source: mainline/uspace/lib/block/libblock.c@ 969585f

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since 969585f was 5716e9a, checked in by Jakub Jermar <jakub@…>, 14 years ago

Make sure that block_get() instantiates each block only once.

  • Property mode set to 100644
File size: 22.7 KB
RevLine 
[fc840d9]1/*
[ed903174]2 * Copyright (c) 2008 Jakub Jermar
3 * Copyright (c) 2008 Martin Decky
[fc840d9]4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 *
10 * - Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * - Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * - The name of the author may not be used to endorse or promote products
16 * derived from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28 */
29
[97c9da8]30/** @addtogroup libblock
[fc840d9]31 * @{
[97c9da8]32 */
[fc840d9]33/**
34 * @file
35 * @brief
36 */
37
[97c9da8]38#include "libblock.h"
[fc840d9]39#include "../../srv/vfs/vfs.h"
[7858bc5f]40#include <ipc/devmap.h>
[c5747fe]41#include <ipc/bd.h>
[7858bc5f]42#include <ipc/services.h>
[fc840d9]43#include <errno.h>
[7858bc5f]44#include <sys/mman.h>
[fc840d9]45#include <async.h>
46#include <as.h>
47#include <assert.h>
[1e4cada]48#include <fibril_synch.h>
[d9c8c81]49#include <adt/list.h>
50#include <adt/hash_table.h>
[1ee00b7]51#include <macros.h>
[d00ae4c]52#include <mem.h>
[16fc3c9]53#include <sys/typefmt.h>
54#include <stacktrace.h>
[fc840d9]55
[916bf1a]56/** Lock protecting the device connection list */
[4e1b57d]57static FIBRIL_MUTEX_INITIALIZE(dcl_lock);
[916bf1a]58/** Device connection list head. */
59static LIST_INITIALIZE(dcl_head);
60
[f1ba5d6]61#define CACHE_BUCKETS_LOG2 10
62#define CACHE_BUCKETS (1 << CACHE_BUCKETS_LOG2)
63
64typedef struct {
[4e1b57d]65 fibril_mutex_t lock;
[1ee00b7]66 size_t lblock_size; /**< Logical block size. */
[f092718]67 unsigned blocks_cluster; /**< Physical blocks per block_t */
[f1ba5d6]68 unsigned block_count; /**< Total number of blocks. */
[d68e4d5]69 unsigned blocks_cached; /**< Number of cached blocks. */
[f1ba5d6]70 hash_table_t block_hash;
71 link_t free_head;
[1fbe064b]72 enum cache_mode mode;
[f1ba5d6]73} cache_t;
74
[916bf1a]75typedef struct {
76 link_t link;
[991f645]77 devmap_handle_t devmap_handle;
[916bf1a]78 int dev_phone;
[a830611]79 fibril_mutex_t comm_area_lock;
80 void *comm_area;
81 size_t comm_size;
[916bf1a]82 void *bb_buf;
[ed903174]83 aoff64_t bb_addr;
[1ee00b7]84 size_t pblock_size; /**< Physical block size. */
[f1ba5d6]85 cache_t *cache;
[916bf1a]86} devcon_t;
87
[ed903174]88static int read_blocks(devcon_t *devcon, aoff64_t ba, size_t cnt);
89static int write_blocks(devcon_t *devcon, aoff64_t ba, size_t cnt);
[00b1d20e]90static int get_block_size(int dev_phone, size_t *bsize);
[ed903174]91static int get_num_blocks(int dev_phone, aoff64_t *nblocks);
[f092718]92static aoff64_t ba_ltop(devcon_t *devcon, aoff64_t lba);
[1fbe064b]93
[991f645]94static devcon_t *devcon_search(devmap_handle_t devmap_handle)
[916bf1a]95{
96 link_t *cur;
97
[4e1b57d]98 fibril_mutex_lock(&dcl_lock);
[916bf1a]99 for (cur = dcl_head.next; cur != &dcl_head; cur = cur->next) {
100 devcon_t *devcon = list_get_instance(cur, devcon_t, link);
[991f645]101 if (devcon->devmap_handle == devmap_handle) {
[4e1b57d]102 fibril_mutex_unlock(&dcl_lock);
[916bf1a]103 return devcon;
104 }
105 }
