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 /* 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 /* PAG garbage collection */
40 /* We induce a compile error if param.h does not define AFS_GCPAGS */
41 afs_int32 afs_gcpags = AFS_GCPAGS;
42 afs_int32 afs_gcpags_procsize = 0;
44 afs_int32 afs_CheckServerDaemonStarted = 0;
45 #ifndef DEFAULT_PROBE_INTERVAL
46 #define DEFAULT_PROBE_INTERVAL 30 /* default to 3 min */
48 afs_int32 afs_probe_interval = DEFAULT_PROBE_INTERVAL;
49 afs_int32 afs_probe_all_interval = 600;
50 afs_int32 afs_nat_probe_interval = 60;
51 afs_int32 afs_preCache = 0;
53 #define PROBE_WAIT() (1000 * (afs_probe_interval - ((afs_random() & 0x7fffffff) \
54 % (afs_probe_interval/2))))
57 afs_SetCheckServerNATmode(int isnat)
59 static afs_int32 old_intvl, old_all_intvl;
62 if (isnat && !wasnat) {
63 old_intvl = afs_probe_interval;
64 old_all_intvl = afs_probe_all_interval;
65 afs_probe_interval = afs_nat_probe_interval;
66 afs_probe_all_interval = afs_nat_probe_interval;
67 afs_osi_CancelWait(&AFS_CSWaitHandler);
68 } else if (!isnat && wasnat) {
69 afs_probe_interval = old_intvl;
70 afs_probe_all_interval = old_all_intvl;
76 afs_CheckServerDaemon(void)
78 afs_int32 now, delay, lastCheck, last10MinCheck;
80 afs_CheckServerDaemonStarted = 1;
82 while (afs_initState < 101)
83 afs_osi_Sleep(&afs_initState);
84 afs_osi_Wait(PROBE_WAIT(), &AFS_CSWaitHandler, 0);
86 last10MinCheck = lastCheck = osi_Time();
88 if (afs_termState == AFSOP_STOP_CS) {
89 afs_termState = AFSOP_STOP_TRUNCDAEMON;
90 afs_osi_Wakeup(&afs_termState);
95 if (afs_probe_interval + lastCheck <= now) {
96 afs_CheckServers(1, NULL); /* check down servers */
97 lastCheck = now = osi_Time();
100 if (afs_probe_all_interval + last10MinCheck <= now) {
101 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, afs_probe_all_interval);
102 afs_CheckServers(0, NULL);
103 last10MinCheck = now = osi_Time();
105 /* shutdown check. */
106 if (afs_termState == AFSOP_STOP_CS) {
107 afs_termState = AFSOP_STOP_TRUNCDAEMON;
108 afs_osi_Wakeup(&afs_termState);
112 /* Compute time to next probe. */
113 delay = afs_probe_interval + lastCheck;
114 if (delay > afs_probe_all_interval + last10MinCheck)
115 delay = afs_probe_all_interval + last10MinCheck;
119 afs_osi_Wait(delay * 1000, &AFS_CSWaitHandler, 0);
121 afs_CheckServerDaemonStarted = 0;
124 extern int vfs_context_ref;
126 /* This function always holds the GLOCK whilst it is running. The caller
127 * gets the GLOCK before invoking it, and afs_osi_Sleep drops the GLOCK
128 * whilst we are sleeping, and regains it when we're woken up.
134 struct afs_exporter *exporter;
136 afs_int32 last3MinCheck, last10MinCheck, last60MinCheck, lastNMinCheck;
137 afs_int32 last1MinCheck, last5MinCheck;
138 afs_uint32 lastCBSlotBump;
140 AFS_STATCNT(afs_Daemon);
142 afs_rootFid.Fid.Volume = 0;
143 while (afs_initState < 101)
144 afs_osi_Sleep(&afs_initState);
146 #ifdef AFS_DARWIN80_ENV
147 if (afs_osi_ctxtp_initialized)
148 osi_Panic("vfs context already initialized");
149 while (afs_osi_ctxtp && vfs_context_ref)
150 afs_osi_Sleep(&afs_osi_ctxtp);
151 if (afs_osi_ctxtp && !vfs_context_ref)
152 vfs_context_rele(afs_osi_ctxtp);
153 afs_osi_ctxtp = vfs_context_create(NULL);
154 afs_osi_ctxtp_initialized = 1;
157 lastCBSlotBump = now;
159 /* when a lot of clients are booted simultaneously, they develop
160 * annoying synchronous VL server bashing behaviors. So we stagger them.
