Subversion Repositories HelenOS

Rev

Rev 3503 | Rev 3506 | Go to most recent revision | Details | Compare with Previous | Last modification | View Log | RSS feed

Rev Author Line No. Line
2627 jermar 1
/*
2793 jermar 2
 * Copyright (c) 2008 Jakub Jermar
2627 jermar 3
 * All rights reserved.
4
 *
5
 * Redistribution and use in source and binary forms, with or without
6
 * modification, are permitted provided that the following conditions
7
 * are met:
8
 *
9
 * - Redistributions of source code must retain the above copyright
10
 *   notice, this list of conditions and the following disclaimer.
11
 * - Redistributions in binary form must reproduce the above copyright
12
 *   notice, this list of conditions and the following disclaimer in the
13
 *   documentation and/or other materials provided with the distribution.
14
 * - The name of the author may not be used to endorse or promote products
15
 *   derived from this software without specific prior written permission.
16
 *
17
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27
 */
28
 
29
/** @addtogroup fs
30
 * @{
31
 */ 
32
 
33
/**
34
 * @file	fat_ops.c
35
 * @brief	Implementation of VFS operations for the FAT file system server.
36
 */
37
 
38
#include "fat.h"
3505 jermar 39
#include "fat_dentry.h"
40
#include "fat_fat.h"
2638 jermar 41
#include "../../vfs/vfs.h"
2793 jermar 42
#include <libfs.h>
2627 jermar 43
#include <ipc/ipc.h>
3257 jermar 44
#include <ipc/services.h>
45
#include <ipc/devmap.h>
2627 jermar 46
#include <async.h>
47
#include <errno.h>
2793 jermar 48
#include <string.h>
2798 jermar 49
#include <byteorder.h>
2831 jermar 50
#include <libadt/hash_table.h>
51
#include <libadt/list.h>
52
#include <assert.h>
2856 jermar 53
#include <futex.h>
3257 jermar 54
#include <sys/mman.h>
3499 jermar 55
#include <align.h>
2627 jermar 56
 
2843 jermar 57
#define BS_BLOCK		0
3253 jermar 58
#define BS_SIZE			512
2843 jermar 59
 
2951 jermar 60
/** Futex protecting the list of cached free FAT nodes. */
61
static futex_t ffn_futex = FUTEX_INITIALIZER;
2843 jermar 62
 
2951 jermar 63
/** List of cached free FAT nodes. */
64
static LIST_INITIALIZE(ffn_head);
65
 
3308 jermar 66
#define min(a, b)		((a) < (b) ? (a) : (b))
67
 
3257 jermar 68
static int dev_phone = -1;		/* FIXME */
69
static void *dev_buffer = NULL;		/* FIXME */
70
 
2910 jermar 71
/* TODO move somewhere else */
2822 jermar 72
typedef struct {
73
	void *data;
3257 jermar 74
	size_t size;
3499 jermar 75
	bool dirty;
2822 jermar 76
} block_t;
77
 
3253 jermar 78
static block_t *block_get(dev_handle_t dev_handle, off_t offset, size_t bs)
2793 jermar 79
{
3257 jermar 80
	/* FIXME */
81
	block_t *b;
82
	off_t bufpos = 0;
83
	size_t buflen = 0;
3272 jermar 84
	off_t pos = offset * bs;
3257 jermar 85
 
86
	assert(dev_phone != -1);
87
	assert(dev_buffer);
88
 
89
	b = malloc(sizeof(block_t));
90
	if (!b)
91
		return NULL;
92
 
93
	b->data = malloc(bs);
94
	if (!b->data) {
95
		free(b);
96
		return NULL;
97
	}
98
	b->size = bs;
99
 
3272 jermar 100
	if (!libfs_blockread(dev_phone, dev_buffer, &bufpos, &buflen, &pos,
3257 jermar 101
	    b->data, bs, bs)) {
102
		free(b->data);
103
		free(b);
104
		return NULL;
105
	}
106
 
107
	return b;
2793 jermar 108
}
109
 
2822 jermar 110
static void block_put(block_t *block)
2793 jermar 111
{
3257 jermar 112
	/* FIXME */
113
	free(block->data);
114
	free(block);
2793 jermar 115
}
116
 
3503 jermar 117
#define FAT1		0
118
 
2859 jermar 119
#define FAT_BS(b)		((fat_bs_t *)((b)->data))
120
 
2864 jermar 121
#define FAT_CLST_RES0	0x0000
3119 jermar 122
#define FAT_CLST_RES1	0x0001
2859 jermar 123
#define FAT_CLST_FIRST	0x0002
124
#define FAT_CLST_BAD	0xfff7
125
#define FAT_CLST_LAST1	0xfff8
126
#define FAT_CLST_LAST8  0xffff
127
 
3119 jermar 128
/* internally used to mark root directory's parent */
129
#define FAT_CLST_ROOTPAR	FAT_CLST_RES0
130
/* internally used to mark root directory */
131
#define FAT_CLST_ROOT		FAT_CLST_RES1
132
 
2891 jermar 133
#define fat_block_get(np, off) \
134
    _fat_block_get((np)->idx->dev_handle, (np)->firstc, (off))
135
 
