[unix-history] / usr / src / sys / nfs / nfs_bio.c

/*
 * Copyright (c) 1989 The Regents of the University of California.
 * All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Rick Macklem at The University of Guelph.
 *
 * %sccs.include.redist.c%
 *
 *	@(#)nfs_bio.c	7.19 (Berkeley) %G%
 */

#include "param.h"
#include "proc.h"
#include "buf.h"
#include "uio.h"
#include "namei.h"
#include "vnode.h"
#include "trace.h"
#include "mount.h"
#include "resourcevar.h"

#include "nfsnode.h"
#include "nfsv2.h"
#include "nfs.h"
#include "nfsiom.h"
#include "nfsmount.h"

/* True and false, how exciting */
#define	TRUE	1
#define	FALSE	0

/*
 * Vnode op for read using bio
 * Any similarity to readip() is purely coincidental
 */
nfs_bioread(vp, uio, ioflag, cred)
	register struct vnode *vp;
	register struct uio *uio;
	int ioflag;
	struct ucred *cred;
{
	register struct nfsnode *np = VTONFS(vp);
	register int biosize;
	struct buf *bp;
	struct vattr vattr;
	daddr_t lbn, bn, rablock;
	int diff, error = 0;
	long n, on;

#ifdef lint
	ioflag = ioflag;
#endif /* lint */
#ifdef DIAGNOSTIC
	if (uio->uio_rw != UIO_READ)
		panic("nfs_read mode");
#endif
	if (uio->uio_resid == 0)
		return (0);
	if (uio->uio_offset < 0 && vp->v_type != VDIR)
		return (EINVAL);
	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
	/*
	 * If the file's modify time on the server has changed since the
	 * last read rpc or you have written to the file,
	 * you may have lost data cache consistency with the
	 * server, so flush all of the file's data out of the cache.
	 * Then force a getattr rpc to ensure that you have up to date
	 * attributes.
	 * NB: This implies that cache data can be read when up to
	 * NFS_ATTRTIMEO seconds out of date. If you find that you need current
	 * attributes this could be forced by setting n_attrstamp to 0 before
	 * the nfs_dogetattr() call.
	 */
	if (vp->v_type != VLNK) {
		if (np->n_flag & NMODIFIED) {
			np->n_flag &= ~NMODIFIED;
			vinvalbuf(vp, TRUE);
			np->n_attrstamp = 0;
			np->n_direofoffset = 0;
			if (error = nfs_dogetattr(vp, &vattr, cred, 1,
			    uio->uio_procp))
				return (error);
			np->n_mtime = vattr.va_mtime.tv_sec;
		} else {
			if (error = nfs_dogetattr(vp, &vattr, cred, 1,
			    uio->uio_procp))
				return (error);
			if (np->n_mtime != vattr.va_mtime.tv_sec) {
				np->n_direofoffset = 0;
				vinvalbuf(vp, TRUE);
				np->n_mtime = vattr.va_mtime.tv_sec;
			}
		}
	}
	do {
	    switch (vp->v_type) {
	    case VREG:
		nfsstats.biocache_reads++;
		lbn = uio->uio_offset / biosize;
		on = uio->uio_offset & (biosize-1);
		n = MIN((unsigned)(biosize - on), uio->uio_resid);
		diff = np->n_size - uio->uio_offset;
		if (diff <= 0)
			return (error);
		if (diff < n)
			n = diff;
		bn = lbn*(biosize/DEV_BSIZE);
		rablock = (lbn+1)*(biosize/DEV_BSIZE);
		if (vp->v_lastr + 1 == lbn &&
		    np->n_size > (rablock * DEV_BSIZE))
			error = breada(vp, bn, biosize, rablock, biosize,
				cred, &bp);
		else
			error = bread(vp, bn, biosize, cred, &bp);
		vp->v_lastr = lbn;
		if (bp->b_resid) {
		   diff = (on >= (biosize-bp->b_resid)) ? 0 :
			(biosize-bp->b_resid-on);
		   n = MIN(n, diff);
		}
		break;
	    case VLNK:
		nfsstats.biocache_readlinks++;
		on = 0;
		error = bread(vp, (daddr_t)0, NFS_MAXPATHLEN, cred, &bp);
		n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
		break;
	    case VDIR:
		nfsstats.biocache_readdirs++;
		on = 0;
		error = bread(vp, uio->uio_offset, NFS_DIRBLKSIZ, cred, &bp);
		n = MIN(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid);
		break;
	    };
	    if (error) {
		brelse(bp);
		return (error);
