root / xseg / peers / user / cached.c @ 21c0e7f3
History | View | Annotate | Download (22.7 kB)
1 |
/*
|
---|---|
2 |
* Copyright 2013 GRNET S.A. All rights reserved.
|
3 |
*
|
4 |
* Redistribution and use in source and binary forms, with or
|
5 |
* without modification, are permitted provided that the following
|
6 |
* conditions are met:
|
7 |
*
|
8 |
* 1. Redistributions of source code must retain the above
|
9 |
* copyright notice, this list of conditions and the following
|
10 |
* disclaimer.
|
11 |
* 2. Redistributions in binary form must reproduce the above
|
12 |
* copyright notice, this list of conditions and the following
|
13 |
* disclaimer in the documentation and/or other materials
|
14 |
* provided with the distribution.
|
15 |
*
|
16 |
* THIS SOFTWARE IS PROVIDED BY GRNET S.A. ``AS IS'' AND ANY EXPRESS
|
17 |
* OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
|
18 |
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
19 |
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GRNET S.A OR
|
20 |
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
21 |
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
22 |
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF
|
23 |
* USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
|
24 |
* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
25 |
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
|
26 |
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
|
27 |
* POSSIBILITY OF SUCH DAMAGE.
|
28 |
*
|
29 |
* The views and conclusions contained in the software and
|
30 |
* documentation are those of the authors and should not be
|
31 |
* interpreted as representing official policies, either expressed
|
32 |
* or implied, of GRNET S.A.
|
33 |
*/
|
34 |
|
35 |
#include <stdio.h> |
36 |
#include <unistd.h> |
37 |
#include <sys/types.h> |
38 |
#include <pthread.h> |
39 |
#include <xseg/xseg.h> |
40 |
#include <peer.h> |
41 |
#include <time.h> |
42 |
#include <xtypes/xlock.h> |
43 |
#include <xtypes/xq.h> |
44 |
#include <xtypes/xhash.h> |
45 |
#include <xtypes/xworkq.h> |
46 |
#include <xtypes/xwaitq.h> |
47 |
#include <xseg/protocol.h> |
48 |
#include <xtypes/xcache.h> |
49 |
|
50 |
#define MAX_ARG_LEN 12 |
51 |
|
52 |
/* bucket states */
|
53 |
#define INVALID 0 |
54 |
#define LOADING 1 |
55 |
#define VALID 2 |
56 |
#define DIRTY 3 |
57 |
#define WRITING 4 |
58 |
|
59 |
#define bucket_readable(__status) \
|
60 |
(__status == VALID || __status == DIRTY || __status == WRITING) |
61 |
|
62 |
/* write policies */
|
63 |
/*FIXME: Shouldn't they be on a flag?*/
|
64 |
#define WRITETHROUGH 1 |
65 |
#define WRITEBACK 2 |
66 |
|
67 |
|
68 |
/* object states */
|
69 |
#define INVALIDATED (1 << 0) |
70 |
|
71 |
|
72 |
/* cio states */
|
73 |
#define CIO_FAILED 1 |
74 |
#define CIO_ACCEPTED 2 |
75 |
#define CIO_READING 3 |
76 |
|
77 |
#define BUCKET_SIZE_QUANTUM 4096 |
78 |
|
79 |
struct cache_io {
|
80 |
uint32_t state; |
81 |
xcache_handler h; |
82 |
uint32_t pending_reqs; |
83 |
struct work work;
|
84 |
}; |
85 |
|
86 |
struct cached {
|
87 |
struct xcache *cache;
|
88 |
uint64_t cache_size; /*Number of objects*/
|
89 |
uint64_t max_req_size; |
90 |
uint32_t object_size; /*Bytes*/ /*TODO: change this to buckets_per_object*/ |
91 |
uint32_t bucket_size; /*In bytes*/
|
92 |
uint32_t buckets_per_object; |
93 |
xport bportno; |
94 |
int write_policy;
|
95 |
//scheduler
|
96 |
}; |
97 |
|
98 |
struct ce {
|
99 |
unsigned char *data; |
100 |
uint32_t *status; |
101 |
struct xwaitq *waitq;
|
102 |
uint32_t flags; |
103 |
struct xlock lock;
|
104 |
struct xworkq workq;
|
105 |
struct peer_req pr;
|
106 |
}; |
107 |
|
108 |
|
109 |
/*
|
110 |
* Helper functions
|
111 |
*/
|
112 |
|
113 |
static inline struct cached * __get_cached(struct peerd *peer) |
114 |
{ |
115 |
return (struct cached *) peer->priv; |
116 |
} |
117 |
|
118 |
static inline struct cache_io * __get_cache_io(struct peer_req *pr) |
119 |
{ |
120 |
return (struct cache_io *) pr->priv; |
121 |
} |
122 |
|
123 |
static inline uint32_t __calculate_size(uint32_t start, uint32_t end, |
124 |
struct cached *cached)
|
125 |
{ |
126 |
