/* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright 2007 Sun Microsystems, Inc. All rights reserved. * Use is subject to license terms. */ #pragma ident "%Z%%M% %I% %E% SMI" /* * This file contains I/O related functions. */ #include "global.h" #include #include #include #include #include #include #include #include #include #include "startup.h" #include "misc.h" #include "menu_partition.h" #include "param.h" #include "menu.h" extern int data_lineno; extern char *space2str(); extern long strtol(); /* * This variable is used to determine whether a token is present in the pipe * already. */ static char token_present = 0; /* * This variable always gives us access to the most recent token type */ int last_token_type = 0; #ifdef __STDC__ /* * Prototypes for ANSI C compilers */ static int sup_get_token(char *); static void pushchar(int c); static int checkeof(void); static void flushline(void); static int strcnt(char *s1, char *s2); static int getbn(char *str, daddr_t *iptr); static void print_input_choices(int type, u_ioparam_t *param); static int slist_widest_str(slist_t *slist); static void ljust_print(char *str, int width); static int sup_inputchar(void); static void sup_pushchar(int c); static int geti64(char *str, uint64_t *iptr, uint64_t *wild); #else /* __STDC__ */ /* * Prototypes for non-ANSI C compilers */ static int sup_get_token(); static void pushchar(int c); static int checkeof(void); static void flushline(void); static int strcnt(char *s1, char *s2); static int getbn(char *str, daddr_t *iptr); static void print_input_choices(int type, u_ioparam_t *param); static int slist_widest_str(slist_t *slist); static void ljust_print(char *str, int width); static int sup_inputchar(void); static void sup_pushchar(int c); static int geti64(char *str, uint64_t *iptr, uint64_t *wild); #endif /* __STDC__ */ /* * This routine pushes the given character back onto the input stream. */ static void pushchar(c) int c; { (void) ungetc(c, stdin); } /* * This routine checks the input stream for an eof condition. */ static int checkeof() { return (feof(stdin)); } /* * This routine gets the next token off the input stream. A token is * basically any consecutive non-white characters. */ char * gettoken(inbuf) char *inbuf; { char *ptr = inbuf; int c, quoted = 0; retoke: /* * Remove any leading white-space. */ while ((isspace(c = getchar())) && (c != '\n')) ; /* * If we are at the beginning of a line and hit the comment character, * flush the line and start again. */ if (!token_present && c == COMMENT_CHAR) { token_present = 1; flushline(); goto retoke; } /* * Loop on each character until we hit unquoted white-space. */ while (!isspace(c) || quoted && (c != '\n')) { /* * If we hit eof, get out. */ if (checkeof()) return (NULL); /* * If we hit a double quote, change the state of quotedness. */ if (c == '"') quoted = !quoted; /* * If there's room in the buffer, add the character to the end. */ else if (ptr - inbuf < TOKEN_SIZE) *ptr++ = (char)c; /* * Get the next character. */ c = getchar(); } /* * Null terminate the token. */ *ptr = '\0'; /* * Peel off white-space still in the pipe. */ while (isspace(c) && (c != '\n')) c = getchar(); /* * If we hit another token, push it back and set state. */ if (c != '\n') { pushchar(c); token_present = 1; } else token_present = 0; /* * Return the token. */ return (inbuf); } /* * This routine removes the leading and trailing spaces from a token. */ void clean_token(cleantoken, token) char *cleantoken, *token; { char *ptr; /* * Strip off leading white-space. */ for (ptr = token; isspace(*ptr); ptr++) ; /* * Copy it into the clean buffer. */ (void) strcpy(cleantoken, ptr); /* * Strip off trailing white-space. */ for (ptr = cleantoken + strlen(cleantoken) - 1; isspace(*ptr) && (ptr >= cleantoken); ptr--) { *ptr = '\0'; } } /* * This routine checks if a token is already present on the input line */ int istokenpresent() { return (token_present); } /* * This routine flushes the rest of an input line if there is known * to be data in it. The flush has to be qualified because the newline * may have already been swallowed by the last gettoken. */ static void flushline() { if (token_present) { /* * Flush the pipe to eol or eof. */ while ((getchar() != '\n') && !checkeof()) ; /* * Mark the pipe empty. */ token_present = 0; } } /* * This routine returns the number of characters that are identical * between s1 and s2, stopping as soon as a mismatch is found. */ static int strcnt(s1, s2) char *s1, *s2; { int i = 0; while ((*s1 != '\0') && (*s1++ == *s2++)) i++; return (i); } /* * This routine converts the given token into an integer. The token * must convert cleanly into an integer with no unknown characters. * If the token is the wildcard string, and the wildcard parameter * is present, the wildcard value will be returned. */ int geti(str, iptr, wild) char *str; int *iptr, *wild; { char *str2; /* * If there's a wildcard value and the string is wild, return the * wildcard value. */ if (wild != NULL && strcmp(str, WILD_STRING) == 0) *iptr = *wild; else { /* * Conver the string to an integer. */ *iptr = (int)strtol(str, &str2, 0); /* * If any characters didn't convert, it's an error. */ if (*str2 != '\0') { err_print("`%s' is not an integer.\n", str); return (-1); } } return (0); } /* * This routine converts the given token into a long long. The token * must convert cleanly into a 64-bit integer with no unknown characters. * If the token is the wildcard string, and the wildcard parameter * is present, the wildcard value will be returned. */ static int geti64(str, iptr, wild) char *str; uint64_t *iptr, *wild; { char *str2; /* * If there's a wildcard value and the string is wild, return the * wildcard value. */ if ((wild != NULL) && (strcmp(str, WILD_STRING)) == 0) { *iptr = *wild; } else { /* * Conver the string to an integer. */ *iptr = (uint64_t)strtoll(str, &str2, 0); /* * If any characters didn't convert, it's an error. */ if (*str2 != '\0') { err_print("`%s' is not an integer.