136
static block_t *
2893 jermar 137
_fat_block_get(dev_handle_t dev_handle, fat_cluster_t firstc, off_t offset)
2859 jermar 138
{
139
	block_t *bb;
140
	block_t *b;
141
	unsigned bps;
142
	unsigned spc;
143
	unsigned rscnt;		/* block address of the first FAT */
144
	unsigned fatcnt;
145
	unsigned rde;
146
	unsigned rds;		/* root directory size */
147
	unsigned sf;
148
	unsigned ssa;		/* size of the system area */
149
	unsigned clusters;
2893 jermar 150
	fat_cluster_t clst = firstc;
2859 jermar 151
	unsigned i;
152
 
3253 jermar 153
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
2859 jermar 154
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
155
	spc = FAT_BS(bb)->spc;
156
	rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
157
	fatcnt = FAT_BS(bb)->fatcnt;
158
	rde = uint16_t_le2host(FAT_BS(bb)->root_ent_max);
159
	sf = uint16_t_le2host(FAT_BS(bb)->sec_per_fat);
160
	block_put(bb);
161
 
162
	rds = (sizeof(fat_dentry_t) * rde) / bps;
163
	rds += ((sizeof(fat_dentry_t) * rde) % bps != 0);
164
	ssa = rscnt + fatcnt * sf + rds;
165
 
3119 jermar 166
	if (firstc == FAT_CLST_ROOT) {
2859 jermar 167
		/* root directory special case */
168
		assert(offset < rds);
3253 jermar 169
		b = block_get(dev_handle, rscnt + fatcnt * sf + offset, bps);
2859 jermar 170
		return b;
171
	}
172
 
173
	clusters = offset / spc; 
174
	for (i = 0; i < clusters; i++) {
175
		unsigned fsec;	/* sector offset relative to FAT1 */
176
		unsigned fidx;	/* FAT1 entry index */
177
 
178
		assert(clst >= FAT_CLST_FIRST && clst < FAT_CLST_BAD);
2864 jermar 179
		fsec = (clst * sizeof(fat_cluster_t)) / bps;
180
		fidx = clst % (bps / sizeof(fat_cluster_t));
2859 jermar 181
		/* read FAT1 */
3253 jermar 182
		b = block_get(dev_handle, rscnt + fsec, bps);
2864 jermar 183
		clst = uint16_t_le2host(((fat_cluster_t *)b->data)[fidx]);
2859 jermar 184
		assert(clst != FAT_CLST_BAD);
185
		assert(clst < FAT_CLST_LAST1);
186
		block_put(b);
187
	}
188
 
2891 jermar 189
	b = block_get(dev_handle, ssa + (clst - FAT_CLST_FIRST) * spc +
3253 jermar 190
	    offset % spc, bps);
2859 jermar 191
 
192
	return b;
193
}
194
 
3325 jermar 195
/** Return number of blocks allocated to a file.
196
 *
197
 * @param dev_handle	Device handle of the device with the file.
198
 * @param firstc	First cluster of the file.
199
 *
200
 * @return		Number of blocks allocated to the file.
201
 */
202
static uint16_t 
203
_fat_blcks_get(dev_handle_t dev_handle, fat_cluster_t firstc)
204
{
205
	block_t *bb;
206
	block_t *b;
207
	unsigned bps;
208
	unsigned spc;
209
	unsigned rscnt;		/* block address of the first FAT */
210
	unsigned clusters = 0;
211
	fat_cluster_t clst = firstc;
212
 
213
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
214
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
215
	spc = FAT_BS(bb)->spc;
216
	rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
217
	block_put(bb);
218
 
219
	if (firstc == FAT_CLST_RES0) {
220
		/* No space allocated to the file. */
221
		return 0;
222
	}
223
 
224
	while (clst < FAT_CLST_LAST1) {
225
		unsigned fsec;	/* sector offset relative to FAT1 */
226
		unsigned fidx;	/* FAT1 entry index */
227
 
228
		assert(clst >= FAT_CLST_FIRST);
229
		fsec = (clst * sizeof(fat_cluster_t)) / bps;
230
		fidx = clst % (bps / sizeof(fat_cluster_t));
231
		/* read FAT1 */
232
		b = block_get(dev_handle, rscnt + fsec, bps);
233
		clst = uint16_t_le2host(((fat_cluster_t *)b->data)[fidx]);
234
		assert(clst != FAT_CLST_BAD);
235
		block_put(b);
236
		clusters++;
237
	}
238
 
239
	return clusters * spc;
240
}
241
 
2831 jermar 242
static void fat_node_initialize(fat_node_t *node)
2793 jermar 243
{
2951 jermar 244
	futex_initialize(&node->lock, 1);
2864 jermar 245
	node->idx = NULL;
2831 jermar 246
	node->type = 0;
247
	link_initialize(&node->ffn_link);
248
	node->size = 0;
249
	node->lnkcnt = 0;
250
	node->refcnt = 0;
251
	node->dirty = false;
2793 jermar 252
}
253
 
2843 jermar 254
static uint16_t fat_bps_get(dev_handle_t dev_handle)
255
{
256
	block_t *bb;
257
	uint16_t bps;
258
 
3253 jermar 259
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
2843 jermar 260
	assert(bb != NULL);
2859 jermar 261
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
2843 jermar 262
	block_put(bb);
263
 
264
	return bps;
265
}
266
 
2893 jermar 267
static void fat_node_sync(fat_node_t *node)
2831 jermar 268
{
269
	/* TODO */
270
}
271
 
