968 /* monotonic infinite-resolution comparison */
969 if (adfs->dlfs_serial < dfs->dlfs_serial)
970 tdfs = adfs;
971 else
972 tdfs = dfs;
973 }
974
975 /* Check the basics. */
976 if (tdfs->dlfs_magic != LFS_MAGIC ||
977 tdfs->dlfs_bsize > MAXBSIZE ||
978 tdfs->dlfs_version > LFS_VERSION ||
979 tdfs->dlfs_bsize < sizeof(struct dlfs)) {
980 #ifdef DEBUG_LFS
981 printf("lfs_mountfs: alt superblock sanity failed\n");
982 #endif
983 error = EINVAL; /* XXX needs translation */
984 goto out;
985 }
986 } else {
987 #ifdef DEBUG_LFS
988 printf("lfs_mountfs: invalid alt superblock daddr=0x%x\n",
989 dfs->dlfs_sboffs[1]);
990 #endif
991 error = EINVAL;
992 goto out;
993 }
994
995 /* Allocate the mount structure, copy the superblock into it. */
996 fs = malloc(sizeof(struct lfs), M_UFSMNT, M_WAITOK | M_ZERO);
997 memcpy(&fs->lfs_dlfs, tdfs, sizeof(struct dlfs));
998
999 /* Compatibility */
1000 if (fs->lfs_version < 2) {
1001 fs->lfs_sumsize = LFS_V1_SUMMARY_SIZE;
1002 fs->lfs_ibsize = fs->lfs_bsize;
1003 fs->lfs_start = fs->lfs_sboffs[0];
1004 fs->lfs_tstamp = fs->lfs_otstamp;
1005 fs->lfs_fsbtodb = 0;
1006 }
1007
1008 /* Before rolling forward, lock so vget will sleep for other procs */
1009 fs->lfs_flags = LFS_NOTYET;
1010 fs->lfs_rfpid = p->p_pid;
1011
1012 ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
1013 ump->um_lfs = fs;
1014 ump->um_fstype = UFS1;
1015 if (sizeof(struct lfs) < LFS_SBPAD) { /* XXX why? */
1016 bp->b_flags |= B_INVAL;
1017 abp->b_flags |= B_INVAL;
1018 }
1019 brelse(bp);
1020 bp = NULL;
1021 brelse(abp);
1022 abp = NULL;
1023
1024 /* Set up the I/O information */
1025 fs->lfs_devbsize = secsize;
1026 fs->lfs_iocount = 0;
1027 fs->lfs_diropwait = 0;
1028 fs->lfs_activesb = 0;
1029 fs->lfs_uinodes = 0;
1030 fs->lfs_ravail = 0;
1031 fs->lfs_sbactive = 0;
1032
1033 /* Set up the ifile and lock aflags */
1034 fs->lfs_doifile = 0;
1035 fs->lfs_writer = 0;
1036 fs->lfs_dirops = 0;
1037 fs->lfs_nadirop = 0;
1038 fs->lfs_seglock = 0;
1039 fs->lfs_pdflush = 0;
1040 fs->lfs_sleepers = 0;
1041 simple_lock_init(&fs->lfs_interlock);
1042 lockinit(&fs->lfs_fraglock, PINOD, "lfs_fraglock", 0, 0);
1043
1044 /* Set the file system readonly/modify bits. */
1045 fs->lfs_ronly = ronly;
1046 if (ronly == 0)
1047 fs->lfs_fmod = 1;
1048
1049 /* Initialize the mount structure. */
1050 dev = devvp->v_rdev;
1051 mp->mnt_data = ump;
1052 mp->mnt_stat.f_fsid.val[0] = (long)dev;
1053 mp->mnt_stat.f_fsid.val[1] = makefstype(MOUNT_LFS);
1054 mp->mnt_stat.f_iosize = fs->lfs_bsize;
1055 mp->mnt_maxsymlinklen = fs->lfs_maxsymlinklen;
1056 mp->mnt_flag |= MNT_LOCAL;
1057 mp->mnt_fs_bshift = fs->lfs_bshift;
1058 ump->um_flags = 0;
1059 ump->um_mountp = mp;
1060 ump->um_dev = dev;
1061 ump->um_devvp = devvp;
1062 ump->um_bptrtodb = fs->lfs_fsbtodb;
1063 ump->um_seqinc = fragstofsb(fs, fs->lfs_frag);
1064 ump->um_nindir = fs->lfs_nindir;
1065 ump->um_lognindir = ffs(fs->lfs_nindir) - 1;
1066 for (i = 0; i < MAXQUOTAS; i++)
1067 ump->um_quotas[i] = NULLVP;
1068 devvp->v_specmountpoint = mp;
1069
1070 /* Set up reserved memory for pageout */
1071 lfs_setup_resblks(fs);
1072 /* Set up vdirop tailq */
1073 TAILQ_INIT(&fs->lfs_dchainhd);
1074 /* and paging tailq */
1075 TAILQ_INIT(&fs->lfs_pchainhd);
1076
1077 /*
1078 * We use the ifile vnode for almost every operation. Instead of
1079 * retrieving it from the hash table each time we retrieve it here,
1080 * artificially increment the reference count and keep a pointer
1081 * to it in the incore copy of the superblock.
