[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	7.28 (Berkeley) %G%
 */

#include <sys/param.h>
#include <sys/resourcevar.h>
#include <sys/proc.h>
#include <sys/buf.h>
#include <sys/vnode.h>
#include <sys/trace.h>
#include <sys/mount.h>
#include <sys/kernel.h>
#include <machine/endian.h>
#include <vm/vm.h>
#include <nfs/nfsnode.h>
#include <nfs/rpcv2.h>
#include <nfs/nfsv2.h>
#include <nfs/nfs.h>
#include <nfs/nfsmount.h>
#include <nfs/nqnfs.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;
{
	USES_VOP_GETATTR;
	register struct nfsnode *np = VTONFS(vp);
	register int biosize;
	struct buf *bp;
	struct vattr vattr;
	struct nfsmount *nmp;
	daddr_t lbn, bn, rablock[NFS_MAXRAHEAD];
	int rasize[NFS_MAXRAHEAD], nra, diff, error = 0;
	int 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);
	nmp = VFSTONFS(vp->v_mount);
	biosize = nmp->nm_rsize;
	/*
	 * For nfs, cache consistency can only be maintained approximately.
	 * Although RFC1094 does not specify the criteria, the following is
	 * believed to be compatible with the reference port.
	 * For nqnfs, full cache consistency is maintained within the loop.
	 * For nfs:
	 * 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.
	 * The mount flag NFSMNT_MYWRITE says "Assume that my writes are
	 * the ones changing the modify time.
	 * 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	 * NFS_ATTRTIMEO seconds out of date.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 VOP_GETATTR() call.
	 */
	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) {
		if (np->n_flag & NMODIFIED) {
			np->n_flag &= ~NMODIFIED;
			if ((nmp->nm_flag & NFSMNT_MYWRITE) == 0 ||
			     vp->v_type != VREG)
				vinvalbuf(vp, TRUE, cred, uio->uio_procp);
			np->n_attrstamp = 0;
			np->n_direofoffset = 0;
			if (error = VOP_GETATTR(vp, &vattr, cred, uio->uio_procp))
				return (error);
			np->n_mtime = vattr.va_mtime.ts_sec;
		} else {
			if (error = VOP_GETATTR(vp, &vattr, cred, uio->uio_procp))
				return (error);
			if (np->n_mtime != vattr.va_mtime.ts_sec) {
				np->n_direofoffset = 0;
				vinvalbuf(vp, TRUE, cred, uio->uio_procp);
				np->n_mtime = vattr.va_mtime.ts_sec;
			}
		}
	}
	do {

	    /*
	     * Get a valid lease. If cached data is stale, flush it.
	     */
	    if ((nmp->nm_flag & NFSMNT_NQNFS) &&
		NQNFS_CKINVALID(vp, np, NQL_READ)) {
		do {
			error = nqnfs_getlease(vp, NQL_READ, cred, uio->uio_procp);
		} while (error == NQNFS_EXPIRED);
		if (error)
			return (error);
		if (np->n_lrev != np->n_brev ||
		    ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
			if (vp->v_type == VDIR) {
				np->n_direofoffset = 0;
				cache_purge(vp);
			}
			np->n_flag &= ~NMODIFIED;
			vinvalbuf(vp, TRUE, cred, uio->uio_procp);
			np->n_brev = np->n_lrev;
		}
	    }
	    if (np->n_flag & NQNFSNONCACHE) {
		switch (vp->v_type) {
		case VREG:
			error = nfs_readrpc(vp, uio, cred);
			break;
		case VLNK:
			error = nfs_readlinkrpc(vp, uio, cred);
			break;
		case VDIR:
			error = nfs_readdirrpc(vp, uio, cred);
			break;
		};
		return (error);
	    }
	    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);
		for (nra = 0; nra < nmp->nm_readahead &&
			(lbn + 1 + nra) * biosize < np->n_size; nra++) {
			rablock[nra] = (lbn + 1 + nra) * (biosize / DEV_BSIZE);
			rasize[nra] = biosize;
		}
again:
		if (nra > 0 && lbn >= vp->v_lastr)
			error = breadn(vp, bn, biosize, rablock, rasize, nra,
				cred, &bp);
		else
			error = bread(vp, bn, biosize, cred, &bp);
		if (bp->b_validend > 0) {
			if (on < bp->b_validoff || (on+n) > bp->b_validend) {
				bp->b_flags |= B_INVAL;
				if (bp->b_dirtyend > 0) {
					if ((bp->b_flags & B_DELWRI) == 0)
						panic("nfsbioread");
					(void) bwrite(bp);
				} else
					brelse(bp);
				goto again;
			}
		} else {
			bp->b_validoff = 0;
			bp->b_validend = biosize - bp->b_resid;
		}
		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);
	    }

	    /*
	     * For nqnfs:
	     * Must check for valid lease, since it may have expired while in
	     * bread(). If expired, get a lease.
