chiark / gitweb /
btrfs: add support for recursive btrfs snapshotting
[elogind.git] / src / shared / machine-image.c
index 117994d6d88c94e2d6884b2e2e6f8795b08e3783..0b41860b5d820b5f0240c5087763da7b56fcf09b 100644 (file)
 #include <linux/fs.h>
 #include <fcntl.h>
 
-#include "strv.h"
 #include "utf8.h"
 #include "btrfs-util.h"
 #include "path-util.h"
 #include "copy.h"
+#include "mkdir.h"
+#include "rm-rf.h"
 #include "machine-image.h"
 
 static const char image_search_path[] =
@@ -72,7 +73,7 @@ static int image_new(
         i->read_only = read_only;
         i->crtime = crtime;
         i->mtime = mtime;
-        i->size = i->size_exclusive = (uint64_t) -1;
+        i->usage = i->usage_exclusive = (uint64_t) -1;
         i->limit = i->limit_exclusive = (uint64_t) -1;
 
         i->name = strdup(pretty);
@@ -109,7 +110,7 @@ static int image_make(
         assert(filename);
 
         /* We explicitly *do* follow symlinks here, since we want to
-         * allow symlinking trees into /var/lib/container/, and treat
+         * allow symlinking trees into /var/lib/machines/, and treat
          * them normally. */
 
         if (fstatat(dfd, filename, &st, 0) < 0)
@@ -163,10 +164,10 @@ static int image_make(
 
                                 r = btrfs_subvol_get_quota_fd(fd, &quota);
                                 if (r >= 0) {
-                                        (*ret)->size = quota.referred;
-                                        (*ret)->size_exclusive = quota.exclusive;
+                                        (*ret)->usage = quota.referenced;
+                                        (*ret)->usage_exclusive = quota.exclusive;
 
-                                        (*ret)->limit = quota.referred_max;
+                                        (*ret)->limit = quota.referenced_max;
                                         (*ret)->limit_exclusive = quota.exclusive_max;
                                 }
 
@@ -193,10 +194,10 @@ static int image_make(
 
                 return 1;
 
-        } else if (S_ISREG(st.st_mode) && endswith(filename, ".gpt")) {
+        } else if (S_ISREG(st.st_mode) && endswith(filename, ".raw")) {
                 usec_t crtime = 0;
 
-                /* It's a GPT block device */
+                /* It's a RAW disk image */
 
                 if (!ret)
                         return 1;
@@ -206,7 +207,7 @@ static int image_make(
                 if (!pretty)
                         pretty = strndupa(filename, strlen(filename) - 4);
 
-                r = image_new(IMAGE_GPT,
+                r = image_new(IMAGE_RAW,
                               pretty,
                               path,
                               filename,
@@ -217,7 +218,7 @@ static int image_make(
                 if (r < 0)
                         return r;
 
-                (*ret)->size = (*ret)->size_exclusive = st.st_blocks * 512;
+                (*ret)->usage = (*ret)->usage_exclusive = st.st_blocks * 512;
                 (*ret)->limit = (*ret)->limit_exclusive = st.st_size;
 
                 return 1;
@@ -249,13 +250,13 @@ int image_find(const char *name, Image **ret) {
 
                 r = image_make(NULL, dirfd(d), path, name, ret);
                 if (r == 0 || r == -ENOENT) {
-                        _cleanup_free_ char *gpt = NULL;
+                        _cleanup_free_ char *raw = NULL;
 
-                        gpt = strappend(name, ".gpt");
-                        if (!gpt)
+                        raw = strappend(name, ".raw");
+                        if (!raw)
                                 return -ENOMEM;
 
-                        r = image_make(NULL, dirfd(d), path, gpt, ret);
+                        r = image_make(NULL, dirfd(d), path, raw, ret);
                         if (r == 0 || r == -ENOENT)
                                 continue;
                 }
@@ -340,32 +341,43 @@ void image_hashmap_free(Hashmap *map) {
 }
 
 int image_remove(Image *i) {
+        _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
+        int r;
+
         assert(i);
 
         if (path_equal(i->path, "/") ||
             path_startswith(i->path, "/usr"))
                 return -EROFS;
 
+        /* Make sure we don't interfere with a running nspawn */
+        r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+        if (r < 0)
+                return r;
+
         switch (i->type) {
 
         case IMAGE_SUBVOLUME:
-                return btrfs_subvol_remove(i->path);
+                return btrfs_subvol_remove(i->path, true);
 
         case IMAGE_DIRECTORY:
                 /* Allow deletion of read-only directories */
                 (void) chattr_path(i->path, false, FS_IMMUTABLE_FL);
+                return rm_rf(i->path, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
 
