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)
84 #define afs_min(A,B) ((A)<(B)) ? (A) : (B)
87 /* conditional GLOCK macros */
88 #define COND_GLOCK(var) \
90 var = ISAFS_GLOCK(); \
95 #define COND_RE_GUNLOCK(var) \
102 /* conditional GUNLOCK macros */
104 #define COND_GUNLOCK(var) \
106 var = ISAFS_GLOCK(); \
111 #define COND_RE_GLOCK(var) \
118 int cache_bypass_strategy = NEVER_BYPASS_CACHE;
119 int cache_bypass_threshold = AFS_CACHE_BYPASS_DISABLED; /* file size > threshold triggers bypass */
120 int cache_bypass_prefetch = 1; /* Should we do prefetching ? */
122 extern afs_rwlock_t afs_xcbhash;
125 * This is almost exactly like the PFlush() routine in afs_pioctl.c,
126 * but that routine is static. We are about to change a file from
127 * normal caching to bypass it's caching. Therefore, we want to
128 * free up any cache space in use by the file, and throw out any
129 * existing VM pages for the file. We keep track of the number of
130 * times we go back and forth from caching to bypass.
133 afs_TransitionToBypass(struct vcache *avc,
134 afs_ucred_t *acred, int aflags)
138 struct vrequest treq;
145 if (aflags & TRANSChangeDesiredBit)
147 if (aflags & TRANSSetManualBit)
150 #ifdef AFS_BOZONLOCK_ENV
151 afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
156 ObtainWriteLock(&avc->lock, 925);
158 * Someone may have beat us to doing the transition - we had no lock
159 * when we checked the flag earlier. No cause to panic, just return.
161 if (avc->f.states & FCSBypass)
164 /* If we never cached this, just change state */
165 if (setDesire && (!(avc->cachingStates & FCSBypass))) {
166 avc->f.states |= FCSBypass;
170 /* cg2v, try to store any chunks not written 20071204 */
171 if (avc->execsOrWriters > 0) {
172 code = afs_InitReq(&treq, acred);
174 code = afs_StoreAllSegments(avc, &treq, AFS_SYNC | AFS_LASTSTORE);
178 /* also cg2v, don't dequeue the callback */
179 ObtainWriteLock(&afs_xcbhash, 956);
180 afs_DequeueCallback(avc);
181 ReleaseWriteLock(&afs_xcbhash);
183 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat */
184 /* now find the disk cache entries */
185 afs_TryToSmush(avc, acred, 1);
186 osi_dnlc_purgedp(avc);
187 if (avc->linkData && !(avc->f.states & CCore)) {
188 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
189 avc->linkData = NULL;
192 avc->cachingStates |= FCSBypass; /* Set the bypass flag */
194 avc->cachingStates |= FCSDesireBypass;
196 avc->cachingStates |= FCSManuallySet;
197 avc->cachingTransitions++;
200 ReleaseWriteLock(&avc->lock);
201 #ifdef AFS_BOZONLOCK_ENV
202 afs_BozonUnlock(&avc->pvnLock, avc);
209 * This is almost exactly like the PFlush() routine in afs_pioctl.c,
210 * but that routine is static. We are about to change a file from
211 * bypassing caching to normal caching. Therefore, we want to
212 * throw out any existing VM pages for the file. We keep track of
213 * the number of times we go back and forth from caching to bypass.
216 afs_TransitionToCaching(struct vcache *avc,
226 if (aflags & TRANSChangeDesiredBit)
228 if (aflags & TRANSSetManualBit)
231 #ifdef AFS_BOZONLOCK_ENV
232 afs_BozonLock(&avc->pvnLock, avc); /* Since afs_TryToSmush will do a pvn_vptrunc */
236 ObtainWriteLock(&avc->lock, 926);
238 * Someone may have beat us to doing the transition - we had no lock
239 * when we checked the flag earlier. No cause to panic, just return.