[4e1b57d]106 fibril_mutex_unlock(&dcl_lock);
[916bf1a]107 return NULL;
108}
109
[991f645]110static int devcon_add(devmap_handle_t devmap_handle, int dev_phone, size_t bsize,
[a830611]111 void *comm_area, size_t comm_size)
[916bf1a]112{
113 link_t *cur;
114 devcon_t *devcon;
115
[a830611]116 if (comm_size < bsize)
[1ee00b7]117 return EINVAL;
118
[916bf1a]119 devcon = malloc(sizeof(devcon_t));
120 if (!devcon)
121 return ENOMEM;
122
123 link_initialize(&devcon->link);
[991f645]124 devcon->devmap_handle = devmap_handle;
[916bf1a]125 devcon->dev_phone = dev_phone;
[a830611]126 fibril_mutex_initialize(&devcon->comm_area_lock);
127 devcon->comm_area = comm_area;
128 devcon->comm_size = comm_size;
[6284978]129 devcon->bb_buf = NULL;
[1ee00b7]130 devcon->bb_addr = 0;
131 devcon->pblock_size = bsize;
[f1ba5d6]132 devcon->cache = NULL;
[916bf1a]133
[4e1b57d]134 fibril_mutex_lock(&dcl_lock);
[916bf1a]135 for (cur = dcl_head.next; cur != &dcl_head; cur = cur->next) {
136 devcon_t *d = list_get_instance(cur, devcon_t, link);
[991f645]137 if (d->devmap_handle == devmap_handle) {
[4e1b57d]138 fibril_mutex_unlock(&dcl_lock);
[916bf1a]139 free(devcon);
140 return EEXIST;
141 }
142 }
143 list_append(&devcon->link, &dcl_head);
[4e1b57d]144 fibril_mutex_unlock(&dcl_lock);
[916bf1a]145 return EOK;
146}
147
148static void devcon_remove(devcon_t *devcon)
149{
[4e1b57d]150 fibril_mutex_lock(&dcl_lock);
[916bf1a]151 list_remove(&devcon->link);
[4e1b57d]152 fibril_mutex_unlock(&dcl_lock);
[916bf1a]153}
[7858bc5f]154
[991f645]155int block_init(devmap_handle_t devmap_handle, size_t comm_size)
[7858bc5f]156{
157 int rc;
[916bf1a]158 int dev_phone;
[a830611]159 void *comm_area;
[1ee00b7]160 size_t bsize;
161
[a830611]162 comm_area = mmap(NULL, comm_size, PROTO_READ | PROTO_WRITE,
[7858bc5f]163 MAP_ANONYMOUS | MAP_PRIVATE, 0, 0);
[a830611]164 if (!comm_area) {
[7858bc5f]165 return ENOMEM;
166 }
167
[991f645]168 dev_phone = devmap_device_connect(devmap_handle, IPC_FLAG_BLOCKING);
[7858bc5f]169 if (dev_phone < 0) {
[a830611]170 munmap(comm_area, comm_size);
[7858bc5f]171 return dev_phone;
172 }
173
[0da4e41]174 rc = async_share_out_start(dev_phone, comm_area,
[7858bc5f]175 AS_AREA_READ | AS_AREA_WRITE);
176 if (rc != EOK) {
[a830611]177 munmap(comm_area, comm_size);
[ffa2c8ef]178 async_hangup(dev_phone);
[7858bc5f]179 return rc;
180 }
[1ee00b7]181
182 if (get_block_size(dev_phone, &bsize) != EOK) {
[a830611]183 munmap(comm_area, comm_size);
[ffa2c8ef]184 async_hangup(dev_phone);
[1ee00b7]185 return rc;
186 }
[916bf1a]187
[991f645]188 rc = devcon_add(devmap_handle, dev_phone, bsize, comm_area, comm_size);
[916bf1a]189 if (rc != EOK) {
[a830611]190 munmap(comm_area, comm_size);
[ffa2c8ef]191 async_hangup(dev_phone);
[916bf1a]192 return rc;
193 }
194
[7858bc5f]195 return EOK;
196}
197
[991f645]198void block_fini(devmap_handle_t devmap_handle)
[7858bc5f]199{
[991f645]200 devcon_t *devcon = devcon_search(devmap_handle);
[916bf1a]201 assert(devcon);
202
[64bc4b6]203 if (devcon->cache)
[991f645]204 (void) block_cache_fini(devmap_handle);
[64bc4b6]205
[916bf1a]206 devcon_remove(devcon);
207
[6284978]208 if (devcon->bb_buf)
209 free(devcon->bb_buf);
[f1ba5d6]210
[a830611]211 munmap(devcon->comm_area, devcon->comm_size);
[ffa2c8ef]212 async_hangup(devcon->dev_phone);
[916bf1a]213
214 free(devcon);
[7858bc5f]215}
216
