Import Upstream version 1.8.5
[hcoop/debian/openafs.git] / src / afs / afs_fetchstore.c
1 /*
2 * Copyright 2000, International Business Machines Corporation and others.
3 * All Rights Reserved.
4 *
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
8 */
9
10 #include <afsconfig.h>
11 #include "afs/param.h"
12
13 #include "afs/sysincludes.h" /* Standard vendor system headers */
14 #ifdef AFS_ALPHA_ENV
15 #undef kmem_alloc
16 #undef kmem_free
17 #undef mem_alloc
18 #undef mem_free
19 #endif /* AFS_ALPHA_ENV */
20 #include "afsincludes.h" /* Afs-based standard headers */
21 #include "afs/afs_stats.h" /* statistics */
22 #include "afs_prototypes.h"
23
24 extern int cacheDiskType;
25
26 #ifndef AFS_NOSTATS
27 static void
28 FillStoreStats(int code, int idx, osi_timeval_t xferStartTime,
29 afs_size_t bytesToXfer, afs_size_t bytesXferred)
30 {
31 struct afs_stats_xferData *xferP;
32 osi_timeval_t xferStopTime;
33 osi_timeval_t elapsedTime;
34
35 xferP = &(afs_stats_cmfullperf.rpc.fsXferTimes[idx]);
36 osi_GetuTime(&xferStopTime);
37 (xferP->numXfers)++;
38 if (!code) {
39 (xferP->numSuccesses)++;
40 afs_stats_XferSumBytes[idx] += bytesXferred;
41 (xferP->sumBytes) += (afs_stats_XferSumBytes[idx] >> 10);
42 afs_stats_XferSumBytes[idx] &= 0x3FF;
43 if (bytesXferred < xferP->minBytes)
44 xferP->minBytes = bytesXferred;
45 if (bytesXferred > xferP->maxBytes)
46 xferP->maxBytes = bytesXferred;
47
48 /*
49 * Tally the size of the object. Note: we tally the actual size,
50 * NOT the number of bytes that made it out over the wire.
51 */
52 if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET0) (xferP->count[0])++;
53 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET1) (xferP->count[1])++;
54 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET2) (xferP->count[2])++;
55 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET3) (xferP->count[3])++;
56 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET4) (xferP->count[4])++;
57 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET5) (xferP->count[5])++;
58 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET6) (xferP->count[6])++;
59 else if (bytesToXfer <= AFS_STATS_MAXBYTES_BUCKET7) (xferP->count[7])++;
60 else
61 (xferP->count[8])++;
62
63 afs_stats_GetDiff(elapsedTime, xferStartTime, xferStopTime);
64 afs_stats_AddTo((xferP->sumTime), elapsedTime);
65 afs_stats_SquareAddTo((xferP->sqrTime), elapsedTime);
66 if (afs_stats_TimeLessThan(elapsedTime, (xferP->minTime))) {
67 afs_stats_TimeAssign((xferP->minTime), elapsedTime);
68 }
69 if (afs_stats_TimeGreaterThan(elapsedTime, (xferP->maxTime))) {
70 afs_stats_TimeAssign((xferP->maxTime), elapsedTime);
71 }
72 }
73 }
74 #endif /* AFS_NOSTATS */
75
76 /* rock and operations for RX_FILESERVER */
77
78
79
80 afs_int32
81 rxfs_storeUfsPrepare(void *r, afs_uint32 size, afs_uint32 *tlen)
82 {
83 *tlen = (size > AFS_LRALLOCSIZ ? AFS_LRALLOCSIZ : size);
84 return 0;
85 }
86
87 afs_int32
88 rxfs_storeMemPrepare(void *r, afs_uint32 size, afs_uint32 *tlen)
89 {
90 afs_int32 code;
91 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *) r;
92
93 RX_AFS_GUNLOCK();
94 code = rx_WritevAlloc(v->call, v->tiov, &v->tnio, RX_MAXIOVECS, size);
95 RX_AFS_GLOCK();
96 if (code <= 0) {
97 code = rx_Error(v->call);
98 if (!code)
99 code = -33;
100 }
101 else {
102 *tlen = code;
103 code = 0;
104 }
105 return code;
106 }
107
108 afs_int32
109 rxfs_storeUfsRead(void *r, struct osi_file *tfile, afs_uint32 offset,
110 afs_uint32 tlen, afs_uint32 *bytesread)
111 {
112 afs_int32 code;
113 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)r;
114
115 *bytesread = 0;
116 code = afs_osi_Read(tfile, -1, v->tbuffer, tlen);
117 if (code < 0)
118 return EIO;
119 *bytesread = code;
120 if (code == tlen)
121 return 0;
122 #if defined(KERNEL_HAVE_UERROR)
123 if (getuerror())
124 return EIO;
125 #endif
126 return 0;
127 }
128
129 afs_int32
130 rxfs_storeMemRead(void *r, struct osi_file *tfile, afs_uint32 offset,
131 afs_uint32 tlen, afs_uint32 *bytesread)
132 {
133 afs_int32 code;
