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 "../afs/param.h" /* Should be always first */
16 #include "../afs/sysincludes.h" /* Standard vendor system headers */
17 #include "../afs/afsincludes.h" /* Afs-based standard headers */
18 #include "../afs/afs_stats.h" /* statistics */
19 #include <vm/vm_mmap.h>
20 #include <vm/vm_ubc.h>
21 #include "../afs/afs_cbqueue.h"
22 #include "../afs/nfsclient.h"
23 #include "../afs/afs_osidnlc.h"
26 extern int afs_lookup(), afs_create(), afs_noop(), afs_open(), afs_close();
27 extern int afs_access(), afs_getattr(), afs_setattr(), afs_badop();
28 extern int afs_fsync(), afs_seek(), afs_remove(), afs_link(), afs_rename();
29 extern int afs_mkdir(), afs_rmdir(), afs_symlink(), afs_readdir();
30 extern int afs_readlink(), afs_lockctl();
31 extern int vn_pathconf_default(), seltrue();
33 int mp_afs_lookup(), mp_afs_create(), mp_afs_open();
34 int mp_afs_access(), mp_afs_getattr(), mp_afs_setattr(), mp_afs_ubcrdwr();
35 int mp_afs_ubcrdwr(), mp_afs_mmap();
36 int mp_afs_fsync(), mp_afs_seek(), mp_afs_remove(), mp_afs_link();
37 int mp_afs_rename(), mp_afs_mkdir(), mp_afs_rmdir(), mp_afs_symlink();
38 int mp_afs_readdir(), mp_afs_readlink(), mp_afs_abortop(), mp_afs_inactive();
39 int mp_afs_reclaim(), mp_afs_bmap(), mp_afs_strategy(), mp_afs_print();
40 int mp_afs_page_read(), mp_afs_page_write(), mp_afs_swap(), mp_afs_bread();
41 int mp_afs_brelse(), mp_afs_lockctl(), mp_afs_syncdata(), mp_afs_close();
45 struct vnodeops Afs_vnodeops = {
48 afs_noop, /* vn_mknod */
56 afs_badop, /* vn_ioctl */
57 seltrue, /* vn_select */
83 afs_noop, /* unLock */
84 afs_noop, /* get ext attrs */
85 afs_noop, /* set ext attrs */
86 afs_noop, /* del ext attrs */
89 struct vnodeops *afs_ops = &Afs_vnodeops;
91 /* vnode file operations, and our own */
93 extern int vn_write();
94 extern int vn_ioctl();
95 extern int vn_select();
96 extern int afs_closex();
98 struct fileops afs_fileops = {
106 mp_afs_lookup(adp, ndp)
108 struct nameidata *ndp;
112 code = afs_lookup(adp, ndp);
117 mp_afs_create(ndp, attrs)
118 struct nameidata *ndp;
123 code = afs_create(ndp, attrs);
128 mp_afs_open(avcp, aflags, acred)
129 struct vcache **avcp;
131 struct AFS_UCRED *acred;
135 code = afs_open(avcp, aflags, acred);
140 mp_afs_access(avc, amode, acred)
143 struct AFS_UCRED *acred;
147 code = afs_access(avc, amode, acred);
152 mp_afs_close(avc, flags, cred)
159 code = afs_close(avc, flags, cred);
164 mp_afs_getattr(avc, attrs, acred)
167 struct AFS_UCRED *acred;
171 code = afs_getattr(avc, attrs, acred);
176 mp_afs_setattr(avc, attrs, acred)
179 struct AFS_UCRED *acred;
183 code = afs_setattr(avc, attrs, acred);
188 mp_afs_fsync(avc, fflags, acred, waitfor)
191 struct AFS_UCRED *acred;
196 code = afs_fsync(avc, fflags, acred, waitfor);
202 struct nameidata *ndp;
206 code = afs_remove(ndp);
211 mp_afs_link(avc, ndp)
213 struct nameidata *ndp;
217 code = afs_link(avc, ndp);
222 mp_afs_rename(fndp, tndp)
223 struct nameidata *fndp, *tndp;
227 code = afs_rename(fndp, tndp);
232 mp_afs_mkdir(ndp, attrs)
233 struct nameidata *ndp;
238 code = afs_mkdir(ndp, attrs);
244 struct nameidata *ndp;
248 code = afs_rmdir(ndp);