162 last1MinCheck = now + ((afs_random() & 0x7fffffff) % 60); /* an extra 30 */
163 last3MinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
164 last60MinCheck = now - 1800 + ((afs_random() & 0x7fffffff) % 3600);
165 last10MinCheck = now - 300 + ((afs_random() & 0x7fffffff) % 600);
166 last5MinCheck = now - 150 + ((afs_random() & 0x7fffffff) % 300);
167 lastNMinCheck = now - 90 + ((afs_random() & 0x7fffffff) % 180);
169 /* start off with afs_initState >= 101 (basic init done) */
171 afs_CheckCallbacks(20); /* unstat anything which will expire soon */
173 /* things to do every 20 seconds or less - required by protocol spec */
175 afs_FlushActiveVcaches(0); /* flush NFS writes */
176 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 */
198 afs_StoreDirtyVcaches();
203 if (last3MinCheck + 180 < now) {
204 afs_CheckTokenCache(); /* check for access cache resets due to expired
209 if (afsd_dynamic_vcaches && (last5MinCheck + 300 < now)) {
210 /* start with trying to drop us back to our base usage */
211 int anumber = VCACHE_FREE + (afs_vcount - afs_cacheStats);
214 ObtainWriteLock(&afs_xvcache, 734);
215 afs_ShakeLooseVCaches(anumber);
216 ReleaseWriteLock(&afs_xvcache);
221 if (!afs_CheckServerDaemonStarted) {
222 if (lastNMinCheck + afs_probe_interval < now) {
223 /* only check down servers */
224 afs_CheckServers(1, NULL);
228 if (last10MinCheck + 600 < now) {
229 #ifdef AFS_USERSPACE_IP_ADDR
230 extern int rxi_GetcbiInfo(void);
232 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEUP, ICL_TYPE_INT32, 600);
233 #ifdef AFS_USERSPACE_IP_ADDR
234 if (rxi_GetcbiInfo()) { /* addresses changed from last time */
237 #else /* AFS_USERSPACE_IP_ADDR */
238 if (rxi_GetIFInfo()) { /* addresses changed from last time */
241 #endif /* else AFS_USERSPACE_IP_ADDR */
242 if (!afs_CheckServerDaemonStarted)
243 afs_CheckServers(0, NULL);
244 afs_GCUserData(); /* gc old conns */
245 /* This is probably the wrong way of doing GC for the various exporters but it will suffice for a while */
246 for (exporter = root_exported; exporter;
247 exporter = exporter->exp_next) {
248 (void)EXP_GC(exporter, 0); /* Generalize params */
253 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
257 afs_CheckVolumeNames(AFS_VOLCHECK_EXPIRED |
262 last10MinCheck = now;
264 if (last60MinCheck + 3600 < now) {
265 afs_Trace1(afs_iclSetp, CM_TRACE_PROBEVOLUME, ICL_TYPE_INT32,
267 afs_CheckRootVolume();
269 if (afs_gcpags == AFS_GCPAGS_OK) {
274 last60MinCheck = now;
276 if (afs_initState < 300) { /* while things ain't rosy */
277 code = afs_CheckRootVolume();
279 afs_initState = 300; /* succeeded */
280 if (afs_initState < 200)
281 afs_initState = 200; /* tried once */
282 afs_osi_Wakeup(&afs_initState);
285 /* 18285 is because we're trying to divide evenly into 128, that is,
286 * CBSlotLen, while staying just under 20 seconds. If CBSlotLen
287 * changes, should probably change this interval, too.
