/* * mkswap.c - set up a linux swap device * * (C) 1991 Linus Torvalds. This file may be redistributed as per * the Linux copyright. */ /* * 20.12.91 - time began. Got VM working yesterday by doing this by hand. * * Usage: mkswap [-c] [-vN] [-f] device [size-in-blocks] * * -c for readability checking. (Use it unless you are SURE!) * -vN for swap areas version N. (Only N=0,1 known today.) * -f for forcing swap creation even if it would smash partition table. * * The device may be a block device or an image of one, but this isn't * enforced (but it's not much fun on a character device :-). * * Patches from jaggy@purplet.demon.co.uk (Mike Jagdis) to make the * size-in-blocks parameter optional added Wed Feb 8 10:33:43 1995. * * Version 1 swap area code (for kernel 2.1.117), aeb, 981010. * * Sparc fixes, jj@ultra.linux.cz (Jakub Jelinek), 981201 - mangled by aeb. * V1_MAX_PAGES fixes, jj, 990325. * sparc64 fixes, jj, 000219. * * 1999-02-22 Arkadiusz Mi¶kiewicz * - added Native Language Support * */ #include #include #include #include #include #include #include /* for _IO */ #include #include #include #ifdef HAVE_LIBSELINUX #include #include #endif #include "linux_version.h" #include "swapheader.h" #include "xstrncpy.h" #include "nls.h" #include "blkdev.h" #ifdef HAVE_LIBUUID #include #endif static char * program_name = "mkswap"; static char * device_name = NULL; static int DEV = -1; static unsigned long PAGES = 0; static unsigned long badpages = 0; static int check = 0; static int version = -1; #define SELINUX_SWAPFILE_TYPE "swapfile_t" #ifdef __sparc__ # ifdef __arch64__ # define is_sparc64() 1 # define is_be64() 1 # else /* sparc32 */ static int is_sparc64(void) { struct utsname un; static int sparc64 = -1; if (sparc64 != -1) return sparc64; sparc64 = 0; if (uname(&un) < 0) return 0; if (! strcmp(un.machine, "sparc64")) { sparc64 = 1; return 1; } if (strcmp(un.machine, "sparc")) return 0; /* Should not happen */ #ifdef HAVE_PERSONALITY { extern int personality(unsigned long); int oldpers; #define PERS_LINUX 0x00000000 #define PERS_LINUX_32BIT 0x00800000 #define PERS_LINUX32 0x00000008 oldpers = personality(PERS_LINUX_32BIT); if (oldpers != -1) { if (personality(PERS_LINUX) != -1) { uname(&un); if (! strcmp(un.machine, "sparc64")) { sparc64 = 1; oldpers = PERS_LINUX32; } } personality(oldpers); } } #endif return sparc64; } # define is_be64() is_sparc64() # endif /* sparc32 */ #else /* !sparc */ # define is_be64() 0 #endif /* * The definition of the union swap_header uses the kernel constant PAGE_SIZE. * Unfortunately, on some architectures this depends on the hardware model, and * can only be found at run time -- we use getpagesize(), so that we do not * need separate binaries e.g. for sun4, sun4c/d/m and sun4u. * * Even more unfortunately, getpagesize() does not always return the right * information. For example, libc4, libc5 and glibc 2.0 do not use the system * call but invent a value themselves (EXEC_PAGESIZE or NBPG * CLSIZE or NBPC), * and thus it may happen that e.g. on a sparc kernel PAGE_SIZE=4096 and * getpagesize() returns 8192. * * What to do? Let us allow the user to specify the pagesize explicitly. * * Update 05-Feb-2007 (kzak): * - use sysconf(_SC_PAGESIZE) to be consistent with the rest of * util-linux code. It is the