134 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)r;
135 struct memCacheEntry *mceP = (struct memCacheEntry *)tfile;
136
137 *bytesread = 0;
138 code = afs_MemReadvBlk(mceP, offset, v->tiov, v->tnio, tlen);
139 if (code != tlen)
140 return -33;
141 *bytesread = code;
142 return 0;
143 }
144
145 afs_int32
146 rxfs_storeMemWrite(void *r, afs_uint32 l, afs_uint32 *byteswritten)
147 {
148 afs_int32 code;
149 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)r;
150
151 RX_AFS_GUNLOCK();
152 code = rx_Writev(v->call, v->tiov, v->tnio, l);
153 RX_AFS_GLOCK();
154 if (code != l) {
155 code = rx_Error(v->call);
156 return (code ? code : -33);
157 }
158 *byteswritten = code;
159 return 0;
160 }
161
162 afs_int32
163 rxfs_storeUfsWrite(void *r, afs_uint32 l, afs_uint32 *byteswritten)
164 {
165 afs_int32 code;
166 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)r;
167
168 RX_AFS_GUNLOCK();
169 code = rx_Write(v->call, v->tbuffer, l);
170 /* writing 0 bytes will
171 * push a short packet. Is that really what we want, just because the
172 * data didn't come back from the disk yet? Let's try it and see. */
173 RX_AFS_GLOCK();
174 if (code != l) {
175 code = rx_Error(v->call);
176 return (code ? code : -33);
177 }
178 *byteswritten = code;
179 return 0;
180 }
181
182 afs_int32
183 rxfs_storePadd(void *rock, afs_uint32 size)
184 {
185 afs_int32 code = 0;
186 afs_uint32 tlen;
187 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)rock;
188
189 if (!v->tbuffer)
190 v->tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
191 memset(v->tbuffer, 0, AFS_LRALLOCSIZ);
192
193 while (size) {
194 tlen = (size > AFS_LRALLOCSIZ ? AFS_LRALLOCSIZ : size);
195 RX_AFS_GUNLOCK();
196 code = rx_Write(v->call, v->tbuffer, tlen);
197 RX_AFS_GLOCK();
198
199 if (code != tlen)
200 return -33; /* XXX */
201 size -= tlen;
202 }
203 return 0;
204 }
205
206 afs_int32
207 rxfs_storeStatus(void *rock)
208 {
209 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)rock;
210
211 if (rx_GetRemoteStatus(v->call) & 1)
212 return 0;
213 return 1;
214 }
215
216 afs_int32
217 rxfs_storeClose(void *r, struct AFSFetchStatus *OutStatus, int *doProcessFS)
218 {
219 afs_int32 code;
220 struct AFSVolSync tsync;
221 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)r;
222
223 if (!v->call)
224 return -1;
225 RX_AFS_GUNLOCK();
226 #ifdef AFS_64BIT_CLIENT
227 if (!v->hasNo64bit)
228 code = EndRXAFS_StoreData64(v->call, OutStatus, &tsync);
229 else
230 #endif
231 code = EndRXAFS_StoreData(v->call, OutStatus, &tsync);
232 RX_AFS_GLOCK();
233 if (!code)
234 *doProcessFS = 1; /* Flag to run afs_ProcessFS() later on */
235
236 return code;
237 }
238
239 afs_int32
240 rxfs_storeDestroy(void **r, afs_int32 code)
241 {
242 struct rxfs_storeVariables *v = (struct rxfs_storeVariables *)*r;
243
244 *r = NULL;
245 if (v->call) {
246 RX_AFS_GUNLOCK();
247 code = rx_EndCall(v->call, code);
248 RX_AFS_GLOCK();
249 }
250 if (v->tbuffer)
251 osi_FreeLargeSpace(v->tbuffer);
252 if (v->tiov)
253 osi_FreeSmallSpace(v->tiov);
254 osi_FreeSmallSpace(v);
255 return code;
256 }
257
258 afs_int32
259 afs_GenericStoreProc(struct storeOps *ops, void *rock,
260 struct dcache *tdc, int *shouldwake,
261 afs_size_t *bytesXferred)
262 {
263 struct rxfs_storeVariables *svar = rock;
264 afs_uint32 tlen, bytesread, byteswritten;
265 afs_int32 code = 0;
266 int offset = 0;
267 afs_size_t size;
268 struct osi_file *tfile;
269
270 size = tdc->f.chunkBytes;
271
272 tfile = afs_CFileOpen(&tdc->f.inode);
273 osi_Assert(tfile);
274
275 while ( size > 0 ) {
276 code = (*ops->prepare)(rock, size, &tlen);
277 if ( code )
278 break;
279
280 code = (*ops->read)(rock, tfile, offset, tlen, &bytesread);
281 if (code)
282 break;
283
284 tlen = bytesread;
285 code = (*ops->write)(rock, tlen, &byteswritten);
286 if (code)
287 break;
288 #ifndef AFS_NOSTATS
289 *bytesXferred += byteswritten;
290 #endif /* AFS_NOSTATS */
291
292 offset += tlen;
293 size -= tlen;
294 /*
295 * if file has been locked on server, can allow
296 * store to continue
297 */
298 if (shouldwake && *shouldwake && ((*ops->status)(rock) == 0)) {