2951 jermar 272
/** Internal version of fat_node_get().
273
 *
274
 * @param idxp		Locked index structure.
275
 */
276
static void *fat_node_get_core(fat_idx_t *idxp)
2831 jermar 277
{
2891 jermar 278
	block_t *b;
279
	fat_dentry_t *d;
3312 jermar 280
	fat_node_t *nodep = NULL;
2891 jermar 281
	unsigned bps;
282
	unsigned dps;
283
 
2951 jermar 284
	if (idxp->nodep) {
2891 jermar 285
		/*
286
		 * We are lucky.
287
		 * The node is already instantiated in memory.
288
		 */
2951 jermar 289
		futex_down(&idxp->nodep->lock);
290
		if (!idxp->nodep->refcnt++)
3312 jermar 291
			list_remove(&idxp->nodep->ffn_link);
2951 jermar 292
		futex_up(&idxp->nodep->lock);
293
		return idxp->nodep;
2891 jermar 294
	}
295
 
296
	/*
297
	 * We must instantiate the node from the file system.
298
	 */
299
 
2951 jermar 300
	assert(idxp->pfc);
2891 jermar 301
 
2951 jermar 302
	futex_down(&ffn_futex);
2893 jermar 303
	if (!list_empty(&ffn_head)) {
2951 jermar 304
		/* Try to use a cached free node structure. */
305
		fat_idx_t *idxp_tmp;
2893 jermar 306
		nodep = list_get_instance(ffn_head.next, fat_node_t, ffn_link);
2951 jermar 307
		if (futex_trydown(&nodep->lock) == ESYNCH_WOULD_BLOCK)
308
			goto skip_cache;
309
		idxp_tmp = nodep->idx;
310
		if (futex_trydown(&idxp_tmp->lock) == ESYNCH_WOULD_BLOCK) {
311
			futex_up(&nodep->lock);
312
			goto skip_cache;
313
		}
314
		list_remove(&nodep->ffn_link);
315
		futex_up(&ffn_futex);
2893 jermar 316
		if (nodep->dirty)
317
			fat_node_sync(nodep);
2951 jermar 318
		idxp_tmp->nodep = NULL;
319
		futex_up(&nodep->lock);
320
		futex_up(&idxp_tmp->lock);
2893 jermar 321
	} else {
2951 jermar 322
skip_cache:
2893 jermar 323
		/* Try to allocate a new node structure. */
2951 jermar 324
		futex_up(&ffn_futex);
2893 jermar 325
		nodep = (fat_node_t *)malloc(sizeof(fat_node_t));
326
		if (!nodep)
327
			return NULL;
328
	}
2891 jermar 329
	fat_node_initialize(nodep);
330
 
2951 jermar 331
	bps = fat_bps_get(idxp->dev_handle);
2891 jermar 332
	dps = bps / sizeof(fat_dentry_t);
333
 
2893 jermar 334
	/* Read the block that contains the dentry of interest. */
2951 jermar 335
	b = _fat_block_get(idxp->dev_handle, idxp->pfc,
336
	    (idxp->pdi * sizeof(fat_dentry_t)) / bps);
2891 jermar 337
	assert(b);
338
 
2951 jermar 339
	d = ((fat_dentry_t *)b->data) + (idxp->pdi % dps);
2893 jermar 340
	if (d->attr & FAT_ATTR_SUBDIR) {
341
		/* 
342
		 * The only directory which does not have this bit set is the
343
		 * root directory itself. The root directory node is handled
344
		 * and initialized elsewhere.
345
		 */
346
		nodep->type = FAT_DIRECTORY;
3282 jermar 347
		/*
3325 jermar 348
		 * Unfortunately, the 'size' field of the FAT dentry is not
349
		 * defined for the directory entry type. We must determine the
350
		 * size of the directory by walking the FAT.
3282 jermar 351
		 */
3325 jermar 352
		nodep->size = bps * _fat_blcks_get(idxp->dev_handle,
353
		    uint16_t_le2host(d->firstc));
2893 jermar 354
	} else {
355
		nodep->type = FAT_FILE;
3282 jermar 356
		nodep->size = uint32_t_le2host(d->size);
2893 jermar 357
	}
358
	nodep->firstc = uint16_t_le2host(d->firstc);
359
	nodep->lnkcnt = 1;
360
	nodep->refcnt = 1;
361
 
362
	block_put(b);
363
 
364
	/* Link the idx structure with the node structure. */
2951 jermar 365
	nodep->idx = idxp;
366
	idxp->nodep = nodep;
2893 jermar 367
 
368
	return nodep;
2831 jermar 369
}
370
 
2951 jermar 371
/** Instantiate a FAT in-core node. */
372
static void *fat_node_get(dev_handle_t dev_handle, fs_index_t index)
373
{
374
	void *node;
375
	fat_idx_t *idxp;
376
 
377
	idxp = fat_idx_get_by_index(dev_handle, index);
378
	if (!idxp)
379
		return NULL;
380
	/* idxp->lock held */
381
	node = fat_node_get_core(idxp);
382
	futex_up(&idxp->lock);
383
	return node;
384
}
385
 
2852 jermar 386
static void fat_node_put(void *node)
387
{
2910 jermar 388
	fat_node_t *nodep = (fat_node_t *)node;
389
 
