| 1 | /* |
| 2 | * Copyright 2000, International Business Machines Corporation and others. |
| 3 | * All Rights Reserved. |
| 4 | * |
| 5 | * This software has been released under the terms of the IBM Public |
| 6 | * License. For details, see the LICENSE file in the top-level source |
| 7 | * directory or online at http://www.openafs.org/dl/license10.html |
| 8 | */ |
| 9 | |
| 10 | /* This is a placeholder for routines unique to the port of AFS to hp-ux*/ |
| 11 | |
| 12 | #include <afsconfig.h> |
| 13 | #include "afs/param.h" |
| 14 | |
| 15 | |
| 16 | #include "afs/sysincludes.h" /* Standard vendor system headers */ |
| 17 | #include "afsincludes.h" /* Afs-based standard headers */ |
| 18 | #include "afs/afs_stats.h" /* statistics stuff */ |
| 19 | |
| 20 | #include <sys/uio.h> |
| 21 | #include <sys/vfs.h> |
| 22 | #include <sys/mount.h> |
| 23 | #include <sys/vnode.h> |
| 24 | #include <sys/pathname.h> |
| 25 | |
| 26 | extern struct vfsops Afs_vfsops; |
| 27 | extern int afs_hp_strategy(); |
| 28 | extern int afs_bmap(), afs_badop(), afs_noop(), afs_lockf(); |
| 29 | extern int afs_pagein(); |
| 30 | extern int afs_pageout(); |
| 31 | extern int afs_ioctl(); |
| 32 | extern int afs_prealloc(); |
| 33 | extern int afs_mapdbd(); |
| 34 | extern int afs_mmap(); |
| 35 | extern int afs_cachelimit(); |
| 36 | extern int afs_vm_checkpage(); |
| 37 | extern int afs_vm_fscontiguous(); |
| 38 | extern int afs_vm_stopio(); |
| 39 | extern int afs_read_ahead(); |
| 40 | extern int afs_unmap(); |
| 41 | extern int afs_release(); |
| 42 | extern int afs_swapfs_len(); |
| 43 | extern int afs_readdir2(); |
| 44 | extern int afs_readdir(); |
| 45 | extern int afs_readdir3(); |
| 46 | extern int afs_pathconf(); |
| 47 | extern int afs_close(); |
| 48 | |
| 49 | #define vtoblksz(vp) ((vp)->v_vfsp->vfs_bsize) |
| 50 | |
| 51 | #if defined(AFS_HPUX110_ENV) |
| 52 | /* We no longer need to lock on the VM Empire, |
| 53 | * or at least that is what is claimed. |
| 54 | * so we will noopt the vmemp_ routines |
| 55 | * This needs to be looked at closer. |
| 56 | */ |
| 57 | #define vmemp_lockx() |
| 58 | #undef vmemp_returnx |
| 59 | #define vmemp_returnx(a) return(a) |
| 60 | #define vmemp_unlockx() |
| 61 | #endif |
| 62 | |
| 63 | #if !defined(AFS_HPUX110_ENV) |
| 64 | /* |
| 65 | * Copy an mbuf to the contiguous area pointed to by cp. |
| 66 | * Skip <off> bytes and copy <len> bytes. |
| 67 | * Returns the number of bytes not transferred. |
| 68 | * The mbuf is NOT changed. |
| 69 | */ |
| 70 | int |
| 71 | m_cpytoc(m, off, len, cp) |
| 72 | struct mbuf *m; |
| 73 | int off, len; |
| 74 | caddr_t cp; |
| 75 | { |
| 76 | int ml; |
| 77 | |
| 78 | if (m == NULL || off < 0 || len < 0 || cp == NULL) |
| 79 | osi_Panic("m_cpytoc"); |
| 80 | while (off && m) |
| 81 | if (m->m_len <= off) { |
| 82 | off -= m->m_len; |
| 83 | m = m->m_next; |
| 84 | continue; |
| 85 | } else |
| 86 | break; |
| 87 | if (m == NULL) |
| 88 | return (len); |
| 89 | |
| 90 | ml = MIN(len, m->m_len - off); |
| 91 | memcpy(cp, mtod(m, caddr_t) + off, (u_int) ml); |
| 92 | cp += ml; |
| 93 | len -= ml; |
| 94 | m = m->m_next; |
| 95 | |
| 96 | while (len && m) { |
| 97 | ml = m->m_len; |
| 98 | memcpy(cp, mtod(m, caddr_t), (u_int) ml); |
| 99 | cp += ml; |
| 100 | len -= ml; |
| 101 | m = m->m_next; |
| 102 | } |
| 103 | |
| 104 | return (len); |
| 105 | } |
| 106 | #endif |
| 107 | |
| 108 | /* |
| 109 | * Note that the standard Sun vnode interface doesn't haven't an vop_lockf(), so this code is |
| 110 | * totally new. This came about because HP-UX has lockf() implemented as |
| 111 | * a system call while Sun has it implemented as a library (apparently). |
| 112 | * To handle this, we have to translate the lockf() request into an |
| 113 | * fcntl() looking request, and then translate the results back if necessary. |
| 114 | * we call afs_lockctl() directly . |
| 115 | */ |
| 116 | afs_lockf(vp, flag, len, cred, fp, LB, UB) |
| 117 | struct vnode *vp; |
| 118 | int flag; |
| 119 | afs_ucred_t *cred; |
| 120 | struct file *fp; |
| 121 | k_off_t len, LB, UB; |
| 122 | { |
| 123 | /*for now, just pretend it works */ |
| 124 | struct k_flock flock; |
| 125 | int cmd, code; |
| 126 | |
| 127 | /* |
| 128 | * Create a flock structure and translate the lockf request |
| 129 | * into an appropriate looking fcntl() type request for afs_lockctl() |
| 130 | */ |
| 131 | flock.l_whence = 0; |
| 132 | flock.l_len = len; |
| 133 | flock.l_start = fp->f_offset; |
| 134 | /* convert negative lengths to positive */ |
| 135 | if (flock.l_len < 0) { |
| 136 | flock.l_start += flock.l_len; |
| 137 | flock.l_len = -(flock.l_len); |
| 138 | } |
| 139 | /* |
| 140 | * Adjust values to look like fcntl() requests. |
| 141 | * All locks are write locks, only F_LOCK requests |
| 142 | * are blocking. F_TEST has to be translated into |
| 143 | * a get lock and then back again. |
| 144 | */ |
| 145 | flock.l_type = F_WRLCK; |
| 146 | cmd = F_SETLK; |
| 147 | switch (flag) { |
| 148 | case F_ULOCK: |
| 149 | flock.l_type = F_UNLCK; |
| 150 | break; |
| 151 | case F_LOCK: |
| 152 | cmd = F_SETLKW; |
| 153 | break; |
| 154 | case F_TEST: |
| 155 | cmd = F_GETLK; |
| 156 | break; |
| 157 | } |
| 158 | u.u_error = mp_afs_lockctl(vp, &flock, cmd, fp->f_cred); |
| 159 | if (u.u_error) { |
| 160 | return (u.u_error); /* some other error code */ |
| 161 | } |
| 162 | /* |
| 163 | * if request is F_TEST, and GETLK changed |
| 164 | * the lock type to ULOCK, then return 0, else |
| 165 | * set errno to EACCESS and return. |
| 166 | */ |
| 167 | if (flag == F_TEST && flock.l_type != F_UNLCK) { |
| 168 | u.u_error = EACCES; |
| 169 | return (u.u_error); |
| 170 | } |
| 171 | return (0); |
| 172 | } |
| 173 | |
| 174 | |
| 175 | #if defined(AFS_HPUX1122_ENV) |
| 176 | #include "machine/vm/vmparam.h" |
| 177 | #else |
| 178 | #include "../machine/vmparam.h" /* For KERNELSPACE */ |
| 179 | #endif |
| 180 | #include "h/debug.h" |
| 181 | #include "h/types.h" |
| 182 | #if !defined(AFS_HPUX1123_ENV) |
| 183 | /* 11.23 is using 64 bit in many cases */ |
| 184 | #define kern_daddr_t daddr_t |
| 185 | #endif |
| 186 | #include "h/param.h" |
| 187 | #include "h/vmmac.h" |
| 188 | #include "h/time.h" |
| 189 | #include "ufs/inode.h" |
| 190 | #include "ufs/fs.h" |
| 191 | #include "h/dbd.h" |
| 192 | #if defined(AFS_HPUX1123_ENV) |
| 193 | dbd_t *finddbd(); |
| 194 | #endif /* AFS_HPUX1123_ENV */ |
| 195 | #include "h/vfd.h" |
| 196 | #include "h/region.h" |
| 197 | #include "h/pregion.h" |
| 198 | #include "h/vmmeter.h" |
| 199 | #include "h/user.h" |
| 200 | #include "h/sysinfo.h" |
| 201 | #include "h/pfdat.h" |
| 202 | #if !defined(AFS_HPUX1123_ENV) |
| 203 | #include "h/tuneable.h" |
| 204 | #endif |
| 205 | #include "h/buf.h" |
| 206 | #include "netinet/in.h" |
| 207 | |
| 208 | /* a freelist of one */ |
| 209 | struct buf *afs_bread_freebp = 0; |
| 210 | |
| 211 | /* |
| 212 | * Only rfs_read calls this, and it only looks at bp->b_un.b_addr. |
| 213 | * Thus we can use fake bufs (ie not from the real buffer pool). |
| 214 | */ |
| 215 | afs_bread(vp, lbn, bpp) |
| 216 | struct vnode *vp; |
| 217 | kern_daddr_t lbn; |
| 218 | struct buf **bpp; |
| 219 | { |
| 220 | int offset, fsbsize, error; |
| 221 | struct buf *bp; |
| 222 | struct iovec iov; |
| 223 | struct uio uio; |
| 224 | |
| 225 | memset(&uio, 0, sizeof(uio)); |
| 226 | memset(&iov, 0, sizeof(iov)); |
| 227 | |
| 228 | AFS_STATCNT(afs_bread); |
| 229 | fsbsize = vp->v_vfsp->vfs_bsize; |
| 230 | offset = lbn * fsbsize; |
| 231 | if (afs_bread_freebp) { |
| 232 | bp = afs_bread_freebp; |
| 233 | afs_bread_freebp = 0; |
| 234 | } else { |
| 235 | bp = (struct buf *)AFS_KALLOC(sizeof(*bp)); |
| 236 | bp->b_un.b_addr = (caddr_t) AFS_KALLOC(fsbsize); |
| 237 | } |
| 238 | |
| 239 | iov.iov_base = bp->b_un.b_addr; |
| 240 | iov.iov_len = fsbsize; |
| 241 | uio.afsio_iov = &iov; |
| 242 | uio.afsio_iovcnt = 1; |
| 243 | uio.afsio_seg = AFS_UIOSYS; |
| 244 | uio.afsio_offset = offset; |
| 245 | uio.afsio_resid = fsbsize; |
| 246 | uio.uio_fpflags = 0; |
| 247 | *bpp = 0; |
| 248 | |
| 249 | error = afs_read(VTOAFS(vp), &uio, p_cred(u.u_procp), 0); |
| 250 | if (error) { |
| 251 | afs_bread_freebp = bp; |
| 252 | return error; |
| 253 | } |
| 254 | if (*bpp) { |
| 255 | afs_bread_freebp = bp; |
| 256 | } else { |
| 257 | *(struct buf **)&bp->b_vp = bp; /* mark as fake */ |
| 258 | *bpp = bp; |
| 259 | } |
| 260 | return 0; |
| 261 | } |
| 262 | |
| 263 | afs_brelse(vp, bp) |
| 264 | struct vnode *vp; |
| 265 | struct buf *bp; |
| 266 | { |
| 267 | AFS_STATCNT(afs_brelse); |
| 268 | |
| 269 | if ((struct buf *)bp->b_vp != bp) { /* not fake */ |
| 270 | ufs_brelse(bp->b_vp, bp); |
| 271 | } else if (afs_bread_freebp) { |
| 272 | AFS_KFREE(bp->b_un.b_addr, vp->v_vfsp->vfs_bsize); |
| 273 | AFS_KFREE(bp, sizeof(*bp)); |
| 274 | } else { |
| 275 | afs_bread_freebp = bp; |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | |
| 280 | afs_bmap(avc, abn, anvp, anbn) |
| 281 | struct vcache *avc; |
| 282 | kern_daddr_t abn, *anbn; |
| 283 | struct vcache **anvp; |
| 284 | { |
| 285 | AFS_STATCNT(afs_bmap); |
| 286 | if (anvp) |
| 287 | *anvp = avc; |
| 288 | if (anbn) |
| 289 | *anbn = abn * (8192 / DEV_BSIZE); /* in 512 byte units */ |
| 290 | return 0; |
| 291 | } |
| 292 | |
| 293 | afs_inactive(avc, acred) |
| 294 | struct vcache *avc; |
| 295 | afs_ucred_t *acred; |
| 296 | { |
| 297 | struct vnode *vp = AFSTOV(avc); |
| 298 | ulong_t context; |
| 299 | lock_t *sv_lock; |
| 300 | if (afs_shuttingdown != AFS_RUNNING) |
| 301 | return; |
| 302 | |
| 303 | /* |
| 304 | * In Solaris and HPUX s800 and HP-UX10.0 they actually call us with |
| 305 | * v_count 1 on last reference! |
| 306 | */ |
| 307 | MP_H_SPINLOCK_USAV(vn_h_sl_pool, vp, &sv_lock, &context); |
| 308 | if (avc->vrefCount < 1) |
| 309 | osi_Panic("afs_inactive : v_count < 1\n"); |
| 310 | |
| 311 | /* |
| 312 | * If more than 1 don't unmap the vnode but do decrement the ref count |
| 313 | */ |
| 314 | vp->v_count--; |
| 315 | if (vp->v_count > 0) { |
| 316 | MP_SPINUNLOCK_USAV(sv_lock, context); |
| 317 | return 0; |
| 318 | } |
| 319 | MP_SPINUNLOCK_USAV(sv_lock, context); |
| 320 | afs_InactiveVCache(avc, acred); |
| 321 | return 0; |
| 322 | } |
| 323 | |
| 324 | |
| 325 | int |
| 326 | mp_afs_open(struct vnode **avcp, int aflags, afs_ucred_t *acred) |
| 327 | { |
| 328 | int code; |
| 329 | |
| 330 | AFS_GLOCK(); |
| 331 | code = afs_open(avcp, aflags, acred); |
| 332 | AFS_GUNLOCK(); |
| 333 | return (code); |
| 334 | } |
| 335 | |
| 336 | int |
| 337 | mp_afs_close(struct vnode *avcp, int aflags, afs_ucred_t *acred) |
| 338 | { |
| 339 | int code; |
| 340 | |
| 341 | AFS_GLOCK(); |
| 342 | code = afs_close(avcp, aflags, acred); |
| 343 | AFS_GUNLOCK(); |
| 344 | return (code); |
| 345 | } |
| 346 | |
| 347 | int |
| 348 | mp_afs_rdwr(struct vnode *avcp, struct uio *uio, enum uio_rw arw, |
| 349 | int aio, afs_ucred_t *acred) |
| 350 | { |
| 351 | int code; |
| 352 | long save_resid; |
| 353 | |
| 354 | AFS_GLOCK(); |
| 355 | save_resid = uio->uio_resid; |
