FreeCalypso > hg > freecalypso-sw
view gsm-fw/services/ffs/ffs.c @ 407:f2cc551e597f
fc-loadtool flash: beginning of Intel command set support
author | Michael Spacefalcon <msokolov@ivan.Harhan.ORG> |
---|---|
date | Tue, 17 Jun 2014 01:39:59 +0000 |
parents | 30a173257f4a |
children |
line wrap: on
line source
/****************************************************************************** * Flash File System (ffs) * Idea, design and coding by Mads Meisner-Jensen, mmj@ti.com * * ffs public API functions * * $Id: ffs.c 1.69.1.24.1.40 Thu, 08 Jan 2004 15:05:23 +0100 tsj $ * f ******************************************************************************/ #include "ffs.h" #if ((TARGET == 1) || (RIV_ENV== 1)) #include "task.h" #endif #if (TARGET == 0) #include <stdlib.h> #endif #include <string.h> #include <limits.h> #include "core.h" #include "ffstrace.h" /****************************************************************************** * ******************************************************************************/ extern struct fs_s fs; // defined in core.c // These dummy defines and struct are only use to simulate FFS on the // PC. The ones that is used in target are located in task.h #if (TARGET == 0) struct ffs_blocking_s {int x; }; #define FFS_BLOCKING_CALL_BEGIN() int result; \ struct ffs_blocking_s fb; #define FFS_BLOCKING_CALL_END() #endif /****************************************************************************** * Create, Read and Write ******************************************************************************/ req_id_t ffs_file_write_b(const char *pathname, void *src, int size, ffs_options_t option, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i, dir; char *name; effs_t error; int chunk_size, size_remaining, bytes_free; tw(tr(TR_FUNC, TrApi, "ffs_file_write('%s', 0x%x, %d, %d) ?\n", pathname, (int) src, size, option)); ttw(ttr(TTrApi, "ffs_file_write('%s', 0x%x, %d, %d) ?" NL, pathname, (int) src, size, option)); // TASKBEGIN effs_t FILE_WRITE(path=pathname, src=src, size=size, value16=option) iref_t i, dir; char *name; effs_t error; int chunk_size, size_remaining, bytes_free; if (fs.initerror) return fs.initerror; if (size < 0) return EFFS_INVALID; ffs_query(Q_BYTES_FREE, &bytes_free); if (bytes_free < size) return EFFS_NOSPACE; chunk_size = (size > fs.chunk_size_max ? fs.chunk_size_max : size); if ((i = object_lookup(pathname, &name, &dir)) < 0) { // Object not found, continue like fcreate() if (i != EFFS_NOTFOUND) return i; if (!is_open_option(option, FFS_O_CREATE)) return EFFS_NOTFOUND; journal_begin(0); if ((dir = object_create(name, src, chunk_size, -dir)) < 0) return dir; journal_end(OT_FILE); } else { // Object found, continue like fupdate() if (is_open_option(option, (FFS_O_CREATE)) && is_open_option(option, (FFS_O_EXCL))) return EFFS_EXISTS; if (get_fdi(i) >= 0) return EFFS_LOCKED; // Even though the ffs architecture allows to have data in // directory objects, we don't want to complicate matters, so we // return an error if (is_object(inode_addr(i), OT_DIR) && !