3 * THE REGENTS OF THE UNIVERSITY OF MICHIGAN
6 * Permission is granted to use, copy, create derivative works
7 * and redistribute this software and such derivative works
8 * for any purpose, so long as the name of The University of
9 * Michigan is not used in any advertising or publicity
10 * pertaining to the use of distribution of this software
11 * without specific, written prior authorization. If the
12 * above copyright notice or any other identification of the
13 * University of Michigan is included in any copy of any
14 * portion of this software, then the disclaimer below must
17 * THIS SOFTWARE IS PROVIDED AS IS, WITHOUT REPRESENTATION
18 * FROM THE UNIVERSITY OF MICHIGAN AS TO ITS FITNESS FOR ANY
19 * PURPOSE, AND WITHOUT WARRANTY BY THE UNIVERSITY O
20 * MICHIGAN OF ANY KIND, EITHER EXPRESS OR IMPLIED, INCLUDING
21 * WITHOUT LIMITATION THE IMPLIED WARRANTIES OF
22 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE. THE
23 * REGENTS OF THE UNIVERSITY OF MICHIGAN SHALL NOT BE LIABLE
24 * FOR ANY DAMAGES, INCLUDING SPECIAL, INDIRECT, INCIDENTAL, OR
25 * CONSEQUENTIAL DAMAGES, WITH RESPECT TO ANY CLAIM ARISING
26 * OUT OF OR IN CONNECTION WITH THE USE OF THE SOFTWARE, EVEN
27 * IF IT HAS BEEN OR IS HEREAFTER ADVISED OF THE POSSIBILITY OF
32 * Portions Copyright (c) 2008
33 * The Linux Box Corporation
36 * Permission is granted to use, copy, create derivative works
37 * and redistribute this software and such derivative works
38 * for any purpose, so long as the name of the Linux Box
39 * Corporation is not used in any advertising or publicity
40 * pertaining to the use or distribution of this software
41 * without specific, written prior authorization. If the
42 * above copyright notice or any other identification of the
43 * Linux Box Corporation is included in any copy of any
44 * portion of this software, then the disclaimer below must
47 * This software is provided as is, without representation
48 * from the Linux Box Corporation as to its fitness for any
49 * purpose, and without warranty by the Linux Box Corporation
50 * of any kind, either express or implied, including
51 * without limitation the implied warranties of
52 * merchantability and fitness for a particular purpose. The
53 * Linux Box Corporation shall not be liable for any damages,
54 * including special, indirect, incidental, or consequential
55 * damages, with respect to any claim arising out of or in
56 * connection with the use of the software, even if it has been
57 * or is hereafter advised of the possibility of such damages.
61 #include <afsconfig.h>
62 #include "afs/param.h"
64 #if defined(AFS_CACHE_BYPASS)
66 #include "afs/afs_bypasscache.h"
72 #include "afs/sysincludes.h" /* Standard vendor system headers */
73 #include "afs/afsincludes.h" /* Afs-based standard headers */
74 #include "afs/afs_stats.h" /* statistics */
75 #include "afs/nfsclient.h"
76 #include "rx/rx_globals.h"
78 #if defined(AFS_LINUX26_ENV)
79 #define LockPage(pp) lock_page(pp)
80 #define UnlockPage(pp) unlock_page(pp)
82 #define AFS_KMAP_ATOMIC
85 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
88 /* conditional GLOCK macros */
89 #define COND_GLOCK(var) \
91 var = ISAFS_GLOCK(); \
96 #define COND_RE_GUNLOCK(var) \
103 /* conditional GUNLOCK macros */
105 #define COND_GUNLOCK(var) \
107 var = ISAFS_GLOCK(); \
112 #define COND_RE_GLOCK(var) \
119 int cache_bypass_strategy = NEVER_BYPASS_CACHE;
120 int cache_bypass_threshold = AFS_CACHE_BYPASS_DISABLED; /* file size > threshold triggers bypass */
121 int cache_bypass_prefetch = 1; /* Should we do prefetching ? */
123 extern afs_rwlock_t afs_xcbhash;
126 * This is almost exactly like the PFlush() routine in afs_pioctl.c,
127 * but that routine is static. We are about to change a file from
128 * normal caching to bypass it's caching. Therefore, we want to
129 * free up any cache space in use by the file, and throw out any
130 * existing VM pages for the file. We keep track of the number of
131 * times we go back and forth from caching to bypass.