standardized and preferred way of * querying page size. */ static int user_pagesize; static int pagesize; static unsigned long *signature_page; struct swap_header_v1 *p; static void init_signature_page(void) { int kernel_pagesize = pagesize = (int) sysconf(_SC_PAGESIZE); if (user_pagesize) { if ((user_pagesize & (user_pagesize-1)) || user_pagesize < 1024) { fprintf(stderr, _("Bad user-specified page size %d\n"), user_pagesize); exit(1); } pagesize = user_pagesize; } if (user_pagesize && user_pagesize != kernel_pagesize) fprintf(stderr, _("Using user-specified page size %d, " "instead of the system value %d\n"), pagesize, kernel_pagesize); signature_page = (unsigned long *) malloc(pagesize); memset(signature_page, 0, pagesize); p = (struct swap_header_v1 *) signature_page; } static void write_signature(char *sig) { char *sp = (char *) signature_page; strncpy(sp+pagesize-10, sig, 10); } static void write_uuid_and_label(unsigned char *uuid, char *volume_name) { struct swap_header_v1_2 *h; /* Sanity check */ if (sizeof(struct swap_header_v1) != sizeof(struct swap_header_v1_2)) { fprintf(stderr, _("Bad swap header size, no label written.\n")); return; } h = (struct swap_header_v1_2 *) signature_page; if (uuid) memcpy(h->uuid, uuid, sizeof(h->uuid)); if (volume_name) { xstrncpy(h->volume_name, volume_name, sizeof(h->volume_name)); if (strlen(volume_name) > strlen(h->volume_name)) fprintf(stderr, _("Label was truncated.\n")); } if (uuid || volume_name) { if (volume_name) printf("LABEL=%s, ", h->volume_name); else printf(_("no label, ")); #ifdef HAVE_LIBUUID if (uuid) { char uuid_string[37]; uuid_unparse(uuid, uuid_string); printf("UUID=%s\n", uuid_string); } else #endif printf(_("no uuid\n")); } } /* * Find out what the maximum amount of swap space is that the kernel will * handle. This wouldn't matter if the kernel just used as much of the * swap space as it can handle, but until 2.3.4 it would return an error * to swapon() if the swapspace was too large. */ #define V0_MAX_PAGES (8 * (pagesize - 10)) /* Before 2.2.0pre9 */ #define V1_OLD_MAX_PAGES ((0x7fffffff / pagesize) - 1) /* Since 2.2.0pre9, before 2.3.4: error if nr of pages >= SWP_OFFSET(SWP_ENTRY(0,~0UL)) with variations on #define SWP_ENTRY(type,offset) (((type) << 1) | ((offset) << 8)) #define SWP_OFFSET(entry) ((entry) >> 8) on the various architectures. Below the result - yuk. Machine pagesize SWP_ENTRY SWP_OFFSET bound+1 oldbound+2 i386 2^12 o<<8 e>>8 1<<24 1<<19 mips 2^12 o<<15 e>>15 1<<17 1<<19 alpha 2^13 o<<40 e>>40 1<<24 1<<18 m68k 2^12 o<<12 e>>12 1<<20 1<<19 sparc 2^{12,13} (o&0x3ffff)<<9 (e>>9)&0x3ffff 1<<18 1<<{19,18} sparc64 2^13 o<<13 e>>13 1<<51 1<<18 ppc 2^12 o<<8 e>>8 1<<24 1<<19 armo 2^{13,14,15} o<<8 e>>8 1<<24 1<<{18,17,16} armv 2^12 o<<9 e>>9 1<<23 1<<19 assuming that longs have 64 bits on alpha and sparc64 and 32 bits elsewhere. The bad part is that we need to know this since the kernel will refuse a swap space if it is too large. */ /* patch from jj - why does this differ from the above? */ /* 32bit kernels have a second limitation of 2GB, sparc64 is limited by the size of virtual address space allocation for vmalloc */ #if defined(__alpha__) #define V1_MAX_PAGES ((1 << 24) - 1) #elif defined(__mips__) #define V1_MAX_PAGES ((1 << 17) - 1) #elif defined(__sparc__) #define