util-linux/lib/loopdev.c

/*
 * No copyright is claimed.  This code is in the public domain; do with
 * it what you wish.
 *
 * Written by Karel Zak <kzak@redhat.com>
 *
 * -- based on mount/losetup.c
 *
 * Simple library for work with loop devices.
 *
 *  - requires kernel 2.6.x
 *  - reads info from /sys/block/loop<N>/loop/<attr> (new kernels)
 *  - reads info by ioctl
 *  - supports *unlimited* number of loop devices
 *  - supports /dev/loop<N> as well as /dev/loop/<N>
 *  - minimize overhead (fd, loopinfo, ... are shared for all operations)
 *  - setup (associate device and backing file)
 *  - delete (dis-associate file)
 *  - old LOOP_{SET,GET}_STATUS (32bit) ioctls are unsupported
 *  - extendible
 */
#include <stdio.h>
#include <stdint.h>
#include <string.h>
#include <ctype.h>
#include <fcntl.h>
#include <stdlib.h>
#include <unistd.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/mman.h>
#include <inttypes.h>
#include <dirent.h>

#include "linux_version.h"
#include "c.h"
#include "sysfs.h"
#include "pathnames.h"
#include "loopdev.h"
#include "canonicalize.h"
#include "blkdev.h"
#include "debug.h"

/*
 * Debug stuff (based on include/debug.h)
 */
static UL_DEBUG_DEFINE_MASK(loopdev);
UL_DEBUG_DEFINE_MASKNAMES(loopdev) = UL_DEBUG_EMPTY_MASKNAMES;

#define LOOPDEV_DEBUG_INIT	(1 << 1)
#define LOOPDEV_DEBUG_CXT	(1 << 2)
#define LOOPDEV_DEBUG_ITER	(1 << 3)
#define LOOPDEV_DEBUG_SETUP	(1 << 4)

#define DBG(m, x)       __UL_DBG(loopdev, LOOPDEV_DEBUG_, m, x)
#define ON_DBG(m, x)    __UL_DBG_CALL(loopdev, LOOPDEV_DEBUG_, m, x)

#define UL_DEBUG_CURRENT_MASK	UL_DEBUG_MASK(loopdev)
#include "debugobj.h"

static void loopdev_init_debug(void)
{
	if (loopdev_debug_mask)
		return;
	__UL_INIT_DEBUG_FROM_ENV(loopdev, LOOPDEV_DEBUG_, 0, LOOPDEV_DEBUG);
}

/*
 * see loopcxt_init()
 */
#define loopcxt_ioctl_enabled(_lc)	(!((_lc)->flags & LOOPDEV_FL_NOIOCTL))
#define loopcxt_sysfs_available(_lc)	(!((_lc)->flags & LOOPDEV_FL_NOSYSFS)) \
					 && !loopcxt_ioctl_enabled(_lc)

/*
 * @lc: context
 * @device: device name, absolute device path or NULL to reset the current setting
 *
 * Sets device, absolute paths (e.g. "/dev/loop<N>") are unchanged, device
 * names ("loop<N>") are converted to the path (/dev/loop<N> or to
 * /dev/loop/<N>)
 *
 * This sets the device name, but does not check if the device exists!
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_set_device(struct loopdev_cxt *lc, const char *device)
{
	if (!lc)
		return -EINVAL;

	if (lc->fd >= 0) {
		close(lc->fd);
		DBG(CXT, ul_debugobj(lc, "closing old open fd"));
	}
	lc->fd = -1;
	lc->mode = 0;
	lc->blocksize = 0;
	lc->has_info = 0;
	lc->info_failed = 0;
	*lc->device = '\0';
	memset(&lc->config, 0, sizeof(lc->config));

	/* set new */
	if (device) {
		if (*device != '/') {
			const char *dir = _PATH_DEV;