288 * Some of the preceding actions may take quite some time, so we
289 * might not want to wait the entire interval */
290 now = 18285 - (osi_Time() - now);
292 afs_osi_Wait(now, &AFS_WaitHandler, 0);
295 if (afs_termState == AFSOP_STOP_AFS) {
296 if (afs_CheckServerDaemonStarted)
297 afs_termState = AFSOP_STOP_CS;
299 afs_termState = AFSOP_STOP_TRUNCDAEMON;
300 afs_osi_Wakeup(&afs_termState);
307 afs_CheckRootVolume(void)
309 char rootVolName[32];
310 struct volume *tvp = NULL;
311 int usingDynroot = afs_GetDynrootEnable();
314 AFS_STATCNT(afs_CheckRootVolume);
315 if (*afs_rootVolumeName == 0) {
316 strcpy(rootVolName, "root.afs");
318 strcpy(rootVolName, afs_rootVolumeName);
322 afs_GetDynrootFid(&afs_rootFid);
323 tvp = afs_GetVolume(&afs_rootFid, NULL, READ_LOCK);
325 struct cell *lc = afs_GetPrimaryCell(READ_LOCK);
329 localcell = lc->cellNum;
330 afs_PutCell(lc, READ_LOCK);
331 tvp = afs_GetVolumeByName(rootVolName, localcell, 1, NULL, READ_LOCK);
334 int len = strlen(rootVolName);
336 if ((len < 9) || strcmp(&rootVolName[len - 9], ".readonly")) {
337 strcpy(buf, rootVolName);
338 afs_strcat(buf, ".readonly");
339 tvp = afs_GetVolumeByName(buf, localcell, 1, NULL, READ_LOCK);
343 int volid = (tvp->roVol ? tvp->roVol : tvp->volume);
344 afs_rootFid.Cell = localcell;
345 if (afs_rootFid.Fid.Volume && afs_rootFid.Fid.Volume != volid
347 /* If we had a root fid before and it changed location we reset
348 * the afs_globalVp so that it will be reevaluated.
349 * Just decrement the reference count. This only occurs during
350 * initial cell setup and can panic the machine if we set the
351 * count to zero and fs checkv is executed when the current
354 #ifdef AFS_LINUX20_ENV
356 struct vrequest *treq = NULL;
362 afs_rootFid.Fid.Volume = volid;
363 afs_rootFid.Fid.Vnode = 1;
364 afs_rootFid.Fid.Unique = 1;
367 if (afs_CreateReq(&treq, credp))
369 vcp = afs_GetVCache(&afs_rootFid, treq, NULL, NULL);
372 afs_getattr(vcp, &vattr, credp);
373 afs_fill_inode(AFSTOV(vcp), &vattr);
375 dp = d_find_alias(AFSTOV(afs_globalVp));
377 #if defined(AFS_LINUX24_ENV)
378 #if defined(HAVE_DCACHE_LOCK)
379 spin_lock(&dcache_lock);
381 spin_lock(&AFSTOV(vcp)->i_lock);
383 #if defined(AFS_LINUX26_ENV)
384 spin_lock(&dp->d_lock);
387 #if defined(D_ALIAS_IS_HLIST)
388 hlist_del_init(&dp->d_alias);
389 hlist_add_head(&dp->d_alias, &(AFSTOV(vcp)->i_dentry));
391 list_del_init(&dp->d_alias);
392 list_add(&dp->d_alias, &(AFSTOV(vcp)->i_dentry));
394 dp->d_inode = AFSTOV(vcp);
395 #if defined(AFS_LINUX24_ENV)
396 #if defined(AFS_LINUX26_ENV)
397 spin_unlock(&dp->d_lock);
399 #if defined(HAVE_DCACHE_LOCK)
400 spin_unlock(&dcache_lock);
402 spin_unlock(&AFSTOV(vcp)->i_lock);
407 AFS_FAST_RELE(afs_globalVp);
411 afs_DestroyReq(treq);
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(struct brequest *ab)
442 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 = NULL;
453 if ((code = afs_CreateReq(&treq, ab->cred))) {
457 #ifdef AFS_LINUX22_ENV
458 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &dp);
460 tvn = (struct vnode *)dp->d_inode;
462 code = gop_lookupname((char *)ab->ptr_parm[0], AFS_UIOSYS, 1, &tvn);
465 osi_FreeLargeSpace((char *)ab->ptr_parm[0]); /* free path name buffer here */
467 afs_DestroyReq(treq);
470 /* now path may not have been in afs, so check that before calling our cache manager */
471 if (!tvn || !IsAfsVnode(tvn)) {
472 /* release it and give up */
474 #ifdef AFS_LINUX22_ENV
480 afs_DestroyReq(treq);
484 /* here we know its an afs vnode, so we can get the data for the chunk */
485 tdc = afs_GetDCache(tvc, ab->size_parm[0], treq, &offset, &len, 1);
489 #ifdef AFS_LINUX22_ENV
494 afs_DestroyReq(treq);
497 /* size_parm 0 to the fetch is the chunk number,
498 * ptr_parm 0 is the dcache entry to wakeup,
499 * size_parm 1 is true iff we should release the dcache entry here.
