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
10 #include <afsconfig.h>
11 #include "afs/param.h"
16 #include <sys/sleep.h>
19 #include "afs/sysincludes.h" /* Standard vendor system headers */
20 #include "afsincludes.h" /* Afs-based standard headers */
21 #include "afs/afs_stats.h" /* statistics gathering code */
22 #include "afs/afs_cbqueue.h"
24 #include <sys/adspace.h> /* for vm_att(), vm_det() */
27 #if defined(AFS_CACHE_BYPASS)
28 #include "afs/afs_bypasscache.h"
29 #endif// defined(AFS_CACHE_BYPASS)
30 /* background request queue size */
31 afs_lock_t afs_xbrs; /* lock for brs */
32 static int brsInit = 0;
33 short afs_brsWaiters = 0; /* number of users waiting for brs buffers */
34 short afs_brsDaemons = 0; /* number of daemons waiting for brs requests */
35 struct brequest afs_brs[NBRS]; /* request structures */
36 struct afs_osi_WaitHandle AFS_WaitHandler, AFS_CSWaitHandler;
37 static int afs_brs_count = 0; /* request counter, to service reqs in order */
39 static int rxepoch_checked = 0;
40 #define afs_CheckRXEpoch() {if (rxepoch_checked == 0 && rxkad_EpochWasSet) { \
41 rxepoch_checked = 1; afs_GCUserData(/* force flag */ 1); } }
43 /* PAG garbage collection */
44 /* We induce a compile error if param.h does not define AFS_GCPAGS */
45 afs_int32 afs_gcpags = AFS_GCPAGS;
46 afs_int32 afs_gcpags_procsize = 0;
48 afs_int32 afs_CheckServerDaemonStarted = 0;
49 #ifndef DEFAULT_PROBE_INTERVAL
50 #define DEFAULT_PROBE_INTERVAL 30 /* default to 3 min */
52 afs_int32 afs_probe_interval = DEFAULT_PROBE_INTERVAL;
53 afs_int32 afs_probe_all_interval = 600;
54 afs_int32 afs_nat_probe_interval = 60;
55 afs_int32 afs_preCache = 0;
57 #define PROBE_WAIT() (1000 * (afs_probe_interval - ((afs_random() & 0x7fffffff) \
58 % (afs_probe_interval/2))))
61 afs_SetCheckServerNATmode(int isnat)
63 static afs_int32 old_intvl, old_all_intvl;
66 if (isnat && !wasnat) {
67 old_intvl = afs_probe_interval;
68 old_all_intvl = afs_probe_all_interval;
69 afs_probe_interval = afs_nat_probe_interval;
70 afs_probe_all_interval = afs_nat_probe_interval;
71 afs_osi_CancelWait(&AFS_CSWaitHandler);
72 } else if (!isnat && wasnat) {
73 afs_probe_interval = old_intvl;
74 afs_probe_all_interval = old_all_intvl;
80 afs_CheckServerDaemon(void)
82 afs_int32 now, delay, lastCheck, last10MinCheck;
84 afs_CheckServerDaemonStarted = 1;
86 while (afs_initState < 101)
87 afs_osi_Sleep(&afs_initState);
88 afs_osi_Wait(PROBE_WAIT(), &AFS_CSWaitHandler, 0);
90 last10MinCheck = lastCheck = osi_Time();
92 if (afs_termState == AFSOP_STOP_CS) {
93 afs_termState = AFSOP_STOP_BKG;
94 afs_osi_Wakeup(&afs_termState);
99 if (afs_probe_interval + lastCheck <= now) {
100 afs_CheckServers(1, NULL); /* check down servers */
101 lastCheck = now = osi_Time();
104 if (afs_probe_all_interval + last10MinCheck <= now) {
105 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, afs_probe_all_interval);
106 afs_CheckServers(0, NULL);
107 last10MinCheck = now = osi_Time();
109 /* shutdown check. */
110 if (afs_termState == AFSOP_STOP_CS) {
111 afs_termState = AFSOP_STOP_BKG;
112 afs_osi_Wakeup(&afs_termState);
116 /* Compute time to next probe. */
117 delay = afs_probe_interval + lastCheck;
118 if (delay > afs_probe_all_interval + last10MinCheck)
119 delay = afs_probe_all_interval + last10MinCheck;
123 afs_osi_Wait(delay * 1000, &AFS_CSWaitHandler, 0);
125 afs_CheckServerDaemonStarted = 0;
128 extern int vfs_context_ref;
130 /* This function always holds the GLOCK whilst it is running. The caller
131 * gets the GLOCK before invoking it, and afs_osi_Sleep drops the GLOCK
132 * whilst we are sleeping, and regains it when we're woken up.