			/* compose device name for /dev/loop<n> or /dev/loop/<n> */
			if (lc->flags & LOOPDEV_FL_DEVSUBDIR) {
				if (strlen(device) < 5)
					return -1;
				device += 4;
				dir = _PATH_DEV_LOOP "/";	/* _PATH_DEV uses tailing slash */
			}
			snprintf(lc->device, sizeof(lc->device), "%s%s",
				dir, device);
		} else
			xstrncpy(lc->device, device, sizeof(lc->device));

		DBG(CXT, ul_debugobj(lc, "%s name assigned", device));
	}

	ul_unref_path(lc->sysfs);
	lc->sysfs = NULL;
	return 0;
}

int loopcxt_has_device(struct loopdev_cxt *lc)
{
	return lc && *lc->device;
}

/*
 * @lc: context
 * @flags: LOOPDEV_FL_* flags
 *
 * Initialize loop handler.
 *
 * We have two sets of the flags:
 *
 *	* LOOPDEV_FL_* flags control loopcxt_* API behavior
 *
 *	* LO_FLAGS_* are kernel flags used for LOOP_{SET,GET}_STAT64 ioctls
 *
 * Note about LOOPDEV_FL_{RDONLY,RDWR} flags. These flags are used for open(2)
 * syscall to open loop device. By default is the device open read-only.
 *
 * The exception is loopcxt_setup_device(), where the device is open read-write
 * if LO_FLAGS_READ_ONLY flags is not set (see loopcxt_set_flags()).
 *
 * Returns: <0 on error, 0 on success.
 */
int loopcxt_init(struct loopdev_cxt *lc, int flags)
{
	int rc;
	struct stat st;
	struct loopdev_cxt dummy = UL_LOOPDEVCXT_EMPTY;

	if (!lc)
		return -EINVAL;

	loopdev_init_debug();
	DBG(CXT, ul_debugobj(lc, "initialize context"));

	memcpy(lc, &dummy, sizeof(dummy));
	lc->flags = flags;

	rc = loopcxt_set_device(lc, NULL);
	if (rc)
		return rc;

	if (stat(_PATH_SYS_BLOCK, &st) || !S_ISDIR(st.st_mode)) {
		lc->flags |= LOOPDEV_FL_NOSYSFS;
		lc->flags &= ~LOOPDEV_FL_NOIOCTL;
		DBG(CXT, ul_debugobj(lc, "init: disable /sys usage"));
	}

	if (!(lc->flags & LOOPDEV_FL_NOSYSFS) &&
	    get_linux_version() >= KERNEL_VERSION(2,6,37)) {
		/*
		 * Use only sysfs for basic information about loop devices
		 */
		lc->flags |= LOOPDEV_FL_NOIOCTL;
		DBG(CXT, ul_debugobj(lc, "init: ignore ioctls"));
	}

	if (!(lc->flags & LOOPDEV_FL_CONTROL) && !stat(_PATH_DEV_LOOPCTL, &st)) {
		lc->flags |= LOOPDEV_FL_CONTROL;
		DBG(CXT, ul_debugobj(lc, "init: loop-control detected "));
	}

	return 0;
}

/*
 * @lc: context
 *
 * Deinitialize loop context
 */
void loopcxt_deinit(struct loopdev_cxt *lc)
{
	int errsv = errno;

	if (!lc)
		return;

	DBG(CXT, ul_debugobj(lc, "de-initialize"));

	free(lc->filename);
	lc->filename = NULL;

	ignore_result( loopcxt_set_device(lc, NULL) );
	loopcxt_deinit_iterator(lc);

	errno = errsv;
}

/*
 * @lc: context
 *
 * Returns newly allocated device path.
 */
char *loopcxt_strdup_device(struct loopdev_cxt *lc)
{
	if (!lc || !*lc->device)
		return NULL;
	return strdup(lc->device);
}

/*
 * @lc: context
 *
 * Returns pointer device name in the @lc struct.
 */
const char *loopcxt_get_device(struct loopdev_cxt *lc)
{
	return lc && *lc->device ? lc->device : NULL;
}