1082 */
1083 if ((error = VFS_VGET(mp, LFS_IFILE_INUM, &vp)) != 0) {
1084 #ifdef DEBUG
1085 printf("lfs_mountfs: ifile vget failed, error=%d\n", error);
1086 #endif
1087 goto out;
1088 }
1089 fs->lfs_ivnode = vp;
1090 VREF(vp);
1091
1092 /* Set up segment usage flags for the autocleaner. */
1093 fs->lfs_nactive = 0;
1094 fs->lfs_suflags = (u_int32_t **)malloc(2 * sizeof(u_int32_t *),
1095 M_SEGMENT, M_WAITOK);
1096 fs->lfs_suflags[0] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1097 M_SEGMENT, M_WAITOK);
1098 fs->lfs_suflags[1] = (u_int32_t *)malloc(fs->lfs_nseg * sizeof(u_int32_t),
1099 M_SEGMENT, M_WAITOK);
1100 memset(fs->lfs_suflags[1], 0, fs->lfs_nseg * sizeof(u_int32_t));
1101 for (i = 0; i < fs->lfs_nseg; i++) {
1102 int changed;
1103
1104 LFS_SEGENTRY(sup, fs, i, bp);
1105 changed = 0;
1106 if (!ronly) {
1107 if (sup->su_nbytes == 0 &&
1108 !(sup->su_flags & SEGUSE_EMPTY)) {
1109 sup->su_flags |= SEGUSE_EMPTY;
1110 ++changed;
1111 } else if (!(sup->su_nbytes == 0) &&
1112 (sup->su_flags & SEGUSE_EMPTY)) {
1113 sup->su_flags &= ~SEGUSE_EMPTY;
1114 ++changed;
1115 }
1116 if (sup->su_flags & SEGUSE_ACTIVE) {
1117 sup->su_flags &= ~SEGUSE_ACTIVE;
1118 ++changed;
1119 }
1120 }
1121 fs->lfs_suflags[0][i] = sup->su_flags;
1122 if (changed)
1123 LFS_WRITESEGENTRY(sup, fs, i, bp);
1124 else
1125 brelse(bp);
1126 }
1127
1128 /*
1129 * Roll forward.
1130 *
1131 * We don't automatically roll forward for v1 filesystems, because
1132 * of the danger that the clock was turned back between the last
1133 * checkpoint and crash. This would roll forward garbage.
1134 *
1135 * v2 filesystems don't have this problem because they use a
1136 * monotonically increasing serial number instead of a timestamp.
1137 */
1138 #ifdef LFS_DO_ROLLFORWARD
1139 do_rollforward = !fs->lfs_ronly;
1140 #else
1141 do_rollforward = (fs->lfs_version > 1 && !fs->lfs_ronly &&
1142 !(fs->lfs_pflags & LFS_PF_CLEAN));
1143 #endif
1144 if (do_rollforward) {
1145 u_int64_t nextserial;
1146 /*
1147 * Phase I: Find the address of the last good partial
1148 * segment that was written after the checkpoint. Mark
1149 * the segments in question dirty, so they won't be
1150 * reallocated.
1151 */
1152 lastgoodpseg = oldoffset = offset = fs->lfs_offset;
1153 flags = 0x0;
1154 #ifdef DEBUG_LFS_RFW
1155 printf("LFS roll forward phase 1: starting at offset 0x%"
1156 PRIx64 "\n", offset);
1157 #endif
1158 LFS_SEGENTRY(sup, fs, dtosn(fs, offset), bp);
1159 if (!(sup->su_flags & SEGUSE_DIRTY))
1160 --fs->lfs_nclean;
1161 sup->su_flags |= SEGUSE_DIRTY;
1162 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, offset), bp);
1163 nextserial = fs->lfs_serial + 1;
1164 while ((offset = check_segsum(fs, offset, nextserial,
1165 cred, CHECK_CKSUM, &flags, p)) > 0) {
1166 nextserial++;
1167 if (sntod(fs, oldoffset) != sntod(fs, offset)) {
1168 LFS_SEGENTRY(sup, fs, dtosn(fs, oldoffset),
1169 bp);
1170 if (!(sup->su_flags & SEGUSE_DIRTY))
1171 --fs->lfs_nclean;
1172 sup->su_flags |= SEGUSE_DIRTY;
1173 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, oldoffset),
1174 bp);
1175 }
1176
1177 #ifdef DEBUG_LFS_RFW
1178 printf("LFS roll forward phase 1: offset=0x%"
1179 PRIx64 "\n", offset);
1180 if (flags & SS_DIROP) {
1181 printf("lfs_mountfs: dirops at 0x%" PRIx64 "\n",
1182 oldoffset);
1183 if (!(flags & SS_CONT))
1184 printf("lfs_mountfs: dirops end "
1185 "at 0x%" PRIx64 "\n", oldoffset);
1186 }
1187 #endif
1188 if (!(flags & SS_CONT))
1189 lastgoodpseg = offset;
1190 oldoffset = offset;
1191 }
1192 #ifdef DEBUG_LFS_RFW
1193 if (flags & SS_CONT) {
1194 printf("LFS roll forward: warning: incomplete "
1195 "dirops discarded\n");
1196 }
1197 printf("LFS roll forward phase 1: completed: "
1198 "lastgoodpseg=0x%" PRIx64 "\n", lastgoodpseg);
1199 #endif
1200 oldoffset = fs->lfs_offset;
1201 if (fs->lfs_offset != lastgoodpseg) {
1202 /* Don't overwrite what we're trying to preserve */
1203 offset = fs->lfs_offset;
1204 fs->lfs_offset = lastgoodpseg;
1205 fs->lfs_curseg = sntod(fs, dtosn(fs, fs->lfs_offset));
1206 for (sn = curseg = dtosn(fs, fs->lfs_curseg);;) {
1207 sn = (sn + 1) % fs->lfs_nseg;
1208 if (sn == curseg)
1209 panic("lfs_mountfs: no clean segments");
1210 LFS_SEGENTRY(sup, fs, sn, bp);
1211 dirty = (sup->su_flags & SEGUSE_DIRTY);
1212 brelse(bp);
1213 if (!dirty)
1214 break;
1215 }
1216 fs->lfs_nextseg = sntod(fs, sn);
1217
1218 /*
1219 * Phase II: Roll forward from the first superblock.
1220 */
1221 while (offset != lastgoodpseg) {
1222 #ifdef DEBUG_LFS_RFW
1223 printf("LFS roll forward phase 2: 0x%"
1224 PRIx64 "\n", offset);
1225 #endif
1226 offset = check_segsum(fs, offset,
1227 fs->lfs_serial + 1, cred, CHECK_UPDATE,
1228 NULL, p);
1229 }
1230
1231 /*
1232 * Finish: flush our changes to disk.