return (end - start + 1) * cached->bucket_size; |
127 |
} |
128 |
|
129 |
static inline uint32_t __calculate_offset(uint32_t start, |
130 |
struct cached *cached)
|
131 |
{ |
132 |
return start * cached->bucket_size;
|
133 |
} |
134 |
|
135 |
static uint32_t __get_bucket(struct cached *cache, uint64_t offset) |
136 |
{ |
137 |
return (offset / cache->bucket_size);
|
138 |
} |
139 |
|
140 |
static int is_loading(void *arg) |
141 |
{ |
142 |
uint32_t *status = (uint32_t *)arg; |
143 |
return (*status == LOADING);
|
144 |
} |
145 |
|
146 |
static void print_cached(struct cached *cached) |
147 |
{ |
148 |
if (!cached) {
|
149 |
XSEGLOG2(&lc, W, "Struct cached is NULL\n");
|
150 |
return;
|
151 |
} |
152 |
|
153 |
XSEGLOG2(&lc, I, "Struct cached fields:\n"
|
154 |
" cache = %p\n"
|
155 |
" cache_size = %lu\n"
|
156 |
" max_req_size = %lu\n"
|
157 |
" object_size = %lu\n"
|
158 |
" bucket_size = %lu\n"
|
159 |
" buckets = %lu\n"
|
160 |
" Bportno = %d\n",
|
161 |
cached->cache, cached->cache_size, cached->max_req_size, |
162 |
cached->object_size, cached->bucket_size, |
163 |
cached->buckets_per_object, cached->bportno); |
164 |
} |
165 |
|
166 |
|
167 |
/*
|
168 |
* Convert string to size in bytes.
|
169 |
* If syntax is invalid, return 0. Values such as zero and non-integer
|
170 |
* multiples of segment's page size should not be accepted.
|
171 |
*/
|
172 |
uint64_t str2num(char *str)
|
173 |
{ |
174 |
char *unit;
|
175 |
uint64_t num; |
176 |
|
177 |
num = strtoll(str, &unit, 10);
|
178 |
if (strlen(unit) > 1) //Invalid syntax |
179 |
return 0; |
180 |
else if (strlen(unit) < 1) //Plain number in bytes |
181 |
return num;
|
182 |
|
183 |
switch (*unit) {
|
184 |
case 'g': |
185 |
case 'G': |
186 |
num *= 1024;
|
187 |
case 'm': |
188 |
case 'M': |
189 |
num *= 1024;
|
190 |
case 'k': |
191 |
case 'K': |
192 |
num *= 1024;
|
193 |
break;
|
194 |
default:
|
195 |
num = 0;
|
196 |
} |
197 |
return num;
|
198 |
} |
199 |
|
200 |
static int rw_range(struct peerd *peer, struct peer_req *pr, int action, |
201 |
uint32_t start, uint32_t end) |
202 |
{ |
203 |
struct cached *cached = __get_cached(peer);
|
204 |
struct cache_io *cio = __get_cache_io(pr);
|
205 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
206 |
/*
|
207 |
* get_req
|
208 |
* prep_req
|
209 |
* read or write data
|
210 |
* associate req with pr
|
211 |
* submit
|
212 |
* signal
|
213 |
*/
|
214 |
struct xseg_request *req;
|
215 |
struct xseg_request *req_old = pr->req;
|
216 |
struct xseg *xseg = peer->xseg;
|
217 |
//FIXME: Which of those for srcport?
|
218 |
//xport srcport = peer->portno_start;
|
219 |
//FIXME: pr->portnumber
|
220 |
xport srcport = pr->portno; |
221 |
xport dstport = cached->bportno; |
222 |
xport p; |
223 |
char *req_target;
|
224 |
int r;
|
225 |
uint32_t i; |
226 |
|
227 |
req = xseg_get_request(xseg, srcport, dstport, X_ALLOC); |
228 |
if (!req) {
|
229 |
XSEGLOG2(&lc, W, "Cannot get request\n");
|
230 |
return -1; |
231 |
} |
232 |
|
233 |
//FIXME: swap args
|
234 |
req->size = __calculate_size(start, end, cached); |
235 |
req->offset = __calculate_offset(start, cached); |
236 |
req->op = req_old->op; |
237 |
|
238 |
//Allocate enough space for the data and the target's name
|
239 |
r = xseg_prep_request(xseg, req, XSEG_MAX_TARGETLEN, req->size); |
240 |
if (r < 0) { |
241 |
XSEGLOG2(&lc, W, "Cannot prepare request! (%lu, %llu)\n",
|
242 |
XSEG_MAX_TARGETLEN, (unsigned long long)req->size); |
243 |
goto put_xseg_request;
|
244 |
} |
245 |
|
246 |
//Paste target name
|
247 |
req_target = xseg_get_target(xseg, req); |
248 |
char *req_old_target = xseg_get_target(xseg, req_old);
|
249 |
strncpy(req_target, req_old_target, XSEG_MAX_TARGETLEN); |
250 |
|
251 |
if (req->op == X_WRITE) {
|
252 |
/*
|
253 |
* //Paste data
|
254 |
* req_data = xseg_get_data(xseg, req);
|
255 |
* memcpy(req_data, req_old->data, size);
|
256 |
*/
|
257 |
} else {
|
258 |
for (i=start; i<=end; i++){
|
259 |
ce->status[i] = LOADING; |
260 |
} |
261 |
} |
262 |
|
263 |
/*
|
264 |
* We can avoid allocating a new peer_request, if we take extra care
|
265 |
* when we receive ours back.
