[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	7.17 (Berkeley) %G%
 */

#include "param.h"
#include "user.h"
#include "buf.h"
#include "vnode.h"
#include "trace.h"
#include "mount.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 */
	if (uio->uio_rw != UIO_READ)
		panic("nfs_read mode");
	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	 * NFS_ATTRTIMEO seconds out of date.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))
				return (error);
			np->n_mtime = vattr.va_mtime.tv_sec;
		} else {
			if (error = nfs_dogetattr(vp, &vattr, cred, 1))
				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, DIRBLKSIZ, cred, &bp);
		n = MIN(uio->uio_resid, 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;
{
	register int biosize;
	struct buf *bp;
	struct nfsnode *np = VTONFS(vp);
	struct vattr vattr;
	daddr_t lbn, bn;
	int n, on, error = 0;

	if (uio->uio_rw != UIO_WRITE)
		panic("nfs_write mode");
	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))
				return (error);
			uio->uio_offset = np->n_size;
		}
		return (nfs_writerpc(vp, uio, cred, u.u_procp));
	}
#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 >
	      u.u_rlimit[RLIMIT_FSIZE].rlim_cur) {
		psignal(u.u_procp, 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 = u.u_procp;
				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	*
8986c97c	10	* @(#)nfs_bio.c 7.17 (Berkeley) %G%
39d108be RM	11	*/
	12
	13	#include "param.h"
	14	#include "user.h"
	15	#include "buf.h"
	16	#include "vnode.h"
	17	#include "trace.h"
	18	#include "mount.h"
	19	#include "nfsnode.h"
e8540f59 KM	20	#include "nfsv2.h"
e8540f59 KM	21	#include "nfs.h"
39d108be	22	#include "nfsiom.h"
f0f1cbaa	23	#include "nfsmount.h"
39d108be RM	24
	25	/* True and false, how exciting */
	26	#define TRUE 1
	27	#define FALSE 0
	28
	29	/*
	30	* Vnode op for read using bio
	31	* Any similarity to readip() is purely coincidental
	32	*/
f0f1cbaa	33	nfs_bioread(vp, uio, ioflag, cred)
39d108be	34	register struct vnode *vp;
170bfd05	35	register struct uio *uio;
39d108be RM	36	int ioflag;
	37	struct ucred *cred;
	38	{
	39	register struct nfsnode *np = VTONFS(vp);
170bfd05	40	register int biosize;
39d108be RM	41	struct buf *bp;
	42	struct vattr vattr;
	43	daddr_t lbn, bn, rablock;
13576453	44	int diff, error = 0;
39d108be	45	long n, on;
39d108be	46
d4e5799e KM	47	#ifdef lint
	48	ioflag = ioflag;
	49	#endif /* lint */
39d108be RM	50	if (uio->uio_rw != UIO_READ)
39d108be RM	51	panic("nfs_read mode");
39d108be	52	if (uio->uio_resid == 0)
b71430cc	53	return (0);
f0f1cbaa	54	if (uio->uio_offset < 0 && vp->v_type != VDIR)
b71430cc	55	return (EINVAL);
170bfd05	56	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
39d108be RM	57	/*
	58	* If the file's modify time on the server has changed since the
	59	* last read rpc or you have written to the file,
	60	* you may have lost data cache consistency with the
	61	* server, so flush all of the file's data out of the cache.
f0f1cbaa KM	62	* Then force a getattr rpc to ensure that you have up to date
f0f1cbaa KM	63	* attributes.
39d108be RM	64	* NB: This implies that cache data can be read when up to
	65	* NFS_ATTRTIMEO seconds out of date. If you find that you need current
	66	* attributes this could be forced by setting n_attrstamp to 0 before
170bfd05	67	* the nfs_dogetattr() call.
