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"
18 #include <sys/sleep.h>
21 #include "afs/sysincludes.h" /* Standard vendor system headers */
22 #include "afsincludes.h" /* Afs-based standard headers */
23 #include "afs/afs_stats.h" /* statistics gathering code */
24 #include "afs/afs_cbqueue.h"
26 #include <sys/adspace.h> /* for vm_att(), vm_det() */
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;
56 #define PROBE_WAIT() (1000 * (afs_probe_interval - ((afs_random() & 0x7fffffff) \
57 % (afs_probe_interval/2))))
60 afs_SetCheckServerNATmode(int isnat)
62 static afs_int32 old_intvl, old_all_intvl;
65 if (isnat && !wasnat) {
66 old_intvl = afs_probe_interval;
67 old_all_intvl = afs_probe_all_interval;
68 afs_probe_interval = afs_nat_probe_interval;
69 afs_probe_all_interval = afs_nat_probe_interval;
70 afs_osi_CancelWait(&AFS_CSWaitHandler);
71 } else if (!isnat && wasnat) {
72 afs_probe_interval = old_intvl;
73 afs_probe_all_interval = old_all_intvl;
79 afs_CheckServerDaemon(void)
81 afs_int32 now, delay, lastCheck, last10MinCheck;
83 afs_CheckServerDaemonStarted = 1;
85 while (afs_initState < 101)
86 afs_osi_Sleep(&afs_initState);
87 afs_osi_Wait(PROBE_WAIT(), &AFS_CSWaitHandler, 0);
89 last10MinCheck = lastCheck = osi_Time();
91 if (afs_termState == AFSOP_STOP_CS) {
92 afs_termState = AFSOP_STOP_BKG;
93 afs_osi_Wakeup(&afs_termState);
98 if (afs_probe_interval + lastCheck <= now) {
99 afs_CheckServers(1, NULL); /* check down servers */
100 lastCheck = now = osi_Time();
103 if (afs_probe_all_interval + last10MinCheck <= now) {
104 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, afs_probe_all_interval);
105 afs_CheckServers(0, NULL);
106 last10MinCheck = now = osi_Time();
108 /* shutdown check. */
109 if (afs_termState == AFSOP_STOP_CS) {
110 afs_termState = AFSOP_STOP_BKG;
111 afs_osi_Wakeup(&afs_termState);
115 /* Compute time to next probe. */
116 delay = afs_probe_interval + lastCheck;
117 if (delay > afs_probe_all_interval + last10MinCheck)
118 delay = afs_probe_all_interval + last10MinCheck;
122 afs_osi_Wait(delay * 1000, &AFS_CSWaitHandler, 0);
124 afs_CheckServerDaemonStarted = 0;
127 extern int vfs_context_ref;
133 struct afs_exporter *exporter;
135 afs_int32 last3MinCheck, last10MinCheck, last60MinCheck, lastNMinCheck;
136 afs_int32 last1MinCheck;
137 afs_uint32 lastCBSlotBump;
140 AFS_STATCNT(afs_Daemon);
141 last1MinCheck = last3MinCheck = last60MinCheck = last10MinCheck =
144 afs_rootFid.Fid.Volume = 0;
145 while (afs_initState < 101)
146 afs_osi_Sleep(&afs_initState);
148 #ifdef AFS_DARWIN80_ENV
149 if (afs_osi_ctxtp_initialized)
150 osi_Panic("vfs context already initialized");
151 while (afs_osi_ctxtp && vfs_context_ref)
152 afs_osi_Sleep(&afs_osi_ctxtp);
153 if (afs_osi_ctxtp && !vfs_context_ref)
154 vfs_context_rele(afs_osi_ctxtp);
155 afs_osi_ctxtp = vfs_context_create(NULL);
156 afs_osi_ctxtp_initialized = 1;
159 lastCBSlotBump = now;
161 /* when a lot of clients are booted simultaneously, they develop
162 * annoying synchronous VL server bashing behaviors. So we stagger them.
