xref: /netbsd-src/sys/ufs/lfs/ulfs_readwrite.c (revision 8ecbf5f02b752fcb7debe1a8fab1dc82602bc760)
1 /*	$NetBSD: ulfs_readwrite.c,v 1.27 2020/04/23 21:47:09 ad 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 1.27 2020/04/23 21:47:09 ad Exp $");
37 
38 #define	FS			struct lfs
39 #define	I_FS			i_lfs
40 #define	READ			lfs_read
41 #define	READ_S			"lfs_read"
42 #define	WRITE			lfs_write
43 #define	WRITE_S			"lfs_write"
44 #define	BUFRD			lfs_bufrd
45 #define	BUFWR			lfs_bufwr
46 #define	fs_sb_getbsize(fs)	lfs_sb_getbsize(fs)
47 #define	fs_bmask		lfs_bmask
48 
49 static int	ulfs_post_read_update(struct vnode *, int, int);
50 static int	ulfs_post_write_update(struct vnode *, struct uio *, int,
51 		    kauth_cred_t, off_t, int, int, int);
52 
53 /*
54  * Vnode op for reading.
55  */
56 /* ARGSUSED */
57 int
58 READ(void *v)
59 {
60 	struct vop_read_args /* {
61 		struct vnode *a_vp;
62 		struct uio *a_uio;
63 		int a_ioflag;
64 		kauth_cred_t a_cred;
65 	} */ *ap = v;
66 	struct vnode *vp;
67 	struct inode *ip;
68 	struct uio *uio;
69 	FS *fs;
70 	vsize_t bytelen;
71 	int error, ioflag, advice;
72 
73 	vp = ap->a_vp;
74 	ip = VTOI(vp);
75 	fs = ip->I_FS;
76 	uio = ap->a_uio;
77 	ioflag = ap->a_ioflag;
78 	error = 0;
79 
80 	KASSERT(uio->uio_rw == UIO_READ);
81 	KASSERT(vp->v_type == VREG || vp->v_type == VDIR);
82 
83 	/* XXX Eliminate me by refusing directory reads from userland.  */
84 	if (vp->v_type == VDIR)
85 		return BUFRD(vp, uio, ioflag, ap->a_cred);
86 	/* XXX Eliminate me by using ufs_bufio in lfs.  */
87 	if (vp->v_type == VREG && ip->i_number == LFS_IFILE_INUM)
88 		return BUFRD(vp, uio, ioflag, ap->a_cred);
89 	if ((u_int64_t)uio->uio_offset > fs->um_maxfilesize)
90 		return (EFBIG);
91 	if (uio->uio_resid == 0)
92 		return (0);
93 
94 
95 	if (uio->uio_offset >= ip->i_size)
96 		goto out;
97 
98 	KASSERT(vp->v_type == VREG);
99 	advice = IO_ADV_DECODE(ap->a_ioflag);
100 	while (uio->uio_resid > 0) {
101 		if (ioflag & IO_DIRECT) {
102 			genfs_directio(vp, uio, ioflag);
103 		}
104 		bytelen = MIN(ip->i_size - uio->uio_offset, uio->uio_resid);
105 		if (bytelen == 0)
106 			break;
107 		error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
108 		    UBC_READ | UBC_PARTIALOK | UBC_VNODE_FLAGS(vp));
109 		if (error)
110 			break;
111 	}
112 
113  out:
114 	error = ulfs_post_read_update(vp, ap->a_ioflag, error);
115 	return (error);
116 }
117 
118 /*
119  * UFS op for reading via the buffer cache
120  */
121 int
122 BUFRD(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred)
123 {
124 	struct inode *ip;
125 	FS *fs;
126 	struct buf *bp;
127 	daddr_t lbn, nextlbn;
128 	off_t bytesinfile;
129 	long size, xfersize, blkoffset;
