| 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 | * |
| 8 | * %sccs.include.redist.c% |
| 9 | * |
| 10 | * @(#)nfs_bio.c 7.19 (Berkeley) %G% |
| 11 | */ |
| 12 | |
| 13 | #include "param.h" |
| 14 | #include "proc.h" |
| 15 | #include "buf.h" |
| 16 | #include "uio.h" |
| 17 | #include "namei.h" |
| 18 | #include "vnode.h" |
| 19 | #include "trace.h" |
| 20 | #include "mount.h" |
| 21 | #include "resourcevar.h" |
| 22 | |
| 23 | #include "nfsnode.h" |
| 24 | #include "nfsv2.h" |
| 25 | #include "nfs.h" |
| 26 | #include "nfsiom.h" |
| 27 | #include "nfsmount.h" |
| 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 | */ |
| 37 | nfs_bioread(vp, uio, ioflag, cred) |
| 38 | register struct vnode *vp; |
| 39 | register struct uio *uio; |
| 40 | int ioflag; |
| 41 | struct ucred *cred; |
| 42 | { |
| 43 | register struct nfsnode *np = VTONFS(vp); |
| 44 | register int biosize; |
| 45 | struct buf *bp; |
| 46 | struct vattr vattr; |
| 47 | daddr_t lbn, bn, rablock; |
| 48 | int diff, error = 0; |
| 49 | long n, on; |
| 50 | |
| 51 | #ifdef lint |
| 52 | ioflag = ioflag; |
| 53 | #endif /* lint */ |
| 54 | #ifdef DIAGNOSTIC |
| 55 | if (uio->uio_rw != UIO_READ) |
| 56 | panic("nfs_read mode"); |
| 57 | #endif |
| 58 | if (uio->uio_resid == 0) |
| 59 | return (0); |
| 60 | if (uio->uio_offset < 0 && vp->v_type != VDIR) |
| 61 | return (EINVAL); |
| 62 | biosize = VFSTONFS(vp->v_mount)->nm_rsize; |
| 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. |
| 68 | * Then force a getattr rpc to ensure that you have up to date |
| 69 | * attributes. |
| 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| 71 | * NFS_ATTRTIMEO seconds out of date.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 |
| 73 | * the nfs_dogetattr() call. |
| 74 | */ |
| 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; |
| 81 | if (error = nfs_dogetattr(vp, &vattr, cred, 1, |
| 82 | uio->uio_procp)) |
| 83 | return (error); |
| 84 | np->n_mtime = vattr.va_mtime.tv_sec; |
| 85 | } else { |
| 86 | if (error = nfs_dogetattr(vp, &vattr, cred, 1, |
| 87 | uio->uio_procp)) |
| 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 | } |
| 94 | } |
| 95 | } |
| 96 | do { |
| 97 | switch (vp->v_type) { |
| 98 | case VREG: |
| 99 | nfsstats.biocache_reads++; |
| 100 | lbn = uio->uio_offset / biosize; |
| 101 | on = uio->uio_offset & (biosize-1); |
| 102 | n = MIN((unsigned)(biosize - on), uio->uio_resid); |
| 103 | diff = np->n_size - uio->uio_offset; |
| 104 | if (diff <= 0) |
| 105 | return (error); |
| 106 | if (diff < n) |
| 107 | n = diff; |
| 108 | bn = lbn*(biosize/DEV_BSIZE); |
| 109 | rablock = (lbn+1)*(biosize/DEV_BSIZE); |
| 110 | if (vp->v_lastr + 1 == lbn && |
| 111 | np->n_size > (rablock * DEV_BSIZE)) |
| 112 | error = breada(vp, bn, biosize, rablock, biosize, |
| 113 | cred, &bp); |
| 114 | else |
| 115 | error = bread(vp, bn, biosize, cred, &bp); |
| 116 | vp->v_lastr = lbn; |
| 117 | if (bp->b_resid) { |
| 118 | diff = (on >= (biosize-bp->b_resid)) ? 0 : |
| 119 | (biosize-bp->b_resid-on); |
| 120 | n = MIN(n, diff); |
| 121 | } |
| 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; |
| 132 | error = bread(vp, uio->uio_offset, NFS_DIRBLKSIZ, cred, &bp); |
| 133 | n = MIN(uio->uio_resid, NFS_DIRBLKSIZ - bp->b_resid); |
| 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: |
| 144 | if (n+on == biosize || uio->uio_offset == np->n_size) |
| 145 | bp->b_flags |= B_AGE; |
| 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); |