164 last1MinCheck = now + ((afs_random() & 0x7fffffff) % 60); /* an extra 30 */
165 last3MinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
166 last60MinCheck = now - 1800 + ((afs_random() & 0x7fffffff) % 3600);
167 last10MinCheck = now - 300 + ((afs_random() & 0x7fffffff) % 600);
168 lastNMinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
170 /* start off with afs_initState >= 101 (basic init done) */
172 afs_CheckCallbacks(20); /* unstat anything which will expire soon */
174 /* things to do every 20 seconds or less - required by protocol spec */
176 afs_FlushActiveVcaches(0); /* flush NFS writes */
177 afs_FlushVCBs(1); /* flush queued callbacks */
178 afs_MaybeWakeupTruncateDaemon(); /* free cache space if have too */
179 rx_CheckPackets(); /* Does RX need more packets? */
182 if (lastCBSlotBump + CBHTSLOTLEN < now) { /* pretty time-dependant */
183 lastCBSlotBump = now;
184 if (afs_BumpBase()) {
185 afs_CheckCallbacks(20); /* unstat anything which will expire soon */
189 if (last1MinCheck + 60 < now) {
190 /* things to do every minute */
191 DFlush(); /* write out dir buffers */
192 afs_WriteThroughDSlots(); /* write through cacheinfo entries */
193 ObtainWriteLock(&afs_xvcache, 736);
194 afs_FlushReclaimedVcaches();
195 ReleaseWriteLock(&afs_xvcache);
196 afs_FlushActiveVcaches(1); /* keep flocks held & flush nfs writes */
197 #ifdef AFS_DISCON_ENV
198 afs_StoreDirtyVcaches();
204 if (last3MinCheck + 180 < now) {
205 afs_CheckTokenCache(); /* check for access cache resets due to expired
209 if (!afs_CheckServerDaemonStarted) {
210 /* Do the check here if the correct afsd is not installed. */
213 printf("Please install afsd with check server daemon.\n");
215 if (lastNMinCheck + afs_probe_interval < now) {
216 /* only check down servers */
217 afs_CheckServers(1, NULL);
221 if (last10MinCheck + 600 < now) {
222 #ifdef AFS_USERSPACE_IP_ADDR
223 extern int rxi_GetcbiInfo(void);
225 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, 600);
226 #ifdef AFS_USERSPACE_IP_ADDR
227 if (rxi_GetcbiInfo()) { /* addresses changed from last time */
230 #else /* AFS_USERSPACE_IP_ADDR */
231 if (rxi_GetIFInfo()) { /* addresses changed from last time */
234 #endif /* else AFS_USERSPACE_IP_ADDR */
235 if (!afs_CheckServerDaemonStarted)
236 afs_CheckServers(0, NULL);
237 afs_GCUserData(0); /* gc old conns */
238 /* This is probably the wrong way of doing GC for the various exporters but it will suffice for a while */
239 for (exporter = root_exported; exporter;
240 exporter = exporter->exp_next) {
241 (void)EXP_GC(exporter, 0); /* Generalize params */
246 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
250 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
255 last10MinCheck = now;
257 if (last60MinCheck + 3600 < now) {
258 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEVOLUME, ICL_TYPE_INT32,
260 afs_CheckRootVolume();
262 if (afs_gcpags == AFS_GCPAGS_OK) {
267 last60MinCheck = now;
269 if (afs_initState < 300) { /* while things ain't rosy */
270 code = afs_CheckRootVolume();
272 afs_initState = 300; /* succeeded */
273 if (afs_initState < 200)
274 afs_initState = 200; /* tried once */
275 afs_osi_Wakeup(&afs_initState);
278 /* 18285 is because we're trying to divide evenly into 128, that is,
279 * CBSlotLen, while staying just under 20 seconds. If CBSlotLen
280 * changes, should probably change this interval, too.