130 	int error;
131 
132 	KASSERT(VOP_ISLOCKED(vp));
133 	KASSERT(vp->v_type == VDIR || vp->v_type == VLNK ||
134 	    vp->v_type == VREG);
135 	KASSERT(uio->uio_rw == UIO_READ);
136 
137 	ip = VTOI(vp);
138 	fs = ip->I_FS;
139 	error = 0;
140 
141 	KASSERT(vp->v_type != VLNK || ip->i_size >= fs->um_maxsymlinklen);
142 	KASSERT(vp->v_type != VLNK || fs->um_maxsymlinklen != 0 ||
143 	    DIP(ip, blocks) == 0);
144 	KASSERT(vp->v_type != VREG || vp == fs->lfs_ivnode);
145 	KASSERT(vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM);
146 
147 	if (uio->uio_offset > fs->um_maxfilesize)
148 		return EFBIG;
149 	if (uio->uio_resid == 0)
150 		return 0;
151 
152 
153 	if (uio->uio_offset >= ip->i_size)
154 		goto out;
155 
156 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
157 		bytesinfile = ip->i_size - uio->uio_offset;
158 		if (bytesinfile <= 0)
159 			break;
160 		lbn = lfs_lblkno(fs, uio->uio_offset);
161 		nextlbn = lbn + 1;
162 		size = lfs_blksize(fs, ip, lbn);
163 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
164 		xfersize = MIN(MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid),
165 		    bytesinfile);
166 
167 		if (lfs_lblktosize(fs, nextlbn) >= ip->i_size)
168 			error = bread(vp, lbn, size, 0, &bp);
169 		else {
170 			int nextsize = lfs_blksize(fs, ip, nextlbn);
171 			error = breadn(vp, lbn,
172 			    size, &nextlbn, &nextsize, 1, 0, &bp);
173 		}
174 		if (error)
175 			break;
176 
177 		/*
178 		 * We should only get non-zero b_resid when an I/O error
179 		 * has occurred, which should cause us to break above.
180 		 * However, if the short read did not cause an error,
181 		 * then we want to ensure that we do not uiomove bad
182 		 * or uninitialized data.
183 		 */
184 		size -= bp->b_resid;
185 		if (size < xfersize) {
186 			if (size == 0)
187 				break;
188 			xfersize = size;
189 		}
190 		error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
191 		if (error)
192 			break;
193 		brelse(bp, 0);
194 	}
195 	if (bp != NULL)
196 		brelse(bp, 0);
197 
198  out:
199 	error = ulfs_post_read_update(vp, ioflag, error);
200 	return (error);
201 }
202 
203 static int
204 ulfs_post_read_update(struct vnode *vp, int ioflag, int oerror)
205 {
206 	struct inode *ip = VTOI(vp);
207 	int error = oerror;
208 
209 	if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
210 		ip->i_state |= IN_ACCESS;
211 		if ((ioflag & IO_SYNC) == IO_SYNC) {
212 			error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
213 		}
214 	}
215 
216 	/* Read error overrides any inode update error.  */
217 	if (oerror)
218 		error = oerror;
219 	return error;
220 }
221 
222 /*
223  * Vnode op for writing.
