1/* $NetBSD: ulfs_readwrite.c,v 1.22 2016/06/20 03:36:09 dholland Exp $ */
2/* from NetBSD: ufs_readwrite.c,v 1.120 2015/04/12 22:48:38 riastradh Exp */
3
4/*-
5 * Copyright (c) 1993
6 * The Regents of the University of California. All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)ufs_readwrite.c 8.11 (Berkeley) 5/8/95
33 */
34
35#include <sys/cdefs.h>
36__KERNEL_RCSID(1, "$NetBSD: ulfs_readwrite.c,v 1.22 2016/06/20 03:36:09 dholland Exp $");
37
38#ifdef LFS_READWRITE
39#define FS struct lfs
40#define I_FS i_lfs
41#define READ lfs_read
42#define READ_S "lfs_read"
43#define WRITE lfs_write
44#define WRITE_S "lfs_write"
45#define BUFRD lfs_bufrd
46#define BUFWR lfs_bufwr
47#define fs_sb_getbsize(fs) lfs_sb_getbsize(fs)
48#define fs_bmask lfs_bmask
49#else
50#define FS struct fs
51#define I_FS i_fs
52#define READ ffs_read
53#define READ_S "ffs_read"
54#define WRITE ffs_write
55#define WRITE_S "ffs_write"
56#define BUFRD ffs_bufrd
57#define BUFWR ffs_bufwr
58#define fs_sb_getbsize(fs) (fs)->fs_bsize
59#endif
60
61static int ulfs_post_read_update(struct vnode *, int, int);
62static int ulfs_post_write_update(struct vnode *, struct uio *, int,
63 kauth_cred_t, off_t, int, int, int);
64
65/*
66 * Vnode op for reading.
67 */
68/* ARGSUSED */
69int
70READ(void *v)
71{
72 struct vop_read_args /* {
73 struct vnode *a_vp;
74 struct uio *a_uio;
75 int a_ioflag;
76 kauth_cred_t a_cred;
77 } */ *ap = v;
78 struct vnode *vp;
79 struct inode *ip;
80 struct uio *uio;
81 FS *fs;
82 vsize_t bytelen;
83 int error, ioflag, advice;
84
85 vp = ap->a_vp;
86 ip = VTOI(vp);
87 fs = ip->I_FS;
88 uio = ap->a_uio;
89 ioflag = ap->a_ioflag;
90 error = 0;
91
92 KASSERT(uio->uio_rw == UIO_READ);
93 KASSERT(vp->v_type == VREG || vp->v_type == VDIR);
94
95 /* XXX Eliminate me by refusing directory reads from userland. */
96 if (vp->v_type == VDIR)
97 return BUFRD(vp, uio, ioflag, ap->a_cred);
98#ifdef LFS_READWRITE
99 /* XXX Eliminate me by using ufs_bufio in lfs. */
100 if (vp->v_type == VREG && ip->i_number == LFS_IFILE_INUM)
101 return BUFRD(vp, uio, ioflag, ap->a_cred);
102#endif
103 if ((u_int64_t)uio->uio_offset > fs->um_maxfilesize)
104 return (EFBIG);
105 if (uio->uio_resid == 0)
106 return (0);
107
108#ifndef LFS_READWRITE
109 if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT)
110 return ffs_snapshot_read(vp, uio, ioflag);
111#endif /* !LFS_READWRITE */
112
113 fstrans_start(vp->v_mount, FSTRANS_SHARED);
114
115 if (uio->uio_offset >= ip->i_size)
116 goto out;
117
118 KASSERT(vp->v_type == VREG);
119 advice = IO_ADV_DECODE(ap->a_ioflag);
120 while (uio->uio_resid > 0) {
121 if (ioflag & IO_DIRECT) {
122 genfs_directio(vp, uio, ioflag);
123 }
124 bytelen = MIN(ip->i_size - uio->uio_offset, uio->uio_resid);
125 if (bytelen == 0)
126 break;
127 error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
128 UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
129 if (error)
130 break;
131 }
132
133 out:
134 error = ulfs_post_read_update(vp, ap->a_ioflag, error);
