X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Futils%2Fmount_utils.c;h=9d8e232f66daa371e2d0493128505cdb923247d0;hp=048aee565a4d3b50ec23d34234060db7d0246f97;hb=85cbe1a3ee6940f7468884bca43cd736a5365694;hpb=71bc053baa961aa3dbbaa5ae89f9cd4fb5a080e7

diff --git a/lustre/utils/mount_utils.c b/lustre/utils/mount_utils.c
index 048aee5..9d8e232 100644
--- a/lustre/utils/mount_utils.c
+++ b/lustre/utils/mount_utils.c
@@ -26,7 +26,8 @@
 /*
  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
- * Copyright (c) 2012 Whamcloud, Inc.
+ *
+ * Copyright (c) 2012, 2014, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -37,15 +38,20 @@
 #  include "config.h"
 #endif /* HAVE_CONFIG_H */
 
+#include "mount_utils.h"
+#include <mntent.h>
 #include <stdio.h>
 #include <errno.h>
 #include <string.h>
+#include <unistd.h>
 #include <config.h>
 #include <lustre_disk.h>
 #include <lustre_ver.h>
+#include <sys/mount.h>
 #include <sys/stat.h>
 #include <sys/utsname.h>
-#include "mount_utils.h"
+#include <linux/loop.h>
+#include <dlfcn.h>
 
 extern char *progname;
 extern int verbose;
@@ -53,6 +59,8 @@ extern int verbose;
 #define vprint(fmt, arg...) if (verbose > 0) printf(fmt, ##arg)
 #define verrprint(fmt, arg...) if (verbose >= 0) fprintf(stderr, fmt, ##arg)
 
+static struct module_backfs_ops *backfs_ops[LDD_MT_LAST];
+
 void fatal(void)
 {
         verbose = 0;
@@ -140,6 +148,43 @@ int get_param(char *buf, char *key, char **val)
 	return ENOENT;
 }
 
+int append_param(char *buf, char *key, char *val, char sep)
+{
+	int key_len, i, offset, old_val_len;
+	char *ptr = NULL, str[1024];
+
+	if (key)
+		ptr = strstr(buf, key);
+
+	/* key doesn't exist yet, so just add it */
+	if (ptr == NULL)
+		return add_param(buf, key, val);
+
+	key_len = strlen(key);
+
+	/* Copy previous values to str */
+	for (i = 0; i < sizeof(str); ++i) {
+		if ((ptr[i+key_len] == ' ') || (ptr[i+key_len] == '\0'))
+			break;
+		str[i] = ptr[i+key_len];
+	}
+	if (i == sizeof(str))
+		return E2BIG;
+	old_val_len = i;
+
+	offset = old_val_len+key_len;
+
+	/* Move rest of buf to overwrite previous key and value */
+	for (i = 0; ptr[i+offset] != '\0'; ++i)
+		ptr[i] = ptr[i+offset];
+
+	ptr[i] = '\0';
+
+	snprintf(str+old_val_len, sizeof(str)-old_val_len, "%c%s", sep, val);
+
+	return add_param(buf, key, str);
+}
+
 char *strscat(char *dst, char *src, int buflen)
 {
 	dst[buflen - 1] = 0;
@@ -180,6 +225,20 @@ int check_mtab_entry(char *spec1, char *spec2, char *mtpt, char *type)
 	return 0;
 }
 
+#define PROC_DIR	"/proc/"
+static int mtab_is_proc(const char *mtab)
+{
+	char path[16];
+
+	if (readlink(mtab, path, sizeof(path)) < 0)
+		return 0;
+
+	if (strncmp(path, PROC_DIR, strlen(PROC_DIR)))
+		return 0;
+
+	return 1;
+}
+
 int update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
 		int flags, int freq, int pass)
 {
@@ -187,6 +246,10 @@ int update_mtab_entry(char *spec, char *mtpt, char *type, char *opts,
 	struct mntent mnt;
 	int rc = 0;
 