	    }
	    if (n > 0)
		error = uiomove(bp->b_un.b_addr + on, (int)n, uio);
	    switch (vp->v_type) {
	    case VREG:
		if (n+on == biosize || uio->uio_offset == np->n_size)
			bp->b_flags |= B_AGE;
		break;
	    case VLNK:
		n = 0;
		break;
	    case VDIR:
		uio->uio_offset = bp->b_blkno;
		break;
	    };
	    brelse(bp);
	} while (error == 0 && uio->uio_resid > 0 && n != 0);
	return (error);
}

/*
 * Vnode op for write using bio
 */
nfs_write(vp, uio, ioflag, cred)
	register struct vnode *vp;
	register struct uio *uio;
	int ioflag;
	struct ucred *cred;
{
	struct proc *p = uio->uio_procp;
	register int biosize;
	struct buf *bp;
	struct nfsnode *np = VTONFS(vp);
	struct vattr vattr;
	daddr_t lbn, bn;
	int n, on, error = 0;

#ifdef DIAGNOSTIC
	if (uio->uio_rw != UIO_WRITE)
		panic("nfs_write mode");
	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
		panic("nfs_write proc");
#endif
	if (vp->v_type != VREG)
		return (EIO);
	/* Should we try and do this ?? */
	if (ioflag & (IO_APPEND | IO_SYNC)) {
		if (np->n_flag & NMODIFIED) {
			np->n_flag &= ~NMODIFIED;
			vinvalbuf(vp, TRUE);
		}
		if (ioflag & IO_APPEND) {
			np->n_attrstamp = 0;
			if (error = nfs_dogetattr(vp, &vattr, cred, 1, p))
				return (error);
			uio->uio_offset = np->n_size;
		}
		return (nfs_writerpc(vp, uio, cred));
	}
#ifdef notdef
	cnt = uio->uio_resid;
	osize = np->n_size;
#endif
	if (uio->uio_offset < 0)
		return (EINVAL);
	if (uio->uio_resid == 0)
		return (0);
	/*
	 * Maybe this should be above the vnode op call, but so long as
	 * file servers have no limits, i don't think it matters
	 */
	if (uio->uio_offset + uio->uio_resid >
	      p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
		psignal(p, SIGXFSZ);
		return (EFBIG);
	}
	/*
	 * I use nm_rsize, not nm_wsize so that all buffer cache blocks
	 * will be the same size within a filesystem. nfs_writerpc will
	 * still use nm_wsize when sizing the rpc's.
	 */
	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
	np->n_flag |= NMODIFIED;
	do {
		nfsstats.biocache_writes++;
		lbn = uio->uio_offset / biosize;
		on = uio->uio_offset & (biosize-1);
		n = MIN((unsigned)(biosize - on), uio->uio_resid);
		if (uio->uio_offset+n > np->n_size) {
			np->n_size = uio->uio_offset+n;
			vnode_pager_setsize(vp, np->n_size);
		}
		bn = lbn*(biosize/DEV_BSIZE);
again:
		bp = getblk(vp, bn, biosize);
		if (bp->b_wcred == NOCRED) {
			crhold(cred);
			bp->b_wcred = cred;
		}
		if (bp->b_dirtyend > 0) {
			/*
			 * If the new write will leave a contiguous dirty
			 * area, just update the b_dirtyoff and b_dirtyend,
			 * otherwise force a write rpc of the old dirty area.
			 */
			if (on <= bp->b_dirtyend && (on+n) >= bp->b_dirtyoff) {
				bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
				bp->b_dirtyend = MAX((on+n), bp->b_dirtyend);
			} else {
				bp->b_proc = p;
				if (error = bwrite(bp))
					return (error);
				goto again;
			}
		} else {
			bp->b_dirtyoff = on;
			bp->b_dirtyend = on+n;
		}
		if (error = uiomove(bp->b_un.b_addr + on, n, uio)) {
			brelse(bp);
			return (error);
		}
		if ((n+on) == biosize) {
			bp->b_flags |= B_AGE;
			bp->b_proc = (struct proc *)0;
			bawrite(bp);
		} else {
			bp->b_proc = (struct proc *)0;
			bdwrite(bp);
		}
	} while (error == 0 && uio->uio_resid > 0 && n != 0);
#ifdef notdef
	/* Should we try and do this for nfs ?? */
	if (error && (ioflag & IO_UNIT)) {
		np->n_size = osize;
		uio->uio_offset -= cnt - uio->uio_resid;
		uio->uio_resid = cnt;
	}
#endif
	return (error);
}
Commit	Line	Data
39d108be RM	1	/*
	2	* Copyright (c) 1989 The Regents of the University of California.
