]> git.proxmox.com Git - mirror_qemu.git/blobdiff - block/gluster.c
block/export: fix blk_size double byteswap
[mirror_qemu.git] / block / gluster.c
index 63d3c37d4c7f8e3bcbf28712efa94eba78b9109b..e8ee14c8e9bfc8b58fd34fc0e3c3bfc2253f8ba6 100644 (file)
@@ -9,16 +9,23 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu/units.h"
 #include <glusterfs/api/glfs.h>
 #include "block/block_int.h"
+#include "block/qdict.h"
 #include "qapi/error.h"
 #include "qapi/qmp/qdict.h"
 #include "qapi/qmp/qerror.h"
 #include "qemu/uri.h"
 #include "qemu/error-report.h"
+#include "qemu/module.h"
 #include "qemu/option.h"
 #include "qemu/cutils.h"
 
+#ifdef CONFIG_GLUSTERFS_FTRUNCATE_HAS_STAT
+# define glfs_ftruncate(fd, offset) glfs_ftruncate(fd, offset, NULL, NULL)
+#endif
+
 #define GLUSTER_OPT_FILENAME        "filename"
 #define GLUSTER_OPT_VOLUME          "volume"
 #define GLUSTER_OPT_PATH            "path"
 #define GLUSTER_DEBUG_MAX           9
 #define GLUSTER_OPT_LOGFILE         "logfile"
 #define GLUSTER_LOGFILE_DEFAULT     "-" /* handled in libgfapi as /dev/stderr */
+/*
+ * Several versions of GlusterFS (3.12? -> 6.0.1) fail when the transfer size
+ * is greater or equal to 1024 MiB, so we are limiting the transfer size to 512
+ * MiB to avoid this rare issue.
+ */
+#define GLUSTER_MAX_TRANSFER        (512 * MiB)
 
 #define GERR_INDEX_HINT "hint: check in 'server' array index '%d'\n"
 
@@ -71,7 +84,7 @@ typedef struct ListElement {
     GlfsPreopened saved;
 } ListElement;
 
-static QLIST_HEAD(glfs_list, ListElement) glfs_list;
+static QLIST_HEAD(, ListElement) glfs_list;
 
 static QemuOptsList qemu_gluster_create_opts = {
     .name = "qemu-gluster-create-opts",
@@ -85,7 +98,14 @@ static QemuOptsList qemu_gluster_create_opts = {
         {
             .name = BLOCK_OPT_PREALLOC,
             .type = QEMU_OPT_STRING,
-            .help = "Preallocation mode (allowed values: off, full)"
+            .help = "Preallocation mode (allowed values: off"
+#ifdef CONFIG_GLUSTERFS_FALLOCATE
+                    ", falloc"
+#endif
+#ifdef CONFIG_GLUSTERFS_ZEROFILL
+                    ", full"
+#endif
+                    ")"
         },
         {
             .name = GLUSTER_OPT_DEBUG,
@@ -167,7 +187,12 @@ static QemuOptsList runtime_unix_opts = {
         {
             .name = GLUSTER_OPT_SOCKET,
             .type = QEMU_OPT_STRING,
-            .help = "socket file path)",
+            .help = "socket file path (legacy)",
+        },
+        {
+            .name = GLUSTER_OPT_PATH,
+            .type = QEMU_OPT_STRING,
+            .help = "socket file path (QAPI)",
         },
         { /* end of list */ }
     },
@@ -334,8 +359,8 @@ static int qemu_gluster_parse_uri(BlockdevOptionsGluster *gconf,
         return -EINVAL;
     }
 
-    gconf->server = g_new0(SocketAddressList, 1);
-    gconf->server->value = gsconf = g_new0(SocketAddress, 1);
+    gsconf = g_new0(SocketAddress, 1);
+    QAPI_LIST_PREPEND(gconf->server, gsconf);
 