253 mp_afs_symlink(ndp, attrs, atargetName)
254 struct nameidata *ndp;
256 register char *atargetName;
260 code = afs_symlink(ndp, attrs, atargetName);
265 mp_afs_readdir(avc, auio, acred, eofp)
268 struct AFS_UCRED *acred;
273 code = afs_readdir(avc, auio, acred, eofp);
278 mp_afs_readlink(avc, auio, acred)
281 struct AFS_UCRED *acred;
285 code = afs_readlink(avc, auio, acred);
290 mp_afs_lockctl(avc, af, flag, acred, clid, offset)
293 struct AFS_UCRED *acred;
300 code = afs_lockctl(avc, af, flag, acred, clid, offset);
310 code = afs_closex(afd);
315 mp_afs_seek(avc, oldoff, newoff, cred)
317 off_t oldoff, newoff;
320 if ((int) newoff < 0)
327 struct nameidata *ndp;
332 mp_afs_inactive(avc, acred)
333 register struct vcache *avc;
334 struct AFS_UCRED *acred;
337 afs_InactiveVCache(avc, acred);
354 mp_afs_page_read(avc, uio, acred)
360 struct vrequest treq;
363 error = afs_rdwr(avc, uio, UIO_READ, 0, acred);
364 afs_Trace3(afs_iclSetp, CM_TRACE_PAGE_READ, ICL_TYPE_POINTER, avc,
365 ICL_TYPE_INT32, error, ICL_TYPE_INT32, avc->states);
368 } else if ((avc->states & CWired) == 0) {
369 afs_InitReq(&treq, acred);
370 ObtainWriteLock(&avc->lock,161);
371 afs_Wire(avc, &treq);
372 ReleaseWriteLock(&avc->lock);
379 mp_afs_page_write(avc, uio, acred, pager, offset)
383 memory_object_t pager;
389 error = afs_rdwr(avc, uio, UIO_WRITE, 0, acred);
390 afs_Trace3(afs_iclSetp, CM_TRACE_PAGE_WRITE, ICL_TYPE_POINTER, avc,
391 ICL_TYPE_INT32, error, ICL_TYPE_INT32, avc->states);
401 mp_afs_ubcrdwr(avc, uio, ioflag, cred)
407 register afs_int32 code;
409 afs_int32 fileBase, size, cnt=0;
411 register afs_int32 tsize;
412 register afs_int32 pageOffset;
414 struct vrequest treq;
415 int rw = uio->uio_rw;
419 afs_int32 save_resid;
425 afs_InitReq(&treq, cred);
426 if (AFS_NFSXLATORREQ(cred) && rw == UIO_READ) {
427 if (!afs_AccessOK(avc, PRSFS_READ, &treq,
428 CHECK_MODE_BITS|CMB_ALLOW_EXEC_AS_READ)) {
433 afs_Trace4(afs_iclSetp, CM_TRACE_VMRW, ICL_TYPE_POINTER, avc,
434 ICL_TYPE_INT32, (rw==UIO_WRITE? 1 : 0),
435 ICL_TYPE_LONG, uio->uio_offset,
436 ICL_TYPE_LONG, uio->uio_resid);
437 code = afs_VerifyVCache(avc, &treq);
439 code = afs_CheckCode(code, &treq, 35);
443 if (vType(avc) != VREG) {
445 return EISDIR; /* can't read or write other things */
447 afs_BozonLock(&avc->pvnLock, avc);
448 osi_FlushPages(avc); /* hold bozon lock, but not basic vnode lock */
449 ObtainWriteLock(&avc->lock,162);
450 /* adjust parameters when appending files */
451 if ((ioflag & IO_APPEND) && uio->uio_rw == UIO_WRITE)
452 uio->uio_offset = avc->m.Length; /* write at EOF position */
453 if (uio->uio_rw == UIO_WRITE) {
454 avc->states |= CDirty;
458 * before starting any I/O, we must ensure that the file is big enough
459 * to hold the results (since afs_putpage will be called to force
462 size = uio->afsio_resid + uio->afsio_offset; /* new file size */
463 if (size > avc->m.Length) avc->m.Length = size; /* file grew */
464 avc->m.Date = osi_Time(); /* Set file date (for ranlib) */
465 if (uio->afsio_resid > PAGE_SIZE)
466 cnt = uio->afsio_resid / PAGE_SIZE;
467 save_resid = uio->afsio_resid;
472 * compute the amount of data to move into this block,
473 * based on uio->afsio_resid.