(fs.flags & FS_DIR_DATA)) { return EFFS_NOTAFILE; } if ((i = is_readonly(i, pathname)) < 0) return i; // Save the segment (if any) in the global variable because this // global variable will be updated if the inode is going to be // relocated if an inode_reclaim() is triggeret by the object_create() fs.i_backup = segment_next(i); journal_begin(i); if ((dir = object_create(name, src, chunk_size, -dir)) < 0) return dir; // Do not link child - we are replacing the complete file! fs.link_child = 0; journal_end(0); // If any other segments exist then remove them FIXME: If we get a // power failure here then the remaining segments wil not be removed // before inode_reclaim() has been executed if (fs.i_backup > 0) if ((error = object_remove(fs.i_backup)) < 0) return error; } // Save dir in fs.i_backup because this will be updated if some of the // chunks below trigger a inode reclaim! fs.i_backup = dir; size_remaining = size - chunk_size; while (size_remaining > 0) { chunk_size = (size_remaining > fs.chunk_size_max ? fs.chunk_size_max : size_remaining); journal_begin(0); if ((i = segment_create((char*) src + size - size_remaining, chunk_size, fs.i_backup)) < 0) return i; journal_end(OT_SEGMENT); size_remaining -= chunk_size; } tw(tr_bstat()); return EFFS_OK; // TASKEND } // Note: ffs_fcreate() is deprecated and should not be used. Use // ffs_file_write(..., FFS_O_CREATE | FFS_O_EXCL) instead. effs_t ffs_fcreate(const char *pathname, void *src, int size) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_file_write_b(pathname, src, size, FFS_O_CREATE | FFS_O_EXCL, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_fcreate_nb(const char *pathname, void *src, int size, T_RV_RETURN *cp) { return ffs_file_write_b(pathname, src, size, FFS_O_CREATE | FFS_O_EXCL, cp, 0); } // Note: ffs_fupdate() is deprecated and should not be used. Use // ffs_file_write(...,FFS_O_TRUNC) instead. effs_t ffs_fupdate(const char *pathname, void *src, int size) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_file_write_b(pathname, src, size, FFS_O_TRUNC, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_fupdate_nb(const char *pathname, void *src, int size, T_RV_RETURN *cp) { return ffs_file_write_b(pathname, src, size, FFS_O_TRUNC, cp, 0); } // Note: ffs_fwrite() is deprecated and should not be used. Use // ffs_file_write(...,FFS_O_CREATE | FFS_O_TRUNC) instead. effs_t ffs_fwrite(const char *pathname, void *src, int size) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_file_write_b(pathname, src, size, FFS_O_CREATE | FFS_O_TRUNC, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_fwrite_nb(const char *pathname, void *src, int size, T_RV_RETURN *cp) { return ffs_file_write_b(pathname, src, size, FFS_O_CREATE | FFS_O_TRUNC, cp, 0); } effs_t ffs_file_write(const char *pathname, void *src, int size, ffs_options_t option) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_file_write_b(pathname, src, size, option, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_file_write_nb(const char *pathname, void *src, int size, ffs_options_t option, T_RV_RETURN *cp) { return ffs_file_write_b(pathname, src, size, option, cp, 0); } // Note important: ffs_fread() is deprecated and should not be used. Use // ffs_file_read() instead. int ffs_fread(const char *name, void *addr, int size) { return ffs_file_read(name, addr, size); } int ffs_file_read(const char *name, void *addr, int size) { int error; tw(tr(TR_BEGIN, TrApi, "file_read('%s', 0x%x, %d) {\n", name, (int) addr, size)); if ((error = ffs_begin()) == EFFS_OK) { error = file_read(name, addr, size); } tw(tr(TR_END, TrApi, "} %d\n", error)); return ffs_end(error); // number of bytes read } /****************************************************************************** * Stat, Symlink, Remove and Rename ******************************************************************************/ effs_t ffs_stat(const char *name, struct stat_s *stat) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_stat('%s', ?) ?