2 * Copyright 2000, International Business Machines Corporation and others.
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
11 * vnodeops structure and Digital Unix specific ops and support routines.
14 #include <afsconfig.h>
15 #include "../afs/param.h"
20 #include "../afs/sysincludes.h" /* Standard vendor system headers */
21 #include "../afs/afsincludes.h" /* Afs-based standard headers */
22 #include "../afs/afs_stats.h" /* statistics */
24 #include <vm/vnode_pager.h>
25 #include <vm/vm_map.h>
26 /* #include <vm/vm_ubc.h> */
27 #include "../afs/afs_cbqueue.h"
28 #include "../afs/nfsclient.h"
29 #include "../afs/afs_osidnlc.h"
32 extern int afs_lookup(), afs_create(), afs_noop(), afs_open(), afs_close();
33 extern int afs_access(), afs_getattr(), afs_setattr(), afs_badop();
34 extern int afs_fsync(), afs_seek(), afs_remove(), afs_link(), afs_rename();
35 extern int afs_mkdir(), afs_rmdir(), afs_symlink(), afs_readdir();
36 extern int afs_readlink(), afs_lockctl();
37 extern int vn_pathconf_default(), seltrue();
39 int mp_afs_lookup(), mp_afs_create(), mp_afs_open();
40 int mp_afs_access(), mp_afs_getattr(), mp_afs_setattr(), mp_afs_ubcrdwr();
41 int mp_afs_ubcrdwr(), mp_afs_mmap();
42 int mp_afs_fsync(), mp_afs_seek(), mp_afs_remove(), mp_afs_link();
43 int mp_afs_rename(), mp_afs_mkdir(), mp_afs_rmdir(), mp_afs_symlink();
44 int mp_afs_readdir(), mp_afs_readlink(), mp_afs_abortop(), mp_afs_inactive();
45 int mp_afs_reclaim(), mp_afs_bmap(), mp_afs_strategy(), mp_afs_print();
46 int mp_afs_page_read(), mp_afs_page_write(), mp_afs_swap(), mp_afs_bread();
47 int mp_afs_brelse(), mp_afs_lockctl(), mp_afs_syncdata(), mp_afs_close();
52 struct vnodeops Afs_vnodeops = {
55 afs_noop, /* vn_mknod */
63 afs_badop, /* vn_ioctl */
64 seltrue, /* vn_select */
90 afs_noop, /* unLock */
91 afs_noop, /* get ext attrs */
92 afs_noop, /* set ext attrs */
93 afs_noop, /* del ext attrs */
96 struct vnodeops *afs_ops = &Afs_vnodeops;
99 /* vnode file operations, and our own */
100 extern int vn_read();
101 extern int vn_write();
102 extern int vn_ioctl();
103 extern int vn_select();
104 extern int afs_closex();
106 struct fileops afs_fileops = {
115 mp_afs_lookup(adp, ndp)
117 struct nameidata *ndp;
121 code = afs_lookup(adp, ndp);
126 mp_afs_create(ndp, attrs)
127 struct nameidata *ndp;
132 code = afs_create(ndp, attrs);
137 mp_afs_open(avcp, aflags, acred)
138 struct vcache **avcp;
140 struct AFS_UCRED *acred;
144 code = afs_open(avcp, aflags, acred);
149 mp_afs_access(avc, amode, acred)
152 struct AFS_UCRED *acred;
156 code = afs_access(avc, amode, acred);
161 mp_afs_close(avc, flags, cred)
168 code = afs_close(avc, flags, cred);
173 mp_afs_getattr(avc, attrs, acred)
176 struct AFS_UCRED *acred;
180 code = afs_getattr(avc, attrs, acred);
185 mp_afs_setattr(avc, attrs, acred)
188 struct AFS_UCRED *acred;
192 code = afs_setattr(avc, attrs, acred);
197 mp_afs_fsync(avc, fflags, acred, waitfor)
200 struct AFS_UCRED *acred;
205 code = afs_fsync(avc, fflags, acred, waitfor);
211 struct nameidata *ndp;
215 code = afs_remove(ndp);
220 mp_afs_link(avc, ndp)
222 struct nameidata *ndp;
226 code = afs_link(avc, ndp);
231 mp_afs_rename(fndp, tndp)
232 struct nameidata *fndp, *tndp;
236 code = afs_rename(fndp, tndp);
241 mp_afs_mkdir(ndp, attrs)
