4 * This file is part of BeRTOS.
6 * Bertos is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 * As a special exception, you may use this file as part of a free software
21 * library without restriction. Specifically, if other files instantiate
22 * templates or use macros or inline functions from this file, or you compile
23 * this file and link it with other files to produce an executable, this
24 * file does not by itself cause the resulting executable to be covered by
25 * the GNU General Public License. This exception does not however
26 * invalidate any other reasons why the executable file might be covered by
27 * the GNU General Public License.
29 * Copyright 2007 Develer S.r.l. (http://www.develer.com/)
33 * \brief BattFS: a filesystem for embedded platforms (implementation).
37 * \author Francesco Sacchi <batt@develer.com>
43 #include <cfg/debug.h>
44 #include <cfg/macros.h> /* MIN, MAX */
45 #include <cpu/byteorder.h> /* cpu_to_xx */
47 #define LOG_LEVEL LOG_LVL_INFO
48 #define LOG_FORMAT LOG_FMT_VERBOSE
51 #include <string.h> /* memset, memmove */
55 * Convert from memory representation to disk structure.
56 * \note filesystem is in little-endian format.
58 INLINE void battfs_to_disk(struct BattFsPageHeader *hdr, uint8_t *buf)
60 STATIC_ASSERT(BATTFS_HEADER_LEN == 12);
64 buf[2] = hdr->fill >> 8;
67 buf[4] = hdr->pgoff >> 8;
70 * Sequence number is 40 bits long.
71 * No need to take care of wraparonds: the memory will die first!
74 buf[6] = hdr->seq >> 8;
75 buf[7] = hdr->seq >> 16;
76 buf[8] = hdr->seq >> 24;
77 buf[9] = hdr->seq >> 32;
80 * This field must be the last one!
81 * This is needed because if the page is only partially
82 * written, we can use this to detect it.
85 buf[11] = hdr->fcs >> 8;
89 * Convert from disk structure to memory representation.
90 * \note filesystem is in little-endian format.
92 INLINE void disk_to_battfs(uint8_t *buf, struct BattFsPageHeader *hdr)
94 STATIC_ASSERT(BATTFS_HEADER_LEN == 12);
96 hdr->fill = buf[2] << 8 | buf[1];
97 hdr->pgoff = buf[4] << 8 | buf[3];
98 hdr->seq = (seq_t)buf[9] << 32 | (seq_t)buf[8] << 24 | (seq_t)buf[7] << 16 | buf[6] << 8 | buf[5];
99 hdr->fcs = buf[11] << 8 | buf[10];
103 * Compute the fcs of the header.
105 static fcs_t computeFcs(struct BattFsPageHeader *hdr)
107 uint8_t buf[BATTFS_HEADER_LEN];
110 battfs_to_disk(hdr, buf);
112 /* fcs is at the end of whole header */
113 rotating_update(buf, BATTFS_HEADER_LEN - sizeof(fcs_t), &cks);
119 * Read header of page \a page.
120 * \return true on success, false otherwise.
122 static bool battfs_readHeader(struct BattFsSuper *disk, pgcnt_t page, struct BattFsPageHeader *hdr)
124 uint8_t buf[BATTFS_HEADER_LEN];
126 * Read header from disk.
127 * Header is actually a footer, and so
128 * resides at page end.
130 if (disk->read(disk, page, disk->page_size - BATTFS_HEADER_LEN, buf, BATTFS_HEADER_LEN)
131 != BATTFS_HEADER_LEN)
133 LOG_ERR("Error: page[%d]\n", page);
138 disk_to_battfs(buf, hdr);
144 * Write header of page \a page.
145 * \return true on success, false otherwise.
147 static bool battfs_writeHeader(struct BattFsSuper *disk, pgcnt_t page, struct BattFsPageHeader *hdr)
149 uint8_t buf[BATTFS_HEADER_LEN];
152 battfs_to_disk(hdr, buf);
155 * write header to disk.
156 * Header is actually a footer, and so
157 * resides at page end.
159 if (disk->write(disk, page, disk->page_size - BATTFS_HEADER_LEN, buf, BATTFS_HEADER_LEN)
160 != BATTFS_HEADER_LEN)
162 LOG_ERR("Error: page[%d]\n", page);
169 * Count the number of pages from
170 * inode 0 to \a inode in \a filelen_table.
