xref: /freebsd/sys/ufs/ffs/ffs_rawread.c (revision 9336e0699bda8a301cd2bfa37106b6ec5e32012e)
1 /*-
2  * Copyright (c) 2000-2003 Tor Egge
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/fcntl.h>
33 #include <sys/file.h>
34 #include <sys/stat.h>
35 #include <sys/proc.h>
36 #include <sys/limits.h>
37 #include <sys/mount.h>
38 #include <sys/namei.h>
39 #include <sys/vnode.h>
40 #include <sys/conf.h>
41 #include <sys/filio.h>
42 #include <sys/ttycom.h>
43 #include <sys/bio.h>
44 #include <sys/buf.h>
45 #include <ufs/ufs/extattr.h>
46 #include <ufs/ufs/quota.h>
47 #include <ufs/ufs/inode.h>
48 #include <ufs/ufs/ufsmount.h>
49 #include <ufs/ufs/ufs_extern.h>
50 #include <ufs/ffs/fs.h>
51 #include <ufs/ffs/ffs_extern.h>
52 
53 #include <vm/vm.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_object.h>
56 #include <sys/kernel.h>
57 #include <sys/sysctl.h>
58 
59 static int ffs_rawread_readahead(struct vnode *vp,
60 				 caddr_t udata,
61 				 off_t offset,
62 				 size_t len,
63 				 struct thread *td,
64 				 struct buf *bp,
65 				 caddr_t sa);
66 static int ffs_rawread_main(struct vnode *vp,
67 			    struct uio *uio);
68 
69 static int ffs_rawread_sync(struct vnode *vp, struct thread *td);
70 
71 int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
72 
73 void ffs_rawread_setup(void);
74 
75 SYSCTL_DECL(_vfs_ffs);
76 
77 static int ffsrawbufcnt = 4;
78 SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
79 	   "Buffers available for raw reads");
80 
81 static int allowrawread = 1;
82 SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
83 	   "Flag to enable raw reads");
84 
85 static int rawreadahead = 1;
86 SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
87 	   "Flag to enable readahead for long raw reads");
88 
89 
90 void
91 ffs_rawread_setup(void)
92 {
93 	ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
94 }
95 
96 
97 static int
98 ffs_rawread_sync(struct vnode *vp, struct thread *td)
99 {
100 	int spl;
101 	int error;
102 	int upgraded;
103 	struct bufobj *bo;
104 	struct mount *mp;
105 
106 	/* Check for dirty mmap, pending writes and dirty buffers */
107 	spl = splbio();
108 	VI_LOCK(vp);
109 	bo = &vp->v_bufobj;
110 	if (bo->bo_numoutput > 0 ||
111 	    bo->bo_dirty.bv_cnt > 0 ||
112 	    (vp->v_iflag & VI_OBJDIRTY) != 0) {
113 		splx(spl);
114 		VI_UNLOCK(vp);
115 
116 		if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
117 			if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE)
118 				upgraded = 1;
119 			else
120 				upgraded = 0;
121 			VOP_UNLOCK(vp, 0, td);
122 			(void) vn_start_write(vp, &mp, V_WAIT);
123 			VOP_LOCK(vp, LK_EXCLUSIVE, td);
124 		} else if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) {
125 			upgraded = 1;
126 			/* Upgrade to exclusive lock, this might block */
127 			VOP_LOCK(vp, LK_UPGRADE, td);
128 		} else
129 			upgraded = 0;
130 
131 
132 		VI_LOCK(vp);
133 		/* Check if vnode was reclaimed while unlocked. */
134 		if ((vp->v_iflag & VI_DOOMED) != 0) {
135 			VI_UNLOCK(vp);
136 			if (upgraded != 0)
137 				VOP_LOCK(vp, LK_DOWNGRADE, td);
138 			vn_finished_write(mp);
139 			return (EIO);
140 		}
141 		/* Attempt to msync mmap() regions to clean dirty mmap */