[991f645]217int block_bb_read(devmap_handle_t devmap_handle, aoff64_t ba)
[6284978]218{
219 void *bb_buf;
[0c243b4]220 int rc;
[6284978]221
[991f645]222 devcon_t *devcon = devcon_search(devmap_handle);
[6284978]223 if (!devcon)
224 return ENOENT;
225 if (devcon->bb_buf)
226 return EEXIST;
[1ee00b7]227 bb_buf = malloc(devcon->pblock_size);
[6284978]228 if (!bb_buf)
229 return ENOMEM;
[1ee00b7]230
[a830611]231 fibril_mutex_lock(&devcon->comm_area_lock);
[1ee00b7]232 rc = read_blocks(devcon, 0, 1);
[0c243b4]233 if (rc != EOK) {
[a830611]234 fibril_mutex_unlock(&devcon->comm_area_lock);
[6284978]235 free(bb_buf);
[0c243b4]236 return rc;
[6284978]237 }
[a830611]238 memcpy(bb_buf, devcon->comm_area, devcon->pblock_size);
239 fibril_mutex_unlock(&devcon->comm_area_lock);
[6408be3]240
[6284978]241 devcon->bb_buf = bb_buf;
[1ee00b7]242 devcon->bb_addr = ba;
[6284978]243
244 return EOK;
245}
246
[991f645]247void *block_bb_get(devmap_handle_t devmap_handle)
[7858bc5f]248{
[991f645]249 devcon_t *devcon = devcon_search(devmap_handle);
[916bf1a]250 assert(devcon);
251 return devcon->bb_buf;
[7858bc5f]252}
253
[f1ba5d6]254static hash_index_t cache_hash(unsigned long *key)
255{
256 return *key & (CACHE_BUCKETS - 1);
257}
258
259static int cache_compare(unsigned long *key, hash_count_t keys, link_t *item)
260{
261 block_t *b = hash_table_get_instance(item, block_t, hash_link);
[a6ba0c9]262 return b->lba == *key;
[f1ba5d6]263}
264
265static void cache_remove_callback(link_t *item)
266{
267}
268
269static hash_table_operations_t cache_ops = {
270 .hash = cache_hash,
271 .compare = cache_compare,
272 .remove_callback = cache_remove_callback
273};
274
[991f645]275int block_cache_init(devmap_handle_t devmap_handle, size_t size, unsigned blocks,
[1fbe064b]276 enum cache_mode mode)
[f1ba5d6]277{
[991f645]278 devcon_t *devcon = devcon_search(devmap_handle);
[f1ba5d6]279 cache_t *cache;
280 if (!devcon)
281 return ENOENT;
282 if (devcon->cache)
283 return EEXIST;
284 cache = malloc(sizeof(cache_t));
285 if (!cache)
286 return ENOMEM;
287
[4e1b57d]288 fibril_mutex_initialize(&cache->lock);
[f1ba5d6]289 list_initialize(&cache->free_head);
[1ee00b7]290 cache->lblock_size = size;
[f1ba5d6]291 cache->block_count = blocks;
[d68e4d5]292 cache->blocks_cached = 0;
[1fbe064b]293 cache->mode = mode;
[f1ba5d6]294
[f092718]295 /* Allow 1:1 or small-to-large block size translation */
[37cf3792]296 if (cache->lblock_size % devcon->pblock_size != 0) {
297 free(cache);
[f092718]298 return ENOTSUP;
[37cf3792]299 }
[f092718]300
301 cache->blocks_cluster = cache->lblock_size / devcon->pblock_size;
[1ee00b7]302
[f1ba5d6]303 if (!hash_table_create(&cache->block_hash, CACHE_BUCKETS, 1,
304 &cache_ops)) {
305 free(cache);
306 return ENOMEM;
307 }
308
309 devcon->cache = cache;
310 return EOK;
311}
312
[991f645]313int block_cache_fini(devmap_handle_t devmap_handle)
[64bc4b6]314{
[991f645]315 devcon_t *devcon = devcon_search(devmap_handle);
[64bc4b6]316 cache_t *cache;
317 int rc;
318
319 if (!devcon)
320 return ENOENT;
321 if (!devcon->cache)
322 return EOK;
323 cache = devcon->cache;
324
325 /*
326 * We are expecting to find all blocks for this device handle on the
327 * free list, i.e. the block reference count should be zero. Do not
328 * bother with the cache and block locks because we are single-threaded.