39d108be	68	*/
f0f1cbaa KM	69	if (vp->v_type != VLNK) {
	70	if (np->n_flag & NMODIFIED) {
	71	np->n_flag &= ~NMODIFIED;
	72	vinvalbuf(vp, TRUE);
	73	np->n_attrstamp = 0;
	74	np->n_direofoffset = 0;
170bfd05	75	if (error = nfs_dogetattr(vp, &vattr, cred, 1))
e8540f59 KM	76	return (error);
e8540f59 KM	77	np->n_mtime = vattr.va_mtime.tv_sec;
f0f1cbaa	78	} else {
170bfd05	79	if (error = nfs_dogetattr(vp, &vattr, cred, 1))
f0f1cbaa KM	80	return (error);
	81	if (np->n_mtime != vattr.va_mtime.tv_sec) {
	82	np->n_direofoffset = 0;
	83	vinvalbuf(vp, TRUE);
	84	np->n_mtime = vattr.va_mtime.tv_sec;
	85	}
39d108be RM	86	}
	87	}
	88	do {
f0f1cbaa KM	89	switch (vp->v_type) {
f0f1cbaa KM	90	case VREG:
e8540f59	91	nfsstats.biocache_reads++;
170bfd05 KM	92	lbn = uio->uio_offset / biosize;
	93	on = uio->uio_offset & (biosize-1);
	94	n = MIN((unsigned)(biosize - on), uio->uio_resid);
39d108be RM	95	diff = np->n_size - uio->uio_offset;
39d108be RM	96	if (diff <= 0)
b71430cc	97	return (error);
39d108be RM	98	if (diff < n)
39d108be RM	99	n = diff;
170bfd05 KM	100	bn = lbn*(biosize/DEV_BSIZE);
170bfd05 KM	101	rablock = (lbn+1)*(biosize/DEV_BSIZE);
d1a28114 KM	102	if (vp->v_lastr + 1 == lbn &&
d1a28114 KM	103	np->n_size > (rablock * DEV_BSIZE))
170bfd05	104	error = breada(vp, bn, biosize, rablock, biosize,
39d108be RM	105	cred, &bp);
39d108be RM	106	else
170bfd05	107	error = bread(vp, bn, biosize, cred, &bp);
d1a28114	108	vp->v_lastr = lbn;
39d108be	109	if (bp->b_resid) {
170bfd05 KM	110	diff = (on >= (biosize-bp->b_resid)) ? 0 :
170bfd05 KM	111	(biosize-bp->b_resid-on);
f0f1cbaa	112	n = MIN(n, diff);
39d108be	113	}
f0f1cbaa KM	114	break;
	115	case VLNK:
	116	nfsstats.biocache_readlinks++;
	117	on = 0;
	118	error = bread(vp, (daddr_t)0, NFS_MAXPATHLEN, cred, &bp);
	119	n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
	120	break;
	121	case VDIR:
	122	nfsstats.biocache_readdirs++;
	123	on = 0;
	124	error = bread(vp, uio->uio_offset, DIRBLKSIZ, cred, &bp);
	125	n = MIN(uio->uio_resid, DIRBLKSIZ - bp->b_resid);
	126	break;
	127	};
	128	if (error) {
	129	brelse(bp);
	130	return (error);
	131	}
	132	if (n > 0)
	133	error = uiomove(bp->b_un.b_addr + on, (int)n, uio);
	134	switch (vp->v_type) {
	135	case VREG:
170bfd05	136	if (n+on == biosize \|\| uio->uio_offset == np->n_size)
39d108be	137	bp->b_flags \|= B_AGE;
f0f1cbaa KM	138	break;
	139	case VLNK:
	140	n = 0;
	141	break;
	142	case VDIR:
	143	uio->uio_offset = bp->b_blkno;
	144	break;
	145	};
	146	brelse(bp);
39d108be	147	} while (error == 0 && uio->uio_resid > 0 && n != 0);
39d108be RM	148	return (error);
	149	}
	150
	151	/*
	152	* Vnode op for write using bio
	153	*/
b71430cc	154	nfs_write(vp, uio, ioflag, cred)
39d108be RM	155	register struct vnode *vp;
39d108be RM	156	register struct uio *uio;
39d108be RM	157	int ioflag;
	158	struct ucred *cred;
	159	{
170bfd05	160	register int biosize;
39d108be RM	161	struct buf *bp;
39d108be RM	162	struct nfsnode *np = VTONFS(vp);
f0f1cbaa	163	struct vattr vattr;
39d108be	164	daddr_t lbn, bn;
3c15394b	165	int n, on, error = 0;
39d108be	166
f0f1cbaa KM	167	if (uio->uio_rw != UIO_WRITE)
	168	panic("nfs_write mode");
	169	if (vp->v_type != VREG)
	170	return (EIO);
39d108be	171	/* Should we try and do this ?? */
170bfd05	172	if (ioflag & (IO_APPEND \| IO_SYNC)) {
