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 afs_StoreDirtyVcaches();
210 if (last3MinCheck + 180 < now) {
211 afs_CheckTokenCache(); /* check for access cache resets due to expired
216 if (afsd_dynamic_vcaches && (last5MinCheck + 300 < now)) {
217 /* start with trying to drop us back to our base usage */
219 if (afs_maxvcount <= afs_cacheStats)
220 anumber = VCACHE_FREE;
222 anumber = VCACHE_FREE + (afs_maxvcount - afs_cacheStats);
224 ObtainWriteLock(&afs_xvcache, 734);
225 afs_ShakeLooseVCaches(anumber);
226 ReleaseWriteLock(&afs_xvcache);
230 if (!afs_CheckServerDaemonStarted) {
231 /* Do the check here if the correct afsd is not installed. */
234 afs_warn("Please install afsd with check server daemon.\n");
236 if (lastNMinCheck + afs_probe_interval < now) {
237 /* only check down servers */
238 afs_CheckServers(1, NULL);
242 if (last10MinCheck + 600 < now) {
243 #ifdef AFS_USERSPACE_IP_ADDR
244 extern int rxi_GetcbiInfo(void);
246 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, 600);
247 #ifdef AFS_USERSPACE_IP_ADDR
248 if (rxi_GetcbiInfo()) { /* addresses changed from last time */
251 #else /* AFS_USERSPACE_IP_ADDR */
252 if (rxi_GetIFInfo()) { /* addresses changed from last time */
255 #endif /* else AFS_USERSPACE_IP_ADDR */
256 if (!afs_CheckServerDaemonStarted)
257 afs_CheckServers(0, NULL);
258 afs_GCUserData(0); /* gc old conns */
259 /* This is probably the wrong way of doing GC for the various exporters but it will suffice for a while */
260 for (exporter = root_exported; exporter;
261 exporter = exporter->exp_next) {
262 (void)EXP_GC(exporter, 0); /* Generalize params */
267 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
271 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
276 last10MinCheck = now;
278 if (last60MinCheck + 3600 < now) {
279 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEVOLUME, ICL_TYPE_INT32,
281 afs_CheckRootVolume();
283 if (afs_gcpags == AFS_GCPAGS_OK) {
288 last60MinCheck = now;
290 if (afs_initState < 300) { /* while things ain't rosy */
291 code = afs_CheckRootVolume();
293 afs_initState = 300; /* succeeded */
294 if (afs_initState < 200)
295 afs_initState = 200; /* tried once */
296 afs_osi_Wakeup(&afs_initState);
299 /* 18285 is because we're trying to divide evenly into 128, that is,
300 * CBSlotLen, while staying just under 20 seconds. If CBSlotLen
301 * changes, should probably change this interval, too.
302 * Some of the preceding actions may take quite some time, so we
303 * might not want to wait the entire interval */
304 now = 18285 - (osi_Time() - now);
306 afs_osi_Wait(now, &AFS_WaitHandler, 0);
309 if (afs_termState == AFSOP_STOP_AFS) {
310 if (afs_CheckServerDaemonStarted)
311 afs_termState = AFSOP_STOP_CS;
313 afs_termState = AFSOP_STOP_BKG;
314 afs_osi_Wakeup(&afs_termState);
321 afs_CheckRootVolume(void)
323 char rootVolName[32];
324 struct volume *tvp = NULL;
325 int usingDynroot = afs_GetDynrootEnable();
328 AFS_STATCNT(afs_CheckRootVolume);
329 if (*afs_rootVolumeName == 0) {
330 strcpy(rootVolName, "root.afs");
332 strcpy(rootVolName, afs_rootVolumeName);
336 afs_GetDynrootFid(&afs_rootFid);
337 tvp = afs_GetVolume(&afs_rootFid, NULL, READ_LOCK);
339 struct cell *lc = afs_GetPrimaryCell(READ_LOCK);
343 localcell = lc->cellNum;
344 afs_PutCell(lc, READ_LOCK);
345 tvp = afs_GetVolumeByName(rootVolName, localcell, 1, NULL, READ_LOCK);
348 int len = strlen(rootVolName);
350 if ((len < 9) || strcmp(&rootVolName[len - 9], ".readonly")) {
351 strcpy(buf, rootVolName);
352 afs_strcat(buf, ".readonly");
353 tvp = afs_GetVolumeByName(buf, localcell, 1, NULL, READ_LOCK);
357 int volid = (tvp->roVol ? tvp->roVol : tvp->volume);
358 afs_rootFid.Cell = localcell;
359 if (afs_rootFid.Fid.Volume && afs_rootFid.Fid.Volume != volid
361 /* If we had a root fid before and it changed location we reset
362 * the afs_globalVp so that it will be reevaluated.