172 static pgcnt_t countPages(pgoff_t *filelen_table, inode_t inode)
176 for (inode_t i = 0; i < inode; i++)
177 cnt += filelen_table[i];
183 * Move all pages in page allocation array from \a src to \a src + \a offset.
184 * The number of pages moved is page_count - MAX(dst, src).
186 static void movePages(struct BattFsSuper *disk, pgcnt_t src, int offset)
188 pgcnt_t dst = src + offset;
189 memmove(&disk->page_array[dst], &disk->page_array[src], (disk->page_count - MAX(dst, src)) * sizeof(pgcnt_t));
193 /* Fill empty space in array with sentinel */
194 for (pgcnt_t page = disk->page_count + offset; page < disk->page_count; page++)
195 disk->page_array[page] = PAGE_UNSET_SENTINEL;
200 * Count number of pages per file on \a disk.
201 * This information is registered in \a filelen_table.
202 * Array index represent file inode, while value contained
203 * is the number of pages used by that file.
205 * \return true if ok, false on disk read errors.
206 * \note The whole disk is scanned once.
208 static bool countDiskFilePages(struct BattFsSuper *disk, pgoff_t *filelen_table)
210 BattFsPageHeader hdr;
211 disk->free_page_start = 0;
213 /* Count the number of disk page per file */
214 for (pgcnt_t page = 0; page < disk->page_count; page++)
216 if (!battfs_readHeader(disk, page, &hdr))
219 /* Increase free space */
220 disk->free_bytes += disk->page_size - BATTFS_HEADER_LEN;
222 /* Check header FCS */
223 if (hdr.fcs == computeFcs(&hdr))
225 ASSERT(hdr.fill <= disk->page_size - BATTFS_HEADER_LEN);
227 /* Page is valid and is owned by a file */
228 filelen_table[hdr.inode]++;
230 /* Keep trace of free space */
231 disk->free_bytes -= hdr.fill;
232 disk->free_page_start++;
235 LOG_INFO("free_bytes:%d, free_page_start:%d\n", disk->free_bytes, disk->free_page_start);
241 * Fill page allocation array of \a disk
242 * using file lenghts in \a filelen_table.
244 * The page allocation array is an array containings all file infos.
245 * Is ordered by file, and within each file is ordered by page offset
247 * e.g. : at page array[0] you will find page address of the first page
248 * of the first file (if present).
249 * Free blocks are allocated after the last file, starting from invalid ones
250 * and continuing with the marked free ones.
252 * \return true if ok, false on disk read errors.
253 * \note The whole disk is scanned once.
255 static bool fillPageArray(struct BattFsSuper *disk, pgoff_t *filelen_table)
257 BattFsPageHeader hdr;
258 pgcnt_t curr_free_page = disk->free_page_start;
259 /* Fill page allocation array */
260 for (pgcnt_t page = 0; page < disk->page_count; page++)
262 if (!battfs_readHeader(disk, page, &hdr))
265 /* Check header FCS */
266 if (hdr.fcs == computeFcs(&hdr))
268 /* Compute array position */
269 pgcnt_t array_pos = countPages(filelen_table, hdr.inode);
270 array_pos += hdr.pgoff;
273 /* Check if position is already used by another page of the same file */
274 if (disk->page_array[array_pos] == PAGE_UNSET_SENTINEL)
275 disk->page_array[array_pos] = page;
278 BattFsPageHeader hdr_prv;
280 if (!battfs_readHeader(disk, disk->page_array[array_pos], &hdr_prv))
283 /* Check header FCS */
284 ASSERT(hdr_prv.fcs == computeFcs(&hdr_prv));
286 /* Only the very same page with a different seq number can be here */
287 ASSERT(hdr.inode == hdr_prv.inode);
288 ASSERT(hdr.pgoff == hdr_prv.pgoff);
289 ASSERT(hdr.seq != hdr_prv.seq);
291 pgcnt_t new_page, old_page;
295 * Sequence number comparison: since
296 * seq is 40 bits wide, it wraps once
297 * every 1.1E12 times.