\n", str); return (-1); } } return (0); } /* * This routine converts the given string into a block number on the * current disk. The format of a block number is either a self-based * number, or a series of self-based numbers separated by slashes. * Any number preceeding the first slash is considered a cylinder value. * Any number succeeding the first slash but preceeding the second is * considered a head value. Any number succeeding the second slash is * considered a sector value. Any of these numbers can be wildcarded * to the highest possible legal value. */ static int getbn(str, iptr) char *str; daddr_t *iptr; { char *cptr, *hptr, *sptr; int cyl, head, sect, wild; TOKEN buf; /* * Set cylinder pointer to beginning of string. */ cptr = str; /* * Look for the first slash. */ while ((*str != '\0') && (*str != '/')) str++; /* * If there wasn't one, convert string to an integer and return it. */ if (*str == '\0') { wild = physsects() - 1; if (geti(cptr, (int *)iptr, &wild)) return (-1); return (0); } /* * Null out the slash and set head pointer just beyond it. */ *str++ = '\0'; hptr = str; /* * Look for the second slash. */ while ((*str != '\0') && (*str != '/')) str++; /* * If there wasn't one, sector pointer points to a . */ if (*str == '\0') sptr = str; /* * If there was, null it out and set sector point just beyond it. */ else { *str++ = '\0'; sptr = str; } /* * Convert the cylinder part to an integer and store it. */ clean_token(buf, cptr); wild = ncyl + acyl - 1; if (geti(buf, &cyl, &wild)) return (-1); if ((cyl < 0) || (cyl >= (ncyl + acyl))) { err_print("`%d' is out of range.\n", cyl); return (-1); } /* * Convert the head part to an integer and store it. */ clean_token(buf, hptr); wild = nhead - 1; if (geti(buf, &head, &wild)) return (-1); if ((head < 0) || (head >= nhead)) { err_print("`%d' is out of range.\n", head); return (-1); } /* * Convert the sector part to an integer and store it. */ clean_token(buf, sptr); wild = sectors(head) - 1; if (geti(buf, §, &wild)) return (-1); if ((sect < 0) || (sect >= sectors(head))) { err_print("`%d' is out of range.\n", sect); return (-1); } /* * Combine the pieces into a block number and return it. */ *iptr = chs2bn(cyl, head, sect); return (0); } /* * This routine is the basis for all input into the program. It * understands the semantics of a set of input types, and provides * consistent error messages for all input. It allows for default * values and prompt strings. */ uint64_t input(type, promptstr, delim, param, deflt, cmdflag) int type; char *promptstr; int delim; u_ioparam_t *param; int *deflt; int cmdflag; { int interactive, help, i, length, index, tied; daddr_t bn; diskaddr_t bn64; char **str, **strings; TOKEN token, cleantoken; TOKEN token2, cleantoken2; char *arg; struct bounds *bounds; char *s; int value; int cyls, cylno; uint64_t blokno; float nmegs; float ngigs; char shell_argv[MAXPATHLEN]; part_deflt_t *part_deflt; efi_deflt_t *efi_deflt; /* * Optional integer input has been added as a hack. * Function result is 1 if user typed anything. * Whatever they typed is returned in *deflt. * This permits us to distinguish between "no value", * and actually entering in some value, for instance. */ if (type == FIO_OPINT) { assert(deflt != NULL); } reprompt: help = interactive = 0; /* * If we are inputting a command, flush any current input in the pipe. */ if (cmdflag == CMD_INPUT) flushline(); /* * Note whether the token is already present. */ if (!token_present) interactive = 1; /* * Print the prompt. */ fmt_print(promptstr); /* * If there is a default value, print it in a format appropriate * for the input type. */ if (deflt != NULL) { switch (type) { case FIO_BN: fmt_print("[%d, ", *deflt); pr_dblock(fmt_print, (daddr_t)*deflt); fmt_print("]"); break; case FIO_INT: fmt_print("[%d]", *deflt); break; case FIO_INT64: #if defined(lint) /* caller is longlong aligned specifying FIO_INT64 */ efi_deflt = NULL; #else efi_deflt = (efi_deflt_t *)deflt; #endif fmt_print("[%llu]", efi_deflt->start_sector); break; case FIO_CSTR: case FIO_MSTR: strings = (char **)param->io_charlist; for (i = 0, str = strings; i < *deflt; i++, str++) ; fmt_print("[%s]", *str); break; case FIO_OSTR: fmt_print("[\"%s\"]", (char *)deflt); break; case FIO_SLIST: /* * Search for a string matching the default * value. If found, use it. Otherwise * assume the default value is actually * an illegal choice, and default to * the first item in the list. */ s = find_string(param->io_slist, *deflt); if (s == (char *)NULL) { s = (param->io_slist)->str; } fmt_print("[%s]", s); break; case FIO_CYL: /* * Old-style partition size input, used to * modify complete partition tables */ fmt_print("[%db, %dc, %1.2fmb, %1.2fgb]", *deflt, bn2c(*deflt), bn2mb(*deflt), bn2gb(*deflt)); break; case FIO_ECYL: /* * set up pointer to partition defaults * structure */ part_deflt = (part_deflt_t *)deflt; /* * Build print format specifier. We use the * starting cylinder number which was entered * before this call to input(), in case the * user has changed it from the value in the * cur_parts->pinfo_map[].dkl_cylno * field for the current parition */ /* * Determine the proper default end cylinder: * Start Cyl Default Size End Cylinder * 0 0 0 * >0 0 Start Cyl * 0 >0 Default Size * (Cyls) - 1 * >0 >0 (Start + * Default Size * (Cyls)) -1 */ if (part_deflt->deflt_size == 0) { cylno = part_deflt->start_cyl; } else if (part_deflt->start_cyl == 0) { cylno = bn2c(part_deflt->deflt_size) - 1; } else { cylno = (bn2c(part_deflt->deflt_size) + part_deflt->start_cyl) - 1; } fmt_print("[%db, %dc, %de, %1.2fmb, %1.2fgb]", part_deflt->deflt_size, bn2c(part_deflt->deflt_size), cylno, bn2mb(part_deflt->deflt_size), bn2gb(part_deflt->deflt_size)); break; case FIO_EFI: #if defined(lint) /* caller is longlong aligned when specifying FIO_EFI */ efi_deflt = NULL; #else efi_deflt = (efi_deflt_t *)deflt; #endif fmt_print("[%llub, %llue, %llumb, %llugb, %llutb]", efi_deflt->end_sector, efi_deflt->start_sector + efi_deflt->end_sector - 1, (efi_deflt->end_sector * DEV_BSIZE) / (1024 * 1024), (efi_deflt->end_sector * DEV_BSIZE) / (1024 * 1024 * 1024), (efi_deflt->end_sector * DEV_BSIZE) / ((uint64_t)1024 * 1024 * 1024 * 1024)); break; case FIO_OPINT: /* no default value for optional input type */ fmt_print("[default]"); break; default: err_print("Error: unknown input type.