+	/* Don't update mtab if it is linked to any file in /proc direcotry.*/
+	if (mtab_is_proc(MOUNTED))
+		return 0;
+
 	mnt.mnt_fsname = spec;
 	mnt.mnt_dir = mtpt;
 	mnt.mnt_type = type;
@@ -235,8 +298,7 @@ static int in_mntlist(char *opt, char *mntlist)
  * present in mountopts.  The justwarn boolean toggles between error and
  * warning message.  Return an error count.
  */
-int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
-			 int justwarn)
+int check_mountfsoptions(char *mountopts, char *wanted_mountopts)
 {
 	char *ml, *mlp, *item, *ctx = NULL;
 	int errors = 0;
@@ -248,9 +310,8 @@ int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
 	mlp = ml;
 	while ((item = strtok_r(mlp, ",", &ctx))) {
 		if (!in_mntlist(item, mountopts)) {
-			fprintf(stderr, "%s: %s mount option `%s' is missing\n",
-				progname, justwarn ? "Warning: default"
-				: "Error: mandatory", item);
+			fprintf(stderr, "%s: Error: mandatory mount option"
+				" '%s' is missing\n", progname, item);
 			errors++;
 		}
 		mlp = NULL;
@@ -289,7 +350,8 @@ int loop_setup(struct mkfs_opts *mop)
 	int i, ret = 0;
 
 	/* Figure out the loop device names */
-	if (!access("/dev/loop0", F_OK | R_OK)) {
+	if (!access("/dev/loop0", F_OK | R_OK) ||
+	    !access("/dev/loop-control", F_OK | R_OK)) {
 		strcpy(loop_base, "/dev/loop\0");
 	} else if (!access("/dev/loop/0", F_OK | R_OK)) {
 		strcpy(loop_base, "/dev/loop/\0");
@@ -303,9 +365,24 @@ int loop_setup(struct mkfs_opts *mop)
 		char cmd[PATH_MAX];
 		int cmdsz = sizeof(cmd);
 
+#ifdef LOOP_CTL_GET_FREE
+		ret = open("/dev/loop-control", O_RDWR);
+		if (ret < 0) {
+			fprintf(stderr, "%s: can't access loop control\n", progname);
+			return EACCES;
+		}
+		/* find or allocate a free loop device to use */
+		i = ioctl(ret, LOOP_CTL_GET_FREE);
+		if (i < 0) {
+			fprintf(stderr, "%s: access loop control error\n", progname);
+			return EACCES;
+		}
+		sprintf(l_device, "%s%d", loop_base, i);
+#else
 		sprintf(l_device, "%s%d", loop_base, i);
 		if (access(l_device, F_OK | R_OK))
 			break;
+#endif
 		snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
 		ret = system(cmd);
 
@@ -321,7 +398,8 @@ int loop_setup(struct mkfs_opts *mop)
 				continue;
 			if (ret) {
 				fprintf(stderr, "%s: error %d on losetup: %s\n",
-					progname, ret, strerror(ret));
+					progname, ret,
+					ret >= 0 ? strerror(ret) : "");
 				return ret;
 			}
 			strscpy(mop->mo_loopdev, l_device,
@@ -337,11 +415,23 @@ int loop_setup(struct mkfs_opts *mop)
 int loop_cleanup(struct mkfs_opts *mop)
 {
 	char cmd[150];
-	int ret = 1;
+	int ret = 0;
+
 	if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
+		int tries;
+
 		sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
-		ret = run_command(cmd, sizeof(cmd));
+		for (tries = 0; tries < 3; tries++) {
+			ret = run_command(cmd, sizeof(cmd));
+			if (ret == 0)
+				break;
+			sleep(1);
+		}
 	}
+
+	if (ret != 0)
+		fprintf(stderr, "cannot cleanup %s: rc = %d\n",
+			mop->mo_loopdev, ret);
 	return ret;
 }
 
@@ -349,7 +439,7 @@ int loop_format(struct mkfs_opts *mop)
 {
 	int fd;
 