502 BPrefetch(struct brequest *ab)
506 afs_size_t offset, len, abyte, totallen = 0;
507 struct vrequest *treq = NULL;
510 AFS_STATCNT(BPrefetch);
511 if ((code = afs_CreateReq(&treq, ab->cred)))
513 abyte = ab->size_parm[0];
516 tdc = afs_GetDCache(tvc, abyte, treq, &offset, &len, 1);
522 } while ((totallen < afs_preCache) && tdc && (len > 0));
523 /* now, dude may be waiting for us to clear DFFetchReq bit; do so. Can't
524 * use tdc from GetDCache since afs_GetDCache may fail, but someone may
525 * be waiting for our wakeup anyway.
527 tdc = (struct dcache *)(ab->ptr_parm[0]);
528 ObtainSharedLock(&tdc->lock, 640);
529 if (tdc->mflags & DFFetchReq) {
530 UpgradeSToWLock(&tdc->lock, 641);
531 tdc->mflags &= ~DFFetchReq;
532 ReleaseWriteLock(&tdc->lock);
534 ReleaseSharedLock(&tdc->lock);
536 afs_osi_Wakeup(&tdc->validPos);
537 if (ab->size_parm[1]) {
538 afs_PutDCache(tdc); /* put this one back, too */
540 afs_DestroyReq(treq);
543 #if defined(AFS_CACHE_BYPASS)
545 BPrefetchNoCache(struct brequest *ab)
547 struct vrequest *treq = NULL;
550 if ((code = afs_CreateReq(&treq, ab->cred)))
554 /* OS-specific prefetch routine */
555 afs_PrefetchNoCache(ab->vc, ab->cred, (struct nocache_read_request *) ab->ptr_parm[0]);
557 afs_DestroyReq(treq);
562 BStore(struct brequest *ab)
566 struct vrequest *treq = NULL;
567 #if defined(AFS_SGI_ENV)
568 struct cred *tmpcred;
572 if ((code = afs_CreateReq(&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) {
601 /* To explain code_raw/code_checkcode:
602 * Anyone that's waiting won't have our treq, so they won't be able to
603 * call afs_CheckCode themselves on the return code we provide here.
604 * But if we give back only the afs_CheckCode value, they won't know
605 * what the "raw" value was. So give back both values, so the waiter
606 * can know the "raw" value for interpreting the value internally, as
607 * well as the afs_CheckCode value to give to the OS. */
609 ab->code_checkcode = afs_CheckCode(code, treq, 430);
611 ab->flags |= BUVALID;
612 if (ab->flags & BUWAIT) {
613 ab->flags &= ~BUWAIT;
617 afs_DestroyReq(treq);
621 BPartialStore(struct brequest *ab)
625 struct vrequest *treq = NULL;
626 int locked, shared_locked = 0;
629 if ((code = afs_CreateReq(&treq, ab->cred)))
632 locked = tvc->lock.excl_locked? 1:0;
634 ObtainWriteLock(&tvc->lock, 1209);
635 else if (!(tvc->lock.excl_locked & WRITE_LOCK)) {
637 ConvertSToRLock(&tvc->lock);
639 code = afs_StoreAllSegments(tvc, treq, AFS_ASYNC);
641 ReleaseWriteLock(&tvc->lock);
642 else if (shared_locked)
643 ConvertSToRLock(&tvc->lock);
644 /* now set final return code, and wakeup anyone waiting */
645 if ((ab->flags & BUVALID) == 0) {
646 /* set final code, since treq doesn't go across processes */
648 ab->code_checkcode = afs_CheckCode(code, treq, 43);
649 ab->flags |= BUVALID;
650 if (ab->flags & BUWAIT) {
651 ab->flags &= ~BUWAIT;
655 afs_DestroyReq(treq);
658 /* release a held request buffer */
660 afs_BRelease(struct brequest *ab)
663 AFS_STATCNT(afs_BRelease);
664 ObtainWriteLock(&afs_xbrs, 294);
665 if (--ab->refCount <= 0) {
669 afs_osi_Wakeup(&afs_brsWaiters);
670 ReleaseWriteLock(&afs_xbrs);
673 /* return true if bkg fetch daemons are all busy */