\n"); fullabort(); } } /* * Print the delimiter character. */ fmt_print("%c ", delim); /* * Get the token. If we hit eof, exit the program gracefully. */ if (gettoken(token) == NULL) fullabort(); /* * check if the user has issued (!) , escape to shell */ if ((cmdflag == CMD_INPUT) && (token[0] == '!')) { /* get the list of arguments to shell command */ (void) memset(shell_argv, 0, sizeof (shell_argv)); /* initialize to the first token... */ arg = &token[1]; /* * ... and then collect all tokens until the end of * the line as arguments */ do { /* skip empty tokens. */ if (*arg == '\0') continue; /* * If either of the following two strlcat() * operations overflows, report an error and * exit gracefully. */ if ((strlcat(shell_argv, arg, sizeof (shell_argv)) >= sizeof (shell_argv)) || (strlcat(shell_argv, " ", sizeof (shell_argv)) >= sizeof (shell_argv))) { err_print("Error: Command line too long.\n"); fullabort(); } } while (token_present && (arg = gettoken(token)) != NULL); /* execute the shell command */ (void) execute_shell(shell_argv, sizeof (shell_argv)); redisplay_menu_list((char **)param->io_charlist); if (interactive) { goto reprompt; } } /* * Certain commands accept up to two tokens * Unfortunately, this is kind of a hack. */ token2[0] = 0; cleantoken2[0] = 0; if (type == FIO_CYL || type == FIO_ECYL) { if (token_present) { if (gettoken(token2) == NULL) fullabort(); clean_token(cleantoken2, token2); } } /* * Echo the token back to the user if it was in the pipe or we * are running out of a command file. */ if (!interactive || option_f) { if (token2[0] == 0) { fmt_print("%s\n", token); } else { fmt_print("%s %s\n", token, token2); } } /* * If we are logging, echo the token to the log file. The else * is necessary here because the above printf will also put the * token in the log file. */ else if (log_file) { log_print("%s %s\n", token, token2); } /* * If the token was not in the pipe and it wasn't a command, flush * the rest of the line to keep things in sync. */ if (interactive && cmdflag != CMD_INPUT) flushline(); /* * Scrub off the white-space. */ clean_token(cleantoken, token); /* * If the input was a blank line and we weren't prompting * specifically for a blank line... */ if ((strcmp(cleantoken, "") == 0) && (type != FIO_BLNK)) { /* * If there's a default, return it. */ if (deflt != NULL) { if (type == FIO_OSTR) { /* * Duplicate and return the default string */ return ((int)alloc_string((char *)deflt)); } else if (type == FIO_SLIST) { /* * If we can find a match for the default * value in the list, return the default * value. If there's no match for the * default value, it's an illegal * choice. Return the first value in * the list. */ s = find_string(param->io_slist, *deflt); if ((cur_label == L_TYPE_EFI) && (s == (char *)NULL)) { return (*deflt); } if (s == (char *)NULL) { return ((param->io_slist)->value); } else { return (*deflt); } } else if (type == FIO_OPINT) { /* * The user didn't enter anything */ return (0); } else if (type == FIO_ECYL) { return (part_deflt->deflt_size); } else if (type == FIO_INT64) { return (efi_deflt->start_sector); } else if (type == FIO_EFI) { return (efi_deflt->end_sector); } else { return (*deflt); } } /* * If the blank was not in the pipe, just reprompt. */ if (interactive) { goto reprompt; } /* * If the blank was in the pipe, it's an error. */ err_print("No default for this entry.\n"); cmdabort(SIGINT); } /* * If token is a '?' or a 'h', it is a request for help. */ if ((strcmp(cleantoken, "?") == 0) || (strcmp(cleantoken, "h") == 0) || (strcmp(cleantoken, "help") == 0)) { help = 1; } /* * Switch on the type of input expected. */ switch (type) { /* * Expecting a disk block number. */ case FIO_BN: /* * Parameter is the bounds of legal block numbers. */ bounds = (struct bounds *)¶m->io_bounds; /* * Print help message if required. */ if (help) { fmt_print("Expecting a block number from %llu (", bounds->lower); pr_dblock(fmt_print, bounds->lower); fmt_print(") to %llu (", bounds->upper); pr_dblock(fmt_print, bounds->upper); fmt_print(")\n"); break; } /* * Convert token to a disk block number. */ if (cur_label == L_TYPE_EFI) { if (geti64(cleantoken, (uint64_t *)&bn64, (uint64_t *)NULL)) break; } else { if (getbn(cleantoken, &bn)) break; } if (cur_label == L_TYPE_EFI) { if ((bn64 < bounds->lower) || (bn64 > bounds->upper)) { err_print("`"); pr_dblock(err_print, bn64); err_print("' is out of range.\n"); break; } return (bn64); } /* * Check to be sure it is within the legal bounds. */ if ((bn < bounds->lower) || (bn > bounds->upper)) { err_print("`"); pr_dblock(err_print, bn); err_print("' is out of range.\n"); break; } /* * If it's ok, return it. */ return (bn); /* * Expecting an integer. */ case FIO_INT: /* * Parameter is the bounds of legal integers. */ bounds = (struct bounds *)¶m->io_bounds; /* * Print help message if required. */ if (help) { fmt_print("Expecting an integer from %llu", bounds->lower); fmt_print(" to %llu\n", bounds->upper); break; } /* * Convert the token into an integer. */ if (geti(cleantoken, (int *)&bn, (int *)NULL)) break; /* * Check to be sure it is within the legal bounds. */ if ((bn < bounds->lower) || (bn > bounds->upper)) { err_print("`%ld' is out of range.\n", bn); break; } /* * If it's ok, return it. */ return (bn); case FIO_INT64: /* * Parameter is the bounds of legal integers. */ bounds = (struct bounds *)¶m->io_bounds; /* * Print help message if required. */ if (help) { fmt_print("Expecting an integer from %llu", bounds->lower); fmt_print(" to %llu\n", bounds->upper); break; } /* * Convert the token into an integer. */ if (geti64(cleantoken, (uint64_t *)&bn64, (uint64_t *)NULL)) { break; } /* * Check to be sure it is within the legal bounds. */ if ((bn64 < bounds->lower) || (bn64 > bounds->upper)) { err_print("`%llu' is out of range.\n", bn64); break; } /* * If it's ok, return it. */ return (bn64); /* * Expecting an integer, or no input. */ case FIO_OPINT: /* * Parameter is the bounds of legal integers. */ bounds = (struct bounds *)¶m->io_bounds; /* * Print help message if required. */ if (help) { fmt_print("Expecting an integer from %llu", bounds->lower); fmt_print(" to %llu, or no input\n", bounds->upper); break; } /* * Convert the token into an integer. */ if (geti(cleantoken, (int *)&bn, (int *)NULL)) break; /* * Check to be sure it is within the legal bounds. */ if ((bn < bounds->lower) || (bn > bounds->upper)) { err_print("`%ld' is out of range.