1233 */
1234 lfs_segwrite(mp, SEGM_CKP | SEGM_SYNC);
1235 printf("lfs_mountfs: roll forward recovered %lld blocks\n",
1236 (long long)(lastgoodpseg - oldoffset));
1237 }
1238 #ifdef DEBUG_LFS_RFW
1239 printf("LFS roll forward complete\n");
1240 #endif
1241 }
1242 /* If writing, sb is not clean; record in case of immediate crash */
1243 if (!fs->lfs_ronly) {
1244 fs->lfs_pflags &= ~LFS_PF_CLEAN;
1245 lfs_writesuper(fs, fs->lfs_sboffs[0]);
1246 lfs_writesuper(fs, fs->lfs_sboffs[1]);
1247 }
1248
1249 /* Allow vget now that roll-forward is complete */
1250 fs->lfs_flags &= ~(LFS_NOTYET);
1251 wakeup(&fs->lfs_flags);
1252
1253 /*
1254 * Initialize the ifile cleaner info with information from
1255 * the superblock.
1256 */
1257 LFS_CLEANERINFO(cip, fs, bp);
1258 cip->clean = fs->lfs_nclean;
1259 cip->dirty = fs->lfs_nseg - fs->lfs_nclean;
1260 cip->avail = fs->lfs_avail;
1261 cip->bfree = fs->lfs_bfree;
1262 (void) LFS_BWRITE_LOG(bp); /* Ifile */
1263
1264 /*
1265 * Mark the current segment as ACTIVE, since we're going to
1266 * be writing to it.
1267 */
1268 LFS_SEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp);
1269 sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE;
1270 fs->lfs_nactive++;
1271 LFS_WRITESEGENTRY(sup, fs, dtosn(fs, fs->lfs_offset), bp); /* Ifile */
1272
1273 /* Now that roll-forward is done, unlock the Ifile */
1274 vput(vp);
1275
1276 /* Comment on ifile size if it is too large */
1277 if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS) {
1278 fs->lfs_flags |= LFS_WARNED;
1279 printf("lfs_mountfs: please consider increasing NBUF to at least %lld\n",
1280 (long long)(fs->lfs_ivnode->v_size / fs->lfs_bsize) * (nbuf / LFS_MAX_BUFS));
1281 }
1282 if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES) {
1283 fs->lfs_flags |= LFS_WARNED;
1284 printf("lfs_mountfs: please consider increasing BUFPAGES to at least %lld\n",
1285 (long long)(fs->lfs_ivnode->v_size * bufpages / LFS_MAX_BYTES));
1286 }
1287
1288 return (0);
1289 out:
1290 if (bp)
1291 brelse(bp);
1292 if (abp)
1293 brelse(abp);
1294 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1295 (void)VOP_CLOSE(devvp, ronly ? FREAD : FREAD|FWRITE, cred, p);
1296 VOP_UNLOCK(devvp, 0);
1297 if (ump) {
1298 free(ump->um_lfs, M_UFSMNT);
1299 free(ump, M_UFSMNT);
1300 mp->mnt_data = NULL;
1301 }
1302
1303 /* Start the pagedaemon-anticipating daemon */
1304 if (lfs_writer_daemon == 0 &&
1305 kthread_create1(lfs_writerd, NULL, NULL, "lfs_writer") != 0)
1306 panic("fork lfs_writer");
1307
1308 return (error);
1309 }
1310
1311 /*
1312 * unmount system call
1313 */
1314 int
1315 lfs_unmount(struct mount *mp, int mntflags, struct proc *p)
1316 {
1317 struct ufsmount *ump;
1318 struct lfs *fs;
1319 int error, flags, ronly;
1320 int s;
1321
1322 flags = 0;
1323 if (mntflags & MNT_FORCE)
1324 flags |= FORCECLOSE;
1325
1326 ump = VFSTOUFS(mp);
1327 fs = ump->um_lfs;
1328
1329 /* wake up the cleaner so it can die */
1330 wakeup(&fs->lfs_nextseg);
1331 wakeup(&lfs_allclean_wakeup);
1332 simple_lock(&fs->lfs_interlock);
1333 while (fs->lfs_sleepers)
1334 ltsleep(&fs->lfs_sleepers, PRIBIO + 1, "lfs_sleepers", 0,
1335 &fs->lfs_interlock);
1336 simple_unlock(&fs->lfs_interlock);
1337
1338 #ifdef QUOTA
1339 if (mp->mnt_flag & MNT_QUOTA) {
1340 int i;
1341 error = vflush(mp, fs->lfs_ivnode, SKIPSYSTEM|flags);
1342 if (error)
1343 return (error);
1344 for (i = 0; i < MAXQUOTAS; i++) {
1345 if (ump->um_quotas[i] == NULLVP)
1346 continue;
1347 quotaoff(p, mp, i);
1348 }
1349 /*
1350 * Here we fall through to vflush again to ensure
1351 * that we have gotten rid of all the system vnodes.
