ffs_rawread.c revision 140782 
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 140782 2005-01-25 00:40:01Z phk $");
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
52#include <vm/vm.h>
53#include <vm/vm_extern.h>
54#include <vm/vm_object.h>
55#include <sys/kernel.h>
56#include <sys/sysctl.h>
57
58static int ffs_rawread_readahead(struct vnode *vp,
59				 caddr_t udata,
60				 off_t offset,
61				 size_t len,
62				 struct thread *td,
63				 struct buf *bp,
64				 caddr_t sa);
65static int ffs_rawread_main(struct vnode *vp,
66			    struct uio *uio);
67
68static int ffs_rawread_sync(struct vnode *vp, struct thread *td);
69
70int ffs_rawread(struct vnode *vp, struct uio *uio, int *workdone);
71
72void ffs_rawread_setup(void);
73
74static void ffs_rawreadwakeup(struct buf *bp);
75
76
77SYSCTL_DECL(_vfs_ffs);
78
79static int ffsrawbufcnt = 4;
80SYSCTL_INT(_vfs_ffs, OID_AUTO, ffsrawbufcnt, CTLFLAG_RD, &ffsrawbufcnt, 0,
81	   "Buffers available for raw reads");
82
83static int allowrawread = 1;
84SYSCTL_INT(_vfs_ffs, OID_AUTO, allowrawread, CTLFLAG_RW, &allowrawread, 0,
85	   "Flag to enable raw reads");
86
87static int rawreadahead = 1;
88SYSCTL_INT(_vfs_ffs, OID_AUTO, rawreadahead, CTLFLAG_RW, &rawreadahead, 0,
89	   "Flag to enable readahead for long raw reads");
90
91
92void
93ffs_rawread_setup(void)
94{
95	ffsrawbufcnt = (nswbuf > 100 ) ? (nswbuf - (nswbuf >> 4)) : nswbuf - 8;
96}
97
98
99static int
100ffs_rawread_sync(struct vnode *vp, struct thread *td)
101{
102	int spl;
103	int error;
104	int upgraded;
105	struct bufobj *bo;
106
107	GIANT_REQUIRED;
108	/* Check for dirty mmap, pending writes and dirty buffers */
109	spl = splbio();
110	VI_LOCK(vp);
111	bo = &vp->v_bufobj;
112	if (bo->bo_numoutput > 0 ||
113	    bo->bo_dirty.bv_cnt > 0 ||
114	    (vp->v_iflag & VI_OBJDIRTY) != 0) {
115		splx(spl);
116		VI_UNLOCK(vp);
117
118		if (VOP_ISLOCKED(vp, td) != LK_EXCLUSIVE) {
119			upgraded = 1;
120			/* Upgrade to exclusive lock, this might block */
121			VOP_LOCK(vp, LK_UPGRADE | LK_NOPAUSE, td);
122		} else
123			upgraded = 0;
124
125
126		/* Attempt to msync mmap() regions to clean dirty mmap */
127		VI_LOCK(vp);
128		if ((vp->v_iflag & VI_OBJDIRTY) != 0) {
129			VI_UNLOCK(vp);
130			if (vp->v_object != NULL) {
131				VM_OBJECT_LOCK(vp->v_object);
132				vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
133				VM_OBJECT_UNLOCK(vp->v_object);
134			}
135			VI_LOCK(vp);
136		}
137
138		/* Wait for pending writes to complete */
139		spl = splbio();
140		error = bufobj_wwait(&vp->v_bufobj, 0, 0);
141		if (error != 0) {
142			/* XXX: can't happen with a zero timeout ??? */
143			splx(spl);
144			VI_UNLOCK(vp);
145			if (upgraded != 0)
146				VOP_LOCK(vp, LK_DOWNGRADE, td);
147			return (error);
148		}
149		/* Flush dirty buffers */
150		if (bo->bo_dirty.bv_cnt > 0) {
151			splx(spl);
152			VI_UNLOCK(vp);
153			if ((error = VOP_FSYNC(vp, MNT_WAIT, td)) != 0) {
154				if (upgraded != 0)
155					VOP_LOCK(vp, LK_DOWNGRADE, td);
156				return (error);
157			}
158			VI_LOCK(vp);
159			spl = splbio();
160			if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
161				panic("ffs_rawread_sync: dirty bufs");
162		}
163		splx(spl);
164		VI_UNLOCK(vp);
165		if (upgraded != 0)
166			VOP_LOCK(vp, LK_DOWNGRADE, td);
167	} else {
168		splx(spl);
169		VI_UNLOCK(vp);
170	}
171	return 0;
172}
173
174
175static int
176ffs_rawread_readahead(struct vnode *vp,
177		      caddr_t udata,
178		      off_t offset,
179		      size_t len,
180		      struct thread *td,
181		      struct buf *bp,
182		      caddr_t sa)
183{
184	int error;
185	u_int iolen;