299 *shouldwake = 0; /* only do this once */
300 afs_wakeup(svar->vcache);
301 }
302 }
303 afs_CFileClose(tfile);
304
305 return code;
306 }
307
308 static
309 struct storeOps rxfs_storeUfsOps = {
310 #ifndef HAVE_STRUCT_LABEL_SUPPORT
311 rxfs_storeUfsPrepare,
312 rxfs_storeUfsRead,
313 rxfs_storeUfsWrite,
314 rxfs_storeStatus,
315 rxfs_storePadd,
316 rxfs_storeClose,
317 rxfs_storeDestroy,
318 afs_GenericStoreProc
319 #else
320 .prepare = rxfs_storeUfsPrepare,
321 .read = rxfs_storeUfsRead,
322 .write = rxfs_storeUfsWrite,
323 .status = rxfs_storeStatus,
324 .padd = rxfs_storePadd,
325 .close = rxfs_storeClose,
326 .destroy = rxfs_storeDestroy,
327 .storeproc = afs_GenericStoreProc
328 #endif
329 };
330
331 static
332 struct storeOps rxfs_storeMemOps = {
333 #ifndef HAVE_STRUCT_LABEL_SUPPORT
334 rxfs_storeMemPrepare,
335 rxfs_storeMemRead,
336 rxfs_storeMemWrite,
337 rxfs_storeStatus,
338 rxfs_storePadd,
339 rxfs_storeClose,
340 rxfs_storeDestroy,
341 afs_GenericStoreProc
342 #else
343 .prepare = rxfs_storeMemPrepare,
344 .read = rxfs_storeMemRead,
345 .write = rxfs_storeMemWrite,
346 .status = rxfs_storeStatus,
347 .padd = rxfs_storePadd,
348 .close = rxfs_storeClose,
349 .destroy = rxfs_storeDestroy,
350 .storeproc = afs_GenericStoreProc
351 #endif
352 };
353
354 afs_int32
355 rxfs_storeInit(struct vcache *avc, struct afs_conn *tc,
356 struct rx_connection *rxconn, afs_size_t base,
357 afs_size_t bytes, afs_size_t length,
358 int sync, struct storeOps **ops, void **rock)
359 {
360 afs_int32 code;
361 struct rxfs_storeVariables *v;
362
363 if ( !tc )
364 return -1;
365
366 v = osi_AllocSmallSpace(sizeof(struct rxfs_storeVariables));
367 if (!v)
368 osi_Panic("rxfs_storeInit: osi_AllocSmallSpace returned NULL\n");
369 memset(v, 0, sizeof(struct rxfs_storeVariables));
370
371 v->InStatus.ClientModTime = avc->f.m.Date;
372 v->InStatus.Mask = AFS_SETMODTIME;
373 v->vcache = avc;
374 if (sync & AFS_SYNC)
375 v->InStatus.Mask |= AFS_FSYNC;
376 RX_AFS_GUNLOCK();
377 v->call = rx_NewCall(rxconn);
378 if (v->call) {
379 #ifdef AFS_64BIT_CLIENT
380 if (!afs_serverHasNo64Bit(tc))
381 code = StartRXAFS_StoreData64(
382 v->call, (struct AFSFid*)&avc->f.fid.Fid,
383 &v->InStatus, base, bytes, length);
384 else {
385 if (length > 0xFFFFFFFF)
386 code = EFBIG;
387 else {
388 afs_int32 t1 = base, t2 = bytes, t3 = length;
389 code = StartRXAFS_StoreData(v->call,
390 (struct AFSFid *) &avc->f.fid.Fid,
391 &v->InStatus, t1, t2, t3);
392 }
393 v->hasNo64bit = 1;
394 }
395 #else /* AFS_64BIT_CLIENT */
396 code = StartRXAFS_StoreData(v->call, (struct AFSFid *)&avc->f.fid.Fid,
397 &v->InStatus, base, bytes, length);
398 #endif /* AFS_64BIT_CLIENT */
399 } else
400 code = -1;
401 RX_AFS_GLOCK();
402 if (code) {
403 *rock = v;
404 return rxfs_storeDestroy(rock, code);
405 }
406 if (cacheDiskType == AFS_FCACHE_TYPE_UFS) {
407 v->tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
408 if (!v->tbuffer)
409 osi_Panic
410 ("rxfs_storeInit: osi_AllocLargeSpace for iovecs returned NULL\n");
411 *ops = (struct storeOps *) &rxfs_storeUfsOps;
412 } else {
413 v->tiov = osi_AllocSmallSpace(sizeof(struct iovec) * RX_MAXIOVECS);
414 if (!v->tiov)
415 osi_Panic
416 ("rxfs_storeInit: osi_AllocSmallSpace for iovecs returned NULL\n");
417 *ops = (struct storeOps *) &rxfs_storeMemOps;
418 #ifdef notdef
419 /* do this at a higher level now -- it's a parameter */
420 /* for now, only do 'continue from close' code if file fits in one
421 * chunk. Could clearly do better: if only one modified chunk
422 * then can still do this. can do this on *last* modified chunk */
423 length = avc->f.m.Length - 1; /* byte position of last byte we'll store */
424 if (shouldWake) {
425 if (AFS_CHUNK(length) != 0)
426 *shouldWake = 0;
427 else
428 *shouldWake = 1;
429 }
430 #endif /* notdef */
431 }
432
433 *rock = (void *)v;
434 return 0;
435 }
436 unsigned int storeallmissing = 0;
437 /*!
438 * Called for each chunk upon store.
439 *
440 * \param avc Ptr to the vcache entry of the file being stored.