| 356 | code = afs_rdwr(avcp, uio, arw, aio, acred); |
| 357 | if (arw == UIO_WRITE && code == ENOSPC) { |
| 358 | /* HP clears code if any data written. */ |
| 359 | uio->uio_resid = save_resid; |
| 360 | } |
| 361 | AFS_GUNLOCK(); |
| 362 | return (code); |
| 363 | } |
| 364 | |
| 365 | int |
| 366 | mp_afs_getattr(struct vnode *avcp, struct vattr *attrs, |
| 367 | afs_ucred_t *acred, enum vsync unused1) |
| 368 | { |
| 369 | int code; |
| 370 | |
| 371 | AFS_GLOCK(); |
| 372 | code = afs_getattr(avcp, attrs, acred); |
| 373 | AFS_GUNLOCK(); |
| 374 | return (code); |
| 375 | } |
| 376 | |
| 377 | int |
| 378 | mp_afs_setattr(struct vnode *avcp, struct vattr *attrs, |
| 379 | afs_ucred_t *acred, int unused1) |
| 380 | { |
| 381 | int code; |
| 382 | |
| 383 | AFS_GLOCK(); |
| 384 | code = afs_setattr(avcp, attrs, acred); |
| 385 | AFS_GUNLOCK(); |
| 386 | return (code); |
| 387 | } |
| 388 | |
| 389 | int |
| 390 | mp_afs_access(struct vnode *avcp, int mode, afs_ucred_t *acred) |
| 391 | { |
| 392 | int code; |
| 393 | |
| 394 | AFS_GLOCK(); |
| 395 | code = afs_access(avcp, mode, acred); |
| 396 | AFS_GUNLOCK(); |
| 397 | return (code); |
| 398 | } |
| 399 | |
| 400 | int |
| 401 | mp_afs_lookup(struct vnode *adp, char *aname, |
| 402 | struct vnode **avcp, afs_ucred_t *acred, |
| 403 | struct vnode *unused1) |
| 404 | { |
| 405 | int code; |
| 406 | |
| 407 | AFS_GLOCK(); |
| 408 | code = afs_lookup(adp, aname, avcp, acred); |
| 409 | AFS_GUNLOCK(); |
| 410 | return (code); |
| 411 | } |
| 412 | |
| 413 | int |
| 414 | mp_afs_create(struct vnode *adp, char *aname, struct vattr *attrs, |
| 415 | enum vcexcl aexcl, int amode, struct vnode **avcp, |
| 416 | afs_ucred_t *acred) |
| 417 | { |
| 418 | int code; |
| 419 | |
| 420 | AFS_GLOCK(); |
| 421 | code = afs_create(adp, aname, attrs, aexcl, amode, avcp, acred); |
| 422 | AFS_GUNLOCK(); |
| 423 | return (code); |
| 424 | } |
| 425 | |
| 426 | |
| 427 | int |
| 428 | mp_afs_remove(struct vnode *adp, char *aname, |
| 429 | afs_ucred_t *acred) |
| 430 | { |
| 431 | int code; |
| 432 | |
| 433 | AFS_GLOCK(); |
| 434 | code = afs_remove(adp, aname, acred); |
| 435 | AFS_GUNLOCK(); |
| 436 | return (code); |
| 437 | } |
| 438 | |
| 439 | int |
| 440 | mp_afs_link(struct vnode *avc, struct vnode *adp, |
| 441 | char *aname, afs_ucred_t *acred) |
| 442 | { |
| 443 | int code; |
| 444 | |
| 445 | AFS_GLOCK(); |
| 446 | code = afs_link(avc, adp, aname, acred); |
| 447 | AFS_GUNLOCK(); |
| 448 | return (code); |
| 449 | } |
| 450 | |
| 451 | int |
| 452 | mp_afs_rename(struct vnode *aodp, char *aname1, |
| 453 | struct vnode *andp, char *aname2, |
| 454 | afs_ucred_t *acred) |
| 455 | { |
| 456 | int code; |
| 457 | |
| 458 | AFS_GLOCK(); |
| 459 | code = afs_rename(aodp, aname1, andp, aname2, acred); |
| 460 | AFS_GUNLOCK(); |
| 461 | return (code); |
| 462 | } |
| 463 | |
| 464 | int |
| 465 | mp_afs_mkdir(struct vnode *adp, char *aname, struct vattr *attrs, |
| 466 | struct vnode **avcp, afs_ucred_t *acred) |
| 467 | { |
| 468 | int code; |
| 469 | |
| 470 | AFS_GLOCK(); |
| 471 | code = afs_mkdir(adp, aname, attrs, avcp, acred); |
| 472 | AFS_GUNLOCK(); |
| 473 | return (code); |
| 474 | } |
| 475 | |
| 476 | |
| 477 | int |
| 478 | mp_afs_rmdir(struct vnode *adp, char *aname, afs_ucred_t *acred) |
| 479 | { |
| 480 | int code; |
| 481 | |
| 482 | AFS_GLOCK(); |
| 483 | code = afs_rmdir(adp, aname, acred); |
| 484 | AFS_GUNLOCK(); |
| 485 | return (code); |
| 486 | } |
| 487 | |
| 488 | |
| 489 | int |
| 490 | mp_afs_readdir(struct vnode *avc, struct uio *auio, |
| 491 | afs_ucred_t *acred) |
| 492 | { |
| 493 | int code; |
| 494 | |
| 495 | AFS_GLOCK(); |
| 496 | code = afs_readdir(avc, auio, acred); |
| 497 | AFS_GUNLOCK(); |
| 498 | return (code); |
| 499 | } |
| 500 | |
| 501 | int |
| 502 | mp_afs_symlink(struct vnode *adp, char *aname, struct vattr *attrs, |
| 503 | char *atargetName, afs_ucred_t *acred) |
| 504 | { |
| 505 | int code; |
| 506 | |
| 507 | AFS_GLOCK(); |
| 508 | code = afs_symlink(adp, aname, attrs, atargetName, NULL, acred); |
| 509 | AFS_GUNLOCK(); |
| 510 | return (code); |
| 511 | } |
| 512 | |
| 513 | |
| 514 | int |
| 515 | mp_afs_readlink(struct vnode *avc, struct uio *auio, |
| 516 | afs_ucred_t *acred) |
| 517 | { |
| 518 | int code; |
| 519 | |
| 520 | AFS_GLOCK(); |
| 521 | code = afs_readlink(avc, auio, acred); |
| 522 | AFS_GUNLOCK(); |
| 523 | return (code); |
| 524 | } |
| 525 | |
| 526 | int |
| 527 | mp_afs_fsync(struct vnode *avc, afs_ucred_t *acred, int unused1) |
| 528 | { |
| 529 | int code; |
| 530 | |
| 531 | AFS_GLOCK(); |
| 532 | code = afs_fsync(avc, acred); |
| 533 | AFS_GUNLOCK(); |
| 534 | return (code); |
| 535 | } |
| 536 | |
| 537 | int |
| 538 | mp_afs_bread(struct vnode *avc, kern_daddr_t lbn, struct buf **bpp, |
| 539 | struct vattr *unused1, struct ucred *unused2) |
| 540 | { |
| 541 | int code; |
| 542 | |
| 543 | AFS_GLOCK(); |
| 544 | code = afs_bread(avc, lbn, bpp); |
| 545 | AFS_GUNLOCK(); |
| 546 | return (code); |
| 547 | } |
| 548 | |
| 549 | int |
| 550 | mp_afs_brelse(struct vnode *avc, struct buf *bp) |
| 551 | { |
| 552 | int code; |
| 553 | |
| 554 | AFS_GLOCK(); |
| 555 | code = afs_brelse(avc, bp); |
| 556 | AFS_GUNLOCK(); |
| 557 | return (code); |
| 558 | } |
| 559 | |
| 560 | |
| 561 | int |
| 562 | mp_afs_inactive(struct vnode *avc, afs_ucred_t *acred) |
| 563 | { |
| 564 | int code; |
| 565 | |
| 566 | AFS_GLOCK(); |
| 567 | code = afs_inactive(avc, acred); |
| 568 | AFS_GUNLOCK(); |
| 569 | return (code); |
| 570 | } |
| 571 | |
| 572 | int |
| 573 | mp_afs_lockctl(struct vnode *avc, struct flock *af, int cmd, |
| 574 | afs_ucred_t *acred, struct file *unused1, off_t unused2, |
| 575 | off_t unused3) |
| 576 | { |
| 577 | int code; |
| 578 | |
| 579 | AFS_GLOCK(); |
| 580 | code = afs_lockctl(avc, af, cmd, acred); |
| 581 | AFS_GUNLOCK(); |
| 582 | return (code); |
| 583 | } |
| 584 | |
| 585 | int |
| 586 | mp_afs_fid(struct vnode *avc, struct fid **fidpp) |
| 587 | { |
| 588 | int code; |
| 589 | |
| 590 | AFS_GLOCK(); |
| 591 | code = afs_fid(avc, fidpp); |
| 592 | AFS_GUNLOCK(); |
| 593 | return (code); |
| 594 | } |
| 595 | |
| 596 | int |
| 597 | mp_afs_readdir2(struct vnode *avc, struct uio *auio, |
| 598 | afs_ucred_t *acred) |
| 599 | { |
| 600 | int code; |
| 601 | |
| 602 | AFS_GLOCK(); |
| 603 | code = afs_readdir2(avc, auio, acred); |
| 604 | AFS_GUNLOCK(); |
| 605 | return (code); |
| 606 | } |
| 607 | |
| 608 | |
| 609 | struct vnodeops Afs_vnodeops = { |
| 610 | mp_afs_open, |
| 611 | mp_afs_close, |
| 612 | mp_afs_rdwr, |
| 613 | afs_ioctl, |
| 614 | afs_noop, |
| 615 | mp_afs_getattr, |
| 616 | mp_afs_setattr, |
| 617 | mp_afs_access, |
| 618 | mp_afs_lookup, |
| 619 | mp_afs_create, |
| 620 | mp_afs_remove, |
| 621 | mp_afs_link, |
| 622 | mp_afs_rename, |
| 623 | mp_afs_mkdir, |
| 624 | mp_afs_rmdir, |
| 625 | afs_readdir, |
| 626 | mp_afs_symlink, |
| 627 | mp_afs_readlink, |
| 628 | mp_afs_fsync, |
| 629 | mp_afs_inactive, |
| 630 | afs_bmap, |
| 631 | afs_hp_strategy, |
| 632 | #if !defined(AFS_NONFSTRANS) |
| 633 | /* on HPUX102 the nfs translator calls afs_bread but does |
| 634 | * not call afs_brelse. Hence we see a memory leak. If the |
| 635 | * VOP_BREAD() call fails, then nfs does VOP_RDWR() to get |
| 636 | * the same data : this is the path we follow now. */ |
| 637 | afs_noop, |
| 638 | afs_noop, |
| 639 | #else |
| 640 | mp_afs_bread, |
| 641 | mp_afs_brelse, |
| 642 | #endif |
| 643 | afs_badop, /* pathsend */ |
| 644 | afs_noop, /* setacl */ |
| 645 | afs_noop, /* getacl */ |
| 646 | afs_pathconf, |
| 647 | afs_pathconf, |
| 648 | mp_afs_lockctl, |
| 649 | afs_lockf, /* lockf */ |
| 650 | mp_afs_fid, |
| 651 | afs_noop, /*fsctl */ |
| 652 | afs_badop, |
| 653 | afs_pagein, |
| 654 | afs_pageout, |
| 655 | NULL, |
| 656 | NULL, |
| 657 | afs_prealloc, |
| 658 | afs_mapdbd, |
| 659 | afs_mmap, |
| 660 | afs_cachelimit, |
| 661 | afs_vm_checkpage, |
| 662 | afs_vm_fscontiguous, |
| 663 | afs_vm_stopio, |
| 664 | afs_read_ahead, |
| 665 | afs_release, |
| 666 | afs_unmap, |
| 667 | afs_swapfs_len, |
| 668 | mp_afs_readdir2, |
| 669 | afs_readdir3, |
| 670 | }; |
| 671 | |
| 672 | struct vnodeops *afs_ops = &Afs_vnodeops; |
| 673 | |
| 674 | /* vnode file operations, and our own */ |
| 675 | extern int vno_rw(); |
| 676 | extern int vno_ioctl(); |
| 677 | extern int vno_select(); |
| 678 | extern int afs_closex(); |
| 679 | extern int vno_close(); |
| 680 | struct fileops afs_fileops = { |
| 681 | vno_rw, |
| 682 | vno_ioctl, |
| 683 | vno_select, |
| 684 | afs_close, |
| 685 | }; |
| 686 | |
| 687 | #define vtoblksz(vp) ((vp)->v_vfsp->vfs_bsize) |
| 688 | |
| 689 | /* |
| 690 | ******************************************************************** |
| 691 | **** |
| 692 | **** afspgin_setup_io_ranges () |
| 693 | **** similar to: nfspgin_setup_io_ranges () |
| 694 | ******************************************************************** |
| 695 | */ |
| 696 | pgcnt_t |
| 697 | afspgin_setup_io_ranges(vfspage_t * vm_info, pgcnt_t bpages, k_off_t isize, |
| 698 | pgcnt_t startindex) |
| 699 | { |
| 700 | pgcnt_t file_offset = VM_FILE_OFFSET(vm_info); |
| 701 | pgcnt_t minpage; /* first page to bring in */ |
| 702 | pgcnt_t maxpage; /* one past last page to bring in */ |
| 703 | pgcnt_t maxpagein; |
| 704 | pgcnt_t multio_maxpage; |
| 705 | kern_daddr_t start_blk; |
| 706 | dbd_t *dbd; |
| 707 | expnd_flags_t up_reason, down_reason; |
| 708 | int count = 1; |
| 709 | int indx = 0; |
| 710 | int max_num_io; |
| 711 | int dbdtype; |
| 712 | preg_t *prp; |
| 713 | |
| 714 | VM_GET_IO_INFO(vm_info, maxpagein, max_num_io); |
| 715 | |
| 716 | /* |
| 717 | * We do not go past the end of the current pregion nor past the end |
| 718 | * of the current file. |
| 719 | */ |
| 720 | |
| 721 | maxpage = startindex + (bpages - (startindex + file_offset) % bpages); |
| 722 | maxpage = vm_reset_maxpage(vm_info, maxpage); |
| 723 | maxpage = MIN(maxpage, (pgcnt_t) btorp(isize) - file_offset); |
| 724 | maxpage = MIN(maxpage, startindex + maxpagein); |
| 725 | multio_maxpage = maxpage = vm_maxpage(vm_info, maxpage); |
| 726 | |
| 727 | if (!maxpage) |
| 728 | return (0); |
| 729 | |
| 730 | VASSERT(maxpage >= startindex); |
| 731 | |
| 732 | /* |
| 733 | * Expanding the fault will create calls to FINDENTRY() for new |
| 734 | * pages, which will obsolete "dbd", so copy what it points to |
| 735 | * and clear it to prevent using stale data. |
| 736 | */ |
| 737 | |
| 738 | prp = VM_PRP(vm_info); |
| 739 | dbdtype = DBD_TYPE(vm_info); |
| 740 | start_blk = DBD_DATA(vm_info); |
| 741 | vm_info->dbd = NULL; |
| 742 | vm_info->vfd = NULL; |
| 743 | VASSERT(dbdtype != DBD_NONE); |
| 744 | |
| 745 | if (max_num_io == 1) { |
| 746 | /* |
| 747 | * We need to set up one I/O: First we attempt to expand the |
| 748 | * I/O forward. Then we expand the I/O backwards. |
| 749 | */ |
| 750 | count = |
| 751 | expand_faultin_up(vm_info, dbdtype, (int)bpages, maxpage, count, |
| 752 | startindex, start_blk, &up_reason); |
| 753 | maxpage = startindex + count; |
