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 afs_warn("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)
540 BPrefetchNoCache(register struct brequest *ab)
542 struct vrequest treq;
545 if ((len = afs_InitReq(&treq, ab->cred)))
549 /* OS-specific prefetch routine */
550 afs_PrefetchNoCache(ab->vc, ab->cred, (struct nocache_read_request *) ab->ptr_parm[0]);
556 BStore(register struct brequest *ab)
558 register struct vcache *tvc;
559 register afs_int32 code;
560 struct vrequest treq;
561 #if defined(AFS_SGI_ENV)
562 struct cred *tmpcred;
566 if ((code = afs_InitReq(&treq, ab->cred)))
570 #if defined(AFS_SGI_ENV)
572 * Since StoreOnLastReference can end up calling osi_SyncVM which
573 * calls into VM code that assumes that u.u_cred has the
574 * correct credentials, we set our to theirs for this xaction
576 tmpcred = OSI_GET_CURRENT_CRED();
577 OSI_SET_CURRENT_CRED(ab->cred);
580 * To avoid recursion since the WriteLock may be released during VM
581 * operations, we hold the VOP_RWLOCK across this transaction as
582 * do the other callers of StoreOnLastReference
584 AFS_RWLOCK((vnode_t *) tvc, 1);
586 ObtainWriteLock(&tvc->lock, 209);
587 code = afs_StoreOnLastReference(tvc, &treq);
588 ReleaseWriteLock(&tvc->lock);
589 #if defined(AFS_SGI_ENV)
590 OSI_SET_CURRENT_CRED(tmpcred);
591 AFS_RWUNLOCK((vnode_t *) tvc, 1);
593 /* now set final return code, and wakeup anyone waiting */
594 if ((ab->flags & BUVALID) == 0) {
595 ab->code = afs_CheckCode(code, &treq, 43); /* set final code, since treq doesn't go across processes */
596 ab->flags |= BUVALID;
597 if (ab->flags & BUWAIT) {
598 ab->flags &= ~BUWAIT;
604 /* release a held request buffer */
606 afs_BRelease(register struct brequest *ab)
609 AFS_STATCNT(afs_BRelease);
610 ObtainWriteLock(&afs_xbrs, 294);
611 if (--ab->refCount <= 0) {
615 afs_osi_Wakeup(&afs_brsWaiters);
616 ReleaseWriteLock(&afs_xbrs);
619 /* return true if bkg fetch daemons are all busy */
623 AFS_STATCNT(afs_BBusy);
624 if (afs_brsDaemons > 0)
630 afs_BQueue(register short aopcode, register struct vcache *avc,
631 afs_int32 dontwait, afs_int32 ause, afs_ucred_t *acred,
632 afs_size_t asparm0, afs_size_t asparm1, void *apparm0,
633 void *apparm1, void *apparm2)
636 register struct brequest *tb;
638 AFS_STATCNT(afs_BQueue);
639 ObtainWriteLock(&afs_xbrs, 296);
642 for (i = 0; i < NBRS; i++, tb++) {
643 if (tb->refCount == 0)
648 tb->opcode = aopcode;
655 tb->refCount = ause + 1;
656 tb->size_parm[0] = asparm0;
657 tb->size_parm[1] = asparm1;
658 tb->ptr_parm[0] = apparm0;
659 tb->ptr_parm[1] = apparm1;
660 tb->ptr_parm[2] = apparm2;
663 tb->ts = afs_brs_count++;
664 /* if daemons are waiting for work, wake them up */
665 if (afs_brsDaemons > 0) {
666 afs_osi_Wakeup(&afs_brsDaemons);
668 ReleaseWriteLock(&afs_xbrs);
672 ReleaseWriteLock(&afs_xbrs);
675 /* no free buffers, sleep a while */
677 ReleaseWriteLock(&afs_xbrs);
678 afs_osi_Sleep(&afs_brsWaiters);
679 ObtainWriteLock(&afs_xbrs, 301);
685 /* AIX 4.1 has a much different sleep/wakeup mechanism available for use.
686 * The modifications here will work for either a UP or MP machine.
688 struct buf *afs_asyncbuf = (struct buf *)0;
689 tid_t afs_asyncbuf_cv = EVENT_NULL;
690 afs_int32 afs_biodcnt = 0;
692 /* in implementing this, I assumed that all external linked lists were
695 * Several places in this code traverse a linked list. The algorithm
696 * used here is probably unfamiliar to most people. Careful examination
697 * will show that it eliminates an assignment inside the loop, as compared
698 * to the standard algorithm, at the cost of occasionally using an extra
704 * This function obtains, and returns, a pointer to a buffer for
705 * processing by a daemon. It sleeps until such a buffer is available.