|
266 |
*/
|
267 |
#if 0
|
268 |
pr = alloc_peer_req(peer);
|
269 |
if (!pr) {
|
270 |
XSEGLOG2(&lc, W, "Cannot allocate peer request (%ld remaining)\n",
|
271 |
peer->nr_ops - xq_count(&peer->free_reqs));
|
272 |
goto put_xseg_request;
|
273 |
}
|
274 |
pr->peer = peer;
|
275 |
pr->portno = srcport;
|
276 |
pr->req = req;
|
277 |
#endif
|
278 |
|
279 |
r = xseg_set_req_data(xseg, req, pr); |
280 |
if (r < 0) { |
281 |
XSEGLOG2(&lc, W, "Cannot set request data\n");
|
282 |
goto put_peer_request;
|
283 |
} |
284 |
|
285 |
p = xseg_submit(xseg, req, srcport, X_ALLOC); |
286 |
if (p == NoPort) {
|
287 |
XSEGLOG2(&lc, W, "Cannot submit request\n");
|
288 |
goto put_peer_request;
|
289 |
} |
290 |
|
291 |
r = xseg_signal(xseg, p); |
292 |
|
293 |
return 0; |
294 |
|
295 |
put_peer_request:
|
296 |
free(pr->priv); |
297 |
free_peer_req(peer, pr); |
298 |
put_xseg_request:
|
299 |
if (xseg_put_request(xseg, req, srcport))
|
300 |
XSEGLOG2(&lc, W, "Cannot put request\n");
|
301 |
return -1; |
302 |
|
303 |
return 0; |
304 |
} |
305 |
|
306 |
|
307 |
int on_init(void *c, void *e) |
308 |
{ |
309 |
uint32_t i; |
310 |
struct peerd *peer = (struct peerd *)c; |
311 |
struct cached *cached = peer->priv;
|
312 |
struct ce *ce = (struct ce *)e; |
313 |
ce->flags = 0;
|
314 |
memset(ce->data, 0, cached->object_size);
|
315 |
for (i = 0; i < cached->buckets_per_object; i++) { |
316 |
ce->status[i] = INVALID; |
317 |
} |
318 |
xlock_release(&ce->lock); |
319 |
return 0; |
320 |
} |
321 |
|
322 |
void on_put(void *c, void *e) |
323 |
{ |
324 |
struct peerd *peer = (struct peerd *)c; |
325 |
struct cached *cached = peer->priv;
|
326 |
struct ce *ce = (struct ce *)e; |
327 |
//since we are the last referrer to the cache entry
|
328 |
//no lock is needed.
|
329 |
|
330 |
uint32_t start, end, i = 0;
|
331 |
if (cached->write_policy == WRITETHROUGH || ce->flags & INVALIDATED)
|
332 |
return;
|
333 |
//write all dirty buckets.
|
334 |
while(i < cached->buckets_per_object){
|
335 |
if (ce->status[i] != DIRTY){
|
336 |
i++; |
337 |
continue;
|
338 |
} |
339 |
start = i; |
340 |
while (i < cached->buckets_per_object &&
|
341 |
(i-start)*cached->bucket_size < cached->max_req_size && |
342 |
ce->status[i] == DIRTY){ |
343 |
i++; |
344 |
} |
345 |
end = i; |
346 |
//problem: no assocciated pr
|
347 |
//maybe put one in cache entry
|
348 |
//rw_range(cached, ce, 1, start, end);
|
349 |
} |
350 |
} |
351 |
|
352 |
void * init_node(void *c) |
353 |
{ |
354 |
int i;
|
355 |
struct peerd *peer = (struct peerd *)c; |
356 |
struct cached *cached = peer->priv;
|
357 |
|
358 |
struct ce *ce = malloc(sizeof(struct ce)); |
359 |
if (!ce)
|
360 |
goto ce_fail;
|
361 |
xlock_release(&ce->lock); |
362 |
|
363 |
ce->data = malloc(sizeof(unsigned char) * cached->object_size); |
364 |
ce->status = malloc(sizeof(uint32_t) * cached->buckets_per_object);
|
365 |
ce->waitq = malloc(sizeof(struct xwaitq) * cached->buckets_per_object); |
366 |
if (!ce->data || !ce->status || !ce->waitq)
|
367 |
goto ce_fields_fail;
|
368 |
|
369 |
ce->pr.peer = peer; |
370 |
for (i = 0; i < cached->buckets_per_object; i++) { |
371 |
xwaitq_init(&ce->waitq[i], is_loading, &ce->status[i], |
372 |
XWAIT_SIGNAL_ONE); |
373 |
} |
374 |
xworkq_init(&ce->workq, &ce->lock, 0);
|
375 |
return ce;
|
376 |
|
377 |
ce_fields_fail:
|
378 |
free(ce->data); |
379 |
free(ce->status); |
380 |
free(ce->waitq); |
381 |
free(ce); |
382 |
ce_fail:
|
383 |
perror("malloc");
|
384 |
return NULL; |
385 |
} |
386 |
|
387 |
struct xcache_ops c_ops = {
|
388 |
.on_init = on_init, |
389 |
.on_put = on_put, |
390 |
.on_node_init = init_node |
391 |
}; |
392 |
|
393 |
static uint32_t __get_next_invalid(struct ce *ce, uint32_t start, |
394 |
uint32_t limit) |
395 |
{ |
396 |
uint32_t end = start + 1;
|
397 |