     /* transport */
     if (!uri->scheme || !strcmp(uri->scheme, "gluster")) {
@@ -489,7 +514,7 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
 {
     QemuOpts *opts;
     SocketAddress *gsconf = NULL;
-    SocketAddressList *curr = NULL;
+    SocketAddressList **tail;
     QDict *backing_options = NULL;
     Error *local_err = NULL;
     char *str = NULL;
@@ -498,8 +523,7 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
 
     /* create opts info from runtime_json_opts list */
     opts = qemu_opts_create(&runtime_json_opts, NULL, 0, &error_abort);
-    qemu_opts_absorb_qdict(opts, options, &local_err);
-    if (local_err) {
+    if (!qemu_opts_absorb_qdict(opts, options, errp)) {
         goto out;
     }
 
@@ -523,6 +547,7 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
     }
     gconf->path = g_strdup(ptr);
     qemu_opts_del(opts);
+    tail = &gconf->server;
 
     for (i = 0; i < num_servers; i++) {
         str = g_strdup_printf(GLUSTER_OPT_SERVER_PATTERN"%d.", i);
@@ -530,8 +555,7 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
 
         /* create opts info from runtime_type_opts list */
         opts = qemu_opts_create(&runtime_type_opts, NULL, 0, &error_abort);
-        qemu_opts_absorb_qdict(opts, backing_options, &local_err);
-        if (local_err) {
+        if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
             goto out;
         }
 
@@ -561,8 +585,7 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
         if (gsconf->type == SOCKET_ADDRESS_TYPE_INET) {
             /* create opts info from runtime_inet_opts list */
             opts = qemu_opts_create(&runtime_inet_opts, NULL, 0, &error_abort);
-            qemu_opts_absorb_qdict(opts, backing_options, &local_err);
-            if (local_err) {
+            if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
                 goto out;
             }
 
@@ -610,15 +633,22 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
         } else {
             /* create opts info from runtime_unix_opts list */
             opts = qemu_opts_create(&runtime_unix_opts, NULL, 0, &error_abort);
-            qemu_opts_absorb_qdict(opts, backing_options, &local_err);
-            if (local_err) {
+            if (!qemu_opts_absorb_qdict(opts, backing_options, errp)) {
                 goto out;
             }
 
-            ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
+            ptr = qemu_opt_get(opts, GLUSTER_OPT_PATH);
+            if (!ptr) {
+                ptr = qemu_opt_get(opts, GLUSTER_OPT_SOCKET);
+            } else if (qemu_opt_get(opts, GLUSTER_OPT_SOCKET)) {
+                error_setg(&local_err,
+                           "Conflicting parameters 'path' and 'socket'");
+                error_append_hint(&local_err, GERR_INDEX_HINT, i);
+                goto out;
+            }
             if (!ptr) {
                 error_setg(&local_err, QERR_MISSING_PARAMETER,
-                           GLUSTER_OPT_SOCKET);
+                           GLUSTER_OPT_PATH);
                 error_append_hint(&local_err, GERR_INDEX_HINT, i);
                 goto out;
             }
@@ -626,18 +656,10 @@ static int qemu_gluster_parse_json(BlockdevOptionsGluster *gconf,
             qemu_opts_del(opts);
         }
 
-        if (gconf->server == NULL) {
-            gconf->server = g_new0(SocketAddressList, 1);
-            gconf->server->value = gsconf;
-            curr = gconf->server;
-        } else {
-            curr->next = g_new0(SocketAddressList, 1);
-            curr->next->value = gsconf;
-            curr = curr->next;
-        }
+        QAPI_LIST_APPEND(tail, gsconf);
         gsconf = NULL;
 