-                /* fall through */
+        case IMAGE_RAW:
+                if (unlink(i->path) < 0)
+                        return -errno;
 
-        case IMAGE_GPT:
-                return rm_rf_dangerous(i->path, false, true, false);
+                return 0;
 
         default:
-                return -ENOTSUP;
+                return -EOPNOTSUPP;
         }
 }
 
 int image_rename(Image *i, const char *new_name) {
+        _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT, name_lock = LOCK_FILE_INIT;
         _cleanup_free_ char *new_path = NULL, *nn = NULL;
         unsigned file_attr = 0;
         int r;
@@ -379,6 +391,18 @@ int image_rename(Image *i, const char *new_name) {
             path_startswith(i->path, "/usr"))
                 return -EROFS;
 
+        /* Make sure we don't interfere with a running nspawn */
+        r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+        if (r < 0)
+                return r;
+
+        /* Make sure nobody takes the new name, between the time we
+         * checked it is currently unused in all search paths, and the
+         * time we take possesion of it */
+        r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
+        if (r < 0)
+                return r;
+
         r = image_find(new_name, NULL);
         if (r < 0)
                 return r;
@@ -400,16 +424,16 @@ int image_rename(Image *i, const char *new_name) {
                 new_path = file_in_same_dir(i->path, new_name);
                 break;
 
-        case IMAGE_GPT: {
+        case IMAGE_RAW: {
                 const char *fn;
 
-                fn = strappenda(new_name, ".gpt");
+                fn = strjoina(new_name, ".raw");
                 new_path = file_in_same_dir(i->path, fn);
                 break;
         }
 
         default:
-                return -ENOTSUP;
+                return -EOPNOTSUPP;
         }
 
         if (!new_path)
@@ -419,8 +443,9 @@ int image_rename(Image *i, const char *new_name) {
         if (!nn)
                 return -ENOMEM;
 
-        if (renameat2(AT_FDCWD, i->path, AT_FDCWD, new_path, RENAME_NOREPLACE) < 0)
-                return -errno;
+        r = rename_noreplace(AT_FDCWD, i->path, AT_FDCWD, new_path);
+        if (r < 0)
+                return r;
 
         /* Restore the immutable bit, if it was set before */
         if (file_attr & FS_IMMUTABLE_FL)
@@ -438,6 +463,7 @@ int image_rename(Image *i, const char *new_name) {
 }
 
 int image_clone(Image *i, const char *new_name, bool read_only) {
+        _cleanup_release_lock_file_ LockFile name_lock = LOCK_FILE_INIT;
         const char *new_path;
         int r;
 
@@ -446,6 +472,13 @@ int image_clone(Image *i, const char *new_name, bool read_only) {
         if (!image_name_is_valid(new_name))
                 return -EINVAL;
 
+        /* Make sure nobody takes the new name, between the time we
+         * checked it is currently unused in all search paths, and the
+         * time we take possesion of it */
+        r = image_name_lock(new_name, LOCK_EX|LOCK_NB, &name_lock);
+        if (r < 0)
+                return r;
+
         r = image_find(new_name, NULL);
         if (r < 0)
                 return r;
@@ -456,19 +489,19 @@ int image_clone(Image *i, const char *new_name, bool read_only) {
 
         case IMAGE_SUBVOLUME:
         case IMAGE_DIRECTORY:
-                new_path = strappenda("/var/lib/container/", new_name);
+                new_path = strjoina("/var/lib/machines/", new_name);
 
-                r = btrfs_subvol_snapshot(i->path, new_path, read_only, true);
+                r = btrfs_subvol_snapshot(i->path, new_path, (read_only ? BTRFS_SNAPSHOT_READ_ONLY : 0) | BTRFS_SNAPSHOT_FALLBACK_COPY | BTRFS_SNAPSHOT_RECURSIVE);
                 break;
 
-        case IMAGE_GPT:
-                new_path = strappenda("/var/lib/container/", new_name, ".gpt");
+        case IMAGE_RAW:
+                new_path = strjoina("/var/lib/machines/", new_name, ".raw");
 
                 r = copy_file_atomic(i->path, new_path, read_only ? 0444 : 0644, false, FS_NOCOW_FL);
                 break;
 
         default:
-                return -ENOTSUP;
+                return -EOPNOTSUPP;
         }
 
         if (r < 0)
@@ -478,6 +511,7 @@ int image_clone(Image *i, const char *new_name, bool read_only) {
 }
 
 int image_read_only(Image *i, bool b) {
+        _cleanup_release_lock_file_ LockFile global_lock = LOCK_FILE_INIT, local_lock = LOCK_FILE_INIT;
         int r;
         assert(i);
 
