-/* Copyright (C) 1995, 1989, 1998 Transarc Corporation - All rights reserved */
/*
- * For copyright information, see IPL which you accepted in order to
- * download this software.
- *
+ * Copyright 2000, International Business Machines Corporation and others.
+ * All Rights Reserved.
+ *
+ * This software has been released under the terms of the IBM Public
+ * License. For details, see the LICENSE file in the top-level source
+ * directory or online at http://www.openafs.org/dl/license10.html
*/
/*
- * afs_vnop_read.c
- *
* Implements:
* afs_MemRead
* afs_PrefetchChunk
- * afs_UFSReadFast
* afs_UFSRead
*
*/
-#include "../afs/param.h" /* Should be always first */
-#include "../afs/sysincludes.h" /* Standard vendor system headers */
-#include "../afs/afsincludes.h" /* Afs-based standard headers */
-#include "../afs/afs_stats.h" /* statistics */
-#include "../afs/afs_cbqueue.h"
-#include "../afs/nfsclient.h"
-#include "../afs/afs_osidnlc.h"
+#include <afsconfig.h>
+#include "afs/param.h"
-extern char afs_zeros[AFS_ZEROS];
+#include "afs/sysincludes.h" /* Standard vendor system headers */
+#include "afsincludes.h" /* Afs-based standard headers */
+#include "afs/afs_stats.h" /* statistics */
+#include "afs/afs_cbqueue.h"
+#include "afs/nfsclient.h"
+#include "afs/afs_osidnlc.h"
+#include "afs/afs_osi.h"
-afs_int32 maxIHint;
-afs_int32 nihints; /* # of above actually in-use */
-afs_int32 usedihint;
+extern char afs_zeros[AFS_ZEROS];
/* Imported variables */
extern afs_rwlock_t afs_xdcache;
extern unsigned char *afs_indexFlags;
-extern afs_hyper_t *afs_indexTimes; /* Dcache entry Access times */
-extern afs_hyper_t afs_indexCounter; /* Fake time for marking index */
+extern afs_hyper_t *afs_indexTimes; /* Dcache entry Access times */
+extern afs_hyper_t afs_indexCounter; /* Fake time for marking index */
/* Forward declarations */
void afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
- struct AFS_UCRED *acred, struct vrequest *areq);
-
-afs_MemRead(avc, auio, acred, albn, abpp, noLock)
- register struct vcache *avc;
- struct uio *auio;
- struct AFS_UCRED *acred;
- daddr_t albn;
- int noLock;
- struct buf **abpp; {
- afs_int32 totalLength;
- afs_int32 transferLength;
- afs_int32 filePos;
- struct dcache *tdc=0;
- afs_int32 offset, len, error, trybusy=1;
- struct uio tuio;
- struct iovec *tvec;
- char *tfile;
+ afs_ucred_t *acred, struct vrequest *areq);
+
+int
+afs_read(struct vcache *avc, struct uio *auio, afs_ucred_t *acred,
+ int noLock)
+{
+ afs_size_t totalLength;
+ afs_size_t transferLength;
+ afs_size_t filePos;
+ afs_size_t offset, len, tlen;
+ afs_int32 trimlen;
+ struct dcache *tdc = 0;
+ afs_int32 error, trybusy = 1;
+ struct uio *tuiop = NULL;
afs_int32 code;
struct vrequest treq;
- AFS_STATCNT(afs_MemRead);
+ AFS_STATCNT(afs_read);
+
if (avc->vc_error)
return EIO;
+ AFS_DISCON_LOCK();
+
/* check that we have the latest status info in the vnode cache */
- if (code = afs_InitReq(&treq, acred)) return code;
+ if ((code = afs_InitReq(&treq, acred)))
+ goto out;
+
if (!noLock) {
+ if (!avc)
+ osi_Panic("null avc in afs_GenericRead");
+
code = afs_VerifyVCache(avc, &treq);
if (code) {
- code = afs_CheckCode(code, &treq, 8); /* failed to get it */
- return code;
+ code = afs_CheckCode(code, &treq, 8); /* failed to get it */
+ goto out;
}
}
-
#ifndef AFS_VM_RDWR_ENV
if (AFS_NFSXLATORREQ(acred)) {
- if (!afs_AccessOK(avc, PRSFS_READ, &treq,
- CHECK_MODE_BITS|CMB_ALLOW_EXEC_AS_READ)) {
- return afs_CheckCode(EACCES, &treq, 9);
+ if (!afs_AccessOK
+ (avc, PRSFS_READ, &treq,
+ CHECK_MODE_BITS | CMB_ALLOW_EXEC_AS_READ)) {
+ code = afs_CheckCode(EACCES, &treq, 9);
+ goto out;
}
}
#endif
- tvec = (struct iovec *) osi_AllocSmallSpace(sizeof(struct iovec));
- totalLength = auio->afsio_resid;
- filePos = auio->afsio_offset;
- afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
- ICL_TYPE_INT32, filePos, ICL_TYPE_INT32, totalLength,
- ICL_TYPE_INT32, avc->m.Length);
+ totalLength = AFS_UIO_RESID(auio);
+ filePos = AFS_UIO_OFFSET(auio);
+ afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(filePos), ICL_TYPE_INT32,
+ totalLength, ICL_TYPE_OFFSET,
+ ICL_HANDLE_OFFSET(avc->f.m.Length));
error = 0;
transferLength = 0;
if (!noLock)
ObtainReadLock(&avc->lock);
#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
- hset(avc->flushDV, avc->m.DataVersion);
+ hset(avc->flushDV, avc->f.m.DataVersion);
}
#endif
- while (totalLength > 0) {
+
+ /*
+ * Locks held:
+ * avc->lock(R)
+ */
+ if (filePos >= avc->f.m.Length) {
+ if (len > AFS_ZEROS)
+ len = sizeof(afs_zeros); /* and in 0 buffer */
+ len = 0;
+ trimlen = len;
+ tuiop = afsio_partialcopy(auio, trimlen);
+ AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
+ }
+
+ while (avc->f.m.Length > 0 && totalLength > 0) {
/* read all of the cached info */
- if (filePos >= avc->m.Length) break; /* all done */
+ if (filePos >= avc->f.m.Length)
+ break; /* all done */
if (noLock) {
- if (tdc) afs_PutDCache(tdc);
+ if (tdc) {
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc);
+ }
tdc = afs_FindDCache(avc, filePos);
if (tdc) {
+ ObtainReadLock(&tdc->lock);
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
- len = tdc->f.chunkBytes - offset;
+ len = tdc->validPos - filePos;
}
} else {
/* a tricky question: does the presence of the DFFetching flag
- mean that we're fetching the latest version of the file? No.