138 struct afs_exporter *exporter;
140 afs_int32 last3MinCheck, last10MinCheck, last60MinCheck, lastNMinCheck;
141 afs_int32 last1MinCheck, last5MinCheck;
142 afs_uint32 lastCBSlotBump;
145 AFS_STATCNT(afs_Daemon);
146 last1MinCheck = last3MinCheck = last60MinCheck = last10MinCheck =
147 last5MinCheck = lastNMinCheck = 0;
149 afs_rootFid.Fid.Volume = 0;
150 while (afs_initState < 101)
151 afs_osi_Sleep(&afs_initState);
153 #ifdef AFS_DARWIN80_ENV
154 if (afs_osi_ctxtp_initialized)
155 osi_Panic("vfs context already initialized");
156 while (afs_osi_ctxtp && vfs_context_ref)
157 afs_osi_Sleep(&afs_osi_ctxtp);
158 if (afs_osi_ctxtp && !vfs_context_ref)
159 vfs_context_rele(afs_osi_ctxtp);
160 afs_osi_ctxtp = vfs_context_create(NULL);
161 afs_osi_ctxtp_initialized = 1;
164 lastCBSlotBump = now;
166 /* when a lot of clients are booted simultaneously, they develop
167 * annoying synchronous VL server bashing behaviors. So we stagger them.
169 last1MinCheck = now + ((afs_random() & 0x7fffffff) % 60); /* an extra 30 */
170 last3MinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
171 last60MinCheck = now - 1800 + ((afs_random() & 0x7fffffff) % 3600);
172 last10MinCheck = now - 300 + ((afs_random() & 0x7fffffff) % 600);
173 last5MinCheck = now - 150 + ((afs_random() & 0x7fffffff) % 300);
174 lastNMinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
176 /* start off with afs_initState >= 101 (basic init done) */
178 afs_CheckCallbacks(20); /* unstat anything which will expire soon */
180 /* things to do every 20 seconds or less - required by protocol spec */
182 afs_FlushActiveVcaches(0); /* flush NFS writes */
183 afs_FlushVCBs(1); /* flush queued callbacks */
184 afs_MaybeWakeupTruncateDaemon(); /* free cache space if have too */
185 rx_CheckPackets(); /* Does RX need more packets? */
188 if (lastCBSlotBump + CBHTSLOTLEN < now) { /* pretty time-dependant */
189 lastCBSlotBump = now;
190 if (afs_BumpBase()) {
191 afs_CheckCallbacks(20); /* unstat anything which will expire soon */
195 if (last1MinCheck + 60 < now) {
196 /* things to do every minute */
197 DFlush(); /* write out dir buffers */
198 afs_WriteThroughDSlots(); /* write through cacheinfo entries */
199 ObtainWriteLock(&afs_xvcache, 736);
200 afs_FlushReclaimedVcaches();
201 ReleaseWriteLock(&afs_xvcache);
202 afs_FlushActiveVcaches(1); /* keep flocks held & flush nfs writes */
204 #ifdef AFS_DISCON_ENV
205 afs_StoreDirtyVcaches();
212 if (last3MinCheck + 180 < now) {
213 afs_CheckTokenCache(); /* check for access cache resets due to expired
218 if (afsd_dynamic_vcaches && (last5MinCheck + 300 < now)) {
219 /* start with trying to drop us back to our base usage */
221 if (afs_maxvcount <= afs_cacheStats)
222 anumber = VCACHE_FREE;
224 anumber = VCACHE_FREE + (afs_maxvcount - afs_cacheStats);
226 ObtainWriteLock(&afs_xvcache, 734);
227 afs_ShakeLooseVCaches(anumber);
228 ReleaseWriteLock(&afs_xvcache);
232 if (!afs_CheckServerDaemonStarted) {
233 /* Do the check here if the correct afsd is not installed. */
236 printf("Please install afsd with check server daemon.\n");
238 if (lastNMinCheck + afs_probe_interval < now) {