706 * The source of buffers for it is the list afs_asyncbuf (see also
707 * afs_gn_strategy). This function may be invoked concurrently by
708 * several processes, that is, several instances of the same daemon.
709 * afs_gn_strategy, which adds buffers to the list, runs at interrupt
710 * level, while get_bioreq runs at process level.
712 * Since AIX 4.1 can wake just one process at a time, the separate sleep
713 * addresses have been removed.
714 * Note that the kernel_lock is held until the e_sleep_thread() occurs.
715 * The afs_asyncbuf_lock is primarily used to serialize access between
716 * process and interrupts.
718 Simple_lock afs_asyncbuf_lock;
722 struct buf *bp = NULL;
724 struct buf **bestlbpP, **lbpP;
726 struct buf *t1P, *t2P; /* temp pointers for list manipulation */
729 struct afs_bioqueue *s;
731 /* ??? Does the forward pointer of the returned buffer need to be NULL?
734 /* Disable interrupts from the strategy function, and save the
735 * prior priority level and lock access to the afs_asyncbuf.
738 oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
742 /* look for oldest buffer */
743 bp = bestbp = afs_asyncbuf;
744 bestage = (long)bestbp->av_back;
745 bestlbpP = &afs_asyncbuf;
751 if ((long)bp->av_back - bestage < 0) {
754 bestage = (long)bp->av_back;
758 *bestlbpP = bp->av_forw;
761 /* If afs_asyncbuf is null, it is necessary to go to sleep.
762 * e_wakeup_one() ensures that only one thread wakes.
765 /* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
766 * lock on an MP machine.
769 e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
770 LOCK_HANDLER | INTERRUPTIBLE);
771 if (interrupted == THREAD_INTERRUPTED) {
772 /* re-enable interrupts from strategy */
773 unlock_enable(oldPriority, &afs_asyncbuf_lock);
777 } /* end of "else asyncbuf is empty" */
778 } /* end of "inner loop" */
782 unlock_enable(oldPriority, &afs_asyncbuf_lock);
785 /* For the convenience of other code, replace the gnodes in
786 * the b_vp field of bp and the other buffers on the b_work
787 * chain with the corresponding vnodes.
789 * ??? what happens to the gnodes? They're not just cut loose,
793 t2P = (struct buf *)t1P->b_work;
794 t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
798 t1P = (struct buf *)t2P->b_work;
799 t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
804 /* If the buffer does not specify I/O, it may immediately
805 * be returned to the caller. This condition is detected
806 * by examining the buffer's flags (the b_flags field). If
807 * the B_PFPROT bit is set, the buffer represents a protection
808 * violation, rather than a request for I/O. The remainder
809 * of the outer loop handles the case where the B_PFPROT bit is clear.
811 if (bp->b_flags & B_PFPROT) {
816 } /* end of function get_bioreq() */
821 * This function is the daemon. It is called from the syscall
822 * interface. Ordinarily, a script or an administrator will run a
823 * daemon startup utility, specifying the number of I/O daemons to
824 * run. The utility will fork off that number of processes,
825 * each making the appropriate syscall, which will cause this
826 * function to be invoked.
828 static int afs_initbiod = 0; /* this is self-initializing code */
831 afs_BioDaemon(afs_int32 nbiods)
833 afs_int32 code, s, pflg = 0;
835 struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
843 /* pin lock, since we'll be using it in an interrupt. */
844 lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
845 simple_lock_init(&afs_asyncbuf_lock);
846 pin(&afs_asyncbuf, sizeof(struct buf *));
847 pin(&afs_asyncbuf_cv, sizeof(afs_int32));
850 /* Ignore HUP signals... */
852 sigset_t sigbits, osigbits;
854 * add SIGHUP to the set of already masked signals
856 SIGFILLSET(sigbits); /* allow all signals */
857 SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
858 limit_sigs(&sigbits, &osigbits); /* and already masked */
860 /* Main body starts here -- this is an intentional infinite loop, and
863 * Now, the loop will exit if get_bioreq() returns NULL, indicating
864 * that we've been interrupted.