329 */
330 while (!list_empty(&cache->free_head)) {
331 block_t *b = list_get_instance(cache->free_head.next,
332 block_t, free_link);
333
334 list_remove(&b->free_link);
335 if (b->dirty) {
336 memcpy(devcon->comm_area, b->data, b->size);
[f092718]337 rc = write_blocks(devcon, b->pba, cache->blocks_cluster);
[64bc4b6]338 if (rc != EOK)
339 return rc;
340 }
341
[a6ba0c9]342 unsigned long key = b->lba;
[64bc4b6]343 hash_table_remove(&cache->block_hash, &key, 1);
344
345 free(b->data);
346 free(b);
347 }
348
349 hash_table_destroy(&cache->block_hash);
350 devcon->cache = NULL;
351 free(cache);
352
353 return EOK;
354}
355
[d68e4d5]356#define CACHE_LO_WATERMARK 10
357#define CACHE_HI_WATERMARK 20
[e1c88d5]358static bool cache_can_grow(cache_t *cache)
[fc840d9]359{
[d68e4d5]360 if (cache->blocks_cached < CACHE_LO_WATERMARK)
361 return true;
362 if (!list_empty(&cache->free_head))
363 return false;
[e1c88d5]364 return true;
365}
366
367static void block_initialize(block_t *b)
368{
[4e1b57d]369 fibril_mutex_initialize(&b->lock);
[e1c88d5]370 b->refcnt = 1;
371 b->dirty = false;
[cd688d9]372 b->toxic = false;
[4e1b57d]373 fibril_rwlock_initialize(&b->contents_lock);
[e1c88d5]374 link_initialize(&b->free_link);
375 link_initialize(&b->hash_link);
376}
377
378/** Instantiate a block in memory and get a reference to it.
379 *
[c91f2d1b]380 * @param block Pointer to where the function will store the
381 * block pointer on success.
[991f645]382 * @param devmap_handle Device handle of the block device.
[a6ba0c9]383 * @param ba Block address (logical).
[1d8cdb1]384 * @param flags If BLOCK_FLAGS_NOREAD is specified, block_get()
385 * will not read the contents of the block from the
386 * device.
[e1c88d5]387 *
[c91f2d1b]388 * @return EOK on success or a negative error code.
[e1c88d5]389 */
[a6ba0c9]390int block_get(block_t **block, devmap_handle_t devmap_handle, aoff64_t ba, int flags)
[e1c88d5]391{
392 devcon_t *devcon;
393 cache_t *cache;
[fc840d9]394 block_t *b;
[e1c88d5]395 link_t *l;
[a6ba0c9]396 unsigned long key = ba;
[b7b3fda]397 int rc;
[e1c88d5]398
[991f645]399 devcon = devcon_search(devmap_handle);
[fc840d9]400
[e1c88d5]401 assert(devcon);
402 assert(devcon->cache);
[fc840d9]403
[e1c88d5]404 cache = devcon->cache;
[02ee6bf5]405
406retry:
[b7b3fda]407 rc = EOK;
[4f690cd]408 b = NULL;
[b7b3fda]409
[4e1b57d]410 fibril_mutex_lock(&cache->lock);
[e1c88d5]411 l = hash_table_find(&cache->block_hash, &key);
412 if (l) {
[5716e9a]413found:
[e1c88d5]414 /*
415 * We found the block in the cache.
416 */
417 b = hash_table_get_instance(l, block_t, hash_link);
[4e1b57d]418 fibril_mutex_lock(&b->lock);
[e1c88d5]419 if (b->refcnt++ == 0)
420 list_remove(&b->free_link);
[402a18f]421 if (b->toxic)
422 rc = EIO;
[4e1b57d]423 fibril_mutex_unlock(&b->lock);
424 fibril_mutex_unlock(&cache->lock);
[e1c88d5]425 } else {
426 /*
427 * The block was not found in the cache.
428 */
429 if (cache_can_grow(cache)) {
430 /*
431 * We can grow the cache by allocating new blocks.
432 * Should the allocation fail, we fail over and try to
433 * recycle a block from the cache.
434 */
435 b = malloc(sizeof(block_t));
436 if (!b)
437 goto recycle;
[1ee00b7]438 b->data = malloc(cache->lblock_size);
[e1c88d5]439 if (!b->data) {
440 free(b);
[0dfaa099]441 b = NULL;
[e1c88d5]442 goto recycle;
443 }
[d68e4d5]444 cache->blocks_cached++;
[e1c88d5]445 } else {
446 /*
447 * Try to recycle a block from the free list.