363 * Just decrement the reference count. This only occurs during
364 * initial cell setup and can panic the machine if we set the
365 * count to zero and fs checkv is executed when the current
368 #ifdef AFS_LINUX20_ENV
370 struct vrequest treq;
376 afs_rootFid.Fid.Volume = volid;
377 afs_rootFid.Fid.Vnode = 1;
378 afs_rootFid.Fid.Unique = 1;
381 if (afs_InitReq(&treq, credp))
383 vcp = afs_GetVCache(&afs_rootFid, &treq, NULL, NULL);
386 afs_getattr(vcp, &vattr, credp);
387 afs_fill_inode(AFSTOV(vcp), &vattr);
389 dp = d_find_alias(AFSTOV(afs_globalVp));
391 #if defined(AFS_LINUX24_ENV)
392 spin_lock(&dcache_lock);
393 #if defined(AFS_LINUX26_ENV)
394 spin_lock(&dp->d_lock);
397 list_del_init(&dp->d_alias);
398 list_add(&dp->d_alias, &(AFSTOV(vcp)->i_dentry));
399 dp->d_inode = AFSTOV(vcp);
400 #if defined(AFS_LINUX24_ENV)
401 #if defined(AFS_LINUX26_ENV)
402 spin_unlock(&dp->d_lock);
404 spin_unlock(&dcache_lock);
408 AFS_FAST_RELE(afs_globalVp);
414 #ifdef AFS_DARWIN80_ENV
415 afs_PutVCache(afs_globalVp);
417 AFS_FAST_RELE(afs_globalVp);
422 afs_rootFid.Fid.Volume = volid;
423 afs_rootFid.Fid.Vnode = 1;
424 afs_rootFid.Fid.Unique = 1;
428 afs_initState = 300; /* won */
429 afs_osi_Wakeup(&afs_initState);
430 afs_PutVolume(tvp, READ_LOCK);
432 if (afs_rootFid.Fid.Volume)
438 /* ptr_parm 0 is the pathname, size_parm 0 to the fetch is the chunk number */
440 BPath(register struct brequest *ab)
442 register struct dcache *tdc = NULL;
443 struct vcache *tvc = NULL;
444 struct vnode *tvn = NULL;
445 #ifdef AFS_LINUX22_ENV
446 struct dentry *dp = NULL;
448 afs_size_t offset, len;
449 struct vrequest treq;
453 if ((code = afs_InitReq(&treq, ab->cred)))
456 #ifdef AFS_LINUX22_ENV
457 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &dp);
459 tvn = (struct vnode *)dp->d_inode;
461 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &tvn);
464 osi_FreeLargeSpace((char *)ab->ptr_parm[0]); /* free path name buffer here */
467 /* now path may not have been in afs, so check that before calling our cache manager */
468 if (!tvn || !IsAfsVnode(tvn)) {
469 /* release it and give up */
471 #ifdef AFS_LINUX22_ENV
480 /* here we know its an afs vnode, so we can get the data for the chunk */
481 tdc = afs_GetDCache(tvc, ab->size_parm[0], &treq, &offset, &len, 1);
485 #ifdef AFS_LINUX22_ENV
492 /* size_parm 0 to the fetch is the chunk number,
493 * ptr_parm 0 is the dcache entry to wakeup,
494 * size_parm 1 is true iff we should release the dcache entry here.