677 AFS_STATCNT(afs_BBusy);
678 if (afs_brsDaemons > 0)
684 afs_BQueue(short aopcode, struct vcache *avc,
685 afs_int32 dontwait, afs_int32 ause, afs_ucred_t *acred,
686 afs_size_t asparm0, afs_size_t asparm1, void *apparm0,
687 void *apparm1, void *apparm2)
692 AFS_STATCNT(afs_BQueue);
693 ObtainWriteLock(&afs_xbrs, 296);
696 for (i = 0; i < NBRS; i++, tb++) {
697 if (tb->refCount == 0)
702 tb->opcode = aopcode;
711 tb->refCount = ause + 1;
712 tb->size_parm[0] = asparm0;
713 tb->size_parm[1] = asparm1;
714 tb->ptr_parm[0] = apparm0;
715 tb->ptr_parm[1] = apparm1;
716 tb->ptr_parm[2] = apparm2;
718 tb->code_raw = tb->code_checkcode = 0;
719 tb->ts = afs_brs_count++;
720 /* if daemons are waiting for work, wake them up */
721 if (afs_brsDaemons > 0) {
722 afs_osi_Wakeup(&afs_brsDaemons);
724 ReleaseWriteLock(&afs_xbrs);
728 ReleaseWriteLock(&afs_xbrs);
731 /* no free buffers, sleep a while */
733 ReleaseWriteLock(&afs_xbrs);
734 afs_osi_Sleep(&afs_brsWaiters);
735 ObtainWriteLock(&afs_xbrs, 301);
741 /* AIX 4.1 has a much different sleep/wakeup mechanism available for use.
742 * The modifications here will work for either a UP or MP machine.
744 struct buf *afs_asyncbuf = (struct buf *)0;
745 tid_t afs_asyncbuf_cv = EVENT_NULL;
746 afs_int32 afs_biodcnt = 0;
748 /* in implementing this, I assumed that all external linked lists were
751 * Several places in this code traverse a linked list. The algorithm
752 * used here is probably unfamiliar to most people. Careful examination
753 * will show that it eliminates an assignment inside the loop, as compared
754 * to the standard algorithm, at the cost of occasionally using an extra
760 * This function obtains, and returns, a pointer to a buffer for
761 * processing by a daemon. It sleeps until such a buffer is available.
762 * The source of buffers for it is the list afs_asyncbuf (see also
763 * afs_gn_strategy). This function may be invoked concurrently by
764 * several processes, that is, several instances of the same daemon.
765 * afs_gn_strategy, which adds buffers to the list, runs at interrupt
766 * level, while get_bioreq runs at process level.
768 * Since AIX 4.1 can wake just one process at a time, the separate sleep
769 * addresses have been removed.
770 * Note that the kernel_lock is held until the e_sleep_thread() occurs.
771 * The afs_asyncbuf_lock is primarily used to serialize access between
772 * process and interrupts.
774 Simple_lock afs_asyncbuf_lock;
778 struct buf *bp = NULL;
780 struct buf **bestlbpP, **lbpP;
782 struct buf *t1P, *t2P; /* temp pointers for list manipulation */
785 struct afs_bioqueue *s;
787 /* ??? Does the forward pointer of the returned buffer need to be NULL?
790 /* Disable interrupts from the strategy function, and save the
791 * prior priority level and lock access to the afs_asyncbuf.
794 oldPriority = disable_lock(INTMAX, &afs_asyncbuf_lock);
798 /* look for oldest buffer */
799 bp = bestbp = afs_asyncbuf;
800 bestage = (long)bestbp->av_back;
801 bestlbpP = &afs_asyncbuf;
807 if ((long)bp->av_back - bestage < 0) {
810 bestage = (long)bp->av_back;
814 *bestlbpP = bp->av_forw;
817 /* If afs_asyncbuf is null, it is necessary to go to sleep.