V1_MAX_PAGES (is_sparc64() ? ((3 << 29) - 1) : ((1 << 18) - 1)) #elif defined(__ia64__) /* * The actual size will depend on the amount of virtual address space * available to vmalloc the swap map. */ #define V1_MAX_PAGES ((1UL << 54) - 1) #else #define V1_MAX_PAGES V1_OLD_MAX_PAGES #endif /* man page now says: The maximum useful size of a swap area now depends on the architecture. It is roughly 2GB on i386, PPC, m68k, ARM, 1GB on sparc, 512MB on mips, 128GB on alpha and 3TB on sparc64. */ #define MAX_BADPAGES ((pagesize-1024-128*sizeof(int)-10)/sizeof(int)) /* * One more point of lossage - Linux swapspace really is a mess. * The definition of the bitmap used is architecture dependent, * and requires one to know whether the machine is bigendian, * and if so, whether it will use 32-bit or 64-bit units in * test_bit(). * davem writes: "... is based upon an unsigned long type of * the cpu and the native endianness". * So, it seems we can write `unsigned long' below. * However, sparc64 uses 64-bit units in the kernel, while * mkswap may have been translated with 32-bit longs. Thus, * we need an explicit test for version 0 swap on sparc64. */ static void bit_set (unsigned long *addr, unsigned int nr) { unsigned int r, m; if(is_be64()) { unsigned long long *bitmap = (unsigned long long *) addr; unsigned long long bitnum = (unsigned long long) nr; unsigned long long rl, ml; bitmap += bitnum / (8 * sizeof(long long)); rl = *bitmap; ml = 1ULL << (bitnum & (8ULL * sizeof(long long) - 1ULL)); *bitmap = rl | ml; return; } addr += nr / (8 * sizeof(unsigned long)); r = *addr; m = 1 << (nr & (8 * sizeof(unsigned long) - 1)); *addr = r | m; } static int bit_test_and_clear (unsigned long *addr, unsigned int nr) { unsigned int r, m; if(is_be64()) { unsigned long long *bitmap = (unsigned long long *) addr; unsigned long long bitnum = (unsigned long long) nr; unsigned long long rl, ml; bitmap += bitnum / (8 * sizeof(long long)); rl = *bitmap; ml = 1ULL << (bitnum & (8ULL * sizeof(long long) - 1ULL)); *bitmap = rl & ~ml; return ((rl & ml) != 0ULL); } addr += nr / (8 * sizeof(unsigned long)); r = *addr; m = 1 << (nr & (8 * sizeof(unsigned long) - 1)); *addr = r & ~m; return (r & m) != 0; } static void usage(void) { fprintf(stderr, _("Usage: %s [-c] [-v0|-v1] [-pPAGESZ] [-L label] [-U UUID] /dev/name [blocks]\n"), program_name); exit(1); } static void die(const char *str) { fprintf(stderr, "%s: %s\n", program_name, str); exit(1); } static void page_ok(int page) { if (version==0) bit_set(signature_page, page); } static void page_bad(int page) { if (version == 0) bit_test_and_clear(signature_page, page); else { if (badpages == MAX_BADPAGES) die(_("too many bad pages")); p->badpages[badpages] = page; } badpages++; } static void check_blocks(void) { unsigned int current_page; int do_seek = 1; char *buffer; buffer = malloc(pagesize); if (!buffer) die(_("Out of memory")); current_page = 0; while (current_page < PAGES) { if (!check) { page_ok(current_page++); continue; } if (do_seek && lseek(DEV,current_page*pagesize,SEEK_SET) != current_page*pagesize) die(_("seek failed in check_blocks")); if ((do_seek = (pagesize != read(DEV, buffer, pagesize)))) { page_bad(current_page++); continue; } page_ok(current_page++); } if (badpages == 1) printf(_("one bad