-        QDECREF(backing_options);
+        qobject_unref(backing_options);
         backing_options = NULL;
         g_free(str);
         str = NULL;
@@ -650,7 +672,7 @@ out:
     qapi_free_SocketAddress(gsconf);
     qemu_opts_del(opts);
     g_free(str);
-    QDECREF(backing_options);
+    qobject_unref(backing_options);
     errno = EINVAL;
     return -errno;
 }
@@ -665,7 +687,7 @@ static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
     if (filename) {
         ret = qemu_gluster_parse_uri(gconf, filename);
         if (ret < 0) {
-            error_setg(errp, "invalid URI");
+            error_setg(errp, "invalid URI %s", filename);
             error_append_hint(errp, "Usage: file=gluster[+transport]://"
                                     "[host[:port]]volume/path[?socket=...]"
                                     "[,file.debug=N]"
@@ -684,7 +706,7 @@ static int qemu_gluster_parse(BlockdevOptionsGluster *gconf,
                              "file.server.0.host=1.2.3.4,"
                              "file.server.0.port=24007,"
                              "file.server.1.transport=unix,"
-                             "file.server.1.socket=/var/run/glusterd.socket ..."
+                             "file.server.1.path=/var/run/glusterd.socket ..."
                              "\n");
             return ret;
         }
@@ -711,7 +733,11 @@ static struct glfs *qemu_gluster_init(BlockdevOptionsGluster *gconf,
 /*
  * AIO callback routine called from GlusterFS thread.
  */
-static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret, void *arg)
+static void gluster_finish_aiocb(struct glfs_fd *fd, ssize_t ret,
+#ifdef CONFIG_GLUSTERFS_IOCB_HAS_STAT
+                                 struct glfs_stat *pre, struct glfs_stat *post,
+#endif
+                                 void *arg)
 {
     GlusterAIOCB *acb = (GlusterAIOCB *)arg;
 
@@ -778,13 +804,10 @@ static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
     int ret = 0;
     BlockdevOptionsGluster *gconf = NULL;
     QemuOpts *opts;
-    Error *local_err = NULL;
     const char *filename, *logfile;
 
     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
-    qemu_opts_absorb_qdict(opts, options, &local_err);
-    if (local_err) {
-        error_propagate(errp, local_err);
+    if (!qemu_opts_absorb_qdict(opts, options, errp)) {
         ret = -EINVAL;
         goto out;
     }
@@ -835,8 +858,16 @@ static int qemu_gluster_open(BlockDriverState *bs,  QDict *options,
     qemu_gluster_parse_flags(bdrv_flags, &open_flags);
 
     s->fd = glfs_open(s->glfs, gconf->path, open_flags);
-    if (!s->fd) {
-        ret = -errno;
+    ret = s->fd ? 0 : -errno;
+
+    if (ret == -EACCES || ret == -EROFS) {
+        /* Try to degrade to read-only, but if it doesn't work, still use the
+         * normal error message. */
+        if (bdrv_apply_auto_read_only(bs, NULL, NULL) == 0) {
+            open_flags = (open_flags & ~O_RDWR) | O_RDONLY;
+            s->fd = glfs_open(s->glfs, gconf->path, open_flags);
+            ret = s->fd ? 0 : -errno;
+        }
     }
 
     s->supports_seek_data = qemu_gluster_test_seek(s->fd);
@@ -857,6 +888,11 @@ out:
     return ret;
 }
 
+static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
+{
+    bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
+}
+
 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
                                        BlockReopenQueue *queue, Error **errp)
 {
@@ -881,7 +917,17 @@ static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
     gconf->has_debug = true;
     gconf->logfile = g_strdup(s->logfile);
     gconf->has_logfile = true;
-    reop_s->glfs = qemu_gluster_init(gconf, state->bs->filename, NULL, errp);
+
+    /*
+     * If 'state->bs->exact_filename' is empty, 'state->options' should contain
+     * the JSON parameters already parsed.
+     */
+    if (state->bs->exact_filename[0] != '\0') {
+        reop_s->glfs = qemu_gluster_init(gconf, state->bs->exact_filename, NULL,
+                                         errp);
+    } else {
+        reop_s->glfs = qemu_gluster_init(gconf, NULL, state->options, errp);
+    }
     if (reop_s->glfs == NULL) {
         ret = -errno;
         goto exit;
@@ -1070,7 +1116,8 @@ out:
     return ret;
 }
 
