xref: /netbsd-src/sys/ufs/lfs/ulfs_readwrite.c (revision 6a493d6bc668897c91594964a732d38505b70cbb)
1 /*	$NetBSD: ulfs_readwrite.c,v 1.7 2013/10/17 21:01:08 christos Exp $	*/
2 /*  from NetBSD: ufs_readwrite.c,v 1.105 2013/01/22 09:39:18 dholland 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 1.7 2013/10/17 21:01:08 christos 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	fs_bsize		lfs_bsize
46 #define	fs_bmask		lfs_bmask
47 #else
48 #define	FS			struct fs
49 #define	I_FS			i_fs
50 #define	READ			ffs_read
51 #define	READ_S			"ffs_read"
52 #define	WRITE			ffs_write
53 #define	WRITE_S			"ffs_write"
54 #endif
55 
56 /*
57  * Vnode op for reading.
58  */
59 /* ARGSUSED */
60 int
61 READ(void *v)
62 {
63 	struct vop_read_args /* {
64 		struct vnode *a_vp;
65 		struct uio *a_uio;
66 		int a_ioflag;
67 		kauth_cred_t a_cred;
68 	} */ *ap = v;
69 	struct vnode *vp;
70 	struct inode *ip;
71 	struct uio *uio;
72 	struct buf *bp;
73 	FS *fs;
74 	vsize_t bytelen;
75 	daddr_t lbn, nextlbn;
76 	off_t bytesinfile;
77 	long size, xfersize, blkoffset;
78 	int error, ioflag;
79 	bool usepc = false;
80 
81 	vp = ap->a_vp;
82 	ip = VTOI(vp);
83 	fs = ip->I_FS;
84 	uio = ap->a_uio;
85 	ioflag = ap->a_ioflag;
86 	error = 0;
87 
88 #ifdef DIAGNOSTIC
89 	if (uio->uio_rw != UIO_READ)
90 		panic("%s: mode", READ_S);
91 
92 	if (vp->v_type == VLNK) {
93 		if (ip->i_size < fs->um_maxsymlinklen ||
94 		    (fs->um_maxsymlinklen == 0 && DIP(ip, blocks) == 0))
95 			panic("%s: short symlink", READ_S);
96 	} else if (vp->v_type != VREG && vp->v_type != VDIR)
97 		panic("%s: type %d", READ_S, vp->v_type);
98 #endif
99 	if ((u_int64_t)uio->uio_offset > fs->um_maxfilesize)
100 		return (EFBIG);
101 	if (uio->uio_resid == 0)
102 		return (0);
103 
104 #ifndef LFS_READWRITE
105 	if ((ip->i_flags & (SF_SNAPSHOT | SF_SNAPINVAL)) == SF_SNAPSHOT)
106 		return ffs_snapshot_read(vp, uio, ioflag);
107 #endif /* !LFS_READWRITE */
108 
109 	fstrans_start(vp->v_mount, FSTRANS_SHARED);
110 
111 	if (uio->uio_offset >= ip->i_size)
112 		goto out;
113 
114 #ifdef LFS_READWRITE
115 	usepc = (vp->v_type == VREG && ip->i_number != LFS_IFILE_INUM);
116 #else /* !LFS_READWRITE */
117 	usepc = vp->v_type == VREG;
118 #endif /* !LFS_READWRITE */
119 	if (usepc) {
120 		const int advice = IO_ADV_DECODE(ap->a_ioflag);
121 
122 		while (uio->uio_resid > 0) {
123 			if (ioflag & IO_DIRECT) {
124 				genfs_directio(vp, uio, ioflag);
125 			}
126 			bytelen = MIN(ip->i_size - uio->uio_offset,
127 			    uio->uio_resid);
128 			if (bytelen == 0)
129 				break;
130 			error = ubc_uiomove(&vp->v_uobj, uio, bytelen, advice,
131 			    UBC_READ | UBC_PARTIALOK | UBC_UNMAP_FLAG(vp));
132 			if (error)
133 				break;
134 		}
135 		goto out;
136 	}
137 
138 	for (error = 0, bp = NULL; uio->uio_resid > 0; bp = NULL) {