135 fstrans_done(vp->v_mount);
136 return (error);
137}
138
139/*
140 * UFS op for reading via the buffer cache
141 */
142int
143BUFRD(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred)
144{
145 struct inode *ip;
146 FS *fs;
147 struct buf *bp;
148 daddr_t lbn, nextlbn;
149 off_t bytesinfile;
150 long size, xfersize, blkoffset;
151 int error;
152
153 KASSERT(VOP_ISLOCKED(vp));
154 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK ||
155 vp->v_type == VREG);
156 KASSERT(uio->uio_rw == UIO_READ);
157
158 ip = VTOI(vp);
159 fs = ip->I_FS;
160 error = 0;
161
162 KASSERT(vp->v_type != VLNK || ip->i_size >= fs->um_maxsymlinklen);
163 KASSERT(vp->v_type != VLNK || fs->um_maxsymlinklen != 0 ||
164 DIP(ip, blocks) == 0);
165 KASSERT(vp->v_type != VREG || vp == fs->lfs_ivnode);
166 KASSERT(vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM);
167
168 if (uio->uio_offset > fs->um_maxfilesize)
169 return EFBIG;
170 if (uio->uio_resid == 0)
171 return 0;
172
173#ifndef LFS_READWRITE
174 KASSERT(!ISSET(ip->i_flags, (SF_SNAPSHOT | SF_SNAPINVAL)));
175#endif
176
177 fstrans_start(vp->v_mount, FSTRANS_SHARED);
178
179 if (uio->uio_offset >= ip->i_size)
180 goto out;
181
182 for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
183 bytesinfile = ip->i_size - uio->uio_offset;
184 if (bytesinfile <= 0)
185 break;
186 lbn = lfs_lblkno(fs, uio->uio_offset);
187 nextlbn = lbn + 1;
188 size = lfs_blksize(fs, ip, lbn);
189 blkoffset = lfs_blkoff(fs, uio->uio_offset);
190 xfersize = MIN(MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid),
191 bytesinfile);
192
193 if (lfs_lblktosize(fs, nextlbn) >= ip->i_size)
194 error = bread(vp, lbn, size, 0, &bp);
195 else {
196 int nextsize = lfs_blksize(fs, ip, nextlbn);
197 error = breadn(vp, lbn,
198 size, &nextlbn, &nextsize, 1, 0, &bp);
199 }
200 if (error)
201 break;
202
203 /*
204 * We should only get non-zero b_resid when an I/O error
205 * has occurred, which should cause us to break above.
206 * However, if the short read did not cause an error,
207 * then we want to ensure that we do not uiomove bad
208 * or uninitialized data.
209 */
210 size -= bp->b_resid;
211 if (size < xfersize) {
212 if (size == 0)
213 break;
214 xfersize = size;
215 }
216 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
217 if (error)
218 break;
219 brelse(bp, 0);
220 }
221 if (bp != NULL)
222 brelse(bp, 0);
223
224 out:
225 error = ulfs_post_read_update(vp, ioflag, error);
226 fstrans_done(vp->v_mount);
227 return (error);
228}
229
230static int
231ulfs_post_read_update(struct vnode *vp, int ioflag, int oerror)
232{
233 struct inode *ip = VTOI(vp);
234 int error = oerror;
235
236 if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
237 ip->i_flag |= IN_ACCESS;
238 if ((ioflag & IO_SYNC) == IO_SYNC) {
239 error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
240 }
241 }
242
243 /* Read error overrides any inode update error. */
244 if (oerror)
245 error = oerror;
246 return error;
247}
248
249/*
250 * Vnode op for writing.