/*
 * @lc: context
 *
 * Returns pointer to the sysfs context (see lib/sysfs.c)
 */
static struct path_cxt *loopcxt_get_sysfs(struct loopdev_cxt *lc)
{
	if (!lc || !*lc->device || (lc->flags & LOOPDEV_FL_NOSYSFS))
		return NULL;

	if (!lc->sysfs) {
		dev_t devno = sysfs_devname_to_devno(lc->device);
		if (!devno) {
			DBG(CXT, ul_debugobj(lc, "sysfs: failed devname to devno"));
			return NULL;
		}

		lc->sysfs = ul_new_sysfs_path(devno, NULL, NULL);
		if (!lc->sysfs)
			DBG(CXT, ul_debugobj(lc, "sysfs: init failed"));
	}

	return lc->sysfs;
}

/*
 * @lc: context
 *
 * Returns: file descriptor to the open loop device or <0 on error. The mode
 *          depends on LOOPDEV_FL_{RDWR,RDONLY} context flags. Default is
 *          read-only.
 */
int loopcxt_get_fd(struct loopdev_cxt *lc)
{
	if (!lc || !*lc->device)
		return -EINVAL;

	if (lc->fd < 0) {
		lc->mode = lc->flags & LOOPDEV_FL_RDWR ? O_RDWR : O_RDONLY;
		lc->fd = open(lc->device, lc->mode | O_CLOEXEC);
		DBG(CXT, ul_debugobj(lc, "open %s [%s]: %m", lc->device,
				lc->flags & LOOPDEV_FL_RDWR ? "rw" : "ro"));
	}
	return lc->fd;
}

int loopcxt_set_fd(struct loopdev_cxt *lc, int fd, int mode)
{
	if (!lc)
		return -EINVAL;

	lc->fd = fd;
	lc->mode = mode;
	return 0;
}

/*
 * @lc: context
 * @flags: LOOPITER_FL_* flags
 *
 * Iterator can be used to scan list of the free or used loop devices.
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_init_iterator(struct loopdev_cxt *lc, int flags)
{
	struct loopdev_iter *iter;
	struct stat st;

	if (!lc)
		return -EINVAL;


	iter = &lc->iter;
	DBG(ITER, ul_debugobj(iter, "initialize"));

	/* always zeroize
	 */
	memset(iter, 0, sizeof(*iter));
	iter->ncur = -1;
	iter->flags = flags;
	iter->default_check = 1;

	if (!lc->extra_check) {
		/*
		 * Check for /dev/loop/<N> subdirectory
		 */
		if (!(lc->flags & LOOPDEV_FL_DEVSUBDIR) &&
		    stat(_PATH_DEV_LOOP, &st) == 0 && S_ISDIR(st.st_mode))
			lc->flags |= LOOPDEV_FL_DEVSUBDIR;

		lc->extra_check = 1;
	}
	return 0;
}

/*
 * @lc: context
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_deinit_iterator(struct loopdev_cxt *lc)
{
	struct loopdev_iter *iter;

	if (!lc)
		return -EINVAL;

	iter = &lc->iter;
	DBG(ITER, ul_debugobj(iter, "de-initialize"));

	free(iter->minors);
	if (iter->proc)
		fclose(iter->proc);
	if (iter->sysblock)
		closedir(iter->sysblock);

	memset(iter, 0, sizeof(*iter));
	return 0;
}

/*
 * Same as loopcxt_set_device, but also checks if the device is
 * associated with any file.
 *
 * Returns: <0 on error, 0 on success, 1 device does not match with
 *         LOOPITER_FL_{USED,FREE} flags.
 */
static int loopiter_set_device(struct loopdev_cxt *lc, const char *device)
{
	int rc = loopcxt_set_device(lc, device);
	int used;

	if (rc)
		return rc;

	if (!(lc->iter.flags & LOOPITER_FL_USED) &&
	    !(lc->iter.flags & LOOPITER_FL_FREE))
		return 0;	/* caller does not care about device status */

	if (!is_loopdev(lc->device)) {
		DBG(ITER, ul_debugobj(&lc->iter, "%s does not exist", lc->device));
		return -errno;
	}