	     * If data is stale, flush and try again.
	     * nb: If a read rpc is done by bread() or breada() and there is
	     *     no valid lease, a get_lease request will be piggy backed.
	     */
	    if (nmp->nm_flag & NFSMNT_NQNFS) {
		if (NQNFS_CKINVALID(vp, np, NQL_READ)) {
			do {
				error = nqnfs_getlease(vp, NQL_READ, cred, uio->uio_procp);
			} while (error == NQNFS_EXPIRED);
			if (error) {
				brelse(bp);
				return (error);
			}
			if ((np->n_flag & NQNFSNONCACHE) ||
			    np->n_lrev != np->n_brev ||
			    ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
				if (vp->v_type == VDIR) {
					np->n_direofoffset = 0;
					cache_purge(vp);
				}
				brelse(bp);
				np->n_flag &= ~NMODIFIED;
				vinvalbuf(vp, TRUE, cred, uio->uio_procp);
				np->n_brev = np->n_lrev;
				continue;
			}
		} else if ((np->n_flag & NQNFSNONCACHE) ||
		    ((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
			np->n_direofoffset = 0;
			brelse(bp);
			np->n_flag &= ~NMODIFIED;
			vinvalbuf(vp, TRUE, cred, uio->uio_procp);
			np->n_brev = np->n_lrev;
			continue;
		}
	    }
	    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 (ap)
	struct vop_write_args /* {
		struct vnode *a_vp;
		struct uio *a_uio;
		int  a_ioflag;
		struct ucred *a_cred;
	} */ *ap;
{
	USES_VOP_GETATTR;
	register int biosize;
	register struct uio *uio = ap->a_uio;
	struct proc *p = uio->uio_procp;
	register struct vnode *vp = ap->a_vp;
	struct nfsnode *np = VTONFS(vp);
	register struct ucred *cred = ap->a_cred;
	int ioflag = ap->a_ioflag;
	struct buf *bp;
	struct vattr vattr;
	struct nfsmount *nmp;
	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);
	if (np->n_flag & NWRITEERR) {
		np->n_flag &= ~NWRITEERR;
		return (np->n_error);
	}
	if (ioflag & (IO_APPEND | IO_SYNC)) {
		if (np->n_flag & NMODIFIED) {
			np->n_flag &= ~NMODIFIED;
			vinvalbuf(vp, TRUE, cred, p);
		}
		if (ioflag & IO_APPEND) {
			np->n_attrstamp = 0;
			if (error = VOP_GETATTR(vp, &vattr, cred, p))
				return (error);
			uio->uio_offset = np->n_size;
		}
	}
	nmp = VFSTONFS(vp->v_mount);
	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 (p && 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 = nmp->nm_rsize;
	np->n_flag |= NMODIFIED;
	do {

		/*
		 * Check for a valid write lease.