| 754 | VASSERT(maxpage <= startindex + maxpagein); |
| 755 | minpage = startindex - (startindex + file_offset) % bpages; |
| 756 | minpage = MAX(minpage, maxpage - maxpagein); |
| 757 | VASSERT(startindex >= VM_BASE_OFFSET(vm_info)); |
| 758 | minpage = vm_minpage(vm_info, minpage); |
| 759 | VASSERT(minpage <= startindex); |
| 760 | count = |
| 761 | expand_faultin_down(vm_info, dbdtype, (int)bpages, minpage, count, |
| 762 | &startindex, &start_blk, &down_reason); |
| 763 | VM_SET_IO_STARTINDX(vm_info, 0, startindex); |
| 764 | VM_SET_IO_STARTBLK(vm_info, 0, start_blk); |
| 765 | VM_SET_IO_COUNT(vm_info, 0, count); |
| 766 | VM_SET_NUM_IO(vm_info, 1); |
| 767 | } |
| 768 | |
| 769 | if (max_num_io > 1) { |
| 770 | /* |
| 771 | * We need to set up multiple I/O information; beginning |
| 772 | * with the startindex, we will expand upwards. The expansion |
| 773 | * could stop for one of 2 reasons; we take the appropriate |
| 774 | * action in each of these cases: |
| 775 | * o VM reasons: abort setting up the multiple I/O |
| 776 | * information and return to our caller indicating |
| 777 | * that "retry" is required. |
| 778 | * o pagelimit: set up the next I/O info [we may have |
| 779 | * reached multio_maxpage at this point]. |
| 780 | * Note that expansion involves no more than a block at a time; |
| 781 | * hence it could never stop due to "discontiguous block" |
| 782 | * reason. |
| 783 | */ |
| 784 | startindex = minpage = vm_minpage(vm_info, 0); |
| 785 | for (indx = 0; (indx < max_num_io) && (startindex < multio_maxpage); |
| 786 | indx++, startindex += count) { |
| 787 | dbd = FINDDBD(prp->p_reg, startindex); |
| 788 | start_blk = dbd->dbd_data; |
| 789 | maxpage = |
| 790 | startindex + (bpages - (startindex + file_offset) % bpages); |
| 791 | maxpage = min(maxpage, multio_maxpage); |
| 792 | count = |
| 793 | expand_faultin_up(vm_info, dbdtype, bpages, maxpage, |
| 794 | 1 /* count */ , |
| 795 | startindex, start_blk, &up_reason); |
| 796 | VM_SET_IO_STARTINDX(vm_info, indx, startindex); |
| 797 | VM_SET_IO_STARTBLK(vm_info, indx, start_blk); |
| 798 | VM_SET_IO_COUNT(vm_info, indx, count); |
| 799 | if (up_reason & VM_REASONS) |
| 800 | break; |
| 801 | VASSERT(!(up_reason & NONCONTIGUOUS_BLOCK)); |
| 802 | VASSERT(up_reason & PAGELIMIT); |
| 803 | } |
| 804 | if (startindex < multio_maxpage) { |
| 805 | VM_MULT_IO_FAILURE(vm_info); |
| 806 | VM_REINIT_FAULT_DBDVFD(vm_info); |
| 807 | return (0); /* retry */ |
| 808 | } |
| 809 | count = maxpagein; |
| 810 | VM_SET_NUM_IO(vm_info, indx); |
| 811 | } |
| 812 | |
| 813 | /* |
| 814 | * Tell VM where the I/O intends to start. This may be different |
| 815 | * from the faulting point. |
| 816 | */ |
| 817 | |
| 818 | VM_SET_STARTINDX(vm_info, VM_GET_IO_STARTINDX(vm_info, 0)); |
| 819 | |
| 820 | return (count); |
| 821 | |
| 822 | } |
| 823 | |
| 824 | /* |
| 825 | ******************************************************************** |
| 826 | **** |
| 827 | **** afspgin_blkflsh () |
| 828 | **** similar to: nfspgin_blkflsh () |
| 829 | ******************************************************************** |
| 830 | */ |
| 831 | retval_t |
| 832 | afspgin_blkflsh(vfspage_t * vm_info, struct vnode * devvp, pgcnt_t * num_4k) |
| 833 | { |
| 834 | int flush_reslt = 0; |
| 835 | pgcnt_t count = *num_4k; |
| 836 | pgcnt_t page_count; |
| 837 | int indx = 0; |
| 838 | int num_io = VM_GET_NUM_IO(vm_info); |
| 839 | |
| 840 | /* |
| 841 | * On this blkflush() we don't want to purge the buffer cache and we do |
| 842 | * want to wait, so the flags are '0'. |
| 843 | */ |
| 844 | |
| 845 | for (indx = 0; indx < num_io; indx++) { |
| 846 | flush_reslt = |
| 847 | blkflush(devvp, (kern_daddr_t) VM_GET_IO_STARTBLK(vm_info, indx), |
| 848 | ptob(VM_GET_IO_COUNT(vm_info, indx)), 0, |
| 849 | VM_REGION(vm_info)); |
| 850 | if (flush_reslt) { |
| 851 | vm_lock(vm_info); |
| 852 | if (vm_page_now_valid(vm_info, &page_count)) { |
| 853 | vm_release_memory(vm_info); |
| 854 | vm_release_structs(vm_info); |
| 855 | *num_4k = page_count; |
| 856 | return (VM_PAGE_PRESENT); |
| 857 | } |
| 858 | return (VM_RETRY); |
| 859 | } |
| 860 | } |
| 861 | return (VM_DONE); |
| 862 | } |
| 863 | |
| 864 | /* |
| 865 | ******************************************************************** |
| 866 | **** |
| 867 | **** afspgin_io () |
| 868 | **** similar to: nfspgin_io () |
| 869 | ******************************************************************** |
| 870 | */ |
| 871 | int |
| 872 | afspgin_io(vfspage_t * vm_info, struct vnode *devvp, pgcnt_t bpages, |
| 873 | pgcnt_t maxpagein, pgcnt_t count) |
| 874 | { |
| 875 | int i; |
| 876 | int error = 0; |
| 877 | caddr_t vaddr = VM_ADDR(vm_info); |
| 878 | caddr_t virt_addr = VM_MAPPED_ADDR(vm_info); |
| 879 | pagein_info_t *io = VM_PAGEIN_INFO(vm_info); |
| 880 | preg_t *prp = VM_PRP(vm_info); |
| 881 | int wrt = VM_WRT(vm_info); |
| 882 | space_t space = VM_SPACE(vm_info); |
| 883 | int num_io = VM_GET_NUM_IO(vm_info); |
| 884 | |
| 885 | #ifdef notdef /* Not used in AFS */ |
| 886 | /* |
| 887 | * With VM_READ_AHEAD_ALLOWED() macro, check if read-ahead should |
| 888 | * be used in this case. |
| 889 | * |
| 890 | * Unlike UFS, NFS does not start the faulting page I/O |
| 891 | * asynchronously. Why? Asynchronous requests are handled by the |
| 892 | * biod's. It doesn't make sense to queue up the faulting request |
| 893 | * behind other asynchrnous requests. This is not true for UFS |
| 894 | * where the asynchrnous request is immediately handled. |
| 895 | */ |
| 896 | |
| 897 | if ((VM_READ_AHEAD_ALLOWED(vm_info)) && (nfs_read_ahead_on) |
| 898 | && (NFS_DO_READ_AHEAD) && (should_do_read_ahead(prp, vaddr))) { |
| 899 | |
| 900 | pgcnt_t max_rhead_io; |
| 901 | caddr_t rhead_vaddr; |
| 902 | pgcnt_t total_rheads_allowed; |
| 903 | |
| 904 | /* |
| 905 | * Determine the maximum amount of read-ahead I/O. |
| 906 | */ |
| 907 | total_rheads_allowed = maxpagein - count; |
| 908 | |
| 909 | /* |
| 910 | * If the count is less than a block, raise it to one. |
| 911 | */ |
| 912 | if (total_rheads_allowed < bpages) |
| 913 | total_rheads_allowed = bpages; |
| 914 | |
| 915 | max_rhead_io = total_rheads_allowed; |
| 916 | rhead_vaddr = VM_MAPPED_ADDR(vm_info) + (count * NBPG); |
| 917 | error = |
| 918 | nfs_read_ahead(vm_info->vp, prp, wrt, space, rhead_vaddr, |
| 919 | &max_rhead_io); |
| 920 | |
| 921 | /* |
| 922 | * Set the next fault location. If read_ahead launches any |
| 923 | * I/O it will adjust it accordingly. |
| 924 | */ |
| 925 | vm_info->prp->p_nextfault = vm_info->startindex + count; |
| 926 | |
| 927 | /* |
| 928 | * Now perform the faulting I/O synchronously. |
| 929 | */ |
| 930 | vm_unlock(vm_info); |
| 931 | |
| 932 | error = |
| 933 | syncpageio((swblk_t) VM_GET_IO_STARTBLK(vm_info, 0), |
| 934 | VM_MAPPED_SPACE(vm_info), VM_MAPPED_ADDR(vm_info), |
| 935 | (int)ptob(count), B_READ, devvp, |
| 936 | B_vfs_pagein | B_pagebf, VM_REGION(vm_info)); |
| 937 | } else |
| 938 | #endif |
| 939 | { |
| 940 | virt_addr = VM_MAPPED_ADDR(vm_info); |
| 941 | vm_unlock(vm_info); |
| 942 | for (i = 0; i < num_io; i++) { |
| 943 | /* |
| 944 | * REVISIT -- investigate doing asyncpageio(). |
| 945 | */ |
| 946 | error |= (io[i].error = |
| 947 | syncpageio((swblk_t) VM_GET_IO_STARTBLK(vm_info, i), |
| 948 | VM_MAPPED_SPACE(vm_info), virt_addr, |
| 949 | (int)ptob(VM_GET_IO_COUNT(vm_info, i)), |
| 950 | B_READ, devvp, B_vfs_pagein | B_pagebf, |
| 951 | VM_REGION(vm_info))); |
| 952 | virt_addr += ptob(VM_GET_IO_COUNT(vm_info, i)); |
| 953 | } |
| 954 | /* |
| 955 | * Set the next fault location. If read_ahead launches any |
| 956 | * I/O it will adjust it accordingly. |
| 957 | */ |
| 958 | vm_info->prp->p_nextfault = vm_info->startindex + count; |
| 959 | } |
| 960 | |
| 961 | return (error); |
| 962 | } |
| 963 | |
| 964 | /* |
| 965 | ******************************************************************** |
| 966 | **** |
| 967 | **** afspgin_update_dbd () |
| 968 | **** similar to: nfspgin_update_dbd () |
| 969 | ******************************************************************** |
| 970 | */ |
| 971 | void |
| 972 | afspgin_update_dbd(vfspage_t * vm_info, int bsize) |
| 973 | { |
| 974 | k_off_t off; |
| 975 | pgcnt_t count = bsize / NBPG; |
| 976 | k_off_t rem; |
| 977 | pgcnt_t m; |
| 978 | pgcnt_t pgindx; |
| 979 | kern_daddr_t blkno; |
| 980 | int num_io = VM_GET_NUM_IO(vm_info); |
| 981 | int i; |
| 982 | |
| 983 | for (i = 0; i < num_io; i++) { |
| 984 | |
| 985 | pgindx = VM_GET_IO_STARTINDX(vm_info, i); |
| 986 | off = vnodindx(VM_REGION(vm_info), pgindx); |
| 987 | rem = off % bsize; |
| 988 | blkno = VM_GET_IO_STARTBLK(vm_info, i); |
| 989 | |
| 990 | VASSERT(bsize % NBPG == 0); |
| 991 | VASSERT(rem % NBPG == 0); |
| 992 | |
| 993 | pgindx -= (pgcnt_t) btop(rem); |
| 994 | blkno -= (kern_daddr_t) btodb(rem); |
| 995 | |
| 996 | /* |
| 997 | * This region could start in mid-block. If so, pgindx |
| 998 | * could be less than 0, so we adjust pgindx and blkno back |
| 999 | * up so that pgindx is 0. |
| 1000 | */ |
| 1001 | |
| 1002 | if (pgindx < 0) { |
| 1003 | pgcnt_t prem; |
| 1004 | prem = 0 - pgindx; |
| 1005 | pgindx = 0; |
| 1006 | count -= prem; |
| 1007 | blkno += btodb(ptob(prem)); |
| 1008 | } |
| 1009 | |
| 1010 | for (m = 0; m < count && pgindx < VM_REGION_SIZE(vm_info); |
| 1011 | m++, pgindx++, blkno += btodb(NBPG)) { |
| 1012 | /* |
| 1013 | * Note: since this only changes one block, it |
| 1014 | * assumes only one block was faulted in. Currently |
| 1015 | * this is always true for remote files, and we only |
| 1016 | * get here for remote files, so everything is ok. |
| 1017 | */ |
| 1018 | vm_mark_dbd(vm_info, pgindx, blkno); |
| 1019 | } |
| 1020 | } |
| 1021 | } |
| 1022 | |
| 1023 | int |
| 1024 | afs_pagein(vp, prp, wrt, space, vaddr, ret_startindex) |
| 1025 | struct vnode *vp; |
| 1026 | preg_t *prp; |
| 1027 | int wrt; |
| 1028 | space_t space; |
| 1029 | caddr_t vaddr; |
| 1030 | pgcnt_t *ret_startindex; |
| 1031 | { |
| 1032 | pgcnt_t startindex; |
| 1033 | pgcnt_t pgindx = *ret_startindex; |
| 1034 | pgcnt_t maxpagein; |
| 1035 | struct vnode *devvp; |
| 1036 | pgcnt_t count; |
| 1037 | kern_daddr_t start_blk = 0; |
| 1038 | int bsize; |
| 1039 | int error; |
| 1040 | k_off_t isize; |
| 1041 | int shared; /* writable memory mapped file */ |
| 1042 | retval_t retval = 0; |
| 1043 | pgcnt_t ok_dbd_limit = 0; /* last dbd that we can trust */ |
| 1044 | pgcnt_t bpages; /* number of pages per block */ |
| 1045 | pgcnt_t page_count; |
| 1046 | vfspage_t *vm_info = NULL; |
| 1047 | int done; |
| 1048 | |
| 1049 | struct vattr va; |
| 1050 | |
| 1051 | caddr_t nvaddr; |
| 1052 | space_t nspace; |
| 1053 | int change_to_fstore = 0; /* need to change dbds to DBD_FSTORE */ |
| 1054 | int flush_start_blk = 0; |
| 1055 | int flush_end_blk = 0; |
| 1056 | |
| 1057 | int i, j; |
| 1058 | |
| 1059 | AFS_STATCNT(afs_pagein); |
| 1060 | vmemp_lockx(); /* lock down VM empire */ |
| 1061 | |
| 1062 | /* Initialize the VM info structure */ |
| 1063 | done = |