	DBG(ITER, ul_debugobj(&lc->iter, "%s exist", lc->device));

	used = loopcxt_get_offset(lc, NULL) == 0;

	if ((lc->iter.flags & LOOPITER_FL_USED) && used)
		return 0;

	if ((lc->iter.flags & LOOPITER_FL_FREE) && !used)
		return 0;

	DBG(ITER, ul_debugobj(&lc->iter, "failed to use %s device", lc->device));

	ignore_result( loopcxt_set_device(lc, NULL) );
	return 1;
}

static int cmpnum(const void *p1, const void *p2)
{
	return (((* (const int *) p1) > (* (const int *) p2)) -
			((* (const int *) p1) < (* (const int *) p2)));
}

/*
 * The classic scandir() is more expensive and less portable.
 * We needn't full loop device names -- loop numbers (loop<N>)
 * are enough.
 */
static int loop_scandir(const char *dirname, int **ary, int hasprefix)
{
	DIR *dir;
	struct dirent *d;
	unsigned int n, count = 0, arylen = 0;

	if (!dirname || !ary)
		return 0;

	DBG(ITER, ul_debug("scan dir: %s", dirname));

	dir = opendir(dirname);
	if (!dir)
		return 0;
	free(*ary);
	*ary = NULL;

	while((d = readdir(dir))) {
#ifdef _DIRENT_HAVE_D_TYPE
		if (d->d_type != DT_BLK && d->d_type != DT_UNKNOWN &&
		    d->d_type != DT_LNK)
			continue;
#endif
		if (!strcmp(d->d_name, ".") || !strcmp(d->d_name, ".."))
			continue;

		if (hasprefix) {
			/* /dev/loop<N> */
			if (sscanf(d->d_name, "loop%u", &n) != 1)
				continue;
		} else {
			/* /dev/loop/<N> */
			char *end = NULL;

			errno = 0;
			n = strtol(d->d_name, &end, 10);
			if (d->d_name == end || (end && *end) || errno)
				continue;
		}
		if (n < LOOPDEV_DEFAULT_NNODES)
			continue;			/* ignore loop<0..7> */

		if (count + 1 > arylen) {
			int *tmp;

			arylen += 1;

			tmp = realloc(*ary, arylen * sizeof(int));
			if (!tmp) {
				free(*ary);
				*ary = NULL;
				closedir(dir);
				return -1;
			}
			*ary = tmp;
		}
		if (*ary)
			(*ary)[count++] = n;
	}
	if (count && *ary)
		qsort(*ary, count, sizeof(int), cmpnum);

	closedir(dir);
	return count;
}

/*
 * Set the next *used* loop device according to /proc/partitions.
 *
 * Loop devices smaller than 512 bytes are invisible for this function.
 */
static int loopcxt_next_from_proc(struct loopdev_cxt *lc)
{
	struct loopdev_iter *iter = &lc->iter;
	char buf[BUFSIZ];

	DBG(ITER, ul_debugobj(iter, "scan /proc/partitions"));

	if (!iter->proc)
		iter->proc = fopen(_PATH_PROC_PARTITIONS, "r" UL_CLOEXECSTR);
	if (!iter->proc)
		return 1;

	while (fgets(buf, sizeof(buf), iter->proc)) {
		unsigned int m;
		char name[128 + 1];


		if (sscanf(buf, " %u %*s %*s %128[^\n ]",
			   &m, name) != 2 || m != LOOPDEV_MAJOR)
			continue;

		DBG(ITER, ul_debugobj(iter, "checking %s", name));

		if (loopiter_set_device(lc, name) == 0)
			return 0;
	}

	return 1;
}

/*
 * Set the next *used* loop device according to
 * /sys/block/loopN/loop/backing_file (kernel >= 2.6.37 is required).
 *
 * This is preferred method.
 */
static int loopcxt_next_from_sysfs(struct loopdev_cxt *lc)
{
	struct loopdev_iter *iter = &lc->iter;
	struct dirent *d;
	int fd;