1352 */
1353 }
1354 #endif
1355 if ((error = vflush(mp, fs->lfs_ivnode, flags)) != 0)
1356 return (error);
1357 if ((error = VFS_SYNC(mp, 1, p->p_ucred, p)) != 0)
1358 return (error);
1359 s = splbio();
1360 if (LIST_FIRST(&fs->lfs_ivnode->v_dirtyblkhd))
1361 panic("lfs_unmount: still dirty blocks on ifile vnode");
1362 splx(s);
1363
1364 /* Comment on ifile size if it has become too large */
1365 if (!(fs->lfs_flags & LFS_WARNED)) {
1366 if (fs->lfs_ivnode->v_size / fs->lfs_bsize > LFS_MAX_BUFS)
1367 printf("lfs_unmount: please consider increasing"
1368 " NBUF to at least %lld\n",
1369 (long long)(fs->lfs_ivnode->v_size /
1370 fs->lfs_bsize) *
1371 (long long)(nbuf / LFS_MAX_BUFS));
1372 if (fs->lfs_ivnode->v_size > LFS_MAX_BYTES)
1373 printf("lfs_unmount: please consider increasing"
1374 " BUFPAGES to at least %lld\n",
1375 (long long)(fs->lfs_ivnode->v_size *
1376 bufpages / LFS_MAX_BYTES));
1377 }
1378
1379 /* Explicitly write the superblock, to update serial and pflags */
1380 fs->lfs_pflags |= LFS_PF_CLEAN;
1381 lfs_writesuper(fs, fs->lfs_sboffs[0]);
1382 lfs_writesuper(fs, fs->lfs_sboffs[1]);
1383 while (fs->lfs_iocount)
1384 tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs_umount", 0);
1385
1386 /* Finish with the Ifile, now that we're done with it */
1387 vrele(fs->lfs_ivnode);
1388 vgone(fs->lfs_ivnode);
1389
1390 ronly = !fs->lfs_ronly;
1391 if (ump->um_devvp->v_type != VBAD)
1392 ump->um_devvp->v_specmountpoint = NULL;
1393 vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1394 error = VOP_CLOSE(ump->um_devvp,
1395 ronly ? FREAD : FREAD|FWRITE, NOCRED, p);
1396 vput(ump->um_devvp);
1397
1398 /* Free per-mount data structures */
1399 free(fs->lfs_suflags[0], M_SEGMENT);
1400 free(fs->lfs_suflags[1], M_SEGMENT);
1401 free(fs->lfs_suflags, M_SEGMENT);
1402 lfs_free_resblks(fs);
1403 free(fs, M_UFSMNT);
1404 free(ump, M_UFSMNT);
1405
1406 mp->mnt_data = NULL;
1407 mp->mnt_flag &= ~MNT_LOCAL;
1408 return (error);
1409 }
1410
1411 /*
1412 * Get file system statistics.
1413 */
1414 int
1415 lfs_statfs(struct mount *mp, struct statfs *sbp, struct proc *p)
1416 {
1417 struct lfs *fs;
1418 struct ufsmount *ump;
1419
1420 ump = VFSTOUFS(mp);
1421 fs = ump->um_lfs;
1422 if (fs->lfs_magic != LFS_MAGIC)
1423 panic("lfs_statfs: magic");
1424
1425 sbp->f_type = 0;
1426 sbp->f_bsize = fs->lfs_fsize;
1427 sbp->f_iosize = fs->lfs_bsize;
1428 sbp->f_blocks = fsbtofrags(fs, LFS_EST_NONMETA(fs));
1429 sbp->f_bfree = fsbtofrags(fs, LFS_EST_BFREE(fs));
1430 sbp->f_bavail = fsbtofrags(fs, (long)LFS_EST_BFREE(fs) -
1431 (long)LFS_EST_RSVD(fs));
1432
1433 sbp->f_files = fs->lfs_bfree / btofsb(fs, fs->lfs_ibsize) * INOPB(fs);
1434 sbp->f_ffree = sbp->f_files - fs->lfs_nfiles;
1435 copy_statfs_info(sbp, mp);
1436 return (0);
1437 }
1438
1439 /*
1440 * Go through the disk queues to initiate sandbagged IO;
1441 * go through the inodes to write those that have been modified;
1442 * initiate the writing of the super block if it has been modified.
1443 *
1444 * Note: we are always called with the filesystem marked `MPBUSY'.
1445 */
1446 int
1447 lfs_sync(struct mount *mp, int waitfor, struct ucred *cred, struct proc *p)
1448 {
1449 int error;
1450 struct lfs *fs;
1451
1452 fs = VFSTOUFS(mp)->um_lfs;
1453 if (fs->lfs_ronly)
1454 return 0;
1455 lfs_writer_enter(fs, "lfs_dirops");
1456
1457 /* All syncs must be checkpoints until roll-forward is implemented. */
1458 error = lfs_segwrite(mp, SEGM_CKP | (waitfor ? SEGM_SYNC : 0));
1459 lfs_writer_leave(fs);
1460 #ifdef QUOTA
1461 qsync(mp);
1462 #endif
1463 return (error);
1464 }
1465
1466 extern struct lock ufs_hashlock;
1467
1468 /*
1469 * Look up an LFS dinode number to find its incore vnode. If not already
1470 * in core, read it in from the specified device. Return the inode locked.
1471 * Detection and handling of mount points must be done by the calling routine.
1472 */
1473 int
1474 lfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
1475 {
1476 struct lfs *fs;
1477 struct ufs1_dinode *dip;
1478 struct inode *ip;
1479 struct buf *bp;
1480 struct ifile *ifp;
1481 struct vnode *vp;
1482 struct ufsmount *ump;
1483 daddr_t daddr;
1484 dev_t dev;
1485 int error, retries;
1486 struct timespec ts;
1487
1488 ump = VFSTOUFS(mp);
1489 dev = ump->um_dev;
1490 fs = ump->um_lfs;
1491
1492 /*
1493 * If the filesystem is not completely mounted yet, suspend
1494 * any access requests (wait for roll-forward to complete).
1495 */
1496 while ((fs->lfs_flags & LFS_NOTYET) && curproc->p_pid != fs->lfs_rfpid)
1497 tsleep(&fs->lfs_flags, PRIBIO+1, "lfs_notyet", 0);
1498
1499 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL)
1500 return (0);
1501
1502 if ((error = getnewvnode(VT_LFS, mp, lfs_vnodeop_p, &vp)) != 0) {
1503 *vpp = NULL;
1504 return (error);
1505 }
1506
1507 do {
1508 if ((*vpp = ufs_ihashget(dev, ino, LK_EXCLUSIVE)) != NULL) {
1509 ungetnewvnode(vp);
1510 return (0);
1511 }
1512 } while (lockmgr(&ufs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
1513
1514 /* Translate the inode number to a disk address. */
1515 if (ino == LFS_IFILE_INUM)
1516 daddr = fs->lfs_idaddr;
1517 else {
1518 /* XXX bounds-check this too */
1519 LFS_IENTRY(ifp, fs, ino, bp);
1520 daddr = ifp->if_daddr;
1521 if (fs->lfs_version > 1) {
1522 ts.tv_sec = ifp->if_atime_sec;
1523 ts.tv_nsec = ifp->if_atime_nsec;
1524 }
1525
1526 brelse(bp);
1527 if (daddr == LFS_UNUSED_DADDR) {
1528 *vpp = NULLVP;
1529 ungetnewvnode(vp);
1530 lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1531 return (ENOENT);
1532 }
1533 }
1534
1535 /* Allocate/init new vnode/inode. */
1536 lfs_vcreate(mp, ino, vp);
1537
1538 /*
1539 * Put it onto its hash chain and lock it so that other requests for
1540 * this inode will block if they arrive while we are sleeping waiting
1541 * for old data structures to be purged or for the contents of the
1542 * disk portion of this inode to be read.