-	if (mop->mo_device_sz == 0) {
+	if (mop->mo_device_kb == 0) {
 		fatal();
 		fprintf(stderr, "loop device requires a --device-size= "
 			"param\n");
@@ -364,7 +454,7 @@ int loop_format(struct mkfs_opts *mop)
 		return errno;
 	}
 
-	if (ftruncate(fd, mop->mo_device_sz * 1024) != 0) {
+	if (ftruncate(fd, mop->mo_device_kb * 1024) != 0) {
 		close(fd);
 		fatal();
 		fprintf(stderr, "%s: Unable to truncate backing store: %s\n",
@@ -376,24 +466,133 @@ int loop_format(struct mkfs_opts *mop)
 	return 0;
 }
 
+#define DLSYM(prefix, sym, func)					\
+	do {								\
+		char _fname[64];					\
+		snprintf(_fname, sizeof(_fname), "%s_%s", prefix, #func); \
+		sym->func = (typeof(sym->func))dlsym(sym->dl_handle, _fname); \
+	} while (0)
+
+/**
+ * Load plugin for a given mount_type from ${pkglibdir}/mount_osd_FSTYPE.so and
+ * return struct of function pointers (will be freed in unloack_backfs_module).
+ *
+ * \param[in] mount_type	Mount type to load module for.
+ * \retval Value of backfs_ops struct
+ * \retval NULL if no module exists
+ */
+struct module_backfs_ops *load_backfs_module(enum ldd_mount_type mount_type)
+{
+	void *handle;
+	char *error, filename[512], fsname[512], *name;
+	struct module_backfs_ops *ops;
+
+	/* This deals with duplicate ldd_mount_types resolving to same OSD layer
+	 * plugin (e.g. ext3/ldiskfs/ldiskfs2 all being ldiskfs) */
+	strncpy(fsname, mt_type(mount_type), sizeof(fsname));
+	name = fsname + sizeof("osd-") - 1;
+
+	/* change osd- to osd_ */
+	fsname[sizeof("osd-") - 2] = '_';
+
+	snprintf(filename, sizeof(filename), PLUGIN_DIR"/mount_%s.so", fsname);
+
+	handle = dlopen(filename, RTLD_LAZY);
+
+	/* Check for $LUSTRE environment variable from test-framework.
+	 * This allows using locally built modules to be used.
+	 */
+	if (handle == NULL) {
+		char *dirname;
+		dirname = getenv("LUSTRE");
+		if (dirname) {
+			snprintf(filename, sizeof(filename),
+				 "%s/utils/.libs/mount_%s.so",
+				 dirname, fsname);
+			handle = dlopen(filename, RTLD_LAZY);
+		}
+	}
+
+	/* Do not clutter up console with missing types */
+	if (handle == NULL)
+		return NULL;
+
+	ops = malloc(sizeof(*ops));
+	if (ops == NULL) {
+		dlclose(handle);
+		return NULL;
+	}
+
+	ops->dl_handle = handle;
+	dlerror(); /* Clear any existing error */
+
+	DLSYM(name, ops, init);
+	DLSYM(name, ops, fini);
+	DLSYM(name, ops, read_ldd);
+	DLSYM(name, ops, write_ldd);
+	DLSYM(name, ops, is_lustre);
+	DLSYM(name, ops, make_lustre);
+	DLSYM(name, ops, prepare_lustre);
+	DLSYM(name, ops, tune_lustre);
+	DLSYM(name, ops, label_lustre);
+	DLSYM(name, ops, enable_quota);
+
+	error = dlerror();
+	if (error != NULL) {
+		fatal();
+		fprintf(stderr, "%s\n", error);
+		dlclose(handle);
+		free(ops);
+		return NULL;
+	}
+
+	/* optional methods */
+	DLSYM(name, ops, fix_mountopts);
+
+	return ops;
+}
+
+/**
+ * Unload plugin and free backfs_ops structure. Must be called the same number
+ * of times as load_backfs_module is.
+ */
+void unload_backfs_module(struct module_backfs_ops *ops)
+{
+	if (ops == NULL)
+		return;
+
+	dlclose(ops->dl_handle);
+	free(ops);
+}
+
+/* Return true if backfs_ops has operations for the given mount_type. */
+int backfs_mount_type_okay(enum ldd_mount_type mount_type)
+{
+	if (unlikely(mount_type >= LDD_MT_LAST || mount_type < 0)) {
+		fatal();
+		fprintf(stderr, "fs type out of range %d\n", mount_type);
+		return 0;
+	}
+	if (backfs_ops[mount_type] == NULL) {
+		fatal();
+		fprintf(stderr, "unhandled/unloaded fs type %d '%s'\n",
+			mount_type, mt_str(mount_type));
+		return 0;
+	}
+	return 1;
+}
+
 /* Write the server config files */
 int osd_write_ldd(struct mkfs_opts *mop)
 {
 	struct lustre_disk_data *ldd = &mop->mo_ldd;
 	int ret;
 