475 size = uio->afsio_resid; /* transfer size */
476 fileBase = uio->afsio_offset; /* start file position */
477 pageBase = fileBase & ~(PAGE_SIZE-1); /* file position of the page */
478 pageOffset = fileBase & (PAGE_SIZE-1); /* start offset within page */
479 tsize = PAGE_SIZE-pageOffset; /* amount left in this page */
481 * we'll read tsize bytes,
482 * but first must make sure tsize isn't too big
484 if (tsize > size) tsize = size; /* don't read past end of request */
485 eof = 0; /* flag telling us if we hit the EOF on the read */
486 if (uio->uio_rw == UIO_READ) { /* we're doing a read operation */
487 /* don't read past EOF */
488 if (tsize + fileBase > avc->m.Length) {
489 tsize = avc->m.Length - fileBase;
490 eof = 1; /* we did hit the EOF */
491 if (tsize < 0) tsize = 0; /* better safe than sorry */
494 if (tsize <= 0) break; /* nothing to transfer, we're done */
496 /* Purge dirty chunks of file if there are too many dirty chunks.
497 * Inside the write loop, we only do this at a chunk boundary.
498 * Clean up partial chunk if necessary at end of loop.
500 if (uio->uio_rw == UIO_WRITE && counter > 0
501 && AFS_CHUNKOFFSET(fileBase) == 0) {
502 code = afs_DoPartialWrite(avc, &treq);
503 avc->states |= CDirty;
511 ReleaseWriteLock(&avc->lock);
513 code = ubc_lookup(((struct vnode *)avc)->v_object, pageBase,
514 PAGE_SIZE, PAGE_SIZE, &page, &flags);
516 ObtainWriteLock(&avc->lock,163);
521 if (flags & B_NOCACHE) {
523 No page found. We should not read the page in if
524 1. the write starts on a page edge (ie, pageoffset == 0)
526 1. we will fill the page (ie, size == PAGESIZE), or
527 2. we are writing past eof
529 if ((uio->uio_rw == UIO_WRITE) &&
530 ((pageOffset == 0 && (size == PAGE_SIZE || fileBase >= avc->m.Length)))) {
531 struct vnode *vp = (struct vnode *)avc;
532 /* we're doing a write operation past eof; no need to read it */
535 ubc_page_zero(page, 0, PAGE_SIZE);
536 ubc_page_release(page, B_DONE);
539 /* page wasn't cached, read it in. */
543 bp = ubc_bufalloc(page, 1, PAGE_SIZE, 1, B_READ);
546 bp->b_vp = (struct vnode *)avc;
547 bp->b_blkno = btodb(pageBase);
548 ReleaseWriteLock(&avc->lock);
549 code = afs_ustrategy(bp, cred); /* do the I/O */
550 ObtainWriteLock(&avc->lock,164);
556 ubc_page_release(page, 0);
564 data = (char *)page->pg_addr; /* DUX 4.0D */
566 data = (char *)PHYS_TO_KSEG(page->pg_phys_addr); /* DUX 4.0E */
568 ReleaseWriteLock(&avc->lock); /* uiomove may page fault */
570 code = uiomove(data+pageOffset, tsize, uio);
571 ubc_unload(page, pageOffset, page_size);
572 if (uio->uio_rw == UIO_WRITE) {
575 /* Mark the page dirty and release it to avoid a deadlock
576 * in ubc_dirty_kluster when more than one process writes
577 * this page at the same time. */
578 toffset = page->pg_offset;
580 ubc_page_release(page, flags);
587 /* We released the page, so we can get a null page
588 * list if another thread calls the strategy routine.