224  */
225 int
226 WRITE(void *v)
227 {
228 	struct vop_write_args /* {
229 		struct vnode *a_vp;
230 		struct uio *a_uio;
231 		int a_ioflag;
232 		kauth_cred_t a_cred;
233 	} */ *ap = v;
234 	struct vnode *vp;
235 	struct uio *uio;
236 	struct inode *ip;
237 	FS *fs;
238 	kauth_cred_t cred;
239 	off_t osize, origoff, oldoff, preallocoff, endallocoff, nsize;
240 	int blkoffset, error, flags, ioflag, resid;
241 	int aflag;
242 	int extended=0;
243 	vsize_t bytelen;
244 	bool async;
245 
246 	cred = ap->a_cred;
247 	ioflag = ap->a_ioflag;
248 	uio = ap->a_uio;
249 	vp = ap->a_vp;
250 	ip = VTOI(vp);
251 
252 	KASSERT(vp->v_size == ip->i_size);
253 	KASSERT(uio->uio_rw == UIO_WRITE);
254 	KASSERT(vp->v_type == VREG);
255 
256 	if (ioflag & IO_APPEND)
257 		uio->uio_offset = ip->i_size;
258 	if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
259 		return (EPERM);
260 
261 	fs = ip->I_FS;
262 	if (uio->uio_offset < 0 ||
263 	    (u_int64_t)uio->uio_offset + uio->uio_resid > fs->um_maxfilesize)
264 		return (EFBIG);
265 	/* Disallow writes to the Ifile, even if noschg flag is removed */
266 	/* XXX can this go away when the Ifile is no longer in the namespace? */
267 	if (vp == fs->lfs_ivnode)
268 		return (EPERM);
269 	if (uio->uio_resid == 0)
270 		return (0);
271 
272 	flags = ioflag & IO_SYNC ? B_SYNC : 0;
273 	async = vp->v_mount->mnt_flag & MNT_ASYNC;
274 	origoff = uio->uio_offset;
275 	resid = uio->uio_resid;
276 	osize = ip->i_size;
277 	error = 0;
278 
279 	KASSERT(vp->v_type == VREG);
280 
281 	async = true;
282 	lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid));
283 	lfs_check(vp, LFS_UNUSED_LBN, 0);
284 
285 	preallocoff = round_page(lfs_blkroundup(fs, MAX(osize, uio->uio_offset)));
286 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
287 	nsize = MAX(osize, uio->uio_offset + uio->uio_resid);
288 	endallocoff = nsize - lfs_blkoff(fs, nsize);
289 
290 	/*
291 	 * if we're increasing the file size, deal with expanding
292 	 * the fragment if there is one.
293 	 */
294 
295 	if (nsize > osize && lfs_lblkno(fs, osize) < ULFS_NDADDR &&
296 	    lfs_lblkno(fs, osize) != lfs_lblkno(fs, nsize) &&
297 	    lfs_blkroundup(fs, osize) != osize) {
298 		off_t eob;
299 
300 		eob = lfs_blkroundup(fs, osize);
301 		uvm_vnp_setwritesize(vp, eob);
302 		error = ulfs_balloc_range(vp, osize, eob - osize, cred, aflag);
303 		if (error)
304 			goto out;
305 		if (flags & B_SYNC) {
306 			rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
307 			VOP_PUTPAGES(vp, trunc_page(osize & lfs_sb_getbmask(fs)),
308 			    round_page(eob),
309 			    PGO_CLEANIT | PGO_SYNCIO);
310 		}
311 	}
312 
313 	while (uio->uio_resid > 0) {
314 		int ubc_flags = UBC_WRITE;
315 		bool overwrite; /* if we're overwrite a whole block */
316 		off_t newoff;
317 
318 		if (ioflag & IO_DIRECT) {
319 			genfs_directio(vp, uio, ioflag);
320 		}
321 
322 		oldoff = uio->uio_offset;
323 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
324 		bytelen = MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid);
325 		if (bytelen == 0) {
326 			break;
327 		}
328 
329 		/*
330 		 * if we're filling in a hole, allocate the blocks now and
331 		 * initialize the pages first.  if we're extending the file,
332 		 * we can safely allocate blocks without initializing pages
333 		 * since the new blocks will be inaccessible until the write
334 		 * is complete.
335 		 */
336 		overwrite = uio->uio_offset >= preallocoff &&
337 		    uio->uio_offset < endallocoff;
338 		if (!overwrite && (vp->v_vflag & VV_MAPPED) == 0 &&
339 		    lfs_blkoff(fs, uio->uio_offset) == 0 &&
340 		    (uio->uio_offset & PAGE_MASK) == 0) {
341 			vsize_t len;
342 
343 			len = trunc_page(bytelen);
344 			len -= lfs_blkoff(fs, len);
345 			if (len > 0) {
346 				overwrite = true;
347 				bytelen = len;
348 			}
349 		}
350 
351 		newoff = oldoff + bytelen;
352 		if (vp->v_size < newoff) {
353 			uvm_vnp_setwritesize(vp, newoff);
354 		}
355 
356 		if (!overwrite) {
357 			error = ulfs_balloc_range(vp, uio->uio_offset, bytelen,
358 			    cred, aflag);
359 			if (error)
360 				break;
361 		} else {
362 			genfs_node_wrlock(vp);
363 			error = GOP_ALLOC(vp, uio->uio_offset, bytelen,
364 			    aflag, cred);
365 			genfs_node_unlock(vp);
366 			if (error)
367 				break;
368 			ubc_flags |= UBC_FAULTBUSY;
369 		}
370 
371 		/*
372 		 * copy the data.