2951 jermar 390
	futex_down(&nodep->lock);
2910 jermar 391
	if (!--nodep->refcnt) {
2951 jermar 392
		futex_down(&ffn_futex);
2910 jermar 393
		list_append(&nodep->ffn_link, &ffn_head);
2951 jermar 394
		futex_up(&ffn_futex);
2910 jermar 395
	}
2951 jermar 396
	futex_up(&nodep->lock);
2852 jermar 397
}
398
 
2857 jermar 399
static void *fat_create(int flags)
400
{
401
	return NULL;	/* not supported at the moment */
402
}
403
 
2858 jermar 404
static int fat_destroy(void *node)
2857 jermar 405
{
2858 jermar 406
	return ENOTSUP;	/* not supported at the moment */
2857 jermar 407
}
408
 
409
static bool fat_link(void *prnt, void *chld, const char *name)
410
{
411
	return false;	/* not supported at the moment */
412
}
413
 
414
static int fat_unlink(void *prnt, void *chld)
415
{
416
	return ENOTSUP;	/* not supported at the moment */
417
}
418
 
2793 jermar 419
static void *fat_match(void *prnt, const char *component)
420
{
421
	fat_node_t *parentp = (fat_node_t *)prnt;
422
	char name[FAT_NAME_LEN + 1 + FAT_EXT_LEN + 1];
2822 jermar 423
	unsigned i, j;
2828 jermar 424
	unsigned bps;		/* bytes per sector */
2822 jermar 425
	unsigned dps;		/* dentries per sector */
426
	unsigned blocks;
2793 jermar 427
	fat_dentry_t *d;
2822 jermar 428
	block_t *b;
2793 jermar 429
 
2953 jermar 430
	futex_down(&parentp->idx->lock);
2864 jermar 431
	bps = fat_bps_get(parentp->idx->dev_handle);
2828 jermar 432
	dps = bps / sizeof(fat_dentry_t);
433
	blocks = parentp->size / bps + (parentp->size % bps != 0);
2822 jermar 434
	for (i = 0; i < blocks; i++) {
435
		unsigned dentries;
2793 jermar 436
 
2864 jermar 437
		b = fat_block_get(parentp, i);
2822 jermar 438
		dentries = (i == blocks - 1) ?
439
		    parentp->size % sizeof(fat_dentry_t) :
440
		    dps;
441
		for (j = 0; j < dentries; j++) { 
442
			d = ((fat_dentry_t *)b->data) + j;
2845 jermar 443
			switch (fat_classify_dentry(d)) {
444
			case FAT_DENTRY_SKIP:
2822 jermar 445
				continue;
2845 jermar 446
			case FAT_DENTRY_LAST:
2822 jermar 447
				block_put(b);
2953 jermar 448
				futex_up(&parentp->idx->lock);
2822 jermar 449
				return NULL;
2845 jermar 450
			default:
451
			case FAT_DENTRY_VALID:
452
				dentry_name_canonify(d, name);
453
				break;
2822 jermar 454
			}
3272 jermar 455
			if (stricmp(name, component) == 0) {
2822 jermar 456
				/* hit */
2951 jermar 457
				void *node;
2953 jermar 458
				/*
459
				 * Assume tree hierarchy for locking.  We
460
				 * already have the parent and now we are going
461
				 * to lock the child.  Never lock in the oposite
462
				 * order.
463
				 */
2890 jermar 464
				fat_idx_t *idx = fat_idx_get_by_pos(
2881 jermar 465
				    parentp->idx->dev_handle, parentp->firstc,
2864 jermar 466
				    i * dps + j);
2953 jermar 467
				futex_up(&parentp->idx->lock);
2890 jermar 468
				if (!idx) {
469
					/*
470
					 * Can happen if memory is low or if we
471
					 * run out of 32-bit indices.
472
					 */
473
					block_put(b);
474
					return NULL;
475
				}
2951 jermar 476
				node = fat_node_get_core(idx);
477
				futex_up(&idx->lock);
2822 jermar 478
				block_put(b);
479
				return node;
480
			}
2793 jermar 481
		}
2822 jermar 482
		block_put(b);
2639 jermar 483
	}
2953 jermar 484
	futex_up(&parentp->idx->lock);
2793 jermar 485
	return NULL;
2638 jermar 486
}
487
 
2831 jermar 488
static fs_index_t fat_index_get(void *node)
489
{
490
	fat_node_t *fnodep = (fat_node_t *)node;
491
	if (!fnodep)
492
		return 0;
2864 jermar 493
	return fnodep->idx->index;
2831 jermar 494
}
495
 
496
static size_t fat_size_get(void *node)
497
{
498
	return ((fat_node_t *)node)->size;
499
}
500
 
501
static unsigned fat_lnkcnt_get(void *node)
502
{
503
	return ((fat_node_t *)node)->lnkcnt;
504
}
505
 
2845 jermar 506
static bool fat_has_children(void *node)
507
{
508
	fat_node_t *nodep = (fat_node_t *)node;
509
	unsigned bps;
510
	unsigned dps;
511
	unsigned blocks;
512
	block_t *b;
513
	unsigned i, j;
514
 
515
	if (nodep->type != FAT_DIRECTORY)
516
		return false;
517
 
2951 jermar 518
	futex_down(&nodep->idx->lock);
2864 jermar 519
	bps = fat_bps_get(nodep->idx->dev_handle);
2845 jermar 520
	dps = bps / sizeof(fat_dentry_t);
521
 