		 * If non-cachable, just do the rpc
		 */
		if ((nmp->nm_flag & NFSMNT_NQNFS) &&
		    NQNFS_CKINVALID(vp, np, NQL_WRITE)) {
			do {
				error = nqnfs_getlease(vp, NQL_WRITE, cred, p);
			} while (error == NQNFS_EXPIRED);
			if (error)
				return (error);
			if (np->n_lrev != np->n_brev ||
			    (np->n_flag & NQNFSNONCACHE)) {
				vinvalbuf(vp, TRUE, cred, p);
				np->n_brev = np->n_lrev;
			}
		}
		if (np->n_flag & NQNFSNONCACHE)
			return (nfs_writerpc(vp, uio, cred));
		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, (u_long)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 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 (bp->b_dirtyend > 0 &&
		    (on > bp->b_dirtyend || (on + n) < bp->b_dirtyoff)) {
			bp->b_proc = p;
			if (error = bwrite(bp))
				return (error);
			goto again;
		}

		/*
		 * Check for valid write lease and get one as required.
		 * In case getblk() and/or bwrite() delayed us.
		 */
		if ((nmp->nm_flag & NFSMNT_NQNFS) &&
		    NQNFS_CKINVALID(vp, np, NQL_WRITE)) {
			do {
				error = nqnfs_getlease(vp, NQL_WRITE, cred, p);
			} while (error == NQNFS_EXPIRED);
			if (error) {
				brelse(bp);
				return (error);
			}
			if (np->n_lrev != np->n_brev ||
			    (np->n_flag & NQNFSNONCACHE)) {
				vinvalbuf(vp, TRUE, cred, p);
				np->n_brev = np->n_lrev;
			}
		}
		if (error = uiomove(bp->b_un.b_addr + on, n, uio)) {
			brelse(bp);
			return (error);
		}
		if (bp->b_dirtyend > 0) {
			bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
			bp->b_dirtyend = MAX((on+n), bp->b_dirtyend);
		} else {
			bp->b_dirtyoff = on;
			bp->b_dirtyend = on+n;
		}
		if (bp->b_validend == 0 || bp->b_validend < bp->b_dirtyoff ||
		    bp->b_validoff > bp->b_dirtyend) {
			bp->b_validoff = bp->b_dirtyoff;
			bp->b_validend = bp->b_dirtyend;
		} else {
			bp->b_validoff = MIN(bp->b_validoff, bp->b_dirtyoff);
			bp->b_validend = MAX(bp->b_validend, bp->b_dirtyend);
		}

		/*
		 * If the lease is non-cachable or IO_SYNC do bwrite().
		 */
		if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) {
			bp->b_proc = p;
			bwrite(bp);
		} else if ((n+on) == biosize &&
			 (nmp->nm_flag & NFSMNT_NQNFS) == 0) {
			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);
	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	*
e19a2ad1	10	* @(#)nfs_bio.c 7.28 (Berkeley) %G%
39d108be RM	11	*/
39d108be RM	12
400a1380 KM	13	#include <sys/param.h>
	14	#include <sys/resourcevar.h>
	15	#include <sys/proc.h>
	16	#include <sys/buf.h>
	17	#include <sys/vnode.h>
	18	#include <sys/trace.h>
	19	#include <sys/mount.h>
	20	#include <sys/kernel.h>
	21	#include <machine/endian.h>
	22	#include <vm/vm.h>
	23	#include <nfs/nfsnode.h>
	24	#include <nfs/rpcv2.h>
	25	#include <nfs/nfsv2.h>
	26	#include <nfs/nfs.h>
	27	#include <nfs/nfsmount.h>
	28	#include <nfs/nqnfs.h>
39d108be RM	29
	30	/* True and false, how exciting */
	31	#define TRUE 1
	32	#define FALSE 0
	33
	34	/*
	35	* Vnode op for read using bio
	36	* Any similarity to readip() is purely coincidental
	37	*/
