ffs_rawread.c revision 116192 
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: head/sys/ufs/ffs/ffs_rawread.c 116192 2003-06-11 06:34:30Z obrien $");
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/quota.h>
46#include <ufs/ufs/inode.h>
47#include <ufs/ffs/fs.h>
48
49#include <vm/vm.h>
50#include <vm/vm_extern.h>
51#include <vm/vm_object.h>
52#include <sys/kernel.h>
53#include <sys/sysctl.h>
54
55static int ffs_rawread_readahead(struct vnode *vp,
56				 caddr_t udata,
57				 off_t offset,
58				 size_t len,
59				 struct thread *td,
60				 struct buf *bp,
61				 caddr_t sa);
62static int ffs_rawread_main(struct vnode *vp,
63			    struct uio *uio);
64
65static int ffs_rawread_sync(struct vnode *vp, struct thread *td);
66
67int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
68
69void ffs_rawread_setup(void);
70
71static void ffs_rawreadwakeup(struct buf *bp);
72
73
74SYSCTL_DECL(_vfs_ffs);
75
76static int ffsrawbufcnt = 4;
77SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
78	   "Buffers available for raw reads");
79
80static int allowrawread = 1;
81SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
82	   "Flag to enable raw reads");
83
84static int rawreadahead = 1;
85SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
86	   "Flag to enable readahead for long raw reads");
87
88
89void
90ffs_rawread_setup(void)
91{
92	ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
93}
94
95
96static int
97ffs_rawread_sync(struct vnode *vp, struct thread *td)
98{
99	int spl;
100	int error;
101	int upgraded;
102
103	GIANT_REQUIRED;
104	/* Check for dirty mmap, pending writes and dirty buffers */
105	spl = splbio();
106	VI_LOCK(vp);
107	if (vp->v_numoutput > 0 ||
108	    !TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
109	    (vp->v_iflag & VI_OBJDIRTY) != 0) {
110		splx(spl);
111		VI_UNLOCK(vp);
112
113		if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) {
114			upgraded = 1;
115			/* Upgrade to exclusive lock, this might block */
116			VOP_LOCK(vp, LK_UPGRADE | LK_NOPAUSE, td);
117		} else
118			upgraded = 0;
119
120
121		/* Attempt to msync mmap() regions to clean dirty mmap */
122		VI_LOCK(vp);
123		if ((vp->v_iflag & VI_OBJDIRTY) != 0) {
124			struct vm_object *obj;
125			VI_UNLOCK(vp);
126			if (VOP_GETVOBJECT(vp, &obj) == 0) {
127				VM_OBJECT_LOCK(obj);
128				vm_object_page_clean(obj, 0, 0, OBJPC_SYNC);
129				VM_OBJECT_UNLOCK(obj);
130			}
131			VI_LOCK(vp);
132		}
133
134		/* Wait for pending writes to complete */
135		spl = splbio();
136		while (vp->v_numoutput) {
137			vp->v_iflag |= VI_BWAIT;
138			error = msleep((caddr_t)&vp->v_numoutput,
139				       VI_MTX(vp),
140				       PRIBIO + 1,
141				       "rawrdfls", 0);
142			if (error != 0) {
143				splx(spl);
144				VI_UNLOCK(vp);
145				if (upgraded != 0)
146					VOP_LOCK(vp, LK_DOWNGRADE, td);
147				return (error);
148			}
149		}
150		/* Flush dirty buffers */
151		if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
152			splx(spl);
153			VI_UNLOCK(vp);
154			if ((error = VOP_FSYNC(vp, NOCRED, MNT_WAIT, td)) != 0) {
155				if (upgraded != 0)
156					VOP_LOCK(vp, LK_DOWNGRADE, td);
157				return (error);
158			}
159			VI_LOCK(vp);
160			spl = splbio();
161			if (vp->v_numoutput > 0 ||
162			    !TAILQ_EMPTY(&vp->v_dirtyblkhd))
163				panic("ffs_rawread_sync: dirty bufs");
164		}
165		splx(spl);
166		VI_UNLOCK(vp);
167		if (upgraded != 0)
168			VOP_LOCK(vp, LK_DOWNGRADE, td);
169	} else {
170		splx(spl);
171		VI_UNLOCK(vp);
172	}
173	return 0;
174}
175
176
177static int
178ffs_rawread_readahead(struct vnode *vp,
179		      caddr_t udata,
180		      off_t offset,
181		      size_t len,
182		      struct thread *td,
183		      struct buf *bp,
184		      caddr_t sa)
185{
186	int error;
187	u_int iolen;
188	off_t blockno;
189	int blockoff;