\n", bn); break; } /* * For optional case, return 1 indicating that * the user actually did enter something. */ *deflt = bn; return (1); /* * Expecting a closed string. This means that the input * string must exactly match one of the strings passed in * as the parameter. */ case FIO_CSTR: /* * The parameter is a null terminated array of character * pointers, each one pointing to a legal input string. */ strings = (char **)param->io_charlist; /* * Walk through the legal strings, seeing if any of them * match the token. If a match is made, return the index * of the string that was matched. */ for (str = strings; *str != NULL; str++) if (strcmp(cleantoken, *str) == 0) return (str - strings); /* * Print help message if required. */ if (help) { print_input_choices(type, param); } else { err_print("`%s' is not expected.\n", cleantoken); } break; /* * Expecting a matched string. This means that the input * string must either match one of the strings passed in, * or be a unique abbreviation of one of them. */ case FIO_MSTR: /* * The parameter is a null terminated array of character * pointers, each one pointing to a legal input string. */ strings = (char **)param->io_charlist; length = index = tied = 0; /* * Loop through the legal input strings. */ for (str = strings; *str != NULL; str++) { /* * See how many characters of the token match * this legal string. */ i = strcnt(cleantoken, *str); /* * If it's not the whole token, then it's not a match. */ if ((uint_t)i < strlen(cleantoken)) continue; /* * If it ties with another input, remember that. */ if (i == length) tied = 1; /* * If it matches the most so far, record that. */ if (i > length) { index = str - strings; tied = 0; length = i; } } /* * Print help message if required. */ if (length == 0) { if (help) { print_input_choices(type, param); } else { err_print("`%s' is not expected.\n", cleantoken); } break; } /* * If the abbreviation was non-unique, it's an error. */ if (tied) { err_print("`%s' is ambiguous.\n", cleantoken); break; } /* * We matched one. Return the index of the string we matched. */ return (index); /* * Expecting an open string. This means that any string is legal. */ case FIO_OSTR: /* * Print a help message if required. */ if (help) { fmt_print("Expecting a string\n"); break; } /* * alloc a copy of the string and return it */ return ((int)alloc_string(token)); /* * Expecting a blank line. */ case FIO_BLNK: /* * We are always in non-echo mode when we are inputting * this type. We echo the newline as a carriage return * only so the prompt string will be covered over. */ nolog_print("\015"); /* * If we are logging, send a newline to the log file. */ if (log_file) log_print("\n"); /* * There is no value returned for this type. */ return (0); /* * Expecting one of the entries in a string list. * Accept unique abbreviations. * Return the value associated with the matched string. */ case FIO_SLIST: i = find_value((slist_t *)param->io_slist, cleantoken, &value); if (i == 1) { return (value); } else { /* * Print help message if required. */ if (help) { print_input_choices(type, param); } else { if (i == 0) err_print("`%s' not expected.\n", cleantoken); else err_print("`%s' is ambiguous.\n", cleantoken); } } break; /* * Cylinder size input when modifying a complete partition map */ case FIO_CYL: /* * Parameter is the bounds of legal block numbers. */ bounds = (struct bounds *)¶m->io_bounds; assert(bounds->lower == 0); /* * Print help message if required. */ if (help) { fmt_print("Expecting up to %llu blocks,", bounds->upper); fmt_print(" %llu cylinders, ", bn2c(bounds->upper)); fmt_print(" %1.2f megabytes, ", bn2mb(bounds->upper)); fmt_print("or %1.2f gigabytes\n", bn2gb(bounds->upper)); break; } /* * Parse the first token: try to find 'b', 'c' or 'm' */ s = cleantoken; while (*s && (isdigit(*s) || (*s == '.') || (*s == '$'))) { s++; } /* * If we found a conversion specifier, second token is unused * Otherwise, the second token should supply it. */ if (*s != 0) { value = *s; *s = 0; } else { value = cleantoken2[0]; } /* * If the token is the wild card, simply supply the max * This order allows the user to specify the maximum in * either blocks/cyls/megabytes - a convenient fiction. */ if (strcmp(cleantoken, WILD_STRING) == 0) { return (bounds->upper); } /* * Allow the user to specify zero with no units, * by just defaulting to cylinders. */ if (strcmp(cleantoken, "0") == 0) { value = 'c'; } /* * If there's a decimal point, but no unit specification, * let's assume megabytes. */ if ((value == 0) && (strchr(cleantoken, '.') != NULL)) { value = 'm'; } /* * Handle each unit type we support */ switch (value) { case 'b': /* * Convert token to a disk block number. */ i = bounds->upper; if (geti(cleantoken, &value, &i)) break; bn = value; /* * Check to be sure it is within the legal bounds. */ if ((bn < bounds->lower) || (bn > bounds->upper)) { err_print( "`%ldb' is out of the range %llu to %llu\n", bn, bounds->lower, bounds->upper); break; } /* * Verify the block lies on a cylinder boundary */ if ((bn % spc()) != 0) { err_print( "partition size must be a multiple of %d blocks to lie on a cylinder \ boundary\n", spc()); err_print( "%ld blocks is approximately %ld cylinders, %1.2f megabytes or %1.2f\ gigabytes\n", bn, bn2c(bn), bn2mb(bn), bn2gb(bn)); break; } return (bn); case 'c': /* * Convert token from a number of cylinders to * a number of blocks. */ i = bn2c(bounds->upper); if (geti(cleantoken, &cyls, &i)) break; /* * Check the bounds - cyls is number of cylinders */ if (cyls > (bounds->upper/spc())) { err_print("`%dc' is out of range\n", cyls); break; } /* * Convert cylinders to blocks and return */ return (cyls * spc()); case 'm': /* * Convert token from megabytes to a block number. */ if (sscanf(cleantoken, "%f2", &nmegs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } /* * Check the bounds */ if (nmegs > bn2mb(bounds->upper)) { err_print("`%1.2fmb' is out of range\n", nmegs); break; } /* * Convert to blocks */ bn = mb2bn(nmegs); /* * Round value up to nearest cylinder */ i = spc(); bn = ((bn + (i-1)) / i) * i; return (bn); case 'g': /* * Convert token from gigabytes to a block number. */ if (sscanf(cleantoken, "%f2", &ngigs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } /* * Check the bounds */ if (ngigs > bn2gb(bounds->upper)) { err_print("`%1.2fgb' is out of range\n", ngigs); break; } /* * Convert to blocks */ bn = gb2bn(ngigs); /* * Round value up to nearest cylinder */ i = spc(); bn = ((bn + (i-1)) / i) * i; return (bn); default: err_print( "Please specify units in either b(blocks), c(cylinders), m(megabytes) \ or g(gigabytes)\n"); break; } break; case FIO_ECYL: /* * Parameter is the bounds of legal block numbers. */ bounds = (struct bounds *)¶m->io_bounds; assert(bounds->lower == 0); /* * Print help message if required. */ if (help) { fmt_print("Expecting up to %llu blocks,", bounds->upper); fmt_print(" %llu cylinders, ", bn2c(bounds->upper)); fmt_print(" %llu end cylinder, ", (bounds->upper / spc())); fmt_print(" %1.2f megabytes, ", bn2mb(bounds->upper)); fmt_print("or %1.2f gigabytes\n", bn2gb(bounds->upper)); break; } /* * Parse the first token: try to find 'b', 'c', 'e' * or 'm' */ s = cleantoken; while (*s && (isdigit(*s) || (*s == '.') || (*s == '$'))) { s++; } /* * If we found a conversion specifier, second token is * unused Otherwise, the second token should supply it. */ if (*s != 0) { value = *s; *s = 0; } else { value = cleantoken2[0]; } /* * If the token is the wild card, simply supply the max * This order allows the user to specify the maximum in * either blocks/cyls/megabytes - a convenient fiction. */ if (strcmp(cleantoken, WILD_STRING) == 0) { return (bounds->upper); } /* * Allow the user to specify zero with no units, * by just defaulting to cylinders. */ if (value != 'e' && strcmp(cleantoken, "0") == 0) { value = 'c'; } /* * If there's a decimal point, but no unit * specification, let's assume megabytes. */ if ((value == 0) && (strchr(cleantoken, '.') != NULL)) { value = 'm'; } /* * Handle each unit type we support */ switch (value) { case 'b': /* * Convert token to a disk block number. */ i = bounds->upper; if (geti(cleantoken, &value, &i)) break; bn = value; /* * Check to be sure it is within the * legal bounds. */ if ((bn < bounds->lower) || (bn > bounds->upper)) { err_print( "`%ldb' is out of the range %llu to %llu\n", bn, bounds->lower, bounds->upper); break; } /* * Verify the block lies on a cylinder * boundary */ if ((bn % spc()) != 0) { err_print( "partition size must be a multiple of %d blocks to lie on a cylinder \ boundary\n", spc()); err_print( "%ld blocks is approximately %ld cylinders, %1.2f \ megabytes or %1.2f gigabytes\n", bn, bn2c(bn), bn2mb(bn), bn2gb(bn)); break; } return (bn); case 'e': /* * Token is ending cylinder */ /* convert token to integer */ if (geti(cleantoken, &cylno, (int *)NULL)) { break; } /* * check that input cylno isn't before the current * starting cylinder number. Note that we are NOT * using the starting cylinder from * cur_parts->pinfo_map[].dkl_cylno! */ if (cylno < part_deflt->start_cyl) { err_print( "End cylinder must fall on or after start cylinder %d\n", part_deflt->start_cyl); break; } /* * calculate cylinder number of upper boundary, and * verify that our input is within range */ i = (bn2c(bounds->upper) + part_deflt->start_cyl - 1); if (cylno > i) { err_print( "End cylinder %d is beyond max cylinder %d\n", cylno, i); break; } /* * calculate number of cylinders based on input */ cyls = ((cylno - part_deflt->start_cyl) + 1); return (cyls * spc()); case 'c': /* * Convert token from a number of * cylinders to a number of blocks. */ i = bn2c(bounds->upper); if (geti(cleantoken, &cyls, &i)) break; /* * Check the bounds - cyls is number of * cylinders */ if (cyls > (bounds->upper/spc())) { err_print("`%dc' is out of range\n", cyls); break; } /* * Convert cylinders to blocks and * return */ return (cyls * spc()); case 'm': /* * Convert token from megabytes to a * block number. */ if (sscanf(cleantoken, "%f2", &nmegs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } /* * Check the bounds */ if (nmegs > bn2mb(bounds->upper)) { err_print("`%1.2fmb' is out of range\n", nmegs); break; } /* * Convert to blocks */ bn = mb2bn(nmegs); /* * Round value up to nearest cylinder */ i = spc(); bn = ((bn + (i-1)) / i) * i; return (bn); case 'g': /* * Convert token from gigabytes to a * block number. */ if (sscanf(cleantoken, "%f2", &ngigs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } /* * Check the bounds */ if (ngigs > bn2gb(bounds->upper)) { err_print("`%1.2fgb' is out of range\n", ngigs); break; } /* * Convert to blocks */ bn = gb2bn(ngigs); /* * Round value up to nearest cylinder */ i = spc(); bn = ((bn + (i-1)) / i) * i; return (bn); default: err_print( "Please specify units in either b(blocks), c(cylinders), e(end cylinder),\n"); err_print("m(megabytes) or g(gigabytes)\n"); break; } break; case FIO_EFI: /* * Parameter is the bounds of legal block numbers. */ bounds = (struct bounds *)¶m->io_bounds; /* * Print help message if required. */ if (help) { fmt_print("Expecting up to %llu sectors,", cur_parts->etoc->efi_last_u_lba); fmt_print("or %llu megabytes,", (cur_parts->etoc->efi_last_u_lba * DEV_BSIZE)/ (1024 * 1024)); fmt_print("or %llu gigabytes\n", (cur_parts->etoc->efi_last_u_lba * DEV_BSIZE)/ (1024 * 1024 * 1024)); fmt_print("or %llu terabytes\n", (cur_parts->etoc->efi_last_u_lba * DEV_BSIZE)/ ((uint64_t)1024 * 1024 * 1024 * 1024)); break; } /* * Parse the first token: try to find 'b', 'c', 'e' * or 'm' */ s = cleantoken; while (*s && (isdigit(*s) || (*s == '.') || (*s == '$'))) { s++; } /* * If we found a conversion specifier, second token is * unused Otherwise, the second token should supply it. */ if (*s != 0) { value = *s; *s = 0; } else { value = cleantoken2[0]; } /* * If the token is the wild card, simply supply the max * This order allows the user to specify the maximum in * either blocks/cyls/megabytes - a convenient fiction. */ if (strcmp(cleantoken, WILD_STRING) == 0) { return (bounds->upper - EFI_MIN_RESV_SIZE - efi_deflt->start_sector); } /* * Allow the user to specify zero with no units, * by just defaulting to sectors. */ if (value != 'e' && strcmp(cleantoken, "0") == 0) { value = 'm'; } /* * If there's a decimal point, but no unit * specification, let's assume megabytes. */ if ((value == 0) && (strchr(cleantoken, '.') != NULL)) { value = 'm'; } /* * Handle each unit type we support */ switch (value) { case 'b': /* * Token is number of blocks */ if (geti64(cleantoken, &blokno, (uint64_t *)NULL)) { break; } if (blokno > bounds->upper) { err_print( "Number of blocks must be less that the total available blocks.