441 * \param dclist pointer to the list of dcaches
442 * \param bytes total number of bytes for the current operation
443 * \param anewDV Ptr to the dataversion after store
444 * \param doProcessFS pointer to the "do process FetchStatus" flag
445 * \param OutStatus pointer to the FetchStatus as returned by the fileserver
446 * \param nchunks number of dcaches to consider
447 * \param nomore copy of the "no more data" flag
448 * \param ops pointer to the block of storeOps to be used for this operation
449 * \param rock pointer to the opaque protocol-specific data of this operation
450 */
451 afs_int32
452 afs_CacheStoreDCaches(struct vcache *avc, struct dcache **dclist,
453 afs_size_t bytes, afs_hyper_t *anewDV, int *doProcessFS,
454 struct AFSFetchStatus *OutStatus, afs_uint32 nchunks,
455 int nomore, struct storeOps *ops, void *rock)
456 {
457 int *shouldwake = NULL;
458 unsigned int i;
459 int stored = 0;
460 afs_int32 code = 0;
461 afs_size_t bytesXferred;
462
463 #ifndef AFS_NOSTATS
464 osi_timeval_t xferStartTime; /*FS xfer start time */
465 afs_size_t bytesToXfer = 10000; /* # bytes to xfer */
466 #endif /* AFS_NOSTATS */
467 XSTATS_DECLS;
468 osi_Assert(nchunks != 0);
469
470 for (i = 0; i < nchunks && !code; i++) {
471 struct dcache *tdc = dclist[i];
472 afs_int32 size;
473
474 if (!tdc) {
475 afs_warn("afs: missing dcache!\n");
476 storeallmissing++;
477 continue; /* panic? */
478 }
479 size = tdc->f.chunkBytes;
480 afs_Trace4(afs_iclSetp, CM_TRACE_STOREALL2, ICL_TYPE_POINTER, avc,
481 ICL_TYPE_INT32, tdc->f.chunk, ICL_TYPE_INT32, tdc->index,
482 ICL_TYPE_INT32, afs_inode2trace(&tdc->f.inode));
483 shouldwake = 0;
484 if (nomore) {
485 if (avc->asynchrony == -1) {
486 if (afs_defaultAsynchrony > (bytes - stored))
487 shouldwake = &nomore;
488 }
489 else if ((afs_uint32) avc->asynchrony >= (bytes - stored))
490 shouldwake = &nomore;
491 }
492
493 afs_Trace4(afs_iclSetp, CM_TRACE_STOREPROC, ICL_TYPE_POINTER, avc,
494 ICL_TYPE_FID, &(avc->f.fid), ICL_TYPE_OFFSET,
495 ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_INT32, size);
496
497 AFS_STATCNT(CacheStoreProc);
498
499 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_STOREDATA);
500 avc->f.truncPos = AFS_NOTRUNC;
501 #ifndef AFS_NOSTATS
502 /*
503 * In this case, size is *always* the amount of data we'll be trying
504 * to ship here.
505 */
506 bytesToXfer = size;
507
508 osi_GetuTime(&xferStartTime);
509 #endif /* AFS_NOSTATS */
510 bytesXferred = 0;
511
512 code = (*ops->storeproc)(ops, rock, tdc, shouldwake,
513 &bytesXferred);
514
515 afs_Trace4(afs_iclSetp, CM_TRACE_STOREPROC, ICL_TYPE_POINTER, avc,
516 ICL_TYPE_FID, &(avc->f.fid), ICL_TYPE_OFFSET,
517 ICL_HANDLE_OFFSET(avc->f.m.Length), ICL_TYPE_INT32, size);
518
519 #ifndef AFS_NOSTATS
520 FillStoreStats(code, AFS_STATS_FS_XFERIDX_STOREDATA,
521 xferStartTime, bytesToXfer, bytesXferred);
522 #endif /* AFS_NOSTATS */
523
524 if ((tdc->f.chunkBytes < afs_OtherCSize)
525 && (i < (nchunks - 1)) && code == 0) {
526 code = (*ops->padd)(rock, afs_OtherCSize - tdc->f.chunkBytes);
527 }
528 stored += tdc->f.chunkBytes;
529 /* ideally, I'd like to unlock the dcache and turn
530 * off the writing bit here, but that would
531 * require being able to retry StoreAllSegments in
532 * the event of a failure. It only really matters
533 * if user can't read from a 'locked' dcache or
534 * one which has the writing bit turned on. */
535 }
536
537 if (!code) {
538 code = (*ops->close)(rock, OutStatus, doProcessFS);
539 /* if this succeeds, dv has been bumped. */
540 if (*doProcessFS) {
541 hadd32(*anewDV, 1);
542 }
543 XSTATS_END_TIME;
544 }
545 if (ops)
546 code = (*ops->destroy)(&rock, code);
547
548 /* if we errored, can't trust this. */
549 if (code)
550 *doProcessFS = 0;
551
552 return code;
553 }
554
555 #define lmin(a,b) (((a) < (b)) ? (a) : (b))
556 /*!
557 * Called upon store.
558 *
559 * \param dclist pointer to the list of dcaches
560 * \param avc Ptr to the vcache entry.
561 * \param areq Ptr to the request structure
562 * \param sync sync flag
563 * \param minj the chunk offset for this call
564 * \param high index of last dcache to store
565 * \param moredata the moredata flag
566 * \param anewDV Ptr to the dataversion after store
567 * \param amaxStoredLength Ptr to the amount of that is actually stored
568 *
569 * \note Environment: Nothing interesting.
570 */
571 int
572 afs_CacheStoreVCache(struct dcache **dcList, struct vcache *avc,
573 struct vrequest *areq, int sync, unsigned int minj,
574 unsigned int high, unsigned int moredata,
575 afs_hyper_t *anewDV, afs_size_t *amaxStoredLength)
576 {
577 afs_int32 code = 0;
578 struct storeOps *ops;
579 void * rock = NULL;
580 unsigned int i, j;
581
582 struct AFSFetchStatus OutStatus;
583 int doProcessFS = 0;
584 afs_size_t base, bytes, length;
585 int nomore;
586 unsigned int first = 0;
587 struct afs_conn *tc;
588 struct rx_connection *rxconn;
589
590 for (bytes = 0, j = 0; !code && j <= high; j++) {
591 if (dcList[j]) {
592 ObtainSharedLock(&(dcList[j]->lock), 629);
593 if (!bytes)
594 first = j;
595 bytes += dcList[j]->f.chunkBytes;
596 if ((dcList[j]->f.chunkBytes < afs_OtherCSize)
597 && (dcList[j]->f.chunk - minj < high)
598 && dcList[j + 1]) {
599 int sbytes = afs_OtherCSize - dcList[j]->f.chunkBytes;
600 bytes += sbytes;
601 }
602 }
603 if (bytes && (j == high || !dcList[j + 1])) {
604 afs_uint32 nchunks;
605 struct dcache **dclist = &dcList[first];
606 /* base = AFS_CHUNKTOBASE(dcList[first]->f.chunk); */
607 base = AFS_CHUNKTOBASE(first + minj);
608 /*
609 *
610 * take a list of dcache structs and send them all off to the server
611 * the list must be in order, and the chunks contiguous.