1543 */
1544 ip = VTOI(vp);
1545 ufs_ihashins(ip);
1546 lockmgr(&ufs_hashlock, LK_RELEASE, 0);
1547
1548 /*
1549 * XXX
1550 * This may not need to be here, logically it should go down with
1551 * the i_devvp initialization.
1552 * Ask Kirk.
1553 */
1554 ip->i_lfs = ump->um_lfs;
1555
1556 /* Read in the disk contents for the inode, copy into the inode. */
1557 retries = 0;
1558 again:
1559 error = bread(ump->um_devvp, fsbtodb(fs, daddr),
1560 (fs->lfs_version == 1 ? fs->lfs_bsize : fs->lfs_ibsize),
1561 NOCRED, &bp);
1562 if (error) {
1563 /*
1564 * The inode does not contain anything useful, so it would
1565 * be misleading to leave it on its hash chain. With mode
1566 * still zero, it will be unlinked and returned to the free
1567 * list by vput().
1568 */
1569 vput(vp);
1570 brelse(bp);
1571 *vpp = NULL;
1572 return (error);
1573 }
1574
1575 dip = lfs_ifind(fs, ino, bp);
1576 if (dip == NULL) {
1577 /* Assume write has not completed yet; try again */
1578 bp->b_flags |= B_INVAL;
1579 brelse(bp);
1580 ++retries;
1581 if (retries > LFS_IFIND_RETRIES) {
1582 #ifdef DEBUG
1583 /* If the seglock is held look at the bpp to see
1584 what is there anyway */
1585 if (fs->lfs_seglock > 0) {
1586 struct buf **bpp;
1587 struct ufs1_dinode *dp;
1588 int i;
1589
1590 for (bpp = fs->lfs_sp->bpp;
1591 bpp != fs->lfs_sp->cbpp; ++bpp) {
1592 if ((*bpp)->b_vp == fs->lfs_ivnode &&
1593 bpp != fs->lfs_sp->bpp) {
1594 /* Inode block */
1595 printf("block 0x%" PRIx64 ": ",
1596 (*bpp)->b_blkno);
1597 dp = (struct ufs1_dinode *)(*bpp)->b_data;
1598 for (i = 0; i < INOPB(fs); i++)
1599 if (dp[i].di_u.inumber)
1600 printf("%d ", dp[i].di_u.inumber);
1601 printf("\n");
1602 }
1603 }
1604 }
1605 #endif
1606 panic("lfs_vget: dinode not found");
1607 }
1608 printf("lfs_vget: dinode %d not found, retrying...\n", ino);
1609 (void)tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs ifind", 1);
1610 goto again;
1611 }
1612 *ip->i_din.ffs1_din = *dip;
1613 brelse(bp);
1614
1615 if (fs->lfs_version > 1) {
1616 ip->i_ffs1_atime = ts.tv_sec;
1617 ip->i_ffs1_atimensec = ts.tv_nsec;
1618 }
1619
1620 lfs_vinit(mp, &vp);
1621
1622 *vpp = vp;
1623
1624 KASSERT(VOP_ISLOCKED(vp));
1625
1626 return (0);
1627 }
1628
1629 /*
1630 * File handle to vnode
1631 */
1632 int
1633 lfs_fhtovp(struct mount *mp, struct fid *fhp, struct vnode **vpp)
1634 {
1635 struct lfid *lfhp;
1636 struct buf *bp;
1637 IFILE *ifp;
1638 int32_t daddr;
1639 struct lfs *fs;
1640
1641 lfhp = (struct lfid *)fhp;
1642 if (lfhp->lfid_ino < LFS_IFILE_INUM)
1643 return ESTALE;
1644
1645 fs = VFSTOUFS(mp)->um_lfs;
1646 if (lfhp->lfid_ident != fs->lfs_ident)
1647 return ESTALE;
1648
1649 if (lfhp->lfid_ino >
1650 ((VTOI(fs->lfs_ivnode)->i_ffs1_size >> fs->lfs_bshift) -
1651 fs->lfs_cleansz - fs->lfs_segtabsz) * fs->lfs_ifpb)
1652 return ESTALE;
1653
1654 if (ufs_ihashlookup(VFSTOUFS(mp)->um_dev, lfhp->lfid_ino) == NULLVP) {
1655 LFS_IENTRY(ifp, fs, lfhp->lfid_ino, bp);
1656 daddr = ifp->if_daddr;
1657 brelse(bp);
1658 if (daddr == LFS_UNUSED_DADDR)
1659 return ESTALE;
1660 }
1661
1662 return (ufs_fhtovp(mp, &lfhp->lfid_ufid, vpp));
1663 }
1664
1665 /*
1666 * Vnode pointer to File handle
1667 */
1668 /* ARGSUSED */
1669 int
1670 lfs_vptofh(struct vnode *vp, struct fid *fhp)
1671 {
1672 struct inode *ip;
1673 struct lfid *lfhp;
1674
1675 ip = VTOI(vp);
1676 lfhp = (struct lfid *)fhp;
1677 lfhp->lfid_len = sizeof(struct lfid);
1678 lfhp->lfid_ino = ip->i_number;
1679 lfhp->lfid_gen = ip->i_gen;
1680 lfhp->lfid_ident = ip->i_lfs->lfs_ident;
1681 return (0);
1682 }
1683
1684 static int
1685 sysctl_lfs_dostats(SYSCTLFN_ARGS)
1686 {
1687 extern struct lfs_stats lfs_stats;
1688 extern int lfs_dostats;
1689 int error;
1690
1691 error = sysctl_lookup(SYSCTLFN_CALL(rnode));
1692 if (error || newp == NULL)
1693 return (error);
1694
1695 if (lfs_dostats == 0)
1696 memset(&lfs_stats,0,sizeof(lfs_stats));
1697
1698 return (0);
1699 }
1700
1701 SYSCTL_SETUP(sysctl_vfs_lfs_setup, "sysctl vfs.lfs subtree setup")
1702 {
1703 extern int lfs_writeindir, lfs_dostats, lfs_clean_vnhead;