818 * e_wakeup_one() ensures that only one thread wakes.
821 /* The LOCK_HANDLER indicates to e_sleep_thread to only drop the
822 * lock on an MP machine.
825 e_sleep_thread(&afs_asyncbuf_cv, &afs_asyncbuf_lock,
826 LOCK_HANDLER | INTERRUPTIBLE);
827 if (interrupted == THREAD_INTERRUPTED) {
828 /* re-enable interrupts from strategy */
829 unlock_enable(oldPriority, &afs_asyncbuf_lock);
833 } /* end of "else asyncbuf is empty" */
834 } /* end of "inner loop" */
838 unlock_enable(oldPriority, &afs_asyncbuf_lock);
841 /* For the convenience of other code, replace the gnodes in
842 * the b_vp field of bp and the other buffers on the b_work
843 * chain with the corresponding vnodes.
845 * ??? what happens to the gnodes? They're not just cut loose,
849 t2P = (struct buf *)t1P->b_work;
850 t1P->b_vp = ((struct gnode *)t1P->b_vp)->gn_vnode;
854 t1P = (struct buf *)t2P->b_work;
855 t2P->b_vp = ((struct gnode *)t2P->b_vp)->gn_vnode;
860 /* If the buffer does not specify I/O, it may immediately
861 * be returned to the caller. This condition is detected
862 * by examining the buffer's flags (the b_flags field). If
863 * the B_PFPROT bit is set, the buffer represents a protection
864 * violation, rather than a request for I/O. The remainder
865 * of the outer loop handles the case where the B_PFPROT bit is clear.
867 if (bp->b_flags & B_PFPROT) {
872 } /* end of function get_bioreq() */
877 * This function is the daemon. It is called from the syscall
878 * interface. Ordinarily, a script or an administrator will run a
879 * daemon startup utility, specifying the number of I/O daemons to
880 * run. The utility will fork off that number of processes,
881 * each making the appropriate syscall, which will cause this
882 * function to be invoked.
884 static int afs_initbiod = 0; /* this is self-initializing code */
887 afs_BioDaemon(afs_int32 nbiods)
889 afs_int32 code, s, pflg = 0;
891 struct buf *bp, *bp1, *tbp1, *tbp2; /* temp pointers only */
899 /* pin lock, since we'll be using it in an interrupt. */
900 lock_alloc(&afs_asyncbuf_lock, LOCK_ALLOC_PIN, 2, 1);
901 simple_lock_init(&afs_asyncbuf_lock);
902 pin(&afs_asyncbuf, sizeof(struct buf *));
903 pin(&afs_asyncbuf_cv, sizeof(afs_int32));
906 /* Ignore HUP signals... */
908 sigset_t sigbits, osigbits;
910 * add SIGHUP to the set of already masked signals
912 SIGFILLSET(sigbits); /* allow all signals */
913 SIGDELSET(sigbits, SIGHUP); /* except SIGHUP */
914 limit_sigs(&sigbits, &osigbits); /* and already masked */
916 /* Main body starts here -- this is an intentional infinite loop, and
919 * Now, the loop will exit if get_bioreq() returns NULL, indicating
920 * that we've been interrupted.
923 bp = afs_get_bioreq();
925 break; /* we were interrupted */
926 if (code = setjmpx(&jmpbuf)) {
927 /* This should not have happend, maybe a lack of resources */
929 s = disable_lock(INTMAX, &afs_asyncbuf_lock);
930 for (bp1 = bp; bp; bp = bp1) {
932 bp1 = (struct buf *)bp1->b_work;
935 bp->b_flags |= B_ERROR;
938 unlock_enable(s, &afs_asyncbuf_lock);
942 vcp = VTOAFS(bp->b_vp);
943 if (bp->b_flags & B_PFSTORE) { /* XXXX */
944 ObtainWriteLock(&vcp->lock, 404);
945 if (vcp->v.v_gnode->gn_mwrcnt) {
946 afs_offs_t newlength =
947 (afs_offs_t) dbtob(bp->b_blkno) + bp->b_bcount;
948 if (vcp->f.m.Length < newlength) {
949 afs_Trace4(afs_iclSetp, CM_TRACE_SETLENGTH,
950 ICL_TYPE_STRING, __FILE__, ICL_TYPE_LONG,
951 __LINE__, ICL_TYPE_OFFSET,
952 ICL_HANDLE_OFFSET(vcp->f.m.Length),
953 ICL_TYPE_OFFSET, ICL_HANDLE_OFFSET(newlength));
954 vcp->f.m.Length = newlength;
957 ReleaseWriteLock(&vcp->lock);
959 /* If the buffer represents a protection violation, rather than
960 * an actual request for I/O, no special action need be taken.