522
	blocks = nodep->size / bps + (nodep->size % bps != 0);
523
 
524
	for (i = 0; i < blocks; i++) {
525
		unsigned dentries;
526
		fat_dentry_t *d;
527
 
2864 jermar 528
		b = fat_block_get(nodep, i);
2845 jermar 529
		dentries = (i == blocks - 1) ?
530
		    nodep->size % sizeof(fat_dentry_t) :
531
		    dps;
532
		for (j = 0; j < dentries; j++) {
533
			d = ((fat_dentry_t *)b->data) + j;
534
			switch (fat_classify_dentry(d)) {
535
			case FAT_DENTRY_SKIP:
536
				continue;
537
			case FAT_DENTRY_LAST:
538
				block_put(b);
2951 jermar 539
				futex_up(&nodep->idx->lock);
2845 jermar 540
				return false;
541
			default:
542
			case FAT_DENTRY_VALID:
543
				block_put(b);
2951 jermar 544
				futex_up(&nodep->idx->lock);
2845 jermar 545
				return true;
546
			}
547
			block_put(b);
2951 jermar 548
			futex_up(&nodep->idx->lock);
2845 jermar 549
			return true;
550
		}
551
		block_put(b);
552
	}
553
 
2951 jermar 554
	futex_up(&nodep->idx->lock);
2845 jermar 555
	return false;
556
}
557
 
2844 jermar 558
static void *fat_root_get(dev_handle_t dev_handle)
559
{
3119 jermar 560
	return fat_node_get(dev_handle, 0);
2844 jermar 561
}
562
 
563
static char fat_plb_get_char(unsigned pos)
564
{
565
	return fat_reg.plb_ro[pos % PLB_SIZE];
566
}
567
 
2831 jermar 568
static bool fat_is_directory(void *node)
569
{
570
	return ((fat_node_t *)node)->type == FAT_DIRECTORY;
571
}
572
 
573
static bool fat_is_file(void *node)
574
{
575
	return ((fat_node_t *)node)->type == FAT_FILE;
576
}
577
 
2793 jermar 578
/** libfs operations */
579
libfs_ops_t fat_libfs_ops = {
580
	.match = fat_match,
581
	.node_get = fat_node_get,
2852 jermar 582
	.node_put = fat_node_put,
2857 jermar 583
	.create = fat_create,
584
	.destroy = fat_destroy,
585
	.link = fat_link,
586
	.unlink = fat_unlink,
2831 jermar 587
	.index_get = fat_index_get,
588
	.size_get = fat_size_get,
589
	.lnkcnt_get = fat_lnkcnt_get,
2845 jermar 590
	.has_children = fat_has_children,
2844 jermar 591
	.root_get = fat_root_get,
592
	.plb_get_char =	fat_plb_get_char,
2831 jermar 593
	.is_directory = fat_is_directory,
594
	.is_file = fat_is_file
2793 jermar 595
};
596
 
3110 jermar 597
void fat_mounted(ipc_callid_t rid, ipc_call_t *request)
598
{
599
	dev_handle_t dev_handle = (dev_handle_t) IPC_GET_ARG1(*request);
3119 jermar 600
	block_t *bb;
3257 jermar 601
	uint16_t bps;
3119 jermar 602
	uint16_t rde;
3110 jermar 603
	int rc;
604
 
3257 jermar 605
	/*
606
	 * For now, we don't bother to remember dev_handle, dev_phone or
607
	 * dev_buffer in some data structure. We use global variables because we
608
	 * know there will be at most one mount on this file system.
609
	 * Of course, this is a huge TODO item.
610
	 */
611
	dev_buffer = mmap(NULL, BS_SIZE, PROTO_READ | PROTO_WRITE,
612
	    MAP_ANONYMOUS | MAP_PRIVATE, 0, 0);
613
 
614
	if (!dev_buffer) {
615
		ipc_answer_0(rid, ENOMEM);
616
		return;
617
	}
618
 
619
	dev_phone = ipc_connect_me_to(PHONE_NS, SERVICE_DEVMAP,
620
	    DEVMAP_CONNECT_TO_DEVICE, dev_handle);
621
 
622
	if (dev_phone < 0) {
623
		munmap(dev_buffer, BS_SIZE);
624
		ipc_answer_0(rid, dev_phone);
625
		return;
626
	}
627
 
628
	rc = ipc_share_out_start(dev_phone, dev_buffer,
629
	    AS_AREA_READ | AS_AREA_WRITE);
630
	if (rc != EOK) {
631
	    	munmap(dev_buffer, BS_SIZE);
632
		ipc_answer_0(rid, rc);
633
		return;
634
	}
635
 
3119 jermar 636
	/* Read the number of root directory entries. */
3253 jermar 637
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
3257 jermar 638
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
3119 jermar 639
	rde = uint16_t_le2host(FAT_BS(bb)->root_ent_max);
640
	block_put(bb);
641
 
3257 jermar 642
	if (bps != BS_SIZE) {
643
		munmap(dev_buffer, BS_SIZE);
644
		ipc_answer_0(rid, ENOTSUP);
645
		return;
646
	}
647
 
3110 jermar 648
	rc = fat_idx_init_by_dev_handle(dev_handle);
649
	if (rc != EOK) {
3257 jermar 650
	    	munmap(dev_buffer, BS_SIZE);
3110 jermar 651
		ipc_answer_0(rid, rc);
652
		return;
653
	}
654
 