448 */
449 unsigned long temp_key;
450recycle:
[7a56b1ed]451 if (list_empty(&cache->free_head)) {
452 fibril_mutex_unlock(&cache->lock);
453 rc = ENOMEM;
454 goto out;
455 }
[e1c88d5]456 l = cache->free_head.next;
[d68e4d5]457 b = list_get_instance(l, block_t, free_link);
[02ee6bf5]458
459 fibril_mutex_lock(&b->lock);
460 if (b->dirty) {
461 /*
462 * The block needs to be written back to the
463 * device before it changes identity. Do this
464 * while not holding the cache lock so that
465 * concurrency is not impeded. Also move the
466 * block to the end of the free list so that we
467 * do not slow down other instances of
468 * block_get() draining the free list.
469 */
470 list_remove(&b->free_link);
471 list_append(&b->free_link, &cache->free_head);
472 fibril_mutex_unlock(&cache->lock);
[a830611]473 fibril_mutex_lock(&devcon->comm_area_lock);
474 memcpy(devcon->comm_area, b->data, b->size);
[f092718]475 rc = write_blocks(devcon, b->pba,
476 cache->blocks_cluster);
[a830611]477 fibril_mutex_unlock(&devcon->comm_area_lock);
[402a18f]478 if (rc != EOK) {
479 /*
480 * We did not manage to write the block
481 * to the device. Keep it around for
482 * another try. Hopefully, we will grab
483 * another block next time.
484 */
485 fibril_mutex_unlock(&b->lock);
486 goto retry;
487 }
[02ee6bf5]488 b->dirty = false;
489 if (!fibril_mutex_trylock(&cache->lock)) {
490 /*
491 * Somebody is probably racing with us.
492 * Unlock the block and retry.
493 */
494 fibril_mutex_unlock(&b->lock);
495 goto retry;
496 }
[5716e9a]497 l = hash_table_find(&cache->block_hash, &key);
498 if (l) {
499 /*
500 * Someone else must have already
501 * instantiated the block while we were
502 * not holding the cache lock.
503 * Leave the recycled block on the
504 * freelist and continue as if we
505 * found the block of interest during
506 * the first try.
507 */
508 fibril_mutex_unlock(&b->lock);
509 goto found;
510 }
[02ee6bf5]511
512 }
513 fibril_mutex_unlock(&b->lock);
514
515 /*
516 * Unlink the block from the free list and the hash
517 * table.
518 */
519 list_remove(&b->free_link);
[a6ba0c9]520 temp_key = b->lba;
[e1c88d5]521 hash_table_remove(&cache->block_hash, &temp_key, 1);
522 }
[fc840d9]523
[e1c88d5]524 block_initialize(b);
[991f645]525 b->devmap_handle = devmap_handle;
[1ee00b7]526 b->size = cache->lblock_size;
[a6ba0c9]527 b->lba = ba;
528 b->pba = ba_ltop(devcon, b->lba);
[a6d97fb9]529 hash_table_insert(&cache->block_hash, &key, &b->hash_link);
530
531 /*
532 * Lock the block before releasing the cache lock. Thus we don't
[5ac8918]533 * kill concurrent operations on the cache while doing I/O on
534 * the block.
[a6d97fb9]535 */
[4e1b57d]536 fibril_mutex_lock(&b->lock);
537 fibril_mutex_unlock(&cache->lock);
[a6d97fb9]538
[1d8cdb1]539 if (!(flags & BLOCK_FLAGS_NOREAD)) {
540 /*
541 * The block contains old or no data. We need to read
542 * the new contents from the device.
543 */
[a830611]544 fibril_mutex_lock(&devcon->comm_area_lock);
[f092718]545 rc = read_blocks(devcon, b->pba, cache->blocks_cluster);
[a830611]546 memcpy(b->data, devcon->comm_area, cache->lblock_size);
547 fibril_mutex_unlock(&devcon->comm_area_lock);
[402a18f]548 if (rc != EOK)
549 b->toxic = true;
550 } else
551 rc = EOK;
[fc840d9]552
[4e1b57d]553 fibril_mutex_unlock(&b->lock);
[a6d97fb9]554 }
[7a56b1ed]555out:
[4f690cd]556 if ((rc != EOK) && b) {
557 assert(b->toxic);
558 (void) block_put(b);
559 b = NULL;
560 }
[c91f2d1b]561 *block = b;
[402a18f]562 return rc;
[fc840d9]563}
564
[d5a720cf]565/** Release a reference to a block.
566 *
[a6d97fb9]567 * If the last reference is dropped, the block is put on the free list.
[d5a720cf]568 *
569 * @param block Block of which a reference is to be released.
[c91f2d1b]570 *
571 * @return EOK on success or a negative error code.