251 */
252int
253WRITE(void *v)
254{
255 struct vop_write_args /* {
256 struct vnode *a_vp;
257 struct uio *a_uio;
258 int a_ioflag;
259 kauth_cred_t a_cred;
260 } */ *ap = v;
261 struct vnode *vp;
262 struct uio *uio;
263 struct inode *ip;
264 FS *fs;
265 kauth_cred_t cred;
266 off_t osize, origoff, oldoff, preallocoff, endallocoff, nsize;
267 int blkoffset, error, flags, ioflag, resid;
268 int aflag;
269 int extended=0;
270 vsize_t bytelen;
271 bool async;
272
273 cred = ap->a_cred;
274 ioflag = ap->a_ioflag;
275 uio = ap->a_uio;
276 vp = ap->a_vp;
277 ip = VTOI(vp);
278
279 KASSERT(vp->v_size == ip->i_size);
280 KASSERT(uio->uio_rw == UIO_WRITE);
281 KASSERT(vp->v_type == VREG);
282
283 if (ioflag & IO_APPEND)
284 uio->uio_offset = ip->i_size;
285 if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
286 return (EPERM);
287
288 fs = ip->I_FS;
289 if (uio->uio_offset < 0 ||
290 (u_int64_t)uio->uio_offset + uio->uio_resid > fs->um_maxfilesize)
291 return (EFBIG);
292#ifdef LFS_READWRITE
293 /* Disallow writes to the Ifile, even if noschg flag is removed */
294 /* XXX can this go away when the Ifile is no longer in the namespace? */
295 if (vp == fs->lfs_ivnode)
296 return (EPERM);
297#endif
298 if (uio->uio_resid == 0)
299 return (0);
300
301 fstrans_start(vp->v_mount, FSTRANS_SHARED);
302
303 flags = ioflag & IO_SYNC ? B_SYNC : 0;
304 async = vp->v_mount->mnt_flag & MNT_ASYNC;
305 origoff = uio->uio_offset;
306 resid = uio->uio_resid;
307 osize = ip->i_size;
308 error = 0;
309
310 KASSERT(vp->v_type == VREG);
311
312#ifdef LFS_READWRITE
313 async = true;
314 lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid));
315 lfs_check(vp, LFS_UNUSED_LBN, 0);
316#endif /* !LFS_READWRITE */
317
318 preallocoff = round_page(lfs_blkroundup(fs, MAX(osize, uio->uio_offset)));
319 aflag = ioflag & IO_SYNC ? B_SYNC : 0;
320 nsize = MAX(osize, uio->uio_offset + uio->uio_resid);
321 endallocoff = nsize - lfs_blkoff(fs, nsize);
322
323 /*
324 * if we're increasing the file size, deal with expanding
325 * the fragment if there is one.
326 */
327
328 if (nsize > osize && lfs_lblkno(fs, osize) < ULFS_NDADDR &&
329 lfs_lblkno(fs, osize) != lfs_lblkno(fs, nsize) &&
330 lfs_blkroundup(fs, osize) != osize) {
331 off_t eob;
332
333 eob = lfs_blkroundup(fs, osize);
334 uvm_vnp_setwritesize(vp, eob);
335 error = ulfs_balloc_range(vp, osize, eob - osize, cred, aflag);
336 if (error)
337 goto out;
338 if (flags & B_SYNC) {
339 mutex_enter(vp->v_interlock);
340 VOP_PUTPAGES(vp, trunc_page(osize & lfs_sb_getbmask(fs)),
341 round_page(eob),
342 PGO_CLEANIT | PGO_SYNCIO);
343 }
344 }
345
346 while (uio->uio_resid > 0) {
347 int ubc_flags = UBC_WRITE;
348 bool overwrite; /* if we're overwrite a whole block */
349 off_t newoff;
350
351 if (ioflag & IO_DIRECT) {
352 genfs_directio(vp, uio, ioflag);
353 }
354
355 oldoff = uio->uio_offset;
356 blkoffset = lfs_blkoff(fs, uio->uio_offset);
357 bytelen = MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid);
358 if (bytelen == 0) {
359 break;
360 }
361
362 /*
363 * if we're filling in a hole, allocate the blocks now and
364 * initialize the pages first. if we're extending the file,
365 * we can safely allocate blocks without initializing pages
366 * since the new blocks will be inaccessible until the write
367 * is complete.
368 */
369 overwrite = uio->uio_offset >= preallocoff &&
370 uio->uio_offset < endallocoff;
371 if (!overwrite && (vp->v_vflag & VV_MAPPED) == 0 &&
372 lfs_blkoff(fs, uio->uio_offset) == 0 &&
373 (uio->uio_offset & PAGE_MASK) == 0) {
374 vsize_t len;
375
376 len = trunc_page(bytelen);
377 len -= lfs_blkoff(fs, len);
378 if (len > 0) {
379 overwrite = true;
380 bytelen = len;
381 }
382 }
383
384 newoff = oldoff + bytelen;
385 if (vp->v_size < newoff) {
386 uvm_vnp_setwritesize(vp, newoff);
387 }
388
389 if (!overwrite) {
390 error = ulfs_balloc_range(vp, uio->uio_offset, bytelen,
391 cred, aflag);
392 if (error)
393 break;
394 } else {
395 genfs_node_wrlock(vp);
396 error = GOP_ALLOC(vp, uio->uio_offset, bytelen,
397 aflag, cred);
398 genfs_node_unlock(vp);
399 if (error)
400 break;
401 ubc_flags |= UBC_FAULTBUSY;
402 }
403
404 /*
405 * copy the data.