139 		bytesinfile = ip->i_size - uio->uio_offset;
140 		if (bytesinfile <= 0)
141 			break;
142 		lbn = lfs_lblkno(fs, uio->uio_offset);
143 		nextlbn = lbn + 1;
144 		size = lfs_blksize(fs, ip, lbn);
145 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
146 		xfersize = MIN(MIN(fs->fs_bsize - blkoffset, uio->uio_resid),
147 		    bytesinfile);
148 
149 		if (lfs_lblktosize(fs, nextlbn) >= ip->i_size)
150 			error = bread(vp, lbn, size, NOCRED, 0, &bp);
151 		else {
152 			int nextsize = lfs_blksize(fs, ip, nextlbn);
153 			error = breadn(vp, lbn,
154 			    size, &nextlbn, &nextsize, 1, NOCRED, 0, &bp);
155 		}
156 		if (error)
157 			break;
158 
159 		/*
160 		 * We should only get non-zero b_resid when an I/O error
161 		 * has occurred, which should cause us to break above.
162 		 * However, if the short read did not cause an error,
163 		 * then we want to ensure that we do not uiomove bad
164 		 * or uninitialized data.
165 		 */
166 		size -= bp->b_resid;
167 		if (size < xfersize) {
168 			if (size == 0)
169 				break;
170 			xfersize = size;
171 		}
172 		error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
173 		if (error)
174 			break;
175 		brelse(bp, 0);
176 	}
177 	if (bp != NULL)
178 		brelse(bp, 0);
179 
180  out:
181 	if (!(vp->v_mount->mnt_flag & MNT_NOATIME)) {
182 		ip->i_flag |= IN_ACCESS;
183 		if ((ap->a_ioflag & IO_SYNC) == IO_SYNC) {
184 			error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
185 		}
186 	}
187 
188 	fstrans_done(vp->v_mount);
189 	return (error);
190 }
191 
192 /*
193  * Vnode op for writing.
194  */
195 int
196 WRITE(void *v)
197 {
198 	struct vop_write_args /* {
199 		struct vnode *a_vp;
200 		struct uio *a_uio;
201 		int a_ioflag;
202 		kauth_cred_t a_cred;
203 	} */ *ap = v;
204 	struct vnode *vp;
205 	struct uio *uio;
206 	struct inode *ip;
207 	FS *fs;
208 	struct buf *bp;
209 	kauth_cred_t cred;
210 	daddr_t lbn;
211 	off_t osize, origoff, oldoff, preallocoff, endallocoff, nsize;
212 	int blkoffset, error, flags, ioflag, resid, size, xfersize;
213 	int aflag;
214 	int extended=0;
215 	vsize_t bytelen;
216 	bool async;
217 	bool usepc = false;
218 #ifdef LFS_READWRITE
219 	bool need_unreserve = false;
220 #endif
221 
222 	cred = ap->a_cred;
223 	ioflag = ap->a_ioflag;
224 	uio = ap->a_uio;
225 	vp = ap->a_vp;
226 	ip = VTOI(vp);
227 
228 	KASSERT(vp->v_size == ip->i_size);
229 #ifdef DIAGNOSTIC
230 	if (uio->uio_rw != UIO_WRITE)
231 		panic("%s: mode", WRITE_S);
232 #endif
233 
234 	switch (vp->v_type) {
235 	case VREG:
236 		if (ioflag & IO_APPEND)
237 			uio->uio_offset = ip->i_size;
238 		if ((ip->i_flags & APPEND) && uio->uio_offset != ip->i_size)
239 			return (EPERM);
240 		/* FALLTHROUGH */
241 	case VLNK:
242 		break;
243 	case VDIR:
244 		if ((ioflag & IO_SYNC) == 0)