142 		if ((vp->v_iflag & VI_OBJDIRTY) != 0) {
143 			VI_UNLOCK(vp);
144 			if (vp->v_object != NULL) {
145 				VM_OBJECT_LOCK(vp->v_object);
146 				vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
147 				VM_OBJECT_UNLOCK(vp->v_object);
148 			}
149 			VI_LOCK(vp);
150 		}
151 
152 		/* Wait for pending writes to complete */
153 		spl = splbio();
154 		error = bufobj_wwait(&vp->v_bufobj, 0, 0);
155 		if (error != 0) {
156 			/* XXX: can't happen with a zero timeout ??? */
157 			splx(spl);
158 			VI_UNLOCK(vp);
159 			if (upgraded != 0)
160 				VOP_LOCK(vp, LK_DOWNGRADE, td);
161 			vn_finished_write(mp);
162 			return (error);
163 		}
164 		/* Flush dirty buffers */
165 		if (bo->bo_dirty.bv_cnt > 0) {
166 			splx(spl);
167 			VI_UNLOCK(vp);
168 			if ((error = ffs_syncvnode(vp, MNT_WAIT)) != 0) {
169 				if (upgraded != 0)
170 					VOP_LOCK(vp, LK_DOWNGRADE, td);
171 				vn_finished_write(mp);
172 				return (error);
173 			}
174 			VI_LOCK(vp);
175 			spl = splbio();
176 			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
177 				panic("ffs_rawread_sync: dirty bufs");
178 		}
179 		splx(spl);
180 		VI_UNLOCK(vp);
181 		if (upgraded != 0)
182 			VOP_LOCK(vp, LK_DOWNGRADE, td);
183 		vn_finished_write(mp);
184 	} else {
185 		splx(spl);
186 		VI_UNLOCK(vp);
187 	}
188 	return 0;
189 }
190 
191 
192 static int
193 ffs_rawread_readahead(struct vnode *vp,
194 		      caddr_t udata,
195 		      off_t offset,
196 		      size_t len,
197 		      struct thread *td,
198 		      struct buf *bp,
199 		      caddr_t sa)
200 {
201 	int error;
202 	u_int iolen;
203 	off_t blockno;
204 	int blockoff;
205 	int bsize;
206 	struct vnode *dp;
207 	int bforwards;
208 	struct inode *ip;
209 	ufs2_daddr_t blkno;
210 
211 	bsize = vp->v_mount->mnt_stat.f_iosize;
212 
213 	ip = VTOI(vp);
214 	dp = ip->i_devvp;
215 
216 	iolen = ((vm_offset_t) udata) & PAGE_MASK;
217 	bp->b_bcount = len;
218 	if (bp->b_bcount + iolen > bp->b_kvasize) {
219 		bp->b_bcount = bp->b_kvasize;
220 		if (iolen != 0)
221 			bp->b_bcount -= PAGE_SIZE;
222 	}
223 	bp->b_flags = 0;	/* XXX necessary ? */
224 	bp->b_iocmd = BIO_READ;
225 	bp->b_iodone = bdone;
226 	bp->b_data = udata;
227 	bp->b_saveaddr = sa;
228 	blockno = offset / bsize;
229 	blockoff = (offset % bsize) / DEV_BSIZE;
230 	if ((daddr_t) blockno != blockno) {
231 		return EINVAL; /* blockno overflow */
232 	}
233 
234 	bp->b_lblkno = bp->b_blkno = blockno;
235 
236 	error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, NULL, &bforwards, NULL);
237 	if (error != 0)
238 		return error;
239 	if (blkno == -1) {
240 
241 		/* Fill holes with NULs to preserve semantics */
242 
243 		if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
244 			bp->b_bcount = bsize - blockoff * DEV_BSIZE;
245 		bp->b_bufsize = bp->b_bcount;
246 
247 		if (vmapbuf(bp) < 0)
248 			return EFAULT;
249 
250 		if (ticks - PCPU_GET(switchticks) >= hogticks)
251 			uio_yield();
252 		bzero(bp->b_data, bp->b_bufsize);
253 
254 		/* Mark operation completed (similar to bufdone()) */
255 
256 		bp->b_resid = 0;
257 		bp->b_flags |= B_DONE;
258 		return 0;
259 	}
260 	bp->b_blkno = blkno + blockoff;