281 * Some of the preceding actions may take quite some time, so we
282 * might not want to wait the entire interval */
283 now = 18285 - (osi_Time() - now);
285 afs_osi_Wait(now, &AFS_WaitHandler, 0);
288 if (afs_termState == AFSOP_STOP_AFS) {
289 if (afs_CheckServerDaemonStarted)
290 afs_termState = AFSOP_STOP_CS;
292 afs_termState = AFSOP_STOP_BKG;
293 afs_osi_Wakeup(&afs_termState);
300 afs_CheckRootVolume(void)
302 char rootVolName[32];
303 struct volume *tvp = NULL;
304 int usingDynroot = afs_GetDynrootEnable();
307 AFS_STATCNT(afs_CheckRootVolume);
308 if (*afs_rootVolumeName == 0) {
309 strcpy(rootVolName, "root.afs");
311 strcpy(rootVolName, afs_rootVolumeName);
315 afs_GetDynrootFid(&afs_rootFid);
316 tvp = afs_GetVolume(&afs_rootFid, NULL, READ_LOCK);
318 struct cell *lc = afs_GetPrimaryCell(READ_LOCK);
322 localcell = lc->cellNum;
323 afs_PutCell(lc, READ_LOCK);
324 tvp = afs_GetVolumeByName(rootVolName, localcell, 1, NULL, READ_LOCK);
327 int len = strlen(rootVolName);
329 if ((len < 9) || strcmp(&rootVolName[len - 9], ".readonly")) {
330 strcpy(buf, rootVolName);
331 afs_strcat(buf, ".readonly");
332 tvp = afs_GetVolumeByName(buf, localcell, 1, NULL, READ_LOCK);
336 int volid = (tvp->roVol ? tvp->roVol : tvp->volume);
337 afs_rootFid.Cell = localcell;
338 if (afs_rootFid.Fid.Volume && afs_rootFid.Fid.Volume != volid
340 struct vcache *tvc = afs_globalVp;
341 /* If we had a root fid before and it changed location we reset
342 * the afs_globalVp so that it will be reevaluated.
343 * Just decrement the reference count. This only occurs during
344 * initial cell setup and can panic the machine if we set the
345 * count to zero and fs checkv is executed when the current
348 #ifdef AFS_LINUX20_ENV
350 struct vrequest treq;
356 afs_rootFid.Fid.Volume = volid;
357 afs_rootFid.Fid.Vnode = 1;
358 afs_rootFid.Fid.Unique = 1;
361 if (afs_InitReq(&treq, credp))
363 vcp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
366 afs_getattr(vcp, &vattr, credp);
367 afs_fill_inode(AFSTOV(vcp), &vattr);
369 dp = d_find_alias(AFSTOV(afs_globalVp));
371 #if defined(AFS_LINUX24_ENV)
372 spin_lock(&dcache_lock);
373 #if defined(AFS_LINUX26_ENV)
374 spin_lock(&dp->d_lock);
377 list_del_init(&dp->d_alias);
378 list_add(&dp->d_alias, &(AFSTOV(vcp)->i_dentry));
379 dp->d_inode = AFSTOV(vcp);
380 #if defined(AFS_LINUX24_ENV)
381 #if defined(AFS_LINUX26_ENV)
382 spin_unlock(&dp->d_lock);
384 spin_unlock(&dcache_lock);
388 AFS_FAST_RELE(afs_globalVp);
394 #ifdef AFS_DARWIN80_ENV
395 afs_PutVCache(afs_globalVp);
397 AFS_FAST_RELE(afs_globalVp);
402 afs_rootFid.Fid.Volume = volid;
403 afs_rootFid.Fid.Vnode = 1;
404 afs_rootFid.Fid.Unique = 1;
408 afs_initState = 300; /* won */
409 afs_osi_Wakeup(&afs_initState);
410 afs_PutVolume(tvp, READ_LOCK);
412 if (afs_rootFid.Fid.Volume)
418 /* ptr_parm 0 is the pathname, size_parm 0 to the fetch is the chunk number */
420 BPath(register struct brequest *ab)
422 register struct dcache *tdc = NULL;
423 struct vcache *tvc = NULL;
424 struct vnode *tvn = NULL;
425 #ifdef AFS_LINUX22_ENV
426 struct dentry *dp = NULL;
428 afs_size_t offset, len;
429 struct vrequest treq;
433 if ((code = afs_InitReq(&treq, ab->cred)))
436 #ifdef AFS_LINUX22_ENV
437 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &dp);
439 tvn = (struct vnode *)dp->d_inode;
441 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &tvn);
444 osi_FreeLargeSpace((char *)ab->ptr_parm[0]); /* free path name buffer here */
447 /* now path may not have been in afs, so check that before calling our cache manager */
448 if (!tvn || !IsAfsVnode(tvn)) {
449 /* release it and give up */
451 #ifdef AFS_LINUX22_ENV
460 /* here we know its an afs vnode, so we can get the data for the chunk */
461 tdc = afs_GetDCache(tvc, ab->size_parm[0], &treq, &offset, &len, 1);
465 #ifdef AFS_LINUX22_ENV
472 /* size_parm 0 to the fetch is the chunk number,
473 * ptr_parm 0 is the dcache entry to wakeup,
474 * size_parm 1 is true iff we should release the dcache entry here.