406 */
407
408 error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
409 IO_ADV_DECODE(ioflag), ubc_flags | UBC_UNMAP_FLAG(vp));
410
411 /*
412 * update UVM's notion of the size now that we've
413 * copied the data into the vnode's pages.
414 *
415 * we should update the size even when uiomove failed.
416 */
417
418 if (vp->v_size < newoff) {
419 uvm_vnp_setsize(vp, newoff);
420 extended = 1;
421 }
422
423 if (error)
424 break;
425
426 /*
427 * flush what we just wrote if necessary.
428 * XXXUBC simplistic async flushing.
429 */
430
431#ifndef LFS_READWRITE
432 if (!async && oldoff >> 16 != uio->uio_offset >> 16) {
433 mutex_enter(vp->v_interlock);
434 error = VOP_PUTPAGES(vp, (oldoff >> 16) << 16,
435 (uio->uio_offset >> 16) << 16,
436 PGO_CLEANIT | PGO_LAZY);
437 if (error)
438 break;
439 }
440#else
441 __USE(async);
442#endif
443 }
444 if (error == 0 && ioflag & IO_SYNC) {
445 mutex_enter(vp->v_interlock);
446 error = VOP_PUTPAGES(vp, trunc_page(origoff & lfs_sb_getbmask(fs)),
447 round_page(lfs_blkroundup(fs, uio->uio_offset)),
448 PGO_CLEANIT | PGO_SYNCIO);
449 }
450
451out:
452 error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid,
453 extended, error);
454 fstrans_done(vp->v_mount);
455
456 return (error);
457}
458
459/*
460 * UFS op for writing via the buffer cache
461 */
462int
463BUFWR(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred)
464{
465 struct inode *ip;
466 FS *fs;
467 int flags;
468 struct buf *bp;
469 off_t osize;
470 int resid, xfersize, size, blkoffset;
471 daddr_t lbn;
472 int extended=0;
473 int error;
474#ifdef LFS_READWRITE
475 bool need_unreserve = false;
476#endif
477
478 KASSERT(ISSET(ioflag, IO_NODELOCKED));
479 KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
480 KASSERT(vp->v_type == VDIR || vp->v_type == VLNK);
481 KASSERT(vp->v_type != VDIR || ISSET(ioflag, IO_SYNC));
482 KASSERT(uio->uio_rw == UIO_WRITE);
483
484 ip = VTOI(vp);
485 fs = ip->I_FS;
486
487 KASSERT(vp->v_size == ip->i_size);
488
489 if (uio->uio_offset < 0 ||
490 uio->uio_resid > fs->um_maxfilesize ||
491 uio->uio_offset > (fs->um_maxfilesize - uio->uio_resid))
492 return EFBIG;
493#ifdef LFS_READWRITE
494 KASSERT(vp != fs->lfs_ivnode);
495#endif
496 if (uio->uio_resid == 0)
497 return 0;
498
499 fstrans_start(vp->v_mount, FSTRANS_SHARED);
500
501 flags = ioflag & IO_SYNC ? B_SYNC : 0;
502 resid = uio->uio_resid;
503 osize = ip->i_size;
504 error = 0;
505
506 KASSERT(vp->v_type != VREG);
507
508#ifdef LFS_READWRITE
509 lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid));
510 lfs_check(vp, LFS_UNUSED_LBN, 0);
511#endif /* !LFS_READWRITE */
512
513 /* XXX Should never have pages cached here. */
514 KASSERT(vp->v_uobj.uo_npages == 0);
515 while (uio->uio_resid > 0) {
516 lbn = lfs_lblkno(fs, uio->uio_offset);
517 blkoffset = lfs_blkoff(fs, uio->uio_offset);
518 xfersize = MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid);
519 if (fs_sb_getbsize(fs) > xfersize)
520 flags |= B_CLRBUF;
521 else
522 flags &= ~B_CLRBUF;
523
524#ifdef LFS_READWRITE
525 error = lfs_reserve(fs, vp, NULL,
526 lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
527 if (error)
528 break;
529 need_unreserve = true;
530#endif
531 error = lfs_balloc(vp, uio->uio_offset, xfersize, cred, flags,
532 &bp);
533
534 if (error)
535 break;
536 if (uio->uio_offset + xfersize > ip->i_size) {
537 ip->i_size = uio->uio_offset + xfersize;
538 DIP_ASSIGN(ip, size, ip->i_size);
539 uvm_vnp_setsize(vp, ip->i_size);
540 extended = 1;
541 }
542 size = lfs_blksize(fs, ip, lbn) - bp->b_resid;
543 if (xfersize > size)
544 xfersize = size;
545
546 error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
547
548 /*
549 * if we didn't clear the block and the uiomove failed,
550 * the buf will now contain part of some other file,
551 * so we need to invalidate it.