[d5a720cf]572 */
[c91f2d1b]573int block_put(block_t *block)
[fc840d9]574{
[991f645]575 devcon_t *devcon = devcon_search(block->devmap_handle);
[d5a720cf]576 cache_t *cache;
[ddfc39a3]577 unsigned blocks_cached;
578 enum cache_mode mode;
[402a18f]579 int rc = EOK;
[d5a720cf]580
581 assert(devcon);
582 assert(devcon->cache);
[0f1cf7a]583 assert(block->refcnt >= 1);
[d5a720cf]584
585 cache = devcon->cache;
[ddfc39a3]586
587retry:
588 fibril_mutex_lock(&cache->lock);
589 blocks_cached = cache->blocks_cached;
590 mode = cache->mode;
591 fibril_mutex_unlock(&cache->lock);
592
593 /*
594 * Determine whether to sync the block. Syncing the block is best done
595 * when not holding the cache lock as it does not impede concurrency.
596 * Since the situation may have changed when we unlocked the cache, the
597 * blocks_cached and mode variables are mere hints. We will recheck the
598 * conditions later when the cache lock is held again.
599 */
600 fibril_mutex_lock(&block->lock);
[402a18f]601 if (block->toxic)
602 block->dirty = false; /* will not write back toxic block */
[ddfc39a3]603 if (block->dirty && (block->refcnt == 1) &&
604 (blocks_cached > CACHE_HI_WATERMARK || mode != CACHE_MODE_WB)) {
[a830611]605 fibril_mutex_lock(&devcon->comm_area_lock);
606 memcpy(devcon->comm_area, block->data, block->size);
[f092718]607 rc = write_blocks(devcon, block->pba, cache->blocks_cluster);
[a830611]608 fibril_mutex_unlock(&devcon->comm_area_lock);
[ddfc39a3]609 block->dirty = false;
610 }
611 fibril_mutex_unlock(&block->lock);
612
[4e1b57d]613 fibril_mutex_lock(&cache->lock);
614 fibril_mutex_lock(&block->lock);
[d5a720cf]615 if (!--block->refcnt) {
616 /*
[d68e4d5]617 * Last reference to the block was dropped. Either free the
[402a18f]618 * block or put it on the free list. In case of an I/O error,
619 * free the block.
[d68e4d5]620 */
[402a18f]621 if ((cache->blocks_cached > CACHE_HI_WATERMARK) ||
622 (rc != EOK)) {
[d68e4d5]623 /*
[402a18f]624 * Currently there are too many cached blocks or there
625 * was an I/O error when writing the block back to the
626 * device.
[d68e4d5]627 */
628 if (block->dirty) {
[ddfc39a3]629 /*
630 * We cannot sync the block while holding the
631 * cache lock. Release everything and retry.
632 */
633 block->refcnt++;
634 fibril_mutex_unlock(&block->lock);
635 fibril_mutex_unlock(&cache->lock);
636 goto retry;
[d68e4d5]637 }
638 /*
639 * Take the block out of the cache and free it.
640 */
[a6ba0c9]641 unsigned long key = block->lba;
[d68e4d5]642 hash_table_remove(&cache->block_hash, &key, 1);
[956d4df8]643 fibril_mutex_unlock(&block->lock);
[d68e4d5]644 free(block->data);
[b9e6205]645 free(block);
[d68e4d5]646 cache->blocks_cached--;
647 fibril_mutex_unlock(&cache->lock);
[402a18f]648 return rc;
[d68e4d5]649 }
650 /*
651 * Put the block on the free list.
[d5a720cf]652 */
[1fbe064b]653 if (cache->mode != CACHE_MODE_WB && block->dirty) {
[ddfc39a3]654 /*
655 * We cannot sync the block while holding the cache
656 * lock. Release everything and retry.
657 */
658 block->refcnt++;
659 fibril_mutex_unlock(&block->lock);
660 fibril_mutex_unlock(&cache->lock);
661 goto retry;
[1fbe064b]662 }
[ddfc39a3]663 list_append(&block->free_link, &cache->free_head);
[d5a720cf]664 }
[4e1b57d]665 fibril_mutex_unlock(&block->lock);
666 fibril_mutex_unlock(&cache->lock);
[c91f2d1b]667
[402a18f]668 return rc;
[d5a720cf]669}
670
[6408be3]671/** Read sequential data from a block device.
[d5a720cf]672 *
[991f645]673 * @param devmap_handle Device handle of the block device.
[d5a720cf]674 * @param bufpos Pointer to the first unread valid offset within the
675 * communication buffer.
676 * @param buflen Pointer to the number of unread bytes that are ready in
677 * the communication buffer.
