nfs_bio.c

早期freebsd实现

	 * 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;
	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)) {
				if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1))
					return (error);
				np->n_brev = np->n_lrev;
			}
		}
		if (np->n_flag & NQNFSNONCACHE)
			return (nfs_writerpc(vp, uio, cred, ioflag));
		nfsstats.biocache_writes++;
		lbn = uio->uio_offset / biosize;
		on = uio->uio_offset & (biosize-1);
		n = min((unsigned)(biosize - on), uio->uio_resid);
		bn = lbn * (biosize / DEV_BSIZE);
again:
		bp = nfs_getcacheblk(vp, bn, biosize, p);
		if (!bp)
			return (EINTR);
		if (bp->b_wcred == NOCRED) {
			crhold(cred);
			bp->b_wcred = cred;
		}
		np->n_flag |= NMODIFIED;
		if (uio->uio_offset + n > np->n_size) {
			np->n_size = uio->uio_offset + n;
			vnode_pager_setsize(vp, (u_long)np->n_size);
		}

		/*
		 * 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 (VOP_BWRITE(bp) == EINTR)
				return (EINTR);
			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)) {
				brelse(bp);
				if (error = nfs_vinvalbuf(vp, V_SAVE, cred, p, 1))
					return (error);
				np->n_brev = np->n_lrev;
				goto again;
			}
		}
		if (error = uiomove((char *)bp->b_data + on, n, uio)) {
			bp->b_flags |= B_ERROR;
			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;
		}
#ifndef notdef
		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);
		}
#else
		bp->b_validoff = bp->b_dirtyoff;
		bp->b_validend = bp->b_dirtyend;
#endif
		if (ioflag & IO_APPEND)
			bp->b_flags |= B_APPENDWRITE;

		/*
		 * If the lease is non-cachable or IO_SYNC do bwrite().
		 */
		if ((np->n_flag & NQNFSNONCACHE) || (ioflag & IO_SYNC)) {
			bp->b_proc = p;
			if (error = VOP_BWRITE(bp))
				return (error);
		} else if ((n + on) == biosize &&
			(nmp->nm_flag & NFSMNT_NQNFS) == 0) {
			bp->b_proc = (struct proc *)0;
			bawrite(bp);
		} else
			bdwrite(bp);
	} while (uio->uio_resid > 0 && n > 0);
	return (0);
}

/*
 * Get an nfs cache block.
 * Allocate a new one if the block isn't currently in the cache
 * and return the block marked busy. If the calling process is
 * interrupted by a signal for an interruptible mount point, return
 * NULL.
 */
struct buf *
nfs_getcacheblk(vp, bn, size, p)
	struct vnode *vp;
	daddr_t bn;
	int size;
	struct proc *p;
{
	register struct buf *bp;
	struct nfsmount *nmp = VFSTONFS(vp->v_mount);

	if (nmp->nm_flag & NFSMNT_INT) {
		bp = getblk(vp, bn, size, PCATCH, 0);
		while (bp == (struct buf *)0) {
			if (nfs_sigintr(nmp, (struct nfsreq *)0, p))
				return ((struct buf *)0);
			bp = getblk(vp, bn, size, 0, 2 * hz);
		}
	} else
		bp = getblk(vp, bn, size, 0, 0);
	return (bp);
}

/*
 * Flush and invalidate all dirty buffers. If another process is already
 * doing the flush, just wait for completion.
 */
nfs_vinvalbuf(vp, flags, cred, p, intrflg)
	struct vnode *vp;
	int flags;
	struct ucred *cred;
	struct proc *p;
	int intrflg;
{
	register struct nfsnode *np = VTONFS(vp);
	struct nfsmount *nmp = VFSTONFS(vp->v_mount);
	int error = 0, slpflag, slptimeo;

	if ((nmp->nm_flag & NFSMNT_INT) == 0)
		intrflg = 0;
	if (intrflg) {
		slpflag = PCATCH;
		slptimeo = 2 * hz;
	} else {
		slpflag = 0;
		slptimeo = 0;
	}
	/*
	 * First wait for any other process doing a flush to complete.
	 */
	while (np->n_flag & NFLUSHINPROG) {
		np->n_flag |= NFLUSHWANT;
		error = tsleep((caddr_t)&np->n_flag, PRIBIO + 2, "nfsvinval",
			slptimeo);
		if (error && intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p))
			return (EINTR);
	}

	/*
	 * Now, flush as required.
	 */
	np->n_flag |= NFLUSHINPROG;
	error = vinvalbuf(vp, flags, cred, p, slpflag, 0);
	while (error) {
		if (intrflg && nfs_sigintr(nmp, (struct nfsreq *)0, p)) {
			np->n_flag &= ~NFLUSHINPROG;
			if (np->n_flag & NFLUSHWANT) {
				np->n_flag &= ~NFLUSHWANT;
				wakeup((caddr_t)&np->n_flag);
			}
			return (EINTR);
		}
		error = vinvalbuf(vp, flags, cred, p, 0, slptimeo);
	}
	np->n_flag &= ~(NMODIFIED | NFLUSHINPROG);
	if (np->n_flag & NFLUSHWANT) {
		np->n_flag &= ~NFLUSHWANT;
		wakeup((caddr_t)&np->n_flag);
	}
	return (0);
}

/*
 * Initiate asynchronous I/O. Return an error if no nfsiods are available.
 * This is mainly to avoid queueing async I/O requests when the nfsiods
 * are all hung on a dead server.
 */
nfs_asyncio(bp, cred)
	register struct buf *bp;
	struct ucred *cred;
{
	register int i;

	if (nfs_numasync == 0)
		return (EIO);
	for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
	    if (nfs_iodwant[i]) {
		if (bp->b_flags & B_READ) {
			if (bp->b_rcred == NOCRED && cred != NOCRED) {
				crhold(cred);
				bp->b_rcred = cred;
			}
		} else {
			if (bp->b_wcred == NOCRED && cred != NOCRED) {
				crhold(cred);
				bp->b_wcred = cred;
			}
		}
	