239 /* only check down servers */
240 afs_CheckServers(1, NULL);
244 if (last10MinCheck + 600 < now) {
245 #ifdef AFS_USERSPACE_IP_ADDR
246 extern int rxi_GetcbiInfo(void);
248 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, 600);
249 #ifdef AFS_USERSPACE_IP_ADDR
250 if (rxi_GetcbiInfo()) { /* addresses changed from last time */
253 #else /* AFS_USERSPACE_IP_ADDR */
254 if (rxi_GetIFInfo()) { /* addresses changed from last time */
257 #endif /* else AFS_USERSPACE_IP_ADDR */
258 if (!afs_CheckServerDaemonStarted)
259 afs_CheckServers(0, NULL);
260 afs_GCUserData(0); /* gc old conns */
261 /* This is probably the wrong way of doing GC for the various exporters but it will suffice for a while */
262 for (exporter = root_exported; exporter;
263 exporter = exporter->exp_next) {
264 (void)EXP_GC(exporter, 0); /* Generalize params */
269 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
273 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
278 last10MinCheck = now;
280 if (last60MinCheck + 3600 < now) {
281 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEVOLUME, ICL_TYPE_INT32,
283 afs_CheckRootVolume();
285 if (afs_gcpags == AFS_GCPAGS_OK) {
290 last60MinCheck = now;
292 if (afs_initState < 300) { /* while things ain't rosy */
293 code = afs_CheckRootVolume();
295 afs_initState = 300; /* succeeded */
296 if (afs_initState < 200)
297 afs_initState = 200; /* tried once */
298 afs_osi_Wakeup(&afs_initState);
301 /* 18285 is because we're trying to divide evenly into 128, that is,
302 * CBSlotLen, while staying just under 20 seconds. If CBSlotLen
303 * changes, should probably change this interval, too.
304 * Some of the preceding actions may take quite some time, so we
305 * might not want to wait the entire interval */
306 now = 18285 - (osi_Time() - now);
308 afs_osi_Wait(now, &AFS_WaitHandler, 0);
311 if (afs_termState == AFSOP_STOP_AFS) {
312 if (afs_CheckServerDaemonStarted)
313 afs_termState = AFSOP_STOP_CS;
315 afs_termState = AFSOP_STOP_BKG;
316 afs_osi_Wakeup(&afs_termState);
323 afs_CheckRootVolume(void)
325 char rootVolName[32];
326 struct volume *tvp = NULL;
327 int usingDynroot = afs_GetDynrootEnable();
330 AFS_STATCNT(afs_CheckRootVolume);
331 if (*afs_rootVolumeName == 0) {
332 strcpy(rootVolName, "root.afs");
334 strcpy(rootVolName, afs_rootVolumeName);
338 afs_GetDynrootFid(&afs_rootFid);
339 tvp = afs_GetVolume(&afs_rootFid, NULL, READ_LOCK);
341 struct cell *lc = afs_GetPrimaryCell(READ_LOCK);
345 localcell = lc->cellNum;
346 afs_PutCell(lc, READ_LOCK);
347 tvp = afs_GetVolumeByName(rootVolName, localcell, 1, NULL, READ_LOCK);
350 int len = strlen(rootVolName);
352 if ((len < 9) || strcmp(&rootVolName[len - 9], ".readonly")) {
353 strcpy(buf, rootVolName);
354 afs_strcat(buf, ".readonly");
355 tvp = afs_GetVolumeByName(buf, localcell, 1, NULL, READ_LOCK);
359 int volid = (tvp->roVol ? tvp->roVol : tvp->volume);
360 afs_rootFid.Cell = localcell;
361 if (afs_rootFid.Fid.Volume && afs_rootFid.Fid.Volume != volid
363 /* If we had a root fid before and it changed location we reset
364 * the afs_globalVp so that it will be reevaluated.