3119 jermar 655
	/* Initialize the root node. */
656
	fat_node_t *rootp = (fat_node_t *)malloc(sizeof(fat_node_t));
657
	if (!rootp) {
3257 jermar 658
	    	munmap(dev_buffer, BS_SIZE);
3119 jermar 659
		fat_idx_fini_by_dev_handle(dev_handle);
660
		ipc_answer_0(rid, ENOMEM);
661
		return;
662
	}
663
	fat_node_initialize(rootp);
664
 
665
	fat_idx_t *ridxp = fat_idx_get_by_pos(dev_handle, FAT_CLST_ROOTPAR, 0);
666
	if (!ridxp) {
3257 jermar 667
	    	munmap(dev_buffer, BS_SIZE);
3119 jermar 668
		free(rootp);
669
		fat_idx_fini_by_dev_handle(dev_handle);
670
		ipc_answer_0(rid, ENOMEM);
671
		return;
672
	}
673
	assert(ridxp->index == 0);
674
	/* ridxp->lock held */
675
 
676
	rootp->type = FAT_DIRECTORY;
677
	rootp->firstc = FAT_CLST_ROOT;
678
	rootp->refcnt = 1;
3352 jermar 679
	rootp->lnkcnt = 0;	/* FS root is not linked */
3119 jermar 680
	rootp->size = rde * sizeof(fat_dentry_t);
681
	rootp->idx = ridxp;
682
	ridxp->nodep = rootp;
683
 
684
	futex_up(&ridxp->lock);
685
 
3352 jermar 686
	ipc_answer_3(rid, EOK, ridxp->index, rootp->size, rootp->lnkcnt);
3110 jermar 687
}
688
 
689
void fat_mount(ipc_callid_t rid, ipc_call_t *request)
690
{
691
	ipc_answer_0(rid, ENOTSUP);
692
}
693
 
2627 jermar 694
void fat_lookup(ipc_callid_t rid, ipc_call_t *request)
695
{
2793 jermar 696
	libfs_lookup(&fat_libfs_ops, fat_reg.fs_handle, rid, request);
2627 jermar 697
}
698
 
3307 jermar 699
void fat_read(ipc_callid_t rid, ipc_call_t *request)
700
{
701
	dev_handle_t dev_handle = (dev_handle_t)IPC_GET_ARG1(*request);
702
	fs_index_t index = (fs_index_t)IPC_GET_ARG2(*request);
703
	off_t pos = (off_t)IPC_GET_ARG3(*request);
704
	fat_node_t *nodep = (fat_node_t *)fat_node_get(dev_handle, index);
3308 jermar 705
	uint16_t bps = fat_bps_get(dev_handle);
706
	size_t bytes;
3335 jermar 707
	block_t *b;
3308 jermar 708
 
3307 jermar 709
	if (!nodep) {
710
		ipc_answer_0(rid, ENOENT);
711
		return;
712
	}
713
 
714
	ipc_callid_t callid;
715
	size_t len;
3314 jermar 716
	if (!ipc_data_read_receive(&callid, &len)) {
3307 jermar 717
		fat_node_put(nodep);
718
		ipc_answer_0(callid, EINVAL);
719
		ipc_answer_0(rid, EINVAL);
720
		return;
721
	}
722
 
723
	if (nodep->type == FAT_FILE) {
3335 jermar 724
		/*
725
		 * Our strategy for regular file reads is to read one block at
726
		 * most and make use of the possibility to return less data than
727
		 * requested. This keeps the code very simple.
728
		 */
3308 jermar 729
		bytes = min(len, bps - pos % bps);
730
		b = fat_block_get(nodep, pos / bps);
731
		(void) ipc_data_read_finalize(callid, b->data + pos % bps,
732
		    bytes);
733
		block_put(b);
3307 jermar 734
	} else {
3335 jermar 735
		unsigned bnum;
736
		off_t spos = pos;
737
		char name[FAT_NAME_LEN + 1 + FAT_EXT_LEN + 1];
738
		fat_dentry_t *d;
739
 
3307 jermar 740
		assert(nodep->type == FAT_DIRECTORY);
3335 jermar 741
		assert(nodep->size % bps == 0);
742
		assert(bps % sizeof(fat_dentry_t) == 0);
743
 
744
		/*
745
		 * Our strategy for readdir() is to use the position pointer as
746
		 * an index into the array of all dentries. On entry, it points
747
		 * to the first unread dentry. If we skip any dentries, we bump
748
		 * the position pointer accordingly.
749
		 */
750
		bnum = (pos * sizeof(fat_dentry_t)) / bps;
751
		while (bnum < nodep->size / bps) {
752
			off_t o;
753
 
754
			b = fat_block_get(nodep, bnum);
755
			for (o = pos % (bps / sizeof(fat_dentry_t));
756
			    o < bps / sizeof(fat_dentry_t);
757
			    o++, pos++) {
758
				d = ((fat_dentry_t *)b->data) + o;
759
				switch (fat_classify_dentry(d)) {
760
				case FAT_DENTRY_SKIP:
761
					continue;
762
				case FAT_DENTRY_LAST:
763
					block_put(b);
764
					goto miss;
765
				default:
766
				case FAT_DENTRY_VALID:
767
					dentry_name_canonify(d, name);
768
					block_put(b);
769
					goto hit;
770
				}
771
			}
772
			block_put(b);
773
			bnum++;
774
		}
775
miss:
3307 jermar 776
		fat_node_put(nodep);
3335 jermar 777
		ipc_answer_0(callid, ENOENT);
778
		ipc_answer_1(rid, ENOENT, 0);
3307 jermar 779
		return;
3335 jermar 780
hit:
781
		(void) ipc_data_read_finalize(callid, name, strlen(name) + 1);
782
		bytes = (pos - spos) + 1;
3307 jermar 783
	}
784
 