497 BPrefetch(register struct brequest *ab)
499 register struct dcache *tdc;
500 register struct vcache *tvc;
501 afs_size_t offset, len, abyte, totallen = 0;
502 struct vrequest treq;
504 AFS_STATCNT(BPrefetch);
505 if ((len = afs_InitReq(&treq, ab->cred)))
507 abyte = ab->size_parm[0];
510 tdc = afs_GetDCache(tvc, abyte, &treq, &offset, &len, 1);
516 } while ((totallen < afs_preCache) && tdc && (len > 0));
517 /* now, dude may be waiting for us to clear DFFetchReq bit; do so. Can't
518 * use tdc from GetDCache since afs_GetDCache may fail, but someone may
519 * be waiting for our wakeup anyway.
521 tdc = (struct dcache *)(ab->ptr_parm[0]);
522 ObtainSharedLock(&tdc->lock, 640);
523 if (tdc->mflags & DFFetchReq) {
524 UpgradeSToWLock(&tdc->lock, 641);
525 tdc->mflags &= ~DFFetchReq;
526 ReleaseWriteLock(&tdc->lock);
528 ReleaseSharedLock(&tdc->lock);
530 afs_osi_Wakeup(&tdc->validPos);
531 if (ab->size_parm[1]) {
532 afs_PutDCache(tdc); /* put this one back, too */
536 #if defined(AFS_CACHE_BYPASS)
538 BPrefetchNoCache(register struct brequest *ab)
540 struct vrequest treq;
543 if ((len = afs_InitReq(&treq, ab->cred)))
547 /* OS-specific prefetch routine */
548 afs_PrefetchNoCache(ab->vc, ab->cred, (struct nocache_read_request *) ab->ptr_parm[0]);
554 BStore(register struct brequest *ab)
556 register struct vcache *tvc;
557 register afs_int32 code;
558 struct vrequest treq;
559 #if defined(AFS_SGI_ENV)
560 struct cred *tmpcred;
564 if ((code = afs_InitReq(&treq, ab->cred)))
568 #if defined(AFS_SGI_ENV)
570 * Since StoreOnLastReference can end up calling osi_SyncVM which
571 * calls into VM code that assumes that u.u_cred has the
572 * correct credentials, we set our to theirs for this xaction
574 tmpcred = OSI_GET_CURRENT_CRED();
575 OSI_SET_CURRENT_CRED(ab->cred);
578 * To avoid recursion since the WriteLock may be released during VM
579 * operations, we hold the VOP_RWLOCK across this transaction as
580 * do the other callers of StoreOnLastReference
582 AFS_RWLOCK((vnode_t *) tvc, 1);
584 ObtainWriteLock(&tvc->lock, 209);
585 code = afs_StoreOnLastReference(tvc, &treq);
586 ReleaseWriteLock(&tvc->lock);
587 #if defined(AFS_SGI_ENV)
588 OSI_SET_CURRENT_CRED(tmpcred);
589 AFS_RWUNLOCK((vnode_t *) tvc, 1);
591 /* now set final return code, and wakeup anyone waiting */
592 if ((ab->flags & BUVALID) == 0) {
593 ab->code = afs_CheckCode(code, &treq, 43); /* set final code, since treq doesn't go across processes */
594 ab->flags |= BUVALID;
595 if (ab->flags & BUWAIT) {
596 ab->flags &= ~BUWAIT;
602 /* release a held request buffer */
604 afs_BRelease(register struct brequest *ab)
607 AFS_STATCNT(afs_BRelease);
608 ObtainWriteLock(&afs_xbrs, 294);
609 if (--ab->refCount <= 0) {
613 afs_osi_Wakeup(&afs_brsWaiters);
614 ReleaseWriteLock(&afs_xbrs);
617 /* return true if bkg fetch daemons are all busy */
621 AFS_STATCNT(afs_BBusy);
622 if (afs_brsDaemons > 0)
628 afs_BQueue(register short aopcode, register struct vcache *avc,
629 afs_int32 dontwait, afs_int32 ause, afs_ucred_t *acred,