241 if (!(avc->f.states & FCSBypass))
244 /* Ok, we actually do need to flush */
245 ObtainWriteLock(&afs_xcbhash, 957);
246 afs_DequeueCallback(avc);
247 avc->f.states &= ~(CStatd | CDirty); /* next reference will re-stat cache entry */
248 ReleaseWriteLock(&afs_xcbhash);
249 /* now find the disk cache entries */
250 afs_TryToSmush(avc, acred, 1);
251 osi_dnlc_purgedp(avc);
252 if (avc->linkData && !(avc->f.states & CCore)) {
253 afs_osi_Free(avc->linkData, strlen(avc->linkData) + 1);
254 avc->linkData = NULL;
257 avc->cachingStates &= ~(FCSBypass); /* Reset the bypass flag */
259 avc->cachingStates &= ~(FCSDesireBypass);
261 avc->cachingStates |= FCSManuallySet;
262 avc->cachingTransitions++;
265 ReleaseWriteLock(&avc->lock);
266 #ifdef AFS_BOZONLOCK_ENV
267 afs_BozonUnlock(&avc->pvnLock, avc);
273 /* In the case where there's an error in afs_NoCacheFetchProc or
274 * afs_PrefetchNoCache, all of the pages they've been passed need
277 #if defined(AFS_LINUX24_ENV)
278 #define unlock_and_release_pages(auio) \
280 struct iovec *ciov; \
283 afs_int32 iovno = 0; \
284 ciov = auio->uio_iov; \
285 iovmax = auio->uio_iovcnt - 1; \
286 pp = (struct page*) ciov->iov_base; \
289 if (PageLocked(pp)) \
291 put_page(pp); /* decrement refcount */ \
296 ciov = (auio->uio_iov + iovno); \
297 pp = (struct page*) ciov->iov_base; \
302 #define unlock_and_release_pages(auio) \
305 #error AFS_CACHE_BYPASS not implemented on this platform
309 /* no-cache prefetch routine */
311 afs_NoCacheFetchProc(struct rx_call *acall,
314 afs_int32 release_pages,
320 int moredata, iovno, iovoff, iovmax, clen, result, locked;
324 char *page_buffer = osi_Alloc(PAGE_SIZE);
326 ciov = auio->uio_iov;
327 pp = (struct page*) ciov->iov_base;
328 iovmax = auio->uio_iovcnt - 1;
329 iovno = iovoff = result = 0;
332 COND_GUNLOCK(locked);
333 code = rx_Read(acall, (char *)&length, sizeof(afs_int32));
334 COND_RE_GLOCK(locked);
336 if (code != sizeof(afs_int32)) {
338 afs_warn("Preread error. code: %d instead of %d\n",
339 code, (int)sizeof(afs_int32));
340 unlock_and_release_pages(auio);
343 length = ntohl(length);
347 afs_warn("Preread error. Got length %d, which is greater than size %d\n",
349 unlock_and_release_pages(auio);
353 /* If we get a 0 length reply, time to cleanup and return */
355 unlock_and_release_pages(auio);
361 * The fetch protocol is extended for the AFS/DFS translator
362 * to allow multiple blocks of data, each with its own length,
363 * to be returned. As long as the top bit is set, there are more
366 * We do not do this for AFS file servers because they sometimes
367 * return large negative numbers as the transfer size.
369 if (avc->f.states & CForeign) {
370 moredata = length & 0x80000000;
371 length &= ~0x80000000;
378 clen = ciov->iov_len - iovoff;
379 tlen = afs_min(length, clen);
380 #ifdef AFS_LINUX24_ENV
381 address = page_buffer;
384 #error AFS_CACHE_BYPASS not implemented on this platform
387 COND_GUNLOCK(locked);
388 code = rx_Read(acall, address, tlen);
389 COND_RE_GLOCK(locked);
392 afs_warn("afs_NoCacheFetchProc: rx_Read error. Return code was %d\n", code);
394 unlock_and_release_pages(auio);
396 } else if (code == 0) {
398 afs_warn("afs_NoCacheFetchProc: rx_Read returned zero. Aborting.\n");
399 unlock_and_release_pages(auio);
409 #ifdef AFS_LINUX24_ENV
411 address = kmap_atomic(pp, KM_USER0);
412 memcpy(address, page_buffer, PAGE_SIZE);
413 kunmap_atomic(address, KM_USER0);
417 #error AFS_CACHE_BYPASS not implemented on this platform
419 #endif /* LINUX 24 */
420 /* we filled a page, conditionally release it */
421 if (release_pages && ciov->iov_base) {
422 /* this is appropriate when no caller intends to unlock
423 * and release the page */
424 #ifdef AFS_LINUX24_ENV
429 afs_warn("afs_NoCacheFetchProc: page not locked at iovno %d!\n", iovno);
430 put_page(pp); /* decrement refcount */
433 #error AFS_CACHE_BYPASS not implemented on this platform
437 /* and carry uio_iov */
442 ciov = (auio->uio_iov + iovno);
443 pp = (struct page*) ciov->iov_base;
451 osi_Free(page_buffer, PAGE_SIZE);
456 /* dispatch a no-cache read request */
458 afs_ReadNoCache(struct vcache *avc,
459 struct nocache_read_request *bparms,
464 struct brequest *breq;
465 struct vrequest *areq;
467 /* the reciever will free this */
468 areq = osi_Alloc(sizeof(struct vrequest));
470 if (avc && avc->vc_error) {
472 afs_warn("afs_ReadNoCache VCache Error!\n");
475 if ((code = afs_InitReq(areq, acred))) {
476 afs_warn("afs_ReadNoCache afs_InitReq error!\n");
481 code = afs_VerifyVCache(avc, areq);
485 code = afs_CheckCode(code, areq, 11); /* failed to get it */
486 afs_warn("afs_ReadNoCache Failed to verify VCache!\n");
492 /* and queue this one */
496 breq = afs_BQueue(BOP_FETCH_NOCACHE, avc, B_DONTWAIT, 0, acred, 1, 1,
497 bparms, (void *)0, (void *)0);
502 afs_osi_Wait(10 * bcnt, 0, 0);
514 /* If there's a problem before we queue the request, we need to
515 * do everything that would normally happen when the request was
516 * processed, like unlocking the pages and freeing memory.