while (end <= limit && ce->status[end] == INVALID)
|
398 |
end++; |
399 |
return end - 1; |
400 |
} |
401 |
|
402 |
static void cached_fail(struct peerd *peer, struct peer_req *pr) |
403 |
{ |
404 |
struct cached *cached = __get_cached(peer);
|
405 |
struct cache_io *cio = __get_cache_io(pr);
|
406 |
if (cio->h != NoEntry){
|
407 |
xcache_put(cached->cache, cio->h); |
408 |
} |
409 |
cio->h = NoEntry; |
410 |
fail(peer, pr); |
411 |
} |
412 |
|
413 |
static void cached_complete(struct peerd *peer, struct peer_req *pr) |
414 |
{ |
415 |
struct cached *cached = __get_cached(peer);
|
416 |
struct cache_io *cio = __get_cache_io(pr);
|
417 |
if (cio->h != NoEntry){
|
418 |
xcache_put(cached->cache, cio->h); |
419 |
} |
420 |
cio->h = NoEntry; |
421 |
complete(peer, pr); |
422 |
} |
423 |
|
424 |
static void handle_read(void *arg); |
425 |
//is this necessary?
|
426 |
static void status_changed(void *arg) |
427 |
{ |
428 |
/*
|
429 |
* In this context we hold a reference to the cache entry.
|
430 |
*
|
431 |
* This function gets called only after the bucket at which the
|
432 |
* current peer_req is waiting, has finished loading of failed.
|
433 |
*
|
434 |
* Assumptions:
|
435 |
* Each pr waits only at one bucket at any time. That means that
|
436 |
* under no circumstances, this function get called simutaneously
|
437 |
* for the same pr.
|
438 |
*/
|
439 |
struct peer_req *pr = (struct peer_req *)arg; |
440 |
struct peerd *peer = pr->peer;
|
441 |
struct cached *cached = __get_cached(peer);
|
442 |
struct cache_io *cio = __get_cache_io(pr);
|
443 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
444 |
|
445 |
if (xworkq_enqueue(&ce->workq, handle_read, (void *)pr) < 0){ |
446 |
//FAIL or mark as failed ? are we the last?
|
447 |
if (cio->pending_reqs){
|
448 |
// cannot put here, since there are outstanding reqs to
|
449 |
// be received.
|
450 |
// Simply mark pr as failed.
|
451 |
cio->state = CIO_FAILED; |
452 |
} else {
|
453 |
//safe to fail here, since there is no pending action on
|
454 |
//this pr.
|
455 |
cached_fail(peer, pr); |
456 |
} |
457 |
} |
458 |
} |
459 |
|
460 |
static void handle_read(void *arg) |
461 |
{ |
462 |
/*
|
463 |
* In this context we hold a reference to the cache entry and
|
464 |
* the assocciated lock
|
465 |
*/
|
466 |
|
467 |
struct peer_req *pr = (struct peer_req *)arg; |
468 |
struct peerd *peer = pr->peer;
|
469 |
struct cached *cached = __get_cached(peer);
|
470 |
struct cache_io *cio = __get_cache_io(pr);
|
471 |
struct xseg_request *req = pr->req;
|
472 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
473 |
|
474 |
uint32_t start_bucket, end_bucket, next; |
475 |
uint32_t i, b, limit; |
476 |
|
477 |
uint32_t pending_buckets = 0;
|
478 |
|
479 |
XSEGLOG2(&lc, E, "Handle read started for %p (ce: %p)", pr, ce );
|
480 |
if (cio->state == CIO_FAILED)
|
481 |
goto out;
|
482 |
|
483 |
b = __get_bucket(cached, req->offset); |
484 |
limit = __get_bucket(cached, req->offset + req->size); |
485 |
//assert limit < cached->object_size
|
486 |
|
487 |
for (i = b; i <= limit; i++) {
|
488 |
if (bucket_readable(ce->status[i]))
|
489 |
continue;
|
490 |
if (ce->status[i] != LOADING){
|
491 |
start_bucket = i; |
492 |
end_bucket = __get_next_invalid(ce, start_bucket, limit); |
493 |
i = end_bucket; |
494 |
if (rw_range(peer, pr, 0, start_bucket, end_bucket) < 0){ |
495 |
cio->state = CIO_FAILED; |
496 |
break;
|
497 |
} |
498 |
cio->pending_reqs++; |
499 |
cio->state = CIO_READING; |
500 |
} |
501 |
pending_buckets++;; |
502 |
} |
503 |
|
504 |
if (pending_buckets) {
|
505 |
/* Do not put cache entry yet */
|
506 |
cio->work.job_fn = handle_read; |
507 |
cio->work.job = pr; |
508 |
/* wait on the last bucket */
|
509 |
xwaitq_enqueue(&ce->waitq[end_bucket], &cio->work); |
510 |
return;
|
511 |
} |
512 |
|
513 |
out:
|