261 	bp->b_offset = bp->b_iooffset = (blkno + blockoff) * DEV_BSIZE;
262 
263 	if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
264 		bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
265 	bp->b_bufsize = bp->b_bcount;
266 
267 	if (vmapbuf(bp) < 0)
268 		return EFAULT;
269 
270 	BO_STRATEGY(&dp->v_bufobj, bp);
271 	return 0;
272 }
273 
274 
275 static int
276 ffs_rawread_main(struct vnode *vp,
277 		 struct uio *uio)
278 {
279 	int error, nerror;
280 	struct buf *bp, *nbp, *tbp;
281 	caddr_t sa, nsa, tsa;
282 	u_int iolen;
283 	int spl;
284 	caddr_t udata;
285 	long resid;
286 	off_t offset;
287 	struct thread *td;
288 
289 	td = uio->uio_td ? uio->uio_td : curthread;
290 	udata = uio->uio_iov->iov_base;
291 	resid = uio->uio_resid;
292 	offset = uio->uio_offset;
293 
294 	/*
295 	 * keep the process from being swapped
296 	 */
297 	PHOLD(td->td_proc);
298 
299 	error = 0;
300 	nerror = 0;
301 
302 	bp = NULL;
303 	nbp = NULL;
304 	sa = NULL;
305 	nsa = NULL;
306 
307 	while (resid > 0) {
308 
309 		if (bp == NULL) { /* Setup first read */
310 			/* XXX: Leave some bufs for swap */
311 			bp = getpbuf(&ffsrawbufcnt);
312 			sa = bp->b_data;
313 			pbgetvp(vp, bp);
314 			error = ffs_rawread_readahead(vp, udata, offset,
315 						     resid, td, bp, sa);
316 			if (error != 0)
317 				break;
318 
319 			if (resid > bp->b_bufsize) { /* Setup fist readahead */
320 				/* XXX: Leave bufs for swap */
321 				if (rawreadahead != 0)
322 					nbp = trypbuf(&ffsrawbufcnt);
323 				else
324 					nbp = NULL;
325 				if (nbp != NULL) {
326 					nsa = nbp->b_data;
327 					pbgetvp(vp, nbp);
328 
329 					nerror = ffs_rawread_readahead(vp,
330 								       udata +
331 								       bp->b_bufsize,
332 								       offset +
333 								       bp->b_bufsize,
334 								       resid -
335 								       bp->b_bufsize,
336 								       td,
337 								       nbp,
338 								       nsa);
339 					if (nerror) {
340 						pbrelvp(nbp);
341 						relpbuf(nbp, &ffsrawbufcnt);
342 						nbp = NULL;
343 					}
344 				}
345 			}
346 		}
347 
348 		spl = splbio();
349 		bwait(bp, PRIBIO, "rawrd");
350 		splx(spl);
351 
352 		vunmapbuf(bp);
353 
354 		iolen = bp->b_bcount - bp->b_resid;
355 		if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) {
356 			nerror = 0;	/* Ignore possible beyond EOF error */
357 			break; /* EOF */
358 		}
359 
360 		if ((bp->b_ioflags & BIO_ERROR) != 0) {
361 			error = bp->b_error;
362 			break;
363 		}
364 		resid -= iolen;
365 		udata += iolen;
366 		offset += iolen;
367 		if (iolen < bp->b_bufsize) {
368 			/* Incomplete read.  Try to read remaining part */
369 			error = ffs_rawread_readahead(vp,
370 						      udata,
371 						      offset,
372 						      bp->b_bufsize - iolen,
373 						      td,
374 						      bp,
375 						      sa);
376 			if (error != 0)
377 				break;
378 		} else if (nbp != NULL) { /* Complete read with readahead */
379 
380 			tbp = bp;
381 			bp = nbp;
382 			nbp = tbp;
383 
384 			tsa = sa;
385 			sa = nsa;
386 			nsa = tsa;
387 
388 			if (resid <= bp->b_bufsize) { /* No more readaheads */
389 				pbrelvp(nbp);
390 				relpbuf(nbp, &ffsrawbufcnt);
391 				nbp = NULL;