- The server could update the file as soon as the fetch responsible
- for the setting of the DFFetching flag completes.
-
- However, the presence of the DFFetching flag (visible under a
- read lock since it is set and cleared only under a write lock)
- means that we're fetching as good a version as was known to this
- client at the time of the last call to afs_VerifyVCache, since
- the latter updates the stat cache's m.DataVersion field under a
- write lock, and from the time that the DFFetching flag goes on
- (before the fetch starts), to the time it goes off (after the
- fetch completes), afs_GetDCache keeps at least a read lock
- (actually it keeps an S lock) on the cache entry.
-
- This means that if the DFFetching flag is set, we can use that
- data for any reads that must come from the current version of
- the file (current == m.DataVersion).
-
- Another way of looking at this same point is this: if we're
- fetching some data and then try do an afs_VerifyVCache, the
- VerifyVCache operation will not complete until after the
- DFFetching flag is turned off and the dcache entry's f.versionNo
- field is updated.
-
- Note, by the way, that if DFFetching is set,
- m.DataVersion > f.versionNo (the latter is not updated until
- after the fetch completes).
+ * mean that we're fetching the latest version of the file? No.
+ * The server could update the file as soon as the fetch responsible
+ * for the setting of the DFFetching flag completes.
+ *
+ * However, the presence of the DFFetching flag (visible under
+ * a dcache read lock since it is set and cleared only under a
+ * dcache write lock) means that we're fetching as good a version
+ * as was known to this client at the time of the last call to
+ * afs_VerifyVCache, since the latter updates the stat cache's
+ * m.DataVersion field under a vcache write lock, and from the
+ * time that the DFFetching flag goes on in afs_GetDCache (before
+ * the fetch starts), to the time it goes off (after the fetch
+ * completes), afs_GetDCache keeps at least a read lock on the
+ * vcache entry.
+ *
+ * This means that if the DFFetching flag is set, we can use that
+ * data for any reads that must come from the current version of
+ * the file (current == m.DataVersion).
+ *
+ * Another way of looking at this same point is this: if we're
+ * fetching some data and then try do an afs_VerifyVCache, the
+ * VerifyVCache operation will not complete until after the
+ * DFFetching flag is turned off and the dcache entry's f.versionNo
+ * field is updated.
+ *
+ * Note, by the way, that if DFFetching is set,
+ * m.DataVersion > f.versionNo (the latter is not updated until
+ * after the fetch completes).
*/
- if (tdc) afs_PutDCache(tdc); /* before reusing tdc */
+ if (tdc) {
+ ReleaseReadLock(&tdc->lock);
+ afs_PutDCache(tdc); /* before reusing tdc */
+ }
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
+ if (!tdc) {
+ error = ENETDOWN;
+ break;
+ }
+
+ ObtainReadLock(&tdc->lock);
/* now, first try to start transfer, if we'll need the data. If
* data already coming, we don't need to do this, obviously. Type
* 2 requests never return a null dcache entry, btw.
*/
- if (!(tdc->flags & DFFetching)
- && !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
+ if (!(tdc->dflags & DFFetching)
+ && !hsame(avc->f.m.DataVersion, tdc->f.versionNo)) {
/* have cache entry, it is not coming in now,
* and we'll need new data */
-tagain:
+ tagain:
if (trybusy && !afs_BBusy()) {
struct brequest *bp;
/* daemon is not busy */
- if (!(tdc->flags & DFFetchReq)) {
+ ObtainSharedLock(&tdc->mflock, 665);
+ if (!(tdc->mflags & DFFetchReq)) {
/* start the daemon (may already be running, however) */
- tdc->flags |= DFFetchReq;
+ UpgradeSToWLock(&tdc->mflock, 666);
+ tdc->mflags |= DFFetchReq;
bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
- (long)filePos, (long) tdc, 0L, 0L);
+ (afs_size_t) filePos, (afs_size_t) 0,
+ tdc, NULL, NULL);
if (!bp) {
- tdc->flags &= ~DFFetchReq;
+ /* Bkg table full; retry deadlocks */
+ tdc->mflags &= ~DFFetchReq;
trybusy = 0; /* Avoid bkg daemon since they're too busy */
+ ReleaseWriteLock(&tdc->mflock);
goto tagain;
}
+ ConvertWToSLock(&tdc->mflock);
/* don't use bp pointer! */
}
- while (tdc->flags & DFFetchReq) {
+ code = 0;
+ ConvertSToRLock(&tdc->mflock);
+ while (!code && tdc->mflags & DFFetchReq) {
+ afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT,
+ ICL_TYPE_STRING, __FILE__, ICL_TYPE_INT32,
+ __LINE__, ICL_TYPE_POINTER, tdc,
+ ICL_TYPE_INT32, tdc->dflags);
/* don't need waiting flag on this one */
+ ReleaseReadLock(&tdc->mflock);
+ ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
- afs_osi_Sleep(&tdc->validPos);
+ code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
+ ObtainReadLock(&tdc->lock);
+ ObtainReadLock(&tdc->mflock);
+ }
+ ReleaseReadLock(&tdc->mflock);
+ if (code) {
+ error = code;
+ break;
}
}
}
/* now data may have started flowing in (if DFFetching is on). If
- * data is now streaming in, then wait for some interesting stuff. */
- while ((tdc->flags & DFFetching) && tdc->validPos <= filePos) {
- /* too early: wait for DFFetching flag to vanish, or data to appear */
- tdc->flags |= DFWaiting;
+ * data is now streaming in, then wait for some interesting stuff.