	3	* All rights reserved.
	4	*
	5	* This code is derived from software contributed to Berkeley by
	6	* Rick Macklem at The University of Guelph.
	7	*
dbf0c423	8	* %sccs.include.redist.c%
39d108be	9	*
b40809cc	10	* @(#)nfs_bio.c 7.19 (Berkeley) %G%
39d108be RM	11	*/
	12
	13	#include "param.h"
4547a923	14	#include "proc.h"
39d108be	15	#include "buf.h"
b40809cc KM	16	#include "uio.h"
b40809cc KM	17	#include "namei.h"
39d108be RM	18	#include "vnode.h"
	19	#include "trace.h"
	20	#include "mount.h"
4547a923 MK	21	#include "resourcevar.h"
4547a923 MK	22
39d108be	23	#include "nfsnode.h"
e8540f59 KM	24	#include "nfsv2.h"
e8540f59 KM	25	#include "nfs.h"
39d108be	26	#include "nfsiom.h"
f0f1cbaa	27	#include "nfsmount.h"
39d108be RM	28
	29	/* True and false, how exciting */
	30	#define TRUE 1
	31	#define FALSE 0
	32
	33	/*
	34	* Vnode op for read using bio
	35	* Any similarity to readip() is purely coincidental
	36	*/
f0f1cbaa	37	nfs_bioread(vp, uio, ioflag, cred)
39d108be	38	register struct vnode *vp;
170bfd05	39	register struct uio *uio;
39d108be RM	40	int ioflag;
	41	struct ucred *cred;
	42	{
	43	register struct nfsnode *np = VTONFS(vp);
170bfd05	44	register int biosize;
39d108be RM	45	struct buf *bp;
	46	struct vattr vattr;
	47	daddr_t lbn, bn, rablock;
13576453	48	int diff, error = 0;
39d108be	49	long n, on;
39d108be	50
d4e5799e KM	51	#ifdef lint
	52	ioflag = ioflag;
	53	#endif /* lint */
b40809cc	54	#ifdef DIAGNOSTIC
39d108be RM	55	if (uio->uio_rw != UIO_READ)
39d108be RM	56	panic("nfs_read mode");
b40809cc	57	#endif
39d108be	58	if (uio->uio_resid == 0)
b71430cc	59	return (0);
f0f1cbaa	60	if (uio->uio_offset < 0 && vp->v_type != VDIR)
b71430cc	61	return (EINVAL);
170bfd05	62	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
39d108be RM	63	/*
	64	* If the file's modify time on the server has changed since the
	65	* last read rpc or you have written to the file,
	66	* you may have lost data cache consistency with the
	67	* server, so flush all of the file's data out of the cache.
f0f1cbaa KM	68	* Then force a getattr rpc to ensure that you have up to date
f0f1cbaa KM	69	* attributes.
39d108be RM	70	* NB: This implies that cache data can be read when up to
	71	* NFS_ATTRTIMEO seconds out of date. If you find that you need current
	72	* attributes this could be forced by setting n_attrstamp to 0 before
170bfd05	73	* the nfs_dogetattr() call.