477 BPrefetch(register struct brequest *ab)
479 register struct dcache *tdc;
480 register struct vcache *tvc;
481 afs_size_t offset, len;
482 struct vrequest treq;
484 AFS_STATCNT(BPrefetch);
485 if ((len = afs_InitReq(&treq, ab->cred)))
488 tdc = afs_GetDCache(tvc, ab->size_parm[0], &treq, &offset, &len, 1);
492 /* now, dude may be waiting for us to clear DFFetchReq bit; do so. Can't
493 * use tdc from GetDCache since afs_GetDCache may fail, but someone may
494 * be waiting for our wakeup anyway.
496 tdc = (struct dcache *)(ab->ptr_parm[0]);
497 ObtainSharedLock(&tdc->lock, 640);
498 if (tdc->mflags & DFFetchReq) {
499 UpgradeSToWLock(&tdc->lock, 641);
500 tdc->mflags &= ~DFFetchReq;
501 ReleaseWriteLock(&tdc->lock);
503 ReleaseSharedLock(&tdc->lock);
505 afs_osi_Wakeup(&tdc->validPos);
506 if (ab->size_parm[1]) {
507 afs_PutDCache(tdc); /* put this one back, too */
513 BStore(register struct brequest *ab)
515 register struct vcache *tvc;
516 register afs_int32 code;
517 struct vrequest treq;
518 #if defined(AFS_SGI_ENV)
519 struct cred *tmpcred;
523 if ((code = afs_InitReq(&treq, ab->cred)))
527 #if defined(AFS_SGI_ENV)
529 * Since StoreOnLastReference can end up calling osi_SyncVM which
530 * calls into VM code that assumes that u.u_cred has the
531 * correct credentials, we set our to theirs for this xaction
533 tmpcred = OSI_GET_CURRENT_CRED();
534 OSI_SET_CURRENT_CRED(ab->cred);
537 * To avoid recursion since the WriteLock may be released during VM
538 * operations, we hold the VOP_RWLOCK across this transaction as
539 * do the other callers of StoreOnLastReference
541 AFS_RWLOCK((vnode_t *) tvc, 1);
543 ObtainWriteLock(&tvc->lock, 209);
544 code = afs_StoreOnLastReference(tvc, &treq);
545 ReleaseWriteLock(&tvc->lock);
546 #if defined(AFS_SGI_ENV)
547 OSI_SET_CURRENT_CRED(tmpcred);
548 AFS_RWUNLOCK((vnode_t *) tvc, 1);
550 /* now set final return code, and wakeup anyone waiting */
551 if ((ab->flags & BUVALID) == 0) {
552 ab->code = afs_CheckCode(code, &treq, 43); /* set final code, since treq doesn't go across processes */
553 ab->flags |= BUVALID;
554 if (ab->flags & BUWAIT) {
555 ab->flags &= ~BUWAIT;
561 /* release a held request buffer */
563 afs_BRelease(register struct brequest *ab)
566 AFS_STATCNT(afs_BRelease);
567 MObtainWriteLock(&afs_xbrs, 294);
568 if (--ab->refCount <= 0) {
572 afs_osi_Wakeup(&afs_brsWaiters);
573 MReleaseWriteLock(&afs_xbrs);
576 /* return true if bkg fetch daemons are all busy */
580 AFS_STATCNT(afs_BBusy);
581 if (afs_brsDaemons > 0)
587 afs_BQueue(register short aopcode, register struct vcache *avc,
588 afs_int32 dontwait, afs_int32 ause, struct AFS_UCRED *acred,
589 afs_size_t asparm0, afs_size_t asparm1, void *apparm0)
592 register struct brequest *tb;
594 AFS_STATCNT(afs_BQueue);
595 MObtainWriteLock(&afs_xbrs, 296);
598 for (i = 0; i < NBRS; i++, tb++) {
599 if (tb->refCount == 0)
604 tb->opcode = aopcode;
609 VN_HOLD(AFSTOV(avc));
611 tb->refCount = ause + 1;
612 tb->size_parm[0] = asparm0;
613 tb->size_parm[1] = asparm1;
614 tb->ptr_parm[0] = apparm0;
617 tb->ts = afs_brs_count++;
618 /* if daemons are waiting for work, wake them up */
619 if (afs_brsDaemons > 0) {
620 afs_osi_Wakeup(&afs_brsDaemons);
622 MReleaseWriteLock(&afs_xbrs);
626 MReleaseWriteLock(&afs_xbrs);
629 /* no free buffers, sleep a while */
631 MReleaseWriteLock(&afs_xbrs);
632 afs_osi_Sleep(&afs_brsWaiters);
633 MObtainWriteLock(&afs_xbrs, 301);
639 /* AIX 4.1 has a much different sleep/wakeup mechanism available for use.