+ */
+ code = 0;
+ while (!code && (tdc->dflags & DFFetching)
+ && tdc->validPos <= filePos) {
+ /* too early: wait for DFFetching flag to vanish,
+ * or data to appear */
+ afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
+ __FILE__, ICL_TYPE_INT32, __LINE__,
+ ICL_TYPE_POINTER, tdc, ICL_TYPE_INT32,
+ tdc->dflags);
+ ReleaseReadLock(&tdc->lock);
ReleaseReadLock(&avc->lock);
- afs_osi_Sleep(&tdc->validPos);
+ code = afs_osi_SleepSig(&tdc->validPos);
ObtainReadLock(&avc->lock);
+ ObtainReadLock(&tdc->lock);
}
- /* fetching flag gone, data is here, or we never tried (BBusy for instance) */
- if (tdc->flags & DFFetching) {
- /* still fetching, some new data is here: compute length and offset */
+ if (code) {
+ error = code;
+ break;
+ }
+ /* fetching flag gone, data is here, or we never tried
+ * (BBusy for instance) */
+ if (tdc->dflags & DFFetching) {
+ /* still fetching, some new data is here:
+ * compute length and offset */
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
len = tdc->validPos - filePos;
- }
- else {
- /* no longer fetching, verify data version (avoid new GetDCache call) */
- if (hsame(avc->m.DataVersion, tdc->f.versionNo)) {
+ } else {
+ /* no longer fetching, verify data version
+ * (avoid new GetDCache call) */
+ if (hsame(avc->f.m.DataVersion, tdc->f.versionNo)
+ && ((len = tdc->validPos - filePos) > 0)) {
offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
- len = tdc->f.chunkBytes - offset;
- }
- else {
+ } else {
/* don't have current data, so get it below */
+ afs_Trace3(afs_iclSetp, CM_TRACE_VERSIONNO,
+ ICL_TYPE_INT64, ICL_HANDLE_OFFSET(filePos),
+ ICL_TYPE_HYPER, &avc->f.m.DataVersion,
+ ICL_TYPE_HYPER, &tdc->f.versionNo);
+ ReleaseReadLock(&tdc->lock);
afs_PutDCache(tdc);
- tdc = (struct dcache *) 0;
+ tdc = NULL;
}
}
if (!tdc) {
+ /* If we get, it was not possible to start the
+ * background daemon. With flag == 1 afs_GetDCache
+ * does the FetchData rpc synchronously.
+ */
ReleaseReadLock(&avc->lock);
tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
ObtainReadLock(&avc->lock);
+ if (tdc)
+ ObtainReadLock(&tdc->lock);
}
}
+ afs_Trace3(afs_iclSetp, CM_TRACE_VNODEREAD, ICL_TYPE_POINTER, tdc,
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(offset),
+ ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(len));
if (!tdc) {
error = EIO;
break;
}
- if (len > totalLength) len = totalLength; /* will read len bytes */
- if (len <= 0) { /* shouldn't get here if DFFetching is on */
+
+ /*
+ * Locks held:
+ * avc->lock(R)
+ * tdc->lock(R)
+ */
+
+ if (len > totalLength)
+ len = totalLength; /* will read len bytes */
+ if (len <= 0) { /* shouldn't get here if DFFetching is on */
/* read past the end of a chunk, may not be at next chunk yet, and yet
- also not at eof, so may have to supply fake zeros */
- len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
- if (len > totalLength) len = totalLength; /* and still within xfr request */
- code = avc->m.Length - offset; /* and still within file */
- if (len > code) len = code;
- if (len > AFS_ZEROS) len = sizeof(afs_zeros); /* and in 0 buffer */
- afsio_copy(auio, &tuio, tvec);
- afsio_trim(&tuio, len);
- AFS_UIOMOVE(afs_zeros, len, UIO_READ, &tuio, code);
+ * also not at eof, so may have to supply fake zeros */
+ len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
+ if (len > totalLength)
+ len = totalLength; /* and still within xfr request */
+ tlen = avc->f.m.Length - offset; /* and still within file */
+ if (len > tlen)
+ len = tlen;
+ if (len > AFS_ZEROS)
+ len = sizeof(afs_zeros); /* and in 0 buffer */
+ trimlen = len;
+ tuiop = afsio_partialcopy(auio, trimlen);
+ AFS_UIOMOVE(afs_zeros, trimlen, UIO_READ, tuiop, code);
if (code) {
error = code;
break;
}
- }
- else {
- /* get the data from the mem cache */
+ } else {
+ /* get the data from the cache */
/* mung uio structure to be right for this transfer */
- afsio_copy(auio, &tuio, tvec);
- afsio_trim(&tuio, len);
- tuio.afsio_offset = offset;
+ trimlen = len;
+ tuiop = afsio_partialcopy(auio, trimlen);
+ AFS_UIO_SETOFFSET(tuiop, offset);
- code = afs_MemReadUIO(tdc->f.inode, &tuio);
+ code = (*(afs_cacheType->vreadUIO))(&tdc->f.inode, tuiop);
if (code) {
error = code;
}
}
/* otherwise we've read some, fixup length, etc and continue with next seg */
- len = len - tuio.afsio_resid; /* compute amount really transferred */
- afsio_skip(auio, len); /* update input uio structure */
+ len = len - AFS_UIO_RESID(tuiop); /* compute amount really transferred */
+ trimlen = len;
+ afsio_skip(auio, trimlen); /* update input uio structure */
totalLength -= len;
transferLength += len;
filePos += len;
- if (len <= 0) break; /* surprise eof */
- } /* the whole while loop */
+ if (len <= 0)
+ break; /* surprise eof */
+ if (tuiop) {
+ afsio_free(tuiop);
+ tuiop = NULL;
+ }
+ } /* the whole while loop */
+
+ /*
+ * Locks held:
+ * avc->lock(R)
+ * tdc->lock(R) if tdc
+ */
/* if we make it here with tdc non-zero, then it is the last chunk we
* dealt with, and we have to release it when we're done. We hold on
* to it in case we need to do a prefetch.