392 			} else { /* Setup next readahead */
393 				nerror = ffs_rawread_readahead(vp,
394 							       udata +
395 							       bp->b_bufsize,
396 							       offset +
397 							       bp->b_bufsize,
398 							       resid -
399 							       bp->b_bufsize,
400 							       td,
401 							       nbp,
402 							       nsa);
403 				if (nerror != 0) {
404 					pbrelvp(nbp);
405 					relpbuf(nbp, &ffsrawbufcnt);
406 					nbp = NULL;
407 				}
408 			}
409 		} else if (nerror != 0) {/* Deferred Readahead error */
410 			break;
411 		}  else if (resid > 0) { /* More to read, no readahead */
412 			error = ffs_rawread_readahead(vp, udata, offset,
413 						      resid, td, bp, sa);
414 			if (error != 0)
415 				break;
416 		}
417 	}
418 
419 	if (bp != NULL) {
420 		pbrelvp(bp);
421 		relpbuf(bp, &ffsrawbufcnt);
422 	}
423 	if (nbp != NULL) {			/* Run down readahead buffer */
424 		spl = splbio();
425 		bwait(nbp, PRIBIO, "rawrd");
426 		splx(spl);
427 		vunmapbuf(nbp);
428 		pbrelvp(nbp);
429 		relpbuf(nbp, &ffsrawbufcnt);
430 	}
431 
432 	if (error == 0)
433 		error = nerror;
434 	PRELE(td->td_proc);
435 	uio->uio_iov->iov_base = udata;
436 	uio->uio_resid = resid;
437 	uio->uio_offset = offset;
438 	return error;
439 }
440 
441 
442 int
443 ffs_rawread(struct vnode *vp,
444 	    struct uio *uio,
445 	    int *workdone)
446 {
447 	if (allowrawread != 0 &&
448 	    uio->uio_iovcnt == 1 &&
449 	    uio->uio_segflg == UIO_USERSPACE &&
450 	    uio->uio_resid == uio->uio_iov->iov_len &&
451 	    (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_pflags &
452 	     TDP_DEADLKTREAT) == 0) {
453 		int secsize;		/* Media sector size */
454 		off_t filebytes;	/* Bytes left of file */
455 		int blockbytes;		/* Bytes left of file in full blocks */
456 		int partialbytes;	/* Bytes in last partial block */
457 		int skipbytes;		/* Bytes not to read in ffs_rawread */
458 		struct inode *ip;
459 		int error;
460 
461 
462 		/* Only handle sector aligned reads */
463 		ip = VTOI(vp);
464 		secsize = ip->i_devvp->v_bufobj.bo_bsize;
465 		if ((uio->uio_offset & (secsize - 1)) == 0 &&
466 		    (uio->uio_resid & (secsize - 1)) == 0) {
467 
468 			/* Sync dirty pages and buffers if needed */
469 			error = ffs_rawread_sync(vp,
470 						 (uio->uio_td != NULL) ?
471 						 uio->uio_td : curthread);
472 			if (error != 0)
473 				return error;
474 
475 			/* Check for end of file */
476 			if (ip->i_size > uio->uio_offset) {
477 				filebytes = ip->i_size - uio->uio_offset;
478 
479 				/* No special eof handling needed ? */
480 				if (uio->uio_resid <= filebytes) {
481 					*workdone = 1;
482 					return ffs_rawread_main(vp, uio);
483 				}
484 
485 				partialbytes = ((unsigned int) ip->i_size) %
486 					ip->i_fs->fs_bsize;
487 				blockbytes = (int) filebytes - partialbytes;
488 				if (blockbytes > 0) {
489 					skipbytes = uio->uio_resid -
490 						blockbytes;
491 					uio->uio_resid = blockbytes;
492 					error = ffs_rawread_main(vp, uio);
493 					uio->uio_resid += skipbytes;
494 					if (error != 0)
495 						return error;
496 					/* Read remaining part using buffer */
497 				}
498 			}
499 		}
500 	}
501 	*workdone = 0;
502 	return 0;
503 }
504