612 * Note - there is no locking done by this code currently. For
613 * safety's sake, xdcache could be locked over the entire call.
614 * However, that pretty well ties up all the threads. Meantime, all
615 * the chunks _MUST_ have their refcounts bumped.
616 * The writes done before a store back will clear setuid-ness
617 * in cache file.
618 * We can permit CacheStoreProc to wake up the user process IFF we
619 * are doing the last RPC for this close, ie, storing back the last
620 * set of contiguous chunks of a file.
621 */
622
623 nchunks = 1 + j - first;
624 nomore = !(moredata || (j != high));
625 length = lmin(avc->f.m.Length, avc->f.truncPos);
626 afs_Trace4(afs_iclSetp, CM_TRACE_STOREDATA64,
627 ICL_TYPE_FID, &avc->f.fid.Fid, ICL_TYPE_OFFSET,
628 ICL_HANDLE_OFFSET(base), ICL_TYPE_OFFSET,
629 ICL_HANDLE_OFFSET(bytes), ICL_TYPE_OFFSET,
630 ICL_HANDLE_OFFSET(length));
631
632 do {
633 tc = afs_Conn(&avc->f.fid, areq, 0, &rxconn);
634
635 #ifdef AFS_64BIT_CLIENT
636 restart:
637 #endif
638 code = rxfs_storeInit(avc, tc, rxconn, base, bytes, length,
639 sync, &ops, &rock);
640 if ( !code ) {
641 code = afs_CacheStoreDCaches(avc, dclist, bytes, anewDV,
642 &doProcessFS, &OutStatus,
643 nchunks, nomore, ops, rock);
644 }
645
646 #ifdef AFS_64BIT_CLIENT
647 if (code == RXGEN_OPCODE && !afs_serverHasNo64Bit(tc)) {
648 afs_serverSetNo64Bit(tc);
649 goto restart;
650 }
651 #endif /* AFS_64BIT_CLIENT */
652 } while (afs_Analyze
653 (tc, rxconn, code, &avc->f.fid, areq,
654 AFS_STATS_FS_RPCIDX_STOREDATA, SHARED_LOCK,
655 NULL));
656
657 /* put back all remaining locked dcache entries */
658 for (i = 0; i < nchunks; i++) {
659 struct dcache *tdc = dclist[i];
660 if (!code) {
661 if (afs_indexFlags[tdc->index] & IFDataMod) {
662 /*
663 * LOCKXXX -- should hold afs_xdcache(W) when
664 * modifying afs_indexFlags.
665 */
666 afs_indexFlags[tdc->index] &= ~IFDataMod;
667 afs_stats_cmperf.cacheCurrDirtyChunks--;
668 afs_indexFlags[tdc->index] &= ~IFDirtyPages;
669 if (sync & AFS_VMSYNC_INVAL) {
670 /* since we have invalidated all the pages of this
671 ** vnode by calling osi_VM_TryToSmush, we can
672 ** safely mark this dcache entry as not having
673 ** any pages. This vnode now becomes eligible for
674 ** reclamation by getDownD.
675 */
676 afs_indexFlags[tdc->index] &= ~IFAnyPages;
677 }
678 }
679 }
680 UpgradeSToWLock(&tdc->lock, 628);
681 tdc->f.states &= ~DWriting; /* correct? */
682 tdc->dflags |= DFEntryMod;
683 ReleaseWriteLock(&tdc->lock);
684 afs_PutDCache(tdc);
685 /* Mark the entry as released */
686 dclist[i] = NULL;
687 }
688
689 if (doProcessFS) {
690 /* Now copy out return params */
691 UpgradeSToWLock(&avc->lock, 28); /* keep out others for a while */
692 afs_ProcessFS(avc, &OutStatus, areq);
693 /* Keep last (max) size of file on server to see if
694 * we need to call afs_StoreMini to extend the file.
695 */
696 if (!moredata)
697 *amaxStoredLength = OutStatus.Length;
698 ConvertWToSLock(&avc->lock);
699 doProcessFS = 0;
700 }
701
702 if (code) {
703 for (j++; j <= high; j++) {
704 if (dcList[j]) {
705 ReleaseSharedLock(&(dcList[j]->lock));
706 afs_PutDCache(dcList[j]);
707 /* Releasing entry */
708 dcList[j] = NULL;
709 }
710 }
711 }
712
713 afs_Trace2(afs_iclSetp, CM_TRACE_STOREALLDCDONE,
714 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code);
715 bytes = 0;
716 }
717 }
718
719 return code;
720 }
721
722 /* rock and operations for RX_FILESERVER */
723
724 struct rxfs_fetchVariables {
725 struct rx_call *call;
726 char *tbuffer;
727 struct iovec *iov;
728 afs_int32 nio;
729 afs_int32 hasNo64bit;
730 afs_int32 iovno;
731 afs_int32 iovmax;
732 };
733
734 afs_int32
735 rxfs_fetchUfsRead(void *r, afs_uint32 size, afs_uint32 *bytesread)
736 {
737 afs_int32 code;
738 afs_uint32 tlen;
739 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
740
741 *bytesread = 0;
742 tlen = (size > AFS_LRALLOCSIZ ? AFS_LRALLOCSIZ : size);
743 RX_AFS_GUNLOCK();
744 code = rx_Read(v->call, v->tbuffer, tlen);
745 RX_AFS_GLOCK();
746 if (code <= 0)
747 return -34;
748 *bytesread = code;
749 return 0;
750 }
751
752 afs_int32
753 rxfs_fetchMemRead(void *r, afs_uint32 tlen, afs_uint32 *bytesread)
754 {
755 afs_int32 code;
756 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
757
758 *bytesread = 0;
759 RX_AFS_GUNLOCK();
760 code = rx_Readv(v->call, v->iov, &v->nio, RX_MAXIOVECS, tlen);
761 RX_AFS_GLOCK();
762 if (code <= 0)
763 return -34;
764 *bytesread = code;
765 return 0;
766 }
767
768
769 afs_int32
770 rxfs_fetchMemWrite(void *r, struct osi_file *fP, afs_uint32 offset,
771 afs_uint32 tlen, afs_uint32 *byteswritten)
772 {
773 afs_int32 code;