\n", name)); ttw(ttr(TTrApi, "ffs_stat('%s', ?) ?" NL, name)); if (name == NULL) return EFFS_BADNAME; if ((i = ffs_begin()) == EFFS_OK) { if ((i = object_stat(name, (struct xstat_s*) stat, 0, 0, 0)) > 0) i = EFFS_OK; } return ffs_end(i); } effs_t ffs_lstat(const char *name, struct stat_s *stat) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_lstat('%s', ?) ?\n", name)); ttw(ttr(TTrApi, "ffs_lstat('%s', ?) ?" NL, name)); if ((i = ffs_begin()) == EFFS_OK) { if ((i = object_stat(name, (struct xstat_s*)stat, 1, 0, 0)) > 0) i = EFFS_OK; } return ffs_end(i); } effs_t ffs_xlstat(const char *name, struct xstat_s *stat) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_xlstat('%s', ?) ?\n", name)); ttw(ttr(TTrApi, "ffs_xlstat('%s', ?) ?" NL, name)); if ((i = ffs_begin()) == EFFS_OK) { if ((i = object_stat(name, stat, 1, 0, 1)) > 0) i = EFFS_OK; } return ffs_end(i); } effs_t ffs_fstat(fd_t fdi, struct stat_s *stat) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_fstat('%d', ?) ?\n", fdi)); ttw(ttr(TTrApi, "ffs_fstat('%d', ?) ?" NL, fdi)); if ((i = ffs_begin()) == EFFS_OK) { if ((i = object_stat( 0, (struct xstat_s*) stat, 0, fdi, 0)) > 0) i = EFFS_OK; } return ffs_end(i); } req_id_t ffs_symlink_b(const char *pathname, const char *src, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i, dir; char *name; int size; tw(tr(TR_FUNC, TrApi, "ffs_symlink('%s', '%s') ?\n", pathname, src)); ttw(ttr(TTrApi, "ffs_symlink('%s', '%s') ?" NL, pathname, src)); // TASKBEGIN effs_t SYMLINK(path=pathname, src=src) iref_t i, dir; int size; char *name; if (fs.initerror) return fs.initerror; if (src == NULL) return EFFS_BADNAME; i = object_lookup(pathname, &name, &dir); if (i > 0) return EFFS_EXISTS; if (i != EFFS_NOTFOUND) return i; size = ffs_strlen(src) + 1; // include null-terminator journal_begin(0); if ((i = object_create(name, src, size, -dir)) < 0) return i; journal_end(OT_LINK); tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_symlink(const char *pathname, const char *actualpath) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_symlink_b(pathname, actualpath, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_symlink_nb(const char *pathname, const char *src, T_RV_RETURN *cp) { return ffs_symlink_b(pathname, src, cp, 0); } int ffs_readlink(const char *name, char *addr, int size) { int error; tw(tr(TR_FUNC, TrApi, "ffs_readlink('%s')\n", name)); if ((error = ffs_begin()) == EFFS_OK) { error = object_read(name, addr, size, 1); } return ffs_end(error); } req_id_t ffs_remove_b(const char *pathname, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_remove('%s')\n", pathname)); ttw(ttr(TTrApi, "ffs_remove('%s') ?" NL, pathname)); // TASKBEGIN effs_t REMOVE(path=pathname) iref_t i; if (fs.initerror) return fs.initerror; if ((i = object_lookup_once(pathname, 0, 0)) < 0) return i; if (get_fdi(i) >= 0) return EFFS_LOCKED; if ((i = is_readonly(i, pathname)) < 0) return i; if ((i = object_remove(i)) < 0) return i; tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_remove(const char *pathname) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_remove_b(pathname, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_remove_nb(const char *pathname, T_RV_RETURN *cp) { return ffs_remove_b(pathname, cp, 0); } req_id_t ffs_fcontrol_b(const char *pathname, int8 action, int param, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i; tw(tr(TR_FUNC, TrApi, "ffs_fcontrol('%s', %d, 0x%x) ?