298 * The memory will not live enough to
299 * see a wraparound, so we can use a simple
302 if (hdr.seq > hdr_prv.seq)
304 /* Current header is newer than the previuos one */
305 old_page = disk->page_array[array_pos];
307 old_fill = hdr_prv.fill;
311 /* Previous header is newer than the current one */
313 new_page = disk->page_array[array_pos];
318 disk->page_array[array_pos] = new_page;
320 disk->free_bytes += old_fill;
321 /* Shift all array one position to the left, overwriting duplicate page */
322 array_pos -= hdr.pgoff;
323 array_pos += filelen_table[hdr.inode];
324 movePages(disk, array_pos, -1);
325 /* Move back all indexes */
326 filelen_table[hdr.inode]--;
327 disk->free_page_start--;
329 /* Set old page as free */
330 ASSERT(disk->page_array[curr_free_page] == PAGE_UNSET_SENTINEL);
331 disk->page_array[curr_free_page++] = old_page;
337 /* Invalid page, keep as free */
338 ASSERT(disk->page_array[curr_free_page] == PAGE_UNSET_SENTINEL);
339 LOG_INFO("Page %d invalid, keeping as free\n", page);
340 disk->page_array[curr_free_page++] = page;
348 * Find the latest version of a page, starting from the
349 * page supplied by \a page_array.
350 * The pages are read from the disk until a different
351 * inode or page offset is found.
352 * The lastest version of the page is moved in the first
353 * position of \a page_array.
354 * \return the number of old versions of the page or PAGE_ERROR
355 * on disk read errors.
357 static pgcnt_t findLastVersion(struct BattFsSuper *disk, pgcnt_t *page_array)
359 pgcnt_t *array_start = page_array;
360 BattFsPageHeader hdr;
361 if (!battfs_readHeader(disk, *page_array++, &hdr))
364 /* Free space: early bailout */
365 if (hdr.fcs != computeFcs(&hdr))
369 * If the first page is valid,
370 * inode and pg_off in the array are taken
371 * as the current page markers.
373 inode_t curr_inode = hdr.inode;
374 pgoff_t curr_pgoff = hdr.pgoff;
376 /* Temps used to find the sequence number range */
377 seq_t minl = HALF_SEQ - 1;
379 seq_t minh = MAX_SEQ;
380 seq_t maxh = MAX_SEQ;
381 pgcnt_t lpos = 0, hpos = 0, dup_cnt = 0;
384 * Find min and max values for the two
385 * half of seq_num range.
386 * With this we can find seqnum wraparounds.
387 * seq_t is a type that has at least 1 bit more than
388 * pgaddr_t. So all version of a page blocks can be numbered using
389 * at most half numbers of a seq_t type.
390 * The sequence number algorithm increments by 1 the previous seq_num
391 * every time a page is rewritten. So the sequence is
392 * guaranteed to be countiguous.
393 * Only wrap arounds may happen, but due to half size sequence limitation,
394 * there are only 4 possible situations:
397 * |------lower half------|-------upper half-------|
399 * 1) |------minl*****maxl---|------------------------|
400 * 2) |------minl********maxl|minh******maxh----------|
401 * 3) |----------------------|----minh*******maxh-----|
402 * 4) |minl******maxl--------|------------minh****maxh|
405 * Situations 1 and 3 are easy to detect, while 2 and 4 require more care.
409 if (hdr.seq < HALF_SEQ)
411 minl = MIN(minl, hdr.seq);
420 minh = MIN(minh, hdr.seq);
428 if (!battfs_readHeader(disk, *page_array++, &hdr))
432 while (curr_inode == hdr.inode && curr_pgoff == hdr.pgoff && hdr.fcs == computeFcs(&hdr));
435 /* Return early if there is only one version of the current page */
439 /* Find the position in the array of the last version of the page */
440 pgcnt_t last_ver = hpos;
443 /* Valid interval found in lower half */
446 /* Valid interval also found in upper half */
447 if (maxl != minh - 1)
449 /* Interval starts in upper half and ends in lower */
451 ASSERT(maxh == MAX_SEQ);
458 * Upper interval is invalid.
464 /* Put last page version at array start position */
465 SWAP(array_start[0], array_start[last_ver]);
471 * Collect old pages, removing empty spaces from \a pg_array, for a maximum len of \a pg_len.
472 * Once the collect task is completed, copy \a old_cnt pages from \a old_pages at the
473 * end of free space in pg_array.