page\n")); else if (badpages > 1) printf(_("%lu bad pages\n"), badpages); } /* return size in pages, to avoid integer overflow */ static unsigned long get_size(const char *file) { int fd; unsigned long long size; fd = open(file, O_RDONLY); if (fd < 0) { perror(file); exit(1); } if (blkdev_get_size(fd, &size) == 0) size /= pagesize; else size = blkdev_find_size(fd) / pagesize; close(fd); return size; } static int isnzdigit(char c) { return (c >= '1' && c <= '9'); } /* * Check to make certain that our new filesystem won't be created on * an already mounted partition. Code adapted from mke2fs, Copyright * (C) 1994 Theodore Ts'o. Also licensed under GPL. * (C) 2006 Karel Zak -- port to mkswap */ static int check_mount(void) { FILE * f; struct mntent * mnt; if ((f = setmntent (MOUNTED, "r")) == NULL) return 0; while ((mnt = getmntent (f)) != NULL) if (strcmp (device_name, mnt->mnt_fsname) == 0) break; endmntent (f); if (!mnt) return 0; return 1; } static int write_all(int fd, const void *buf, size_t count) { while(count) { ssize_t tmp; errno = 0; tmp = write(fd, buf, count); if (tmp > 0) { count -= tmp; if (count) buf += tmp; } else if (errno != EINTR && errno != EAGAIN) return -1; } return 0; } int main(int argc, char ** argv) { struct stat statbuf; int i; unsigned long maxpages; unsigned long goodpages; unsigned long sz; off_t offset; int force = 0; char *block_count = 0; char *pp; char *opt_label = NULL; unsigned char *uuid = NULL; #ifdef HAVE_LIBUUID const char *opt_uuid = NULL; uuid_t uuid_dat; #endif program_name = (argc && *argv) ? argv[0] : "mkswap"; if ((pp = strrchr(program_name, '/')) != NULL) program_name = pp+1; setlocale(LC_ALL, ""); bindtextdomain(PACKAGE, LOCALEDIR); textdomain(PACKAGE); if (argc == 2 && (!strcmp(argv[1], "-V") || !strcmp(argv[1], "--version"))) { printf(_("%s (%s)\n"), program_name, PACKAGE_STRING); exit(0); } for (i=1; i sz && !force) { fprintf(stderr, _("%s: error: " "size %lu KiB is larger than device size %lu KiB\n"), program_name, PAGES*(pagesize/1024), sz*(pagesize/1024)); exit(1); } if (version == -1) { /* labels only for v1 */ if (opt_label) version = 1; else /* use version 1 as default, if possible */ #ifdef __linux__ if (get_linux_version() < KERNEL_VERSION(2,1,117)) version = 0; else #endif if (pagesize < 2048) version = 0; else version = 1; } if (version != 0 && version != 1) { fprintf(stderr, _("%s: error: unknown version %d\n"), program_name, version); usage(); } if (PAGES < 10) { fprintf(stderr, _("%s: error: swap area needs to be at least %ld KiB\n"), program_name, (long)(10 * pagesize/1024)); usage(); } if (version == 0) maxpages = V0_MAX_PAGES; #ifdef __linux__ else if (get_linux_version() >= KERNEL_VERSION(2,3,4)) maxpages = PAGES; else if (get_linux_version() >= KERNEL_VERSION(2,2,1)) maxpages = V1_MAX_PAGES; #endif else maxpages = V1_OLD_MAX_PAGES; if (PAGES > maxpages) { PAGES = maxpages; fprintf(stderr, _("%s: warning: truncating swap area to %ld KiB\n"), program_name, PAGES * pagesize / 1024); } if (opt_label && version == 0) { fprintf(stderr, _("%s: error: label only with v1 swap area\n"), program_name); usage(); } if (stat(device_name, &statbuf) < 0) { perror(device_name); exit(EXIT_FAILURE); } if (S_ISBLK(statbuf.st_mode)) DEV = open(device_name, O_RDWR | O_EXCL); else DEV = open(device_name, O_RDWR); if (DEV < 0) { perror(device_name); exit(1); } /* Want a block device. Probably not /dev/hda or /dev/hdb. */ if (!S_ISBLK(statbuf.st_mode)) check=0; else if (statbuf.st_rdev == 0x0300 || statbuf.st_rdev == 0x0340) { fprintf(stderr, _("%s: error: " "will not try to make swapdevice on '%s'\n"), program_name, device_name); exit(1); } else if (check_mount()) { fprintf(stderr, _("%s: error: " "%s is mounted; will not make swapspace.