245 			panic("%s: nonsync dir write", WRITE_S);
246 		break;
247 	default:
248 		panic("%s: type", WRITE_S);
249 	}
250 
251 	fs = ip->I_FS;
252 	if (uio->uio_offset < 0 ||
253 	    (u_int64_t)uio->uio_offset + uio->uio_resid > fs->um_maxfilesize)
254 		return (EFBIG);
255 #ifdef LFS_READWRITE
256 	/* Disallow writes to the Ifile, even if noschg flag is removed */
257 	/* XXX can this go away when the Ifile is no longer in the namespace? */
258 	if (vp == fs->lfs_ivnode)
259 		return (EPERM);
260 #endif
261 	if (uio->uio_resid == 0)
262 		return (0);
263 
264 	fstrans_start(vp->v_mount, FSTRANS_SHARED);
265 
266 	flags = ioflag & IO_SYNC ? B_SYNC : 0;
267 	async = vp->v_mount->mnt_flag & MNT_ASYNC;
268 	origoff = uio->uio_offset;
269 	resid = uio->uio_resid;
270 	osize = ip->i_size;
271 	error = 0;
272 
273 	usepc = vp->v_type == VREG;
274 
275 #ifdef LFS_READWRITE
276 	async = true;
277 	lfs_availwait(fs, lfs_btofsb(fs, uio->uio_resid));
278 	lfs_check(vp, LFS_UNUSED_LBN, 0);
279 #endif /* !LFS_READWRITE */
280 	if (!usepc)
281 		goto bcache;
282 
283 	preallocoff = round_page(lfs_blkroundup(fs, MAX(osize, uio->uio_offset)));
284 	aflag = ioflag & IO_SYNC ? B_SYNC : 0;
285 	nsize = MAX(osize, uio->uio_offset + uio->uio_resid);
286 	endallocoff = nsize - lfs_blkoff(fs, nsize);
287 
288 	/*
289 	 * if we're increasing the file size, deal with expanding
290 	 * the fragment if there is one.
291 	 */
292 
293 	if (nsize > osize && lfs_lblkno(fs, osize) < ULFS_NDADDR &&
294 	    lfs_lblkno(fs, osize) != lfs_lblkno(fs, nsize) &&
295 	    lfs_blkroundup(fs, osize) != osize) {
296 		off_t eob;
297 
298 		eob = lfs_blkroundup(fs, osize);
299 		uvm_vnp_setwritesize(vp, eob);
300 		error = ulfs_balloc_range(vp, osize, eob - osize, cred, aflag);
301 		if (error)
302 			goto out;
303 		if (flags & B_SYNC) {
304 			mutex_enter(vp->v_interlock);
305 			VOP_PUTPAGES(vp, trunc_page(osize & fs->fs_bmask),
306 			    round_page(eob),
307 			    PGO_CLEANIT | PGO_SYNCIO | PGO_JOURNALLOCKED);
308 		}
309 	}
310 
311 	while (uio->uio_resid > 0) {
312 		int ubc_flags = UBC_WRITE;
313 		bool overwrite; /* if we're overwrite a whole block */
314 		off_t newoff;
315 
316 		if (ioflag & IO_DIRECT) {
317 			genfs_directio(vp, uio, ioflag | IO_JOURNALLOCKED);
318 		}
319 
320 		oldoff = uio->uio_offset;
321 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
322 		bytelen = MIN(fs->fs_bsize - blkoffset, uio->uio_resid);
323 		if (bytelen == 0) {
324 			break;
325 		}
326 
327 		/*
328 		 * if we're filling in a hole, allocate the blocks now and
329 		 * initialize the pages first.  if we're extending the file,
330 		 * we can safely allocate blocks without initializing pages
331 		 * since the new blocks will be inaccessible until the write
332 		 * is complete.