365 * Just decrement the reference count. This only occurs during
366 * initial cell setup and can panic the machine if we set the
367 * count to zero and fs checkv is executed when the current
370 #ifdef AFS_LINUX20_ENV
372 struct vrequest treq;
378 afs_rootFid.Fid.Volume = volid;
379 afs_rootFid.Fid.Vnode = 1;
380 afs_rootFid.Fid.Unique = 1;
383 if (afs_InitReq(&treq, credp))
385 vcp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
388 afs_getattr(vcp, &vattr, credp);
389 afs_fill_inode(AFSTOV(vcp), &vattr);
391 dp = d_find_alias(AFSTOV(afs_globalVp));
393 #if defined(AFS_LINUX24_ENV)
394 spin_lock(&dcache_lock);
395 #if defined(AFS_LINUX26_ENV)
396 spin_lock(&dp->d_lock);
399 list_del_init(&dp->d_alias);
400 list_add(&dp->d_alias, &(AFSTOV(vcp)->i_dentry));
401 dp->d_inode = AFSTOV(vcp);
402 #if defined(AFS_LINUX24_ENV)
403 #if defined(AFS_LINUX26_ENV)
404 spin_unlock(&dp->d_lock);
406 spin_unlock(&dcache_lock);
410 AFS_FAST_RELE(afs_globalVp);
416 #ifdef AFS_DARWIN80_ENV
417 afs_PutVCache(afs_globalVp);
419 AFS_FAST_RELE(afs_globalVp);
424 afs_rootFid.Fid.Volume = volid;
425 afs_rootFid.Fid.Vnode = 1;
426 afs_rootFid.Fid.Unique = 1;
430 afs_initState = 300; /* won */
431 afs_osi_Wakeup(&afs_initState);
432 afs_PutVolume(tvp, READ_LOCK);
434 if (afs_rootFid.Fid.Volume)
440 /* ptr_parm 0 is the pathname, size_parm 0 to the fetch is the chunk number */
442 BPath(register struct brequest *ab)
444 register struct dcache *tdc = NULL;
445 struct vcache *tvc = NULL;
446 struct vnode *tvn = NULL;
447 #ifdef AFS_LINUX22_ENV
448 struct dentry *dp = NULL;
450 afs_size_t offset, len;
451 struct vrequest treq;
455 if ((code = afs_InitReq(&treq, ab->cred)))
458 #ifdef AFS_LINUX22_ENV
459 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &dp);
461 tvn = (struct vnode *)dp->d_inode;
463 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &tvn);
466 osi_FreeLargeSpace((char *)ab->ptr_parm[0]); /* free path name buffer here */
469 /* now path may not have been in afs, so check that before calling our cache manager */
470 if (!tvn || !IsAfsVnode(tvn)) {
471 /* release it and give up */
473 #ifdef AFS_LINUX22_ENV
482 /* here we know its an afs vnode, so we can get the data for the chunk */
483 tdc = afs_GetDCache(tvc, ab->size_parm[0], &treq, &offset, &len, 1);
487 #ifdef AFS_LINUX22_ENV
494 /* size_parm 0 to the fetch is the chunk number,
495 * ptr_parm 0 is the dcache entry to wakeup,
496 * size_parm 1 is true iff we should release the dcache entry here.
499 BPrefetch(register struct brequest *ab)
501 register struct dcache *tdc;
502 register struct vcache *tvc;
503 afs_size_t offset, len, abyte, totallen = 0;
504 struct vrequest treq;
506 AFS_STATCNT(BPrefetch);
507 if ((len = afs_InitReq(&treq, ab->cred)))
509 abyte = ab->size_parm[0];
512 tdc = afs_GetDCache(tvc, abyte, &treq, &offset, &len, 1);
518 } while ((totallen < afs_preCache) && tdc && (len > 0));
519 /* now, dude may be waiting for us to clear DFFetchReq bit; do so. Can't
520 * use tdc from GetDCache since afs_GetDCache may fail, but someone may
521 * be waiting for our wakeup anyway.