242 struct nameidata *ndp;
247 code = afs_mkdir(ndp, attrs);
253 struct nameidata *ndp;
257 code = afs_rmdir(ndp);
262 mp_afs_symlink(ndp, attrs, atargetName)
263 struct nameidata *ndp;
265 register char *atargetName;
269 code = afs_symlink(ndp, attrs, atargetName);
274 mp_afs_readdir(avc, auio, acred, eofp)
277 struct AFS_UCRED *acred;
282 code = afs_readdir(avc, auio, acred, eofp);
287 mp_afs_readlink(avc, auio, acred)
290 struct AFS_UCRED *acred;
294 code = afs_readlink(avc, auio, acred);
299 mp_afs_lockctl(avc, af, flag, acred, clid, offset)
302 struct AFS_UCRED *acred;
309 code = afs_lockctl(avc, af, flag, acred, clid, offset);
319 code = afs_closex(afd);
324 mp_afs_seek(avc, oldoff, newoff, cred)
326 off_t oldoff, newoff;
329 if ((int) newoff < 0)
336 struct nameidata *ndp;
341 mp_afs_inactive(avc, acred)
342 register struct vcache *avc;
343 struct AFS_UCRED *acred;
346 afs_InactiveVCache(avc, acred);
363 mp_afs_page_read(avc, uio, acred)
369 struct vrequest treq;
372 error = afs_rdwr(avc, uio, UIO_READ, 0, acred);
373 afs_Trace3(afs_iclSetp, CM_TRACE_PAGE_READ, ICL_TYPE_POINTER, avc,
374 ICL_TYPE_INT32, error, ICL_TYPE_INT32, avc->states);
377 } else if ((avc->states) == 0) {
378 afs_InitReq(&treq, acred);
379 ObtainWriteLock(&avc->lock,161);
380 afs_Wire(avc, &treq);
381 ReleaseWriteLock(&avc->lock);
388 mp_afs_page_write(avc, uio, acred, pager, offset)
392 memory_object_t pager;
398 error = afs_rdwr(avc, uio, UIO_WRITE, 0, acred);
399 afs_Trace3(afs_iclSetp, CM_TRACE_PAGE_WRITE, ICL_TYPE_POINTER, avc,
400 ICL_TYPE_INT32, error, ICL_TYPE_INT32, avc->states);
410 mp_afs_ubcrdwr(avc, uio, ioflag, cred)
416 register afs_int32 code;
418 afs_int32 fileBase, size, cnt=0;
420 register afs_int32 tsize;
421 register afs_int32 pageOffset;
423 struct vrequest treq;
424 int rw = uio->uio_rw;
428 afs_int32 save_resid;
434 afs_InitReq(&treq, cred);
435 if (AFS_NFSXLATORREQ(cred) && rw == UIO_READ) {
436 if (!afs_AccessOK(avc, PRSFS_READ, &treq,
437 CHECK_MODE_BITS|CMB_ALLOW_EXEC_AS_READ)) {
442 afs_Trace4(afs_iclSetp, CM_TRACE_VMRW, ICL_TYPE_POINTER, avc,
443 ICL_TYPE_INT32, (rw==UIO_WRITE? 1 : 0),
444 ICL_TYPE_LONG, uio->uio_offset,
445 ICL_TYPE_LONG, uio->uio_resid);
446 code = afs_VerifyVCache(avc, &treq);
448 code = afs_CheckCode(code, &treq, 35);
452 if (vType(avc) != VREG) {
454 return EISDIR; /* can't read or write other things */
456 afs_BozonLock(&avc->pvnLock, avc);
457 osi_FlushPages(avc); /* hold bozon lock, but not basic vnode lock */
458 ObtainWriteLock(&avc->lock,162);
459 /* adjust parameters when appending files */
460 if ((ioflag & IO_APPEND) && uio->uio_rw == UIO_WRITE)
461 uio->uio_offset = avc->m.Length; /* write at EOF position */
462 if (uio->uio_rw == UIO_WRITE) {
463 avc->states |= CDirty;
467 * before starting any I/O, we must ensure that the file is big enough
468 * to hold the results (since afs_putpage will be called to force
471 size = uio->afsio_resid + uio->afsio_offset; /* new file size */
472 if (size > avc->m.Length) avc->m.Length = size; /* file grew */
473 avc->m.Date = osi_Time(); /* Set file date (for ranlib) */
474 if (uio->afsio_resid > PAGE_SIZE)
475 cnt = uio->afsio_resid / PAGE_SIZE;
476 save_resid = uio->afsio_resid;
481 * compute the amount of data to move into this block,
482 * based on uio->afsio_resid.