| 1064 | vm_pagein_init(&vm_info, prp, pgindx, space, vaddr, wrt, 0, |
| 1065 | LGPG_ENABLE); |
| 1066 | |
| 1067 | /* Check to see if we slept and the page was falted in. */ |
| 1068 | if (done) { |
| 1069 | vm_release_structs(vm_info); |
| 1070 | vmemp_returnx(1); |
| 1071 | } |
| 1072 | |
| 1073 | vp = VM_GET_PAGEIN_VNODE(vm_info); |
| 1074 | VASSERT(vp != NULL); |
| 1075 | shared = VM_SHARED_OBJECT(vm_info); |
| 1076 | VASSERT(DBD_TYPE(vm_info) != DBD_NONE); |
| 1077 | |
| 1078 | /* |
| 1079 | * Get the devvp and block size for this vnode type |
| 1080 | */ |
| 1081 | devvp = vp; |
| 1082 | bsize = vp->v_vfsp->vfs_bsize; |
| 1083 | if (bsize <= 0 || (bsize & (DEV_BSIZE - 1))) |
| 1084 | osi_Panic("afs_pagein: bsize is zero or not a multiple of DEV_BSIZE"); |
| 1085 | |
| 1086 | bpages = (pgcnt_t) btop(bsize); |
| 1087 | VASSERT(bpages > 0); |
| 1088 | VM_SET_FS_MAX_PAGES(vm_info, bpages); |
| 1089 | |
| 1090 | /* this trace cannot be here because the afs_global lock might not be |
| 1091 | * held at this point. We hold the vm global lock throughout |
| 1092 | * this procedure ( and not the AFS global lock ) |
| 1093 | * afs_Trace4(afs_iclSetp, CM_TRACE_HPPAGEIN, ICL_TYPE_POINTER, (afs_int32) vp, |
| 1094 | * ICL_TYPE_LONG, DBD_TYPE(vm_info), ICL_TYPE_LONG, bpages, |
| 1095 | * ICL_TYPE_LONG, shared); |
| 1096 | */ |
| 1097 | /* Come here if we have to release the region lock before |
| 1098 | * locking pages. This can happen in memreserve() and |
| 1099 | * blkflush(). |
| 1100 | */ |
| 1101 | retry: |
| 1102 | /* |
| 1103 | * For remote files like ours, we want to check to see if the file has shrunk. |
| 1104 | * If so, we should invalidate any pages past the end. In the name |
| 1105 | * of efficiency, we only do this if the page we want to fault is |
| 1106 | * past the end of the file. |
| 1107 | */ |
| 1108 | { |
| 1109 | if (VOP_GETATTR(vp, &va, kt_cred(u.u_kthreadp), VIFSYNC) != 0) { |
| 1110 | VM_ZOMBIE_OBJECT(vm_info); |
| 1111 | vm_release_memory(vm_info); |
| 1112 | vm_release_structs(vm_info); |
| 1113 | vmemp_returnx(0); |
| 1114 | } |
| 1115 | isize = va.va_size; |
| 1116 | if (vnodindx(VM_REGION(vm_info), pgindx) >= isize) { |
| 1117 | /* |
| 1118 | * The file has shrunk and someone is trying to access a |
| 1119 | * page past the end of the object. Shrink the object back |
| 1120 | * to its currrent size, send a SIGBUS to the faulting |
| 1121 | * process and return. |
| 1122 | * |
| 1123 | * We must release the region lock before calling mtrunc(), |
| 1124 | * since mtrunc() locks all the regions that are using this |
| 1125 | * file. |
| 1126 | */ |
| 1127 | vm_release_memory(vm_info); |
| 1128 | vm_truncate_region(vm_info, isize); |
| 1129 | vm_release_structs(vm_info); |
| 1130 | vmemp_returnx(-SIGBUS); |
| 1131 | } |
| 1132 | } |
| 1133 | |
| 1134 | maxpagein = vm_pick_maxpagein(vm_info); |
| 1135 | if (vm_wait_for_memory(vm_info, maxpagein, 1)) { |
| 1136 | /* Check to see if we should continue faulting. */ |
| 1137 | if (vm_page_now_valid(vm_info, &page_count)) { |
| 1138 | vm_release_memory(vm_info); |
| 1139 | vm_release_structs(vm_info); |
| 1140 | vmemp_returnx(page_count); |
| 1141 | } |
| 1142 | } |
| 1143 | if (count = vm_no_io_required(vm_info)) { |
| 1144 | /* Release any excess memory. */ |
| 1145 | vm_release_memory(vm_info); |
| 1146 | vm_release_structs(vm_info); |
| 1147 | vmemp_returnx(count); |
| 1148 | } |
| 1149 | #ifdef OSDEBUG |
| 1150 | /* |
| 1151 | * We should never have DBD_HOLE pages in a non-MMF region. |
| 1152 | */ |
| 1153 | if (!shared) |
| 1154 | VASSERT(dbd->dbd_type != DBD_HOLE); |
| 1155 | #endif |
| 1156 | VASSERT(DBD_TYPE(vm_info) != DBD_NONE); |
| 1157 | |
| 1158 | startindex = *ret_startindex; |
| 1159 | |
| 1160 | /* |
| 1161 | * If the page we want is in memory already, take it |
| 1162 | */ |
| 1163 | if (VM_MEMORY_RESERVED(vm_info) < maxpagein) { |
| 1164 | /* pick up the rest of memory now. */ |
| 1165 | if (vm_wait_for_memory(vm_info, maxpagein, 0)) { |
| 1166 | if (vm_page_now_valid(vm_info, &page_count)) { |
| 1167 | vm_release_memory(vm_info); |
| 1168 | vm_release_structs(vm_info); |
| 1169 | vmemp_returnx(page_count); |
| 1170 | } |
| 1171 | goto retry; |
| 1172 | } |
| 1173 | } |
| 1174 | |
| 1175 | if (! |
| 1176 | (count = |
| 1177 | afspgin_setup_io_ranges(vm_info, bpages, isize, startindex))) { |
| 1178 | goto retry; |
| 1179 | } |
| 1180 | |
| 1181 | startindex = VM_GET_STARTINDX(vm_info); |
| 1182 | |
| 1183 | VASSERT(maxpagein >= count); |
| 1184 | |
| 1185 | /* |
| 1186 | * Release the memory we won't need. |
| 1187 | */ |
| 1188 | if (count < maxpagein) { |
| 1189 | vm_release_excess_memory(vm_info, |
| 1190 | (VM_MEMORY_RESERVED(vm_info) - count)); |
| 1191 | } |
| 1192 | |
| 1193 | retval = afspgin_blkflsh(vm_info, devvp, &count); |
| 1194 | |
| 1195 | if (retval == VM_RETRY) { |
| 1196 | goto retry; |
| 1197 | } |
| 1198 | |
| 1199 | if (retval == VM_PAGE_PRESENT) |
| 1200 | return (count); |
| 1201 | |
| 1202 | #if 0 |
| 1203 | /* |
| 1204 | * The definition of krusage_cntr_t is in h/kmetric.h, which |
| 1205 | * is not shipped. Since it's just statistics, we punt and do |
| 1206 | * not update it. If it's a problem we'll need to get HP to export |
| 1207 | * an interface that we can use to increment the counter. |
| 1208 | */ |
| 1209 | |
| 1210 | /* It's a real fault, not a reclaim */ |
| 1211 | { |
| 1212 | krusage_cntr_t *temp; |
| 1213 | temp = kt_cntrp(u.u_kthreadp); |
| 1214 | temp->krc_majflt++; |
| 1215 | } |
| 1216 | #endif |
| 1217 | |
| 1218 | /* |
| 1219 | * Tell VM where the I/O intends to start. This may be different |
| 1220 | * from the faulting point. |
| 1221 | */ |
| 1222 | |
| 1223 | /* |
| 1224 | * vm_prepare_io will fill the region with pages and release the |
| 1225 | * region lock. |
| 1226 | */ |
| 1227 | vm_prepare_io(vm_info, &count); |
| 1228 | |
| 1229 | /* |
| 1230 | * Count may have been adjusted, check to make sure it's non-zero. |
| 1231 | */ |
| 1232 | if (count == 0) { |
| 1233 | if (vm_retry(vm_info)) { |
| 1234 | goto retry; |
| 1235 | } |
| 1236 | |
| 1237 | /* |
| 1238 | * Release resources and retry the fault. Release any excess |
| 1239 | * memory. |
| 1240 | */ |
| 1241 | |
| 1242 | vm_release_memory(vm_info); |
| 1243 | vm_release_structs(vm_info); |
| 1244 | vmemp_returnx(0); |
| 1245 | } |
| 1246 | |
| 1247 | error = afspgin_io(vm_info, devvp, bpages, maxpagein, count); |
| 1248 | |
| 1249 | if ((VM_IS_ZOMBIE(vm_info)) || (error)) { |
| 1250 | retval = -SIGBUS; |
| 1251 | VM_ZOMBIE_OBJECT(vm_info); |
| 1252 | goto backout; |
| 1253 | } |
| 1254 | /* |
| 1255 | * For a writable memory mapped file that is remote we must |
| 1256 | * detect potential holes in the file and force allocation of |
| 1257 | * disk space on the remote system. Unfortunately, there is |
| 1258 | * no easy way to do this, so this gets a little ugly. |
| 1259 | */ |
| 1260 | if (shared && wrt) { |
| 1261 | /* |
| 1262 | * See if The user wants to write to this page. Write some |
| 1263 | * minimal amount of data back to the remote file to |
| 1264 | * force allocation of file space. We only need to |
| 1265 | * write a small amount, since holes are always at |
| 1266 | * least one filesystem block in size. |
| 1267 | */ |
| 1268 | error = vm_alloc_hole(vm_info); |
| 1269 | |
| 1270 | /* |
| 1271 | * If some sort of I/O error occurred we generate a |
| 1272 | * SIGBUS for the process that caused the write, |
| 1273 | * undo our page locks, etc and return. |
| 1274 | */ |
| 1275 | if ((VM_IS_ZOMBIE(vm_info)) || (error)) { |
| 1276 | VM_ZOMBIE_OBJECT(vm_info); |
| 1277 | retval = -SIGBUS; |
| 1278 | goto backout; |
| 1279 | } |
| 1280 | |
| 1281 | /* |
| 1282 | * Change these dbds to DBD_FSTORE. We cannot do it here, |
| 1283 | * since the region must be locked, and it is not locked |
| 1284 | * at the moment. We cannot lock the region yet, as we |
| 1285 | * first have to release the page locks. |
| 1286 | */ |
| 1287 | change_to_fstore = 1; |
| 1288 | } |
| 1289 | |
| 1290 | vm_finish_io(vm_info, count); |
| 1291 | |
| 1292 | /* |
| 1293 | * Acquire the lock before we play around with changing the vfd's. |
| 1294 | */ |
| 1295 | vm_lock(vm_info); |
| 1296 | |
| 1297 | if (change_to_fstore) |
| 1298 | afspgin_update_dbd(vm_info, bsize); |
| 1299 | |
| 1300 | #if defined(AFS_HPUX110_ENV) |
| 1301 | getppdp()->cnt.v_exfod += count; |
| 1302 | #else |
| 1303 | mpproc_info[getprocindex()].cnt.v_exfod += count; |
| 1304 | #endif |
| 1305 | vmemp_unlockx(); /* free up VM empire */ |
| 1306 | *ret_startindex = startindex; |
| 1307 | |
| 1308 | /* |
| 1309 | * In case we have any excess memory... |
| 1310 | */ |
| 1311 | if (VM_MEMORY_RESERVED(vm_info)) |
| 1312 | vm_release_memory(vm_info); |
| 1313 | vm_release_structs(vm_info); |
| 1314 | |
| 1315 | return count; |
| 1316 | |
| 1317 | backout: |
| 1318 | |
| 1319 | vm_finish_io_failed(vm_info, count); |
| 1320 | |
| 1321 | vm_lock(vm_info); |
| 1322 | |
| 1323 | vm_undo_validation(vm_info, count); |
| 1324 | |
| 1325 | /* |
| 1326 | * In case we have any excess memory... |
| 1327 | */ |
| 1328 | if (VM_MEMORY_RESERVED(vm_info)) |
| 1329 | vm_release_memory(vm_info); |
| 1330 | vm_release_structs(vm_info); |
| 1331 | |
| 1332 | vmemp_unlockx(); /* free up VM empire */ |
| 1333 | return retval; |
| 1334 | } |
| 1335 | |
| 1336 | int |
| 1337 | afs_pageout(vp, prp, start, end, flags) |
| 1338 | struct vnode *vp; /* not used */ |
| 1339 | preg_t *prp; |
| 1340 | pgcnt_t start; |
| 1341 | pgcnt_t end; |
| 1342 | int flags; |
| 1343 | { |
| 1344 | struct vnode *filevp; |
| 1345 | struct vnode *devvp; |
| 1346 | pgcnt_t i; |
| 1347 | int steal; |
| 1348 | int vhand; |
| 1349 | int hard; |
| 1350 | int *piocnt; /* wakeup counter used if PAGEOUT_WAIT */ |
| 1351 | struct ucred *old_cred; |
| 1352 | vfspage_t vm_info; |
| 1353 | fsdata_t args; |
| 1354 | |
| 1355 | int inode_changed = 0; |
| 1356 | int file_is_remote; |
| 1357 | struct inode *ip; |
| 1358 | |
| 1359 | AFS_STATCNT(afs_pageout); |
| 1360 | |
| 1361 | steal = (flags & PAGEOUT_FREE); |
| 1362 | vhand = (flags & PAGEOUT_VHAND); |
| 1363 | hard = (flags & PAGEOUT_HARD); |
| 1364 | |
| 1365 | vmemp_lockx(); |
| 1366 | |
| 1367 | /* Initialize the VM info structure. */ |
| 1368 | vm_pageout_init(&vm_info, prp, start, end, 0, 0, 0, flags); |
| 1369 | |
| 1370 | /* |
| 1371 | * If the region is marked "don't swap", then don't steal any pages |
| 1372 | * from it. We can, however, write dirty pages out to disk (only if |
| 1373 | * PAGEOUT_FREE is not set). |
| 1374 | */ |
| 1375 | if (vm_no_pageout(&vm_info)) { |
| 1376 | vmemp_unlockx(); |
| 1377 | return (0); |
| 1378 | } |
| 1379 | |
| 1380 | /* |
| 1381 | * If caller wants to wait until the I/O is complete. |
| 1382 | */ |
| 1383 | vm_setup_wait_for_io(&vm_info); |
| 1384 | |
| 1385 | filevp = VM_GET_PAGEOUT_VNODE(&vm_info); /* always page out to back store */ |
| 1386 | VASSERT(filevp != NULL); |
| 1387 | |
| 1388 | memset((caddr_t) & args, 0, sizeof(fsdata_t)); |
| 1389 | args.remote_down = 0; /* assume remote file servers are up */ |