552 */
553 if (error && (flags & B_CLRBUF) == 0) {
554 brelse(bp, BC_INVAL);
555 break;
556 }
557#ifdef LFS_READWRITE
558 (void)VOP_BWRITE(bp->b_vp, bp);
559 lfs_reserve(fs, vp, NULL,
560 -lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
561 need_unreserve = false;
562#else
563 if (ioflag & IO_SYNC)
564 (void)bwrite(bp);
565 else if (xfersize + blkoffset == fs->fs_bsize)
566 bawrite(bp);
567 else
568 bdwrite(bp);
569#endif
570 if (error || xfersize == 0)
571 break;
572 }
573#ifdef LFS_READWRITE
574 if (need_unreserve) {
575 lfs_reserve(fs, vp, NULL,
576 -lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
577 }
578#endif
579
580 error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid,
581 extended, error);
582 fstrans_done(vp->v_mount);
583
584 return (error);
585}
586
587static int
588ulfs_post_write_update(struct vnode *vp, struct uio *uio, int ioflag,
589 kauth_cred_t cred, off_t osize, int resid, int extended, int oerror)
590{
591 struct inode *ip = VTOI(vp);
592 int error = oerror;
593
594 /* Trigger ctime and mtime updates, and atime if MNT_RELATIME. */
595 ip->i_flag |= IN_CHANGE | IN_UPDATE;
596 if (vp->v_mount->mnt_flag & MNT_RELATIME)
597 ip->i_flag |= IN_ACCESS;
598
599 /*
600 * If we successfully wrote any data and we are not the superuser,
601 * we clear the setuid and setgid bits as a precaution against
602 * tampering.
603 */
604 if (resid > uio->uio_resid && cred) {
605 if (ip->i_mode & ISUID) {
606 if (kauth_authorize_vnode(cred,
607 KAUTH_VNODE_RETAIN_SUID, vp, NULL, EPERM) != 0) {
608 ip->i_mode &= ~ISUID;
609 DIP_ASSIGN(ip, mode, ip->i_mode);
610 }
611 }
612
613 if (ip->i_mode & ISGID) {
614 if (kauth_authorize_vnode(cred,
615 KAUTH_VNODE_RETAIN_SGID, vp, NULL, EPERM) != 0) {
616 ip->i_mode &= ~ISGID;
617 DIP_ASSIGN(ip, mode, ip->i_mode);
618 }
619 }
620 }
621
622 /* If we successfully wrote anything, notify kevent listeners. */
623 if (resid > uio->uio_resid)
624 VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
625
626 /*
627 * Update the size on disk: truncate back to original size on
628 * error, or reflect the new size on success.
629 */
630 if (error) {
631 (void) lfs_truncate(vp, osize, ioflag & IO_SYNC, cred);
632 uio->uio_offset -= resid - uio->uio_resid;
633 uio->uio_resid = resid;
634 } else if (resid > uio->uio_resid && (ioflag & IO_SYNC) == IO_SYNC) {
635 error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
636 } else {
637 /* nothing */
638 }
639
640 /* Make sure the vnode uvm size matches the inode file size. */
641 KASSERT(vp->v_size == ip->i_size);
642
643 /* Write error overrides any inode update error. */
644 if (oerror)
645 error = oerror;
646 return error;
647}
648