X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fovsdb-data.c;h=76d046f17a38440e5fc25d91e7e780189cc8832a;hb=0d0f05b909b6428d44eb147bd4edd73782d2a137;hp=0d7749b17fc3912606128384c781d250e32212ee;hpb=e9f8f9367e9e654b9582279608af86ea3744a2a0;p=openvswitch

diff --git a/lib/ovsdb-data.c b/lib/ovsdb-data.c
index 0d7749b1..76d046f1 100644
--- a/lib/ovsdb-data.c
+++ b/lib/ovsdb-data.c
@@ -1,4 +1,4 @@
-/* Copyright (c) 2009 Nicira Networks
+/* Copyright (c) 2009, 2010 Nicira Networks
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,13 +18,18 @@
 #include "ovsdb-data.h"
 
 #include <assert.h>
+#include <ctype.h>
+#include <float.h>
+#include <inttypes.h>
 #include <limits.h>
 
+#include "dynamic-string.h"
 #include "hash.h"
 #include "ovsdb-error.h"
 #include "json.h"
 #include "shash.h"
 #include "sort.h"
+#include "unicode.h"
 
 static struct json *
 wrap_json(const char *name, struct json *wrapped)
@@ -65,6 +70,35 @@ ovsdb_atom_init_default(union ovsdb_atom *atom, enum ovsdb_atomic_type type)
     }
 }
 
+bool
+ovsdb_atom_is_default(const union ovsdb_atom *atom,
+                      enum ovsdb_atomic_type type)
+{
+    switch (type) {
+    case OVSDB_TYPE_VOID:
+        NOT_REACHED();
+
+    case OVSDB_TYPE_INTEGER:
+        return atom->integer == 0;
+
+    case OVSDB_TYPE_REAL:
+        return atom->real == 0.0;
+
+    case OVSDB_TYPE_BOOLEAN:
+        return atom->boolean == false;
+
+    case OVSDB_TYPE_STRING:
+        return atom->string[0] == '\0';
+
+    case OVSDB_TYPE_UUID:
+        return uuid_is_zero(&atom->uuid);
+
+    case OVSDB_N_TYPES:
+    default:
+        NOT_REACHED();
+    }
+}
+
 void
 ovsdb_atom_clone(union ovsdb_atom *new, const union ovsdb_atom *old,
                  enum ovsdb_atomic_type type)
@@ -239,10 +273,10 @@ ovsdb_atom_parse_uuid(struct uuid *uuid, const struct json *json,
     return error0;
 }
 
-struct ovsdb_error *
-ovsdb_atom_from_json(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
-                     const struct json *json,
-                     const struct ovsdb_symbol_table *symtab)
+static struct ovsdb_error * WARN_UNUSED_RESULT
+ovsdb_atom_from_json__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
+                       const struct json *json,
+                       const struct ovsdb_symbol_table *symtab)
 {
     switch (type) {
     case OVSDB_TYPE_VOID:
@@ -294,6 +328,26 @@ ovsdb_atom_from_json(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
                               ovsdb_atomic_type_to_string(type));
 }
 