39d108be	74	*/
f0f1cbaa KM	75	if (vp->v_type != VLNK) {
	76	if (np->n_flag & NMODIFIED) {
	77	np->n_flag &= ~NMODIFIED;
	78	vinvalbuf(vp, TRUE);
	79	np->n_attrstamp = 0;
	80	np->n_direofoffset = 0;
b40809cc KM	81	if (error = nfs_dogetattr(vp, &vattr, cred, 1,
b40809cc KM	82	uio->uio_procp))
e8540f59 KM	83	return (error);
e8540f59 KM	84	np->n_mtime = vattr.va_mtime.tv_sec;
f0f1cbaa	85	} else {
b40809cc KM	86	if (error = nfs_dogetattr(vp, &vattr, cred, 1,
b40809cc KM	87	uio->uio_procp))
f0f1cbaa KM	88	return (error);
	89	if (np->n_mtime != vattr.va_mtime.tv_sec) {
	90	np->n_direofoffset = 0;
	91	vinvalbuf(vp, TRUE);
	92	np->n_mtime = vattr.va_mtime.tv_sec;
	93	}
39d108be RM	94	}
	95	}
	96	do {
f0f1cbaa KM	97	switch (vp->v_type) {
f0f1cbaa KM	98	case VREG:
e8540f59	99	nfsstats.biocache_reads++;
170bfd05 KM	100	lbn = uio->uio_offset / biosize;
	101	on = uio->uio_offset & (biosize-1);
	102	n = MIN((unsigned)(biosize - on), uio->uio_resid);
39d108be RM	103	diff = np->n_size - uio->uio_offset;
39d108be RM	104	if (diff <= 0)
b71430cc	105	return (error);
39d108be RM	106	if (diff < n)
39d108be RM	107	n = diff;
170bfd05 KM	108	bn = lbn*(biosize/DEV_BSIZE);
170bfd05 KM	109	rablock = (lbn+1)*(biosize/DEV_BSIZE);
d1a28114 KM	110	if (vp->v_lastr + 1 == lbn &&
d1a28114 KM	111	np->n_size > (rablock * DEV_BSIZE))
170bfd05	112	error = breada(vp, bn, biosize, rablock, biosize,
39d108be RM	113	cred, &bp);
39d108be RM	114	else
170bfd05	115	error = bread(vp, bn, biosize, cred, &bp);
d1a28114	116	vp->v_lastr = lbn;
39d108be	117	if (bp->b_resid) {
170bfd05 KM	118	diff = (on >= (biosize-bp->b_resid)) ? 0 :
170bfd05 KM	119	(biosize-bp->b_resid-on);
f0f1cbaa	120	n = MIN(n, diff);
39d108be	121	}
f0f1cbaa KM	122	break;
	123	case VLNK:
	124	nfsstats.biocache_readlinks++;
	125	on = 0;
	126	error = bread(vp, (daddr_t)0, NFS_MAXPATHLEN, cred, &bp);
	127	n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
	128	break;
	129	case VDIR:
	130	nfsstats.biocache_readdirs++;
	131	on = 0;
b40809cc KM	132	error = bread(vp, uio->uio_offset, NFS_DIRBLKSIZ, cred, &bp);
b40809cc KM	133	n = MIN(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid);
f0f1cbaa KM	134	break;
	135	};
	136	if (error) {
	137	brelse(bp);
	138	return (error);
	139	}
	140	if (n > 0)
	141	error = uiomove(bp->b_un.b_addr + on, (int)n, uio);
	142	switch (vp->v_type) {
	143	case VREG:
170bfd05	144	if (n+on == biosize \|\| uio->uio_offset == np->n_size)
39d108be	145	bp->b_flags \|= B_AGE;
f0f1cbaa KM	146	break;
	147	case VLNK:
	148	n = 0;
	149	break;
	150	case VDIR:
	151	uio->uio_offset = bp->b_blkno;
	152	break;
	153	};
	154	brelse(bp);
39d108be	155	} while (error == 0 && uio->uio_resid > 0 && n != 0);
39d108be RM	156	return (error);
	157	}
	158
	159	/*
	160	* Vnode op for write using bio
	161	*/
b71430cc	162	nfs_write(vp, uio, ioflag, cred)
39d108be RM	163	register struct vnode *vp;
39d108be RM	164	register struct uio *uio;
39d108be RM	165	int ioflag;
	166	struct ucred *cred;
	167	{
b40809cc	168	struct proc *p = uio->uio_procp;
170bfd05	169	register int biosize;
39d108be RM	170	struct buf *bp;
39d108be RM	171	struct nfsnode *np = VTONFS(vp);
f0f1cbaa	172	struct vattr vattr;
39d108be	173	daddr_t lbn, bn;
3c15394b	174	int n, on, error = 0;
39d108be	175
b40809cc	176	#ifdef DIAGNOSTIC
f0f1cbaa KM	177	if (uio->uio_rw != UIO_WRITE)
f0f1cbaa KM	178	panic("nfs_write mode");
b40809cc KM	179	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
	180	panic("nfs_write proc");
	181	#endif