373 		 */
374 
375 		error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
376 		    IO_ADV_DECODE(ioflag), ubc_flags | UBC_VNODE_FLAGS(vp));
377 
378 		/*
379 		 * update UVM's notion of the size now that we've
380 		 * copied the data into the vnode's pages.
381 		 *
382 		 * we should update the size even when uiomove failed.
383 		 */
384 
385 		if (vp->v_size < newoff) {
386 			uvm_vnp_setsize(vp, newoff);
387 			extended = 1;
388 		}
389 
390 		if (error)
391 			break;
392 
393 		/*
394 		 * flush what we just wrote if necessary.
395 		 * XXXUBC simplistic async flushing.
396 		 */
397 
398 		__USE(async);
399 	}
400 	if (error == 0 && ioflag & IO_SYNC) {
401 		rw_enter(vp->v_uobj.vmobjlock, RW_WRITER);
402 		error = VOP_PUTPAGES(vp, trunc_page(origoff & lfs_sb_getbmask(fs)),
403 		    round_page(lfs_blkroundup(fs, uio->uio_offset)),
404 		    PGO_CLEANIT | PGO_SYNCIO);
405 	}
406 
407 out:
408 	error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid,
409 	    extended, error);
410 
411 	return (error);
412 }
413 
414 /*
415  * UFS op for writing via the buffer cache
416  */
417 int
418 BUFWR(struct vnode *vp, struct uio *uio, int ioflag, kauth_cred_t cred)
419 {
420 	struct inode *ip;
421 	FS *fs;
422 	int flags;
423 	struct buf *bp;
424 	off_t osize;
425 	int resid, xfersize, size, blkoffset;
426 	daddr_t lbn;
427 	int extended=0;
428 	int error;
429 	bool need_unreserve = false;
430 
431 	KASSERT(ISSET(ioflag, IO_NODELOCKED));
432 	KASSERT(VOP_ISLOCKED(vp) == LK_EXCLUSIVE);
433 	KASSERT(vp->v_type == VDIR || vp->v_type == VLNK);
434 	KASSERT(vp->v_type != VDIR || ISSET(ioflag, IO_SYNC));
435 	KASSERT(uio->uio_rw == UIO_WRITE);
436 
437 	ip = VTOI(vp);
438 	fs = ip->I_FS;
439 
440 	KASSERT(vp->v_size == ip->i_size);
441 
442 	if (uio->uio_offset < 0 ||
443 	    uio->uio_resid > fs->um_maxfilesize ||
444 	    uio->uio_offset > (fs->um_maxfilesize - uio->uio_resid))
445 		return EFBIG;
446 	KASSERT(vp != fs->lfs_ivnode);
447 	if (uio->uio_resid == 0)
448 		return 0;
449 
450 	flags = ioflag & IO_SYNC ? B_SYNC : 0;
451 	resid = uio->uio_resid;
452 	osize = ip->i_size;
453 	error = 0;
454 
455 	KASSERT(vp->v_type != VREG);
456 
457 	lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid));
458 	lfs_check(vp, LFS_UNUSED_LBN, 0);
459 
460 	/* XXX Should never have pages cached here.  */
461 	KASSERT(vp->v_uobj.uo_npages == 0);
462 	while (uio->uio_resid > 0) {
463 		lbn = lfs_lblkno(fs, uio->uio_offset);
464 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
465 		xfersize = MIN(fs_sb_getbsize(fs) - blkoffset, uio->uio_resid);
466 		if (fs_sb_getbsize(fs) > xfersize)
467 			flags |= B_CLRBUF;
468 		else
469 			flags &= ~B_CLRBUF;
470 
471 		error = lfs_reserve(fs, vp, NULL,
472 		    lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
473 		if (error)
474 			break;
475 		need_unreserve = true;
476 		error = lfs_balloc(vp, uio->uio_offset, xfersize, cred, flags,
477 		    &bp);
478 
479 		if (error)
480 			break;
481 		if (uio->uio_offset + xfersize > ip->i_size) {
482 			ip->i_size = uio->uio_offset + xfersize;
483 			DIP_ASSIGN(ip, size, ip->i_size);
484 			uvm_vnp_setsize(vp, ip->i_size);
485 			extended = 1;
486 		}
487 		size = lfs_blksize(fs, ip, lbn) - bp->b_resid;
488 		if (xfersize > size)
489 			xfersize = size;
490 
491 		error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
492 
493 		/*
494 		 * if we didn't clear the block and the uiomove failed,
495 		 * the buf will now contain part of some other file,
496 		 * so we need to invalidate it.