*/
if (tdc) {
-#ifndef AFS_VM_RDWR_ENV
+ ReleaseReadLock(&tdc->lock);
+#if !defined(AFS_VM_RDWR_ENV)
/* try to queue prefetch, if needed */
- if (!(tdc->flags & DFNextStarted) && !noLock) {
- afs_PrefetchChunk(avc, tdc, acred, &treq);
+ if (!noLock) {
+ if (!(tdc->mflags &DFNextStarted))
+ afs_PrefetchChunk(avc, tdc, acred, &treq);
}
#endif
afs_PutDCache(tdc);
}
if (!noLock)
ReleaseReadLock(&avc->lock);
- osi_FreeSmallSpace(tvec);
- error = afs_CheckCode(error, &treq, 10);
- return error;
+
+ code = afs_CheckCode(error, &treq, 10);
+
+ if (tuiop)
+ afsio_free(tuiop);
+
+out:
+ AFS_DISCON_UNLOCK();
+ return code;
}
/* called with the dcache entry triggering the fetch, the vcache entry involved,
* flag in the prefetched block, so that the next call to read knows to wait
* for the daemon to start doing things.
*
- * This function must be called with the vnode at least read-locked
- * because it plays around with dcache entries.
+ * This function must be called with the vnode at least read-locked, and
+ * no locks on the dcache, because it plays around with dcache entries.
*/
-void afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
- struct AFS_UCRED *acred, struct vrequest *areq)
+void
+afs_PrefetchChunk(struct vcache *avc, struct dcache *adc,
+ afs_ucred_t *acred, struct vrequest *areq)
{
- register struct dcache *tdc;
- register afs_int32 offset;
- afs_int32 j1, j2; /* junk vbls for GetDCache to trash */
-
- offset = adc->f.chunk+1; /* next chunk we'll need */
- offset = AFS_CHUNKTOBASE(offset); /* base of next chunk */
- if (offset < avc->m.Length && !afs_BBusy()) {
+ struct dcache *tdc;
+ afs_size_t offset;
+ afs_size_t j1, j2; /* junk vbls for GetDCache to trash */
+
+ offset = adc->f.chunk + 1; /* next chunk we'll need */
+ offset = AFS_CHUNKTOBASE(offset); /* base of next chunk */
+ ObtainReadLock(&adc->lock);
+ ObtainSharedLock(&adc->mflock, 662);
+ if (offset < avc->f.m.Length && !(adc->mflags & DFNextStarted)
+ && !afs_BBusy()) {
struct brequest *bp;
- adc->flags |= DFNextStarted; /* we've tried to prefetch for this guy */
+
+ UpgradeSToWLock(&adc->mflock, 663);
+ adc->mflags |= DFNextStarted; /* we've tried to prefetch for this guy */
+ ReleaseWriteLock(&adc->mflock);
+ ReleaseReadLock(&adc->lock);
+
tdc = afs_GetDCache(avc, offset, areq, &j1, &j2, 2); /* type 2 never returns 0 */
- if (!(tdc->flags & DFFetchReq)) {
+ /*
+ * In disconnected mode, type 2 can return 0 because it doesn't
+ * make any sense to allocate a dcache we can never fill
+ */
+ if (tdc == NULL)
+ return;
+
+ ObtainSharedLock(&tdc->mflock, 651);
+ if (!(tdc->mflags & DFFetchReq)) {
/* ask the daemon to do the work */
- tdc->flags |= DFFetchReq; /* guaranteed to be cleared by BKG or GetDCache */
+ UpgradeSToWLock(&tdc->mflock, 652);
+ tdc->mflags |= DFFetchReq; /* guaranteed to be cleared by BKG or GetDCache */
/* last parm (1) tells bkg daemon to do an afs_PutDCache when it is done,
* since we don't want to wait for it to finish before doing so ourselves.
*/
-#ifdef AFS_SUN5_ENVX
- mutex_exit(&tdc->lock);
-#endif
- bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred, (long)offset,
- (long) tdc, 1L, 0L, 0L);
+ bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
+ (afs_size_t) offset, (afs_size_t) 1, tdc,
+ (void *)0, (void *)0);
if (!bp) {
/* Bkg table full; just abort non-important prefetching to avoid deadlocks */
- tdc->flags &= ~(DFNextStarted | DFFetchReq);
+ tdc->mflags &= ~DFFetchReq;
+ ReleaseWriteLock(&tdc->mflock);
afs_PutDCache(tdc);
- return;
- }
- }
- else
- afs_PutDCache(tdc);
- }
-}
-
-/* if the vcache is up-to-date, and the request fits entirely into the chunk
- * that the hint here references, then we just use it quickly, otherwise we
- * have to call the slow read.