514 |
if (cio->state == CIO_FAILED){
|
515 |
if (!cio->pending_reqs)
|
516 |
cached_fail(peer, pr); |
517 |
} |
518 |
else{
|
519 |
//serve req;
|
520 |
cached_complete(peer, pr); |
521 |
} |
522 |
return;
|
523 |
} |
524 |
|
525 |
static void handle_write(void *arg) |
526 |
{ |
527 |
//if writeback
|
528 |
// for each bucket
|
529 |
// write all buckets
|
530 |
// mark them as dirty
|
531 |
// cache_put(h)
|
532 |
// complete
|
533 |
//else
|
534 |
// send write to blocker
|
535 |
|
536 |
/*
|
537 |
* In this context we hold a reference to the cache entry and
|
538 |
* the assocciated lock
|
539 |
*/
|
540 |
|
541 |
int r;
|
542 |
struct peer_req *pr = (struct peer_req *)arg; |
543 |
struct peerd *peer = pr->peer;
|
544 |
struct cached *cached = __get_cached(peer);
|
545 |
struct cache_io *cio = __get_cache_io(pr);
|
546 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
547 |
(void)ce;
|
548 |
|
549 |
if (cached->write_policy == WRITETHROUGH){
|
550 |
//send write to blocker
|
551 |
//return
|
552 |
} else if (cached->write_policy == WRITEBACK) { |
553 |
//for each bucket
|
554 |
// write all buckets
|
555 |
// mark them as dirty
|
556 |
r = 0;
|
557 |
} else {
|
558 |
r = -1;
|
559 |
} |
560 |
|
561 |
out:
|
562 |
if (r < 0) |
563 |
cached_fail(peer, pr); |
564 |
else
|
565 |
cached_complete(peer, pr); |
566 |
return;
|
567 |
} |
568 |
|
569 |
static int handle_readwrite(struct peerd *peer, struct peer_req *pr) |
570 |
{ |
571 |
int r = -1; |
572 |
struct ce *ce;
|
573 |
struct cached *cached = __get_cached(peer);
|
574 |
struct cache_io *cio = __get_cache_io(pr);
|
575 |
char name[XSEG_MAX_TARGETLEN + 1]; |
576 |
struct xseg_request *req = pr->req;
|
577 |
char *target = xseg_get_target(peer->xseg, req);
|
578 |
|
579 |
xcache_handler h = NoEntry; |
580 |
XSEGLOG2(&lc, D, "In handle readwrite");
|
581 |
|
582 |
strncpy(name, target, req->targetlen); |
583 |
name[req->targetlen] = 0;
|
584 |
XSEGLOG2(&lc, D, "All good till here1. Name: %s\n", name);
|
585 |
|
586 |
h = xcache_lookup(cached->cache, name); |
587 |
if (h == NoEntry){
|
588 |
h = xcache_alloc_init(cached->cache, name); |
589 |
if (h == NoEntry){
|
590 |
goto out;
|
591 |
} |
592 |
r = xcache_insert(cached->cache, h); |
593 |
if (r < 0){ |
594 |
goto out;
|
595 |
} |
596 |
} |
597 |
|
598 |
ce = (struct ce *)get_cache_entry(cached->cache, h);
|
599 |
if (!ce){
|
600 |
r = -1;
|
601 |
goto out;
|
602 |
} |
603 |
cio->h = h; |
604 |
|
605 |
if (req->op == X_WRITE)
|
606 |
r = xworkq_enqueue(&ce->workq, handle_write, (void *)pr);
|
607 |
else if (req->op == X_READ) |
608 |
r = xworkq_enqueue(&ce->workq, handle_read, (void *)pr);
|
609 |
else {
|
610 |
r = -1;
|
611 |
goto out;
|
612 |
} |
613 |
|
614 |
out:
|
615 |
if (r < 0){ |
616 |
cached_fail(peer, pr); |
617 |
} |
618 |
return r;
|
619 |
|
620 |
} |
621 |
|
622 |
struct req_completion{
|
623 |
struct peer_req *pr;
|
624 |
struct xseg_request *req;
|
625 |
}; |
626 |
|
627 |
static void complete_read(void *arg) |
628 |
{ |
629 |
/*
|
630 |
* In this context we hold a reference to the cache entry and
|
631 |
* the assocciated lock
|
632 |
*/
|
633 |
|
634 |
struct req_completion *rc = (struct req_completion *)arg; |
635 |
struct peer_req *pr = rc->pr;
|
636 |
struct xseg_request *req = rc->req;
|
637 |
struct peerd *peer = pr->peer;
|
638 |
struct cached *cached = __get_cached(peer);
|
639 |
struct cache_io *cio = __get_cache_io(pr);
|
640 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
641 |
uint32_t start, end, i; |
642 |
int success;
|
643 |
char *data = xseg_get_data(peer->xseg, req);
|
644 |
|
645 |
/*
|
646 |
* Synchronize pending_reqs of the cache_io here, since each cache_io
|
647 |
* refers to only one object, and therefore we can use the object lock
|
648 |
* to synchronize between receive contextes.