| 1390 | args.remote = 1; /* we are remote */ |
| 1391 | args.bsize = 0; /* filled up later by afs_vm_checkpage() */ |
| 1392 | |
| 1393 | if (filevp->v_fstype == VUFS) { |
| 1394 | ip = VTOI(filevp); |
| 1395 | devvp = ip->i_devvp; |
| 1396 | file_is_remote = 0; |
| 1397 | } else { |
| 1398 | file_is_remote = 1; |
| 1399 | devvp = filevp; |
| 1400 | |
| 1401 | /* |
| 1402 | * If we are vhand(), and this is an NFS file, we need to |
| 1403 | * see if the NFS server is "down". If so, we decide |
| 1404 | * if we will try to talk to it again, or defer pageouts |
| 1405 | * of dirty NFS pages until a future time. |
| 1406 | */ |
| 1407 | #ifdef notdef |
| 1408 | if (vhand && filevp->v_fstype == VNFS && vtomi(filevp)->mi_down |
| 1409 | && vtomi(filevp)->mi_hard) { |
| 1410 | extern afs_int32 vhand_nfs_retry; |
| 1411 | /* |
| 1412 | * If there is still time left on our timer, we will |
| 1413 | * not talk to this server right now. |
| 1414 | */ |
| 1415 | if (vhand_nfs_retry > 0) |
| 1416 | args.remote_down = 1; |
| 1417 | } |
| 1418 | #endif |
| 1419 | } |
| 1420 | |
| 1421 | /* |
| 1422 | * Initialize args. We set bsize to 0 to tell vfs_vfdcheck() that |
| 1423 | * it must get the file size and other attributes if it comes across |
| 1424 | * a dirty page. |
| 1425 | */ |
| 1426 | vm_info.fs_data = (caddr_t) & args; |
| 1427 | |
| 1428 | /* this trace cannot be here because the afs_global lock might not be |
| 1429 | * held at this point. We hold the vm global lock throughout |
| 1430 | * this procedure ( and not the AFS global lock ) |
| 1431 | * afs_Trace4(afs_iclSetp, CM_TRACE_HPPAGEOUT, ICL_TYPE_POINTER, (afs_int32) filevp, |
| 1432 | * ICL_TYPE_LONG, start, ICL_TYPE_LONG, end, ICL_TYPE_LONG, flags); |
| 1433 | */ |
| 1434 | |
| 1435 | i = start; |
| 1436 | |
| 1437 | while (i <= end) { |
| 1438 | struct buf *bp; |
| 1439 | k_off_t start; |
| 1440 | pgcnt_t npages; |
| 1441 | k_off_t nbytes; |
| 1442 | int error; |
| 1443 | |
| 1444 | extern int pageiodone(); |
| 1445 | space_t nspace; |
| 1446 | caddr_t nvaddr; |
| 1447 | |
| 1448 | /* |
| 1449 | * Ask the VM system to find the next run of pages. |
| 1450 | */ |
| 1451 | vm_find_next_range(&vm_info, i, end); |
| 1452 | |
| 1453 | /* |
| 1454 | * It's possible that the remote file shrunk in size. Check the flags |
| 1455 | * to see if the request was beyond the end of the file. If it was, |
| 1456 | * truncate the region to the file size and continue. We could be on a |
| 1457 | * run so after trunction continue, there may be some I/O to write |
| 1458 | * out. |
| 1459 | */ |
| 1460 | if (VM_FS_FLAGS(&vm_info) & PAGEOUT_TRUNCATE) { |
| 1461 | pgcnt_t pglen = (pgcnt_t) btorp(args.isize); |
| 1462 | |
| 1463 | /* |
| 1464 | * This page is past the end of the file. Unlock this page |
| 1465 | * (region_trunc will throw it away) and then call |
| 1466 | * region_trunc() to invalidate all pages past the new end of |
| 1467 | * the file. |
| 1468 | */ |
| 1469 | region_trunc(VM_REGION(&vm_info), pglen, pglen + 1); |
| 1470 | |
| 1471 | /* |
| 1472 | * remove the truncation flag. |
| 1473 | */ |
| 1474 | VM_UNSETFS_FLAGS(&vm_info, PAGEOUT_TRUNCATE); |
| 1475 | } |
| 1476 | |
| 1477 | if (VM_NO_PAGEOUT_RUN(&vm_info)) |
| 1478 | break; |
| 1479 | |
| 1480 | /* |
| 1481 | * We have a run of dirty pages [args.start...args.end]. |
| 1482 | */ |
| 1483 | VASSERT(filevp->v_fstype != VCDFS); |
| 1484 | VASSERT((filevp->v_vfsp->vfs_flag & VFS_RDONLY) == 0); |
| 1485 | VASSERT(VM_GET_NUM_IO(&vm_info) == 1); |
| 1486 | |
| 1487 | /* |
| 1488 | * We will be doing an I/O on the region, let the VM system know. |
| 1489 | */ |
| 1490 | (void)vm_up_physio_count(&vm_info); |
| 1491 | |
| 1492 | /* |
| 1493 | * Okay, get set to perform the I/O. |
| 1494 | */ |
| 1495 | inode_changed = 1; |
| 1496 | npages = |
| 1497 | (VM_END_PAGEOUT_INDX(&vm_info) + 1) - |
| 1498 | VM_START_PAGEOUT_INDX(&vm_info); |
| 1499 | |
| 1500 | /* |
| 1501 | * Allocate and initialize an I/O buffer. |
| 1502 | */ |
| 1503 | bp = bswalloc(); |
| 1504 | vm_init_bp(&vm_info, bp); /* Let the VM system initialize */ |
| 1505 | |
| 1506 | /* Identify this buffer for KI */ |
| 1507 | bp->b_bptype = B_vfs_pageout | B_pagebf; |
| 1508 | |
| 1509 | if (steal) |
| 1510 | bp->b_flags = B_CALL | B_BUSY | B_PAGEOUT; /* steal pages */ |
| 1511 | else |
| 1512 | bp->b_flags = B_CALL | B_BUSY; /* keep pages */ |
| 1513 | |
| 1514 | /* |
| 1515 | * If we are vhand paging over NFS, we will wait for the I/O |
| 1516 | * to complete. |
| 1517 | */ |
| 1518 | if (vhand && filevp->v_fstype == VNFS) { |
| 1519 | bp->b_flags &= ~B_CALL; |
| 1520 | } else { |
| 1521 | bp->b_iodone = (int (*)())pageiodone; |
| 1522 | } |
| 1523 | |
| 1524 | /* |
| 1525 | * Make sure we do not write past the end of the file. |
| 1526 | */ |
| 1527 | nbytes = ptob(npages); |
| 1528 | start = vnodindx(VM_REGION(&vm_info), vm_info.start); |
| 1529 | if (start + nbytes > args.isize) { |
| 1530 | #ifdef OSDEBUG |
| 1531 | /* |
| 1532 | * The amount we are off better not be bigger than a |
| 1533 | * filesystem block. |
| 1534 | */ |
| 1535 | if (start + nbytes - args.isize >= args.bsize) { |
| 1536 | osi_Panic("afs_pageout: remainder too large"); |
| 1537 | } |
| 1538 | #endif |
| 1539 | /* |
| 1540 | * Reset the size of the I/O as necessary. For remote |
| 1541 | * files, we set the size to the exact number of bytes to |
| 1542 | * the end of the file. For local files, we round this up |
| 1543 | * to the nearest DEV_BSIZE chunk since disk I/O must always |
| 1544 | * be in multiples of DEV_BSIZE. In this case, we do not |
| 1545 | * bother to zero out the data past the "real" end of the |
| 1546 | * file, this is done when the data is read (either through |
| 1547 | * mmap() or by normal file system access). |
| 1548 | */ |
| 1549 | if (file_is_remote) |
| 1550 | nbytes = args.isize - start; |
| 1551 | else |
| 1552 | nbytes = roundup(args.isize - start, DEV_BSIZE); |
| 1553 | } |
| 1554 | |
| 1555 | /* |
| 1556 | * Now get ready to perform the I/O |
| 1557 | */ |
| 1558 | if (!vm_protect_pageout(&vm_info, npages)) { |
| 1559 | VASSERT(vhand); |
| 1560 | vm_undo_invalidation(&vm_info, vm_info.start, vm_info.end); |
| 1561 | vm_finish_io_failed(&vm_info, npages); |
| 1562 | bswfree(bp); |
| 1563 | break; |
| 1564 | } |
| 1565 | /* |
| 1566 | * If this is an NFS write by vhand(), we will not be calling |
| 1567 | * pageiodone(). asyncpageio() increments parolemem for us |
| 1568 | * if bp->b_iodone is pageiodone, so we must do it manually |
| 1569 | * if pageiodone() will not be called automatically. |
| 1570 | */ |
| 1571 | if (!(bp->b_flags & B_CALL) && steal) { |
| 1572 | ulong_t context; |
| 1573 | |
| 1574 | SPINLOCK_USAV(pfdat_lock, context); |
| 1575 | parolemem += btorp(nbytes); |
| 1576 | SPINUNLOCK_USAV(pfdat_lock, context); |
| 1577 | } |
| 1578 | blkflush(devvp, VM_START_PAGEOUT_BLK(&vm_info), (long)nbytes, |
| 1579 | (BX_NOBUFWAIT | BX_PURGE), VM_REGION(&vm_info)); |
| 1580 | |
| 1581 | /* |
| 1582 | * If vhand is the one paging things out, and this is an NFS |
| 1583 | * file, we need to temporarily become a different user so |
| 1584 | * that we are not trying to page over NFS as root. We use |
| 1585 | * the user credentials associated with the writable file |
| 1586 | * pointer that is in the psuedo-vas for this MMF. |
| 1587 | * |
| 1588 | * NOTE: we are currently using "va_rss" to store the ucred |
| 1589 | * value in the vas (this should be fixed in 10.0). |
| 1590 | */ |
| 1591 | old_cred = kt_cred(u.u_kthreadp); |
| 1592 | if (vhand) { |
| 1593 | #if defined(AFS_HPUX1123_ENV) |
| 1594 | /* |
| 1595 | * DEE - 1123 does not have the vas.h, and it looks |
| 1596 | * we should never be called with a NFS type file anyway. |
| 1597 | * so where did this come from? Was it copied from NFS? |
| 1598 | * I assume it was, so we will add an assert for now |
| 1599 | * and see if the code runs at all. |
| 1600 | */ |
| 1601 | VASSERT(filevp->v_fstype != VNFS); |
| 1602 | #else |
| 1603 | set_kt_cred(u.u_kthreadp, filevp->v_vas->va_cred); |
| 1604 | |
| 1605 | /* |
| 1606 | * If root was the one who opened the mmf for write, |
| 1607 | * va_cred will be NULL. So reset kt_cred(u.u_kthreadp) to what it |
| 1608 | * was. We will page out as root, but that is the |
| 1609 | * correct thing to do in this case anyway. |
| 1610 | */ |
| 1611 | if (kt_cred(u.u_kthreadp) == NULL) |
| 1612 | set_kt_cred(u.u_kthreadp, old_cred); |
| 1613 | #endif |
| 1614 | } |
| 1615 | |
| 1616 | /* |
| 1617 | * Really do the I/O. |
| 1618 | */ |
| 1619 | error = |
| 1620 | asyncpageio(bp, VM_START_PAGEOUT_BLK(&vm_info), |
| 1621 | VM_MAPPED_SPACE(&vm_info), VM_MAPPED_ADDR(&vm_info), |
| 1622 | (int)nbytes, B_WRITE, devvp); |
| 1623 | |
| 1624 | VASSERT(error == 0); |
| 1625 | |
| 1626 | #ifdef notdef |
| 1627 | /* |
| 1628 | * If we are vhand paging over NFS we want to wait for the |
| 1629 | * I/O to complete and take the appropriate actions if an |
| 1630 | * error is encountered. |
| 1631 | */ |
| 1632 | if (vhand) { |
| 1633 | if (waitforpageio(bp) && nfs_mi_harddown(filevp)) { |
| 1634 | /* |
| 1635 | * The server is down, ignore this failure, and |
| 1636 | * try again later. (rfscall() has set our retry |
| 1637 | * timer). |
| 1638 | */ |
| 1639 | fsdata.remote_down = 1; |
| 1640 | pageiocleanup(bp, 0); |
| 1641 | |
| 1642 | /* |
| 1643 | * vm_vfdcheck() has cleared the valid bit on the |
| 1644 | * vfds for these pages. We must go back and set the |
| 1645 | * valid bit, as the pages are really not gone. |
| 1646 | * |
| 1647 | * NOTE: we can do this because we still hold (and have |
| 1648 | * not released) the region lock. |
| 1649 | */ |
| 1650 | if (steal) |
| 1651 | vm_undo_invalidation(&vm_info, vm_info.start, |
| 1652 | vm_info.end); |
| 1653 | } else { |
| 1654 | /* |
| 1655 | * The I/O succeeded, or we had an error that we do |
| 1656 | * not want to defer until later. Call pageidone() |
| 1657 | * to handle things. |
| 1658 | */ |
| 1659 | pageiodone(bp); |
| 1660 | } |
| 1661 | } |
| 1662 | #endif |
| 1663 | |
| 1664 | /* |
| 1665 | * And restore our credentials to what they were. |
| 1666 | */ |
| 1667 | set_kt_cred(u.u_kthreadp, old_cred); |
| 1668 | |
| 1669 | /* |
| 1670 | * If we reserved memory in vfs_vfdcheck(), (only for NFS) we |
| 1671 | * can now unreserve it. |
| 1672 | */ |
| 1673 | if (vm_info.vm_flags & PAGEOUT_RESERVED) { |
| 1674 | vm_info.vm_flags &= ~PAGEOUT_RESERVED; |
| 1675 | vm_release_malloc_memory(); |
| 1676 | } |
| 1677 | |
| 1678 | /* |
| 1679 | * Update statistics |
| 1680 | */ |
| 1681 | if (steal) { |
| 1682 | if (flags & PF_DEACT) { |
| 1683 | #if defined(AFS_HPUX110_ENV) |
| 1684 | getppdp()->cnt.v_pswpout += npages; |
| 1685 | #else |
| 1686 | mpproc_info[getprocindex()].cnt.v_pswpout += npages; |
| 1687 | #endif |
| 1688 | /* sar_bswapout += ptod(npages);*/ |
| 1689 | } else if (vhand) { |
| 1690 | #if defined(AFS_HPUX110_ENV) |
| 1691 | getppdp()->cnt.v_pgout++; |
| 1692 | getppdp()->cnt.v_pgpgout += npages; |
| 1693 | #else |
| 1694 | mpproc_info[getprocindex()].cnt.v_pgout++; |