- *
- * This could be generalized in several ways to take advantage of partial
- * state even when all the chips don't fall the right way. For instance,
- * if the hint is good and the first part of the read request can be
- * satisfied from the chunk, then just do the read. After the read has
- * completed, check to see if there's more. (Chances are there won't be.)
- * If there is more, then just call afs_UFSReadSlow and let it do the rest.
- *
- * For the time being, I'm ignoring quick.f, but it should be used at
- * some future date.
- * do this in the future avc->quick.f = tfile; but I think it
- * has to be done under a write lock, but don't want to wait on the
- * write lock
- */
- /* everywhere that a dcache can be freed (look for NULLIDX)
- * probably does it under a write lock on xdcache. Need to invalidate
- * stamp there, too.
- * Also need to worry about DFFetching, and IFFree, I think. */
-static struct dcache *savedc = 0;
-
-afs_UFSReadFast(avc, auio, acred, albn, abpp, noLock)
- register struct vcache *avc;
- struct uio *auio;
- struct AFS_UCRED *acred;
- int noLock;
- daddr_t albn;
- struct buf **abpp; {
- struct vrequest treq;
- int offDiff;
- struct dcache *tdc;
- struct vnode *vp;
- struct osi_file *tfile;
- afs_int32 code = 0;
-
- if (!noLock)
- ObtainReadLock(&avc->lock);
- ObtainReadLock(&afs_xdcache);
-
- if ((avc->states & CStatd) /* up to date */
- && (tdc = avc->quick.dc) && (tdc->index != NULLIDX)
- && !(afs_indexFlags[tdc->index] & IFFree)) {
-
- tdc->refCount++;
- ReleaseReadLock(&afs_xdcache);
-
- if ((tdc->stamp == avc->quick.stamp) /* hint matches */
- && ((offDiff = (auio->afsio_offset - avc->quick.minLoc)) >= 0)
- && (tdc->f.chunkBytes >= auio->afsio_resid + offDiff)
- && !(tdc->flags & DFFetching)) { /* fits in chunk */
-
- auio->afsio_offset -= avc->quick.minLoc;
-
- tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
-
-#ifdef AFS_AIX_ENV
-#ifdef AFS_AIX41_ENV
- AFS_GUNLOCK();
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, auio, NULL, NULL, NULL, &afs_osi_cred);
- AFS_GLOCK();
-#else
-#ifdef AFS_AIX32_ENV
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, auio, NULL, NULL);
-#else
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t)&auio->afsio_offset, auio, NULL, NULL, -1);
-#endif
-#endif
-#else
-#ifdef AFS_SUN5_ENV
- AFS_GUNLOCK();
- VOP_RWLOCK(tfile->vnode, 0);
- code = VOP_READ(tfile->vnode, auio, 0, &afs_osi_cred);
- VOP_RWUNLOCK(tfile->vnode, 0);
- AFS_GLOCK();
-#else
-#if defined(AFS_SGI_ENV)
- AFS_GUNLOCK();
- AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
- AFS_VOP_READ(tfile->vnode, auio, IO_ISLOCKED, &afs_osi_cred, code);
- AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
- AFS_GLOCK();
-#else
-#ifdef AFS_OSF_ENV
- auio->uio_rw = UIO_READ;
- AFS_GUNLOCK();
- VOP_READ(tfile->vnode, auio, 0, &afs_osi_cred, code);
- AFS_GLOCK();
-#else /* AFS_OSF_ENV */
-#if defined(AFS_HPUX100_ENV)
- AFS_GUNLOCK();
- code = VOP_RDWR(tfile->vnode, auio, UIO_READ, 0, &afs_osi_cred);
- AFS_GLOCK();
-#else
-#if defined(AFS_LINUX20_ENV)
- AFS_GUNLOCK();
- code = osi_file_uio_rdwr(tfile, auio, UIO_READ);
- AFS_GLOCK();
-#else
- code = VOP_RDWR(tfile->vnode, auio, UIO_READ, 0, &afs_osi_cred);
-#endif
-#endif
-#endif
-#endif
-#endif
-#endif
- auio->afsio_offset += avc->quick.minLoc;
- osi_UFSClose(tfile);
- /* Fix up LRU info */
- hset(afs_indexTimes[tdc->index], afs_indexCounter);
- hadd32(afs_indexCounter, 1);
-
- if (!noLock) {
-#ifndef AFS_VM_RDWR_ENV
- if (!(code = afs_InitReq(&treq, acred))&& (!(tdc->flags & DFNextStarted)))
- afs_PrefetchChunk(avc, tdc, acred, &treq);
-#endif
- ReleaseReadLock(&avc->lock);
+ /*
+ * DCLOCKXXX: This is a little sketchy, since someone else
+ * could have already started a prefetch.. In practice,
+ * this probably doesn't matter; at most it would cause an
+ * extra slot in the BKG table to be used up when someone
+ * prefetches this for the second time.