785
	fat_node_put(nodep);
3308 jermar 786
	ipc_answer_1(rid, EOK, (ipcarg_t)bytes);
3307 jermar 787
}
788
 
3501 jermar 789
/** Fill the gap between EOF and a new file position.
790
 *
791
 * @param nodep		FAT node with the gap.
792
 * @param mcl		First cluster in an independent cluster chain that will
793
 *			be later appended to the end of the node's own cluster
794
 *			chain. If pos is still in the last allocated cluster,
795
 *			this argument is ignored.
796
 * @param pos		Position in the last node block.
797
 */
3500 jermar 798
static void
3501 jermar 799
fat_fill_gap(fat_node_t *nodep, fat_cluster_t mcl, off_t pos)
3499 jermar 800
{
3501 jermar 801
	uint16_t bps;
802
	unsigned spc;
803
	block_t *bb, *b;
804
	off_t o, boundary;
805
 
806
	bb = block_get(nodep->idx->dev_handle, BS_BLOCK, BS_SIZE);
807
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
808
	spc = FAT_BS(bb)->spc;
809
	block_put(bb);
810
 
811
	boundary = ROUND_UP(nodep->size, bps * spc);
812
 
813
	/* zero out already allocated space */
814
	for (o = nodep->size - 1; o < pos && o < boundary;
815
	    o = ALIGN_DOWN(o + bps, bps)) {
816
		b = fat_block_get(nodep, o / bps);
817
		memset(b->data + o % bps, 0, bps - o % bps);
818
		b->dirty = true;		/* need to sync node */
819
		block_put(b);
820
	}
821
 
822
	if (o >= pos)
823
		return;
824
 
825
	/* zero out the initial part of the new cluster chain */
826
	for (o = boundary; o < pos; o += bps) {
827
		b = _fat_block_get(nodep->idx->dev_handle, mcl,
828
		    (o - boundary) / bps);
829
		memset(b->data, 0, min(bps, pos - o));
3503 jermar 830
		b->dirty = true;		/* need to sync node */
3501 jermar 831
		block_put(b);
832
	}
3499 jermar 833
}
834
 
3503 jermar 835
static void
836
fat_mark_cluster(dev_handle_t dev_handle, unsigned fatno, fat_cluster_t clst,
837
    fat_cluster_t value)
838
{
839
	/* TODO */
840
}
841
 
842
static void
843
fat_alloc_shadow_clusters(dev_handle_t dev_handle, fat_cluster_t *lifo,
844
    unsigned nclsts)
845
{
846
	/* TODO */
847
}
848
 
3500 jermar 849
static int
3503 jermar 850
fat_alloc_clusters(dev_handle_t dev_handle, unsigned nclsts, fat_cluster_t *mcl,
851
    fat_cluster_t *lcl)
3500 jermar 852
{
3503 jermar 853
	uint16_t bps;
854
	uint16_t rscnt;
855
	uint16_t sf;
856
	block_t *bb, *blk;
857
	fat_cluster_t *lifo;	/* stack for storing free cluster numbers */ 
858
	unsigned found = 0;	/* top of the free cluster number stack */
859
	unsigned b, c, cl; 
860
 
861
	lifo = (fat_cluster_t *) malloc(nclsts * sizeof(fat_cluster_t));
862
	if (lifo)
863
		return ENOMEM;
864
 
865
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
866
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
867
	rscnt = uint16_t_le2host(FAT_BS(bb)->rscnt);
868
	sf = uint16_t_le2host(FAT_BS(bb)->sec_per_fat);
869
	block_put(bb);
870
 
871
	/*
872
	 * Search FAT1 for unused clusters.
873
	 */
874
	for (b = 0, cl = 0; b < sf; blk++) {
875
		blk = block_get(dev_handle, rscnt + b, bps);
876
		for (c = 0; c < bps / sizeof(fat_cluster_t); c++, cl++) {
877
			fat_cluster_t *clst = (fat_cluster_t *)blk->data + c;
878
			if (*clst == FAT_CLST_RES0) {
879
				/*
880
				 * The cluster is free. Put it into our stack
881
				 * of found clusters and mark it as non-free.
882
				 */
883
				lifo[found] = cl;
884
				if (found == 0)
885
					*clst = FAT_CLST_LAST1;
886
				else
887
					*clst = lifo[found - 1];
888
				blk->dirty = true;	/* need to sync block */
889
				if (++found == nclsts) {
890
					/* we are almost done */
891
					block_put(blk);
892
					/* update the shadow copies of FAT */
893
					fat_alloc_shadow_clusters(dev_handle,
894
					    lifo, nclsts);
895
					*mcl = lifo[found - 1];
896
					*lcl = lifo[0];
897
					free(lifo);
898
					return EOK;
899
				}
900
			}
901
		}
902
		block_put(blk);
903
	}
904
 