\n"), program_name, device_name); exit(1); } #ifdef __sparc__ if (!force && version == 0) { /* Don't overwrite partition table unless forced */ unsigned char *buffer = (unsigned char *)signature_page; unsigned short *q, sum; if (read(DEV, buffer, 512) != 512) die(_("fatal: first page unreadable")); if (buffer[508] == 0xDA && buffer[509] == 0xBE) { q = (unsigned short *)(buffer + 510); for (sum = 0; q >= (unsigned short *) buffer;) sum ^= *q--; if (!sum) { fprintf(stderr, _("\ %s: Device '%s' contains a valid Sun disklabel.\n\ This probably means creating v0 swap would destroy your partition table.\n\ No swap was created. If you really want to create v0 swap on that device,\n\ use the -f option to force it.\n"), program_name, device_name); exit(1); } } } #endif if (version == 0 || check) check_blocks(); if (version == 0 && !bit_test_and_clear(signature_page,0)) die(_("fatal: first page unreadable")); if (version == 1) { p->version = version; p->last_page = PAGES-1; p->nr_badpages = badpages; } goodpages = PAGES - badpages - 1; if ((long) goodpages <= 0) die(_("Unable to set up swap-space: unreadable")); printf(_("Setting up swapspace version %d, size = %llu KiB\n"), version, (unsigned long long)goodpages * pagesize / 1024); write_signature((version == 0) ? "SWAP-SPACE" : "SWAPSPACE2"); if (version == 1) write_uuid_and_label(uuid, opt_label); offset = ((version == 0) ? 0 : 1024); if (lseek(DEV, offset, SEEK_SET) != offset) die(_("unable to rewind swap-device")); if (write_all(DEV, (char *) signature_page + offset, pagesize - offset) == -1) { fprintf(stderr, _("%s: %s: unable to write signature page: %s"), program_name, device_name, strerror(errno)); exit(1); } /* * A subsequent swapon() will fail if the signature * is not actually on disk. (This is a kernel bug.) */ #ifdef HAVE_FSYNC if (fsync(DEV)) die(_("fsync failed")); #endif #ifdef HAVE_LIBSELINUX if (S_ISREG(statbuf.st_mode) && is_selinux_enabled() > 0) { security_context_t context_string; security_context_t oldcontext; context_t newcontext; if (fgetfilecon(DEV, &oldcontext) < 0) { if (errno != ENODATA) { fprintf(stderr, _("%s: %s: unable to obtain selinux file label: %s\n"), program_name, device_name, strerror(errno)); exit(1); } if (matchpathcon(device_name, statbuf.st_mode, &oldcontext)) die(_("unable to matchpathcon()")); } if (!(newcontext = context_new(oldcontext))) die(_("unable to create new selinux context")); if (context_type_set(newcontext, SELINUX_SWAPFILE_TYPE)) die(_("couldn't compute selinux context")); context_string = context_str(newcontext); if (strcmp(context_string, oldcontext)!=0) { if (fsetfilecon(DEV, context_string)) { fprintf(stderr, _("%s: unable to relabel %s to %s: %s\n"), program_name, device_name, context_string, strerror(errno)); exit(1); } } context_free(newcontext); freecon(oldcontext); } #endif return 0; }