678 * @param pos Device position to be read.
679 * @param dst Destination buffer.
680 * @param size Size of the destination buffer.
681 * @param block_size Block size to be used for the transfer.
682 *
683 * @return EOK on success or a negative return code on failure.
684 */
[991f645]685int block_seqread(devmap_handle_t devmap_handle, size_t *bufpos, size_t *buflen,
[ed903174]686 aoff64_t *pos, void *dst, size_t size)
[d5a720cf]687{
[ed903174]688 size_t offset = 0;
[d5a720cf]689 size_t left = size;
[1ee00b7]690 size_t block_size;
691 devcon_t *devcon;
692
[991f645]693 devcon = devcon_search(devmap_handle);
[d5a720cf]694 assert(devcon);
[1ee00b7]695 block_size = devcon->pblock_size;
[e1c88d5]696
[a830611]697 fibril_mutex_lock(&devcon->comm_area_lock);
[d5a720cf]698 while (left > 0) {
699 size_t rd;
700
701 if (*bufpos + left < *buflen)
702 rd = left;
703 else
704 rd = *buflen - *bufpos;
705
706 if (rd > 0) {
707 /*
708 * Copy the contents of the communication buffer to the
709 * destination buffer.
710 */
[a830611]711 memcpy(dst + offset, devcon->comm_area + *bufpos, rd);
[d5a720cf]712 offset += rd;
713 *bufpos += rd;
714 *pos += rd;
715 left -= rd;
716 }
717
[ed903174]718 if (*bufpos == *buflen) {
[d5a720cf]719 /* Refill the communication buffer with a new block. */
[6408be3]720 int rc;
721
[1ee00b7]722 rc = read_blocks(devcon, *pos / block_size, 1);
[d68e4d5]723 if (rc != EOK) {
[a830611]724 fibril_mutex_unlock(&devcon->comm_area_lock);
[6408be3]725 return rc;
[d68e4d5]726 }
[d5a720cf]727
728 *bufpos = 0;
729 *buflen = block_size;
730 }
731 }
[a830611]732 fibril_mutex_unlock(&devcon->comm_area_lock);
[d5a720cf]733
734 return EOK;
[fc840d9]735}
736
[00b1d20e]737/** Read blocks directly from device (bypass cache).
738 *
[991f645]739 * @param devmap_handle Device handle of the block device.
[a6ba0c9]740 * @param ba Address of first block (physical).
[00b1d20e]741 * @param cnt Number of blocks.
742 * @param src Buffer for storing the data.
743 *
744 * @return EOK on success or negative error code on failure.
745 */
[991f645]746int block_read_direct(devmap_handle_t devmap_handle, aoff64_t ba, size_t cnt, void *buf)
[00b1d20e]747{
748 devcon_t *devcon;
749 int rc;
750
[991f645]751 devcon = devcon_search(devmap_handle);
[00b1d20e]752 assert(devcon);
753
754 fibril_mutex_lock(&devcon->comm_area_lock);
755
756 rc = read_blocks(devcon, ba, cnt);
757 if (rc == EOK)
758 memcpy(buf, devcon->comm_area, devcon->pblock_size * cnt);
759
760 fibril_mutex_unlock(&devcon->comm_area_lock);
761
762 return rc;
763}
764
765/** Write blocks directly to device (bypass cache).
766 *
[991f645]767 * @param devmap_handle Device handle of the block device.
[a6ba0c9]768 * @param ba Address of first block (physical).
[00b1d20e]769 * @param cnt Number of blocks.
770 * @param src The data to be written.
771 *
772 * @return EOK on success or negative error code on failure.
773 */
[991f645]774int block_write_direct(devmap_handle_t devmap_handle, aoff64_t ba, size_t cnt,
[00b1d20e]775 const void *data)
776{
777 devcon_t *devcon;
778 int rc;
779
[991f645]780 devcon = devcon_search(devmap_handle);
[00b1d20e]781 assert(devcon);
782
783 fibril_mutex_lock(&devcon->comm_area_lock);
784
785 memcpy(devcon->comm_area, data, devcon->pblock_size * cnt);
[dccf721]786 rc = write_blocks(devcon, ba, cnt);
[00b1d20e]787
788 fibril_mutex_unlock(&devcon->comm_area_lock);
789
790 return rc;
791}
792
793/** Get device block size.
794 *
[991f645]795 * @param devmap_handle Device handle of the block device.
[00b1d20e]796 * @param bsize Output block size.
797 *
798 * @return EOK on success or negative error code on failure.