962 if (bp->b_flags & B_PFPROT) {
963 iodone(bp); /* Notify all users of the buffer that we're done */
968 ObtainWriteLock(&vcp->pvmlock, 211);
970 * First map its data area to a region in the current address space
971 * by calling vm_att with the subspace identifier, and a pointer to
972 * the data area. vm_att returns a new data area pointer, but we
973 * also want to hang onto the old one.
975 tmpaddr = bp->b_baddr;
976 bp->b_baddr = (caddr_t) vm_att(bp->b_xmemd.subspace_id, tmpaddr);
977 tmperr = afs_ustrategy(bp); /* temp variable saves offset calculation */
978 if (tmperr) { /* in non-error case */
979 bp->b_flags |= B_ERROR; /* should other flags remain set ??? */
980 bp->b_error = tmperr;
983 /* Unmap the buffer's data area by calling vm_det. Reset data area
984 * to the value that we saved above.
987 bp->b_baddr = tmpaddr;
990 * buffer may be linked with other buffers via the b_work field.
991 * See also afs_gn_strategy. For each buffer in the chain (including
992 * bp) notify all users of the buffer that the daemon is finished
993 * using it by calling iodone.
994 * assumes iodone can modify the b_work field.
997 tbp2 = (struct buf *)tbp1->b_work;
1002 tbp1 = (struct buf *)tbp2->b_work;
1008 ReleaseWriteLock(&vcp->pvmlock); /* Unlock the vnode. */
1010 } /* infinite loop (unless we're interrupted) */
1011 } /* end of afs_BioDaemon() */
1013 #endif /* AFS_AIX41_ENV */
1018 afs_BackgroundDaemon_once(void)
1020 LOCK_INIT(&afs_xbrs, "afs_xbrs");
1021 memset(afs_brs, 0, sizeof(afs_brs));
1023 #if defined (AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1025 * steal the first daemon for doing delayed DSlot flushing
1026 * (see afs_GetDownDSlot)
1030 exit(CLD_EXITED, 0);
1035 brequest_release(struct brequest *tb)
1038 AFS_RELE(AFSTOV(tb->vc)); /* MUST call vnode layer or could lose vnodes */
1043 tb->cred = (afs_ucred_t *)0;
1045 afs_BRelease(tb); /* this grabs and releases afs_xbrs lock */
1050 afs_BackgroundDaemon(struct afs_uspc_param *uspc, void *param1, void *param2)
1053 afs_BackgroundDaemon(void)
1056 struct brequest *tb;
1059 AFS_STATCNT(afs_BackgroundDaemon);
1060 /* initialize subsystem */
1062 /* Irix with "short stack" exits */
1063 afs_BackgroundDaemon_once();
1066 /* If it's a re-entering syscall, complete the request and release */
1067 if (uspc->ts > -1) {
1069 for (i = 0; i < NBRS; i++, tb++) {
1070 if (tb->ts == uspc->ts) {
1071 /* copy the userspace status back in */
1072 ((struct afs_uspc_param *) tb->ptr_parm[0])->retval =
1074 /* mark it valid and notify our caller */
1075 tb->flags |= BUVALID;
1076 if (tb->flags & BUWAIT) {
1077 tb->flags &= ~BUWAIT;
1080 brequest_release(tb);
1085 afs_osi_MaskUserLoop();
1087 /* Otherwise it's a new one */
1093 ObtainWriteLock(&afs_xbrs, 302);
1096 struct brequest *min_tb = NULL;
1098 if (afs_termState == AFSOP_STOP_BKG) {
1099 if (--afs_nbrs <= 0)
1100 afs_termState = AFSOP_STOP_RXCALLBACK;
1101 ReleaseWriteLock(&afs_xbrs);
1102 afs_osi_Wakeup(&afs_termState);