523 tdc = (struct dcache *)(ab->ptr_parm[0]);
524 ObtainSharedLock(&tdc->lock, 640);
525 if (tdc->mflags & DFFetchReq) {
526 UpgradeSToWLock(&tdc->lock, 641);
527 tdc->mflags &= ~DFFetchReq;
528 ReleaseWriteLock(&tdc->lock);
530 ReleaseSharedLock(&tdc->lock);
532 afs_osi_Wakeup(&tdc->validPos);
533 if (ab->size_parm[1]) {
534 afs_PutDCache(tdc); /* put this one back, too */
538 #if defined(AFS_CACHE_BYPASS)
539 #if 1 /* XXX Matt debugging */
543 BPrefetchNoCache(register struct brequest *ab)
545 struct vrequest treq;
548 if ((len = afs_InitReq(&treq, ab->cred)))
552 /* OS-specific prefetch routine */
553 afs_PrefetchNoCache(ab->vc, ab->cred, (struct nocache_read_request *) ab->ptr_parm[0]);
555 #warning Cache-bypass code path not implemented in UKERNEL
561 BStore(register struct brequest *ab)
563 register struct vcache *tvc;
564 register afs_int32 code;
565 struct vrequest treq;
566 #if defined(AFS_SGI_ENV)
567 struct cred *tmpcred;
571 if ((code = afs_InitReq(&treq, ab->cred)))
575 #if defined(AFS_SGI_ENV)
577 * Since StoreOnLastReference can end up calling osi_SyncVM which
578 * calls into VM code that assumes that u.u_cred has the
579 * correct credentials, we set our to theirs for this xaction
581 tmpcred = OSI_GET_CURRENT_CRED();
582 OSI_SET_CURRENT_CRED(ab->cred);
585 * To avoid recursion since the WriteLock may be released during VM
586 * operations, we hold the VOP_RWLOCK across this transaction as
587 * do the other callers of StoreOnLastReference
589 AFS_RWLOCK((vnode_t *) tvc, 1);
591 ObtainWriteLock(&tvc->lock, 209);
592 code = afs_StoreOnLastReference(tvc, &treq);
593 ReleaseWriteLock(&tvc->lock);
594 #if defined(AFS_SGI_ENV)
595 OSI_SET_CURRENT_CRED(tmpcred);
596 AFS_RWUNLOCK((vnode_t *) tvc, 1);
598 /* now set final return code, and wakeup anyone waiting */
599 if ((ab->flags & BUVALID) == 0) {
600 ab->code = afs_CheckCode(code, &treq, 43); /* set final code, since treq doesn't go across processes */
601 ab->flags |= BUVALID;
602 if (ab->flags & BUWAIT) {
603 ab->flags &= ~BUWAIT;
609 /* release a held request buffer */
611 afs_BRelease(register struct brequest *ab)
614 AFS_STATCNT(afs_BRelease);
615 ObtainWriteLock(&afs_xbrs, 294);
616 if (--ab->refCount <= 0) {
620 afs_osi_Wakeup(&afs_brsWaiters);
621 ReleaseWriteLock(&afs_xbrs);
624 /* return true if bkg fetch daemons are all busy */
628 AFS_STATCNT(afs_BBusy);
629 if (afs_brsDaemons > 0)
635 afs_BQueue(register short aopcode, register struct vcache *avc,
636 afs_int32 dontwait, afs_int32 ause, afs_ucred_t *acred,
637 afs_size_t asparm0, afs_size_t asparm1, void *apparm0)
640 register struct brequest *tb;
642 AFS_STATCNT(afs_BQueue);
643 ObtainWriteLock(&afs_xbrs, 296);
646 for (i = 0; i < NBRS; i++, tb++) {
647 if (tb->refCount == 0)
652 tb->opcode = aopcode;
657 VN_HOLD(AFSTOV(avc));
659 tb->refCount = ause + 1;
660 tb->size_parm[0] = asparm0;
661 tb->size_parm[1] = asparm1;
662 tb->ptr_parm[0] = apparm0;
665 tb->ts = afs_brs_count++;
666 /* if daemons are waiting for work, wake them up */
667 if (afs_brsDaemons > 0) {
668 afs_osi_Wakeup(&afs_brsDaemons);
670 ReleaseWriteLock(&afs_xbrs);
674 ReleaseWriteLock(&afs_xbrs);
677 /* no free buffers, sleep a while */
679 ReleaseWriteLock(&afs_xbrs);
680 afs_osi_Sleep(&afs_brsWaiters);
681 ObtainWriteLock(&afs_xbrs, 301);
687 /* AIX 4.1 has a much different sleep/wakeup mechanism available for use.