@@ -485,6 +519,11 @@ int image_read_only(Image *i, bool b) {
             path_startswith(i->path, "/usr"))
                 return -EROFS;
 
+        /* Make sure we don't interfere with a running nspawn */
+        r = image_path_lock(i->path, LOCK_EX|LOCK_NB, &global_lock, &local_lock);
+        if (r < 0)
+                return r;
+
         switch (i->type) {
 
         case IMAGE_SUBVOLUME:
@@ -509,7 +548,7 @@ int image_read_only(Image *i, bool b) {
 
                 break;
 
-        case IMAGE_GPT: {
+        case IMAGE_RAW: {
                 struct stat st;
 
                 if (stat(i->path, &st) < 0)
@@ -527,16 +566,111 @@ int image_read_only(Image *i, bool b) {
         }
 
         default:
-                return -ENOTSUP;
+                return -EOPNOTSUPP;
+        }
+
+        return 0;
+}
+
+int image_path_lock(const char *path, int operation, LockFile *global, LockFile *local) {
+        _cleanup_free_ char *p = NULL;
+        LockFile t = LOCK_FILE_INIT;
+        struct stat st;
+        int r;
+
+        assert(path);
+        assert(global);
+        assert(local);
+
+        /* Locks an image path. This actually creates two locks: one
+         * "local" one, next to the image path itself, which might be
+         * shared via NFS. And another "global" one, in /run, that
+         * uses the device/inode number. This has the benefit that we
+         * can even lock a tree that is a mount point, correctly. */
+
+        if (path_equal(path, "/"))
+                return -EBUSY;
+
+        if (!path_is_absolute(path))
+                return -EINVAL;
+
+        if (stat(path, &st) >= 0) {
+                if (asprintf(&p, "/run/systemd/nspawn/locks/inode-%lu:%lu", (unsigned long) st.st_dev, (unsigned long) st.st_ino) < 0)
+                        return -ENOMEM;
         }
 
+        r = make_lock_file_for(path, operation, &t);
+        if (r < 0)
+                return r;
+
+        if (p) {
+                mkdir_p("/run/systemd/nspawn/locks", 0600);
+
+                r = make_lock_file(p, operation, global);
+                if (r < 0) {
+                        release_lock_file(&t);
+                        return r;
+                }
+        }
+
+        *local = t;
         return 0;
 }
 
+int image_set_limit(Image *i, uint64_t referenced_max) {
+        assert(i);
+
+        if (path_equal(i->path, "/") ||
+            path_startswith(i->path, "/usr"))
+                return -EROFS;
+
+        if (i->type != IMAGE_SUBVOLUME)
+                return -EOPNOTSUPP;
+
+        return btrfs_quota_limit(i->path, referenced_max);
+}
+
+int image_name_lock(const char *name, int operation, LockFile *ret) {
+        const char *p;
+
+        assert(name);
+        assert(ret);
+
+        /* Locks an image name, regardless of the precise path used. */
+
+        if (!image_name_is_valid(name))
+                return -EINVAL;
+
+        if (streq(name, ".host"))
+                return -EBUSY;
+
+        mkdir_p("/run/systemd/nspawn/locks", 0600);
+        p = strjoina("/run/systemd/nspawn/locks/name-", name);
+
+        return make_lock_file(p, operation, ret);
+}
+
+bool image_name_is_valid(const char *s) {
+        if (!filename_is_valid(s))
+                return false;
+
+        if (string_has_cc(s, NULL))
+                return false;
+
+        if (!utf8_is_valid(s))
+                return false;
+
+        /* Temporary files for atomically creating new files */
+        if (startswith(s, ".#"))
+                return false;
+
+        return true;
+}
+
 static const char* const image_type_table[_IMAGE_TYPE_MAX] = {
         [IMAGE_DIRECTORY] = "directory",
         [IMAGE_SUBVOLUME] = "subvolume",
-        [IMAGE_GPT] = "gpt",
+        [IMAGE_RAW] = "raw",
 };
 
 DEFINE_STRING_TABLE_LOOKUP(image_type, ImageType);