	DBG(ITER, ul_debugobj(iter, "scanning /sys/block"));

	if (!iter->sysblock)
		iter->sysblock = opendir(_PATH_SYS_BLOCK);

	if (!iter->sysblock)
		return 1;

	fd = dirfd(iter->sysblock);

	while ((d = readdir(iter->sysblock))) {
		char name[NAME_MAX + 18 + 1];
		struct stat st;

		DBG(ITER, ul_debugobj(iter, "check %s", d->d_name));

		if (strcmp(d->d_name, ".") == 0
		    || strcmp(d->d_name, "..") == 0
		    || strncmp(d->d_name, "loop", 4) != 0)
			continue;

		snprintf(name, sizeof(name), "%s/loop/backing_file", d->d_name);
		if (fstatat(fd, name, &st, 0) != 0)
			continue;

		if (loopiter_set_device(lc, d->d_name) == 0)
			return 0;
	}

	return 1;
}

/*
 * @lc: context, has to initialized by loopcxt_init_iterator()
 *
 * Returns: 0 on success, -1 on error, 1 at the end of scanning. The details
 *          about the current loop device are available by
 *          loopcxt_get_{fd,backing_file,device,offset, ...} functions.
 */
int loopcxt_next(struct loopdev_cxt *lc)
{
	struct loopdev_iter *iter;

	if (!lc)
		return -EINVAL;


	iter = &lc->iter;
	if (iter->done)
		return 1;

	DBG(ITER, ul_debugobj(iter, "next"));

	/* A) Look for used loop devices in /proc/partitions ("losetup -a" only)
	 */
	if (iter->flags & LOOPITER_FL_USED) {
		int rc;

		if (loopcxt_sysfs_available(lc))
			rc = loopcxt_next_from_sysfs(lc);
		else
			rc = loopcxt_next_from_proc(lc);
		if (rc == 0)
			return 0;
		goto done;
	}

	/* B) Classic way, try first eight loop devices (default number
	 *    of loop devices). This is enough for 99% of all cases.
	 */
	if (iter->default_check) {
		DBG(ITER, ul_debugobj(iter, "next: default check"));
		for (++iter->ncur; iter->ncur < LOOPDEV_DEFAULT_NNODES;
							iter->ncur++) {
			char name[16];
			snprintf(name, sizeof(name), "loop%d", iter->ncur);

			if (loopiter_set_device(lc, name) == 0)
				return 0;
		}
		iter->default_check = 0;
	}

	/* C) the worst possibility, scan whole /dev or /dev/loop/<N>
	 */
	if (!iter->minors) {
		DBG(ITER, ul_debugobj(iter, "next: scanning /dev"));
		iter->nminors = (lc->flags & LOOPDEV_FL_DEVSUBDIR) ?
			loop_scandir(_PATH_DEV_LOOP, &iter->minors, 0) :
			loop_scandir(_PATH_DEV, &iter->minors, 1);
		iter->ncur = -1;
	}
	for (++iter->ncur; iter->ncur < iter->nminors; iter->ncur++) {
		char name[16];
		snprintf(name, sizeof(name), "loop%d", iter->minors[iter->ncur]);

		if (loopiter_set_device(lc, name) == 0)
			return 0;
	}
done:
	loopcxt_deinit_iterator(lc);
	return 1;
}

/*
 * @device: path to device
 */
int is_loopdev(const char *device)
{
	struct stat st;

	if (device && stat(device, &st) == 0 &&
		S_ISBLK(st.st_mode) &&
		major(st.st_rdev) == LOOPDEV_MAJOR)
		return 1;

	errno = ENODEV;
	return 0;
}

/*
 * @lc: context
 *
 * Returns result from LOOP_GET_STAT64 ioctl or NULL on error.
 */
struct loop_info64 *loopcxt_get_info(struct loopdev_cxt *lc)
{
	int fd;

	if (!lc || lc->info_failed) {
		errno = EINVAL;
		return NULL;
	}
	errno = 0;
	if (lc->has_info)
		return &lc->config.info;

	fd = loopcxt_get_fd(lc);
	if (fd < 0)
		return NULL;