-static int coroutine_fn qemu_gluster_co_create_opts(const char *filename,
+static int coroutine_fn qemu_gluster_co_create_opts(BlockDriver *drv,
+                                                    const char *filename,
                                                     QemuOpts *opts,
                                                     Error **errp)
 {
@@ -1163,8 +1210,12 @@ static coroutine_fn int qemu_gluster_co_rw(BlockDriverState *bs,
     return acb.ret;
 }
 
-static int qemu_gluster_truncate(BlockDriverState *bs, int64_t offset,
-                                 PreallocMode prealloc, Error **errp)
+static coroutine_fn int qemu_gluster_co_truncate(BlockDriverState *bs,
+                                                 int64_t offset,
+                                                 bool exact,
+                                                 PreallocMode prealloc,
+                                                 BdrvRequestFlags flags,
+                                                 Error **errp)
 {
     BDRVGlusterState *s = bs->opaque;
     return qemu_gluster_do_truncate(s->fd, offset, prealloc, errp);
@@ -1181,8 +1232,10 @@ static coroutine_fn int qemu_gluster_co_readv(BlockDriverState *bs,
 static coroutine_fn int qemu_gluster_co_writev(BlockDriverState *bs,
                                                int64_t sector_num,
                                                int nb_sectors,
-                                               QEMUIOVector *qiov)
+                                               QEMUIOVector *qiov,
+                                               int flags)
 {
+    assert(!flags);
     return qemu_gluster_co_rw(bs, sector_num, nb_sectors, qiov, 1);
 }
 
@@ -1292,12 +1345,6 @@ static int64_t qemu_gluster_allocated_file_size(BlockDriverState *bs)
     }
 }
 
-static int qemu_gluster_has_zero_init(BlockDriverState *bs)
-{
-    /* GlusterFS volume could be backed by a block device */
-    return 0;
-}
-
 /*
  * Find allocation range in @bs around offset @start.
  * May change underlying file descriptor's file offset.
@@ -1308,7 +1355,7 @@ static int qemu_gluster_has_zero_init(BlockDriverState *bs)
  * If @start is in a trailing hole or beyond EOF, return -ENXIO.
  * If we can't find out, return a negative errno other than -ENXIO.
  *
- * (Shamefully copied from file-posix.c, only miniscule adaptions.)
+ * (Shamefully copied from file-posix.c, only minuscule adaptions.)
  */
 static int find_allocation(BlockDriverState *bs, off_t start,
                            off_t *data, off_t *hole)
@@ -1469,6 +1516,21 @@ static int coroutine_fn qemu_gluster_co_block_status(BlockDriverState *bs,
 }
 
 
+static const char *const gluster_strong_open_opts[] = {
+    GLUSTER_OPT_VOLUME,
+    GLUSTER_OPT_PATH,
+    GLUSTER_OPT_TYPE,
+    GLUSTER_OPT_SERVER_PATTERN,
+    GLUSTER_OPT_HOST,
+    GLUSTER_OPT_PORT,
+    GLUSTER_OPT_TO,
+    GLUSTER_OPT_IPV4,
+    GLUSTER_OPT_IPV6,
+    GLUSTER_OPT_SOCKET,
+
+    NULL
+};
+
 static BlockDriver bdrv_gluster = {
     .format_name                  = "gluster",
     .protocol_name                = "gluster",
@@ -1483,11 +1545,10 @@ static BlockDriver bdrv_gluster = {
     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
     .bdrv_getlength               = qemu_gluster_getlength,
     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
-    .bdrv_truncate                = qemu_gluster_truncate,
+    .bdrv_co_truncate             = qemu_gluster_co_truncate,
     .bdrv_co_readv                = qemu_gluster_co_readv,
     .bdrv_co_writev               = qemu_gluster_co_writev,
     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
-    .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
 #ifdef CONFIG_GLUSTERFS_DISCARD
     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
 #endif
@@ -1495,7 +1556,9 @@ static BlockDriver bdrv_gluster = {
     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
 #endif
     .bdrv_co_block_status         = qemu_gluster_co_block_status,
+    .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
     .create_opts                  = &qemu_gluster_create_opts,
+    .strong_runtime_opts          = gluster_strong_open_opts,
 };
 