f0f1cbaa	38	nfs_bioread(vp, uio, ioflag, cred)
39d108be	39	register struct vnode *vp;
170bfd05	40	register struct uio *uio;
39d108be RM	41	int ioflag;
	42	struct ucred *cred;
	43	{
9342689a	44	USES_VOP_GETATTR;
39d108be	45	register struct nfsnode *np = VTONFS(vp);
170bfd05	46	register int biosize;
39d108be RM	47	struct buf *bp;
39d108be RM	48	struct vattr vattr;
2c5b44a2 KM	49	struct nfsmount *nmp;
	50	daddr_t lbn, bn, rablock[NFS_MAXRAHEAD];
	51	int rasize[NFS_MAXRAHEAD], nra, diff, error = 0;
	52	int n, on;
39d108be	53
d4e5799e KM	54	#ifdef lint
	55	ioflag = ioflag;
	56	#endif /* lint */
b40809cc	57	#ifdef DIAGNOSTIC
39d108be RM	58	if (uio->uio_rw != UIO_READ)
39d108be RM	59	panic("nfs_read mode");
b40809cc	60	#endif
39d108be	61	if (uio->uio_resid == 0)
b71430cc	62	return (0);
f0f1cbaa	63	if (uio->uio_offset < 0 && vp->v_type != VDIR)
b71430cc	64	return (EINVAL);
2c5b44a2 KM	65	nmp = VFSTONFS(vp->v_mount);
2c5b44a2 KM	66	biosize = nmp->nm_rsize;
39d108be	67	/*
2c5b44a2 KM	68	* For nfs, cache consistency can only be maintained approximately.
	69	* Although RFC1094 does not specify the criteria, the following is
	70	* believed to be compatible with the reference port.
	71	* For nqnfs, full cache consistency is maintained within the loop.
	72	* For nfs:
39d108be RM	73	* If the file's modify time on the server has changed since the
	74	* last read rpc or you have written to the file,
	75	* you may have lost data cache consistency with the
	76	* server, so flush all of the file's data out of the cache.
f0f1cbaa KM	77	* Then force a getattr rpc to ensure that you have up to date
f0f1cbaa KM	78	* attributes.
2c5b44a2 KM	79	* The mount flag NFSMNT_MYWRITE says "Assume that my writes are
2c5b44a2 KM	80	* the ones changing the modify time.
39d108be RM	81	* NB: This implies that cache data can be read when up to
	82	* NFS_ATTRTIMEO seconds out of date. If you find that you need current
	83	* attributes this could be forced by setting n_attrstamp to 0 before
9342689a	84	* the VOP_GETATTR() call.
39d108be	85	*/
2c5b44a2	86	if ((nmp->nm_flag & NFSMNT_NQNFS) == 0 && vp->v_type != VLNK) {
f0f1cbaa KM	87	if (np->n_flag & NMODIFIED) {
f0f1cbaa KM	88	np->n_flag &= ~NMODIFIED;
2c5b44a2 KM	89	if ((nmp->nm_flag & NFSMNT_MYWRITE) == 0 \|\|
2c5b44a2 KM	90	vp->v_type != VREG)
e19a2ad1	91	vinvalbuf(vp, TRUE, cred, uio->uio_procp);
f0f1cbaa KM	92	np->n_attrstamp = 0;
f0f1cbaa KM	93	np->n_direofoffset = 0;
9342689a	94	if (error = VOP_GETATTR(vp, &vattr, cred, uio->uio_procp))
e8540f59	95	return (error);
7e11a0c9	96	np->n_mtime = vattr.va_mtime.ts_sec;
f0f1cbaa	97	} else {
9342689a	98	if (error = VOP_GETATTR(vp, &vattr, cred, uio->uio_procp))
f0f1cbaa	99	return (error);
7e11a0c9	100	if (np->n_mtime != vattr.va_mtime.ts_sec) {
f0f1cbaa	101	np->n_direofoffset = 0;
e19a2ad1	102	vinvalbuf(vp, TRUE, cred, uio->uio_procp);
7e11a0c9	103	np->n_mtime = vattr.va_mtime.ts_sec;
f0f1cbaa	104	}
39d108be RM	105	}
	106	}
	107	do {
2c5b44a2 KM	108
	109	/*
	110	* Get a valid lease. If cached data is stale, flush it.