| 1695 | mpproc_info[getprocindex()].cnt.v_pgpgout += npages; |
| 1696 | #endif |
| 1697 | } |
| 1698 | } |
| 1699 | |
| 1700 | /* |
| 1701 | * If time and patience have delivered enough |
| 1702 | * pages, then quit now while we are ahead. |
| 1703 | */ |
| 1704 | if (VM_STOP_PAGING(&vm_info)) |
| 1705 | break; |
| 1706 | |
| 1707 | i = VM_END_PAGEOUT_INDX(&vm_info) - VM_BASE_OFFSET(&vm_info) + 1; |
| 1708 | } |
| 1709 | |
| 1710 | vm_finish_pageout(&vm_info); /* update vhand's stealscan */ |
| 1711 | |
| 1712 | vmemp_unlockx(); |
| 1713 | |
| 1714 | /* |
| 1715 | * If we wanted to wait for the I/O to complete, sleep on piocnt. |
| 1716 | * We must decrement it by one first, and then make sure that it |
| 1717 | * is non-zero before going to sleep. |
| 1718 | */ |
| 1719 | vm_wait_for_io(&vm_info); |
| 1720 | |
| 1721 | if (inode_changed && !file_is_remote) { |
| 1722 | imark(ip, IUPD | ICHG); |
| 1723 | iupdat(ip, 0, 0); |
| 1724 | } |
| 1725 | return 0; |
| 1726 | } |
| 1727 | |
| 1728 | int |
| 1729 | afs_mapdbd(filevp, offset, bn, flags, hole, startidx, endidx) |
| 1730 | struct vnode *filevp; |
| 1731 | off_t offset; |
| 1732 | kern_daddr_t *bn; /* Block number. */ |
| 1733 | int flags; /* B_READ or B_WRITE */ |
| 1734 | int *hole; /* To be used for read-ahead. */ |
| 1735 | pgcnt_t *startidx; /* To be used for read-ahead. */ |
| 1736 | pgcnt_t *endidx; /* To be used for read-ahead. */ |
| 1737 | { |
| 1738 | kern_daddr_t lbn, local_bn; |
| 1739 | int on; |
| 1740 | int err; |
| 1741 | long bsize = vtoblksz(filevp) & ~(DEV_BSIZE - 1); |
| 1742 | |
| 1743 | if (startidx) |
| 1744 | *startidx = (pgcnt_t) (offset / NBPG); |
| 1745 | if (endidx) |
| 1746 | *endidx = (pgcnt_t) (offset / NBPG); |
| 1747 | if (hole) |
| 1748 | *hole = 0; /* Can't have holes. */ |
| 1749 | if (bsize <= 0) |
| 1750 | osi_Panic("afs_mapdbd: zero size"); |
| 1751 | |
| 1752 | lbn = (kern_daddr_t) (offset / bsize); |
| 1753 | on = offset % bsize; |
| 1754 | |
| 1755 | err = VOP_BMAP(filevp, lbn, NULL, &local_bn, flags); |
| 1756 | VASSERT(err == 0); |
| 1757 | |
| 1758 | /* |
| 1759 | * We can never get a bn less than zero on remote files. |
| 1760 | */ |
| 1761 | VASSERT(local_bn >= 0); |
| 1762 | |
| 1763 | local_bn = local_bn + btodb(on); |
| 1764 | *bn = local_bn; |
| 1765 | |
| 1766 | return (0); |
| 1767 | } |
| 1768 | |
| 1769 | /* |
| 1770 | * Return values: |
| 1771 | * 1: The blocks are contiguous. |
| 1772 | * 0: The blocks are not contiguous. |
| 1773 | */ |
| 1774 | int |
| 1775 | afs_vm_fscontiguous(vp, args, cur_data) |
| 1776 | struct vnode *vp; |
| 1777 | vfspage_t *args; |
| 1778 | u_int cur_data; |
| 1779 | { |
| 1780 | if (cur_data == (VM_END_PAGEOUT_BLK(args) + btodb(NBPG))) { |
| 1781 | return (1); |
| 1782 | } else { |
| 1783 | return (0); |
| 1784 | } |
| 1785 | } |
| 1786 | |
| 1787 | /* |
| 1788 | * Return values: |
| 1789 | * 1: Stop, this page is the last in the block. |
| 1790 | * 0: Continue on |
| 1791 | * Terminate requests at filesystem block boundaries |
| 1792 | */ |
| 1793 | afs_vm_stopio(vp, args) |
| 1794 | struct vnode *vp; |
| 1795 | vfspage_t *args; |
| 1796 | { |
| 1797 | fsdata_t *fsdata = (fsdata_t *) args->fs_data; |
| 1798 | |
| 1799 | #if defined(AFS_HPUX1123_ENV) |
| 1800 | uint64_t tmpdb; |
| 1801 | tmpdb = VM_END_PAGEOUT_BLK(args); |
| 1802 | |
| 1803 | if ((dbtob(tmpdb) + NBPG) % (fsdata->bsize) == 0) |
| 1804 | #else |
| 1805 | if ((dbtob(VM_END_PAGEOUT_BLK(args)) + NBPG) % (fsdata->bsize) == 0) |
| 1806 | #endif /* AFS_HPUX1123_ENV */ |
| 1807 | { |
| 1808 | return (1); |
| 1809 | } else { |
| 1810 | return (0); |
| 1811 | } |
| 1812 | } |
| 1813 | |
| 1814 | /* |
| 1815 | * afs_vm_checkpage is called by the VM while collecting a run of |
| 1816 | * pages on a pageout. afs_vm_checkpage() is called for each page |
| 1817 | * VM wants to write to disk. |
| 1818 | */ |
| 1819 | afs_vm_checkpage(vp, args, pgindx, cur_data) |
| 1820 | struct vnode *vp; |
| 1821 | vfspage_t *args; |
| 1822 | pgcnt_t pgindx; |
| 1823 | int cur_data; |
| 1824 | { |
| 1825 | fsdata_t *fsdata = (fsdata_t *) args->fs_data; |
| 1826 | |
| 1827 | if (fsdata->remote_down) { /* never happens for AFS */ |
| 1828 | /* |
| 1829 | * The remote system is down. |
| 1830 | */ |
| 1831 | VASSERT(args->run == 0); |
| 1832 | return 1; |
| 1833 | } |
| 1834 | /* |
| 1835 | * A dirty page. If we have not yet determined the file size and |
| 1836 | * other attributes that we need to write out pages (the block |
| 1837 | * size and ok_dbd_limit), get that information now. |
| 1838 | */ |
| 1839 | if (fsdata->bsize == 0) { |
| 1840 | k_off_t isize; |
| 1841 | long bsize; |
| 1842 | struct vattr va; |
| 1843 | struct vnode *filevp; |
| 1844 | /* |
| 1845 | * Get the various attributes about the file. Store them |
| 1846 | * in args for the next time around. |
| 1847 | */ |
| 1848 | filevp = args->vp; |
| 1849 | |
| 1850 | bsize = vtoblksz(filevp); |
| 1851 | args->maxpgs = (pgcnt_t) btop(bsize); |
| 1852 | |
| 1853 | if (VOP_GETATTR(filevp, &va, kt_cred(u.u_kthreadp), VIFSYNC) != 0) { |
| 1854 | /* |
| 1855 | * The VOP_GETATTR() failed. |
| 1856 | * we are vhand, and this is a hard mount, we will |
| 1857 | * skip dirty pages for a while and try again later. |
| 1858 | */ |
| 1859 | if (args->vm_flags & PAGEOUT_VHAND) { |
| 1860 | VASSERT(args->run == 0); |
| 1861 | return 1; |
| 1862 | } |
| 1863 | /* |
| 1864 | * This is a "soft" mount, or some other error was |
| 1865 | * returned from the server. Mark this region |
| 1866 | * as a zombie, and free this dirty page. |
| 1867 | */ |
| 1868 | VM_ZOMBIE_OBJECT(args); |
| 1869 | |
| 1870 | /* |
| 1871 | * The caller will see r_zomb and remove the page |
| 1872 | * appropriately. |
| 1873 | */ |
| 1874 | return (1); |
| 1875 | } |
| 1876 | isize = va.va_size; |
| 1877 | fsdata->isize = isize; |
| 1878 | fsdata->bsize = bsize; |
| 1879 | fsdata->remote = 1; |
| 1880 | } |
| 1881 | /* |
| 1882 | * See if the file has shrunk (this could have happened |
| 1883 | * asynchronously because of NFS or DUX). If so, invalidate |
| 1884 | * all of the pages past the end of the file. This is only |
| 1885 | * needed for remote files, as local files are truncated |
| 1886 | * synchronously. |
| 1887 | */ |
| 1888 | |
| 1889 | if (vnodindx(VM_REGION(args), pgindx) > fsdata->isize) { |
| 1890 | /* |
| 1891 | * This page is past the end of the file. Unlock this page |
| 1892 | * (region_trunc will throw it away) and then call region_trunc() |
| 1893 | * to invalidate all pages past the new end of the file. |
| 1894 | */ |
| 1895 | VM_SETFS_FLAGS(args, PAGEOUT_TRUNCATE); |
| 1896 | return (1); |
| 1897 | } |
| 1898 | #ifdef notdef |
| 1899 | if ((args->vm_flags & PAGEOUT_VHAND) |
| 1900 | && (!(args->vm_flags & PAGEOUT_RESERVED)) |
| 1901 | && (!(VM_IS_ZOMBIE(args)))) { |
| 1902 | VASSERT(args->run == 0); |
| 1903 | if (vm_reserve_malloc_memory(NFS_PAGEOUT_MEM)) { |
| 1904 | /* |
| 1905 | * Got enough memory to pageout. Mark the fact that we did |
| 1906 | * a sysprocmemreserve(), so that we can sysprocmemunreserve() it |
| 1907 | * later (in remote_pageout()). |
| 1908 | */ |
| 1909 | args->vm_flags |= PAGEOUT_RESERVED; |
| 1910 | } else { |
| 1911 | /* |
| 1912 | * We do not have enough memory to do this pageout. By |
| 1913 | * definition, we do not yet have a run, so we just unlock |
| 1914 | * this page and tell foreach_valid() to continue scanning. |
| 1915 | * If we come across another dirty page, we will try to |
| 1916 | * reserve memory again. That is okay, in fact some memory |
| 1917 | * may have freed up (as earlier pageouts complete under |
| 1918 | * interrupt). |
| 1919 | */ |
| 1920 | return 1; |
| 1921 | } |
| 1922 | } |
| 1923 | #endif |
| 1924 | return (0); |
| 1925 | } |
| 1926 | |
| 1927 | afs_swapfs_len(bp) |
| 1928 | struct buf *bp; |
| 1929 | { |
| 1930 | long fs_bsize; |
| 1931 | long max_size; |
| 1932 | long bnrem; |
| 1933 | |
| 1934 | fs_bsize = vtoblksz(bp->b_vp); |
| 1935 | /* |
| 1936 | * Check to see if we are starting mid block. If so, then |
| 1937 | * we must return the remainder of the block or less depending |
| 1938 | * on the length. |
| 1939 | */ |
| 1940 | bnrem = bp->b_offset % fs_bsize; |
| 1941 | if (bnrem) { |
| 1942 | max_size = fs_bsize - bnrem; |
| 1943 | } else { |
| 1944 | max_size = fs_bsize; |
| 1945 | } |
| 1946 | |
| 1947 | if (bp->b_bcount > max_size) { |
| 1948 | return (max_size); |
| 1949 | } else { |
| 1950 | return (bp->b_bcount); |
| 1951 | } |
| 1952 | } |
| 1953 | |
| 1954 | afs_mmap(vp, off, size_bytes, access) |
| 1955 | struct vnode *vp; |
| 1956 | u_int off; |
| 1957 | #if defined(AFS_HPUX1111_ENV) |
| 1958 | u_long size_bytes; |
| 1959 | #else |
| 1960 | u_int size_bytes; |
| 1961 | #endif |
| 1962 | int access; |
| 1963 | { |
| 1964 | long bsize = vtoblksz(vp); |
| 1965 | |
| 1966 | if (bsize % NBPG != 0) { |
| 1967 | return (EINVAL); |
| 1968 | } |
| 1969 | |
| 1970 | return (0); |
| 1971 | } |
| 1972 | |
| 1973 | afs_cachelimit(vp, len, location) |
| 1974 | struct vnode *vp; |
| 1975 | k_off_t len; |
| 1976 | int *location; |
| 1977 | { |
| 1978 | /* |
| 1979 | * Disk addresses are logical, not physical, so fragments are |
| 1980 | * transparent. |
| 1981 | */ |
| 1982 | *location = btorp(len) + 1; |
| 1983 | } |
| 1984 | |
| 1985 | afs_release(vp) |
| 1986 | struct vnode *vp; |
| 1987 | { |
| 1988 | return (0); |
| 1989 | } |
| 1990 | |
| 1991 | int |
| 1992 | afs_unmap(vp, off, size_bytes, access) |
| 1993 | struct vnode *vp; |
| 1994 | u_int off; |
| 1995 | #if defined(AFS_HPUX1111_ENV) |
| 1996 | u_long size_bytes; |
| 1997 | #else |
| 1998 | u_int size_bytes; |
| 1999 | #endif |
| 2000 | int access; |
| 2001 | { |
| 2002 | return 0; |
| 2003 | } |
| 2004 | |
| 2005 | int |
| 2006 | afs_read_ahead(vp, prp, wrt, space, vaddr, rhead_cnt) |
| 2007 | struct vnode *vp; |
| 2008 | preg_t *prp; |
| 2009 | int wrt; |
| 2010 | space_t space; |
| 2011 | caddr_t vaddr; |
| 2012 | pgcnt_t *rhead_cnt; |
| 2013 | { |
| 2014 | printf("afs_read_ahead returning 0 \n"); |
| 2015 | return 0; |
| 2016 | } |
| 2017 | |
| 2018 | int |
| 2019 | afs_prealloc(vp, size, ignore_minfree, reserved) |
| 2020 | struct vnode *vp; |
| 2021 | /* DEE on 11.22 following is off_t */ |
| 2022 | size_t size; |
| 2023 | int ignore_minfree; |
| 2024 | int reserved; |
| 2025 | { |
| 2026 | printf("afs_prealloc returning ENOSPC\n"); |
| 2027 | return ENOSPC; |
| 2028 | } |
| 2029 | |
| 2030 | int |
| 2031 | afs_ioctl(vp, com, data, flag, cred) |
| 2032 | struct vnode *vp; |
| 2033 | int com; |
| 2034 | caddr_t data; |
| 2035 | int flag; |
| 2036 | struct ucred *cred; |
| 2037 | { |
| 2038 | int error; |
| 2039 | struct afs_ioctl afsioctl, *ai; |
| 2040 | |
| 2041 | AFS_STATCNT(afs_ioctl); |
| 2042 | |
| 2043 | /* The call must be a VICEIOCTL call */ |
| 2044 | if (((com >> 8) & 0xff) == 'V') { |
| 2045 | #ifdef notdef |
| 2046 | /* AFS_COPYIN returns error 14. Copy data in instead */ |
| 2047 | AFS_COPYIN(data, (caddr_t) & afsioctl, sizeof(afsioctl), error); |
| 2048 | if (error) |
| 2049 | return (error); |
| 2050 | #endif |