\n", pathname, action, param)); ttw(ttr(TTrApi, "ffs_fcontrol('%s', %d, 0x%x) ?" NL, pathname, action, param)); // TASKBEGIN effs_t FCONTROL(path=pathname, value16=action, size=param) iref_t i; if (fs.initerror) return fs.initerror; if (pathname == NULL) return EFFS_BADNAME; if ((i = ffs_strcmp(pathname, "/dev/ffs")) != 0) { if ((i = object_lookup_once(pathname, 0, 0)) < 0) return i; if ((i = is_readonly(i, pathname)) < 0) return i; } if ((i = object_control(i, action, param)) < 0) return i; tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_fcontrol(const char *pathname, int8 action, int param) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_fcontrol_b(pathname, action, param, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_fcontrol_nb(const char *pathname, int8 action, int param, T_RV_RETURN *cp) { return ffs_fcontrol_b(pathname, action, param, cp, 0); } req_id_t ffs_rename_b(const char *pathname, const char *newname, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i, oldi, dir; char *name; struct inode_s *ip; tw(tr(TR_FUNC, TrApi, "ffs_rename('%s', '%s') ?\n", pathname, newname)); ttw(ttr(TTrApi, "ffs_rename('%s', '%s') ?" NL, pathname, newname)); // TASKBEGIN effs_t RENAME(path=pathname, src=newname) iref_t i, oldi, dir; char *name; struct inode_s *ip; if (fs.initerror) return fs.initerror; // pathname MUST exist, not be open and MUST be writable if ((oldi = object_lookup_once(pathname, 0, 0)) < 0) return oldi; if ((oldi = is_readonly(oldi, pathname)) < 0) return oldi; if (get_fdi(oldi) >= 0) return EFFS_LOCKED; journal_begin(oldi); if ((i = object_lookup_once(newname, &name, &dir)) < 0) { if (i != EFFS_NOTFOUND) return i; } else { // newname obj exist ip = inode_addr(oldi); if (is_object(ip, OT_FILE)) { // is old obj a file? if ((i = is_readonly(i, newname)) < 0) return i; ip = inode_addr(i); if (!is_object(ip, OT_FILE)) // newname MUST be a file return EFFS_NOTAFILE; fs.journal.repli = i; } else return EFFS_EXISTS; } if ((i = object_rename(oldi, name, -dir)) < 0) return i; journal_end(0); tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_rename(const char *pathname, const char *newname) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_rename_b(pathname, newname, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_rename_nb(const char *pathname, const char *newname, T_RV_RETURN *cp) { return ffs_rename_b(pathname, newname, cp, 0); } /****************************************************************************** * Directory Operations ******************************************************************************/ // All directory operations are more or less similar to unix // semantics. req_id_t ffs_mkdir_b(const char *pathname, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i, dir; char *name; tw(tr(TR_FUNC, TrApi, "ffs_mkdir('%s')\n", pathname)); ttw(ttr(TTrApi, "ffs_mkdir('%s') ?" NL, pathname)); // TASKBEGIN effs_t MKDIR(path=pathname) iref_t i, dir; char *name; if (fs.initerror) return fs.initerror; i = object_lookup(pathname, &name, &dir); if (i > 0) return EFFS_EXISTS; if (i != EFFS_NOTFOUND) return i; journal_begin(0); if ((i = object_create(name, 0, 0, -dir)) < 0) return i; journal_end(OT_DIR); tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_mkdir(const char *pathname) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_mkdir_b(pathname, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_mkdir_nb(const char *pathname, T_RV_RETURN *cp) { return ffs_mkdir_b(pathname, cp, 0); } int ffs_opendir(const char *name, struct dir_s *dir) { int i; tw(tr(TR_FUNC, TrApi, "ffs_opendir('%s', ?)