475 void collectOldPages(pgcnt_t *pg_array, pgcnt_t pg_len, pgcnt_t *old_pages, pgcnt_t old_cnt)
480 for (pgcnt_t curr_page = 0; curr_page < pg_len; curr_page++)
484 if (pg_array[curr_page] == PAGE_UNSET_SENTINEL)
488 pg_array[curr_page - gap] = pg_array[curr_page];
494 if (pg_array[curr_page] != PAGE_UNSET_SENTINEL)
495 pg_array[curr_page - gap] = pg_array[curr_page];
503 ASSERT(gap == old_cnt);
504 pg_array += pg_len - old_cnt;
506 memcpy(pg_array, old_pages, old_cnt * sizeof(pgcnt_t));
510 * This function scan the page array of \a disk looking for
511 * old versions of the same page.
513 * Only the last version is kept as valid, the old ones are inserted
514 * in the free blocks heap.
515 * \return true if ok, false on disk read errors.
516 * \note The whole disk is scanned once.
518 static bool dropOldPages(struct BattFsSuper *disk)
520 #define OLD_PAGE_BUFLEN 64
521 pgcnt_t old_pages[OLD_PAGE_BUFLEN];
524 pgcnt_t *curr_page = disk->page_array;
525 pgcnt_t *collect_start = disk->page_array;
526 pgcnt_t collect_len = disk->page_count;
531 dup_pages = findLastVersion(disk, curr_page);
532 if (dup_pages == PAGE_ERROR)
534 /* The first page is the last version */
538 if (old_cnt >= OLD_PAGE_BUFLEN)
540 collectOldPages(collect_start, collect_len, old_pages, old_cnt);
541 collect_len -= old_cnt;
542 disk->free_bytes += old_cnt * (disk->page_size - BATTFS_HEADER_LEN);
543 disk->free_page_start -= old_cnt;
544 curr_page -= old_cnt;
545 collect_start = curr_page;
549 old_pages[old_cnt++] = *curr_page;
550 *curr_page++ = PAGE_UNSET_SENTINEL;
553 while (curr_page < disk->page_array + disk->free_page_start);
555 collectOldPages(collect_start, collect_len, old_pages, old_cnt);
556 disk->free_bytes += old_cnt * (disk->page_size - BATTFS_HEADER_LEN);
557 disk->free_page_start -= old_cnt;
564 * Initialize and mount disk described by
566 * \return false on errors, true otherwise.
568 bool battfs_init(struct BattFsSuper *disk)
570 pgoff_t filelen_table[BATTFS_MAX_FILES];
575 /* Init disk device */
576 if (!disk->open(disk))
578 LOG_ERR("open error\n");
582 /* Disk open must set all of these */
587 ASSERT(disk->page_size);
588 ASSERT(disk->page_count);
589 ASSERT(disk->page_count < PAGE_UNSET_SENTINEL - 1);
590 ASSERT(disk->page_array);
592 memset(filelen_table, 0, BATTFS_MAX_FILES * sizeof(pgoff_t));
594 disk->free_bytes = 0;
595 disk->disk_size = (disk_size_t)(disk->page_size - BATTFS_HEADER_LEN) * disk->page_count;
597 /* Count pages per file */
598 if (!countDiskFilePages(disk, filelen_table))
600 LOG_ERR("error counting file pages\n");
604 /* Once here, we have filelen_table filled with file lengths */
606 /* Fill page array with sentinel */
607 for (pgcnt_t page = 0; page < disk->page_count; page++)
608 disk->page_array[page] = PAGE_UNSET_SENTINEL;
610 /* Fill page allocation array using filelen_table */
611 if (!fillPageArray(disk, filelen_table))
613 LOG_ERR("error filling page array\n");
617 /* Init list for opened files. */
618 LIST_INIT(&disk->file_opened_list);
624 * \return 0 if ok, EOF on errors.
626 static int battfs_flush(struct KFile *fd)
635 * \return 0 if ok, EOF on errors.
637 static int battfs_fileclose(struct KFile *fd)
639 BattFs *fdb = BATTFS_CAST(fd);
647 * Read from file \a fd \a size bytes in \a buf.
648 * \return The number of bytes read.