497 		 */
498 		if (error && (flags & B_CLRBUF) == 0) {
499 			brelse(bp, BC_INVAL);
500 			break;
501 		}
502 		(void)VOP_BWRITE(bp->b_vp, bp);
503 		lfs_reserve(fs, vp, NULL,
504 		    -lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
505 		need_unreserve = false;
506 		if (error || xfersize == 0)
507 			break;
508 	}
509 	if (need_unreserve) {
510 		lfs_reserve(fs, vp, NULL,
511 		    -lfs_btofsb(fs, (ULFS_NIADDR + 1) << lfs_sb_getbshift(fs)));
512 	}
513 
514 	error = ulfs_post_write_update(vp, uio, ioflag, cred, osize, resid,
515 	    extended, error);
516 
517 	return (error);
518 }
519 
520 static int
521 ulfs_post_write_update(struct vnode *vp, struct uio *uio, int ioflag,
522     kauth_cred_t cred, off_t osize, int resid, int extended, int oerror)
523 {
524 	struct inode *ip = VTOI(vp);
525 	int error = oerror;
526 
527 	/* Trigger ctime and mtime updates, and atime if MNT_RELATIME.  */
528 	ip->i_state |= IN_CHANGE | IN_UPDATE;
529 	if (vp->v_mount->mnt_flag & MNT_RELATIME)
530 		ip->i_state |= IN_ACCESS;
531 
532 	/*
533 	 * If we successfully wrote any data and we are not the superuser,
534 	 * we clear the setuid and setgid bits as a precaution against
535 	 * tampering.
536 	 */
537 	if (resid > uio->uio_resid && cred) {
538 		if (ip->i_mode & ISUID) {
539 			if (kauth_authorize_vnode(cred,
540 			    KAUTH_VNODE_RETAIN_SUID, vp, NULL, EPERM) != 0) {
541 				ip->i_mode &= ~ISUID;
542 				DIP_ASSIGN(ip, mode, ip->i_mode);
543 			}
544 		}
545 
546 		if (ip->i_mode & ISGID) {
547 			if (kauth_authorize_vnode(cred,
548 			    KAUTH_VNODE_RETAIN_SGID, vp, NULL, EPERM) != 0) {
549 				ip->i_mode &= ~ISGID;
550 				DIP_ASSIGN(ip, mode, ip->i_mode);
551 			}
552 		}
553 	}
554 
555 	/* If we successfully wrote anything, notify kevent listeners.  */
556 	if (resid > uio->uio_resid)
557 		VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
558 
559 	/*
560 	 * Update the size on disk: truncate back to original size on
561 	 * error, or reflect the new size on success.
562 	 */
563 	if (error) {
564 		(void) lfs_truncate(vp, osize, ioflag & IO_SYNC, cred);
565 		uio->uio_offset -= resid - uio->uio_resid;
566 		uio->uio_resid = resid;
567 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC) == IO_SYNC) {
568 		error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
569 	} else {
570 		/* nothing */
571 	}
572 
573 	/* Make sure the vnode uvm size matches the inode file size.  */
574 	KASSERT(vp->v_size == ip->i_size);
575 
576 	/* Write error overrides any inode update error.  */
577 	if (oerror)
578 		error = oerror;
579 	return error;
580 }
581