867 bp = afs_get_bioreq();
869 break; /* we were interrupted */
870 if (code = setjmpx(&jmpbuf)) {
871 /* This should not have happend, maybe a lack of resources */
873 s = disable_lock(INTMAX, &afs_asyncbuf_lock);
874 for (bp1 = bp; bp; bp = bp1) {
876 bp1 = (struct buf *)bp1->b_work;
879 bp->b_flags |= B_ERROR;
882 unlock_enable(s, &afs_asyncbuf_lock);
886 vcp = VTOAFS(bp->b_vp);
887 if (bp->b_flags & B_PFSTORE) { /* XXXX */
888 ObtainWriteLock(&vcp->lock, 404);
889 if (vcp->v.v_gnode->gn_mwrcnt) {
890 afs_offs_t newlength =
891 (afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
892 if (vcp->f.m.Length < newlength) {
893 afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
894 ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
895 __LINE__, ICL_TYPE_OFFSET,
896 ICL_HANDLE_OFFSET(vcp->f.m.Length),
897 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
898 vcp->f.m.Length = newlength;
901 ReleaseWriteLock(&vcp->lock);
903 /* If the buffer represents a protection violation, rather than
904 * an actual request for I/O, no special action need be taken.
906 if (bp->b_flags & B_PFPROT) {
907 iodone(bp); /* Notify all users of the buffer that we're done */
912 ObtainWriteLock(&vcp->pvmlock, 211);
914 * First map its data area to a region in the current address space
915 * by calling vm_att with the subspace identifier, and a pointer to
916 * the data area. vm_att returns a new data area pointer, but we
917 * also want to hang onto the old one.
919 tmpaddr = bp->b_baddr;
920 bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
921 tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
922 if (tmperr) { /* in non-error case */
923 bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
924 bp->b_error = tmperr;
927 /* Unmap the buffer's data area by calling vm_det. Reset data area
928 * to the value that we saved above.
931 bp->b_baddr = tmpaddr;
934 * buffer may be linked with other buffers via the b_work field.
935 * See also afs_gn_strategy. For each buffer in the chain (including
936 * bp) notify all users of the buffer that the daemon is finished
937 * using it by calling iodone.
938 * assumes iodone can modify the b_work field.
941 tbp2 = (struct buf *)tbp1->b_work;
946 tbp1 = (struct buf *)tbp2->b_work;
952 ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
954 } /* infinite loop (unless we're interrupted) */
955 } /* end of afs_BioDaemon() */
957 #endif /* AFS_AIX41_ENV */
962 afs_BackgroundDaemon_once(void)
964 LOCK_INIT(&afs_xbrs, "afs_xbrs");
965 memset(afs_brs, 0, sizeof(afs_brs));
967 #if defined (AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
969 * steal the first daemon for doing delayed DSlot flushing
970 * (see afs_GetDownDSlot)
979 brequest_release(struct brequest *tb)
982 AFS_RELE(AFSTOV(tb->vc)); /* MUST call vnode layer or could lose vnodes */
987 tb->cred = (afs_ucred_t *)0;
989 afs_BRelease(tb); /* this grabs and releases afs_xbrs lock */
992 #ifdef AFS_DARWIN80_ENV
994 afs_BackgroundDaemon(struct afs_uspc_param *uspc, void *param1, void *param2)
997 afs_BackgroundDaemon(void)
1000 struct brequest *tb;
1003 AFS_STATCNT(afs_BackgroundDaemon);
1004 /* initialize subsystem */
1006 /* Irix with "short stack" exits */
1007 afs_BackgroundDaemon_once();
1009 #ifdef AFS_DARWIN80_ENV
1010 /* If it's a re-entering syscall, complete the request and release */
1011 if (uspc->ts > -1) {
1013 for (i = 0; i < NBRS; i++, tb++) {
1014 if (tb->ts == uspc->ts) {
1015 /* copy the userspace status back in */
1016 ((struct afs_uspc_param *) tb->ptr_parm[0])->retval =