333 		 */
334 		overwrite = uio->uio_offset >= preallocoff &&
335 		    uio->uio_offset < endallocoff;
336 		if (!overwrite && (vp->v_vflag & VV_MAPPED) == 0 &&
337 		    lfs_blkoff(fs, uio->uio_offset) == 0 &&
338 		    (uio->uio_offset & PAGE_MASK) == 0) {
339 			vsize_t len;
340 
341 			len = trunc_page(bytelen);
342 			len -= lfs_blkoff(fs, len);
343 			if (len > 0) {
344 				overwrite = true;
345 				bytelen = len;
346 			}
347 		}
348 
349 		newoff = oldoff + bytelen;
350 		if (vp->v_size < newoff) {
351 			uvm_vnp_setwritesize(vp, newoff);
352 		}
353 
354 		if (!overwrite) {
355 			error = ulfs_balloc_range(vp, uio->uio_offset, bytelen,
356 			    cred, aflag);
357 			if (error)
358 				break;
359 		} else {
360 			genfs_node_wrlock(vp);
361 			error = GOP_ALLOC(vp, uio->uio_offset, bytelen,
362 			    aflag, cred);
363 			genfs_node_unlock(vp);
364 			if (error)
365 				break;
366 			ubc_flags |= UBC_FAULTBUSY;
367 		}
368 
369 		/*
370 		 * copy the data.
371 		 */
372 
373 		error = ubc_uiomove(&vp->v_uobj, uio, bytelen,
374 		    IO_ADV_DECODE(ioflag), ubc_flags | UBC_UNMAP_FLAG(vp));
375 
376 		/*
377 		 * update UVM's notion of the size now that we've
378 		 * copied the data into the vnode's pages.
379 		 *
380 		 * we should update the size even when uiomove failed.
381 		 */
382 
383 		if (vp->v_size < newoff) {
384 			uvm_vnp_setsize(vp, newoff);
385 			extended = 1;
386 		}
387 
388 		if (error)
389 			break;
390 
391 		/*
392 		 * flush what we just wrote if necessary.
393 		 * XXXUBC simplistic async flushing.
394 		 */
395 
396 #ifndef LFS_READWRITE
397 		if (!async && oldoff >> 16 != uio->uio_offset >> 16) {
398 			mutex_enter(vp->v_interlock);
399 			error = VOP_PUTPAGES(vp, (oldoff >> 16) << 16,
400 			    (uio->uio_offset >> 16) << 16,
401 			    PGO_CLEANIT | PGO_JOURNALLOCKED | PGO_LAZY);
402 			if (error)
403 				break;
404 		}
405 #else
406 		__USE(async);
407 #endif
408 	}
409 	if (error == 0 && ioflag & IO_SYNC) {
410 		mutex_enter(vp->v_interlock);
411 		error = VOP_PUTPAGES(vp, trunc_page(origoff & fs->fs_bmask),
412 		    round_page(lfs_blkroundup(fs, uio->uio_offset)),
413 		    PGO_CLEANIT | PGO_SYNCIO | PGO_JOURNALLOCKED);
414 	}
415 	goto out;
416 
417  bcache:
418 	mutex_enter(vp->v_interlock);
419 	VOP_PUTPAGES(vp, trunc_page(origoff), round_page(origoff + resid),
420 	    PGO_CLEANIT | PGO_FREE | PGO_SYNCIO | PGO_JOURNALLOCKED);
421 	while (uio->uio_resid > 0) {
422 		lbn = lfs_lblkno(fs, uio->uio_offset);
423 		blkoffset = lfs_blkoff(fs, uio->uio_offset);
424 		xfersize = MIN(fs->fs_bsize - blkoffset, uio->uio_resid);
425 		if (fs->fs_bsize > xfersize)
426 			flags |= B_CLRBUF;
427 		else
428 			flags &= ~B_CLRBUF;
429 
430 #ifdef LFS_READWRITE
431 		error = lfs_reserve(fs, vp, NULL,
432 		    lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift));
433 		if (error)
434 			break;
435 		need_unreserve = true;
436 #endif
437 		error = lfs_balloc(vp, uio->uio_offset, xfersize,
438 		    ap->a_cred, flags, &bp);
439 
440 		if (error)
441 			break;
442 		if (uio->uio_offset + xfersize > ip->i_size) {
443 			ip->i_size = uio->uio_offset + xfersize;
444 			DIP_ASSIGN(ip, size, ip->i_size);
445 			uvm_vnp_setsize(vp, ip->i_size);
446 			extended = 1;
447 		}
448 		size = lfs_blksize(fs, ip, lbn) - bp->b_resid;
449 		if (xfersize > size)
450 			xfersize = size;
451 
452 		error = uiomove((char *)bp->b_data + blkoffset, xfersize, uio);
453 
454 		/*
455 		 * if we didn't clear the block and the uiomove failed,
456 		 * the buf will now contain part of some other file,
457 		 * so we need to invalidate it.