+ */
+ ObtainReadLock(&adc->lock);
+ ObtainWriteLock(&adc->mflock, 664);
+ adc->mflags &= ~DFNextStarted;
+ ReleaseWriteLock(&adc->mflock);
+ ReleaseReadLock(&adc->lock);
+ } else {
+ ReleaseWriteLock(&tdc->mflock);
}
- tdc->refCount--;
- return (code);
- }
- if (!tdc->f.chunkBytes) { /* debugging f.chunkBytes == 0 problem */
- savedc = tdc;
+ } else {
+ ReleaseSharedLock(&tdc->mflock);
+ afs_PutDCache(tdc);
}
- tdc->refCount--;
} else {
- ReleaseReadLock(&afs_xdcache);
+ ReleaseSharedLock(&adc->mflock);
+ ReleaseReadLock(&adc->lock);
}
-
- /* come here if fast path doesn't work for some reason or other */
- if (!noLock)
- ReleaseReadLock(&avc->lock);
- return afs_UFSRead(avc, auio, acred, albn, abpp, noLock);
}
-afs_UFSRead(avc, auio, acred, albn, abpp, noLock)
- struct vcache *avc;
- struct uio *auio;
- struct AFS_UCRED *acred;
- daddr_t albn;
- int noLock;
- struct buf **abpp; {
- afs_int32 totalLength;
- afs_int32 transferLength;
- afs_int32 filePos;
- struct dcache *tdc=0;
- afs_int32 offset, len, error;
- struct uio tuio;
- struct iovec *tvec;
+int
+afs_UFSReadUIO(afs_dcache_id_t *cacheId, struct uio *tuiop)
+{
+ int code;
struct osi_file *tfile;
- afs_int32 code;
- int munlocked, trybusy=1;
- struct vnode *vp;
- struct vrequest treq;
- AFS_STATCNT(afs_UFSRead);
- if (avc->vc_error)
- return EIO;
-
- /* check that we have the latest status info in the vnode cache */
- if (code = afs_InitReq(&treq, acred)) return code;
- if (!noLock) {
- if (!avc)
- osi_Panic ("null avc in afs_UFSRead");
- else {
- code = afs_VerifyVCache(avc, &treq);
- if (code) {
- code = afs_CheckCode(code, &treq, 11); /* failed to get it */
- return code;
- }
- }
- }
-
-#ifndef AFS_VM_RDWR_ENV
- if (AFS_NFSXLATORREQ(acred)) {
- if (!afs_AccessOK(avc, PRSFS_READ, &treq,
- CHECK_MODE_BITS|CMB_ALLOW_EXEC_AS_READ)) {
- return afs_CheckCode(EACCES, &treq, 12);
- }
- }
-#endif
-
- tvec = (struct iovec *) osi_AllocSmallSpace(sizeof(struct iovec));
- totalLength = auio->afsio_resid;
- filePos = auio->afsio_offset;
- afs_Trace4(afs_iclSetp, CM_TRACE_READ, ICL_TYPE_POINTER, avc,
- ICL_TYPE_INT32, filePos, ICL_TYPE_INT32, totalLength,
- ICL_TYPE_INT32, avc->m.Length);
- error = 0;
- transferLength = 0;
- if (!noLock)
- ObtainReadLock(&avc->lock);
-#if defined(AFS_TEXT_ENV) && !defined(AFS_VM_RDWR_ENV)
- if (avc->flushDV.high == AFS_MAXDV && avc->flushDV.low == AFS_MAXDV) {
- hset(avc->flushDV, avc->m.DataVersion);
- }
-#endif
-
- while (totalLength > 0) {
- /* read all of the cached info */
- if (filePos >= avc->m.Length) break; /* all done */
- if (noLock) {
- if (tdc) afs_PutDCache(tdc);
- tdc = afs_FindDCache(avc, filePos);
- if (tdc) {
- offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
- len = tdc->f.chunkBytes - offset;
- }
- } else {
- /* a tricky question: does the presence of the DFFetching flag
- mean that we're fetching the latest version of the file? No.
- The server could update the file as soon as the fetch responsible
- for the setting of the DFFetching flag completes.
-
- However, the presence of the DFFetching flag (visible under a
- read lock since it is set and cleared only under a write lock)
- means that we're fetching as good a version as was known to this
- client at the time of the last call to afs_VerifyVCache, since
- the latter updates the stat cache's m.DataVersion field under a
- write lock, and from the time that the DFFetching flag goes on
- (before the fetch starts), to the time it goes off (after the
- fetch completes), afs_GetDCache keeps at least a read lock
- (actually it keeps an S lock) on the cache entry.
-
- This means that if the DFFetching flag is set, we can use that
- data for any reads that must come from the current version of
- the file (current == m.DataVersion).
-
- Another way of looking at this same point is this: if we're
- fetching some data and then try do an afs_VerifyVCache, the
- VerifyVCache operation will not complete until after the
- DFFetching flag is turned off and the dcache entry's f.versionNo
- field is updated.
-
- Note, by the way, that if DFFetching is set,
- m.DataVersion > f.versionNo (the latter is not updated until
- after the fetch completes).