|
649 |
*/
|
650 |
cio->pending_reqs--; |
651 |
success = (req->state == XS_SERVED && req->serviced == req->size); |
652 |
if (!success)
|
653 |
cio->state = CIO_FAILED; |
654 |
//assert (req->offset % cached->bucket_size) == 0;
|
655 |
//assert ((req->offset+req->serviced) % cached->bucket_size) == 0;
|
656 |
start = __get_bucket(cached, req->offset); |
657 |
end = __get_bucket(cached, req->offset + req->serviced); |
658 |
for (i = start; i <= end; i++) {
|
659 |
if (ce->status[i] == LOADING){
|
660 |
if (success){
|
661 |
memcpy(ce->data+(i*cached->bucket_size), data, |
662 |
cached->bucket_size); |
663 |
ce->status[i] = VALID; |
664 |
} |
665 |
else {
|
666 |
//reset status
|
667 |
XSEGLOG2(&lc, E, "Before ce %p, i %lu, ce->status[i] %u", ce, i, ce->status[i]);
|
668 |
ce->status[i] = INVALID; |
669 |
XSEGLOG2(&lc, E, "After ce %p, i %lu, ce->status[i] %u", ce, i, ce->status[i]);
|
670 |
} |
671 |
xwaitq_signal(&ce->waitq[i]); |
672 |
} |
673 |
} |
674 |
free(rc); |
675 |
} |
676 |
|
677 |
void complete_write(void *arg) |
678 |
{ |
679 |
//for each bucket
|
680 |
// if WRITETHROUGH
|
681 |
// copy data to bucket
|
682 |
// mark as valid
|
683 |
// else if WRITEBACK
|
684 |
// if status writing
|
685 |
// mark as valid
|
686 |
//
|
687 |
/*
|
688 |
* In this context we hold a reference to the cache entry and
|
689 |
* the assocciated lock
|
690 |
*/
|
691 |
return;
|
692 |
} |
693 |
|
694 |
static int handle_receive_read(struct peerd *peer, struct peer_req *pr, |
695 |
struct xseg_request *req)
|
696 |
{ |
697 |
/*
|
698 |
* Should be rentrant
|
699 |
*/
|
700 |
struct cached *cached = __get_cached(peer);
|
701 |
struct cache_io *cio = __get_cache_io(pr);
|
702 |
struct ce *ce = get_cache_entry(cached->cache, cio->h);
|
703 |
|
704 |
struct req_completion *rc;
|
705 |
|
706 |
rc = malloc(sizeof(struct req_completion)); |
707 |
|
708 |
rc->pr = pr; |
709 |
rc->req = req; |
710 |
if (xworkq_enqueue(&ce->workq, complete_read, (void *)rc) < 0){ |
711 |
free(rc); |
712 |
//TODO WHAT?