	111	*/
	112	if ((nmp->nm_flag & NFSMNT_NQNFS) &&
	113	NQNFS_CKINVALID(vp, np, NQL_READ)) {
	114	do {
	115	error = nqnfs_getlease(vp, NQL_READ, cred, uio->uio_procp);
	116	} while (error == NQNFS_EXPIRED);
	117	if (error)
	118	return (error);
e19a2ad1	119	if (np->n_lrev != np->n_brev \|\|
2c5b44a2 KM	120	((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
	121	if (vp->v_type == VDIR) {
	122	np->n_direofoffset = 0;
	123	cache_purge(vp);
	124	}
	125	np->n_flag &= ~NMODIFIED;
e19a2ad1	126	vinvalbuf(vp, TRUE, cred, uio->uio_procp);
2c5b44a2 KM	127	np->n_brev = np->n_lrev;
	128	}
	129	}
	130	if (np->n_flag & NQNFSNONCACHE) {
	131	switch (vp->v_type) {
	132	case VREG:
	133	error = nfs_readrpc(vp, uio, cred);
	134	break;
	135	case VLNK:
	136	error = nfs_readlinkrpc(vp, uio, cred);
	137	break;
	138	case VDIR:
	139	error = nfs_readdirrpc(vp, uio, cred);
	140	break;
	141	};
	142	return (error);
	143	}
f0f1cbaa KM	144	switch (vp->v_type) {
f0f1cbaa KM	145	case VREG:
e8540f59	146	nfsstats.biocache_reads++;
170bfd05 KM	147	lbn = uio->uio_offset / biosize;
	148	on = uio->uio_offset & (biosize-1);
	149	n = MIN((unsigned)(biosize - on), uio->uio_resid);
39d108be RM	150	diff = np->n_size - uio->uio_offset;
39d108be RM	151	if (diff <= 0)
b71430cc	152	return (error);
39d108be RM	153	if (diff < n)
39d108be RM	154	n = diff;
170bfd05	155	bn = lbn*(biosize/DEV_BSIZE);
2c5b44a2 KM	156	for (nra = 0; nra < nmp->nm_readahead &&
	157	(lbn + 1 + nra) * biosize < np->n_size; nra++) {
	158	rablock[nra] = (lbn + 1 + nra) * (biosize / DEV_BSIZE);
	159	rasize[nra] = biosize;
	160	}
	161	again:
	162	if (nra > 0 && lbn >= vp->v_lastr)
	163	error = breadn(vp, bn, biosize, rablock, rasize, nra,
39d108be RM	164	cred, &bp);
39d108be RM	165	else
170bfd05	166	error = bread(vp, bn, biosize, cred, &bp);
2c5b44a2 KM	167	if (bp->b_validend > 0) {
	168	if (on < bp->b_validoff \|\| (on+n) > bp->b_validend) {
	169	bp->b_flags \|= B_INVAL;
	170	if (bp->b_dirtyend > 0) {
	171	if ((bp->b_flags & B_DELWRI) == 0)
	172	panic("nfsbioread");
	173	(void) bwrite(bp);
	174	} else
	175	brelse(bp);
	176	goto again;
	177	}
	178	} else {
	179	bp->b_validoff = 0;
	180	bp->b_validend = biosize - bp->b_resid;
	181	}
d1a28114	182	vp->v_lastr = lbn;
39d108be	183	if (bp->b_resid) {
170bfd05 KM	184	diff = (on >= (biosize-bp->b_resid)) ? 0 :
170bfd05 KM	185	(biosize-bp->b_resid-on);
f0f1cbaa	186	n = MIN(n, diff);
39d108be	187	}
f0f1cbaa KM	188	break;
	189	case VLNK:
	190	nfsstats.biocache_readlinks++;
	191	on = 0;
	192	error = bread(vp, (daddr_t)0, NFS_MAXPATHLEN, cred, &bp);
	193	n = MIN(uio->uio_resid, NFS_MAXPATHLEN - bp->b_resid);
	194	break;
	195	case VDIR:
	196	nfsstats.biocache_readdirs++;
	197	on = 0;
b40809cc KM	198	error = bread(vp, uio->uio_offset, NFS_DIRBLKSIZ, cred, &bp);
b40809cc KM	199	n = MIN(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid);
f0f1cbaa KM	200	break;
	201	};
	202	if (error) {
	203	brelse(bp);
	204	return (error);
	205	}
2c5b44a2 KM	206
	207	/*
	208	* For nqnfs:
	209	* Must check for valid lease, since it may have expired while in
	210	* bread(). If expired, get a lease.