640 * The modifications here will work for either a UP or MP machine.
642 struct buf *afs_asyncbuf = (struct buf *)0;
643 tid_t afs_asyncbuf_cv = EVENT_NULL;
644 afs_int32 afs_biodcnt = 0;
646 /* in implementing this, I assumed that all external linked lists were
649 * Several places in this code traverse a linked list. The algorithm
650 * used here is probably unfamiliar to most people. Careful examination
651 * will show that it eliminates an assignment inside the loop, as compared
652 * to the standard algorithm, at the cost of occasionally using an extra
658 * This function obtains, and returns, a pointer to a buffer for
659 * processing by a daemon. It sleeps until such a buffer is available.
660 * The source of buffers for it is the list afs_asyncbuf (see also
661 * afs_gn_strategy). This function may be invoked concurrently by
662 * several processes, that is, several instances of the same daemon.
663 * afs_gn_strategy, which adds buffers to the list, runs at interrupt
664 * level, while get_bioreq runs at process level.
666 * Since AIX 4.1 can wake just one process at a time, the separate sleep
667 * addresses have been removed.
668 * Note that the kernel_lock is held until the e_sleep_thread() occurs.
669 * The afs_asyncbuf_lock is primarily used to serialize access between
670 * process and interrupts.
672 Simple_lock afs_asyncbuf_lock;
676 struct buf *bp = NULL;
678 struct buf **bestlbpP, **lbpP;
680 struct buf *t1P, *t2P; /* temp pointers for list manipulation */
683 struct afs_bioqueue *s;
685 /* ??? Does the forward pointer of the returned buffer need to be NULL?
688 /* Disable interrupts from the strategy function, and save the
689 * prior priority level and lock access to the afs_asyncbuf.
692 oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
696 /* look for oldest buffer */
697 bp = bestbp = afs_asyncbuf;
698 bestage = (long)bestbp->av_back;
699 bestlbpP = &afs_asyncbuf;
705 if ((long)bp->av_back - bestage < 0) {
708 bestage = (long)bp->av_back;
712 *bestlbpP = bp->av_forw;
715 /* If afs_asyncbuf is null, it is necessary to go to sleep.
716 * e_wakeup_one() ensures that only one thread wakes.
719 /* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
720 * lock on an MP machine.
723 e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
724 LOCK_HANDLER | INTERRUPTIBLE);
725 if (interrupted == THREAD_INTERRUPTED) {
726 /* re-enable interrupts from strategy */
727 unlock_enable(oldPriority, &afs_asyncbuf_lock);
731 } /* end of "else asyncbuf is empty" */
732 } /* end of "inner loop" */
736 unlock_enable(oldPriority, &afs_asyncbuf_lock);
739 /* For the convenience of other code, replace the gnodes in
740 * the b_vp field of bp and the other buffers on the b_work
741 * chain with the corresponding vnodes.
743 * ??? what happens to the gnodes? They're not just cut loose,
747 t2P = (struct buf *)t1P->b_work;
748 t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
752 t1P = (struct buf *)t2P->b_work;
753 t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
758 /* If the buffer does not specify I/O, it may immediately
759 * be returned to the caller. This condition is detected
760 * by examining the buffer's flags (the b_flags field). If
761 * the B_PFPROT bit is set, the buffer represents a protection
762 * violation, rather than a request for I/O. The remainder
763 * of the outer loop handles the case where the B_PFPROT bit is clear.