630 afs_size_t asparm0, afs_size_t asparm1, void *apparm0,
631 void *apparm1, void *apparm2)
634 register struct brequest *tb;
636 AFS_STATCNT(afs_BQueue);
637 ObtainWriteLock(&afs_xbrs, 296);
640 for (i = 0; i < NBRS; i++, tb++) {
641 if (tb->refCount == 0)
646 tb->opcode = aopcode;
653 tb->refCount = ause + 1;
654 tb->size_parm[0] = asparm0;
655 tb->size_parm[1] = asparm1;
656 tb->ptr_parm[0] = apparm0;
657 tb->ptr_parm[1] = apparm1;
658 tb->ptr_parm[2] = apparm2;
661 tb->ts = afs_brs_count++;
662 /* if daemons are waiting for work, wake them up */
663 if (afs_brsDaemons > 0) {
664 afs_osi_Wakeup(&afs_brsDaemons);
666 ReleaseWriteLock(&afs_xbrs);
670 ReleaseWriteLock(&afs_xbrs);
673 /* no free buffers, sleep a while */
675 ReleaseWriteLock(&afs_xbrs);
676 afs_osi_Sleep(&afs_brsWaiters);
677 ObtainWriteLock(&afs_xbrs, 301);
683 /* AIX 4.1 has a much different sleep/wakeup mechanism available for use.
684 * The modifications here will work for either a UP or MP machine.
686 struct buf *afs_asyncbuf = (struct buf *)0;
687 tid_t afs_asyncbuf_cv = EVENT_NULL;
688 afs_int32 afs_biodcnt = 0;
690 /* in implementing this, I assumed that all external linked lists were
693 * Several places in this code traverse a linked list. The algorithm
694 * used here is probably unfamiliar to most people. Careful examination
695 * will show that it eliminates an assignment inside the loop, as compared
696 * to the standard algorithm, at the cost of occasionally using an extra
702 * This function obtains, and returns, a pointer to a buffer for
703 * processing by a daemon. It sleeps until such a buffer is available.
704 * The source of buffers for it is the list afs_asyncbuf (see also
705 * afs_gn_strategy). This function may be invoked concurrently by
706 * several processes, that is, several instances of the same daemon.
707 * afs_gn_strategy, which adds buffers to the list, runs at interrupt
708 * level, while get_bioreq runs at process level.
710 * Since AIX 4.1 can wake just one process at a time, the separate sleep
711 * addresses have been removed.
712 * Note that the kernel_lock is held until the e_sleep_thread() occurs.
713 * The afs_asyncbuf_lock is primarily used to serialize access between
714 * process and interrupts.
716 Simple_lock afs_asyncbuf_lock;
720 struct buf *bp = NULL;
722 struct buf **bestlbpP, **lbpP;
724 struct buf *t1P, *t2P; /* temp pointers for list manipulation */
727 struct afs_bioqueue *s;
729 /* ??? Does the forward pointer of the returned buffer need to be NULL?
732 /* Disable interrupts from the strategy function, and save the
733 * prior priority level and lock access to the afs_asyncbuf.
736 oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
740 /* look for oldest buffer */
741 bp = bestbp = afs_asyncbuf;
742 bestage = (long)bestbp->av_back;
743 bestlbpP = &afs_asyncbuf;
749 if ((long)bp->av_back - bestage < 0) {
752 bestage = (long)bp->av_back;
756 *bestlbpP = bp->av_forw;
759 /* If afs_asyncbuf is null, it is necessary to go to sleep.