	if (ioctl(fd, LOOP_GET_STATUS64, &lc->config.info) == 0) {
		lc->has_info = 1;
		lc->info_failed = 0;
		DBG(CXT, ul_debugobj(lc, "reading loop_info64 OK"));
		return &lc->config.info;
	}

	lc->info_failed = 1;
	DBG(CXT, ul_debugobj(lc, "reading loop_info64 FAILED"));

	return NULL;
}

/*
 * @lc: context
 *
 * Returns (allocated) string with path to the file associated
 * with the current loop device.
 */
char *loopcxt_get_backing_file(struct loopdev_cxt *lc)
{
	struct path_cxt *sysfs = loopcxt_get_sysfs(lc);
	char *res = NULL;

	if (sysfs)
		/*
		 * This is always preferred, the loop_info64
		 * has too small buffer for the filename.
		 */
		ul_path_read_string(sysfs, &res, "loop/backing_file");

	if (!res && loopcxt_ioctl_enabled(lc)) {
		struct loop_info64 *lo = loopcxt_get_info(lc);

		if (lo) {
			lo->lo_file_name[LO_NAME_SIZE - 2] = '*';
			lo->lo_file_name[LO_NAME_SIZE - 1] = '\0';
			res = strdup((char *) lo->lo_file_name);
		}
	}

	DBG(CXT, ul_debugobj(lc, "get_backing_file [%s]", res));
	return res;
}

/*
 * @lc: context
 * @offset: returns offset number for the given device
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_get_offset(struct loopdev_cxt *lc, uint64_t *offset)
{
	struct path_cxt *sysfs = loopcxt_get_sysfs(lc);
	int rc = -EINVAL;

	if (sysfs)
		rc = ul_path_read_u64(sysfs, offset, "loop/offset");

	if (rc && loopcxt_ioctl_enabled(lc)) {
		struct loop_info64 *lo = loopcxt_get_info(lc);
		if (lo) {
			if (offset)
				*offset = lo->lo_offset;
			rc = 0;
		} else
			rc = -errno;
	}

	DBG(CXT, ul_debugobj(lc, "get_offset [rc=%d]", rc));
	return rc;
}

/*
 * @lc: context
 * @blocksize: returns logical blocksize for the given device
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_get_blocksize(struct loopdev_cxt *lc, uint64_t *blocksize)
{
	struct path_cxt *sysfs = loopcxt_get_sysfs(lc);
	int rc = -EINVAL;

	if (sysfs)
		rc = ul_path_read_u64(sysfs, blocksize, "queue/logical_block_size");

	/* Fallback based on BLKSSZGET ioctl */
	if (rc) {
		int fd = loopcxt_get_fd(lc);
		int sz = 0;

		if (fd < 0)
			return -EINVAL;
		rc = blkdev_get_sector_size(fd, &sz);
		if (rc)
			return rc;

		*blocksize = sz;
	}

	DBG(CXT, ul_debugobj(lc, "get_blocksize [rc=%d]", rc));
	return rc;
}

/*
 * @lc: context
 * @sizelimit: returns size limit for the given device
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_get_sizelimit(struct loopdev_cxt *lc, uint64_t *size)
{
	struct path_cxt *sysfs = loopcxt_get_sysfs(lc);
	int rc = -EINVAL;

	if (sysfs)
		rc = ul_path_read_u64(sysfs, size, "loop/sizelimit");

	if (rc && loopcxt_ioctl_enabled(lc)) {
		struct loop_info64 *lo = loopcxt_get_info(lc);
		if (lo) {
			if (size)
				*size = lo->lo_sizelimit;
			rc = 0;
		} else
			rc = -errno;
	}

	DBG(CXT, ul_debugobj(lc, "get_sizelimit [rc=%d]", rc));
	return rc;
}

/*
 * @lc: context
 * @devno: returns encryption type
 *
 * Cryptoloop is DEPRECATED!
 *
 * Returns: <0 on error, 0 on success
 */
int loopcxt_get_encrypt_type(struct loopdev_cxt *lc, uint32_t *type)
{
	struct loop_info64 *lo = loopcxt_get_info(lc);
	int rc;

	/* not provided by sysfs */