1704
1705 sysctl_createv(clog, 0, NULL, NULL,
1706 CTLFLAG_PERMANENT,
1707 CTLTYPE_NODE, "vfs", NULL,
1708 NULL, 0, NULL, 0,
1709 CTL_VFS, CTL_EOL);
1710 sysctl_createv(clog, 0, NULL, NULL,
1711 CTLFLAG_PERMANENT,
1712 CTLTYPE_NODE, "lfs",
1713 SYSCTL_DESCR("Log-structured file system"),
1714 NULL, 0, NULL, 0,
1715 CTL_VFS, 5, CTL_EOL);
1716 /*
1717 * XXX the "5" above could be dynamic, thereby eliminating one
1718 * more instance of the "number to vfs" mapping problem, but
1719 * "2" is the order as taken from sys/mount.h
1720 */
1721
1722 sysctl_createv(clog, 0, NULL, NULL,
1723 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1724 CTLTYPE_INT, "flushindir", NULL,
1725 NULL, 0, &lfs_writeindir, 0,
1726 CTL_VFS, 5, LFS_WRITEINDIR, CTL_EOL);
1727 sysctl_createv(clog, 0, NULL, NULL,
1728 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1729 CTLTYPE_INT, "clean_vnhead", NULL,
1730 NULL, 0, &lfs_clean_vnhead, 0,
1731 CTL_VFS, 5, LFS_CLEAN_VNHEAD, CTL_EOL);
1732 sysctl_createv(clog, 0, NULL, NULL,
1733 CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
1734 CTLTYPE_INT, "dostats",
1735 SYSCTL_DESCR("Maintain statistics on LFS operations"),
1736 sysctl_lfs_dostats, 0, &lfs_dostats, 0,
1737 CTL_VFS, 5, LFS_DOSTATS, CTL_EOL);
1738 }
1739
1740 /*
1741 * ufs_bmaparray callback function for writing.
1742 *
1743 * Since blocks will be written to the new segment anyway,
1744 * we don't care about current daddr of them.
1745 */
1746 static boolean_t
1747 lfs_issequential_hole(const struct ufsmount *ump,
1748 daddr_t daddr0, daddr_t daddr1)
1749 {
1750
1751 KASSERT(daddr0 == UNWRITTEN ||
1752 (0 <= daddr0 && daddr0 <= LFS_MAX_DADDR));
1753 KASSERT(daddr1 == UNWRITTEN ||
1754 (0 <= daddr1 && daddr1 <= LFS_MAX_DADDR));
1755
1756 /* NOTE: all we want to know here is 'hole or not'. */
1757 /* NOTE: UNASSIGNED is converted to 0 by ufs_bmaparray. */
1758
1759 /*
1760 * treat UNWRITTENs and all resident blocks as 'contiguous'
1761 */
1762 if (daddr0 != 0 && daddr1 != 0)
1763 return TRUE;
1764
1765 /*
1766 * both are in hole?
1767 */
1768 if (daddr0 == 0 && daddr1 == 0)
1769 return TRUE; /* all holes are 'contiguous' for us. */
1770
1771 return FALSE;
1772 }
1773
1774 /*
1775 * lfs_gop_write functions exactly like genfs_gop_write, except that
1776 * (1) it requires the seglock to be held by its caller, and sp->fip
1777 * to be properly initialized (it will return without re-initializing
1778 * sp->fip, and without calling lfs_writeseg).
1779 * (2) it uses the remaining space in the segment, rather than VOP_BMAP,
1780 * to determine how large a block it can write at once (though it does
1781 * still use VOP_BMAP to find holes in the file);
1782 * (3) it calls lfs_gatherblock instead of VOP_STRATEGY on its blocks
1783 * (leaving lfs_writeseg to deal with the cluster blocks, so we might
1784 * now have clusters of clusters, ick.)
1785 */
1786 static int
1787 lfs_gop_write(struct vnode *vp, struct vm_page **pgs, int npages, int flags)
1788 {
1789 int i, s, error, run;
1790 int fs_bshift;
1791 vaddr_t kva;
1792 off_t eof, offset, startoffset;
1793 size_t bytes, iobytes, skipbytes;
1794 daddr_t lbn, blkno;
1795 struct vm_page *pg;
1796 struct buf *mbp, *bp;
1797 struct vnode *devvp = VTOI(vp)->i_devvp;
1798 struct inode *ip = VTOI(vp);
1799 struct lfs *fs = ip->i_lfs;
1800 struct segment *sp = fs->lfs_sp;
1801 UVMHIST_FUNC("lfs_gop_write"); UVMHIST_CALLED(ubchist);
1802
1803 /* The Ifile lives in the buffer cache */
1804 if (vp == fs->lfs_ivnode)
1805 return genfs_compat_gop_write(vp, pgs, npages, flags);
1806
1807 /*
1808 * Sometimes things slip past the filters in lfs_putpages,
1809 * and the pagedaemon tries to write pages---problem is
1810 * that the pagedaemon never acquires the segment lock.
1811 *
1812 * Unbusy and unclean the pages, and put them on the ACTIVE
1813 * queue under the hypothesis that they couldn't have got here
1814 * unless they were modified *quite* recently.
1815 *
1816 * XXXUBC that last statement is an oversimplification of course.