484 size = uio->afsio_resid; /* transfer size */
485 fileBase = uio->afsio_offset; /* start file position */
486 pageBase = fileBase & ~(PAGE_SIZE-1); /* file position of the page */
487 pageOffset = fileBase & (PAGE_SIZE-1); /* start offset within page */
488 tsize = PAGE_SIZE-pageOffset; /* amount left in this page */
490 * we'll read tsize bytes,
491 * but first must make sure tsize isn't too big
493 if (tsize > size) tsize = size; /* don't read past end of request */
494 eof = 0; /* flag telling us if we hit the EOF on the read */
495 if (uio->uio_rw == UIO_READ) { /* we're doing a read operation */
496 /* don't read past EOF */
497 if (tsize + fileBase > avc->m.Length) {
498 tsize = avc->m.Length - fileBase;
499 eof = 1; /* we did hit the EOF */
500 if (tsize < 0) tsize = 0; /* better safe than sorry */
503 if (tsize <= 0) break; /* nothing to transfer, we're done */
505 /* Purge dirty chunks of file if there are too many dirty chunks.
506 * Inside the write loop, we only do this at a chunk boundary.
507 * Clean up partial chunk if necessary at end of loop.
509 if (uio->uio_rw == UIO_WRITE && counter > 0
510 && AFS_CHUNKOFFSET(fileBase) == 0) {
511 code = afs_DoPartialWrite(avc, &treq);
512 avc->states |= CDirty;
520 ReleaseWriteLock(&avc->lock);
522 code = ubc_lookup(((struct vnode *)avc)->v_object, pageBase,
523 PAGE_SIZE, PAGE_SIZE, &page, &flags);
525 ObtainWriteLock(&avc->lock,163);
530 if (flags & B_NOCACHE) {
532 No page found. We should not read the page in if
533 1. the write starts on a page edge (ie, pageoffset == 0)
535 1. we will fill the page (ie, size == PAGESIZE), or
536 2. we are writing past eof
538 if ((uio->uio_rw == UIO_WRITE) &&
539 ((pageOffset == 0 && (size == PAGE_SIZE || fileBase >= avc->m.Length)))) {
540 struct vnode *vp = (struct vnode *)avc;
541 /* we're doing a write operation past eof; no need to read it */
544 ubc_page_zero(page, 0, PAGE_SIZE);
545 ubc_page_release(page, B_DONE);
548 /* page wasn't cached, read it in. */
552 bp = ubc_bufalloc(page, 1, PAGE_SIZE, 1, B_READ);
555 bp->b_vp = (struct vnode *)avc;
556 bp->b_blkno = btodb(pageBase);
557 ReleaseWriteLock(&avc->lock);
558 code = afs_ustrategy(bp, cred); /* do the I/O */
559 ObtainWriteLock(&avc->lock,164);
565 ubc_page_release(page, 0);
573 data = (char *)page->pg_addr; /* DUX 4.0D */
575 data = (char *)PHYS_TO_KSEG(page->pg_phys_addr); /* DUX 4.0E */
577 ReleaseWriteLock(&avc->lock); /* uiomove may page fault */
579 code = uiomove(data+pageOffset, tsize, uio);
580 ubc_unload(page, pageOffset, page_size);
581 if (uio->uio_rw == UIO_WRITE) {
584 /* Mark the page dirty and release it to avoid a deadlock
585 * in ubc_dirty_kluster when more than one process writes
586 * this page at the same time. */
587 toffset = page->pg_offset;
589 ubc_page_release(page, flags);
596 /* We released the page, so we can get a null page
597 * list if another thread calls the strategy routine.