\n", name)); ttw(ttr(TTrApi, "ffs_opendir('%s', ?) ?" NL, name)); if (dir == NULL) return EFFS_INVALID; if ((i = ffs_begin()) == EFFS_OK) { if ((i = dir_open(name)) >= 0) { dir->this = i; dir->index = i; // Now count the number of entries in the directory dir_traverse(-i, (iref_t *) &i); } } return ffs_end(i); } int ffs_readdir(struct dir_s *dir, char *name, int size) { iref_t i; tw(tr(TR_BEGIN, TrApi, "ffs_readdir(?, ?, ?) {\n")); ttw(ttr(TTrApi, "ffs_readdir(?, ?, ?) ?" NL)); if (dir == NULL || name == NULL || size < 0) { tw(tr(TR_END, TrApi, "} %d\n", EFFS_INVALID)); return EFFS_INVALID; } if ((i = ffs_begin()) == EFFS_OK) { if ((i = dir_next(dir->this, dir->index, name, size))) dir->index = i; } tw(tr(TR_END, TrApi, "} ('%s') %d\n", name, i)); return ffs_end(i); } /****************************************************************************** * Preformat and Format ******************************************************************************/ // Note that we do NOT call ffs_begin() because it will just return // EFFS_NOFORMAT! req_id_t ffs_format_b(const char *name, uint16 magic, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { effs_t i; tw(tr(TR_BEGIN, TrApi, "ffs_format('%s', 0x%x) {\n", name, magic)); ttw(ttr(TTrApi, "ffs_format('%s', 0x%x) ?" NL, name, magic)); // TASKBEGIN effs_t FORMAT(path=name, size=magic) iref_t i; if (magic != 0x2BAD) { tw(tr(TR_END, TrApi, "} %d\n", EFFS_INVALID)); return EFFS_INVALID; } if (name == NULL) { name = "/ffs-5.54"; } if (*name != '/') { tw(tr(TR_END, TrApi, "} %d\n", EFFS_BADNAME)); return EFFS_BADNAME; } if ((i = is_formattable(1)) < 0) { tw(tr(TR_END, TrApi, "} %d\n", i)); return i; } if ((i = fs_format(name)) < 0) return i; tw(tr(TR_END, TrApi, "} %d\n", i)); tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_format(const char *name, uint16 magic) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_format_b(name, magic, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_format_nb(const char *name, uint16 magic, T_RV_RETURN *cp) { return ffs_format_b(name, magic, cp, 0); } req_id_t ffs_preformat_b(uint16 magic, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { effs_t i; tw(tr(TR_BEGIN, TrApi, "ffs_preformat(0x%x) {\n", magic)); ttw(ttr(TTrApi, "ffs_preformat(0x%x) ?" NL, magic)); // TASKBEGIN effs_t PREFORMAT(path="/", size=magic) effs_t i; if (magic != 0xDEAD) { tw(tr(TR_END, TrApi, "} %d\n", EFFS_INVALID)); return EFFS_INVALID; } if (!ffs_is_modifiable("")) { tw(tr(TR_END, TrApi, "} %d\n", EFFS_ACCESS)); return EFFS_ACCESS; } if ((i = is_formattable(0)) < 0) { tw(tr(TR_END, TrApi, "} %d\n", i)); return i; } if ((i = fs_preformat()) < 0) return i; tw(tr(TR_END, TrApi, "} %d\n", i)); tw(tr_bstat()); return EFFS_OK; // TASKEND } effs_t ffs_preformat(uint16 magic) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_preformat_b(magic, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_preformat_nb(uint16 magic, T_RV_RETURN *cp) { return ffs_preformat_b(magic, cp, 0); } /****************************************************************************** * Open, Read, Write, Close ******************************************************************************/ req_id_t ffs_open_b(const char *pathname, ffs_options_t option, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { iref_t i, dir, dummy; char *name; fd_t other_fdi, fdi = 0; int error; struct inode_s *ip; tw(tr(TR_FUNC, TrApi, "ffs_open('%s', 0x%x) ?