f0f1cbaa KM	182	if (vp->v_type != VREG)
f0f1cbaa KM	183	return (EIO);
39d108be	184	/* Should we try and do this ?? */
170bfd05	185	if (ioflag & (IO_APPEND \| IO_SYNC)) {
f0f1cbaa KM	186	if (np->n_flag & NMODIFIED) {
	187	np->n_flag &= ~NMODIFIED;
	188	vinvalbuf(vp, TRUE);
	189	}
170bfd05 KM	190	if (ioflag & IO_APPEND) {
170bfd05 KM	191	np->n_attrstamp = 0;
b40809cc	192	if (error = nfs_dogetattr(vp, &vattr, cred, 1, p))
170bfd05 KM	193	return (error);
	194	uio->uio_offset = np->n_size;
	195	}
b40809cc	196	return (nfs_writerpc(vp, uio, cred));
f0f1cbaa	197	}
39d108be	198	#ifdef notdef
b71430cc	199	cnt = uio->uio_resid;
39d108be RM	200	osize = np->n_size;
39d108be RM	201	#endif
b71430cc KM	202	if (uio->uio_offset < 0)
b71430cc KM	203	return (EINVAL);
39d108be	204	if (uio->uio_resid == 0)
b71430cc	205	return (0);
39d108be RM	206	/*
	207	* Maybe this should be above the vnode op call, but so long as
	208	* file servers have no limits, i don't think it matters
	209	*/
f0f1cbaa	210	if (uio->uio_offset + uio->uio_resid >
4547a923 MK	211	p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
4547a923 MK	212	psignal(p, SIGXFSZ);
b71430cc	213	return (EFBIG);
39d108be	214	}
170bfd05 KM	215	/*
	216	* I use nm_rsize, not nm_wsize so that all buffer cache blocks
	217	* will be the same size within a filesystem. nfs_writerpc will
	218	* still use nm_wsize when sizing the rpc's.
	219	*/
	220	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
f0f1cbaa	221	np->n_flag \|= NMODIFIED;
39d108be	222	do {
e8540f59	223	nfsstats.biocache_writes++;
170bfd05 KM	224	lbn = uio->uio_offset / biosize;
	225	on = uio->uio_offset & (biosize-1);
	226	n = MIN((unsigned)(biosize - on), uio->uio_resid);
8986c97c	227	if (uio->uio_offset+n > np->n_size) {
39d108be	228	np->n_size = uio->uio_offset+n;
8986c97c KM	229	vnode_pager_setsize(vp, np->n_size);
8986c97c KM	230	}
170bfd05	231	bn = lbn*(biosize/DEV_BSIZE);
141671b8	232	again:
170bfd05	233	bp = getblk(vp, bn, biosize);
39d108be RM	234	if (bp->b_wcred == NOCRED) {
	235	crhold(cred);
	236	bp->b_wcred = cred;
	237	}
	238	if (bp->b_dirtyend > 0) {
	239	/*
141671b8 KM	240	* If the new write will leave a contiguous dirty
	241	* area, just update the b_dirtyoff and b_dirtyend,
	242	* otherwise force a write rpc of the old dirty area.
39d108be RM	243	*/
	244	if (on <= bp->b_dirtyend && (on+n) >= bp->b_dirtyoff) {
	245	bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
	246	bp->b_dirtyend = MAX((on+n), bp->b_dirtyend);
	247	} else {
4547a923	248	bp->b_proc = p;
141671b8	249	if (error = bwrite(bp))
b71430cc	250	return (error);
141671b8	251	goto again;
39d108be RM	252	}
	253	} else {
	254	bp->b_dirtyoff = on;
	255	bp->b_dirtyend = on+n;
	256	}
141671b8 KM	257	if (error = uiomove(bp->b_un.b_addr + on, n, uio)) {
141671b8 KM	258	brelse(bp);
b71430cc	259	return (error);
141671b8	260	}
170bfd05	261	if ((n+on) == biosize) {
39d108be	262	bp->b_flags \|= B_AGE;
f0f1cbaa	263	bp->b_proc = (struct proc *)0;
39d108be RM	264	bawrite(bp);
39d108be RM	265	} else {
f0f1cbaa	266	bp->b_proc = (struct proc *)0;
39d108be RM	267	bdwrite(bp);
	268	}
	269	} while (error == 0 && uio->uio_resid > 0 && n != 0);
	270	#ifdef notdef
	271	/* Should we try and do this for nfs ?? */
b71430cc	272	if (error && (ioflag & IO_UNIT)) {
39d108be	273	np->n_size = osize;
b71430cc KM	274	uio->uio_offset -= cnt - uio->uio_resid;
	275	uio->uio_resid = cnt;
	276	}
39d108be	277	#endif
39d108be RM	278	return (error);
39d108be RM	279	}