590 pl = ubc_dirty_kluster(((struct vnode *)avc)->v_object,
591 NULL, toffset, 0, B_WANTED, FALSE, &kpcnt);
593 bp = ubc_bufalloc(pl, 1, PAGE_SIZE, 1, B_WRITE);
595 bp->b_vp = (struct vnode *)avc;
596 bp->b_blkno = btodb(pageBase);
598 code = afs_ustrategy(bp, cred); /* do the I/O */
603 ObtainWriteLock(&avc->lock,415);
609 ubc_page_release(page, flags);
612 ObtainWriteLock(&avc->lock,165);
614 * If reading at a chunk boundary, start prefetch of next chunk.
616 if (uio->uio_rw == UIO_READ
617 && (counter == 0 || AFS_CHUNKOFFSET(fileBase) == 0)) {
618 tdc = afs_FindDCache(avc, fileBase);
620 if (!(tdc->flags & DFNextStarted))
621 afs_PrefetchChunk(avc, tdc, cred, &treq);
629 afs_FakeClose(avc, cred);
630 if (uio->uio_rw == UIO_WRITE && code == 0 && (avc->states & CDirty)) {
631 code = afs_DoPartialWrite(avc, &treq);
633 ReleaseWriteLock(&avc->lock);
634 afs_BozonUnlock(&avc->pvnLock, avc);
635 if (DO_FLUSH || (!newpage && (cnt < 10))) {
637 ubc_flush_dirty(((struct vnode *)avc)->v_object, flags);
641 ObtainSharedLock(&avc->lock, 409);
644 code = avc->vc_error;
647 /* This is required since we may still have dirty pages after the write.
648 * I could just let close do the right thing, but stat's before the close
649 * return the wrong length.
651 if (code == EDQUOT || code == ENOSPC) {
652 uio->uio_resid = save_resid;
653 UpgradeSToWLock(&avc->lock, 410);
654 osi_ReleaseVM(avc, cred);
655 ConvertWToSLock(&avc->lock);
657 ReleaseSharedLock(&avc->lock);
659 if (!code && (ioflag & IO_SYNC) && (uio->uio_rw == UIO_WRITE)
660 && !AFS_NFSXLATORREQ(cred)) {
661 code = afs_fsync(avc, 0, cred, 0);
664 code = afs_CheckCode(code, &treq, 36);
671 * Now for some bad news. Since we artificially hold on to vnodes by doing
672 * and extra VNHOLD in afs_NewVCache(), there is no way for us to know
673 * when we need to flush the pages when a program exits. Particularly
674 * if it closes the file after mapping it R/W.
678 mp_afs_mmap(avc, offset, map, addrp, len, prot, maxprot, flags, cred)
679 register struct vcache *avc;
689 struct vp_mmap_args args;
690 register struct vp_mmap_args *ap = &args;
691 struct vnode *vp = (struct vnode *)avc;
693 struct vrequest treq;
695 extern kern_return_t u_vp_create();
699 afs_InitReq(&treq, cred);
700 code = afs_VerifyVCache(avc, &treq);
702 code = afs_CheckCode(code, &treq, 37);
706 afs_BozonLock(&avc->pvnLock, avc);
707 osi_FlushPages(avc); /* ensure old pages are gone */
708 afs_BozonUnlock(&avc->pvnLock, avc);
709 ObtainWriteLock(&avc->lock,166);
710 avc->states |= CMAPPED;
711 ReleaseWriteLock(&avc->lock);
712 ap->a_offset = offset;
716 ap->a_maxprot = maxprot;
719 code = u_vp_create(map, vp->v_object, (vm_offset_t) ap);
721 code = afs_CheckCode(code, &treq, 38);
727 int mp_afs_getpage(vop, offset, len, protp, pl, plsz, mape, addr, rw, cred)
739 register afs_int32 code;
740 struct vrequest treq;
742 int i, pages = (len + PAGE_SIZE - 1) >> page_shift;
746 struct vcache *avc = (struct vcache *)vop->vu_vp;
748 /* first, obtain the proper lock for the VM system */
751 afs_InitReq(&treq, cred);
752 code = afs_VerifyVCache(avc, &treq);