|
713 |
} |
714 |
return 0; |
715 |
} |
716 |
|
717 |
static int handle_receive_write(struct peerd *peer, struct peer_req *pr) |
718 |
{ |
719 |
//enqueue_work
|
720 |
return 0; |
721 |
} |
722 |
|
723 |
static int handle_delete(struct peerd *peer, struct peer_req *pr) |
724 |
{ |
725 |
//h = cache_lookup
|
726 |
//if h
|
727 |
// cio->h = h
|
728 |
//
|
729 |
//send delete to blocker
|
730 |
return 0; |
731 |
} |
732 |
|
733 |
static int handle_receive_delete(struct peerd *peer, struct peer_req *pr) |
734 |
{ |
735 |
//if success
|
736 |
// if cio->h
|
737 |
// //this should not write any dirty data
|
738 |
// xcache_remove(h)
|
739 |
return 0; |
740 |
} |
741 |
|
742 |
static int forward_req(struct peerd *peer, struct peer_req *pr) |
743 |
{ |
744 |
/*
|
745 |
struct cached *cached = __get_cached(peer);
|
746 |
struct cache_io *cio = __get_cache_io(pr);
|
747 |
|
748 |
xport p;
|
749 |
p = xseg_forward(peer->xseg, req, pr->portno, X_ALLOC);
|
750 |
|
751 |
xseg_signal(peer->xseg, p);
|
752 |
|
753 |
*/
|
754 |
return 0; |
755 |
} |
756 |
|
757 |
static int handle_receive(struct peerd *peer, struct peer_req *pr, |
758 |
struct xseg_request *req)
|
759 |
{ |
760 |
//if not read/write/delete
|
761 |
// put req;
|
762 |
// complete or fail pr;
|
763 |
int r;
|
764 |
xport p; |
765 |
|
766 |
switch (req->op){
|
767 |
case X_READ: r = handle_receive_read(peer, pr, req); break; |
768 |
// case X_WRITE: r = handle_receive_write(peer, pr, req); break;
|
769 |
// case X_DELETE: r = handle_receive_delete(peer, pr, req); break;
|
770 |
default:
|
771 |
p = xseg_respond(peer->xseg, req, pr->portno, X_ALLOC); |
772 |
if (p == NoPort)
|
773 |
xseg_put_request(peer->xseg, req, pr->portno); |
774 |
break;
|
775 |
} |
776 |
|
777 |
return 0; |
778 |
} |
779 |
|
780 |
int dispatch(struct peerd *peer, struct peer_req *pr, struct xseg_request *req, |
781 |
enum dispatch_reason reason)
|
782 |
{ |
783 |
struct cached *cached = __get_cached(peer);
|
784 |
(void) cached;
|
785 |
struct cache_io *cio = __get_cache_io(pr);
|
786 |
(void) cio;
|
787 |
|
788 |
switch (reason) {
|
789 |
case dispatch_accept:
|
790 |
XSEGLOG2(&lc, D, "In dispatch accept");
|
791 |
if (req->op == X_READ || req->op == X_WRITE) {
|
792 |
cio->state = CIO_ACCEPTED; |
793 |
handle_readwrite(peer, pr); |
794 |
} else {
|
795 |
fail(peer, pr); |
796 |
} |
797 |
break;
|
798 |
case dispatch_receive:
|
799 |
XSEGLOG2(&lc, D, "In dispatch receive");
|
800 |
handle_receive(peer, pr, req); |
801 |
break;
|
802 |
case dispatch_internal:
|
803 |
default:
|
804 |
XSEGLOG2(&lc, E, "Invalid dispatch reason");
|
805 |
} |
806 |
return 0; |
807 |
} |
808 |
|
809 |
int custom_peer_init(struct peerd *peer, int argc, char *argv[]) |
810 |
{ |
811 |
int i;
|
812 |
char cache_size[MAX_ARG_LEN + 1]; |
813 |
char bucket_size[MAX_ARG_LEN + 1]; |
814 |
char object_size[MAX_ARG_LEN + 1]; |
815 |
char max_req_size[MAX_ARG_LEN + 1]; |
816 |
char write_policy[MAX_ARG_LEN + 1]; |
817 |
long bportno = -1; |
818 |
|
819 |
cache_size[0] = 0; |
820 |
bucket_size[0] = 0; |
821 |
object_size[0] = 0; |
822 |
max_req_size[0] = 0; |
823 |
write_policy[0] = 0; |
824 |
|
825 |
/*Allocate enough space for needed structs*/
|
826 |
struct cached *cached = malloc(sizeof(struct cached)); |
827 |
if (!cached) {
|
828 |
perror("malloc");
|
829 |
goto fail;
|
830 |
} |
831 |
cached->cache = malloc(sizeof(struct xcache)); |
832 |
if (!cached->cache) {
|
833 |
perror("malloc");
|
834 |
goto cache_fail;
|
835 |
} |
836 |
peer->priv = cached; |
837 |
|
838 |
for (i = 0; i < peer->nr_ops; i++) { |
839 |
struct cache_io *cio = malloc(sizeof(struct cache_io)); |
840 |
if (!cio) {
|
841 |
perror("malloc");
|
842 |
goto cio_fail;
|
843 |
} |
844 |
cio->h = NoEntry; |
845 |
cio->pending_reqs = 0;
|
846 |
peer->peer_reqs[i].priv = cio; |
847 |
} |
848 |
|
849 |
/*Read arguments*/
|
850 |
/*FIXME: change actual size to relative size for cache_size*/