774 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
775 struct memCacheEntry *mceP = (struct memCacheEntry *)fP;
776
777 code = afs_MemWritevBlk(mceP, offset, v->iov, v->nio, tlen);
778 if (code != tlen) {
779 return EIO;
780 }
781 *byteswritten = code;
782 return 0;
783 }
784
785 afs_int32
786 rxfs_fetchUfsWrite(void *r, struct osi_file *fP, afs_uint32 offset,
787 afs_uint32 tlen, afs_uint32 *byteswritten)
788 {
789 afs_int32 code;
790 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
791
792 code = afs_osi_Write(fP, -1, v->tbuffer, tlen);
793 if (code != tlen) {
794 return EIO;
795 }
796 *byteswritten = code;
797 return 0;
798 }
799
800
801 afs_int32
802 rxfs_fetchClose(void *r, struct vcache *avc, struct dcache * adc,
803 struct afs_FetchOutput *o)
804 {
805 afs_int32 code;
806 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
807
808 if (!v->call)
809 return -1;
810
811 RX_AFS_GUNLOCK();
812 #ifdef AFS_64BIT_CLIENT
813 if (!v->hasNo64bit)
814 code = EndRXAFS_FetchData64(v->call, &o->OutStatus, &o->CallBack,
815 &o->tsync);
816 else
817 #endif
818 code = EndRXAFS_FetchData(v->call, &o->OutStatus, &o->CallBack,
819 &o->tsync);
820 code = rx_EndCall(v->call, code);
821 RX_AFS_GLOCK();
822
823 v->call = NULL;
824
825 return code;
826 }
827
828 afs_int32
829 rxfs_fetchDestroy(void **r, afs_int32 code)
830 {
831 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)*r;
832
833 *r = NULL;
834 if (v->call) {
835 RX_AFS_GUNLOCK();
836 code = rx_EndCall(v->call, code);
837 RX_AFS_GLOCK();
838 }
839 if (v->tbuffer)
840 osi_FreeLargeSpace(v->tbuffer);
841 if (v->iov)
842 osi_FreeSmallSpace(v->iov);
843 osi_FreeSmallSpace(v);
844 return code;
845 }
846
847 afs_int32
848 rxfs_fetchMore(void *r, afs_int32 *length, afs_uint32 *moredata)
849 {
850 afs_int32 code;
851 struct rxfs_fetchVariables *v = (struct rxfs_fetchVariables *)r;
852
853 /*
854 * The fetch protocol is extended for the AFS/DFS translator
855 * to allow multiple blocks of data, each with its own length,
856 * to be returned. As long as the top bit is set, there are more
857 * blocks expected.
858 *
859 * We do not do this for AFS file servers because they sometimes
860 * return large negative numbers as the transfer size.
861 */
862 if (*moredata) {
863 RX_AFS_GUNLOCK();
864 code = rx_Read(v->call, (void *)length, sizeof(afs_int32));
865 RX_AFS_GLOCK();
866 *length = ntohl(*length);
867 if (code != sizeof(afs_int32)) {
868 code = rx_Error(v->call);
869 *moredata = 0;
870 return (code ? code : -1); /* try to return code, not -1 */
871 }
872 }
873 *moredata = *length & 0x80000000;
874 *length &= ~0x80000000;
875 return 0;
876 }
877
878 static
879 struct fetchOps rxfs_fetchUfsOps = {
880 rxfs_fetchMore,
881 rxfs_fetchUfsRead,
882 rxfs_fetchUfsWrite,
883 rxfs_fetchClose,
884 rxfs_fetchDestroy
885 };
886
887 static
888 struct fetchOps rxfs_fetchMemOps = {
889 rxfs_fetchMore,
890 rxfs_fetchMemRead,
891 rxfs_fetchMemWrite,
892 rxfs_fetchClose,
893 rxfs_fetchDestroy
894 };
895
896 afs_int32
897 rxfs_fetchInit(struct afs_conn *tc, struct rx_connection *rxconn,
898 struct vcache *avc, afs_offs_t base,
899 afs_uint32 size, afs_int32 *alength, struct dcache *adc,
900 struct osi_file *fP, struct fetchOps **ops, void **rock)
901 {
902 struct rxfs_fetchVariables *v;
903 int code = 0;
904 #ifdef AFS_64BIT_CLIENT
905 afs_uint32 length_hi = 0;
906 #endif
907 afs_uint32 length = 0, bytes;
908
909 v = (struct rxfs_fetchVariables *)
910 osi_AllocSmallSpace(sizeof(struct rxfs_fetchVariables));
911 if (!v)
912 osi_Panic("rxfs_fetchInit: osi_AllocSmallSpace returned NULL\n");
913 memset(v, 0, sizeof(struct rxfs_fetchVariables));
914
915 RX_AFS_GUNLOCK();
916 v->call = rx_NewCall(rxconn);
917 RX_AFS_GLOCK();
918 if (v->call) {
919 #ifdef AFS_64BIT_CLIENT
920 afs_size_t length64; /* as returned from server */
921 if (!afs_serverHasNo64Bit(tc)) {
922 afs_uint64 llbytes = size;
923 RX_AFS_GUNLOCK();
924 code = StartRXAFS_FetchData64(v->call,
925 (struct AFSFid *) &avc->f.fid.Fid,
926 base, llbytes);
927 if (code != 0) {
928 RX_AFS_GLOCK();
929 afs_Trace2(afs_iclSetp, CM_TRACE_FETCH64CODE,
930 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code);
931 } else {
932 bytes = rx_Read(v->call, (char *)&length_hi, sizeof(afs_int32));
933 RX_AFS_GLOCK();
934 if (bytes == sizeof(afs_int32)) {
935 length_hi = ntohl(length_hi);
936 } else {
937 RX_AFS_GUNLOCK();
938 code = rx_EndCall(v->call, RX_PROTOCOL_ERROR);
939 RX_AFS_GLOCK();
940 v->call = NULL;
941 }
942 }
943 }
944 if (code == RXGEN_OPCODE || afs_serverHasNo64Bit(tc)) {
945 if (base > 0x7FFFFFFF) {
946 code = EFBIG;
947 } else {
948 afs_uint32 pos;
949 pos = base;
950 RX_AFS_GUNLOCK();
951 if (!v->call)