905
	/*
906
	 * We could not find enough clusters. Now we need to free the clusters
907
	 * we have allocated so far.
908
	 */
909
	while (found--)
910
		fat_mark_cluster(dev_handle, FAT1, lifo[found], FAT_CLST_RES0);
911
 
912
	free(lifo);
913
	return ENOSPC;
3500 jermar 914
}
915
 
916
static void
917
fat_append_clusters(fat_node_t *nodep, fat_cluster_t mcl)
918
{
919
}
920
 
3497 jermar 921
void fat_write(ipc_callid_t rid, ipc_call_t *request)
922
{
3499 jermar 923
	dev_handle_t dev_handle = (dev_handle_t)IPC_GET_ARG1(*request);
924
	fs_index_t index = (fs_index_t)IPC_GET_ARG2(*request);
925
	off_t pos = (off_t)IPC_GET_ARG3(*request);
926
	fat_node_t *nodep = (fat_node_t *)fat_node_get(dev_handle, index);
927
	size_t bytes;
928
	block_t *b, *bb;
929
	uint16_t bps;
930
	unsigned spc;
3501 jermar 931
	off_t boundary;
3499 jermar 932
 
933
	if (!nodep) {
934
		ipc_answer_0(rid, ENOENT);
935
		return;
936
	}
937
 
938
	/* XXX remove me when you are ready */
939
	{
940
		ipc_answer_0(rid, ENOTSUP);
941
		fat_node_put(nodep);
942
		return;
943
	}
944
 
945
	ipc_callid_t callid;
946
	size_t len;
947
	if (!ipc_data_write_receive(&callid, &len)) {
948
		fat_node_put(nodep);
949
		ipc_answer_0(callid, EINVAL);
950
		ipc_answer_0(rid, EINVAL);
951
		return;
952
	}
953
 
954
	/*
955
	 * In all scenarios, we will attempt to write out only one block worth
956
	 * of data at maximum. There might be some more efficient approaches,
957
	 * but this one greatly simplifies fat_write(). Note that we can afford
958
	 * to do this because the client must be ready to handle the return
959
	 * value signalizing a smaller number of bytes written. 
960
	 */ 
961
	bytes = min(len, bps - pos % bps);
962
 
963
	bb = block_get(dev_handle, BS_BLOCK, BS_SIZE);
964
	bps = uint16_t_le2host(FAT_BS(bb)->bps);
965
	spc = FAT_BS(bb)->spc;
966
	block_put(bb);
967
 
3501 jermar 968
	boundary = ROUND_UP(nodep->size, bps * spc);
969
	if (pos < boundary) {
3499 jermar 970
		/*
971
		 * This is the easier case - we are either overwriting already
972
		 * existing contents or writing behind the EOF, but still within
973
		 * the limits of the last cluster. The node size may grow to the
974
		 * next block size boundary.
975
		 */
3500 jermar 976
		fat_fill_gap(nodep, FAT_CLST_RES0, pos);
3499 jermar 977
		b = fat_block_get(nodep, pos / bps);
978
		(void) ipc_data_write_finalize(callid, b->data + pos % bps,
979
		    bytes);
980
		b->dirty = true;		/* need to sync block */
3500 jermar 981
		block_put(b);
3499 jermar 982
		if (pos + bytes > nodep->size) {
983
			nodep->size = pos + bytes;
984
			nodep->dirty = true;	/* need to sync node */
985
		}
986
		fat_node_put(nodep);
987
		ipc_answer_1(rid, EOK, bytes);	
988
		return;
989
	} else {
990
		/*
991
		 * This is the more difficult case. We must allocate new
992
		 * clusters for the node and zero them out.
993
		 */
3500 jermar 994
		int status;
3499 jermar 995
		unsigned nclsts;
3500 jermar 996
		fat_cluster_t mcl, lcl;
997
 
3501 jermar 998
		nclsts = (ROUND_UP(pos + bytes, bps * spc) - boundary) /
3500 jermar 999
		    bps * spc;
1000
		/* create an independent chain of nclsts clusters in all FATs */
3503 jermar 1001
		status = fat_alloc_clusters(dev_handle, nclsts, &mcl, &lcl);
3500 jermar 1002
		if (status != EOK) {
1003
			/* could not allocate a chain of nclsts clusters */
1004
			fat_node_put(nodep);
1005
			ipc_answer_0(callid, status);
1006
			ipc_answer_0(rid, status);
1007
			return;
1008
		}
1009
		/* zero fill any gaps */
1010
		fat_fill_gap(nodep, mcl, pos);
1011
		b = _fat_block_get(dev_handle, lcl, (pos / bps) % spc);
1012
		(void) ipc_data_write_finalize(callid, b->data + pos % bps,
1013
		    bytes);
3501 jermar 1014
		b->dirty = true;		/* need to sync block */
3500 jermar 1015
		block_put(b);
1016
		/*
1017
		 * Append the cluster chain starting in mcl to the end of the
1018
		 * node's cluster chain.
1019
		 */
1020
		fat_append_clusters(nodep, mcl);
1021
		nodep->size = pos + bytes;
3501 jermar 1022
		nodep->dirty = true;		/* need to sync node */
3500 jermar 1023
		fat_node_put(nodep);
1024
		ipc_answer_1(rid, EOK, bytes);
1025
		return;
3499 jermar 1026
	}
3497 jermar 1027
}
1028
 
2627 jermar 1029
/**
1030
 * @}
1031
 */