|
851 |
BEGIN_READ_ARGS(argc, argv); |
852 |
READ_ARG_ULONG("-bp", bportno);
|
853 |
READ_ARG_STRING("-cs", cache_size, MAX_ARG_LEN);
|
854 |
READ_ARG_STRING("-mrs", max_req_size, MAX_ARG_LEN);
|
855 |
READ_ARG_STRING("-os", object_size, MAX_ARG_LEN);
|
856 |
READ_ARG_STRING("-bs", bucket_size, MAX_ARG_LEN);
|
857 |
READ_ARG_STRING("-wcp", write_policy, MAX_ARG_LEN);
|
858 |
END_READ_ARGS(); |
859 |
|
860 |
/*Parse arguments for:*/
|
861 |
|
862 |
/*Bucket size*/
|
863 |
if (!bucket_size[0]) { |
864 |
cached->bucket_size = BUCKET_SIZE_QUANTUM; /*Default value*/
|
865 |
} else {
|
866 |
cached->bucket_size = str2num(bucket_size); |
867 |
if (!cached->bucket_size) {
|
868 |
XSEGLOG2(&lc, E, "Invalid syntax: -bs %s\n", bucket_size);
|
869 |
goto arg_fail;
|
870 |
} |
871 |
if (cached->bucket_size % BUCKET_SIZE_QUANTUM) {
|
872 |
XSEGLOG2(&lc, E, "Misaligned bucket size: %s\n", bucket_size);
|
873 |
goto arg_fail;
|
874 |
} |
875 |
} |
876 |
|
877 |
/*Object size*/
|
878 |
if (!object_size[0]) |
879 |
strcpy(object_size, "4M"); /*Default value*/ |
880 |
|
881 |
cached->object_size = str2num(object_size); |
882 |
if (!cached->object_size) {
|
883 |
XSEGLOG2(&lc, E, "Invalid syntax: -os %s\n", object_size);
|
884 |
goto arg_fail;
|
885 |
} |
886 |
if (cached->object_size % cached->bucket_size) {
|
887 |
XSEGLOG2(&lc, E, "Misaligned object size: %s\n", object_size);
|
888 |
goto arg_fail;
|
889 |
} |
890 |
|
891 |
/*Max request size*/
|
892 |
if (!max_req_size[0]) { |
893 |
XSEGLOG2(&lc, E, "Maximum request size must be provided\n");
|
894 |
goto arg_fail;
|
895 |
} |
896 |
cached->max_req_size = str2num(max_req_size); |
897 |
if (!cached->max_req_size) {
|
898 |
XSEGLOG2(&lc, E, "Invalid syntax: -mrs %s\n", max_req_size);
|
899 |
goto arg_fail;
|
900 |
} |
901 |
if (cached->max_req_size % BUCKET_SIZE_QUANTUM) {
|
902 |
XSEGLOG2(&lc, E, "Misaligned maximum request size: %s\n",
|
903 |
max_req_size); |
904 |
goto arg_fail;
|
905 |
} |
906 |
|
907 |
/*Cache size*/
|
908 |
if (!cache_size[0]) { |
909 |
XSEGLOG2(&lc, E, "Cache size must be provided\n");
|
910 |
goto arg_fail;
|
911 |
} |
912 |
cached->cache_size = str2num(cache_size); |
913 |
if (!cached->cache_size) {
|
914 |
XSEGLOG2(&lc, E, "Invalid syntax: -cs %s\n", cache_size);
|
915 |
goto arg_fail;
|
916 |
} |
917 |
if (cached->cache_size % cached->object_size) {
|
918 |
XSEGLOG2(&lc, E, "Misaligned cache size: %s\n", cache_size);
|
919 |
goto arg_fail;
|
920 |
} |
921 |
|
922 |
/*Blocker port*/
|
923 |
if (bportno < 0){ |
924 |
XSEGLOG2(&lc, E, "Blocker port must be provided");
|
925 |
goto arg_fail;
|
926 |
} |
927 |
cached->bportno = bportno; |
928 |
|
929 |
|
930 |
/*Change cache_size to number of objects*/
|
931 |
cached->cache_size = cached->cache_size / cached->object_size; |
932 |
cached->buckets_per_object = cached->object_size / cached->bucket_size; |
933 |
|
934 |
print_cached(cached); |
935 |
|
936 |
/*Initialize xcache*/
|
937 |
xcache_init(cached->cache, cached->cache_size, &c_ops, peer); |
938 |
|
939 |
return 0; |
940 |
|
941 |
arg_fail:
|
942 |
custom_peer_usage(); |
943 |
cio_fail:
|
944 |
for (i = 0; i < peer->nr_ops && peer->peer_reqs[i].priv != NULL; i++) |
945 |
free(peer->peer_reqs[i].priv); |
946 |
free(cached->cache); |
947 |
cache_fail:
|
948 |
free(cached); |
949 |
fail:
|
950 |
return -1; |
951 |
} |
952 |
|
953 |
void custom_peer_finalize(struct peerd *peer) |
954 |
{ |
955 |
//write dirty objects
|
956 |
//or cache_close(cached->cache);
|
957 |
return;
|
958 |
} |
959 |
|
960 |
void custom_peer_usage()
|
961 |
{ |
962 |
fprintf(stderr, "Custom peer options: \n"
|
963 |
" --------------------------------------------\n"
|
964 |
" -cs | None | Cache size\n"
|
965 |
" -mrs | None | Max request size\n"
|
966 |
" -os | 4M | Object size\n"
|
967 |
" -bs | 4K | Bucket size\n"
|
968 |
" -bp | None | Blocker port\n"
|
969 |
"\n");
|
970 |
} |