- */
- if (tdc) afs_PutDCache(tdc); /* before reusing tdc */
- munlocked = 0;
- tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 2);
- if (tdc == savedc) {
- savedc = 0;
- }
- /* now, first try to start transfer, if we'll need the data. If
- * data already coming, we don't need to do this, obviously. Type
- * 2 requests never return a null dcache entry, btw. */
- if (!(tdc->flags & DFFetching)
- && !hsame(avc->m.DataVersion, tdc->f.versionNo)) {
- /* have cache entry, it is not coming in now, and we'll need new data */
-tagain:
- if (trybusy && !afs_BBusy()) {
- struct brequest *bp;
- /* daemon is not busy */
- if (!(tdc->flags & DFFetchReq)) {
- tdc->flags |= DFFetchReq;
-#ifdef AFS_SUN5_ENVX
- mutex_exit(&tdc->lock);
- munlocked = 1;
-#endif
- bp = afs_BQueue(BOP_FETCH, avc, B_DONTWAIT, 0, acred,
- (long)filePos, (long) tdc, 0L, 0L);
- if (!bp) {
- /* Bkg table full; retry deadlocks */
- tdc->flags &= ~DFFetchReq;
- trybusy = 0; /* Avoid bkg daemon since they're too busy */
- goto tagain;
- }
- }
- while (tdc->flags & DFFetchReq) {
- /* don't need waiting flag on this one */
- ReleaseReadLock(&avc->lock);
- afs_osi_Sleep(&tdc->validPos);
- ObtainReadLock(&avc->lock);
- }
- }
- }
- /* now data may have started flowing in (if DFFetching is on). If
- * data is now streaming in, then wait for some interesting stuff. */
- while ((tdc->flags & DFFetching) && tdc->validPos <= filePos) {
- /* too early: wait for DFFetching flag to vanish, or data to appear */
- tdc->flags |= DFWaiting;
- ReleaseReadLock(&avc->lock);
- afs_osi_Sleep(&tdc->validPos);
- ObtainReadLock(&avc->lock);
- }
- /* fetching flag gone, data is here, or we never tried (BBusy for instance) */
- if (tdc->flags & DFFetching) {
- /* still fetching, some new data is here: compute length and offset */
- offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
- len = tdc->validPos - filePos;
- }
- else {
- /* no longer fetching, verify data version (avoid new GetDCache call) */
- if (hsame(avc->m.DataVersion, tdc->f.versionNo)) {
- offset = filePos - AFS_CHUNKTOBASE(tdc->f.chunk);
- len = tdc->f.chunkBytes - offset;
- }
- else {
- /* don't have current data, so get it below */
- afs_PutDCache(tdc);
- tdc = (struct dcache *) 0;
- }
- }
-
- if (!tdc) {
- ReleaseReadLock(&avc->lock);
- tdc = afs_GetDCache(avc, filePos, &treq, &offset, &len, 1);
- ObtainReadLock(&avc->lock);
- }
- }
-
- if (!tdc) {
- error = EIO;
- break;
- }
- if (len > totalLength) len = totalLength; /* will read len bytes */
- if (len <= 0) { /* shouldn't get here if DFFetching is on */
- /* read past the end of a chunk, may not be at next chunk yet, and yet
- also not at eof, so may have to supply fake zeros */
- len = AFS_CHUNKTOSIZE(tdc->f.chunk) - offset; /* bytes left in chunk addr space */
- if (len > totalLength) len = totalLength; /* and still within xfr request */
- code = avc->m.Length - offset; /* and still within file */
- if (len > code) len = code;
- if (len > AFS_ZEROS) len = sizeof(afs_zeros); /* and in 0 buffer */
- afsio_copy(auio, &tuio, tvec);
- afsio_trim(&tuio, len);
- AFS_UIOMOVE(afs_zeros, len, UIO_READ, &tuio, code);
- if (code) {
- error = code;
- break;
- }
- }
- else {
- /* get the data from the file */
-#ifdef IHINT
- if (tfile = tdc->ihint) {
- if (tdc->f.inode != tfile->inum){
- afs_warn( "afs_UFSRead: %x hint mismatch tdc %d inum %d\n",
- tdc, tdc->f.inode, tfile->inum );
- osi_UFSClose(tfile);
- tdc->ihint = tfile = 0;
- nihints--;
- }
- }
- if (tfile != 0) {
- usedihint++;
- }
- else
-#endif /* IHINT */
-
- tfile = (struct osi_file *)osi_UFSOpen(tdc->f.inode);
- /* mung uio structure to be right for this transfer */
- afsio_copy(auio, &tuio, tvec);
- afsio_trim(&tuio, len);
- tuio.afsio_offset = offset;
-#ifdef AFS_AIX_ENV
-#ifdef AFS_AIX41_ENV
- AFS_GUNLOCK();
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL,
- NULL, &afs_osi_cred);
- AFS_GLOCK();
-#else
-#ifdef AFS_AIX32_ENV
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, &tuio, NULL, NULL);
- /* Flush all JFS pages now for big performance gain in big file cases
- * If we do something like this, must check to be sure that AFS file
- * isn't mmapped... see afs_gn_map() for why.
- */
-/*
- if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
- many different ways to do similar things:
+ tfile = (struct osi_file *) osi_UFSOpen(cacheId);
+
+#if defined(AFS_AIX41_ENV)
+ AFS_GUNLOCK();
+ code =
+ VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, tuiop, NULL, NULL,
+ NULL, afs_osi_credp);
+ AFS_GLOCK();
+#elif defined(AFS_AIX32_ENV)
+ code =
+ VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, tuiop, NULL, NULL);
+ /* Flush all JFS pages now for big performance gain in big file cases
+ * If we do something like this, must check to be sure that AFS file
+ * isn't mmapped... see afs_gn_map() for why.
+ */
+ /*
+ if (tfile->vnode->v_gnode && tfile->vnode->v_gnode->gn_seg) {
+ any different ways to do similar things:
so far, the best performing one is #2, but #1 might match it if we
straighten out the confusion regarding which pages to flush. It
really does matter.