190	int bsize;
191	struct vnode *dp;
192	int bforwards;
193
194	GIANT_REQUIRED;
195	bsize = vp->v_mount->mnt_stat.f_iosize;
196
197	iolen = ((vm_offset_t) udata) & PAGE_MASK;
198	bp->b_bcount = len;
199	if (bp->b_bcount + iolen > bp->b_kvasize) {
200		bp->b_bcount = bp->b_kvasize;
201		if (iolen != 0)
202			bp->b_bcount -= PAGE_SIZE;
203	}
204	bp->b_flags = B_PHYS;
205	bp->b_iocmd = BIO_READ;
206	bp->b_iodone = ffs_rawreadwakeup;
207	bp->b_data = udata;
208	bp->b_saveaddr = sa;
209	bp->b_offset = offset;
210	blockno = bp->b_offset / bsize;
211	blockoff = (bp->b_offset % bsize) / DEV_BSIZE;
212	if ((daddr_t) blockno != blockno) {
213		return EINVAL; /* blockno overflow */
214	}
215
216	bp->b_lblkno = bp->b_blkno = blockno;
217
218	error = VOP_BMAP(vp, bp->b_lblkno, &dp, &bp->b_blkno, &bforwards,
219			 NULL);
220	if (error != 0) {
221		return error;
222	}
223	if (bp->b_blkno == -1) {
224
225		/* Fill holes with NULs to preserve semantics */
226
227		if (bp->b_bcount + blockoff * DEV_BSIZE > bsize)
228			bp->b_bcount = bsize - blockoff * DEV_BSIZE;
229		bp->b_bufsize = bp->b_bcount;
230
231		if (vmapbuf(bp) < 0)
232			return EFAULT;
233
234		if (ticks - PCPU_GET(switchticks) >= hogticks)
235			uio_yield();
236		bzero(bp->b_data, bp->b_bufsize);
237
238		/* Mark operation completed (similar to bufdone()) */
239
240		bp->b_resid = 0;
241		bp->b_flags |= B_DONE;
242		return 0;
243	}
244
245	if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
246		bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
247	bp->b_bufsize = bp->b_bcount;
248	bp->b_blkno += blockoff;
249	bp->b_dev = dp->v_rdev;
250
251	if (vmapbuf(bp) < 0)
252		return EFAULT;
253
254	if (dp->v_type == VCHR)
255		(void) VOP_SPECSTRATEGY(dp, bp);
256	else
257		(void) VOP_STRATEGY(dp, bp);
258	return 0;
259}
260
261
262static int
263ffs_rawread_main(struct vnode *vp,
264		 struct uio *uio)
265{
266	int error, nerror;
267	struct buf *bp, *nbp, *tbp;
268	caddr_t sa, nsa, tsa;
269	u_int iolen;
270	int spl;
271	caddr_t udata;
272	long resid;
273	off_t offset;
274	struct thread *td;
275
276	GIANT_REQUIRED;
277	td = uio->uio_td ? uio->uio_td : curthread;
278	udata = uio->uio_iov->iov_base;
279	resid = uio->uio_resid;
280	offset = uio->uio_offset;
281
282	/*
283	 * keep the process from being swapped
284	 */
285	PHOLD(td->td_proc);
286
287	error = 0;
288	nerror = 0;
289
290	bp = NULL;
291	nbp = NULL;
292	sa = NULL;
293	nsa = NULL;
294
295	while (resid > 0) {
296
297		if (bp == NULL) { /* Setup first read */
298			/* XXX: Leave some bufs for swap */
299			bp = getpbuf(&ffsrawbufcnt);
300			sa = bp->b_data;
301			bp->b_vp = vp;
302			error = ffs_rawread_readahead(vp, udata, offset,
303						     resid, td, bp, sa);
304			if (error != 0)
305				break;
306
307			if (resid > bp->b_bufsize) { /* Setup fist readahead */
308				/* XXX: Leave bufs for swap */
309				if (rawreadahead != 0)
310					nbp = trypbuf(&ffsrawbufcnt);
311				else
312					nbp = NULL;
313				if (nbp != NULL) {
314					nsa = nbp->b_data;
315					nbp->b_vp = vp;
316
317					nerror = ffs_rawread_readahead(vp,
318								       udata +
319								       bp->b_bufsize,
320								       offset +
321								       bp->b_bufsize,
322								       resid -
323								       bp->b_bufsize,
324								       td,
325								       nbp,
326								       nsa);
327					if (nerror) {
328						relpbuf(nbp, &ffsrawbufcnt);
329						nbp = NULL;
330					}
331				}
332			}
333		}
334
335		spl = splbio();
336		bwait(bp, PRIBIO, "rawrd");
337		splx(spl);
338
339		vunmapbuf(bp);
340
341		iolen = bp->b_bcount - bp->b_resid;
342		if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) {
343			nerror = 0;	/* Ignore possible beyond EOF error */
344			break; /* EOF */
345		}
346
347		if ((bp->b_ioflags & BIO_ERROR) != 0) {
348			error = bp->b_error;