765 if (bp->b_flags & B_PFPROT) {
770 } /* end of function get_bioreq() */
775 * This function is the daemon. It is called from the syscall
776 * interface. Ordinarily, a script or an administrator will run a
777 * daemon startup utility, specifying the number of I/O daemons to
778 * run. The utility will fork off that number of processes,
779 * each making the appropriate syscall, which will cause this
780 * function to be invoked.
782 static int afs_initbiod = 0; /* this is self-initializing code */
785 afs_BioDaemon(afs_int32 nbiods)
787 afs_int32 code, s, pflg = 0;
789 struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
797 /* pin lock, since we'll be using it in an interrupt. */
798 lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
799 simple_lock_init(&afs_asyncbuf_lock);
800 pin(&afs_asyncbuf, sizeof(struct buf *));
801 pin(&afs_asyncbuf_cv, sizeof(afs_int32));
804 /* Ignore HUP signals... */
806 sigset_t sigbits, osigbits;
808 * add SIGHUP to the set of already masked signals
810 SIGFILLSET(sigbits); /* allow all signals */
811 SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
812 limit_sigs(&sigbits, &osigbits); /* and already masked */
814 /* Main body starts here -- this is an intentional infinite loop, and
817 * Now, the loop will exit if get_bioreq() returns NULL, indicating
818 * that we've been interrupted.
821 bp = afs_get_bioreq();
823 break; /* we were interrupted */
824 if (code = setjmpx(&jmpbuf)) {
825 /* This should not have happend, maybe a lack of resources */
827 s = disable_lock(INTMAX, &afs_asyncbuf_lock);
828 for (bp1 = bp; bp; bp = bp1) {
830 bp1 = (struct buf *)bp1->b_work;
833 bp->b_flags |= B_ERROR;
836 unlock_enable(s, &afs_asyncbuf_lock);
840 vcp = VTOAFS(bp->b_vp);
841 if (bp->b_flags & B_PFSTORE) { /* XXXX */
842 ObtainWriteLock(&vcp->lock, 404);
843 if (vcp->v.v_gnode->gn_mwrcnt) {
844 afs_offs_t newlength =
845 (afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
846 if (vcp->m.Length < newlength) {
847 afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
848 ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
849 __LINE__, ICL_TYPE_OFFSET,
850 ICL_HANDLE_OFFSET(vcp->m.Length),
851 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
852 vcp->m.Length = newlength;
855 ReleaseWriteLock(&vcp->lock);
857 /* If the buffer represents a protection violation, rather than
858 * an actual request for I/O, no special action need be taken.
860 if (bp->b_flags & B_PFPROT) {
861 iodone(bp); /* Notify all users of the buffer that we're done */
866 ObtainWriteLock(&vcp->pvmlock, 211);
868 * First map its data area to a region in the current address space
869 * by calling vm_att with the subspace identifier, and a pointer to
870 * the data area. vm_att returns a new data area pointer, but we
871 * also want to hang onto the old one.
873 tmpaddr = bp->b_baddr;
874 bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
875 tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
876 if (tmperr) { /* in non-error case */
877 bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
878 bp->b_error = tmperr;
881 /* Unmap the buffer's data area by calling vm_det. Reset data area
882 * to the value that we saved above.
885 bp->b_baddr = tmpaddr;
888 * buffer may be linked with other buffers via the b_work field.
889 * See also afs_gn_strategy. For each buffer in the chain (including
890 * bp) notify all users of the buffer that the daemon is finished
891 * using it by calling iodone.
892 * assumes iodone can modify the b_work field.