688 * The modifications here will work for either a UP or MP machine.
690 struct buf *afs_asyncbuf = (struct buf *)0;
691 tid_t afs_asyncbuf_cv = EVENT_NULL;
692 afs_int32 afs_biodcnt = 0;
694 /* in implementing this, I assumed that all external linked lists were
697 * Several places in this code traverse a linked list. The algorithm
698 * used here is probably unfamiliar to most people. Careful examination
699 * will show that it eliminates an assignment inside the loop, as compared
700 * to the standard algorithm, at the cost of occasionally using an extra
706 * This function obtains, and returns, a pointer to a buffer for
707 * processing by a daemon. It sleeps until such a buffer is available.
708 * The source of buffers for it is the list afs_asyncbuf (see also
709 * afs_gn_strategy). This function may be invoked concurrently by
710 * several processes, that is, several instances of the same daemon.
711 * afs_gn_strategy, which adds buffers to the list, runs at interrupt
712 * level, while get_bioreq runs at process level.
714 * Since AIX 4.1 can wake just one process at a time, the separate sleep
715 * addresses have been removed.
716 * Note that the kernel_lock is held until the e_sleep_thread() occurs.
717 * The afs_asyncbuf_lock is primarily used to serialize access between
718 * process and interrupts.
720 Simple_lock afs_asyncbuf_lock;
724 struct buf *bp = NULL;
726 struct buf **bestlbpP, **lbpP;
728 struct buf *t1P, *t2P; /* temp pointers for list manipulation */
731 struct afs_bioqueue *s;
733 /* ??? Does the forward pointer of the returned buffer need to be NULL?
736 /* Disable interrupts from the strategy function, and save the
737 * prior priority level and lock access to the afs_asyncbuf.
740 oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
744 /* look for oldest buffer */
745 bp = bestbp = afs_asyncbuf;
746 bestage = (long)bestbp->av_back;
747 bestlbpP = &afs_asyncbuf;
753 if ((long)bp->av_back - bestage < 0) {
756 bestage = (long)bp->av_back;
760 *bestlbpP = bp->av_forw;
763 /* If afs_asyncbuf is null, it is necessary to go to sleep.
764 * e_wakeup_one() ensures that only one thread wakes.
767 /* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
768 * lock on an MP machine.
771 e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
772 LOCK_HANDLER | INTERRUPTIBLE);
773 if (interrupted == THREAD_INTERRUPTED) {
774 /* re-enable interrupts from strategy */
775 unlock_enable(oldPriority, &afs_asyncbuf_lock);
779 } /* end of "else asyncbuf is empty" */
780 } /* end of "inner loop" */
784 unlock_enable(oldPriority, &afs_asyncbuf_lock);
787 /* For the convenience of other code, replace the gnodes in
788 * the b_vp field of bp and the other buffers on the b_work
789 * chain with the corresponding vnodes.
791 * ??? what happens to the gnodes? They're not just cut loose,
795 t2P = (struct buf *)t1P->b_work;
796 t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
800 t1P = (struct buf *)t2P->b_work;
801 t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
806 /* If the buffer does not specify I/O, it may immediately
807 * be returned to the caller. This condition is detected
808 * by examining the buffer's flags (the b_flags field). If
809 * the B_PFPROT bit is set, the buffer represents a protection
810 * violation, rather than a request for I/O. The remainder
811 * of the outer loop handles the case where the B_PFPROT bit is clear.