1817 */
1818 if (!(fs->lfs_seglock) || fs->lfs_lockpid != curproc->p_pid) {
1819 simple_lock(&vp->v_interlock);
1820 #ifdef DEBUG
1821 printf("lfs_gop_write: seglock not held\n");
1822 #endif
1823 uvm_lock_pageq();
1824 for (i = 0; i < npages; i++) {
1825 pg = pgs[i];
1826
1827 if (pg->flags & PG_PAGEOUT)
1828 uvmexp.paging--;
1829 if (pg->flags & PG_DELWRI) {
1830 uvm_pageunwire(pg);
1831 }
1832 uvm_pageactivate(pg);
1833 pg->flags &= ~(PG_CLEAN|PG_DELWRI|PG_PAGEOUT|PG_RELEASED);
1834 #ifdef DEBUG_LFS
1835 printf("pg[%d]->flags = %x\n", i, pg->flags);
1836 printf("pg[%d]->pqflags = %x\n", i, pg->pqflags);
1837 printf("pg[%d]->uanon = %p\n", i, pg->uanon);
1838 printf("pg[%d]->uobject = %p\n", i, pg->uobject);
1839 printf("pg[%d]->wire_count = %d\n", i, pg->wire_count);
1840 printf("pg[%d]->loan_count = %d\n", i, pg->loan_count);
1841 #endif
1842 }
1843 /* uvm_pageunbusy takes care of PG_BUSY, PG_WANTED */
1844 uvm_page_unbusy(pgs, npages);
1845 uvm_unlock_pageq();
1846 simple_unlock(&vp->v_interlock);
1847 return EAGAIN;
1848 }
1849
1850 UVMHIST_LOG(ubchist, "vp %p pgs %p npages %d flags 0x%x",
1851 vp, pgs, npages, flags);
1852
1853 GOP_SIZE(vp, vp->v_size, &eof, GOP_SIZE_WRITE);
1854
1855 if (vp->v_type == VREG)
1856 fs_bshift = vp->v_mount->mnt_fs_bshift;
1857 else
1858 fs_bshift = DEV_BSHIFT;
1859 error = 0;
1860 pg = pgs[0];
1861 startoffset = pg->offset;
1862 bytes = MIN(npages << PAGE_SHIFT, eof - startoffset);
1863 skipbytes = 0;
1864
1865 /* KASSERT(bytes != 0); */
1866 if (bytes == 0)
1867 printf("ino %d bytes == 0 offset %" PRId64 "\n",
1868 VTOI(vp)->i_number, pgs[0]->offset);
1869
1870 /* Swap PG_DELWRI for PG_PAGEOUT */
1871 for (i = 0; i < npages; i++)
1872 if (pgs[i]->flags & PG_DELWRI) {
1873 KASSERT(!(pgs[i]->flags & PG_PAGEOUT));
1874 pgs[i]->flags &= ~PG_DELWRI;
1875 pgs[i]->flags |= PG_PAGEOUT;
1876 uvmexp.paging++;
1877 uvm_lock_pageq();
1878 uvm_pageunwire(pgs[i]);
1879 uvm_unlock_pageq();
1880 }
1881
1882 /*
1883 * Check to make sure we're starting on a block boundary.
1884 * We'll check later to make sure we always write entire
1885 * blocks (or fragments).
1886 */
1887 if (startoffset & fs->lfs_bmask)
1888 printf("%" PRId64 " & %" PRId64 " = %" PRId64 "\n",
1889 startoffset, fs->lfs_bmask,
1890 startoffset & fs->lfs_bmask);
1891 KASSERT((startoffset & fs->lfs_bmask) == 0);
1892 if (bytes & fs->lfs_ffmask) {
1893 printf("lfs_gop_write: asked to write %ld bytes\n", (long)bytes);
1894 panic("lfs_gop_write: non-integer blocks");
1895 }
1896
1897 kva = uvm_pagermapin(pgs, npages,
1898 UVMPAGER_MAPIN_WRITE | UVMPAGER_MAPIN_WAITOK);
1899
1900 s = splbio();
1901 simple_lock(&global_v_numoutput_slock);
1902 vp->v_numoutput += 2; /* one for biodone, one for aiodone */
1903 simple_unlock(&global_v_numoutput_slock);
1904 mbp = pool_get(&bufpool, PR_WAITOK);
1905 splx(s);
1906
1907 memset(mbp, 0, sizeof(*bp));
1908 BUF_INIT(mbp);
1909 UVMHIST_LOG(ubchist, "vp %p mbp %p num now %d bytes 0x%x",
1910 vp, mbp, vp->v_numoutput, bytes);
1911 mbp->b_bufsize = npages << PAGE_SHIFT;
1912 mbp->b_data = (void *)kva;
1913 mbp->b_resid = mbp->b_bcount = bytes;
1914 mbp->b_flags = B_BUSY|B_WRITE|B_AGE|B_CALL;
1915 mbp->b_iodone = uvm_aio_biodone;
1916 mbp->b_vp = vp;
1917
1918 bp = NULL;
1919 for (offset = startoffset;
1920 bytes > 0;
1921 offset += iobytes, bytes -= iobytes) {
1922 lbn = offset >> fs_bshift;
1923 error = ufs_bmaparray(vp, lbn, &blkno, NULL, NULL, &run,
1924 lfs_issequential_hole);
1925 if (error) {
1926 UVMHIST_LOG(ubchist, "ufs_bmaparray() -> %d",
1927 error,0,0,0);
1928 skipbytes += bytes;
1929 bytes = 0;
1930 break;
1931 }
1932
1933 iobytes = MIN((((off_t)lbn + 1 + run) << fs_bshift) - offset,
1934 bytes);
1935 if (blkno == (daddr_t)-1) {
1936 skipbytes += iobytes;
1937 continue;
1938 }
1939
1940 /*
1941 * Discover how much we can really pack into this buffer.