186	off_t blockno;
187	int blockoff;
188	int bsize;
189	struct vnode *dp;
190	int bforwards;
191	struct inode *ip;
192	ufs2_daddr_t blkno;
193
194	GIANT_REQUIRED;
195	bsize = vp->v_mount->mnt_stat.f_iosize;
196
197	ip = VTOI(vp);
198	dp = ip->i_devvp;
199
200	iolen = ((vm_offset_t) udata) & PAGE_MASK;
201	bp->b_bcount = len;
202	if (bp->b_bcount + iolen > bp->b_kvasize) {
203		bp->b_bcount = bp->b_kvasize;
204		if (iolen != 0)
205			bp->b_bcount -= PAGE_SIZE;
206	}
207	bp->b_flags = 0;	/* XXX necessary ? */
208	bp->b_iocmd = BIO_READ;
209	bp->b_iodone = ffs_rawreadwakeup;
210	bp->b_data = udata;
211	bp->b_saveaddr = sa;
212	blockno = offset / bsize;
213	blockoff = (offset % bsize) / DEV_BSIZE;
214	if ((daddr_t) blockno != blockno) {
215		return EINVAL; /* blockno overflow */
216	}
217
218	bp->b_lblkno = bp->b_blkno = blockno;
219
220	error = ufs_bmaparray(vp, bp->b_lblkno, &blkno, NULL, &bforwards, NULL);
221	if (error != 0)
222		return error;
223	if (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	bp->b_blkno = blkno + blockoff;
245	bp->b_offset = bp->b_iooffset = (blkno + blockoff) * DEV_BSIZE;
246
247	if (bp->b_bcount + blockoff * DEV_BSIZE > bsize * (1 + bforwards))
248		bp->b_bcount = bsize * (1 + bforwards) - blockoff * DEV_BSIZE;
249	bp->b_bufsize = bp->b_bcount;
250
251	if (vmapbuf(bp) < 0)
252		return EFAULT;
253
254	BO_STRATEGY(&dp->v_bufobj, bp);
255	return 0;
256}
257
258
259static int
260ffs_rawread_main(struct vnode *vp,
261		 struct uio *uio)
262{
263	int error, nerror;
264	struct buf *bp, *nbp, *tbp;
265	caddr_t sa, nsa, tsa;
266	u_int iolen;
267	int spl;
268	caddr_t udata;
269	long resid;
270	off_t offset;
271	struct thread *td;
272
273	GIANT_REQUIRED;
274	td = uio->uio_td ? uio->uio_td : curthread;
275	udata = uio->uio_iov->iov_base;
276	resid = uio->uio_resid;
277	offset = uio->uio_offset;
278
279	/*
280	 * keep the process from being swapped
281	 */
282	PHOLD(td->td_proc);
283
284	error = 0;
285	nerror = 0;
286
287	bp = NULL;
288	nbp = NULL;
289	sa = NULL;
290	nsa = NULL;
291
292	while (resid > 0) {
293
294		if (bp == NULL) { /* Setup first read */
295			/* XXX: Leave some bufs for swap */
296			bp = getpbuf(&ffsrawbufcnt);
297			sa = bp->b_data;
298			bp->b_vp = vp;
299			error = ffs_rawread_readahead(vp, udata, offset,
300						     resid, td, bp, sa);
301			if (error != 0)
302				break;
303
304			if (resid > bp->b_bufsize) { /* Setup fist readahead */
305				/* XXX: Leave bufs for swap */
306				if (rawreadahead != 0)
307					nbp = trypbuf(&ffsrawbufcnt);
308				else
309					nbp = NULL;
310				if (nbp != NULL) {
311					nsa = nbp->b_data;
312					nbp->b_vp = vp;
313
314					nerror = ffs_rawread_readahead(vp,
315								       udata +
316								       bp->b_bufsize,
317								       offset +
318								       bp->b_bufsize,
319								       resid -
320								       bp->b_bufsize,
321								       td,
322								       nbp,
323								       nsa);
324					if (nerror) {
325						relpbuf(nbp, &ffsrawbufcnt);
326						nbp = NULL;
327					}
328				}
329			}
330		}
331
332		spl = splbio();
333		bwait(bp, PRIBIO, "rawrd");
334		splx(spl);
335
336		vunmapbuf(bp);
337
338		iolen = bp->b_bcount - bp->b_resid;
339		if (iolen == 0 && (bp->b_ioflags & BIO_ERROR) == 0) {
340			nerror = 0;	/* Ignore possible beyond EOF error */
341			break; /* EOF */
342		}
343
344		if ((bp->b_ioflags & BIO_ERROR) != 0) {
345			error = bp->b_error;
346			break;
347		}
348		resid -= iolen;
349		udata += iolen;
350		offset += iolen;
351		if (iolen < bp->b_bufsize) {
352			/* Incomplete read.  Try to read remaining part */