760 * e_wakeup_one() ensures that only one thread wakes.
763 /* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
764 * lock on an MP machine.
767 e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
768 LOCK_HANDLER | INTERRUPTIBLE);
769 if (interrupted == THREAD_INTERRUPTED) {
770 /* re-enable interrupts from strategy */
771 unlock_enable(oldPriority, &afs_asyncbuf_lock);
775 } /* end of "else asyncbuf is empty" */
776 } /* end of "inner loop" */
780 unlock_enable(oldPriority, &afs_asyncbuf_lock);
783 /* For the convenience of other code, replace the gnodes in
784 * the b_vp field of bp and the other buffers on the b_work
785 * chain with the corresponding vnodes.
787 * ??? what happens to the gnodes? They're not just cut loose,
791 t2P = (struct buf *)t1P->b_work;
792 t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
796 t1P = (struct buf *)t2P->b_work;
797 t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
802 /* If the buffer does not specify I/O, it may immediately
803 * be returned to the caller. This condition is detected
804 * by examining the buffer's flags (the b_flags field). If
805 * the B_PFPROT bit is set, the buffer represents a protection
806 * violation, rather than a request for I/O. The remainder
807 * of the outer loop handles the case where the B_PFPROT bit is clear.
809 if (bp->b_flags & B_PFPROT) {
814 } /* end of function get_bioreq() */
819 * This function is the daemon. It is called from the syscall
820 * interface. Ordinarily, a script or an administrator will run a
821 * daemon startup utility, specifying the number of I/O daemons to
822 * run. The utility will fork off that number of processes,
823 * each making the appropriate syscall, which will cause this
824 * function to be invoked.
826 static int afs_initbiod = 0; /* this is self-initializing code */
829 afs_BioDaemon(afs_int32 nbiods)
831 afs_int32 code, s, pflg = 0;
833 struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
841 /* pin lock, since we'll be using it in an interrupt. */
842 lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
843 simple_lock_init(&afs_asyncbuf_lock);
844 pin(&afs_asyncbuf, sizeof(struct buf *));
845 pin(&afs_asyncbuf_cv, sizeof(afs_int32));
848 /* Ignore HUP signals... */
850 sigset_t sigbits, osigbits;
852 * add SIGHUP to the set of already masked signals
854 SIGFILLSET(sigbits); /* allow all signals */
855 SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
856 limit_sigs(&sigbits, &osigbits); /* and already masked */
858 /* Main body starts here -- this is an intentional infinite loop, and
861 * Now, the loop will exit if get_bioreq() returns NULL, indicating
862 * that we've been interrupted.
865 bp = afs_get_bioreq();
867 break; /* we were interrupted */
868 if (code = setjmpx(&jmpbuf)) {
869 /* This should not have happend, maybe a lack of resources */
871 s = disable_lock(INTMAX, &afs_asyncbuf_lock);
872 for (bp1 = bp; bp; bp = bp1) {
874 bp1 = (struct buf *)bp1->b_work;
877 bp->b_flags |= B_ERROR;
880 unlock_enable(s, &afs_asyncbuf_lock);
884 vcp = VTOAFS(bp->b_vp);
885 if (bp->b_flags & B_PFSTORE) { /* XXXX */
886 ObtainWriteLock(&vcp->lock, 404);
887 if (vcp->v.v_gnode->gn_mwrcnt) {
888 afs_offs_t newlength =
889 (afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
890 if (vcp->f.m.Length < newlength) {
891 afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
892 ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
893 __LINE__, ICL_TYPE_OFFSET,
894 ICL_HANDLE_OFFSET(vcp->f.m.Length),
895 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
896 vcp->f.m.Length = newlength;
899 ReleaseWriteLock(&vcp->lock);
901 /* If the buffer represents a protection violation, rather than
902 * an actual request for I/O, no special action need be taken.