813 if (bp->b_flags & B_PFPROT) {
818 } /* end of function get_bioreq() */
823 * This function is the daemon. It is called from the syscall
824 * interface. Ordinarily, a script or an administrator will run a
825 * daemon startup utility, specifying the number of I/O daemons to
826 * run. The utility will fork off that number of processes,
827 * each making the appropriate syscall, which will cause this
828 * function to be invoked.
830 static int afs_initbiod = 0; /* this is self-initializing code */
833 afs_BioDaemon(afs_int32 nbiods)
835 afs_int32 code, s, pflg = 0;
837 struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
845 /* pin lock, since we'll be using it in an interrupt. */
846 lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
847 simple_lock_init(&afs_asyncbuf_lock);
848 pin(&afs_asyncbuf, sizeof(struct buf *));
849 pin(&afs_asyncbuf_cv, sizeof(afs_int32));
852 /* Ignore HUP signals... */
854 sigset_t sigbits, osigbits;
856 * add SIGHUP to the set of already masked signals
858 SIGFILLSET(sigbits); /* allow all signals */
859 SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
860 limit_sigs(&sigbits, &osigbits); /* and already masked */
862 /* Main body starts here -- this is an intentional infinite loop, and
865 * Now, the loop will exit if get_bioreq() returns NULL, indicating
866 * that we've been interrupted.
869 bp = afs_get_bioreq();
871 break; /* we were interrupted */
872 if (code = setjmpx(&jmpbuf)) {
873 /* This should not have happend, maybe a lack of resources */
875 s = disable_lock(INTMAX, &afs_asyncbuf_lock);
876 for (bp1 = bp; bp; bp = bp1) {
878 bp1 = (struct buf *)bp1->b_work;
881 bp->b_flags |= B_ERROR;
884 unlock_enable(s, &afs_asyncbuf_lock);
888 vcp = VTOAFS(bp->b_vp);
889 if (bp->b_flags & B_PFSTORE) { /* XXXX */
890 ObtainWriteLock(&vcp->lock, 404);
891 if (vcp->v.v_gnode->gn_mwrcnt) {
892 afs_offs_t newlength =
893 (afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
894 if (vcp->f.m.Length < newlength) {
895 afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
896 ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
897 __LINE__, ICL_TYPE_OFFSET,
898 ICL_HANDLE_OFFSET(vcp->f.m.Length),
899 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
900 vcp->f.m.Length = newlength;
903 ReleaseWriteLock(&vcp->lock);
905 /* If the buffer represents a protection violation, rather than
906 * an actual request for I/O, no special action need be taken.
908 if (bp->b_flags & B_PFPROT) {
909 iodone(bp); /* Notify all users of the buffer that we're done */
914 ObtainWriteLock(&vcp->pvmlock, 211);
916 * First map its data area to a region in the current address space
917 * by calling vm_att with the subspace identifier, and a pointer to
918 * the data area. vm_att returns a new data area pointer, but we
919 * also want to hang onto the old one.
921 tmpaddr = bp->b_baddr;
922 bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
923 tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
924 if (tmperr) { /* in non-error case */
925 bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
926 bp->b_error = tmperr;
929 /* Unmap the buffer's data area by calling vm_det. Reset data area
930 * to the value that we saved above.
933 bp->b_baddr = tmpaddr;
936 * buffer may be linked with other buffers via the b_work field.
937 * See also afs_gn_strategy. For each buffer in the chain (including
938 * bp) notify all users of the buffer that the daemon is finished
939 * using it by calling iodone.
940 * assumes iodone can modify the b_work field.