349			break;
350		}
351		resid -= iolen;
352		udata += iolen;
353		offset += iolen;
354		if (iolen < bp->b_bufsize) {
355			/* Incomplete read.  Try to read remaining part */
356			error = ffs_rawread_readahead(vp,
357						      udata,
358						      offset,
359						      bp->b_bufsize - iolen,
360						      td,
361						      bp,
362						      sa);
363			if (error != 0)
364				break;
365		} else if (nbp != NULL) { /* Complete read with readahead */
366
367			tbp = bp;
368			bp = nbp;
369			nbp = tbp;
370
371			tsa = sa;
372			sa = nsa;
373			nsa = tsa;
374
375			if (resid <= bp->b_bufsize) { /* No more readaheads */
376				relpbuf(nbp, &ffsrawbufcnt);
377				nbp = NULL;
378			} else { /* Setup next readahead */
379				nerror = ffs_rawread_readahead(vp,
380							       udata +
381							       bp->b_bufsize,
382							       offset +
383							       bp->b_bufsize,
384							       resid -
385							       bp->b_bufsize,
386							       td,
387							       nbp,
388							       nsa);
389				if (nerror != 0) {
390					relpbuf(nbp, &ffsrawbufcnt);
391					nbp = NULL;
392				}
393			}
394		} else if (nerror != 0) {/* Deferred Readahead error */
395			break;
396		}  else if (resid > 0) { /* More to read, no readahead */
397			error = ffs_rawread_readahead(vp, udata, offset,
398						      resid, td, bp, sa);
399			if (error != 0)
400				break;
401		}
402	}
403
404	if (bp != NULL)
405		relpbuf(bp, &ffsrawbufcnt);
406	if (nbp != NULL) {			/* Run down readahead buffer */
407		spl = splbio();
408		bwait(nbp, PRIBIO, "rawrd");
409		splx(spl);
410		vunmapbuf(nbp);
411		relpbuf(nbp, &ffsrawbufcnt);
412	}
413
414	if (error == 0)
415		error = nerror;
416	PRELE(td->td_proc);
417	uio->uio_iov->iov_base = udata;
418	uio->uio_resid = resid;
419	uio->uio_offset = offset;
420	return error;
421}
422
423
424int
425ffs_rawread(struct vnode *vp,
426	    struct uio *uio,
427	    int *workdone)
428{
429	if (allowrawread != 0 &&
430	    uio->uio_iovcnt == 1 &&
431	    uio->uio_segflg == UIO_USERSPACE &&
432	    uio->uio_resid == uio->uio_iov->iov_len &&
433	    (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_flags &
434	     TDF_DEADLKTREAT) == 0) {
435		int secsize;		/* Media sector size */
436		off_t filebytes;	/* Bytes left of file */
437		int blockbytes;		/* Bytes left of file in full blocks */
438		int partialbytes;	/* Bytes in last partial block */
439		int skipbytes;		/* Bytes not to read in ffs_rawread */
440		struct inode *ip;
441		int error;
442
443
444		/* Only handle sector aligned reads */
445		ip = VTOI(vp);
446		secsize = ip->i_devvp->v_rdev->si_bsize_phys;
447		if ((uio->uio_offset & (secsize - 1)) == 0 &&
448		    (uio->uio_resid & (secsize - 1)) == 0) {
449
450			/* Sync dirty pages and buffers if needed */
451			error = ffs_rawread_sync(vp,
452						 (uio->uio_td != NULL) ?
453						 uio->uio_td : curthread);
454			if (error != 0)
455				return error;
456
457			/* Check for end of file */
458			if (ip->i_size > uio->uio_offset) {
459				filebytes = ip->i_size - uio->uio_offset;
460
461				/* No special eof handling needed ? */
462				if (uio->uio_resid <= filebytes) {
463					*workdone = 1;
464					return ffs_rawread_main(vp, uio);
465				}
466
467				partialbytes = ((unsigned int) ip->i_size) %
468					ip->i_fs->fs_bsize;
469				blockbytes = (int) filebytes - partialbytes;
470				if (blockbytes > 0) {
471					skipbytes = uio->uio_resid -
472						blockbytes;
473					uio->uio_resid = blockbytes;
474					error = ffs_rawread_main(vp, uio);
475					uio->uio_resid += skipbytes;
476					if (error != 0)
477						return error;
478					/* Read remaining part using buffer */
479				}
480			}
481		}
482	}
483	*workdone = 0;
484	return 0;
485}
486
487
488static void
489ffs_rawreadwakeup(struct buf *bp)
490{
491	bdone(bp);
492}
493