\n"); break; } return (blokno); case 'e': /* * Token is ending block number */ /* convert token to integer */ if (geti64(cleantoken, &blokno, (uint64_t *)NULL)) { break; } /* * Some sanity check */ if (blokno < efi_deflt->start_sector) { err_print( "End Sector must fall on or after start sector %llu\n", efi_deflt->start_sector); break; } /* * verify that our input is within range */ if (blokno > cur_parts->etoc->efi_last_u_lba) { err_print( "End Sector %llu is beyond max Sector %llu\n", blokno, cur_parts->etoc->efi_last_u_lba); break; } /* * calculate number of blocks based on input */ return (blokno - efi_deflt->start_sector + 1); case 'm': /* * Convert token from megabytes to a * block number. */ if (sscanf(cleantoken, "%f2", &nmegs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } /* * Check the bounds */ if (nmegs > bn2mb(bounds->upper - bounds->lower)) { err_print("`%1.2fmb' is out of range\n", nmegs); break; } return (mb2bn(nmegs)); case 'g': if (sscanf(cleantoken, "%f2", &nmegs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } if (nmegs > bn2gb(bounds->upper - bounds->lower)) { err_print("`%1.2fgb' is out of range\n", nmegs); break; } return (gb2bn(nmegs)); case 't': if (sscanf(cleantoken, "%f2", &nmegs) != 1) { err_print("`%s' is not recognized\n", cleantoken); break; } if (nmegs > bn2tb(bounds->upper - bounds->lower)) { err_print("`%1.2ftb' is out of range\n", nmegs); break; } return (uint64_t)((float)nmegs * 1024.0 * 1024.0 * 1024.0 * 1024.0 / DEV_BSIZE); default: err_print( "Please specify units in either b(number of blocks), e(end sector),\n"); err_print(" g(gigabytes), m(megabytes)"); err_print(" or t(terabytes)\n"); break; } break; /* * If we don't recognize the input type, it's bad news. */ default: err_print("Error: unknown input type.\n"); fullabort(); } /* * If we get here, it's because some error kept us from accepting * the token. If we are running out of a command file, gracefully * leave the program. If we are interacting with the user, simply * reprompt. If the token was in the pipe, abort the current command. */ if (option_f) fullabort(); else if (interactive) goto reprompt; else cmdabort(SIGINT); /* * Never actually reached. */ return (-1); } /* * Print input choices */ static void print_input_choices(type, param) int type; u_ioparam_t *param; { char **sp; slist_t *lp; int width; int col; int ncols; switch (type) { case FIO_CSTR: fmt_print("Expecting one of the following:\n"); goto common; case FIO_MSTR: fmt_print("Expecting one of the following: "); fmt_print("(abbreviations ok):\n"); common: for (sp = (char **)param->io_charlist; *sp != NULL; sp++) { fmt_print("\t%s\n", *sp); } break; case FIO_SLIST: fmt_print("Expecting one of the following: "); fmt_print("(abbreviations ok):\n"); /* * Figure out the width of the widest string */ width = slist_widest_str((slist_t *)param->io_slist); width += 4; /* * If the help messages are empty, print the * possible choices in left-justified columns */ lp = (slist_t *)param->io_slist; if (*lp->help == 0) { col = 0; ncols = 60 / width; for (; lp->str != NULL; lp++) { if (col == 0) fmt_print("\t"); ljust_print(lp->str, (++col == ncols) ? 0 : width); if (col == ncols) { col = 0; fmt_print("\n"); } } if (col != 0) fmt_print("\n"); } else { /* * With help messages, print each choice, * and help message, on its own line. */ for (; lp->str != NULL; lp++) { fmt_print("\t"); ljust_print(lp->str, width); fmt_print("- %s\n", lp->help); } } break; default: err_print("Error: unknown input type.\n"); fullabort(); } fmt_print("\n"); } /* * Search a string list for a particular string. * Use minimum recognition, to accept unique abbreviations * Return the number of possible matches found. * If only one match was found, return the arbitrary value * associated with the matched string in match_value. */ int find_value(slist, match_str, match_value) slist_t *slist; char *match_str; int *match_value; { int i; int nmatches; int length; int match_length; nmatches = 0; length = 0; match_length = strlen(match_str); for (; slist->str != NULL; slist++) { /* * See how many characters of the token match */ i = strcnt(match_str, slist->str); /* * If it's not the whole token, then it's not a match. */ if (i < match_length) continue; /* * If it ties with another input, remember that. */ if (i == length) nmatches++; /* * If it matches the most so far, record that. */ if (i > length) { *match_value = slist->value; nmatches = 1; length = i; } } return (nmatches); } /* * Search a string list for a particular value. * Return the string associated with that value. */ char * find_string(slist, match_value) slist_t *slist; int match_value; { for (; slist->str != NULL; slist++) { if (slist->value == match_value) { return (slist->str); } } return ((char *)NULL); } /* * Return the width of the widest string in an slist */ static int slist_widest_str(slist) slist_t *slist; { int i; int width; width = 0; for (; slist->str != NULL; slist++) { if ((i = strlen(slist->str)) > width) width = i; } return (width); } /* * Print a string left-justified to a fixed width. */ static void ljust_print(str, width) char *str; int width; { int i; fmt_print("%s", str); for (i = width - strlen(str); i > 0; i--) { fmt_print(" "); } } /* * This routine is a modified version of printf. It handles the cases * of silent mode and logging; other than that it is identical to the * library version. */ /*PRINTFLIKE1*/ void fmt_print(char *format, ...) { va_list ap; va_start(ap, format); /* * If we are running silent, skip it. */ if (option_s == 0) { /* * Do the print to standard out. */ if (need_newline) { (void) printf("\n"); } (void) vprintf(format, ap); /* * If we are logging, also