134 afs_TransitionToBypass(struct vcache *avc,
135 afs_ucred_t *acred, int aflags)
139 struct vrequest treq;
146 if (aflags & TRANSChangeDesiredBit)
148 if (aflags & TRANSSetManualBit)
151 #ifdef AFS_BOZONLOCK_ENV
152 afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
157 ObtainWriteLock(&avc->lock, 925);
159 * Someone may have beat us to doing the transition - we had no lock
160 * when we checked the flag earlier. No cause to panic, just return.
162 if (avc->f.states & FCSBypass)
165 /* If we never cached this, just change state */
166 if (setDesire && (!(avc->cachingStates & FCSBypass))) {
167 avc->f.states |= FCSBypass;
171 /* cg2v, try to store any chunks not written 20071204 */
172 if (avc->execsOrWriters > 0) {
173 code = afs_InitReq(&treq, acred);
175 code = afs_StoreAllSegments(avc, &treq, AFS_SYNC | AFS_LASTSTORE);
179 /* also cg2v, don't dequeue the callback */
180 ObtainWriteLock(&afs_xcbhash, 956);
181 afs_DequeueCallback(avc);
182 ReleaseWriteLock(&afs_xcbhash);
184 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
185 /* now find the disk cache entries */
186 afs_TryToSmush(avc, acred, 1);
187 osi_dnlc_purgedp(avc);
188 if (avc->linkData && !(avc->f.states & CCore)) {
189 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
190 avc->linkData = NULL;
193 avc->cachingStates |= FCSBypass; /* Set the bypass flag */
195 avc->cachingStates |= FCSDesireBypass;
197 avc->cachingStates |= FCSManuallySet;
198 avc->cachingTransitions++;
201 ReleaseWriteLock(&avc->lock);
202 #ifdef AFS_BOZONLOCK_ENV
203 afs_BozonUnlock(&avc->pvnLock, avc);
210 * This is almost exactly like the PFlush() routine in afs_pioctl.c,
211 * but that routine is static. We are about to change a file from
212 * bypassing caching to normal caching. Therefore, we want to
213 * throw out any existing VM pages for the file. We keep track of
214 * the number of times we go back and forth from caching to bypass.
217 afs_TransitionToCaching(struct vcache *avc,
227 if (aflags & TRANSChangeDesiredBit)
229 if (aflags & TRANSSetManualBit)
232 #ifdef AFS_BOZONLOCK_ENV
233 afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
237 ObtainWriteLock(&avc->lock, 926);
239 * Someone may have beat us to doing the transition - we had no lock
240 * when we checked the flag earlier. No cause to panic, just return.
242 if (!(avc->f.states & FCSBypass))
245 /* Ok, we actually do need to flush */
246 ObtainWriteLock(&afs_xcbhash, 957);
247 afs_DequeueCallback(avc);
248 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat cache entry */
249 ReleaseWriteLock(&afs_xcbhash);
250 /* now find the disk cache entries */
251 afs_TryToSmush(avc, acred, 1);
252 osi_dnlc_purgedp(avc);
253 if (avc->linkData && !(avc->f.states & CCore)) {
254 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
255 avc->linkData = NULL;
258 avc->cachingStates &= ~(FCSBypass); /* Reset the bypass flag */
260 avc->cachingStates &= ~(FCSDesireBypass);
262 avc->cachingStates |= FCSManuallySet;
263 avc->cachingTransitions++;
266 ReleaseWriteLock(&avc->lock);
267 #ifdef AFS_BOZONLOCK_ENV
268 afs_BozonUnlock(&avc->pvnLock, avc);
274 /* In the case where there's an error in afs_NoCacheFetchProc or
275 * afs_PrefetchNoCache, all of the pages they've been passed need
278 #if defined(AFS_LINUX24_ENV)
279 #define unlock_and_release_pages(auio) \
281 struct iovec *ciov; \
284 afs_int32 iovno = 0; \
285 ciov = auio->uio_iov; \
286 iovmax = auio->uio_iovcnt - 1; \
287 pp = (struct page*) ciov->iov_base; \
288 afs_warn("BYPASS: Unlocking pages..."); \
290 if(pp != NULL && PageLocked(pp)) \
292 put_page(pp); /* decrement refcount */ \
296 ciov = (auio->uio_iov + iovno); \
297 pp = (struct page*) ciov->iov_base; \
299 afs_warn("Pages Unlocked.\n"); \
303 #define unlock_and_release_pages(auio) \
306 #error AFS_CACHE_BYPASS not implemented on this platform
310 /* no-cache prefetch routine */
312 afs_NoCacheFetchProc(struct rx_call *acall,
315 afs_int32 release_pages,
321 int moredata, iovno, iovoff, iovmax, clen, result, locked;
325 #ifdef AFS_KMAP_ATOMIC
326 char *page_buffer = osi_Alloc(PAGE_SIZE);
328 char *page_buffer = NULL;
331 ciov = auio->uio_iov;
332 pp = (struct page*) ciov->iov_base;
333 iovmax = auio->uio_iovcnt - 1;
334 iovno = iovoff = result = 0;
337 COND_GUNLOCK(locked);
338 code = rx_Read(acall, (char *)&length, sizeof(afs_int32));
339 COND_RE_GLOCK(locked);
341 if (code != sizeof(afs_int32)) {
343 afs_warn("Preread error. code: %d instead of %d\n",
344 code, (int)sizeof(afs_int32));
345 unlock_and_release_pages(auio);
348 length = ntohl(length);
352 afs_warn("Preread error. Got length %d, which is greater than size %d\n",
354 unlock_and_release_pages(auio);
359 * The fetch protocol is extended for the AFS/DFS translator
360 * to allow multiple blocks of data, each with its own length,
361 * to be returned. As long as the top bit is set, there are more
364 * We do not do this for AFS file servers because they sometimes
365 * return large negative numbers as the transfer size.