\n", pathname, option)); ttw(ttr(TTrApi, "ffs_open('%s', 0x%x) ?" NL, pathname, option)); // TASKBEGIN fd_t OPEN(path=pathname, value16=option) iref_t i, dir, dummy; char *name; fd_t other_fdi, fdi = 0; int error; struct inode_s *ip; if (fs.initerror) return fs.initerror; // Minimum one of the flags RD or WR must be specifyed if (!is_open_option(option, FFS_O_RDONLY) && !is_open_option(option, FFS_O_WRONLY)) return EFFS_INVALID; // RDONLY must not be combined with any other options if not together // with WR! if (is_open_option(option, FFS_O_RDONLY) && !is_open_option(option, FFS_O_WRONLY)) if (!(option == FFS_O_RDONLY)) return EFFS_INVALID; for (fdi = 0; fdi < fs.fd_max; fdi++) { // Find free fd if (fs.fd[fdi].options == 0) { break; } } if (fdi >= fs.fd_max) return EFFS_NUMFD; // Too many open files in system i = object_lookup(pathname, &name, &dir); if (i < 0 && i != EFFS_NOTFOUND) return i; // Open one file several times in RD is okay but only one time in WR if (i != EFFS_NOTFOUND && (other_fdi = get_fdi(i)) >= 0) { if (is_open_option(fs.fd[other_fdi].options, FFS_O_WRONLY) || is_open_option(option, FFS_O_WRONLY)) return EFFS_LOCKED; } // Init default values fs.fd[fdi].fp = fs.fd[fdi].size = fs.fd[fdi].wfp = fs.fd[fdi].dirty = 0; if (i == EFFS_NOTFOUND) { if (is_open_option(option, (FFS_O_CREATE | FFS_O_WRONLY))) { if ((error = is_filename(name)) < 0) return error; // Create segmenthead journal_begin(0); if ((i = object_create(name, 0, 0, -dir)) < 0) return i; journal_end(OT_FILE); tw(tr_bstat()); fs.fd[fdi].seghead = i; } else return EFFS_NOTFOUND; } else { if (is_open_option(option, FFS_O_WRONLY)) { if (is_open_option(option, (FFS_O_CREATE | FFS_O_EXCL))) return EFFS_EXISTS; if ((i = is_readonly(i, pathname)) < 0) return i; } ip = inode_addr(i); if (is_object(ip, OT_DIR)) return EFFS_NOTAFILE; if (is_open_option(option, FFS_O_TRUNC)) { // Save the segment (if any) in the global variable because this // global variable will be updated if the inode is relocated by // an inode_reclaim() triggeret by object_create() fs.i_backup = segment_next(i); // Replace old seghead with a new and remove all old segments journal_begin(i); if ((i = object_create(name, 0, 0, -dir)) < 0) return i; // Do not link child fs.link_child = 0; journal_end(0); // If any further segments exist then remove them now if (fs.i_backup > 0) if ((error = object_remove(fs.i_backup)) < 0) return error; tw(tr_bstat()); } else { // Get total size of the file. fs.fd[fdi].size = segfile_seek(i, INT_MAX, &dummy, 0); } if (is_open_option(option, FFS_O_APPEND)) { fs.fd[fdi].fp = fs.fd[fdi].size; } } if (is_open_option(option, FFS_O_WRONLY)) { #if (TARGET == 1) if ((fs.fd[fdi].buf = (char *) target_malloc(fs.fd_buf_size)) == 0) return EFFS_MEMORY; #else if ((fs.fd[fdi].buf = malloc(fs.fd_buf_size)) == 0) return EFFS_MEMORY; #endif } // Save data in file descriptor fs.fd[fdi].seghead = i; fs.fd[fdi].options = option; return fdi + FFS_FD_OFFSET; // TASKEND } fd_t ffs_open(const char *pathname, ffs_options_t option) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_open_b(pathname, option, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_open_nb(const char *pathname, ffs_options_t option, T_RV_RETURN *cp) { return ffs_open_b(pathname, option, cp, 0); } req_id_t ffs_close_b(fd_t fdi, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { int error; tw(tr(TR_FUNC, TrApi, "ffs_close(%d) ?