| 155 | } while (error == 0 && uio->uio_resid > 0 && n != 0); |
| 156 | return (error); |
| 157 | } |
| 158 | |
| 159 | /* |
| 160 | * Vnode op for write using bio |
| 161 | */ |
| 162 | nfs_write(vp, uio, ioflag, cred) |
| 163 | register struct vnode *vp; |
| 164 | register struct uio *uio; |
| 165 | int ioflag; |
| 166 | struct ucred *cred; |
| 167 | { |
| 168 | struct proc *p = uio->uio_procp; |
| 169 | register int biosize; |
| 170 | struct buf *bp; |
| 171 | struct nfsnode *np = VTONFS(vp); |
| 172 | struct vattr vattr; |
| 173 | daddr_t lbn, bn; |
| 174 | int n, on, error = 0; |
| 175 | |
| 176 | #ifdef DIAGNOSTIC |
| 177 | if (uio->uio_rw != UIO_WRITE) |
| 178 | panic("nfs_write mode"); |
| 179 | if (uio->uio_segflg == UIO_USERSPACE && uio->uio_procp != curproc) |
| 180 | panic("nfs_write proc"); |
| 181 | #endif |
| 182 | if (vp->v_type != VREG) |
| 183 | return (EIO); |
| 184 | /* Should we try and do this ?? */ |
| 185 | if (ioflag & (IO_APPEND | IO_SYNC)) { |
| 186 | if (np->n_flag & NMODIFIED) { |
| 187 | np->n_flag &= ~NMODIFIED; |
| 188 | vinvalbuf(vp, TRUE); |
| 189 | } |
| 190 | if (ioflag & IO_APPEND) { |
| 191 | np->n_attrstamp = 0; |
| 192 | if (error = nfs_dogetattr(vp, &vattr, cred, 1, p)) |
| 193 | return (error); |
| 194 | uio->uio_offset = np->n_size; |
| 195 | } |
| 196 | return (nfs_writerpc(vp, uio, cred)); |
| 197 | } |
| 198 | #ifdef notdef |
| 199 | cnt = uio->uio_resid; |
| 200 | osize = np->n_size; |
| 201 | #endif |
| 202 | if (uio->uio_offset < 0) |
| 203 | return (EINVAL); |
| 204 | if (uio->uio_resid == 0) |
| 205 | return (0); |
| 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 | */ |
| 210 | if (uio->uio_offset + uio->uio_resid > |
| 211 | p->p_rlimit[RLIMIT_FSIZE].rlim_cur) { |
| 212 | psignal(p, SIGXFSZ); |
| 213 | return (EFBIG); |
| 214 | } |
| 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; |
| 221 | np->n_flag |= NMODIFIED; |
| 222 | do { |
| 223 | nfsstats.biocache_writes++; |
| 224 | lbn = uio->uio_offset / biosize; |
| 225 | on = uio->uio_offset & (biosize-1); |
| 226 | n = MIN((unsigned)(biosize - on), uio->uio_resid); |
| 227 | if (uio->uio_offset+n > np->n_size) { |
| 228 | np->n_size = uio->uio_offset+n; |
| 229 | vnode_pager_setsize(vp, np->n_size); |
| 230 | } |
| 231 | bn = lbn*(biosize/DEV_BSIZE); |
| 232 | again: |
| 233 | bp = getblk(vp, bn, biosize); |
| 234 | if (bp->b_wcred == NOCRED) { |
| 235 | crhold(cred); |
| 236 | bp->b_wcred = cred; |
| 237 | } |
| 238 | if (bp->b_dirtyend > 0) { |
| 239 | /* |
| 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. |
| 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 { |
| 248 | bp->b_proc = p; |
| 249 | if (error = bwrite(bp)) |
| 250 | return (error); |
| 251 | goto again; |
| 252 | } |
| 253 | } else { |
| 254 | bp->b_dirtyoff = on; |
| 255 | bp->b_dirtyend = on+n; |
| 256 | } |
| 257 | if (error = uiomove(bp->b_un.b_addr + on, n, uio)) { |
| 258 | brelse(bp); |
| 259 | return (error); |
| 260 | } |
| 261 | if ((n+on) == biosize) { |
| 262 | bp->b_flags |= B_AGE; |
| 263 | bp->b_proc = (struct proc *)0; |
| 264 | bawrite(bp); |
| 265 | } else { |
| 266 | bp->b_proc = (struct proc *)0; |
| 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 ?? */ |
| 272 | if (error && (ioflag & IO_UNIT)) { |
| 273 | np->n_size = osize; |
| 274 | uio->uio_offset -= cnt - uio->uio_resid; |
| 275 | uio->uio_resid = cnt; |
| 276 | } |
| 277 | #endif |
| 278 | return (error); |
| 279 | } |