1942 */
1943 /* If no room in the current segment, finish it up */
1944 if (sp->sum_bytes_left < sizeof(int32_t) ||
1945 sp->seg_bytes_left < (1 << fs->lfs_bshift)) {
1946 int version;
1947
1948 lfs_updatemeta(sp);
1949
1950 version = sp->fip->fi_version;
1951 (void) lfs_writeseg(fs, sp);
1952
1953 sp->fip->fi_version = version;
1954 sp->fip->fi_ino = ip->i_number;
1955 /* Add the current file to the segment summary. */
1956 ++((SEGSUM *)(sp->segsum))->ss_nfinfo;
1957 sp->sum_bytes_left -= FINFOSIZE;
1958 }
1959 /* Check both for space in segment and space in segsum */
1960 iobytes = MIN(iobytes, (sp->seg_bytes_left >> fs_bshift)
1961 << fs_bshift);
1962 iobytes = MIN(iobytes, (sp->sum_bytes_left / sizeof(int32_t))
1963 << fs_bshift);
1964 KASSERT(iobytes > 0);
1965
1966 /* if it's really one i/o, don't make a second buf */
1967 if (offset == startoffset && iobytes == bytes) {
1968 bp = mbp;
1969 /* printf("bp is mbp\n"); */
1970 /* correct overcount if there is no second buffer */
1971 s = splbio();
1972 simple_lock(&global_v_numoutput_slock);
1973 --vp->v_numoutput;
1974 simple_unlock(&global_v_numoutput_slock);
1975 splx(s);
1976 } else {
1977 /* printf("bp is not mbp\n"); */
1978 s = splbio();
1979 bp = pool_get(&bufpool, PR_WAITOK);
1980 UVMHIST_LOG(ubchist, "vp %p bp %p num now %d",
1981 vp, bp, vp->v_numoutput, 0);
1982 splx(s);
1983 memset(bp, 0, sizeof(*bp));
1984 BUF_INIT(bp);
1985 bp->b_data = (char *)kva +
1986 (vaddr_t)(offset - pg->offset);
1987 bp->b_resid = bp->b_bcount = iobytes;
1988 bp->b_flags = B_BUSY|B_WRITE|B_CALL;
1989 bp->b_iodone = uvm_aio_biodone1;
1990 }
1991
1992 /* XXX This is silly ... is this necessary? */
1993 bp->b_vp = NULL;
1994 s = splbio();
1995 bgetvp(vp, bp);
1996 splx(s);
1997
1998 bp->b_lblkno = lblkno(fs, offset);
1999 bp->b_private = mbp;
2000 if (devvp->v_type == VBLK) {
2001 bp->b_dev = devvp->v_rdev;
2002 }
2003 VOP_BWRITE(bp);
2004 while (lfs_gatherblock(sp, bp, NULL))
2005 continue;
2006 }
2007
2008 if (skipbytes) {
2009 UVMHIST_LOG(ubchist, "skipbytes %d", skipbytes, 0,0,0);
2010 s = splbio();
2011 if (error) {
2012 mbp->b_flags |= B_ERROR;
2013 mbp->b_error = error;
2014 }
2015 mbp->b_resid -= skipbytes;
2016 if (mbp->b_resid == 0) {
2017 biodone(mbp);
2018 }
2019 splx(s);
2020 }
2021 UVMHIST_LOG(ubchist, "returning 0", 0,0,0,0);
2022 return (0);
2023 }
2024
2025 /*
2026 * finish vnode/inode initialization.
2027 * used by lfs_vget and lfs_fastvget.
2028 */
2029 void
2030 lfs_vinit(struct mount *mp, struct vnode **vpp)
2031 {
2032 struct vnode *vp = *vpp;
2033 struct inode *ip = VTOI(vp);
2034 struct ufsmount *ump = VFSTOUFS(mp);
2035 int i;
2036
2037 ip->i_mode = ip->i_ffs1_mode;
2038 ip->i_ffs_effnlink = ip->i_nlink = ip->i_ffs1_nlink;
2039 ip->i_lfs_osize = ip->i_size = ip->i_ffs1_size;
2040 ip->i_flags = ip->i_ffs1_flags;
2041 ip->i_gen = ip->i_ffs1_gen;
2042 ip->i_uid = ip->i_ffs1_uid;
2043 ip->i_gid = ip->i_ffs1_gid;
2044
2045 ip->i_lfs_effnblks = ip->i_ffs1_blocks;
2046
2047 /*
2048 * Initialize the vnode from the inode, check for aliases. In all
2049 * cases re-init ip, the underlying vnode/inode may have changed.
2050 */
2051 ufs_vinit(mp, lfs_specop_p, lfs_fifoop_p, &vp);
2052
2053 memset(ip->i_lfs_fragsize, 0, NDADDR * sizeof(*ip->i_lfs_fragsize));
2054 if (vp->v_type != VLNK ||
2055 VTOI(vp)->i_size >= vp->v_mount->mnt_maxsymlinklen) {
2056 struct lfs *fs = ump->um_lfs;
2057 #ifdef DEBUG
2058 for (i = (ip->i_size + fs->lfs_bsize - 1) >> fs->lfs_bshift;
2059 i < NDADDR; i++) {
2060 if (ip->i_ffs1_db[i] != 0) {
2061 inconsistent:
2062 lfs_dump_dinode(ip->i_din.ffs1_din);
2063 panic("inconsistent inode");
2064 }
2065 }
2066 for ( ; i < NDADDR + NIADDR; i++) {
2067 if (ip->i_ffs1_ib[i - NDADDR] != 0) {
2068 goto inconsistent;
2069 }
2070 }
2071 #endif /* DEBUG */
2072 for (i = 0; i < NDADDR; i++)
2073 if (ip->i_ffs1_db[i] != 0)
2074 ip->i_lfs_fragsize[i] = blksize(fs, ip, i);
2075 }
2076
2077 #ifdef DEBUG
2078 if (vp->v_type == VNON) {
2079 printf("lfs_vinit: ino %d is type VNON! (ifmt=%o)\n",
2080 ip->i_number, (ip->i_mode & IFMT) >> 12);
2081 lfs_dump_dinode(ip->i_din.ffs1_din);
2082 #ifdef DDB
2083 Debugger();
2084 #endif /* DDB */
2085 }
2086 #endif /* DEBUG */
2087
2088 /*
2089 * Finish inode initialization now that aliasing has been resolved.
2090 */
2091
2092 ip->i_devvp = ump->um_devvp;
2093 VREF(ip->i_devvp);
2094 genfs_node_init(vp, &lfs_genfsops);
2095 uvm_vnp_setsize(vp, ip->i_size);
2096
2097 *vpp = vp;
2098 }
Cache object: 505194fd78716f2d8448d246c599e25f
|