-	switch (ldd->ldd_mount_type) {
-	case LDD_MT_LDISKFS:
-	case LDD_MT_LDISKFS2:
-		ret = ldiskfs_write_ldd(mop);
-		break;
-	default:
-		fatal();
-		fprintf(stderr, "unknown fs type %d '%s'\n",
-			ldd->ldd_mount_type, MT_STR(ldd));
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->write_ldd(mop);
+
+	else
 		ret = EINVAL;
-		break;
-	}
 
 	return ret;
 }
@@ -403,18 +602,11 @@ int osd_read_ldd(char *dev, struct lustre_disk_data *ldd)
 {
 	int ret;
 
-	switch (ldd->ldd_mount_type) {
-	case LDD_MT_LDISKFS:
-	case LDD_MT_LDISKFS2:
-		ret = ldiskfs_read_ldd(dev, ldd);
-		break;
-	default:
-		fatal();
-		fprintf(stderr, "unknown fs type %d '%s'\n",
-			ldd->ldd_mount_type, MT_STR(ldd));
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->read_ldd(dev, ldd);
+
+	else
 		ret = EINVAL;
-		break;
-	}
 
 	return ret;
 }
@@ -422,11 +614,16 @@ int osd_read_ldd(char *dev, struct lustre_disk_data *ldd)
 /* Was this device formatted for Lustre */
 int osd_is_lustre(char *dev, unsigned *mount_type)
 {
+	int i;
+
 	vprint("checking for existing Lustre data: ");
 
-	if (ldiskfs_is_lustre(dev, mount_type)) {
-		vprint("found\n");
-		return 1;
+	for (i = 0; i < LDD_MT_LAST; ++i) {
+		if (backfs_ops[i] != NULL &&
+		    backfs_ops[i]->is_lustre(dev, mount_type)) {
+			vprint("found\n");
+			return 1;
+		}
 	}
 
 	vprint("not found\n");
@@ -439,61 +636,119 @@ int osd_make_lustre(struct mkfs_opts *mop)
 	struct lustre_disk_data *ldd = &mop->mo_ldd;
 	int ret;
 
-	switch (ldd->ldd_mount_type) {
-	case LDD_MT_LDISKFS:
-	case LDD_MT_LDISKFS2:
-		ret = ldiskfs_make_lustre(mop);
-		break;
-	default:
-		fatal();
-		fprintf(stderr, "unknown fs type %d '%s'\n",
-			ldd->ldd_mount_type, MT_STR(ldd));
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->make_lustre(mop);
+
+	else
 		ret = EINVAL;
-		break;
-	}
 
 	return ret;
 }
 
 int osd_prepare_lustre(struct mkfs_opts *mop,
-		char *default_mountopts, int default_len,
-		char *always_mountopts, int always_len)
+		       char *wanted_mountopts, size_t len)
 {
 	struct lustre_disk_data *ldd = &mop->mo_ldd;
 	int ret;
 