f0f1cbaa KM	173	if (np->n_flag & NMODIFIED) {
	174	np->n_flag &= ~NMODIFIED;
	175	vinvalbuf(vp, TRUE);
	176	}
170bfd05 KM	177	if (ioflag & IO_APPEND) {
	178	np->n_attrstamp = 0;
	179	if (error = nfs_dogetattr(vp, &vattr, cred, 1))
	180	return (error);
	181	uio->uio_offset = np->n_size;
	182	}
f0f1cbaa KM	183	return (nfs_writerpc(vp, uio, cred, u.u_procp));
f0f1cbaa KM	184	}
39d108be	185	#ifdef notdef
b71430cc	186	cnt = uio->uio_resid;
39d108be RM	187	osize = np->n_size;
39d108be RM	188	#endif
b71430cc KM	189	if (uio->uio_offset < 0)
b71430cc KM	190	return (EINVAL);
39d108be	191	if (uio->uio_resid == 0)
b71430cc	192	return (0);
39d108be RM	193	/*
	194	* Maybe this should be above the vnode op call, but so long as
	195	* file servers have no limits, i don't think it matters
	196	*/
f0f1cbaa	197	if (uio->uio_offset + uio->uio_resid >
39d108be RM	198	u.u_rlimit[RLIMIT_FSIZE].rlim_cur) {
39d108be RM	199	psignal(u.u_procp, SIGXFSZ);
b71430cc	200	return (EFBIG);
39d108be	201	}
170bfd05 KM	202	/*
	203	* I use nm_rsize, not nm_wsize so that all buffer cache blocks
	204	* will be the same size within a filesystem. nfs_writerpc will
	205	* still use nm_wsize when sizing the rpc's.
	206	*/
	207	biosize = VFSTONFS(vp->v_mount)->nm_rsize;
f0f1cbaa	208	np->n_flag \|= NMODIFIED;
39d108be	209	do {
e8540f59	210	nfsstats.biocache_writes++;
170bfd05 KM	211	lbn = uio->uio_offset / biosize;
	212	on = uio->uio_offset & (biosize-1);
	213	n = MIN((unsigned)(biosize - on), uio->uio_resid);
8986c97c	214	if (uio->uio_offset+n > np->n_size) {
39d108be	215	np->n_size = uio->uio_offset+n;
8986c97c KM	216	vnode_pager_setsize(vp, np->n_size);
8986c97c KM	217	}
170bfd05	218	bn = lbn*(biosize/DEV_BSIZE);
141671b8	219	again:
170bfd05	220	bp = getblk(vp, bn, biosize);
39d108be RM	221	if (bp->b_wcred == NOCRED) {
	222	crhold(cred);
	223	bp->b_wcred = cred;
	224	}
	225	if (bp->b_dirtyend > 0) {
	226	/*
141671b8 KM	227	* If the new write will leave a contiguous dirty
	228	* area, just update the b_dirtyoff and b_dirtyend,
	229	* otherwise force a write rpc of the old dirty area.
39d108be RM	230	*/
	231	if (on <= bp->b_dirtyend && (on+n) >= bp->b_dirtyoff) {
	232	bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
	233	bp->b_dirtyend = MAX((on+n), bp->b_dirtyend);
	234	} else {
f0f1cbaa	235	bp->b_proc = u.u_procp;
141671b8	236	if (error = bwrite(bp))
b71430cc	237	return (error);
141671b8	238	goto again;
39d108be RM	239	}
	240	} else {
	241	bp->b_dirtyoff = on;
	242	bp->b_dirtyend = on+n;
	243	}
141671b8 KM	244	if (error = uiomove(bp->b_un.b_addr + on, n, uio)) {
141671b8 KM	245	brelse(bp);
b71430cc	246	return (error);
141671b8	247	}
170bfd05	248	if ((n+on) == biosize) {
39d108be	249	bp->b_flags \|= B_AGE;
f0f1cbaa	250	bp->b_proc = (struct proc *)0;
39d108be RM	251	bawrite(bp);
39d108be RM	252	} else {
f0f1cbaa	253	bp->b_proc = (struct proc *)0;
39d108be RM	254	bdwrite(bp);
	255	}
	256	} while (error == 0 && uio->uio_resid > 0 && n != 0);
	257	#ifdef notdef
	258	/* Should we try and do this for nfs ?? */
b71430cc	259	if (error && (ioflag & IO_UNIT)) {
39d108be	260	np->n_size = osize;
b71430cc KM	261	uio->uio_offset -= cnt - uio->uio_resid;
	262	uio->uio_resid = cnt;
	263	}
39d108be	264	#endif
39d108be RM	265	return (error);
39d108be RM	266	}