print to the log file. */ if (log_file) { if (need_newline) { (void) fprintf(log_file, "\n"); } (void) vfprintf(log_file, format, ap); (void) fflush(log_file); } } need_newline = 0; va_end(ap); } /* * This routine is a modified version of printf. It handles the cases * of silent mode; other than that it is identical to the * library version. It differs from the above printf in that it does * not print the message to a log file. */ /*PRINTFLIKE1*/ void nolog_print(char *format, ...) { va_list ap; va_start(ap, format); /* * If we are running silent, skip it. */ if (option_s == 0) { /* * Do the print to standard out. */ if (need_newline) { (void) printf("\n"); } (void) vprintf(format, ap); } va_end(ap); need_newline = 0; } /* * This routine is a modified version of printf. It handles the cases * of silent mode, and only prints the message to the log file, not * stdout. Other than that is identical to the library version. */ /*PRINTFLIKE1*/ void log_print(char *format, ...) { va_list ap; va_start(ap, format); /* * If we are running silent, skip it. */ if (option_s == 0) { /* * Do the print to the log file. */ if (need_newline) { (void) fprintf(log_file, "\n"); } (void) vfprintf(log_file, format, ap); (void) fflush(log_file); } va_end(ap); need_newline = 0; } /* * This routine is a modified version of printf. It prints the message * to stderr, and to the log file is appropriate. * Other than that is identical to the library version. */ /*PRINTFLIKE1*/ void err_print(char *format, ...) { va_list ap; va_start(ap, format); /* * Flush anything pending to stdout */ if (need_newline) { (void) printf("\n"); } (void) fflush(stdout); /* * Do the print to stderr. */ (void) vfprintf(stderr, format, ap); /* * If we are logging, also print to the log file. */ if (log_file) { if (need_newline) { (void) fprintf(log_file, "\n"); } (void) vfprintf(log_file, format, ap); (void) fflush(log_file); } va_end(ap); need_newline = 0; } /* * Print a number of characters from a buffer. The buffer * does not need to be null-terminated. Since the data * may be coming from a device, we cannot be sure the * data is not crud, so be rather defensive. */ void print_buf(buf, nbytes) char *buf; int nbytes; { int c; while (nbytes-- > 0) { c = *buf++; if (isascii(c) && isprint(c)) { fmt_print("%c", c); } else break; } } #ifdef not /* * This routine prints out a message describing the given ctlr. * The message is identical to the one printed by the kernel during * booting. */ void pr_ctlrline(ctlr) register struct ctlr_info *ctlr; { fmt_print(" %s%d at %s 0x%x ", ctlr->ctlr_cname, ctlr->ctlr_num, space2str(ctlr->ctlr_space), ctlr->ctlr_addr); if (ctlr->ctlr_vec != 0) fmt_print("vec 0x%x ", ctlr->ctlr_vec); else fmt_print("pri %d ", ctlr->ctlr_prio); fmt_print("\n"); } #endif /* not */ /* * This routine prints out a message describing the given disk. * The message is identical to the one printed by the kernel during * booting. */ void pr_diskline(disk, num) register struct disk_info *disk; int num; { struct ctlr_info *ctlr = disk->disk_ctlr; struct disk_type *type = disk->disk_type; fmt_print(" %4d. %s ", num, disk->disk_name); if ((type != NULL) && (disk->label_type == L_TYPE_SOLARIS)) { fmt_print("<%s cyl %d alt %d hd %d sec %d>", type->dtype_asciilabel, type->dtype_ncyl, type->dtype_acyl, type->dtype_nhead, type->dtype_nsect); } else if ((type != NULL) && (disk->label_type == L_TYPE_EFI)) { print_efi_string(type->vendor, type->product, type->revision, type->capacity); } else if (disk->disk_flags & DSK_RESERVED) { fmt_print(""); } else if (disk->disk_flags & DSK_UNAVAILABLE) { fmt_print(""); } else { fmt_print(""); } if (chk_volname(disk)) { fmt_print(" "); print_volname(disk); } fmt_print("\n"); if (disk->devfs_name != NULL) { fmt_print(" %s\n", disk->devfs_name); } else { fmt_print(" %s%d at %s%d slave %d\n", ctlr->ctlr_dname, disk->disk_dkinfo.dki_unit, ctlr->ctlr_cname, ctlr->ctlr_num, disk->disk_dkinfo.dki_slave); } #ifdef OLD fmt_print(" %4d. %s at %s%d slave %d", num, disk->disk_name, ctlr->ctlr_cname, ctlr->ctlr_num, disk->disk_dkinfo.dki_slave); if (chk_volname(disk)) { fmt_print(": "); print_volname(disk); } fmt_print("\n"); if (type != NULL) { fmt_print( " %s%d: <%s cyl %d alt %d hd %d sec %d>\n", ctlr->ctlr_dname, disk->disk_dkinfo.dki_unit, type->dtype_asciilabel, type->dtype_ncyl, type->dtype_acyl, type->dtype_nhead, type->dtype_nsect); } else { fmt_print(" %s%d: \n", ctlr->ctlr_dname, disk->disk_dkinfo.dki_unit); } #endif /* OLD */ } /* * This routine prints out a given disk block number in cylinder/head/sector * format. It uses the printing routine passed in to do the actual output. * Downcasting bn is okay for L_TYPE_SOLARIS because the number of blocks * on a Solaris (VTOC) label will never exceed 4 bytes (daddr_t). */ void pr_dblock(void (*func)(char *, ...), diskaddr_t bn) { if (cur_label == L_TYPE_SOLARIS) { (*func)("%d/%d/%d", bn2c((daddr_t)bn), bn2h((daddr_t)bn), bn2s((daddr_t)bn)); } else { (*func)("%llu", bn); } } /* * This routine inputs a character from the data file. It understands * the use of '\' to prevent interpretation of a newline. It also keeps * track of the current line in the data file via a global variable. */ static int sup_inputchar() { int c; /* * Input the character. */ c = getc(data_file); /* * If it's not a backslash, return it. */ if (c != '\\') return (c); /* * It was a backslash. Get the next character. */ c = getc(data_file); /* * If it was a newline, update the line counter and get the next * character. */ if (c == '\n') { data_lineno++; c = getc(data_file); } /* * Return the character. */ return (c); } /* * This routine pushes a character back onto the input pipe for the data file. */ static void sup_pushchar(c) int c; { (void) ungetc(c, data_file); } /* * Variables to support pushing back tokens */ static int have_pushed_token = 0; static TOKEN pushed_buf; static int pushed_token; /* * This routine inputs a token from the data file. A token is a series * of contiguous non-white characters or a recognized special delimiter * character. Use of the wrapper lets us always have the value of the * last token around, which is useful for error recovery. */ int sup_gettoken(buf) char *buf; { last_token_type = sup_get_token(buf); return (last_token_type); } static int sup_get_token(buf) char *buf; { char *ptr = buf; int c, quoted = 0; /* * First check for presence of push-backed token. * If so, return it. */ if (have_pushed_token) { have_pushed_token = 0; bcopy(pushed_buf, buf, TOKEN_SIZE+1); return (pushed_token); } /* * Zero out the returned token buffer */ bzero(buf, TOKEN_SIZE + 1); /* * Strip off leading white-space. */ while ((isspace(c = sup_inputchar())) && (c != '\n')) ; /* * Read in characters until we hit unquoted white-space. */ for (; !isspace(c) || quoted; c = sup_inputchar()) { /* * If we hit eof, that's a token. */ if (feof(data_file)) return (SUP_EOF); /* * If we hit a double quote, change the state of quoting. */ if (c == '"') { quoted = !quoted; continue; } /* * If we hit a newline, that delimits a token. */ if (c == '\n') break; /* * If we hit any nonquoted special delimiters, that delimits * a token. */ if (!quoted && (c == '=' || c == ',' || c == ':' || c == '#' || c == '|' || c == '&' || c == '~')) break; /* * Store the character if there's room left. */ if (ptr - buf < TOKEN_SIZE) *ptr++ = (char)c; } /* * If we stored characters in the buffer, then we inputted a string. * Push the delimiter back into the pipe and return the string. */ if (ptr - buf > 0) { sup_pushchar(c); return (SUP_STRING); } /* * We didn't input a string, so we must have inputted a known delimiter. * store the delimiter in the buffer, so it will get returned. */ buf[0] = c; /* * Switch on the delimiter. Return the appropriate value for each one. */ switch (c) { case '=': return (SUP_EQL); case ':': return (SUP_COLON); case ',': return (SUP_COMMA); case '\n': return (SUP_EOL); case '|': return (SUP_OR); case '&': return (SUP_AND); case '~': return (SUP_TILDE); case '#': /* * For comments, we flush out the rest of the line and return * an EOL. */ while ((c = sup_inputchar()) != '\n' && !feof(data_file)) ; if (feof(data_file)) return (SUP_EOF); else return (SUP_EOL); /* * Shouldn't ever get here. */ default: return (SUP_STRING); } } /* * Push back a token */ void sup_pushtoken(token_buf, token_type) char *token_buf; int token_type; { /* * We can only push one token back at a time */ assert(have_pushed_token == 0); have_pushed_token = 1; bcopy(token_buf, pushed_buf, TOKEN_SIZE+1); pushed_token = token_type; } /* * Get an entire line of input. Handles logging, comments, * and EOF. */ void get_inputline(line, nbytes) char *line; int nbytes; { char *p = line; int c; /* * Remove any leading white-space and comments */ do { while ((isspace(c = getchar())) && (c != '\n')) ; } while (c == COMMENT_CHAR); /* * Loop on each character until end of line */ while (c != '\n') { /* * If we hit eof, get out. */ if (checkeof()) { fullabort(); } /* * Add the character to the buffer. */ if (nbytes > 1) { *p++ = (char)c; nbytes --; } /* * Get the next character. */ c = getchar(); } /* * Null terminate the token. */ *p = 0; /* * Indicate that we've emptied the pipe */ token_present = 0; /* * If we're running out of a file, echo the line to * the user, otherwise if we're logging, copy the * input to the log file. */ if (option_f) { fmt_print("%s\n", line); } else if (log_file) { log_print("%s\n", line); } } /* * execute the shell escape command */ int execute_shell(s, buff_size) char *s; size_t buff_size; { struct termio termio; struct termios tty; int tty_flag, i, j; char *shell_name; static char *default_shell = "/bin/sh"; tty_flag = -1; if (*s == NULL) { shell_name = getenv("SHELL"); if (shell_name == NULL) { shell_name = default_shell; } if (strlcpy(s, shell_name, buff_size) >= buff_size) { err_print("Error: Shell command ($SHELL) too long.\n"); fullabort(); } } /* save tty information */ if (isatty(0)) { if (ioctl(0, TCGETS, &tty) == 0) tty_flag = 1; else { if (ioctl(0, TCGETA, &termio) == 0) { tty_flag = 0; tty.c_iflag = termio.c_iflag; tty.c_oflag = termio.c_oflag; tty.c_cflag = termio.c_cflag; tty.c_lflag = termio.c_lflag; for (i = 0; i < NCC; i++) tty.c_cc[i] = termio.c_cc[i]; } } } /* close the current file descriptor */ if (cur_disk != NULL) { (void) close(cur_file); } /* execute the shell escape */ (void) system(s); /* reopen file descriptor if one was open before */ if (cur_disk != NULL) { if ((cur_file = open_disk(cur_disk->disk_path, O_RDWR | O_NDELAY)) < 0) { err_print("Error: can't reopen selected disk '%s'. \n", cur_disk->disk_name); fullabort(); } } /* Restore tty information */ if (isatty(0)) { if (tty_flag > 0) (void) ioctl(0, TCSETSW, &tty); else if (tty_flag == 0) { termio.c_iflag = tty.c_iflag; termio.c_oflag = tty.c_oflag; termio.c_cflag = tty.c_cflag; termio.c_lflag = tty.c_lflag; for (j = 0; j < NCC; j++) termio.c_cc[j] = tty.c_cc[j]; (void) ioctl(0, TCSETAW, &termio); } if (isatty(1)) { fmt_print("\n[Hit Return to continue] \n"); (void) fflush(stdin); if (getchar() == EOF) fullabort(); } } return (0); } void print_efi_string(char *vendor, char *product, char *revision, uint64_t capacity) { char new_vendor[9]; char new_product[17]; char new_revision[5]; char capacity_string[10]; float scaled; int i; /* Strip whitespace from the end of inquiry strings */ (void) strlcpy(new_vendor, vendor, sizeof (new_vendor)); for (i = (strlen(new_vendor) - 1); i >= 0; i--) { if (new_vendor[i] != 0x20) { new_vendor[i+1] = '\0'; break; } } (void) strlcpy(new_product, product, sizeof (new_product)); for (i = (strlen(new_product) - 1); i >= 0; i--) { if (new_product[i] != 0x20) { new_product[i+1] = '\0'; break; } } (void) strlcpy(new_revision, revision, sizeof (new_revision)); for (i = (strlen(new_revision) - 1); i >= 0; i--) { if (new_revision[i] != 0x20) { new_revision[i+1] = '\0'; break; } } /* Now build size string */ scaled = bn2mb(capacity); if (scaled >= (float)1024.0 * 1024) { (void) snprintf(capacity_string, sizeof (capacity_string), "%.2fTB", scaled/((float)1024.0 * 1024)); } else if (scaled >= (float)1024.0) { (void) snprintf(capacity_string, sizeof (capacity_string), "%.2fGB", scaled/(float)1024.0); } else { (void) snprintf(capacity_string, sizeof (capacity_string), "%.2fMB", scaled); } fmt_print("<%s-%s-%s-%s>", new_vendor, new_product, new_revision, capacity_string); }