904 if (bp->b_flags & B_PFPROT) {
905 iodone(bp); /* Notify all users of the buffer that we're done */
910 ObtainWriteLock(&vcp->pvmlock, 211);
912 * First map its data area to a region in the current address space
913 * by calling vm_att with the subspace identifier, and a pointer to
914 * the data area. vm_att returns a new data area pointer, but we
915 * also want to hang onto the old one.
917 tmpaddr = bp->b_baddr;
918 bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
919 tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
920 if (tmperr) { /* in non-error case */
921 bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
922 bp->b_error = tmperr;
925 /* Unmap the buffer's data area by calling vm_det. Reset data area
926 * to the value that we saved above.
929 bp->b_baddr = tmpaddr;
932 * buffer may be linked with other buffers via the b_work field.
933 * See also afs_gn_strategy. For each buffer in the chain (including
934 * bp) notify all users of the buffer that the daemon is finished
935 * using it by calling iodone.
936 * assumes iodone can modify the b_work field.
939 tbp2 = (struct buf *)tbp1->b_work;
944 tbp1 = (struct buf *)tbp2->b_work;
950 ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
952 } /* infinite loop (unless we're interrupted) */
953 } /* end of afs_BioDaemon() */
955 #endif /* AFS_AIX41_ENV */
960 afs_BackgroundDaemon_once(void)
962 LOCK_INIT(&afs_xbrs, "afs_xbrs");
963 memset(afs_brs, 0, sizeof(afs_brs));
965 #if defined (AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
967 * steal the first daemon for doing delayed DSlot flushing
968 * (see afs_GetDownDSlot)
977 brequest_release(struct brequest *tb)
980 AFS_RELE(AFSTOV(tb->vc)); /* MUST call vnode layer or could lose vnodes */
985 tb->cred = (afs_ucred_t *)0;
987 afs_BRelease(tb); /* this grabs and releases afs_xbrs lock */
990 #ifdef AFS_DARWIN80_ENV
992 afs_BackgroundDaemon(struct afs_uspc_param *uspc, void *param1, void *param2)
995 afs_BackgroundDaemon(void)
1001 AFS_STATCNT(afs_BackgroundDaemon);
1002 /* initialize subsystem */
1004 /* Irix with "short stack" exits */
1005 afs_BackgroundDaemon_once();
1007 #ifdef AFS_DARWIN80_ENV
1008 /* If it's a re-entering syscall, complete the request and release */
1009 if (uspc->ts > -1) {
1011 for (i = 0; i < NBRS; i++, tb++) {
1012 if (tb->ts == uspc->ts) {
1013 /* copy the userspace status back in */
1014 ((struct afs_uspc_param *) tb->ptr_parm[0])->retval =
1016 /* mark it valid and notify our caller */
1017 tb->flags |= BUVALID;
1018 if (tb->flags & BUWAIT) {
1019 tb->flags &= ~BUWAIT;
1022 brequest_release(tb);
1027 afs_osi_MaskUserLoop();
1029 /* Otherwise it's a new one */
1031 #ifdef AFS_DARWIN80_ENV
1035 ObtainWriteLock(&afs_xbrs, 302);
1038 struct brequest *min_tb = NULL;
1040 if (afs_termState == AFSOP_STOP_BKG) {
1041 if (--afs_nbrs <= 0)
1042 afs_termState = AFSOP_STOP_TRUNCDAEMON;
1043 ReleaseWriteLock(&afs_xbrs);
1044 afs_osi_Wakeup(&afs_termState);