\n", fdi)); ttw(ttr(TTrApi, "ffs_close(%d) ?" NL, fdi)); // TASKBEGIN effs_t CLOSE(fdi=fdi) iref_t i; int error; if (fs.initerror) return fs.initerror; fdi -= FFS_FD_OFFSET; if (!is_fd_valid(fdi)) return EFFS_BADFD; if (is_open_option(fs.fd[fdi].options, FFS_O_WRONLY )) { if ((error = datasync(fdi)) < 0) return error; #if (TARGET == 1) target_free(fs.fd[fdi].buf); #else free(fs.fd[fdi].buf); #endif } // Clear all data in file descriptor fs.fd[fdi].seghead = 0; fs.fd[fdi].options = fs.fd[fdi].fp = 0; return EFFS_OK; // TASKEND } effs_t ffs_close(fd_t fdi) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_close_b(fdi, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_close_nb(fd_t fdi, T_RV_RETURN *cp) { return ffs_close_b( fdi, cp, 0); } req_id_t ffs_write_b(fd_t fdi, void *src, int amount, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { effs_t error; iref_t i; int size_remaining, fp_offset; int size, size_done; offset_t chunk_offset; tw(tr(TR_BEGIN, TrApi, "ffs_write_b(%d, 0x%x, %d) ?{\n", fdi, src, amount)); ttw(ttr(TTrApi, "ffs_write_b(%d, 0x%x, %d) ?" NL, fdi, src, amount)); // TASKBEGIN int WRITE(fdi=fdi, src=src, size=amount) effs_t error; iref_t i; int size_remaining, fp_offset; int size, size_done; offset_t chunk_offset; if (fs.initerror) return fs.initerror; if (amount < 0 || src == NULL) return EFFS_INVALID; fdi -= FFS_FD_OFFSET; if (!is_fd_valid(fdi)) return EFFS_BADFD; if (!is_open_option(fs.fd[fdi].options, FFS_O_WRONLY )) return EFFS_INVALID; // not opened with write flag // If FFS_O_APPEEND is specified move fp to eof if (is_open_option(fs.fd[fdi].options, FFS_O_APPEND )) fs.fd[fdi].fp = fs.fd[fdi].size; // If fp has been moved outside the write buf (by a read) then flush the // write buffer. if (fs.fd[fdi].fp >= (fs.fd[fdi].wfp + fs.chunk_size_max)) { if ((error = datasync(fdi)) < 0) return error; } size_done = 0; size_remaining = amount; do { if (!fs.fd[fdi].dirty ) { // Buffer is not dirty so find the chunk that fp points to. segfile_seek(fs.fd[fdi].seghead, fs.fd[fdi].fp, &i, &chunk_offset); if ((fs.fd[fdi].size == fs.fd[fdi].fp && chunk_offset == fs.chunk_size_max) || fs.fd[fdi].size == 0 ) { // End of file and last chunk is full or empty seghead. fs.fd[fdi].wfp = fs.fd[fdi].size; fs.fd[fdi].wch = 0; // Create new chunk (not update). } else { // Work on this chunk and update it later by datasyns segment_read(i, fs.fd[fdi].buf, fs.fd_buf_size, 0); fs.fd[fdi].wfp = fs.fd[fdi].fp - chunk_offset; fs.fd[fdi].wch = i; } } fs.fd[fdi].dirty = 1; fp_offset = fs.fd[fdi].fp - fs.fd[fdi].wfp; // Fill the buffer to max or just add the rest size = fs.chunk_size_max - fp_offset; if (size_remaining <= fs.chunk_size_max - fp_offset) size = size_remaining; tw(tr(TR_FUNC, TrApi, "Copy data to buffer (size: %d)\n", size)); memcpy(fs.fd[fdi].buf + fp_offset, (uint8*)src + size_done, size); fs.fd[fdi].fp += size; if (fs.fd[fdi].fp > fs.fd[fdi].size) fs.fd[fdi].size = fs.fd[fdi].fp; size_done += size; // FIXME: remove size_done or size_remaining size_remaining -= size; // If wrbuf is full (size = chunk_size_max) so create a chunk. if (fs.fd[fdi].fp >= (fs.fd[fdi].wfp + fs.chunk_size_max)) { if ((error = datasync(fdi)) < 0) return error; } } while(size_remaining > 0); tw(tr(TR_END, TrApi, "} %d\n", amount)); return amount; // TASKEND } int ffs_write(fd_t fdi, void *src, int amount) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_write_b(fdi, src, amount, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_write_nb(fd_t fdi, void *src, int amount, T_RV_RETURN *cp) { tw(tr(TR_FUNC, TrApi, "ffs_write_nb(%d, 0x%x, %d) ?