353			error = ffs_rawread_readahead(vp,
354						      udata,
355						      offset,
356						      bp->b_bufsize - iolen,
357						      td,
358						      bp,
359						      sa);
360			if (error != 0)
361				break;
362		} else if (nbp != NULL) { /* Complete read with readahead */
363
364			tbp = bp;
365			bp = nbp;
366			nbp = tbp;
367
368			tsa = sa;
369			sa = nsa;
370			nsa = tsa;
371
372			if (resid <= bp->b_bufsize) { /* No more readaheads */
373				relpbuf(nbp, &ffsrawbufcnt);
374				nbp = NULL;
375			} else { /* Setup next readahead */
376				nerror = ffs_rawread_readahead(vp,
377							       udata +
378							       bp->b_bufsize,
379							       offset +
380							       bp->b_bufsize,
381							       resid -
382							       bp->b_bufsize,
383							       td,
384							       nbp,
385							       nsa);
386				if (nerror != 0) {
387					relpbuf(nbp, &ffsrawbufcnt);
388					nbp = NULL;
389				}
390			}
391		} else if (nerror != 0) {/* Deferred Readahead error */
392			break;
393		}  else if (resid > 0) { /* More to read, no readahead */
394			error = ffs_rawread_readahead(vp, udata, offset,
395						      resid, td, bp, sa);
396			if (error != 0)
397				break;
398		}
399	}
400
401	if (bp != NULL)
402		relpbuf(bp, &ffsrawbufcnt);
403	if (nbp != NULL) {			/* Run down readahead buffer */
404		spl = splbio();
405		bwait(nbp, PRIBIO, "rawrd");
406		splx(spl);
407		vunmapbuf(nbp);
408		relpbuf(nbp, &ffsrawbufcnt);
409	}
410
411	if (error == 0)
412		error = nerror;
413	PRELE(td->td_proc);
414	uio->uio_iov->iov_base = udata;
415	uio->uio_resid = resid;
416	uio->uio_offset = offset;
417	return error;
418}
419
420
421int
422ffs_rawread(struct vnode *vp,
423	    struct uio *uio,
424	    int *workdone)
425{
426	if (allowrawread != 0 &&
427	    uio->uio_iovcnt == 1 &&
428	    uio->uio_segflg == UIO_USERSPACE &&
429	    uio->uio_resid == uio->uio_iov->iov_len &&
430	    (((uio->uio_td != NULL) ? uio->uio_td : curthread)->td_pflags &
431	     TDP_DEADLKTREAT) == 0) {
432		int secsize;		/* Media sector size */
433		off_t filebytes;	/* Bytes left of file */
434		int blockbytes;		/* Bytes left of file in full blocks */
435		int partialbytes;	/* Bytes in last partial block */
436		int skipbytes;		/* Bytes not to read in ffs_rawread */
437		struct inode *ip;
438		int error;
439
440
441		/* Only handle sector aligned reads */
442		ip = VTOI(vp);
443		secsize = ip->i_devvp->v_bufobj.bo_bsize;
444		if ((uio->uio_offset & (secsize - 1)) == 0 &&
445		    (uio->uio_resid & (secsize - 1)) == 0) {
446
447			/* Sync dirty pages and buffers if needed */
448			error = ffs_rawread_sync(vp,
449						 (uio->uio_td != NULL) ?
450						 uio->uio_td : curthread);
451			if (error != 0)
452				return error;
453
454			/* Check for end of file */
455			if (ip->i_size > uio->uio_offset) {
456				filebytes = ip->i_size - uio->uio_offset;
457
458				/* No special eof handling needed ? */
459				if (uio->uio_resid <= filebytes) {
460					*workdone = 1;
461					return ffs_rawread_main(vp, uio);
462				}
463
464				partialbytes = ((unsigned int) ip->i_size) %
465					ip->i_fs->fs_bsize;
466				blockbytes = (int) filebytes - partialbytes;
467				if (blockbytes > 0) {
468					skipbytes = uio->uio_resid -
469						blockbytes;
470					uio->uio_resid = blockbytes;
471					error = ffs_rawread_main(vp, uio);
472					uio->uio_resid += skipbytes;
473					if (error != 0)
474						return error;
475					/* Read remaining part using buffer */
476				}
477			}
478		}
479	}
480	*workdone = 0;
481	return 0;
482}
483
484
485static void
486ffs_rawreadwakeup(struct buf *bp)
487{
488	bdone(bp);
489}
490