1018 /* mark it valid and notify our caller */
1019 tb->flags |= BUVALID;
1020 if (tb->flags & BUWAIT) {
1021 tb->flags &= ~BUWAIT;
1024 brequest_release(tb);
1029 afs_osi_MaskUserLoop();
1031 /* Otherwise it's a new one */
1033 #ifdef AFS_DARWIN80_ENV
1037 ObtainWriteLock(&afs_xbrs, 302);
1040 struct brequest *min_tb = NULL;
1042 if (afs_termState == AFSOP_STOP_BKG) {
1043 if (--afs_nbrs <= 0)
1044 afs_termState = AFSOP_STOP_TRUNCDAEMON;
1045 ReleaseWriteLock(&afs_xbrs);
1046 afs_osi_Wakeup(&afs_termState);
1047 #ifdef AFS_DARWIN80_ENV
1054 /* find a request */
1057 for (i = 0; i < NBRS; i++, tb++) {
1058 /* look for request with smallest ts */
1059 if ((tb->refCount > 0) && !(tb->flags & BSTARTED)) {
1060 /* new request, not yet picked up */
1061 if ((min_tb && (min_ts - tb->ts > 0)) || !min_tb) {
1067 if ((tb = min_tb)) {
1068 /* claim and process this request */
1069 tb->flags |= BSTARTED;
1070 ReleaseWriteLock(&afs_xbrs);
1072 afs_Trace1(afs_iclSetp, CM_TRACE_BKG1, ICL_TYPE_INT32,
1074 if (tb->opcode == BOP_FETCH)
1076 #if defined(AFS_CACHE_BYPASS)
1077 else if (tb->opcode == BOP_FETCH_NOCACHE)
1078 BPrefetchNoCache(tb);
1080 else if (tb->opcode == BOP_STORE)
1082 else if (tb->opcode == BOP_PATH)
1084 #ifdef AFS_DARWIN80_ENV
1085 else if (tb->opcode == BOP_MOVE) {
1086 memcpy(uspc, (struct afs_uspc_param *) tb->ptr_parm[0],
1087 sizeof(struct afs_uspc_param));
1089 /* string lengths capped in move vop; copy NUL tho */
1090 memcpy(param1, (char *)tb->ptr_parm[1],
1091 strlen(tb->ptr_parm[1])+1);
1092 memcpy(param2, (char *)tb->ptr_parm[2],
1093 strlen(tb->ptr_parm[2])+1);
1098 panic("background bop");
1099 brequest_release(tb);
1100 ObtainWriteLock(&afs_xbrs, 305);
1103 /* wait for new request */
1105 ReleaseWriteLock(&afs_xbrs);
1106 afs_osi_Sleep(&afs_brsDaemons);
1107 ObtainWriteLock(&afs_xbrs, 307);
1111 #ifdef AFS_DARWIN80_ENV
1118 shutdown_daemons(void)
1120 AFS_STATCNT(shutdown_daemons);
1121 if (afs_cold_shutdown) {
1122 afs_brsDaemons = brsInit = 0;
1123 rxepoch_checked = afs_nbrs = 0;
1124 memset(afs_brs, 0, sizeof(afs_brs));
1125 memset(&afs_xbrs, 0, sizeof(afs_lock_t));
1127 #ifdef AFS_AIX41_ENV
1128 lock_free(&afs_asyncbuf_lock);
1129 unpin(&afs_asyncbuf, sizeof(struct buf *));
1130 unpin(&afs_asyncbuf_cv, sizeof(afs_int32));
1136 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1138 * sgi - daemon - handles certain operations that otherwise
1139 * would use up too much kernel stack space
1141 * This all assumes that since the caller must have the xdcache lock
1142 * exclusively that the list will never be more than one long
1143 * and noone else can attempt to add anything until we're done.
1145 SV_TYPE afs_sgibksync;
1146 SV_TYPE afs_sgibkwait;
1147 lock_t afs_sgibklock;
1148 struct dcache *afs_sgibklist;
1156 if (afs_sgibklock == NULL) {
1157 SV_INIT(&afs_sgibksync, "bksync", 0, 0);
1158 SV_INIT(&afs_sgibkwait, "bkwait", 0, 0);
1159 SPINLOCK_INIT(&afs_sgibklock, "bklock");
1161 s = SPLOCK(afs_sgibklock);
1163 /* wait for something to do */
1164 SP_WAIT(afs_sgibklock, s, &afs_sgibksync, PINOD);
1165 osi_Assert(afs_sgibklist);
1167 /* XX will probably need to generalize to real list someday */
1168 s = SPLOCK(afs_sgibklock);
1169 while (afs_sgibklist) {
1170 tdc = afs_sgibklist;
1171 afs_sgibklist = NULL;
1172 SPUNLOCK(afs_sgibklock, s);
1174 tdc->dflags &= ~DFEntryMod;
1175 afs_WriteDCache(tdc, 1);
1177 s = SPLOCK(afs_sgibklock);
1180 /* done all the work - wake everyone up */
1181 while (SV_SIGNAL(&afs_sgibkwait));