952 v->call = rx_NewCall(rxconn);
953 code =
954 StartRXAFS_FetchData(
955 v->call, (struct AFSFid*)&avc->f.fid.Fid,
956 pos, size);
957 RX_AFS_GLOCK();
958 }
959 afs_serverSetNo64Bit(tc);
960 v->hasNo64bit = 1;
961 }
962 if (!code) {
963 RX_AFS_GUNLOCK();
964 bytes = rx_Read(v->call, (char *)&length, sizeof(afs_int32));
965 RX_AFS_GLOCK();
966 if (bytes == sizeof(afs_int32))
967 length = ntohl(length);
968 else {
969 RX_AFS_GUNLOCK();
970 code = rx_EndCall(v->call, RX_PROTOCOL_ERROR);
971 v->call = NULL;
972 length = 0;
973 RX_AFS_GLOCK();
974 }
975 }
976 FillInt64(length64, length_hi, length);
977
978 if (!code) {
979 /* Check if the fileserver said our length is bigger than can fit
980 * in a signed 32-bit integer. If it is, we can't handle that, so
981 * error out. */
982 if (length64 > MAX_AFS_INT32) {
983 static int warned;
984 if (!warned) {
985 warned = 1;
986 afs_warn("afs: Warning: FetchData64 returned too much data "
987 "(length64 %u.%u); this should not happen! "
988 "Aborting fetch request.\n",
989 length_hi, length);
990 }
991 RX_AFS_GUNLOCK();
992 code = rx_EndCall(v->call, RX_PROTOCOL_ERROR);
993 v->call = NULL;
994 length = 0;
995 RX_AFS_GLOCK();
996 code = code != 0 ? code : EIO;
997 }
998 }
999
1000 if (!code) {
1001 /* Check if the fileserver said our length was negative. If it
1002 * is, just treat it as a 0 length, since some older fileservers
1003 * returned negative numbers when they meant to return 0. Note
1004 * that we must do this in this 64-bit-specific block, since
1005 * length64 being negative will screw up our conversion to the
1006 * 32-bit 'alength' below. */
1007 if (length64 < 0) {
1008 length_hi = length = 0;
1009 FillInt64(length64, 0, 0);
1010 }
1011 }
1012
1013 afs_Trace3(afs_iclSetp, CM_TRACE_FETCH64LENG,
1014 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code,
1015 ICL_TYPE_OFFSET,
1016 ICL_HANDLE_OFFSET(length64));
1017 if (!code)
1018 *alength = length;
1019 #else /* AFS_64BIT_CLIENT */
1020 RX_AFS_GUNLOCK();
1021 code = StartRXAFS_FetchData(v->call, (struct AFSFid *)&avc->f.fid.Fid,
1022 base, size);
1023 RX_AFS_GLOCK();
1024 if (code == 0) {
1025 RX_AFS_GUNLOCK();
1026 bytes =
1027 rx_Read(v->call, (char *)&length, sizeof(afs_int32));
1028 RX_AFS_GLOCK();
1029 if (bytes == sizeof(afs_int32)) {
1030 *alength = ntohl(length);
1031 if (*alength < 0) {
1032 /* Older fileservers can return a negative length when they
1033 * meant to return 0; just assume negative lengths were
1034 * meant to be 0 lengths. */
1035 *alength = 0;
1036 }
1037 } else {
1038 code = rx_EndCall(v->call, RX_PROTOCOL_ERROR);
1039 v->call = NULL;
1040 }
1041 }
1042 #endif /* AFS_64BIT_CLIENT */
1043 } else
1044 code = -1;
1045
1046 /* We need to cast here, in order to avoid issues if *alength is
1047 * negative. Some, older, fileservers can return a negative length,
1048 * which the rest of the code deals correctly with. */
1049 if (code == 0 && *alength > (afs_int32) size) {
1050 /* The fileserver told us it is going to send more data than we
1051 * requested. It shouldn't do that, and accepting that much data
1052 * can make us take up more cache space than we're supposed to,
1053 * so error. */
1054 static int warned;
1055 if (!warned) {
1056 warned = 1;
1057 afs_warn("afs: Warning: FetchData64 returned more data than "
1058 "requested (requested %ld, got %ld); this should not "
1059 "happen! Aborting fetch request.\n",
1060 (long)size, (long)*alength);
1061 }
1062 RX_AFS_GUNLOCK();
1063 code = rx_EndCall(v->call, RX_PROTOCOL_ERROR);
1064 RX_AFS_GLOCK();
1065 v->call = NULL;
1066 code = EIO;
1067 }
1068
1069 if (code) {
1070 *rock = v;
1071 return rxfs_fetchDestroy(rock, code);
1072 }
1073 if (cacheDiskType == AFS_FCACHE_TYPE_UFS) {
1074 v->tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
1075 if (!v->tbuffer)
1076 osi_Panic("rxfs_fetchInit: osi_AllocLargeSpace for iovecs returned NULL\n");
1077 osi_Assert(WriteLocked(&adc->lock));
1078 fP->offset = 0;
1079 *ops = (struct fetchOps *) &rxfs_fetchUfsOps;
1080 }
1081 else {
1082 afs_Trace4(afs_iclSetp, CM_TRACE_MEMFETCH, ICL_TYPE_POINTER, avc,
1083 ICL_TYPE_POINTER, fP, ICL_TYPE_OFFSET,
1084 ICL_HANDLE_OFFSET(base), ICL_TYPE_INT32, length);
1085 /*
1086 * We need to alloc the iovecs on the heap so that they are "pinned"
1087 * rather than declare them on the stack - defect 11272
1088 */
1089 v->iov = osi_AllocSmallSpace(sizeof(struct iovec) * RX_MAXIOVECS);
1090 if (!v->iov)
1091 osi_Panic("rxfs_fetchInit: osi_AllocSmallSpace for iovecs returned NULL\n");
1092 *ops = (struct fetchOps *) &rxfs_fetchMemOps;
1093 }
1094 *rock = (void *)v;
1095 return 0;
1096 }
1097
1098
1099 /*!