(len + PAGESIZE-1)/PAGESIZE);
}
*/
+#elif defined(AFS_AIX_ENV)
+ code =
+ VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t) & offset,
+ tuiop, NULL, NULL, -1);
+#elif defined(AFS_SUN5_ENV)
+ AFS_GUNLOCK();
+#ifdef AFS_SUN510_ENV
+ {
+ caller_context_t ct;
+ VOP_RWLOCK(tfile->vnode, 0, &ct);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp, &ct);
+ VOP_RWUNLOCK(tfile->vnode, 0, &ct);
+ }
#else
- code = VNOP_RDWR(tfile->vnode, UIO_READ, FREAD, (off_t)&offset, &tuio, NULL, NULL, -1);
-#endif
+ VOP_RWLOCK(tfile->vnode, 0);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+ VOP_RWUNLOCK(tfile->vnode, 0);
#endif
+ AFS_GLOCK();
+#elif defined(AFS_SGI_ENV)
+ AFS_GUNLOCK();
+ AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
+ AFS_VOP_READ(tfile->vnode, tuiop, IO_ISLOCKED, afs_osi_credp,
+ code);
+ AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
+ AFS_GLOCK();
+#elif defined(AFS_HPUX100_ENV)
+ AFS_GUNLOCK();
+ code = VOP_RDWR(tfile->vnode, tuiop, UIO_READ, 0, afs_osi_credp);
+ AFS_GLOCK();
+#elif defined(AFS_LINUX20_ENV)
+ AFS_GUNLOCK();
+ code = osi_rdwr(tfile, tuiop, UIO_READ);
+ AFS_GLOCK();
+#elif defined(AFS_DARWIN80_ENV)
+ AFS_GUNLOCK();
+ code = VNOP_READ(tfile->vnode, tuiop, 0, afs_osi_ctxtp);
+ AFS_GLOCK();
+#elif defined(AFS_DARWIN_ENV)
+ AFS_GUNLOCK();
+ VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, current_proc());
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+ VOP_UNLOCK(tfile->vnode, 0, current_proc());
+ AFS_GLOCK();
+#elif defined(AFS_FBSD80_ENV)
+ AFS_GUNLOCK();
+ VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+ VOP_UNLOCK(tfile->vnode, 0);
+ AFS_GLOCK();
+#elif defined(AFS_FBSD_ENV)
+ AFS_GUNLOCK();
+ VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curthread);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+ VOP_UNLOCK(tfile->vnode, 0, curthread);
+ AFS_GLOCK();
+#elif defined(AFS_NBSD_ENV)
+ tuiop->uio_rw = UIO_READ;
+ AFS_GUNLOCK();
+ VOP_LOCK(tfile->vnode, LK_EXCLUSIVE);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+# if defined(AFS_NBSD60_ENV)
+ VOP_UNLOCK(tfile->vnode);
+# else
+ VOP_UNLOCK(tfile->vnode, 0);
+# endif
+ AFS_GLOCK();
+#elif defined(AFS_XBSD_ENV)
+ AFS_GUNLOCK();
+ VOP_LOCK(tfile->vnode, LK_EXCLUSIVE, curproc);
+ code = VOP_READ(tfile->vnode, tuiop, 0, afs_osi_credp);
+ VOP_UNLOCK(tfile->vnode, 0, curproc);
+ AFS_GLOCK();
#else
-#ifdef AFS_SUN5_ENV
- AFS_GUNLOCK();
- VOP_RWLOCK(tfile->vnode, 0);
- code = VOP_READ(tfile->vnode, &tuio, 0, &afs_osi_cred);
- VOP_RWUNLOCK(tfile->vnode, 0);
- AFS_GLOCK();
-#else
-#if defined(AFS_SGI_ENV)
- AFS_GUNLOCK();
- AFS_VOP_RWLOCK(tfile->vnode, VRWLOCK_READ);
- AFS_VOP_READ(tfile->vnode, &tuio, IO_ISLOCKED, &afs_osi_cred,
- code);
- AFS_VOP_RWUNLOCK(tfile->vnode, VRWLOCK_READ);
- AFS_GLOCK();
-#else
-#ifdef AFS_OSF_ENV
- tuio.uio_rw = UIO_READ;
- AFS_GUNLOCK();
- VOP_READ(tfile->vnode, &tuio, 0, &afs_osi_cred, code);
- AFS_GLOCK();
-#else /* AFS_OSF_ENV */
-#ifdef AFS_SUN_ENV
- code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, &afs_osi_cred);
-#else
-#if defined(AFS_HPUX100_ENV)
- AFS_GUNLOCK();
- code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, &afs_osi_cred);
- AFS_GLOCK();
-#else
-#if defined(AFS_LINUX20_ENV)
- AFS_GUNLOCK();
- code = osi_file_uio_rdwr(tfile, &tuio, UIO_READ);
- AFS_GLOCK();
-#else
- code = VOP_RDWR(tfile->vnode, &tuio, UIO_READ, 0, &afs_osi_cred);
-#endif
-#endif
-#endif
+ code = VOP_RDWR(tfile->vnode, tuiop, UIO_READ, 0, afs_osi_credp);
#endif
-#endif
-#endif
-#endif
-
-#ifdef IHINT
- if (!tdc->ihint && nihints < maxIHint) {
- tdc->ihint = tfile;
- nihints++;
- }
- else
-#endif /* IHINT */
- osi_UFSClose(tfile);
-
- if (code) {
- error = code;
- break;
- }
- }
- /* otherwise we've read some, fixup length, etc and continue with next seg */
- len = len - tuio.afsio_resid; /* compute amount really transferred */
- afsio_skip(auio, len); /* update input uio structure */
- totalLength -= len;
- transferLength += len;
- filePos += len;
- if (len <= 0) break; /* surprise eof */
- }
-
- /* if we make it here with tdc non-zero, then it is the last chunk we
- * dealt with, and we have to release it when we're done. We hold on
- * to it in case we need to do a prefetch, obviously.
- */
- if (tdc) {
-#ifndef AFS_VM_RDWR_ENV
- /* try to queue prefetch, if needed */
- if (!(tdc->flags & DFNextStarted) && !noLock) {
- afs_PrefetchChunk(avc, tdc, acred, &treq);
- }
-#endif
- afs_PutDCache(tdc);
- }
- if (!noLock)
- ReleaseReadLock(&avc->lock);
+ osi_UFSClose(tfile);
- osi_FreeSmallSpace(tvec);
- error = afs_CheckCode(error, &treq, 13);
- return error;
+ return code;
}