| 2051 | ai = (struct afs_ioctl *)data; |
| 2052 | afsioctl.in = ai->in; |
| 2053 | afsioctl.out = ai->out; |
| 2054 | afsioctl.in_size = ai->in_size; |
| 2055 | afsioctl.out_size = ai->out_size; |
| 2056 | error = HandleIoctl(VTOAFS(vp), com, &afsioctl); |
| 2057 | return (error); |
| 2058 | } |
| 2059 | return (ENOTTY); |
| 2060 | } |
| 2061 | |
| 2062 | #if defined(AFS_HPUX1111_ENV) |
| 2063 | /* looks like even if appl is 32 bit, we need to round to 8 bytes */ |
| 2064 | /* This had no effect, it must not be being used */ |
| 2065 | |
| 2066 | #define roundtoint(x) (((x) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) |
| 2067 | #define reclen(dp) roundtoint(((dp)->d_namlen + 1 + (sizeof(u_long)) +\ |
| 2068 | sizeof(u_int) + 2 * sizeof(u_short))) |
| 2069 | #else |
| 2070 | |
| 2071 | #define roundtoint(x) (((x) + (sizeof(int) - 1)) & ~(sizeof(int) - 1)) |
| 2072 | #define reclen(dp) roundtoint(((dp)->d_namlen + 1 + (sizeof(u_long)) +\ |
| 2073 | 2 * sizeof(u_short))) |
| 2074 | #endif |
| 2075 | |
| 2076 | int |
| 2077 | afs_readdir(vp, uiop, cred) |
| 2078 | struct vnode *vp; |
| 2079 | struct uio *uiop; |
| 2080 | struct ucred *cred; |
| 2081 | { |
| 2082 | struct uio auio; |
| 2083 | struct iovec aiov; |
| 2084 | caddr_t ibuf, obuf, ibufend, obufend; |
| 2085 | struct __dirent32 *idp; |
| 2086 | struct dirent *odp; |
| 2087 | int count, outcount; |
| 2088 | dir_off_t offset; |
| 2089 | uint64_t tmp_offset; |
| 2090 | |
| 2091 | memset(&auio, 0, sizeof(auio)); |
| 2092 | memset(&aiov, 0, sizeof(aiov)); |
| 2093 | |
| 2094 | count = uiop->uio_resid; |
| 2095 | /* Allocate temporary space for format conversion */ |
| 2096 | ibuf = kmem_alloc(2 * count); /* overkill - fix later */ |
| 2097 | obuf = kmem_alloc(count + sizeof(struct dirent)); |
| 2098 | aiov.iov_base = ibuf; |
| 2099 | aiov.iov_len = count; |
| 2100 | auio.uio_iov = &aiov; |
| 2101 | auio.uio_iovcnt = 1; |
| 2102 | offset = auio.uio_offset = uiop->uio_offset; |
| 2103 | auio.uio_seg = UIOSEG_KERNEL; |
| 2104 | auio.uio_resid = count; |
| 2105 | auio.uio_fpflags = 0; |
| 2106 | |
| 2107 | u.u_error = mp_afs_readdir2(vp, &auio, cred); |
| 2108 | if (u.u_error) |
| 2109 | goto out; |
| 2110 | |
| 2111 | /* Convert entries from __dirent32 to dirent format */ |
| 2112 | |
| 2113 | for (idp = (struct __dirent32 *)ibuf, odp = |
| 2114 | (struct dirent *)obuf, ibufend = |
| 2115 | ibuf + (count - auio.uio_resid), obufend = obuf + count; |
| 2116 | (caddr_t) idp < ibufend; |
| 2117 | idp = (struct __dirent32 *)((caddr_t) idp + idp->__d_reclen), odp = |
| 2118 | (struct dirent *)((caddr_t) odp + odp->d_reclen)) { |
| 2119 | odp->d_ino = idp->__d_ino; |
| 2120 | odp->d_namlen = idp->__d_namlen; |
| 2121 | (void)strcpy(odp->d_name, idp->__d_name); |
| 2122 | odp->d_reclen = reclen(odp); |
| 2123 | if ((caddr_t) odp + odp->d_reclen > obufend) |
| 2124 | break; |
| 2125 | /* record offset *after* we're sure to use this entry */ |
| 2126 | memcpy((char *)&tmp_offset, (char *)&idp->__d_off, sizeof tmp_offset); |
| 2127 | offset = tmp_offset; |
| 2128 | } |
| 2129 | |
| 2130 | outcount = (caddr_t) odp - obuf; |
| 2131 | AFS_UIOMOVE(obuf, outcount, UIO_READ, uiop, u.u_error); |
| 2132 | if (u.u_error) |
| 2133 | goto out; |
| 2134 | uiop->uio_offset = offset; |
| 2135 | out: |
| 2136 | kmem_free(ibuf, count); |
| 2137 | kmem_free(obuf, count + sizeof(struct dirent)); |
| 2138 | return u.u_error; |
| 2139 | } |
| 2140 | |
| 2141 | |
| 2142 | #define roundtolong(x) (((x) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) |
| 2143 | #define reclen_dirent64(dp) roundtolong(((dp)->__d_namlen + 1 + (2*sizeof(u_long)) +\ |
| 2144 | 2 * sizeof(u_short))) |
| 2145 | |
| 2146 | int |
| 2147 | afs_readdir3(vp, uiop, cred) |
| 2148 | struct vnode *vp; |
| 2149 | struct uio *uiop; |
| 2150 | struct ucred *cred; |
| 2151 | { |
| 2152 | struct uio auio; |
| 2153 | struct iovec aiov; |
| 2154 | caddr_t ibuf, obuf, ibufend, obufend; |
| 2155 | struct __dirent32 *idp; |
| 2156 | struct __dirent64 *odp; |
| 2157 | int count, outcount; |
| 2158 | dir_off_t offset; |
| 2159 | |
| 2160 | memset(&auio, 0, sizeof(auio)); |
| 2161 | memset(&aiov, 0, sizeof(aiov)); |
| 2162 | |
| 2163 | count = uiop->uio_resid; |
| 2164 | /* Allocate temporary space for format conversion */ |
| 2165 | ibuf = kmem_alloc(2 * count); /* overkill - fix later */ |
| 2166 | obuf = kmem_alloc(count + sizeof(struct __dirent64)); |
| 2167 | aiov.iov_base = ibuf; |
| 2168 | aiov.iov_len = count; |
| 2169 | auio.uio_iov = &aiov; |
| 2170 | auio.uio_iovcnt = 1; |
| 2171 | offset = auio.uio_offset = uiop->uio_offset; |
| 2172 | auio.uio_seg = UIOSEG_KERNEL; |
| 2173 | auio.uio_resid = count; |
| 2174 | auio.uio_fpflags = 0; |
| 2175 | |
| 2176 | u.u_error = mp_afs_readdir2(vp, &auio, cred); |
| 2177 | if (u.u_error) |
| 2178 | goto out; |
| 2179 | |
| 2180 | /* Convert entries from __dirent32 to __dirent64 format */ |
| 2181 | |
| 2182 | for (idp = (struct __dirent32 *)ibuf, odp = |
| 2183 | (struct __dirent64 *)obuf, ibufend = |
| 2184 | ibuf + (count - auio.uio_resid), obufend = obuf + count; |
| 2185 | (caddr_t) idp < ibufend; |
| 2186 | idp = (struct __dirent32 *)((caddr_t) idp + idp->__d_reclen), odp = |
| 2187 | (struct __dirent64 *)((caddr_t) odp + odp->__d_reclen)) { |
| 2188 | memcpy((char *)&odp->__d_off, (char *)&idp->__d_off, |
| 2189 | sizeof odp->__d_off); |
| 2190 | odp->__d_ino = idp->__d_ino; |
| 2191 | odp->__d_namlen = idp->__d_namlen; |
| 2192 | (void)strcpy(odp->__d_name, idp->__d_name); |
| 2193 | odp->__d_reclen = reclen_dirent64(odp); |
| 2194 | if ((caddr_t) odp + odp->__d_reclen > obufend) |
| 2195 | break; |
| 2196 | /* record offset *after* we're sure to use this entry */ |
| 2197 | offset = odp->__d_off; |
| 2198 | } |
| 2199 | |
| 2200 | outcount = (caddr_t) odp - obuf; |
| 2201 | AFS_UIOMOVE(obuf, outcount, UIO_READ, uiop, u.u_error); |
| 2202 | if (u.u_error) |
| 2203 | goto out; |
| 2204 | uiop->uio_offset = offset; |
| 2205 | out: |
| 2206 | kmem_free(ibuf, count); |
| 2207 | kmem_free(obuf, count + sizeof(struct __dirent64)); |
| 2208 | return u.u_error; |
| 2209 | } |
| 2210 | |
| 2211 | #define AFS_SV_SEMA_HASH 1 |
| 2212 | #define AFS_SV_SEMA_HASH_DEBUG 0 |
| 2213 | |
| 2214 | #if AFS_SV_SEMA_HASH |
| 2215 | /* This portion of the code was originally used to implement |
| 2216 | * thread specific storage for the semaphore save area. However, |
| 2217 | * there were some spare fields in the proc structure, this is |
| 2218 | * now being used for the saving semapores. Hence, this portion of |
| 2219 | * the code is no longer used. |
| 2220 | */ |
| 2221 | |
| 2222 | /* This portion of the code implements thread specific information. |
| 2223 | * The thread id is passed in as the key. The semaphore saved area |
| 2224 | * is hashed on this key. |
| 2225 | */ |
| 2226 | |
| 2227 | /* why is this hash table required ? |
| 2228 | * The AFS code is written in such a way that a GLOCK() is done in |
| 2229 | * one function and the GUNLOCK() is done in another function further |
| 2230 | * down the call chain. The GLOCK() call has to save the current |
| 2231 | * semaphore status before acquiring afs_global_sema. The GUNLOCK |
| 2232 | * has to release afs_global_sema and reacquire the sempahore status |
| 2233 | * that existed before the corresponding GLOCK. If GLOCK() and |
| 2234 | * GUNLOCK() were called in the same function, the GLOCK call could |
| 2235 | * have stored the saved sempahore status in a local variable and the |
| 2236 | * corresponding GUNLOCK() call could have restored the original |
| 2237 | * status from this local variable. But this is not the case with |
| 2238 | * AFS code. Hence, we have to implement a thread specific semaphore |
| 2239 | * save area. This is implemented as a hash table. The key is the |
| 2240 | * thread id. |
| 2241 | */ |
| 2242 | |
| 2243 | /* In order for multithreaded processes to work, the sv_sema structures |
| 2244 | * must be saved on a per-thread basis, not a per-process basis. There |
| 2245 | * is no per-thread storage available to hijack in the OS per-thread |
| 2246 | * data structures (e.g. struct user) so we revive this code. |
| 2247 | * I removed the upper limit on the memory consumption since we don't |
| 2248 | * know how many threads there will be. Now the code first checks the |
| 2249 | * freeList. If that fails it then tries garbage collecting. If that |
| 2250 | * doesn't free up anything then it allocs what it needs. |
| 2251 | */ |
| 2252 | |
| 2253 | #define ELEMENT sv_sema_t |
| 2254 | #define KEY tid_t |
| 2255 | #define Hash(xx) ( (xx) % sizeOfHashTable ) |
| 2256 | #define hashLockInit(xx) initsema(&xx,1, FILESYS_SEMA_PRI, FILESYS_SEMA_ORDER) |
| 2257 | #define hashLock(xx) MP_PSEMA(&xx) |
| 2258 | #define hashUnlock(xx) MP_VSEMA(&xx) |
| 2259 | |
| 2260 | typedef struct elem { |
| 2261 | struct elem *next; |
| 2262 | ELEMENT element; |
| 2263 | KEY key; |
| 2264 | int refCnt; |
| 2265 | } Element; |
| 2266 | |
| 2267 | typedef struct bucket { |
| 2268 | sema_t lock; |
| 2269 | Element *element; |
| 2270 | } Bucket; |
| 2271 | |
| 2272 | static int sizeOfHashTable; |
| 2273 | static Bucket *hashTable; |
| 2274 | |
| 2275 | static int currentSize = 0; |
| 2276 | static Element *freeList; /* free list */ |
| 2277 | |
| 2278 | #pragma align 64 |
| 2279 | static sema_t afsHashLock = { 0 }; /* global lock for hash table */ |
| 2280 | |
| 2281 | static void afsHashGarbageCollect(); |
| 2282 | |
| 2283 | /* |
| 2284 | ** The global lock protects the global data structures, |
| 2285 | ** e.g. freeList and currentSize. |
| 2286 | ** The bucket lock protects the link list hanging off that bucket. |
| 2287 | ** The lock hierarchy : one can obtain the bucket lock while holding |
| 2288 | ** the global lock, but not vice versa. |
| 2289 | */ |
| 2290 | |
| 2291 | |
| 2292 | void |
| 2293 | afsHash(int nbuckets) |
| 2294 | { /* allocate the hash table */ |
| 2295 | int i; |
| 2296 | |
| 2297 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2298 | printf("afsHash: enter\n"); |
| 2299 | #endif |
| 2300 | |
| 2301 | sizeOfHashTable = nbuckets; |
| 2302 | currentSize = nbuckets * sizeof(Bucket); |
| 2303 | |
| 2304 | if (hashTable) |
| 2305 | osi_Panic("afs: SEMA Hashtable already created\n"); |
| 2306 | |
| 2307 | hashTable = (Bucket *) AFS_KALLOC(sizeOfHashTable * sizeof(Bucket)); |
| 2308 | if (!hashTable) |
| 2309 | osi_Panic("afs: cannot create SEMA Hashtable\n"); |
| 2310 | |
| 2311 | /* initialize the hash table and associated locks */ |
| 2312 | memset(hashTable, 0, sizeOfHashTable * sizeof(Bucket)); |
| 2313 | for (i = 0; i < sizeOfHashTable; i++) |
| 2314 | hashLockInit(hashTable[i].lock); |
| 2315 | hashLockInit(afsHashLock); |
| 2316 | |
| 2317 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2318 | printf("afsHash: exit\n"); |
| 2319 | #endif |
| 2320 | } |
| 2321 | |
| 2322 | ELEMENT * |
| 2323 | afsHashInsertFind(KEY key) |
| 2324 | { |
| 2325 | int index; |
| 2326 | Element *ptr; |
| 2327 | |
| 2328 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2329 | printf("afsHashInsertFind: %d\n", key); |