1100 * Routine called on fetch; also tells people waiting for data
1101 * that more has arrived.
1102 *
1103 * \param tc Ptr to the AFS connection structure.
1104 * \param rxconn Ptr to the Rx connection structure.
1105 * \param fP File descriptor for the cache file.
1106 * \param base Base offset to fetch.
1107 * \param adc Ptr to the dcache entry for the file, write-locked.
1108 * \param avc Ptr to the vcache entry for the file.
1109 * \param size Amount of data that should be fetched.
1110 * \param tsmall Ptr to the afs_FetchOutput structure.
1111 *
1112 * \note Environment: Nothing interesting.
1113 */
1114 int
1115 afs_CacheFetchProc(struct afs_conn *tc, struct rx_connection *rxconn,
1116 struct osi_file *fP, afs_size_t base,
1117 struct dcache *adc, struct vcache *avc, afs_int32 size,
1118 struct afs_FetchOutput *tsmall)
1119 {
1120 afs_int32 code;
1121 afs_int32 length;
1122 afs_uint32 bytesread, byteswritten;
1123 struct fetchOps *ops = NULL;
1124 void *rock = NULL;
1125 afs_uint32 moredata = 0;
1126 int offset = 0;
1127
1128 XSTATS_DECLS;
1129 #ifndef AFS_NOSTATS
1130 osi_timeval_t xferStartTime; /*FS xfer start time */
1131 afs_size_t bytesToXfer = 0, bytesXferred = 0;
1132 #endif
1133
1134 AFS_STATCNT(CacheFetchProc);
1135
1136 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_FETCHDATA);
1137
1138 /*
1139 * Locks held:
1140 * avc->lock(R) if setLocks && !slowPass
1141 * avc->lock(W) if !setLocks || slowPass
1142 * adc->lock(W)
1143 */
1144 code = rxfs_fetchInit(
1145 tc, rxconn, avc, base, size, &length, adc, fP, &ops, &rock);
1146
1147 #ifndef AFS_NOSTATS
1148 osi_GetuTime(&xferStartTime);
1149 #endif /* AFS_NOSTATS */
1150
1151 adc->validPos = base;
1152
1153 if (code) {
1154 goto done;
1155 }
1156
1157 do {
1158 if (avc->f.states & CForeign) {
1159 code = (*ops->more)(rock, &length, &moredata);
1160 if ( code )
1161 goto done;
1162 }
1163 #ifndef AFS_NOSTATS
1164 bytesToXfer += length;
1165 #endif /* AFS_NOSTATS */
1166 while (length > 0) {
1167 #ifdef RX_KERNEL_TRACE
1168 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1169 "before rx_Read");
1170 #endif
1171 code = (*ops->read)(rock, length, &bytesread);
1172 #ifdef RX_KERNEL_TRACE
1173 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1174 "after rx_Read");
1175 #endif
1176 #ifndef AFS_NOSTATS
1177 bytesXferred += bytesread;
1178 #endif /* AFS_NOSTATS */
1179 if ( code ) {
1180 afs_Trace3(afs_iclSetp, CM_TRACE_FETCH64READ,
1181 ICL_TYPE_POINTER, avc, ICL_TYPE_INT32, code,
1182 ICL_TYPE_INT32, length);
1183 code = -34;
1184 goto done;
1185 }
1186 code = (*ops->write)(rock, fP, offset, bytesread, &byteswritten);
1187 if ( code )
1188 goto done;
1189 offset += bytesread;
1190 base += bytesread;
1191 length -= bytesread;
1192 adc->validPos = base;
1193 if (afs_osi_Wakeup(&adc->validPos) == 0)
1194 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAKE, ICL_TYPE_STRING,
1195 __FILE__, ICL_TYPE_INT32, __LINE__,
1196 ICL_TYPE_POINTER, adc, ICL_TYPE_INT32,
1197 adc->dflags);
1198 }
1199 code = 0;
1200 } while (moredata);
1201 done:
1202 if (!code)
1203 code = (*ops->close)(rock, avc, adc, tsmall);
1204 if (ops)
1205 code = (*ops->destroy)(&rock, code);
1206
1207 #ifndef AFS_NOSTATS
1208 FillStoreStats(code, AFS_STATS_FS_XFERIDX_FETCHDATA, xferStartTime,
1209 bytesToXfer, bytesXferred);
1210 #endif
1211 XSTATS_END_TIME;
1212 return code;
1213 }