895 tbp2 = (struct buf *)tbp1->b_work;
900 tbp1 = (struct buf *)tbp2->b_work;
906 ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
908 } /* infinite loop (unless we're interrupted) */
909 } /* end of afs_BioDaemon() */
911 #endif /* AFS_AIX41_ENV */
916 afs_BackgroundDaemon(void)
921 AFS_STATCNT(afs_BackgroundDaemon);
922 /* initialize subsystem */
924 LOCK_INIT(&afs_xbrs, "afs_xbrs");
925 memset((char *)afs_brs, 0, sizeof(afs_brs));
927 #if defined (AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
929 * steal the first daemon for doing delayed DSlot flushing
930 * (see afs_GetDownDSlot)
939 MObtainWriteLock(&afs_xbrs, 302);
942 struct brequest *min_tb = NULL;
944 if (afs_termState == AFSOP_STOP_BKG) {
946 afs_termState = AFSOP_STOP_TRUNCDAEMON;
947 MReleaseWriteLock(&afs_xbrs);
948 afs_osi_Wakeup(&afs_termState);
955 for (i = 0; i < NBRS; i++, tb++) {
956 /* look for request with smallest ts */
957 if ((tb->refCount > 0) && !(tb->flags & BSTARTED)) {
958 /* new request, not yet picked up */
959 if ((min_tb && (min_ts - tb->ts > 0)) || !min_tb) {
966 /* claim and process this request */
967 tb->flags |= BSTARTED;
968 MReleaseWriteLock(&afs_xbrs);
970 afs_Trace1(afs_iclSetp, CM_TRACE_BKG1, ICL_TYPE_INT32,
972 if (tb->opcode == BOP_FETCH)
974 else if (tb->opcode == BOP_STORE)
976 else if (tb->opcode == BOP_PATH)
979 panic("background bop");
981 AFS_RELE(AFSTOV(tb->vc)); /* MUST call vnode layer or could lose vnodes */
986 tb->cred = (struct AFS_UCRED *)0;
988 afs_BRelease(tb); /* this grabs and releases afs_xbrs lock */
989 MObtainWriteLock(&afs_xbrs, 305);
992 /* wait for new request */
994 MReleaseWriteLock(&afs_xbrs);
995 afs_osi_Sleep(&afs_brsDaemons);
996 MObtainWriteLock(&afs_xbrs, 307);
1004 shutdown_daemons(void)
1006 AFS_STATCNT(shutdown_daemons);
1007 if (afs_cold_shutdown) {
1008 afs_brsDaemons = brsInit = 0;
1009 rxepoch_checked = afs_nbrs = 0;
1010 memset((char *)afs_brs, 0, sizeof(afs_brs));
1011 memset((char *)&afs_xbrs, 0, sizeof(afs_lock_t));
1013 #ifdef AFS_AIX41_ENV
1014 lock_free(&afs_asyncbuf_lock);
1015 unpin(&afs_asyncbuf, sizeof(struct buf *));
1016 unpin(&afs_asyncbuf_cv, sizeof(afs_int32));
1022 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1024 * sgi - daemon - handles certain operations that otherwise
1025 * would use up too much kernel stack space
1027 * This all assumes that since the caller must have the xdcache lock
1028 * exclusively that the list will never be more than one long
1029 * and noone else can attempt to add anything until we're done.
1031 SV_TYPE afs_sgibksync;
1032 SV_TYPE afs_sgibkwait;
1033 lock_t afs_sgibklock;
1034 struct dcache *afs_sgibklist;
1042 if (afs_sgibklock == NULL) {
1043 SV_INIT(&afs_sgibksync, "bksync", 0, 0);
1044 SV_INIT(&afs_sgibkwait, "bkwait", 0, 0);
1045 SPINLOCK_INIT(&afs_sgibklock, "bklock");
1047 s = SPLOCK(afs_sgibklock);
1049 /* wait for something to do */
1050 SP_WAIT(afs_sgibklock, s, &afs_sgibksync, PINOD);
1051 osi_Assert(afs_sgibklist);
1053 /* XX will probably need to generalize to real list someday */
1054 s = SPLOCK(afs_sgibklock);
1055 while (afs_sgibklist) {
1056 tdc = afs_sgibklist;
1057 afs_sgibklist = NULL;
1058 SPUNLOCK(afs_sgibklock, s);
1060 tdc->dflags &= ~DFEntryMod;
1061 afs_WriteDCache(tdc, 1);
1063 s = SPLOCK(afs_sgibklock);
1066 /* done all the work - wake everyone up */
1067 while (SV_SIGNAL(&afs_sgibkwait));