 static BlockDriver bdrv_gluster_tcp = {
@@ -1512,11 +1575,10 @@ static BlockDriver bdrv_gluster_tcp = {
     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
     .bdrv_getlength               = qemu_gluster_getlength,
     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
-    .bdrv_truncate                = qemu_gluster_truncate,
+    .bdrv_co_truncate             = qemu_gluster_co_truncate,
     .bdrv_co_readv                = qemu_gluster_co_readv,
     .bdrv_co_writev               = qemu_gluster_co_writev,
     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
-    .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
 #ifdef CONFIG_GLUSTERFS_DISCARD
     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
 #endif
@@ -1524,7 +1586,9 @@ static BlockDriver bdrv_gluster_tcp = {
     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
 #endif
     .bdrv_co_block_status         = qemu_gluster_co_block_status,
+    .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
     .create_opts                  = &qemu_gluster_create_opts,
+    .strong_runtime_opts          = gluster_strong_open_opts,
 };
 
 static BlockDriver bdrv_gluster_unix = {
@@ -1541,11 +1605,10 @@ static BlockDriver bdrv_gluster_unix = {
     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
     .bdrv_getlength               = qemu_gluster_getlength,
     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
-    .bdrv_truncate                = qemu_gluster_truncate,
+    .bdrv_co_truncate             = qemu_gluster_co_truncate,
     .bdrv_co_readv                = qemu_gluster_co_readv,
     .bdrv_co_writev               = qemu_gluster_co_writev,
     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
-    .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
 #ifdef CONFIG_GLUSTERFS_DISCARD
     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
 #endif
@@ -1553,7 +1616,9 @@ static BlockDriver bdrv_gluster_unix = {
     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
 #endif
     .bdrv_co_block_status         = qemu_gluster_co_block_status,
+    .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
     .create_opts                  = &qemu_gluster_create_opts,
+    .strong_runtime_opts          = gluster_strong_open_opts,
 };
 
 /* rdma is deprecated (actually never supported for volfile fetch).
@@ -1576,11 +1641,10 @@ static BlockDriver bdrv_gluster_rdma = {
     .bdrv_co_create_opts          = qemu_gluster_co_create_opts,
     .bdrv_getlength               = qemu_gluster_getlength,
     .bdrv_get_allocated_file_size = qemu_gluster_allocated_file_size,
-    .bdrv_truncate                = qemu_gluster_truncate,
+    .bdrv_co_truncate             = qemu_gluster_co_truncate,
     .bdrv_co_readv                = qemu_gluster_co_readv,
     .bdrv_co_writev               = qemu_gluster_co_writev,
     .bdrv_co_flush_to_disk        = qemu_gluster_co_flush_to_disk,
-    .bdrv_has_zero_init           = qemu_gluster_has_zero_init,
 #ifdef CONFIG_GLUSTERFS_DISCARD
     .bdrv_co_pdiscard             = qemu_gluster_co_pdiscard,
 #endif
@@ -1588,7 +1652,9 @@ static BlockDriver bdrv_gluster_rdma = {
     .bdrv_co_pwrite_zeroes        = qemu_gluster_co_pwrite_zeroes,
 #endif
     .bdrv_co_block_status         = qemu_gluster_co_block_status,
+    .bdrv_refresh_limits          = qemu_gluster_refresh_limits,
     .create_opts                  = &qemu_gluster_create_opts,
+    .strong_runtime_opts          = gluster_strong_open_opts,
 };
 
 static void bdrv_gluster_init(void)