+struct ovsdb_error *
+ovsdb_atom_from_json(union ovsdb_atom *atom,
+                     const struct ovsdb_base_type *base,
+                     const struct json *json,
+                     const struct ovsdb_symbol_table *symtab)
+{
+    struct ovsdb_error *error;
+
+    error = ovsdb_atom_from_json__(atom, base->type, json, symtab);
+    if (error) {
+        return error;
+    }
+
+    error = ovsdb_atom_check_constraints(atom, base);
+    if (error) {
+        ovsdb_atom_destroy(atom, base->type);
+    }
+    return error;
+}
+
 struct json *
 ovsdb_atom_to_json(const union ovsdb_atom *atom, enum ovsdb_atomic_type type)
 {
@@ -322,6 +376,327 @@ ovsdb_atom_to_json(const union ovsdb_atom *atom, enum ovsdb_atomic_type type)
         NOT_REACHED();
     }
 }
+
+static char *
+ovsdb_atom_from_string__(union ovsdb_atom *atom, enum ovsdb_atomic_type type,
+                         const char *s)
+{
+    switch (type) {
+    case OVSDB_TYPE_VOID:
+        NOT_REACHED();
+
+    case OVSDB_TYPE_INTEGER: {
+        long long int integer;
+        if (!str_to_llong(s, 10, &integer)) {
+            return xasprintf("\"%s\" is not a valid integer", s);
+        }
+        atom->integer = integer;
+    }
+        break;
+
+    case OVSDB_TYPE_REAL:
+        if (!str_to_double(s, &atom->real)) {
+            return xasprintf("\"%s\" is not a valid real number", s);
+        }
+        /* Our JSON input routines map negative zero to zero, so do that here
+         * too for consistency. */
+        if (atom->real == 0.0) {
+            atom->real = 0.0;
+        }
+        break;
+
+    case OVSDB_TYPE_BOOLEAN:
+        if (!strcmp(s, "true") || !strcmp(s, "yes") || !strcmp(s, "on")
+            || !strcmp(s, "1")) {
+            atom->boolean = true;
+        } else if (!strcmp(s, "false") || !strcmp(s, "no") || !strcmp(s, "off")
+                   || !strcmp(s, "0")) {
+            atom->boolean = false;
+        } else {
+            return xasprintf("\"%s\" is not a valid boolean "
+                             "(use \"true\" or \"false\")", s);
+        }
+        break;
+
+    case OVSDB_TYPE_STRING:
+        if (*s == '\0') {
+            return xstrdup("An empty string is not valid as input; "
+                           "use \"\" to represent the empty string");
+        } else if (*s == '"') {
+            size_t s_len = strlen(s);
+
+            if (s_len < 2 || s[s_len - 1] != '"') {
+                return xasprintf("%s: missing quote at end of "
+                                 "quoted string", s);
+            } else if (!json_string_unescape(s + 1, s_len - 2,
+                                             &atom->string)) {
+                char *error = xasprintf("%s: %s", s, atom->string);
+                free(atom->string);
+                return error;
+            }
+        } else {
+            atom->string = xstrdup(s);
+        }
+        break;
+
+    case OVSDB_TYPE_UUID:
+        if (!uuid_from_string(&atom->uuid, s)) {
+            return xasprintf("\"%s\" is not a valid UUID", s);
+        }
+        break;
+
+    case OVSDB_N_TYPES:
+    default:
+        NOT_REACHED();
+    }
+
+    return NULL;
+}
+
+/* Initializes 'atom' to a value of type 'base' parsed from 's', which takes
+ * one of the following forms:
+ *
+ *      - OVSDB_TYPE_INTEGER: A decimal integer optionally preceded by a sign.
+ *
+ *      - OVSDB_TYPE_REAL: A floating-point number in the format accepted by
+ *        strtod().
+ *
+ *      - OVSDB_TYPE_BOOLEAN: "true", "yes", "on", "1" for true, or "false",
+ *        "no", "off", or "0" for false.
+ *
+ *      - OVSDB_TYPE_STRING: A JSON string if it begins with a quote, otherwise
+ *        an arbitrary string.
+ *
+ *      - OVSDB_TYPE_UUID: A UUID in RFC 4122 format.
+ *
+ * Returns a null pointer if successful, otherwise an error message describing
+ * the problem.  The caller is responsible for freeing the error.
+ */
+char *
+ovsdb_atom_from_string(union ovsdb_atom *atom,
+                       const struct ovsdb_base_type *base, const char *s)
+{
+    struct ovsdb_error *error;
+    char *msg;
+
+    msg = ovsdb_atom_from_string__(atom, base->type, s);
+    if (msg) {
+        return msg;
+    }
+
+    error = ovsdb_atom_check_constraints(atom, base);
+    if (error) {
+        msg = ovsdb_error_to_string(error);
+        ovsdb_error_destroy(error);
+    }
+    return msg;
+}
+
+static bool
+string_needs_quotes(const char *s)
+{
+    const char *p = s;
+    unsigned char c;
+
+    c = *p++;
+    if (!isalpha(c) && c != '_') {
+        return true;
+    }
+
+    while ((c = *p++) != '\0') {
+        if (!isalpha(c) && c != '_' && c != '-' && c != '.') {
+            return true;
+        }
+    }
+
+    if (!strcmp(s, "true") || !strcmp(s, "false")) {
+        return true;
+    }
+
+    return false;
+}
+
+/* Appends 'atom' (which has the given 'type') to 'out', in a format acceptable
+ * to ovsdb_atom_from_string().  */
+void
+ovsdb_atom_to_string(const union ovsdb_atom *atom, enum ovsdb_atomic_type type,
+                     struct ds *out)
+{
+    switch (type) {
+    case OVSDB_TYPE_VOID:
+        NOT_REACHED();
+
+    case OVSDB_TYPE_INTEGER:
+        ds_put_format(out, "%"PRId64, atom->integer);
+        break;
+
+    case OVSDB_TYPE_REAL:
+        ds_put_format(out, "%.*g", DBL_DIG, atom->real);