		TAILQ_INSERT_TAIL(&nfs_bufq, bp, b_freelist);
		nfs_iodwant[i] = (struct proc *)0;
		wakeup((caddr_t)&nfs_iodwant[i]);
		return (0);
	    }
	return (EIO);
}

/*
 * Do an I/O operation to/from a cache block. This may be called
 * synchronously or from an nfsiod.
 */
int
nfs_doio(bp, cr, p)
	register struct buf *bp;
	struct cred *cr;
	struct proc *p;
{
	register struct uio *uiop;
	register struct vnode *vp;
	struct nfsnode *np;
	struct nfsmount *nmp;
	int error, diff, len;
	struct uio uio;
	struct iovec io;

	vp = bp->b_vp;
	np = VTONFS(vp);
	nmp = VFSTONFS(vp->v_mount);
	uiop = &uio;
	uiop->uio_iov = &io;
	uiop->uio_iovcnt = 1;
	uiop->uio_segflg = UIO_SYSSPACE;
	uiop->uio_procp = p;

	/*
	 * Historically, paging was done with physio, but no more.
	 */
	if (bp->b_flags & B_PHYS)
	    panic("doio phys");
	if (bp->b_flags & B_READ) {
	    io.iov_len = uiop->uio_resid = bp->b_bcount;
	    io.iov_base = bp->b_data;
	    uiop->uio_rw = UIO_READ;
	    switch (vp->v_type) {
	    case VREG:
		uiop->uio_offset = bp->b_blkno * DEV_BSIZE;
		nfsstats.read_bios++;
		error = nfs_readrpc(vp, uiop, cr);
		if (!error) {
		    bp->b_validoff = 0;
		    if (uiop->uio_resid) {
			/*
			 * If len > 0, there is a hole in the file and
			 * no writes after the hole have been pushed to
			 * the server yet.
			 * Just zero fill the rest of the valid area.
			 */
			diff = bp->b_bcount - uiop->uio_resid;
			len = np->n_size - (bp->b_blkno * DEV_BSIZE
				+ diff);
			if (len > 0) {
			    len = min(len, uiop->uio_resid);
			    bzero((char *)bp->b_data + diff, len);
			    bp->b_validend = diff + len;
			} else
			    bp->b_validend = diff;
		    } else
			bp->b_validend = bp->b_bcount;
		}
		if (p && (vp->v_flag & VTEXT) &&
			(((nmp->nm_flag & NFSMNT_NQNFS) &&
			  np->n_lrev != np->n_brev) ||
			 (!(nmp->nm_flag & NFSMNT_NQNFS) &&
			  np->n_mtime != np->n_vattr.va_mtime.ts_sec))) {
			uprintf("Process killed due to text file modification\n");
			psignal(p, SIGKILL);
			p->p_flag |= P_NOSWAP;
		}
		break;
	    case VLNK:
		uiop->uio_offset = 0;
		nfsstats.readlink_bios++;
		error = nfs_readlinkrpc(vp, uiop, cr);
		break;
	    case VDIR:
		uiop->uio_offset = bp->b_lblkno;
		nfsstats.readdir_bios++;
		if (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NQNFS)
		    error = nfs_readdirlookrpc(vp, uiop, cr);
		else
		    error = nfs_readdirrpc(vp, uiop, cr);
		/*
		 * Save offset cookie in b_blkno.
		 */
		bp->b_blkno = uiop->uio_offset;
		break;
	    };
	    if (error) {
		bp->b_flags |= B_ERROR;
		bp->b_error = error;
	    }
	} else {
	    io.iov_len = uiop->uio_resid = bp->b_dirtyend
		- bp->b_dirtyoff;
	    uiop->uio_offset = (bp->b_blkno * DEV_BSIZE)
		+ bp->b_dirtyoff;
	    io.iov_base = (char *)bp->b_data + bp->b_dirtyoff;
	    uiop->uio_rw = UIO_WRITE;
	    nfsstats.write_bios++;
	    if (bp->b_flags & B_APPENDWRITE)
		error = nfs_writerpc(vp, uiop, cr, IO_APPEND);
	    else
		error = nfs_writerpc(vp, uiop, cr, 0);
	    bp->b_flags &= ~(B_WRITEINPROG | B_APPENDWRITE);

	    /*
	     * For an interrupted write, the buffer is still valid and the
	     * write hasn't been pushed to the server yet, so we can't set
	     * B_ERROR and report the interruption by setting B_EINTR. For
	     * the B_ASYNC case, B_EINTR is not relevant, so the rpc attempt
	     * is essentially a noop.
	     */
	    if (error == EINTR) {
		bp->b_flags &= ~B_INVAL;
		bp->b_flags |= B_DELWRI;

		/*
		 * Since for the B_ASYNC case, nfs_bwrite() has reassigned the
		 * buffer to the clean list, we have to reassign it back to the
		 * dirty one. Ugh.
		 */
		if (bp->b_flags & B_ASYNC)
		    reassignbuf(bp, vp);
		else
		    bp->b_flags |= B_EINTR;
	    } else {
		if (error) {
		    bp->b_flags |= B_ERROR;
		    bp->b_error = np->n_error = error;
		    np->n_flag |= NWRITEERR;
		}
		bp->b_dirtyoff = bp->b_dirtyend = 0;
	    }
	}
	bp->b_resid = uiop->uio_resid;
	biodone(bp);
	return (error);
}