755 code = afs_CheckCode(code, &treq, 39); /* failed to get it */
760 /* clean all dirty pages for this vnode */
762 ubc_flush_dirty(vop,0);
765 afs_BozonLock(&avc->pvnLock, avc);
766 ObtainWriteLock(&avc->lock,167);
767 afs_Trace4(afs_iclSetp, CM_TRACE_PAGEIN, ICL_TYPE_POINTER, avc,
768 ICL_TYPE_LONG, offset, ICL_TYPE_LONG, len,
769 ICL_TYPE_INT32, (int) rw);
770 for (i = 0; i < pages; i++) {
772 off = offset + PAGE_SIZE * i;
773 if (protp) protp[i] = 0;
775 ReleaseWriteLock(&avc->lock);
777 code = ubc_lookup(((struct vnode *)avc)->v_object, off,
778 PAGE_SIZE, PAGE_SIZE, pagep, &flags);
780 ObtainWriteLock(&avc->lock,168);
784 if(flags & B_NOCACHE) { /* if (page) */
785 if ((rw & B_WRITE) && (offset+len >= avc->m.Length)) {
786 struct vnode *vp = (struct vnode *)avc;
787 /* we're doing a write operation past eof; no need to read it */
789 ubc_page_zero(*pagep, 0, PAGE_SIZE);
790 ubc_page_release(*pagep, B_DONE);
793 /* page wasn't cached, read it in. */
797 bp = ubc_bufalloc(*pagep, 1, PAGE_SIZE, 1, B_READ);
800 bp->b_vp = (struct vnode *)avc;
801 bp->b_blkno = btodb(off);
802 ReleaseWriteLock(&avc->lock);
803 code = afs_ustrategy(bp, cred); /* do the I/O */
804 ObtainWriteLock(&avc->lock,169);
810 ubc_page_release(pl[i], 0);
816 if ((rw & B_READ) == 0) {
818 ubc_page_dirty(pl[i]);
821 if (protp && (flags & B_DIRTY) == 0) {
822 protp[i] = VM_PROT_WRITE;
827 pl[i] = VM_PAGE_NULL;
828 ReleaseWriteLock(&avc->lock);
829 afs_BozonUnlock(&avc->pvnLock, avc);
830 afs_Trace3(afs_iclSetp, CM_TRACE_PAGEINDONE, ICL_TYPE_INT32, code,
831 ICL_TYPE_POINTER, *pagep, ICL_TYPE_INT32, flags);
832 code = afs_CheckCode(code, &treq, 40);
838 int mp_afs_putpage(vop, pl, pcnt, flags, cred)
845 register afs_int32 code=0;
846 struct vcache *avc = (struct vcache *)vop->vu_vp;
847 struct vnode *vp = (struct vnode *)avc;
851 afs_Trace4(afs_iclSetp, CM_TRACE_PAGEOUT, ICL_TYPE_POINTER, avc,
852 ICL_TYPE_INT32, pcnt, ICL_TYPE_INT32, vp->v_flag,
853 ICL_TYPE_INT32, flags);
857 if (vp->v_flag & VXLOCK) {
859 for (i = 0; i < pcnt; i++) {
860 ubc_page_release(pl[i], B_DONE|B_DIRTY);
861 pl[i] = VM_PAGE_NULL;
870 /* first, obtain the proper lock for the VM system */
871 afs_BozonLock(&avc->pvnLock, avc);
872 ObtainWriteLock(&avc->lock,170);
873 for (i = 0; i < pcnt; i++) {
874 vm_page_t page = pl[i];
879 bp = ubc_bufalloc(page, 1, PAGE_SIZE, 1, B_WRITE);
882 bp->b_vp = (struct vnode *)avc;
883 bp->b_blkno = btodb(page->pg_offset);
884 ReleaseWriteLock(&avc->lock);
885 code = afs_ustrategy(bp, cred); /* do the I/O */
886 ObtainWriteLock(&avc->lock,171);
893 pl[i] = VM_PAGE_NULL;
897 ReleaseWriteLock(&avc->lock);
898 afs_BozonUnlock(&avc->pvnLock, avc);
899 afs_Trace2(afs_iclSetp, CM_TRACE_PAGEOUTDONE, ICL_TYPE_INT32, code,
900 ICL_TYPE_INT32, avc->m.Length);
906 int mp_afs_swap(avc, swapop, argp)
914 int mp_afs_syncdata(avc, flag, offset, length, cred)
921 /* NFS V3 makes this call, ignore it. We'll sync the data in afs_fsync. */
922 if (AFS_NFSXLATORREQ(cred))
928 /* a freelist of one */
929 struct buf *afs_bread_freebp = 0;
932 * Only rfs_read calls this, and it only looks at bp->b_un.b_addr.