\n", fdi, src, amount)); return ffs_write_b(fdi, src, amount, cp, 0); } int ffs_read(fd_t fdi, void *src, int size) { int error; tw(tr(TR_BEGIN, TrApi, "ffs_read(%d, 0x%x, %d) {\n", fdi, src, size)); ttw(ttr(TTrApi, "ffs_read(%d, 0x%x, %d) ?" NL, fdi, src, size)); if ((error = ffs_begin()) == EFFS_OK) { error = stream_read(fdi - FFS_FD_OFFSET, src, size); } tw(tr(TR_END, TrApi, "} %d\n", error)); return ffs_end(error); // number of bytes read } // The seek function will not allow the file offset to be set beyond the end // of the existing data in the file or the final offset to be negative. req_id_t ffs_seek_b(fd_t fdi, int offset, int whence, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { effs_t error; int fp_new; tw(tr(TR_FUNC, TrApi, "ffs_seek(%d, %d, %d) ?\n", fdi, offset, whence)); ttw(ttr(TTrApi, "ffs_seek(%d, %d, %d) ?" NL, fdi, offset, whence)); // TASKBEGIN int SEEK(fdi=fdi, size=offset, value16=whence) effs_t error; iref_t i; int fp_new, foffset; if (fs.initerror) return fs.initerror; fdi -= FFS_FD_OFFSET; if (!is_fd_valid(fdi)) return EFFS_BADFD; switch(whence) { case FFS_SEEK_SET: if (offset < 0 || offset > fs.fd[fdi].size) return EFFS_INVALID; fp_new = offset; break; case FFS_SEEK_CUR: if (fs.fd[fdi].fp + offset < 0 || fs.fd[fdi].fp + offset > fs.fd[fdi].size) return EFFS_INVALID; fp_new = fs.fd[fdi].fp + offset; break; case FFS_SEEK_END: if (offset > 0 || fs.fd[fdi].size < -offset) return EFFS_INVALID; fp_new = (offset + fs.fd[fdi].size); break; default: return EFFS_INVALID; } if (!is_offset_in_buf(fp_new, fdi)) if ((error = datasync(fdi)) < 0) return error; return fs.fd[fdi].fp = fp_new; // TASKEND } int ffs_seek(fd_t fdi, int offset, int whence) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_seek_b(fdi, offset, whence, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_seek_nb(fd_t fdi, int offset, int whence, T_RV_RETURN *cp) { return ffs_seek_b(fdi, offset, whence, cp, 0); } req_id_t ffs_truncate_b(const char *path, offset_t length, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { tw(tr(TR_FUNC, TrApi, "ffs_truncate('%s', %d) \n", path, length)); ttw(ttr(TTrApi, "ffs_ftruncate('%s', %d) ?" NL, path, length)); // TASKBEGIN effs_t TRUNC(path=path, size=length) iref_t i; if (fs.initerror) return fs.initerror; if (path == NULL) return EFFS_BADNAME; return object_truncate(path, -1, length); // TASKEND } effs_t ffs_truncate(const char *path, offset_t length) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_truncate_b(path, length, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_truncate_nb(const char *path, offset_t length, T_RV_RETURN *cp) { return ffs_truncate_b(path, length, cp, 0); } req_id_t ffs_ftruncate_b(fd_t fdi, offset_t length, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { tw(tr(TR_FUNC, TrApi, "ffs_ftruncate(%d, %d) \n", fdi, length)); ttw(ttr(TTrApi, "ffs_ftruncate(%d, %d) ?" NL, fdi, length)); // TASKBEGIN effs_t FTRUNC(fdi=fdi, size=length) iref_t i; if (fs.initerror) return fs.initerror; return object_truncate(0, fdi - FFS_FD_OFFSET, length); // TASKEND } effs_t ffs_ftruncate(fd_t fdi, offset_t length) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_ftruncate_b(fdi, length, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_ftruncate_nb(fd_t fdi, offset_t length, T_RV_RETURN *cp) { return ffs_ftruncate_b(fdi, length, cp, 0); } req_id_t ffs_fdatasync_b(fd_t fdi, T_RV_RETURN *cp, struct ffs_blocking_s *fb) { tw(tr(TR_FUNC, TrApi, "ffs_fdatasync(%d) \n", fdi)); ttw(ttr(TTrApi, "ffs_fdatasync(%d) ?" NL, fdi)); // TASKBEGIN effs_t FDATASYNC(fdi=fdi) effs_t error; if (fs.initerror) return fs.initerror; return datasync(fdi - FFS_FD_OFFSET); // TASKEND } effs_t ffs_fdatasync(fd_t fdi) { FFS_BLOCKING_CALL_BEGIN(); result = ffs_fdatasync_b(fdi, 0, &fb); FFS_BLOCKING_CALL_END(); return result; } req_id_t ffs_fdatasync_nb(fd_t fdi, T_RV_RETURN *cp) { return ffs_fdatasync_b(fdi, cp, 0); }