650 static size_t battfs_read(struct KFile *fd, void *_buf, size_t size)
652 BattFs *fdb = BATTFS_CAST(fd);
653 uint8_t *buf = (uint8_t *)_buf;
655 size_t total_read = 0;
657 pgaddr_t addr_offset;
660 size = MIN((kfile_off_t)size, fd->size - fd->seek_pos);
664 pg_offset = fd->seek_pos / (fdb->disk->page_size - BATTFS_HEADER_LEN);
665 addr_offset = fd->seek_pos % (fdb->disk->page_size - BATTFS_HEADER_LEN);
666 read_len = MIN(size, (size_t)(fdb->disk->page_size - BATTFS_HEADER_LEN - addr_offset));
669 if (fdb->disk->read(fdb->disk, fdb->start[pg_offset], addr_offset, buf, read_len) != read_len)
671 #warning TODO set error?
675 fd->seek_pos += read_len;
676 total_read += read_len;
684 * Search file \a inode in \a disk using a binary search.
685 * \return pointer to file start in disk->page_array
686 * if file exists, NULL otherwise.
688 static pgcnt_t *findFile(BattFsSuper *disk, inode_t inode)
690 BattFsPageHeader hdr;
691 pgcnt_t first = 0, page, last = disk->page_count -1;
694 while (first <= last)
696 page = (first + last) / 2;
698 if (!battfs_readHeader(disk, disk->page_array[page], &hdr))
701 fcs = computeFcs(&hdr);
702 if (hdr.fcs == fcs && hdr.inode == inode)
703 return (&disk->page_array[page]) - hdr.pgoff;
704 else if (hdr.fcs == fcs && hdr.inode < inode)
714 * \return true if file \a inode exists on \a disk, false otherwise.
716 bool battfs_fileExists(BattFsSuper *disk, inode_t inode)
718 return findFile(disk, inode) != NULL;
722 * Count size of file \a inode on \a disk, starting at pointer \a start
723 * in disk->page_array. Size is written in \a size.
724 * \return true if all s ok, false on disk read errors.
726 static bool countFileSize(BattFsSuper *disk, pgcnt_t *start, inode_t inode, file_size_t *size)
729 BattFsPageHeader hdr;
733 if (!battfs_readHeader(disk, *start++, &hdr))
735 if (hdr.fcs == computeFcs(&hdr) && hdr.inode == inode)
743 * Open file \a inode from \a disk in \a mode.
744 * File context is stored in \a fd.
745 * \return true if ok, false otherwise.
747 bool battfs_fileopen(BattFsSuper *disk, BattFs *fd, inode_t inode, filemode_t mode)
751 memset(fd, 0, sizeof(*fd));
753 /* Search file start point in disk page array */
754 fd->start = findFile(disk, inode);
755 if (fd->start == NULL)
757 if (!(mode & BATTFS_CREATE))
760 /* File does not exist, create it */
761 BattFsPageHeader hdr;
766 hdr.fcs = computeFcs(&hdr);
767 #warning TODO: get a free block and write on disk!
771 if (!countFileSize(disk, fd->start, inode, &fd->fd.size))
774 /* Reset seek position */
777 /* Insert file handle in list, ordered by inode, ascending. */
778 FOREACH_NODE(n, &disk->file_opened_list)
780 BattFs *file = containerof(n, BattFs, link);
781 if (file->inode >= inode)
784 INSERT_BEFORE(&fd->link, n);
791 fd->fd.close = battfs_fileclose;
792 fd->fd.flush = battfs_flush;
793 fd->fd.read = battfs_read;
794 fd->fd.reopen = kfile_genericReopen;
795 fd->fd.seek = kfile_genericSeek;
797 #warning TODO battfs_write, battfs_error, battfs_clearerr
799 fd->fd.write = battfs_write;
800 fd->fd.error = battfs_error;
801 fd->fd.clearerr = battfs_clearerr;
804 DB(fd->fd._type = KFT_BATTFS);
812 bool battfs_close(struct BattFsSuper *disk)
817 /* Close all open files */
818 FOREACH_NODE(n, &disk->file_opened_list)
820 BattFs *file = containerof(n, BattFs, link);
821 res += battfs_fileclose(&file->fd);
825 return disk->close(disk) && (res == 0);
828 bool battfs_writeTestBlock(struct BattFsSuper *disk, pgcnt_t page, inode_t inode, seq_t seq, fill_t fill, pgoff_t pgoff)
830 BattFsPageHeader hdr;
836 hdr.fcs = computeFcs(&hdr);
838 if (!battfs_writeHeader(disk, page, &hdr))
840 LOG_ERR("error writing hdr\n");