599 pl = ubc_dirty_kluster(((struct vnode *)avc)->v_object,
600 NULL, toffset, 0, B_WANTED, FALSE, &kpcnt);
602 bp = ubc_bufalloc(pl, 1, PAGE_SIZE, 1, B_WRITE);
604 bp->b_vp = (struct vnode *)avc;
605 bp->b_blkno = btodb(pageBase);
607 code = afs_ustrategy(bp, cred); /* do the I/O */
612 ObtainWriteLock(&avc->lock,415);
618 ubc_page_release(page, flags);
621 ObtainWriteLock(&avc->lock,165);
623 * If reading at a chunk boundary, start prefetch of next chunk.
625 if (uio->uio_rw == UIO_READ
626 && (counter == 0 || AFS_CHUNKOFFSET(fileBase) == 0)) {
627 tdc = afs_FindDCache(avc, fileBase);
629 if (!(tdc->mflags & DFNextStarted))
630 afs_PrefetchChunk(avc, tdc, cred, &treq);
638 afs_FakeClose(avc, cred);
639 if (uio->uio_rw == UIO_WRITE && code == 0 && (avc->states & CDirty)) {
640 code = afs_DoPartialWrite(avc, &treq);
642 ReleaseWriteLock(&avc->lock);
643 afs_BozonUnlock(&avc->pvnLock, avc);
644 if (DO_FLUSH || (!newpage && (cnt < 10))) {
646 ubc_flush_dirty(((struct vnode *)avc)->v_object, flags);
650 ObtainSharedLock(&avc->lock, 409);
653 code = avc->vc_error;
656 /* This is required since we may still have dirty pages after the write.
657 * I could just let close do the right thing, but stat's before the close
658 * return the wrong length.
660 if (code == EDQUOT || code == ENOSPC) {
661 uio->uio_resid = save_resid;
662 UpgradeSToWLock(&avc->lock, 410);
663 osi_ReleaseVM(avc, cred);
664 ConvertWToSLock(&avc->lock);
666 ReleaseSharedLock(&avc->lock);
668 if (!code && (ioflag & IO_SYNC) && (uio->uio_rw == UIO_WRITE)
669 && !AFS_NFSXLATORREQ(cred)) {
670 code = afs_fsync(avc, 0, cred, 0);
673 code = afs_CheckCode(code, &treq, 36);
680 * Now for some bad news. Since we artificially hold on to vnodes by doing
681 * and extra VNHOLD in afs_NewVCache(), there is no way for us to know
682 * when we need to flush the pages when a program exits. Particularly
683 * if it closes the file after mapping it R/W.
687 mp_afs_mmap(avc, offset, map, addrp, len, prot, maxprot, flags, cred)
688 register struct vcache *avc;
698 struct vp_mmap_args args;
699 register struct vp_mmap_args *ap = &args;
700 struct vnode *vp = (struct vnode *)avc;
702 struct vrequest treq;
704 extern kern_return_t u_vp_create();
708 afs_InitReq(&treq, cred);
709 code = afs_VerifyVCache(avc, &treq);
711 code = afs_CheckCode(code, &treq, 37);
715 afs_BozonLock(&avc->pvnLock, avc);
716 osi_FlushPages(avc); /* ensure old pages are gone */
717 afs_BozonUnlock(&avc->pvnLock, avc);
718 ObtainWriteLock(&avc->lock,166);
719 avc->states |= CMAPPED;
720 ReleaseWriteLock(&avc->lock);
721 ap->a_offset = offset;
725 ap->a_maxprot = maxprot;
728 code = u_vp_create(map, vp->v_object, (vm_offset_t) ap);
730 code = afs_CheckCode(code, &treq, 38);
736 int mp_afs_getpage(vop, offset, len, protp, pl, plsz, mape, addr, rw, cred)
748 register afs_int32 code;
749 struct vrequest treq;
751 int i, pages = (len + PAGE_SIZE - 1) >> page_shift;
755 struct vcache *avc = (struct vcache *)vop->vu_vp;
757 /* first, obtain the proper lock for the VM system */
760 afs_InitReq(&treq, cred);
761 code = afs_VerifyVCache(avc, &treq);