799 */
[991f645]800int block_get_bsize(devmap_handle_t devmap_handle, size_t *bsize)
[00b1d20e]801{
802 devcon_t *devcon;
803
[991f645]804 devcon = devcon_search(devmap_handle);
[00b1d20e]805 assert(devcon);
806
807 return get_block_size(devcon->dev_phone, bsize);
808}
809
[08232ee]810/** Get number of blocks on device.
811 *
[991f645]812 * @param devmap_handle Device handle of the block device.
[08232ee]813 * @param nblocks Output number of blocks.
814 *
815 * @return EOK on success or negative error code on failure.
816 */
[991f645]817int block_get_nblocks(devmap_handle_t devmap_handle, aoff64_t *nblocks)
[08232ee]818{
819 devcon_t *devcon;
820
[991f645]821 devcon = devcon_search(devmap_handle);
[08232ee]822 assert(devcon);
823
824 return get_num_blocks(devcon->dev_phone, nblocks);
825}
826
[1ee00b7]827/** Read blocks from block device.
[6408be3]828 *
829 * @param devcon Device connection.
[1ee00b7]830 * @param ba Address of first block.
831 * @param cnt Number of blocks.
[6408be3]832 * @param src Buffer for storing the data.
833 *
834 * @return EOK on success or negative error code on failure.
835 */
[ed903174]836static int read_blocks(devcon_t *devcon, aoff64_t ba, size_t cnt)
[6408be3]837{
838 int rc;
839
840 assert(devcon);
[1ee00b7]841 rc = async_req_3_0(devcon->dev_phone, BD_READ_BLOCKS, LOWER32(ba),
842 UPPER32(ba), cnt);
[16fc3c9]843 if (rc != EOK) {
[7e752b2]844 printf("Error %d reading %zu blocks starting at block %" PRIuOFF64
845 " from device handle %" PRIun "\n", rc, cnt, ba,
[991f645]846 devcon->devmap_handle);
[16fc3c9]847#ifndef NDEBUG
848 stacktrace_print();
849#endif
850 }
[1ee00b7]851 return rc;
[6408be3]852}
853
[1fbe064b]854/** Write block to block device.
855 *
856 * @param devcon Device connection.
[1ee00b7]857 * @param ba Address of first block.
858 * @param cnt Number of blocks.
[1fbe064b]859 * @param src Buffer containing the data to write.
860 *
861 * @return EOK on success or negative error code on failure.
862 */
[ed903174]863static int write_blocks(devcon_t *devcon, aoff64_t ba, size_t cnt)
[1fbe064b]864{
865 int rc;
866
867 assert(devcon);
[1ee00b7]868 rc = async_req_3_0(devcon->dev_phone, BD_WRITE_BLOCKS, LOWER32(ba),
869 UPPER32(ba), cnt);
[16fc3c9]870 if (rc != EOK) {
[7e752b2]871 printf("Error %d writing %zu blocks starting at block %" PRIuOFF64
872 " to device handle %" PRIun "\n", rc, cnt, ba, devcon->devmap_handle);
[16fc3c9]873#ifndef NDEBUG
874 stacktrace_print();
875#endif
876 }
[1ee00b7]877 return rc;
878}
[1fbe064b]879
[1ee00b7]880/** Get block size used by the device. */
[00b1d20e]881static int get_block_size(int dev_phone, size_t *bsize)
[1ee00b7]882{
[96b02eb9]883 sysarg_t bs;
[1ee00b7]884 int rc;
885
886 rc = async_req_0_1(dev_phone, BD_GET_BLOCK_SIZE, &bs);
887 if (rc == EOK)
888 *bsize = (size_t) bs;
889
890 return rc;
[1fbe064b]891}
892
[08232ee]893/** Get total number of blocks on block device. */
[ed903174]894static int get_num_blocks(int dev_phone, aoff64_t *nblocks)
[08232ee]895{
[96b02eb9]896 sysarg_t nb_l, nb_h;
[08232ee]897 int rc;
898
899 rc = async_req_0_2(dev_phone, BD_GET_NUM_BLOCKS, &nb_l, &nb_h);
900 if (rc == EOK) {
[ed903174]901 *nblocks = (aoff64_t) MERGE_LOUP32(nb_l, nb_h);
[08232ee]902 }
903
904 return rc;
905}
906
[f092718]907/** Convert logical block address to physical block address. */
908static aoff64_t ba_ltop(devcon_t *devcon, aoff64_t lba)
909{
910 assert(devcon->cache != NULL);
911 return lba * devcon->cache->blocks_cluster;
912}
913
[fc840d9]914/** @}
915 */
Note: See TracBrowser for help on using the repository browser.