458 		 */
459 		if (error && (flags & B_CLRBUF) == 0) {
460 			brelse(bp, BC_INVAL);
461 			break;
462 		}
463 #ifdef LFS_READWRITE
464 		(void)VOP_BWRITE(bp->b_vp, bp);
465 		lfs_reserve(fs, vp, NULL,
466 		    -lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift));
467 		need_unreserve = false;
468 #else
469 		if (ioflag & IO_SYNC)
470 			(void)bwrite(bp);
471 		else if (xfersize + blkoffset == fs->fs_bsize)
472 			bawrite(bp);
473 		else
474 			bdwrite(bp);
475 #endif
476 		if (error || xfersize == 0)
477 			break;
478 	}
479 #ifdef LFS_READWRITE
480 	if (need_unreserve) {
481 		lfs_reserve(fs, vp, NULL,
482 		    -lfs_btofsb(fs, (ULFS_NIADDR + 1) << fs->lfs_bshift));
483 	}
484 #endif
485 
486 	/*
487 	 * If we successfully wrote any data, and we are not the superuser
488 	 * we clear the setuid and setgid bits as a precaution against
489 	 * tampering.
490 	 */
491 out:
492 	ip->i_flag |= IN_CHANGE | IN_UPDATE;
493 	if (vp->v_mount->mnt_flag & MNT_RELATIME)
494 		ip->i_flag |= IN_ACCESS;
495 	if (resid > uio->uio_resid && ap->a_cred) {
496 		if (ip->i_mode & ISUID) {
497 			if (kauth_authorize_vnode(ap->a_cred,
498 			    KAUTH_VNODE_RETAIN_SUID, vp, NULL, EPERM) != 0) {
499 				ip->i_mode &= ~ISUID;
500 				DIP_ASSIGN(ip, mode, ip->i_mode);
501 			}
502 		}
503 
504 		if (ip->i_mode & ISGID) {
505 			if (kauth_authorize_vnode(ap->a_cred,
506 			    KAUTH_VNODE_RETAIN_SGID, vp, NULL, EPERM) != 0) {
507 				ip->i_mode &= ~ISGID;
508 				DIP_ASSIGN(ip, mode, ip->i_mode);
509 			}
510 		}
511 	}
512 	if (resid > uio->uio_resid)
513 		VN_KNOTE(vp, NOTE_WRITE | (extended ? NOTE_EXTEND : 0));
514 	if (error) {
515 		(void) lfs_truncate(vp, osize, ioflag & IO_SYNC, ap->a_cred);
516 		uio->uio_offset -= resid - uio->uio_resid;
517 		uio->uio_resid = resid;
518 	} else if (resid > uio->uio_resid && (ioflag & IO_SYNC) == IO_SYNC) {
519 		error = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
520 	} else {
521 		/* nothing */
522 	}
523 	KASSERT(vp->v_size == ip->i_size);
524 	fstrans_done(vp->v_mount);
525 
526 	return (error);
527 }
528