	211	* If data is stale, flush and try again.
	212	* nb: If a read rpc is done by bread() or breada() and there is
	213	* no valid lease, a get_lease request will be piggy backed.
	214	*/
	215	if (nmp->nm_flag & NFSMNT_NQNFS) {
	216	if (NQNFS_CKINVALID(vp, np, NQL_READ)) {
	217	do {
	218	error = nqnfs_getlease(vp, NQL_READ, cred, uio->uio_procp);
	219	} while (error == NQNFS_EXPIRED);
	220	if (error) {
	221	brelse(bp);
	222	return (error);
	223	}
	224	if ((np->n_flag & NQNFSNONCACHE) \|\|
e19a2ad1	225	np->n_lrev != np->n_brev \|\|
2c5b44a2 KM	226	((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
	227	if (vp->v_type == VDIR) {
	228	np->n_direofoffset = 0;
	229	cache_purge(vp);
	230	}
	231	brelse(bp);
	232	np->n_flag &= ~NMODIFIED;
e19a2ad1	233	vinvalbuf(vp, TRUE, cred, uio->uio_procp);
2c5b44a2 KM	234	np->n_brev = np->n_lrev;
	235	continue;
	236	}
	237	} else if ((np->n_flag & NQNFSNONCACHE) \|\|
	238	((np->n_flag & NMODIFIED) && vp->v_type == VDIR)) {
	239	np->n_direofoffset = 0;
	240	brelse(bp);
	241	np->n_flag &= ~NMODIFIED;
e19a2ad1	242	vinvalbuf(vp, TRUE, cred, uio->uio_procp);
2c5b44a2 KM	243	np->n_brev = np->n_lrev;
	244	continue;
	245	}
	246	}
f0f1cbaa KM	247	if (n > 0)
	248	error = uiomove(bp->b_un.b_addr + on, (int)n, uio);
	249	switch (vp->v_type) {
	250	case VREG:
170bfd05	251	if (n+on == biosize \|\| uio->uio_offset == np->n_size)
39d108be	252	bp->b_flags \|= B_AGE;
f0f1cbaa KM	253	break;
	254	case VLNK:
	255	n = 0;
	256	break;
	257	case VDIR:
	258	uio->uio_offset = bp->b_blkno;
	259	break;
	260	};
	261	brelse(bp);
39d108be	262	} while (error == 0 && uio->uio_resid > 0 && n != 0);
39d108be RM	263	return (error);
	264	}
	265
	266	/*
	267	* Vnode op for write using bio
	268	*/
9342689a	269	nfs_write (ap)
e19a2ad1 KM	270	struct vop_write_args /* {
	271	struct vnode *a_vp;
	272	struct uio *a_uio;
	273	int a_ioflag;
	274	struct ucred *a_cred;
	275	} / ap;
39d108be	276	{
9342689a	277	USES_VOP_GETATTR;
170bfd05	278	register int biosize;
e19a2ad1 KM	279	register struct uio *uio = ap->a_uio;
	280	struct proc *p = uio->uio_procp;
	281	register struct vnode *vp = ap->a_vp;
	282	struct nfsnode *np = VTONFS(vp);
	283	register struct ucred *cred = ap->a_cred;
	284	int ioflag = ap->a_ioflag;
39d108be	285	struct buf *bp;
f0f1cbaa	286	struct vattr vattr;
2c5b44a2	287	struct nfsmount *nmp;
39d108be	288	daddr_t lbn, bn;
3c15394b	289	int n, on, error = 0;
39d108be	290
b40809cc	291	#ifdef DIAGNOSTIC
e19a2ad1	292	if (uio->uio_rw != UIO_WRITE)
f0f1cbaa	293	panic("nfs_write mode");
e19a2ad1	294	if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc)
b40809cc KM	295	panic("nfs_write proc");
b40809cc KM	296	#endif
e19a2ad1	297	if (vp->v_type != VREG)
f0f1cbaa	298	return (EIO);
9b61ab4a KM	299	if (np->n_flag & NWRITEERR) {