933 * Thus we can use fake bufs (ie not from the real buffer pool).
935 mp_afs_bread(vp, lbn, bpp, cred)
941 int offset, fsbsize, error;
947 AFS_STATCNT(afs_bread);
948 fsbsize = vp->v_vfsp->vfs_bsize;
949 offset = lbn * fsbsize;
950 if (afs_bread_freebp) {
951 bp = afs_bread_freebp;
952 afs_bread_freebp = 0;
954 bp = (struct buf *) AFS_KALLOC(sizeof(*bp));
955 bp->b_un.b_addr = (caddr_t) AFS_KALLOC(fsbsize);
958 iov.iov_base = bp->b_un.b_addr;
959 iov.iov_len = fsbsize;
960 uio.afsio_iov = &iov;
961 uio.afsio_iovcnt = 1;
962 uio.afsio_seg = AFS_UIOSYS;
963 uio.afsio_offset = offset;
964 uio.afsio_resid = fsbsize;
966 error = afs_read((struct vcache *)vp, &uio, cred, lbn, bpp, 0);
968 afs_bread_freebp = bp;
973 afs_bread_freebp = bp;
975 *(struct buf **)&bp->b_vp = bp; /* mark as fake */
983 mp_afs_brelse(vp, bp)
988 AFS_STATCNT(afs_brelse);
989 if ((struct buf *)bp->b_vp != bp) { /* not fake */
991 } else if (afs_bread_freebp) {
992 AFS_KFREE(bp->b_un.b_addr, vp->v_vfsp->vfs_bsize);
993 AFS_KFREE(bp, sizeof(*bp));
995 afs_bread_freebp = bp;
1001 mp_afs_bmap(avc, abn, anvp, anbn)
1002 register struct vcache *avc;
1003 afs_int32 abn, *anbn;
1004 struct vcache **anvp;
1007 AFS_STATCNT(afs_bmap);
1011 *anbn = abn * (8192 / DEV_BSIZE); /* in 512 byte units */
1018 mp_afs_strategy (abp)
1019 register struct buf *abp;
1021 register afs_int32 code;
1024 AFS_STATCNT(afs_strategy);
1025 code = afs_osi_MapStrategy(afs_ustrategy, abp);
1031 mp_afs_refer(vm_ubc_object_t vop)
1037 mp_afs_release(vm_ubc_object_t vop)
1043 mp_afs_write_check(vm_ubc_object_t vop, vm_page_t pp)
1050 struct vfs_ubcops afs_ubcops = {
1051 mp_afs_refer, /* refer vnode */
1052 mp_afs_release, /* release vnode */
1053 mp_afs_getpage, /* get page */
1054 mp_afs_putpage, /* put page */
1055 mp_afs_write_check, /* check writablity */
1060 * Cover function for lookup name using OSF equivalent, namei()
1062 * Note, the result vnode (ni_vp) in the namei data structure is remains
1063 * locked after return.
1065 lookupname(namep, seg, follow, dvpp, cvpp)
1066 char *namep; /* path name */
1067 int seg; /* address space containing name */
1068 int follow; /* follow symbolic links */
1069 struct vnode **dvpp; /* result, containing parent vnode */
1070 struct vnode **cvpp; /* result, containing final component vnode */
1072 /* Should I use free-bee in u-area? */
1073 struct nameidata *ndp = &u.u_nd;
1076 ndp->ni_nameiop = ((follow) ? (LOOKUP|FOLLOW) : (LOOKUP));
1077 ndp->ni_segflg = seg;
1078 ndp->ni_dirp = namep;
1080 if (dvpp != (struct vnode **)0)
1081 *dvpp = ndp->ni_dvp;
1082 if (cvpp != (struct vnode **)0)