764 code = afs_CheckCode(code, &treq, 39); /* failed to get it */
769 /* clean all dirty pages for this vnode */
771 ubc_flush_dirty(vop,0);
774 afs_BozonLock(&avc->pvnLock, avc);
775 ObtainWriteLock(&avc->lock,167);
776 afs_Trace4(afs_iclSetp, CM_TRACE_PAGEIN, ICL_TYPE_POINTER, avc,
777 ICL_TYPE_LONG, offset, ICL_TYPE_LONG, len,
778 ICL_TYPE_INT32, (int) rw);
779 for (i = 0; i < pages; i++) {
781 off = offset + PAGE_SIZE * i;
782 if (protp) protp[i] = 0;
784 ReleaseWriteLock(&avc->lock);
786 code = ubc_lookup(((struct vnode *)avc)->v_object, off,
787 PAGE_SIZE, PAGE_SIZE, pagep, &flags);
789 ObtainWriteLock(&avc->lock,168);
793 if(flags & B_NOCACHE) { /* if (page) */
794 if ((rw & B_WRITE) && (offset+len >= avc->m.Length)) {
795 struct vnode *vp = (struct vnode *)avc;
796 /* we're doing a write operation past eof; no need to read it */
798 ubc_page_zero(*pagep, 0, PAGE_SIZE);
799 ubc_page_release(*pagep, B_DONE);
802 /* page wasn't cached, read it in. */
806 bp = ubc_bufalloc(*pagep, 1, PAGE_SIZE, 1, B_READ);
809 bp->b_vp = (struct vnode *)avc;
810 bp->b_blkno = btodb(off);
811 ReleaseWriteLock(&avc->lock);
812 code = afs_ustrategy(bp, cred); /* do the I/O */
813 ObtainWriteLock(&avc->lock,169);
819 ubc_page_release(pl[i], 0);
825 if ((rw & B_READ) == 0) {
827 ubc_page_dirty(pl[i]);
830 if (protp && (flags & B_DIRTY) == 0) {
831 protp[i] = VM_PROT_WRITE;
836 pl[i] = VM_PAGE_NULL;
837 ReleaseWriteLock(&avc->lock);
838 afs_BozonUnlock(&avc->pvnLock, avc);
839 afs_Trace3(afs_iclSetp, CM_TRACE_PAGEINDONE, ICL_TYPE_INT32, code,
840 ICL_TYPE_POINTER, *pagep, ICL_TYPE_INT32, flags);
841 code = afs_CheckCode(code, &treq, 40);
847 int mp_afs_putpage(vop, pl, pcnt, flags, cred)
854 register afs_int32 code=0;
855 struct vcache *avc = (struct vcache *)vop->vu_vp;
856 struct vnode *vp = (struct vnode *)avc;
860 afs_Trace4(afs_iclSetp, CM_TRACE_PAGEOUT, ICL_TYPE_POINTER, avc,
861 ICL_TYPE_INT32, pcnt, ICL_TYPE_INT32, vp->v_flag,
862 ICL_TYPE_INT32, flags);
866 if (vp->v_flag & VXLOCK) {
868 for (i = 0; i < pcnt; i++) {
869 ubc_page_release(pl[i], B_DONE|B_DIRTY);
870 pl[i] = VM_PAGE_NULL;
879 /* first, obtain the proper lock for the VM system */
880 afs_BozonLock(&avc->pvnLock, avc);
881 ObtainWriteLock(&avc->lock,170);
882 for (i = 0; i < pcnt; i++) {
883 vm_page_t page = pl[i];
888 bp = ubc_bufalloc(page, 1, PAGE_SIZE, 1, B_WRITE);
891 bp->b_vp = (struct vnode *)avc;
892 bp->b_blkno = btodb(page->pg_offset);
893 ReleaseWriteLock(&avc->lock);
894 code = afs_ustrategy(bp, cred); /* do the I/O */
895 ObtainWriteLock(&avc->lock,171);
902 pl[i] = VM_PAGE_NULL;
906 ReleaseWriteLock(&avc->lock);
907 afs_BozonUnlock(&avc->pvnLock, avc);
908 afs_Trace2(afs_iclSetp, CM_TRACE_PAGEOUTDONE, ICL_TYPE_INT32, code,
909 ICL_TYPE_INT32, avc->m.Length);
915 int mp_afs_swap(avc, swapop, argp)
923 int mp_afs_syncdata(avc, flag, offset, length, cred)
930 /* NFS V3 makes this call, ignore it. We'll sync the data in afs_fsync. */
931 if (AFS_NFSXLATORREQ(cred))
937 /* a freelist of one */
938 struct buf *afs_bread_freebp = 0;
941 * Only rfs_read calls this, and it only looks at bp->b_un.b_addr.