	300	np->n_flag &= ~NWRITEERR;
	301	return (np->n_error);
	302	}
e19a2ad1	303	if (ioflag & (IO_APPEND \| IO_SYNC)) {
225498c1 KM	304	if (np->n_flag & NMODIFIED) {
225498c1 KM	305	np->n_flag &= ~NMODIFIED;
e19a2ad1	306	vinvalbuf(vp, TRUE, cred, p);
225498c1	307	}
e19a2ad1	308	if (ioflag & IO_APPEND) {
225498c1	309	np->n_attrstamp = 0;
e19a2ad1	310	if (error = VOP_GETATTR(vp, &vattr, cred, p))
225498c1	311	return (error);
e19a2ad1	312	uio->uio_offset = np->n_size;
225498c1 KM	313	}
225498c1 KM	314	}
e19a2ad1 KM	315	nmp = VFSTONFS(vp->v_mount);
e19a2ad1 KM	316	if (uio->uio_offset < 0)
b71430cc	317	return (EINVAL);
e19a2ad1	318	if (uio->uio_resid == 0)
b71430cc	319	return (0);
39d108be RM	320	/*
	321	* Maybe this should be above the vnode op call, but so long as
	322	* file servers have no limits, i don't think it matters
	323	*/
e19a2ad1	324	if (p && uio->uio_offset + uio->uio_resid >
4547a923 MK	325	p->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
4547a923 MK	326	psignal(p, SIGXFSZ);
b71430cc	327	return (EFBIG);
39d108be	328	}
170bfd05 KM	329	/*
	330	* I use nm_rsize, not nm_wsize so that all buffer cache blocks
	331	* will be the same size within a filesystem. nfs_writerpc will
	332	* still use nm_wsize when sizing the rpc's.
	333	*/
2c5b44a2	334	biosize = nmp->nm_rsize;
f0f1cbaa	335	np->n_flag \|= NMODIFIED;
39d108be	336	do {
2c5b44a2 KM	337
	338	/*
	339	* Check for a valid write lease.
	340	* If non-cachable, just do the rpc
	341	*/
	342	if ((nmp->nm_flag & NFSMNT_NQNFS) &&
e19a2ad1	343	NQNFS_CKINVALID(vp, np, NQL_WRITE)) {
2c5b44a2	344	do {
e19a2ad1	345	error = nqnfs_getlease(vp, NQL_WRITE, cred, p);
2c5b44a2 KM	346	} while (error == NQNFS_EXPIRED);
	347	if (error)
	348	return (error);
e19a2ad1	349	if (np->n_lrev != np->n_brev \|\|
2c5b44a2	350	(np->n_flag & NQNFSNONCACHE)) {
e19a2ad1	351	vinvalbuf(vp, TRUE, cred, p);
2c5b44a2 KM	352	np->n_brev = np->n_lrev;
	353	}
	354	}
	355	if (np->n_flag & NQNFSNONCACHE)
e19a2ad1	356	return (nfs_writerpc(vp, uio, cred));
e8540f59	357	nfsstats.biocache_writes++;
e19a2ad1 KM	358	lbn = uio->uio_offset / biosize;
	359	on = uio->uio_offset & (biosize-1);
	360	n = MIN((unsigned)(biosize - on), uio->uio_resid);
	361	if (uio->uio_offset + n > np->n_size) {
	362	np->n_size = uio->uio_offset + n;
	363	vnode_pager_setsize(vp, (u_long)np->n_size);
8986c97c	364	}
2c5b44a2	365	bn = lbn * (biosize / DEV_BSIZE);
141671b8	366	again:
e19a2ad1	367	bp = getblk(vp, bn, biosize);
39d108be	368	if (bp->b_wcred == NOCRED) {
e19a2ad1 KM	369	crhold(cred);
e19a2ad1 KM	370	bp->b_wcred = cred;
39d108be	371	}
2c5b44a2 KM	372
	373	/*
	374	* If the new write will leave a contiguous dirty
	375	* area, just update the b_dirtyoff and b_dirtyend,
	376	* otherwise force a write rpc of the old dirty area.