367 if (avc->f.states & CForeign) {
368 moredata = length & 0x80000000;
369 length &= ~0x80000000;
376 clen = ciov->iov_len - iovoff;
377 tlen = afs_min(length, clen);
378 #ifdef AFS_LINUX24_ENV
379 #ifndef AFS_KMAP_ATOMIC
383 /* rx doesn't provide an interface to simply advance
384 or consume n bytes. for now, allocate a PAGE_SIZE
385 region of memory to receive bytes in the case that
386 there were holes in readpages */
387 if(page_buffer == NULL)
388 page_buffer = osi_Alloc(PAGE_SIZE);
389 address = page_buffer;
392 address = page_buffer;
396 #error AFS_CACHE_BYPASS not implemented on this platform
399 COND_GUNLOCK(locked);
400 code = rx_Read(acall, address, tlen);
401 COND_RE_GLOCK(locked);
404 afs_warn("afs_NoCacheFetchProc: rx_Read error. Return code was %d\n", code);
406 unlock_and_release_pages(auio);
408 } else if (code == 0) {
410 afs_warn("afs_NoCacheFetchProc: rx_Read returned zero. Aborting.\n");
411 unlock_and_release_pages(auio);
421 #ifdef AFS_LINUX24_ENV
422 #ifdef AFS_KMAP_ATOMIC
424 address = kmap_atomic(pp, KM_USER0);
425 memcpy(address, page_buffer, PAGE_SIZE);
426 kunmap_atomic(address, KM_USER0);
431 #error AFS_CACHE_BYPASS not implemented on this platform
433 #endif /* LINUX 24 */
434 /* we filled a page, conditionally release it */
435 if (release_pages && ciov->iov_base) {
436 /* this is appropriate when no caller intends to unlock
437 * and release the page */
438 #ifdef AFS_LINUX24_ENV
443 afs_warn("afs_NoCacheFetchProc: page not locked at iovno %d!\n", iovno);
444 put_page(pp); /* decrement refcount */
445 #ifndef AFS_KMAP_ATOMIC
450 #error AFS_CACHE_BYPASS not implemented on this platform
454 /* and carry uio_iov */
459 ciov = (auio->uio_iov + iovno);
460 pp = (struct page*) ciov->iov_base;
468 osi_Free(page_buffer, PAGE_SIZE);
473 /* dispatch a no-cache read request */
475 afs_ReadNoCache(struct vcache *avc,
476 struct nocache_read_request *bparms,
481 struct brequest *breq;
482 struct vrequest *areq;
484 /* the reciever will free this */
485 areq = osi_Alloc(sizeof(struct vrequest));
487 if (avc && avc->vc_error) {
489 afs_warn("afs_ReadNoCache VCache Error!\n");
492 if ((code = afs_InitReq(areq, acred))) {
493 afs_warn("afs_ReadNoCache afs_InitReq error!\n");
498 code = afs_VerifyVCache(avc, areq);
502 code = afs_CheckCode(code, areq, 11); /* failed to get it */
503 afs_warn("afs_ReadNoCache Failed to verify VCache!\n");
509 /* and queue this one */
513 breq = afs_BQueue(BOP_FETCH_NOCACHE, avc, B_DONTWAIT, 0, acred, 1, 1,
514 bparms, (void *)0, (void *)0);
519 afs_osi_Wait(10 * bcnt, 0, 0);
531 /* If there's a problem before we queue the request, we need to
532 * do everything that would normally happen when the request was
533 * processed, like unlocking the pages and freeing memory.