942 * Thus we can use fake bufs (ie not from the real buffer pool).
944 mp_afs_bread(vp, lbn, bpp, cred)
950 int offset, fsbsize, error;
956 AFS_STATCNT(afs_bread);
957 fsbsize = vp->v_vfsp->vfs_bsize;
958 offset = lbn * fsbsize;
959 if (afs_bread_freebp) {
960 bp = afs_bread_freebp;
961 afs_bread_freebp = 0;
963 bp = (struct buf *) AFS_KALLOC(sizeof(*bp));
964 bp->b_un.b_addr = (caddr_t) AFS_KALLOC(fsbsize);
967 iov.iov_base = bp->b_un.b_addr;
968 iov.iov_len = fsbsize;
969 uio.afsio_iov = &iov;
970 uio.afsio_iovcnt = 1;
971 uio.afsio_seg = AFS_UIOSYS;
972 uio.afsio_offset = offset;
973 uio.afsio_resid = fsbsize;
975 error = afs_read((struct vcache *)vp, &uio, cred, lbn, bpp, 0);
977 afs_bread_freebp = bp;
982 afs_bread_freebp = bp;
984 *(struct buf **)&bp->b_vp = bp; /* mark as fake */
992 mp_afs_brelse(vp, bp)
997 AFS_STATCNT(afs_brelse);
998 if ((struct buf *)bp->b_vp != bp) { /* not fake */
1000 } else if (afs_bread_freebp) {
1001 AFS_KFREE(bp->b_un.b_addr, vp->v_vfsp->vfs_bsize);
1002 AFS_KFREE(bp, sizeof(*bp));
1004 afs_bread_freebp = bp;
1010 mp_afs_bmap(avc, abn, anvp, anbn)
1011 register struct vcache *avc;
1012 afs_int32 abn, *anbn;
1013 struct vcache **anvp;
1016 AFS_STATCNT(afs_bmap);
1020 *anbn = abn * (8192 / DEV_BSIZE); /* in 512 byte units */
1027 mp_afs_strategy (abp)
1028 register struct buf *abp;
1030 register afs_int32 code;
1033 AFS_STATCNT(afs_strategy);
1034 code = afs_osi_MapStrategy(afs_ustrategy, abp);
1040 mp_afs_refer(vm_ubc_object_t vop)
1046 mp_afs_release(vm_ubc_object_t vop)
1052 mp_afs_write_check(vm_ubc_object_t vop, vm_page_t pp)
1059 struct vfs_ubcops afs_ubcops = {
1060 mp_afs_refer, /* refer vnode */
1061 mp_afs_release, /* release vnode */
1062 mp_afs_getpage, /* get page */
1063 mp_afs_putpage, /* put page */
1064 mp_afs_write_check, /* check writablity */
1069 * Cover function for lookup name using OSF equivalent, namei()
1071 * Note, the result vnode (ni_vp) in the namei data structure is remains
1072 * locked after return.
1074 lookupname(namep, seg, follow, dvpp, cvpp)
1075 char *namep; /* path name */
1076 int seg; /* address space containing name */
1077 int follow; /* follow symbolic links */
1078 struct vnode **dvpp; /* result, containing parent vnode */
1079 struct vnode **cvpp; /* result, containing final component vnode */
1081 /* Should I use free-bee in u-area? */
1082 struct nameidata *ndp = &u.u_nd;
1085 ndp->ni_nameiop = ((follow) ? (LOOKUP|FOLLOW) : (LOOKUP));
1086 ndp->ni_segflg = seg;
1087 ndp->ni_dirp = namep;
1089 if (dvpp != (struct vnode **)0)
1090 *dvpp = ndp->ni_dvp;
1091 if (cvpp != (struct vnode **)0)