-	switch (ldd->ldd_mount_type) {
-	case LDD_MT_LDISKFS:
-	case LDD_MT_LDISKFS2:
-		ret = ldiskfs_prepare_lustre(mop,
-					     default_mountopts, default_len,
-					     always_mountopts, always_len);
-		break;
-	default:
-		fatal();
-		fprintf(stderr, "unknown fs type %d '%s'\n",
-			ldd->ldd_mount_type, MT_STR(ldd));
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->prepare_lustre(mop,
+							wanted_mountopts, len);
+
+	else
 		ret = EINVAL;
-		break;
-	}
 
 	return ret;
 }
 
-int osd_init(void)
+int osd_fix_mountopts(struct mkfs_opts *mop, char *mountopts, size_t len)
 {
-	int ret = 0;
+	struct lustre_disk_data *ldd = &mop->mo_ldd;
+
+	if (!backfs_mount_type_okay(ldd->ldd_mount_type))
+		return EINVAL;
+
+	if (backfs_ops[ldd->ldd_mount_type]->fix_mountopts == NULL)
+		return 0;
+
+	return backfs_ops[ldd->ldd_mount_type]->fix_mountopts(mop, mountopts,
+							      len);
+}
+
+int osd_tune_lustre(char *dev, struct mount_opts *mop)
+{
+	struct lustre_disk_data *ldd = &mop->mo_ldd;
+	int ret;
+
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->tune_lustre(dev, mop);
 
-	ret = ldiskfs_init();
-	if (ret)
-		return ret;
+	else
+		ret = EINVAL;
+
+	return ret;
+}
+
+int osd_label_lustre(struct mount_opts *mop)
+{
+	struct lustre_disk_data *ldd = &mop->mo_ldd;
+	int ret;
+
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->label_lustre(mop);
+
+	else
+		ret = EINVAL;
+
+	return ret;
+}
+
+/* Enable quota accounting */
+int osd_enable_quota(struct mkfs_opts *mop)
+{
+	struct lustre_disk_data *ldd = &mop->mo_ldd;
+	int ret;
+
+	if (backfs_mount_type_okay(ldd->ldd_mount_type))
+		ret = backfs_ops[ldd->ldd_mount_type]->enable_quota(mop);
+
+	else
+		ret = EINVAL;
+
+	return ret;
+}
+
+int osd_init(void)
+{
+	int i, rc, ret = EINVAL;
+
+	for (i = 0; i < LDD_MT_LAST; ++i) {
+		rc = 0;
+		backfs_ops[i] = load_backfs_module(i);
+		if (backfs_ops[i] != NULL)
+			rc = backfs_ops[i]->init();
+		if (rc != 0) {
+			backfs_ops[i]->fini();
+			unload_backfs_module(backfs_ops[i]);
+			backfs_ops[i] = NULL;
+		} else
+			ret = 0;
+	}
 
 	return ret;
 }
 
 void osd_fini(void)
 {
-	ldiskfs_fini();
+	int i;
+
+	for (i = 0; i < LDD_MT_LAST; ++i) {
+		if (backfs_ops[i] != NULL) {
+			backfs_ops[i]->fini();
+			unload_backfs_module(backfs_ops[i]);
+			backfs_ops[i] = NULL;
+		}
+	}
 }
 
 __u64 get_device_size(char* device)
@@ -531,11 +786,23 @@ __u64 get_device_size(char* device)
 	return size >> 10;
 }
 
-int file_create(char *path, int size)
+int file_create(char *path, __u64 size)
 {
+	__u64 size_max;
 	int ret;
 	int fd;
 
+	/*
+	 * Since "size" is in KB, the file offset it represents could overflow
+	 * off_t.
+	 */
+	size_max = (off_t)1 << (_FILE_OFFSET_BITS - 1 - 10);
+	if (size >= size_max) {
+		fprintf(stderr, "%s: "LPU64" KB: Backing store size must be "
+			"smaller than "LPU64" KB\n", progname, size, size_max);
+		return EFBIG;
+	}
+
 	ret = access(path, F_OK);
 	if (ret == 0) {
 		ret = unlink(path);