| 2330 | #endif |
| 2331 | if (!hashTable) |
| 2332 | osi_Panic("afs: afsHashInsertFind: no hashTable\n"); |
| 2333 | |
| 2334 | index = Hash(key); /* get bucket number */ |
| 2335 | hashLock(hashTable[index].lock); /* lock this bucket */ |
| 2336 | ptr = hashTable[index].element; |
| 2337 | |
| 2338 | /* if it is already there */ |
| 2339 | while (ptr) { |
| 2340 | if (ptr->key == key) { |
| 2341 | ptr->refCnt++; /* hold it */ |
| 2342 | hashUnlock(hashTable[index].lock); |
| 2343 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2344 | printf("afsHashInsertFind: %d FOUND\n", key); |
| 2345 | #endif |
| 2346 | return &(ptr->element); |
| 2347 | } else { |
| 2348 | ptr = ptr->next; |
| 2349 | } |
| 2350 | } |
| 2351 | |
| 2352 | hashUnlock(hashTable[index].lock); |
| 2353 | |
| 2354 | /* if something exists in the freeList, take it from there */ |
| 2355 | ptr = NULL; |
| 2356 | hashLock(afsHashLock); |
| 2357 | |
| 2358 | if (freeList) { |
| 2359 | ptr = freeList; /* reuse entry */ |
| 2360 | freeList = freeList->next; |
| 2361 | } else { |
| 2362 | afsHashGarbageCollect(); /* afsHashLock locked */ |
| 2363 | if (freeList) { |
| 2364 | ptr = freeList; /* reuse entry */ |
| 2365 | freeList = freeList->next; |
| 2366 | } else { |
| 2367 | ptr = (Element *) AFS_KALLOC(sizeof(Element)); |
| 2368 | } |
| 2369 | } |
| 2370 | |
| 2371 | currentSize += sizeof(Element); /* update memory used */ |
| 2372 | hashUnlock(afsHashLock); |
| 2373 | |
| 2374 | if (!ptr) |
| 2375 | osi_Panic("afs: SEMA Hashtable cannot create new entry\n"); |
| 2376 | /* create new entry */ |
| 2377 | ptr->key = key; |
| 2378 | memset(&ptr->element, 0, sizeof(ptr->element)); |
| 2379 | ptr->refCnt = 1; /* this guy */ |
| 2380 | |
| 2381 | /* insert new entry in bucket */ |
| 2382 | hashLock(hashTable[index].lock); /* lock this bucket */ |
| 2383 | ptr->next = hashTable[index].element; |
| 2384 | hashTable[index].element = ptr; |
| 2385 | hashUnlock(hashTable[index].lock); |
| 2386 | |
| 2387 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2388 | printf("afsHashInsertFind: %d MADE\n", key); |
| 2389 | #endif |
| 2390 | |
| 2391 | return &(ptr->element); |
| 2392 | } |
| 2393 | |
| 2394 | ELEMENT * |
| 2395 | afsHashFind(KEY key) |
| 2396 | { |
| 2397 | int index; |
| 2398 | Element *ptr; |
| 2399 | |
| 2400 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2401 | printf("afsHashFind: %d\n", key); |
| 2402 | #endif |
| 2403 | if (!hashTable) |
| 2404 | osi_Panic("afs: afsHashFind: no hashTable\n"); |
| 2405 | |
| 2406 | index = Hash(key); /* get bucket number */ |
| 2407 | hashLock(hashTable[index].lock); /* lock this bucket */ |
| 2408 | ptr = hashTable[index].element; |
| 2409 | |
| 2410 | /* it should be in the hash table */ |
| 2411 | while (ptr) { |
| 2412 | if (ptr->key == key) { |
| 2413 | if (ptr->refCnt <= 0) |
| 2414 | osi_Panic("afs: SEMA HashTable entry already released\n"); |
| 2415 | hashUnlock(hashTable[index].lock); |
| 2416 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2417 | printf("afsHashFind: %d FOUND\n", key); |
| 2418 | #endif |
| 2419 | return &(ptr->element); |
| 2420 | } else { |
| 2421 | ptr = ptr->next; |
| 2422 | } |
| 2423 | } |
| 2424 | |
| 2425 | hashUnlock(hashTable[index].lock); |
| 2426 | /* it better be in the hash table */ |
| 2427 | osi_Panic("afs: SEMA HashTable wants non-existent entry \n"); |
| 2428 | return 0; |
| 2429 | } |
| 2430 | |
| 2431 | void |
| 2432 | afsHashRelease(KEY key) |
| 2433 | { |
| 2434 | int index; |
| 2435 | Element *ptr; |
| 2436 | |
| 2437 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2438 | printf("afsHashRelease: %d\n", key); |
| 2439 | #endif |
| 2440 | if (!hashTable) |
| 2441 | osi_Panic("afs: afsHashRelease: no hashTable\n"); |
| 2442 | |
| 2443 | index = Hash(key); /* get bucket number */ |
| 2444 | hashLock(hashTable[index].lock); /* lock this bucket */ |
| 2445 | ptr = hashTable[index].element; |
| 2446 | |
| 2447 | /* it should be in the hash table */ |
| 2448 | while (ptr) { |
| 2449 | if (ptr->key == key) { |
| 2450 | if (ptr->refCnt <= 0) |
| 2451 | osi_Panic("afs: SEMA HashTable entry already released\n"); |
| 2452 | ptr->refCnt--; /* release this guy */ |
| 2453 | hashUnlock(hashTable[index].lock); |
| 2454 | #if AFS_SV_SEMA_HASH_DEBUG |
| 2455 | printf("afsHashRelease: %d FOUND\n", key); |
| 2456 | #endif |
| 2457 | return; |
| 2458 | } else { |
| 2459 | ptr = ptr->next; |
| 2460 | } |
| 2461 | } |
| 2462 | |
| 2463 | hashUnlock(hashTable[index].lock); |
| 2464 | /* it better be in the hash table */ |
| 2465 | osi_Panic("afs: SEMA HashTable deleting non-existent entry \n"); |
| 2466 | } |
| 2467 | |
| 2468 | /* this should be called with afsHashLock WRITE locked */ |
| 2469 | static void |
| 2470 | afsHashGarbageCollect() |
| 2471 | { |
| 2472 | int index; |
| 2473 | Element *ptr; |
| 2474 | int foundFlag = 0; |
| 2475 | |
| 2476 | if (!hashTable) |
| 2477 | osi_Panic("afs: afsHashGarbageCollect: no hashTable\n"); |
| 2478 | |
| 2479 | for (index = 0; index < sizeOfHashTable; index++) { |
| 2480 | hashLock(hashTable[index].lock); |
| 2481 | ptr = hashTable[index].element; /* pick up bucket */ |
| 2482 | |
| 2483 | while (ptr && !ptr->refCnt) { |
| 2484 | /* insert this element into free list */ |
| 2485 | Element *temp; |
| 2486 | temp = ptr->next; |
| 2487 | ptr->next = freeList; |
| 2488 | freeList = ptr; |
| 2489 | |
| 2490 | foundFlag = 1; /* found at least one */ |
| 2491 | currentSize -= sizeof(Element); |
| 2492 | ptr = temp; |
| 2493 | } |
| 2494 | hashTable[index].element = ptr; |
| 2495 | |
| 2496 | /* scan thru the remaining list */ |
| 2497 | if (ptr) { |
| 2498 | while (ptr->next) { |
| 2499 | if (ptr->next->refCnt == 0) { |
| 2500 | /* collect this element */ |
| 2501 | Element *temp; |
| 2502 | temp = ptr->next; |
| 2503 | ptr->next = ptr->next->next; |
| 2504 | temp->next = freeList; |
| 2505 | freeList = temp; |
| 2506 | foundFlag = 1; |
| 2507 | currentSize -= sizeof(Element); |
| 2508 | } else { |
| 2509 | ptr = ptr->next; |
| 2510 | } |
| 2511 | } |
| 2512 | } |
| 2513 | hashUnlock(hashTable[index].lock); |
| 2514 | } |
| 2515 | #if 0 |
| 2516 | if (!foundFlag) |
| 2517 | osi_Panic("afs: SEMA HashTable full\n"); |
| 2518 | #endif |
| 2519 | } |
| 2520 | |
| 2521 | #endif /* AFS_SV_SEMA_HASH */ |
| 2522 | |
| 2523 | |
| 2524 | afs_hp_strategy(bp) |
| 2525 | struct buf *bp; |
| 2526 | { |
| 2527 | afs_int32 code; |
| 2528 | struct uio tuio; |
| 2529 | struct iovec tiovec[1]; |
| 2530 | extern caddr_t hdl_kmap_bp(); |
| 2531 | struct kthread *t = u.u_kthreadp; |
| 2532 | |
| 2533 | memset(&tuio, 0, sizeof(tuio)); |
| 2534 | memset(&tiovec, 0, sizeof(tiovec)); |
| 2535 | |
| 2536 | AFS_STATCNT(afs_hp_strategy); |
| 2537 | /* |
| 2538 | * hdl_kmap_bp() saves "b_bcount" and restores it in hdl_remap_bp() after |
| 2539 | * the I/O. We must save and restore the count because pageiodone() |
| 2540 | * uses b_bcount to determine how many pages to unlock. |
| 2541 | * |
| 2542 | * Remap the entire range. |
| 2543 | */ |
| 2544 | hdl_kmap_bp(bp); |
| 2545 | |
| 2546 | AFS_GLOCK(); |
| 2547 | afs_Trace4(afs_iclSetp, CM_TRACE_HPSTRAT, ICL_TYPE_POINTER, bp->b_vp, |
| 2548 | ICL_TYPE_LONG, (int)bp->b_blkno * DEV_BSIZE, ICL_TYPE_LONG, |
| 2549 | bp->b_bcount, ICL_TYPE_LONG, 0); |
| 2550 | |
| 2551 | /* Set up the uio structure */ |
| 2552 | tuio.afsio_iov = tiovec; |
| 2553 | tuio.afsio_iovcnt = 1; |
| 2554 | tuio.afsio_offset = DEV_BSIZE * bp->b_blkno; |
| 2555 | tuio.afsio_seg = AFS_UIOSYS; |
| 2556 | tuio.afsio_resid = bp->b_bcount; |
| 2557 | tuio.uio_fpflags = 0; |
| 2558 | tiovec[0].iov_base = bp->b_un.b_addr; |
| 2559 | tiovec[0].iov_len = bp->b_bcount; |
| 2560 | |
| 2561 | /* Do the I/O */ |
| 2562 | if ((bp->b_flags & B_READ) == B_READ) { |
| 2563 | /* read b_bcount bytes into kernel address b_un.b_addr |
| 2564 | * starting at byte DEV_BSIZE * b_blkno. Bzero anything |
| 2565 | * we can't read, and finally call iodone(bp). File is |
| 2566 | * in bp->b_vp. Credentials are from u area?? |
| 2567 | */ |
| 2568 | code = afs_rdwr(VTOAFS(bp->b_vp), &tuio, UIO_READ, 0, kt_cred(t)); |
| 2569 | if (code == 0) |
| 2570 | if (tuio.afsio_resid > 0) { |
| 2571 | privlbzero(bvtospace(bp, bp->b_un.b_addr), |
| 2572 | bp->b_un.b_addr + bp->b_bcount - tuio.afsio_resid, |
| 2573 | (size_t) tuio.afsio_resid); |
| 2574 | |
| 2575 | } |
| 2576 | } else |
| 2577 | code = afs_rdwr(VTOAFS(bp->b_vp), &tuio, UIO_WRITE, 0, kt_cred(t)); |
| 2578 | |
| 2579 | /* Remap back to the user's space */ |
| 2580 | hdl_remap_bp(bp); |
| 2581 | |
| 2582 | AFS_GUNLOCK(); |
| 2583 | |
| 2584 | iodone(bp); |
| 2585 | return code; |
| 2586 | } |
| 2587 | |
| 2588 | afs_pathconf(vp, name, resultp, cred) |
| 2589 | struct vnode *vp; |
| 2590 | int name; |
| 2591 | int *resultp; |
| 2592 | struct ucred *cred; /* unused */ |
| 2593 | { |
| 2594 | switch (name) { |
| 2595 | case _PC_LINK_MAX: /* Maximum number of links to a file */ |
| 2596 | *resultp = 255; /* an unsigned short on the fileserver */ |
| 2597 | break; /* a unsigned char in the client.... */ |
| 2598 | |
| 2599 | case _PC_NAME_MAX: /* Max length of file name */ |
| 2600 | *resultp = 255; |
| 2601 | break; |
| 2602 | |
| 2603 | case _PC_PATH_MAX: /* Maximum length of Path Name */ |
| 2604 | *resultp = 1024; |
| 2605 | break; |
| 2606 | |
| 2607 | case _PC_PIPE_BUF: /* Max atomic write to pipe. See fifo_vnops */ |
| 2608 | case _PC_CHOWN_RESTRICTED: /* Anybody can chown? */ |
| 2609 | case _PC_NO_TRUNC: /* No file name truncation on overflow? */ |
| 2610 | u.u_error = EOPNOTSUPP; |
| 2611 | return (EOPNOTSUPP); |
| 2612 | break; |
| 2613 | |
| 2614 | case _PC_MAX_CANON: /* TTY buffer size for canonical input */ |
| 2615 | /* need more work here for pty, ite buffer size, if differ */ |
| 2616 | if (vp->v_type != VCHR) { |
| 2617 | u.u_error = EINVAL; |
| 2618 | return (EINVAL); |
| 2619 | } |
| 2620 | *resultp = CANBSIZ; /*for tty */ |
| 2621 | break; |
| 2622 | |
| 2623 | case _PC_MAX_INPUT: |
| 2624 | /* need more work here for pty, ite buffer size, if differ */ |
| 2625 | if (vp->v_type != VCHR) { /* TTY buffer size */ |
| 2626 | u.u_error = EINVAL; |
| 2627 | return (EINVAL); |
| 2628 | } |
| 2629 | *resultp = TTYHOG; /*for tty */ |
| 2630 | break; |
| 2631 | |
| 2632 | case _PC_VDISABLE: |
| 2633 | /* Terminal special characters can be disabled? */ |
| 2634 | if (vp->v_type != VCHR) { |
| 2635 | u.u_error = EINVAL; |
| 2636 | return (EINVAL); |
| 2637 | } |
| 2638 | *resultp = 1; |
| 2639 | break; |
| 2640 | |
| 2641 | case _PC_SYNC_IO: |
| 2642 | if ((vp->v_type != VREG) && (vp->v_type != VBLK)) { |
| 2643 | *resultp = -1; |
| 2644 | return EINVAL; |
| 2645 | } |
| 2646 | *resultp = 1; /* Synchronized IO supported for this file */ |
| 2647 | break; |
| 2648 | |
| 2649 | case _PC_FILESIZEBITS: |
| 2650 | if (vp->v_type != VDIR) |
| 2651 | return (EINVAL); |
| 2652 | *resultp = MAX_SMALL_FILE_BITS; |
| 2653 | break; |
| 2654 | |
| 2655 | default: |
| 2656 | return (EINVAL); |
| 2657 | } |
| 2658 | |
| 2659 | return (0); |
| 2660 | } |