518 #ifdef AFS_LINUX24_ENV
519 unlock_and_release_pages(bparms->auio);
522 #error AFS_CACHE_BYPASS not implemented on this platform
525 osi_Free(areq, sizeof(struct vrequest));
526 osi_Free(bparms->auio->uio_iov,
527 bparms->auio->uio_iovcnt * sizeof(struct iovec));
528 osi_Free(bparms->auio, sizeof(uio_t));
529 osi_Free(bparms, sizeof(struct nocache_read_request));
534 /* Cannot have static linkage--called from BPrefetch (afs_daemons) */
536 afs_PrefetchNoCache(struct vcache *avc,
538 struct nocache_read_request *bparms)
541 struct iovec *iovecp;
542 struct vrequest *areq;
544 #ifdef AFS_64BIT_CLIENT
545 afs_int32 length_hi, bytes, locked;
550 struct rx_call *tcall;
552 struct AFSVolSync tsync;
553 struct AFSFetchStatus OutStatus;
554 struct AFSCallBack CallBack;
556 struct tlocal1 *tcallspec;
560 iovecp = auio->uio_iov;
562 tcallspec = (struct tlocal1 *) osi_Alloc(sizeof(struct tlocal1));
564 tc = afs_Conn(&avc->f.fid, areq, SHARED_LOCK /* ignored */);
566 avc->callback = tc->srvr->server;
568 tcall = rx_NewCall(tc->id);
569 #ifdef AFS_64BIT_CLIENT
570 if (!afs_serverHasNo64Bit(tc)) {
571 code = StartRXAFS_FetchData64(tcall,
572 (struct AFSFid *) &avc->f.fid.Fid,
576 COND_GUNLOCK(locked);
577 bytes = rx_Read(tcall, (char *)&length_hi,
579 COND_RE_GLOCK(locked);
581 if (bytes != sizeof(afs_int32)) {
583 code = rx_Error(tcall);
584 COND_GUNLOCK(locked);
585 code = rx_EndCall(tcall, code);
586 COND_RE_GLOCK(locked);
590 } /* afs_serverHasNo64Bit */
591 if (code == RXGEN_OPCODE || afs_serverHasNo64Bit(tc)) {
592 if (auio->uio_offset > 0x7FFFFFFF) {
596 pos = auio->uio_offset;
597 COND_GUNLOCK(locked);
599 tcall = rx_NewCall(tc->id);
600 code = StartRXAFS_FetchData(tcall,
601 (struct AFSFid *) &avc->f.fid.Fid,
602 pos, bparms->length);
603 COND_RE_GLOCK(locked);
605 afs_serverSetNo64Bit(tc);
608 code = StartRXAFS_FetchData(tcall,
609 (struct AFSFid *) &avc->f.fid.Fid,
610 auio->uio_offset, bparms->length);
613 code = afs_NoCacheFetchProc(tcall, avc, auio,
614 1 /* release_pages */,
617 afs_warn("BYPASS: StartRXAFS_FetchData failed: %d\n", code);
618 unlock_and_release_pages(auio);
622 code = EndRXAFS_FetchData(tcall, &tcallspec->OutStatus,
623 &tcallspec->CallBack,
626 afs_warn("BYPASS: NoCacheFetchProc failed: %d\n", code);
628 code = rx_EndCall(tcall, code);
630 afs_warn("BYPASS: No connection.\n");
632 #ifdef AFS_LINUX24_ENV
633 unlock_and_release_pages(auio);
636 #error AFS_CACHE_BYPASS not implemented on this platform
641 } while (afs_Analyze(tc, code, &avc->f.fid, areq,
642 AFS_STATS_FS_RPCIDX_FETCHDATA,
646 * Copy appropriate fields into vcache
649 afs_ProcessFS(avc, &tcallspec->OutStatus, areq);
651 osi_Free(areq, sizeof(struct vrequest));
652 osi_Free(tcallspec, sizeof(struct tlocal1));
653 osi_Free(iovecp, auio->uio_iovcnt * sizeof(struct iovec));
654 osi_Free(bparms, sizeof(struct nocache_read_request));
655 osi_Free(auio, sizeof(uio_t));
659 #endif /* AFS_CACHE_BYPASS */