535 #ifdef AFS_LINUX24_ENV
536 unlock_and_release_pages(bparms->auio);
539 #error AFS_CACHE_BYPASS not implemented on this platform
542 osi_Free(areq, sizeof(struct vrequest));
543 osi_Free(bparms->auio->uio_iov,
544 bparms->auio->uio_iovcnt * sizeof(struct iovec));
545 osi_Free(bparms->auio, sizeof(uio_t));
546 osi_Free(bparms, sizeof(struct nocache_read_request));
551 /* Cannot have static linkage--called from BPrefetch (afs_daemons) */
553 afs_PrefetchNoCache(struct vcache *avc,
555 struct nocache_read_request *bparms)
558 struct iovec *iovecp;
559 struct vrequest *areq;
561 #ifdef AFS_64BIT_CLIENT
562 afs_int32 length_hi, bytes, locked;
567 struct rx_call *tcall;
569 struct AFSVolSync tsync;
570 struct AFSFetchStatus OutStatus;
571 struct AFSCallBack CallBack;
573 struct tlocal1 *tcallspec;
577 iovecp = auio->uio_iov;
579 tcallspec = (struct tlocal1 *) osi_Alloc(sizeof(struct tlocal1));
581 tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK /* ignored */);
583 avc->callback = tc->srvr->server;
585 tcall = rx_NewCall(tc->id);
586 #ifdef AFS_64BIT_CLIENT
587 if (!afs_serverHasNo64Bit(tc)) {
588 code = StartRXAFS_FetchData64(tcall,
589 (struct AFSFid *) &avc->f.fid.Fid,
593 COND_GUNLOCK(locked);
594 bytes = rx_Read(tcall, (char *)&length_hi,
596 COND_RE_GLOCK(locked);
598 if (bytes != sizeof(afs_int32)) {
600 code = rx_Error(tcall);
601 COND_GUNLOCK(locked);
602 code = rx_EndCall(tcall, code);
603 COND_RE_GLOCK(locked);
607 } /* afs_serverHasNo64Bit */
608 if (code == RXGEN_OPCODE || afs_serverHasNo64Bit(tc)) {
609 if (auio->uio_offset > 0x7FFFFFFF) {
613 pos = auio->uio_offset;
614 COND_GUNLOCK(locked);
616 tcall = rx_NewCall(tc->id);
617 code = StartRXAFS_FetchData(tcall,
618 (struct AFSFid *) &avc->f.fid.Fid,
619 pos, bparms->length);
620 COND_RE_GLOCK(locked);
622 afs_serverSetNo64Bit(tc);
625 code = StartRXAFS_FetchData(tcall,
626 (struct AFSFid *) &avc->f.fid.Fid,
627 auio->uio_offset, bparms->length);
630 code = afs_NoCacheFetchProc(tcall, avc, auio,
631 1 /* release_pages */,
634 afs_warn("BYPASS: StartRXAFS_FetchData failed: %d\n", code);
635 unlock_and_release_pages(auio);
639 code = EndRXAFS_FetchData(tcall, &tcallspec->OutStatus,
640 &tcallspec->CallBack,
643 afs_warn("BYPASS: NoCacheFetchProc failed: %d\n", code);
645 code = rx_EndCall(tcall, code);
647 afs_warn("BYPASS: No connection.\n");
649 #ifdef AFS_LINUX24_ENV
650 unlock_and_release_pages(auio);
653 #error AFS_CACHE_BYPASS not implemented on this platform
658 } while (afs_Analyze(tc, code, &avc->f.fid, areq,
659 AFS_STATS_FS_RPCIDX_FETCHDATA,
663 * Copy appropriate fields into vcache
666 afs_ProcessFS(avc, &tcallspec->OutStatus, areq);
668 osi_Free(areq, sizeof(struct vrequest));
669 osi_Free(tcallspec, sizeof(struct tlocal1));
670 osi_Free(iovecp, auio->uio_iovcnt * sizeof(struct iovec));
671 osi_Free(bparms, sizeof(struct nocache_read_request));
672 osi_Free(auio, sizeof(uio_t));
676 #endif /* AFS_CACHE_BYPASS */