1110 /* find a request */
1113 for (i = 0; i < NBRS; i++, tb++) {
1114 /* look for request with smallest ts */
1115 if ((tb->refCount > 0) && !(tb->flags & BSTARTED)) {
1116 /* new request, not yet picked up */
1117 if ((min_tb && (min_ts - tb->ts > 0)) || !min_tb) {
1123 if ((tb = min_tb)) {
1124 /* claim and process this request */
1125 tb->flags |= BSTARTED;
1126 ReleaseWriteLock(&afs_xbrs);
1128 afs_Trace1(afs_iclSetp, CM_TRACE_BKG1, ICL_TYPE_INT32,
1130 if (tb->opcode == BOP_FETCH)
1132 #if defined(AFS_CACHE_BYPASS)
1133 else if (tb->opcode == BOP_FETCH_NOCACHE)
1134 BPrefetchNoCache(tb);
1136 else if (tb->opcode == BOP_STORE)
1138 else if (tb->opcode == BOP_PATH)
1140 #ifdef AFS_DARWIN80_ENV
1141 else if (tb->opcode == BOP_MOVE) {
1142 memcpy(uspc, (struct afs_uspc_param *) tb->ptr_parm[0],
1143 sizeof(struct afs_uspc_param));
1145 /* string lengths capped in move vop; copy NUL tho */
1146 memcpy(param1, (char *)tb->ptr_parm[1],
1147 strlen(tb->ptr_parm[1])+1);
1148 memcpy(param2, (char *)tb->ptr_parm[2],
1149 strlen(tb->ptr_parm[2])+1);
1153 else if (tb->opcode == BOP_PARTIAL_STORE)
1156 panic("background bop");
1157 brequest_release(tb);
1158 ObtainWriteLock(&afs_xbrs, 305);
1161 /* wait for new request */
1163 ReleaseWriteLock(&afs_xbrs);
1164 afs_osi_Sleep(&afs_brsDaemons);
1165 ObtainWriteLock(&afs_xbrs, 307);
1176 shutdown_daemons(void)
1178 AFS_STATCNT(shutdown_daemons);
1179 if (afs_cold_shutdown) {
1180 afs_brsDaemons = brsInit = 0;
1182 memset(afs_brs, 0, sizeof(afs_brs));
1183 memset(&afs_xbrs, 0, sizeof(afs_lock_t));
1185 #ifdef AFS_AIX41_ENV
1186 lock_free(&afs_asyncbuf_lock);
1187 unpin(&afs_asyncbuf, sizeof(struct buf *));
1188 unpin(&afs_asyncbuf_cv, sizeof(afs_int32));
1194 #if defined(AFS_SGI_ENV) && defined(AFS_SGI_SHORTSTACK)
1196 * sgi - daemon - handles certain operations that otherwise
1197 * would use up too much kernel stack space
1199 * This all assumes that since the caller must have the xdcache lock
1200 * exclusively that the list will never be more than one long
1201 * and noone else can attempt to add anything until we're done.
1203 SV_TYPE afs_sgibksync;
1204 SV_TYPE afs_sgibkwait;
1205 lock_t afs_sgibklock;
1206 struct dcache *afs_sgibklist;
1214 if (afs_sgibklock == NULL) {
1215 SV_INIT(&afs_sgibksync, "bksync", 0, 0);
1216 SV_INIT(&afs_sgibkwait, "bkwait", 0, 0);
1217 SPINLOCK_INIT(&afs_sgibklock, "bklock");
1219 s = SPLOCK(afs_sgibklock);
1221 /* wait for something to do */
1222 SP_WAIT(afs_sgibklock, s, &afs_sgibksync, PINOD);
1223 osi_Assert(afs_sgibklist);
1225 /* XX will probably need to generalize to real list someday */
1226 s = SPLOCK(afs_sgibklock);
1227 while (afs_sgibklist) {
1228 tdc = afs_sgibklist;
1229 afs_sgibklist = NULL;
1230 SPUNLOCK(afs_sgibklock, s);
1232 tdc->dflags &= ~DFEntryMod;
1233 osi_Assert(afs_WriteDCache(tdc, 1) == 0);
1235 s = SPLOCK(afs_sgibklock);
1238 /* done all the work - wake everyone up */
1239 while (SV_SIGNAL(&afs_sgibkwait));