943 tbp2 = (struct buf *)tbp1->b_work;
948 tbp1 = (struct buf *)tbp2->b_work;
954 ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
956 } /* infinite loop (unless we're interrupted) */
957 } /* end of afs_BioDaemon() */
959 #endif /* AFS_AIX41_ENV */
964 afs_BackgroundDaemon(void)
969 AFS_STATCNT(afs_BackgroundDaemon);
970 /* initialize subsystem */
972 LOCK_INIT(&afs_xbrs, "afs_xbrs");
973 memset(afs_brs, 0, sizeof(afs_brs));
975 #if defined (AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
977 * steal the first daemon for doing delayed DSlot flushing
978 * (see afs_GetDownDSlot)
987 ObtainWriteLock(&afs_xbrs, 302);
990 struct brequest *min_tb = NULL;
992 if (afs_termState == AFSOP_STOP_BKG) {
994 afs_termState = AFSOP_STOP_TRUNCDAEMON;
995 ReleaseWriteLock(&afs_xbrs);
996 afs_osi_Wakeup(&afs_termState);
1000 /* find a request */
1003 for (i = 0; i < NBRS; i++, tb++) {
1004 /* look for request with smallest ts */
1005 if ((tb->refCount > 0) && !(tb->flags & BSTARTED)) {
1006 /* new request, not yet picked up */
1007 if ((min_tb && (min_ts - tb->ts > 0)) || !min_tb) {
1013 if ((tb = min_tb)) {
1014 /* claim and process this request */
1015 tb->flags |= BSTARTED;
1016 ReleaseWriteLock(&afs_xbrs);
1018 afs_Trace1(afs_iclSetp, CM_TRACE_BKG1, ICL_TYPE_INT32,
1020 if (tb->opcode == BOP_FETCH)
1022 #if defined(AFS_CACHE_BYPASS)
1023 else if (tb->opcode == BOP_FETCH_NOCACHE)
1024 BPrefetchNoCache(tb);
1026 else if (tb->opcode == BOP_STORE)
1028 else if (tb->opcode == BOP_PATH)
1031 panic("background bop");
1033 AFS_RELE(AFSTOV(tb->vc)); /* MUST call vnode layer or could lose vnodes */
1038 tb->cred = (afs_ucred_t *)0;
1040 afs_BRelease(tb); /* this grabs and releases afs_xbrs lock */
1041 ObtainWriteLock(&afs_xbrs, 305);
1044 /* wait for new request */
1046 ReleaseWriteLock(&afs_xbrs);
1047 afs_osi_Sleep(&afs_brsDaemons);
1048 ObtainWriteLock(&afs_xbrs, 307);
1056 shutdown_daemons(void)
1058 AFS_STATCNT(shutdown_daemons);
1059 if (afs_cold_shutdown) {
1060 afs_brsDaemons = brsInit = 0;
1061 rxepoch_checked = afs_nbrs = 0;
1062 memset(afs_brs, 0, sizeof(afs_brs));
1063 memset(&afs_xbrs, 0, sizeof(afs_lock_t));
1065 #ifdef AFS_AIX41_ENV
1066 lock_free(&afs_asyncbuf_lock);
1067 unpin(&afs_asyncbuf, sizeof(struct buf *));
1068 unpin(&afs_asyncbuf_cv, sizeof(afs_int32));
1074 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1076 * sgi - daemon - handles certain operations that otherwise
1077 * would use up too much kernel stack space
1079 * This all assumes that since the caller must have the xdcache lock
1080 * exclusively that the list will never be more than one long
1081 * and noone else can attempt to add anything until we're done.
1083 SV_TYPE afs_sgibksync;
1084 SV_TYPE afs_sgibkwait;
1085 lock_t afs_sgibklock;
1086 struct dcache *afs_sgibklist;
1094 if (afs_sgibklock == NULL) {
1095 SV_INIT(&afs_sgibksync, "bksync", 0, 0);
1096 SV_INIT(&afs_sgibkwait, "bkwait", 0, 0);
1097 SPINLOCK_INIT(&afs_sgibklock, "bklock");
1099 s = SPLOCK(afs_sgibklock);
1101 /* wait for something to do */
1102 SP_WAIT(afs_sgibklock, s, &afs_sgibksync, PINOD);
1103 osi_Assert(afs_sgibklist);
1105 /* XX will probably need to generalize to real list someday */
1106 s = SPLOCK(afs_sgibklock);
1107 while (afs_sgibklist) {
1108 tdc = afs_sgibklist;
1109 afs_sgibklist = NULL;
1110 SPUNLOCK(afs_sgibklock, s);
1112 tdc->dflags &= ~DFEntryMod;
1113 afs_WriteDCache(tdc, 1);
1115 s = SPLOCK(afs_sgibklock);
1118 /* done all the work - wake everyone up */
1119 while (SV_SIGNAL(&afs_sgibkwait));