	377	*/
	378	if (bp->b_dirtyend > 0 &&
	379	(on > bp->b_dirtyend \|\| (on + n) < bp->b_dirtyoff)) {
	380	bp->b_proc = p;
	381	if (error = bwrite(bp))
	382	return (error);
	383	goto again;
	384	}
	385
	386	/*
	387	* Check for valid write lease and get one as required.
	388	* In case getblk() and/or bwrite() delayed us.
	389	*/
	390	if ((nmp->nm_flag & NFSMNT_NQNFS) &&
e19a2ad1	391	NQNFS_CKINVALID(vp, np, NQL_WRITE)) {
2c5b44a2	392	do {
e19a2ad1	393	error = nqnfs_getlease(vp, NQL_WRITE, cred, p);
2c5b44a2 KM	394	} while (error == NQNFS_EXPIRED);
	395	if (error) {
	396	brelse(bp);
	397	return (error);
	398	}
e19a2ad1	399	if (np->n_lrev != np->n_brev \|\|
2c5b44a2	400	(np->n_flag & NQNFSNONCACHE)) {
e19a2ad1	401	vinvalbuf(vp, TRUE, cred, p);
2c5b44a2	402	np->n_brev = np->n_lrev;
39d108be	403	}
39d108be	404	}
e19a2ad1	405	if (error = uiomove(bp->b_un.b_addr + on, n, uio)) {
141671b8	406	brelse(bp);
b71430cc	407	return (error);
141671b8	408	}
2c5b44a2 KM	409	if (bp->b_dirtyend > 0) {
	410	bp->b_dirtyoff = MIN(on, bp->b_dirtyoff);
	411	bp->b_dirtyend = MAX((on+n), bp->b_dirtyend);
	412	} else {
	413	bp->b_dirtyoff = on;
	414	bp->b_dirtyend = on+n;
	415	}
	416	if (bp->b_validend == 0 \|\| bp->b_validend < bp->b_dirtyoff \|\|
	417	bp->b_validoff > bp->b_dirtyend) {
	418	bp->b_validoff = bp->b_dirtyoff;
	419	bp->b_validend = bp->b_dirtyend;
	420	} else {
	421	bp->b_validoff = MIN(bp->b_validoff, bp->b_dirtyoff);
	422	bp->b_validend = MAX(bp->b_validend, bp->b_dirtyend);
	423	}
	424
	425	/*
	426	* If the lease is non-cachable or IO_SYNC do bwrite().
	427	*/
e19a2ad1	428	if ((np->n_flag & NQNFSNONCACHE) \|\| (ioflag & IO_SYNC)) {
2c5b44a2 KM	429	bp->b_proc = p;
	430	bwrite(bp);
	431	} else if ((n+on) == biosize &&
	432	(nmp->nm_flag & NFSMNT_NQNFS) == 0) {
39d108be	433	bp->b_flags \|= B_AGE;
f0f1cbaa	434	bp->b_proc = (struct proc *)0;
39d108be RM	435	bawrite(bp);
39d108be RM	436	} else {
f0f1cbaa	437	bp->b_proc = (struct proc *)0;
39d108be RM	438	bdwrite(bp);
39d108be RM	439	}
e19a2ad1	440	} while (error == 0 && uio->uio_resid > 0 && n != 0);
39d108be RM	441	return (error);
39d108be RM	442	}