1045 #ifdef AFS_DARWIN80_ENV
1052 /* find a request */
1055 for (i = 0; i < NBRS; i++, tb++) {
1056 /* look for request with smallest ts */
1057 if ((tb->refCount > 0) && !(tb->flags & BSTARTED)) {
1058 /* new request, not yet picked up */
1059 if ((min_tb && (min_ts - tb->ts > 0)) || !min_tb) {
1065 if ((tb = min_tb)) {
1066 /* claim and process this request */
1067 tb->flags |= BSTARTED;
1068 ReleaseWriteLock(&afs_xbrs);
1070 afs_Trace1(afs_iclSetp, CM_TRACE_BKG1, ICL_TYPE_INT32,
1072 if (tb->opcode == BOP_FETCH)
1074 #if defined(AFS_CACHE_BYPASS)
1075 else if (tb->opcode == BOP_FETCH_NOCACHE)
1076 BPrefetchNoCache(tb);
1078 else if (tb->opcode == BOP_STORE)
1080 else if (tb->opcode == BOP_PATH)
1082 #ifdef AFS_DARWIN80_ENV
1083 else if (tb->opcode == BOP_MOVE) {
1084 memcpy(uspc, (struct afs_uspc_param *) tb->ptr_parm[0],
1085 sizeof(struct afs_uspc_param));
1087 /* string lengths capped in move vop; copy NUL tho */
1088 memcpy(param1, (char *)tb->ptr_parm[1],
1089 strlen(tb->ptr_parm[1])+1);
1090 memcpy(param2, (char *)tb->ptr_parm[2],
1091 strlen(tb->ptr_parm[2])+1);
1096 panic("background bop");
1097 brequest_release(tb);
1098 ObtainWriteLock(&afs_xbrs, 305);
1101 /* wait for new request */
1103 ReleaseWriteLock(&afs_xbrs);
1104 afs_osi_Sleep(&afs_brsDaemons);
1105 ObtainWriteLock(&afs_xbrs, 307);
1109 #ifdef AFS_DARWIN80_ENV
1116 shutdown_daemons(void)
1118 AFS_STATCNT(shutdown_daemons);
1119 if (afs_cold_shutdown) {
1120 afs_brsDaemons = brsInit = 0;
1121 rxepoch_checked = afs_nbrs = 0;
1122 memset(afs_brs, 0, sizeof(afs_brs));
1123 memset(&afs_xbrs, 0, sizeof(afs_lock_t));
1125 #ifdef AFS_AIX41_ENV
1126 lock_free(&afs_asyncbuf_lock);
1127 unpin(&afs_asyncbuf, sizeof(struct buf *));
1128 unpin(&afs_asyncbuf_cv, sizeof(afs_int32));
1134 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1136 * sgi - daemon - handles certain operations that otherwise
1137 * would use up too much kernel stack space
1139 * This all assumes that since the caller must have the xdcache lock
1140 * exclusively that the list will never be more than one long
1141 * and noone else can attempt to add anything until we're done.
1143 SV_TYPE afs_sgibksync;
1144 SV_TYPE afs_sgibkwait;
1145 lock_t afs_sgibklock;
1146 struct dcache *afs_sgibklist;
1154 if (afs_sgibklock == NULL) {
1155 SV_INIT(&afs_sgibksync, "bksync", 0, 0);
1156 SV_INIT(&afs_sgibkwait, "bkwait", 0, 0);
1157 SPINLOCK_INIT(&afs_sgibklock, "bklock");
1159 s = SPLOCK(afs_sgibklock);
1161 /* wait for something to do */
1162 SP_WAIT(afs_sgibklock, s, &afs_sgibksync, PINOD);
1163 osi_Assert(afs_sgibklist);
1165 /* XX will probably need to generalize to real list someday */
1166 s = SPLOCK(afs_sgibklock);
1167 while (afs_sgibklist) {
1168 tdc = afs_sgibklist;
1169 afs_sgibklist = NULL;
1170 SPUNLOCK(afs_sgibklock, s);
1172 tdc->dflags &= ~DFEntryMod;
1173 afs_WriteDCache(tdc, 1);
1175 s = SPLOCK(afs_sgibklock);
1178 /* done all the work - wake everyone up */
1179 while (SV_SIGNAL(&afs_sgibkwait));