+        break;
+
+    case OVSDB_TYPE_BOOLEAN:
+        ds_put_cstr(out, atom->boolean ? "true" : "false");
+        break;
+
+    case OVSDB_TYPE_STRING:
+        if (string_needs_quotes(atom->string)) {
+            struct json json;
+
+            json.type = JSON_STRING;
+            json.u.string = atom->string;
+            json_to_ds(&json, 0, out);
+        } else {
+            ds_put_cstr(out, atom->string);
+        }
+        break;
+
+    case OVSDB_TYPE_UUID:
+        ds_put_format(out, UUID_FMT, UUID_ARGS(&atom->uuid));
+        break;
+
+    case OVSDB_N_TYPES:
+    default:
+        NOT_REACHED();
+    }
+}
+
+static struct ovsdb_error *
+check_string_constraints(const char *s,
+                         const struct ovsdb_string_constraints *c)
+{
+    size_t n_chars;
+    char *msg;
+
+    msg = utf8_validate(s, &n_chars);
+    if (msg) {
+        struct ovsdb_error *error;
+
+        error = ovsdb_error("constraint violation",
+                            "\"%s\" is not a valid UTF-8 string: %s",
+                            s, msg);
+        free(msg);
+        return error;
+    }
+
+    if (n_chars < c->minLen) {
+        return ovsdb_error(
+            "constraint violation",
+            "\"%s\" length %zu is less than minimum allowed "
+            "length %u", s, n_chars, c->minLen);
+    } else if (n_chars > c->maxLen) {
+        return ovsdb_error(
+            "constraint violation",
+            "\"%s\" length %zu is greater than maximum allowed "
+            "length %u", s, n_chars, c->maxLen);
+    }
+
+#if HAVE_PCRE
+    if (c->re) {
+        int retval;
+
+        retval = pcre_exec(c->re, NULL, s, strlen(s), 0,
+                           PCRE_ANCHORED | PCRE_NO_UTF8_CHECK, NULL, 0);
+        if (retval == PCRE_ERROR_NOMATCH) {
+            if (c->reComment) {
+                return ovsdb_error("constraint violation",
+                                   "\"%s\" is not a %s", s, c->reComment);
+            } else {
+                return ovsdb_error("constraint violation",
+                                   "\"%s\" does not match regular expression "
+                                   "/%s/", s, c->reMatch);
+            }
+        } else if (retval < 0) {
+            /* PCRE doesn't have a function to translate an error code to a
+             * description.  Bizarre.  See pcreapi(3) for error details. */
+            return ovsdb_error("internal error", "PCRE returned error %d",
+                               retval);
+        }
+    }
+#endif  /* HAVE_PCRE */
+
+    return NULL;
+}
+
+/* Checks whether 'atom' meets the constraints (if any) defined in 'base'.
+ * (base->type must specify 'atom''s type.)  Returns a null pointer if the
+ * constraints are met, otherwise an error that explains the violation.
+ *
+ * Checking UUID constraints is deferred to transaction commit time, so this
+ * function does nothing for UUID constraints. */
+struct ovsdb_error *
+ovsdb_atom_check_constraints(const union ovsdb_atom *atom,
+                             const struct ovsdb_base_type *base)
+{
+    switch (base->type) {
+    case OVSDB_TYPE_VOID:
+        NOT_REACHED();
+
+    case OVSDB_TYPE_INTEGER:
+        if (atom->integer >= base->u.integer.min
+            && atom->integer <= base->u.integer.max) {
+            return NULL;
+        } else if (base->u.integer.min != INT64_MIN) {
+            if (base->u.integer.max != INT64_MAX) {
+                return ovsdb_error("constraint violation",
+                                   "%"PRId64" is not in the valid range "
+                                   "%"PRId64" to %"PRId64" (inclusive)",
+                                   atom->integer,
+                                   base->u.integer.min, base->u.integer.max);
+            } else {
+                return ovsdb_error("constraint violation",
+                                   "%"PRId64" is less than minimum allowed "
+                                   "value %"PRId64,
+                                   atom->integer, base->u.integer.min);
+            }
+        } else {
+            return ovsdb_error("constraint violation",
+                               "%"PRId64" is greater than maximum allowed "
+                               "value %"PRId64,
+                               atom->integer, base->u.integer.max);
+        }
+        NOT_REACHED();
+
+    case OVSDB_TYPE_REAL:
+        if (atom->real >= base->u.real.min && atom->real <= base->u.real.max) {
+            return NULL;
+        } else if (base->u.real.min != -DBL_MAX) {
+            if (base->u.real.max != DBL_MAX) {
+                return ovsdb_error("constraint violation",
+                                   "%.*g is not in the valid range "
+                                   "%.*g to %.*g (inclusive)",
+                                   DBL_DIG, atom->real,
+                                   DBL_DIG, base->u.real.min,
+                                   DBL_DIG, base->u.real.max);
+            } else {
+                return ovsdb_error("constraint violation",
+                                   "%.*g is less than minimum allowed "
+                                   "value %.*g",
+                                   DBL_DIG, atom->real,
+                                   DBL_DIG, base->u.real.min);
+            }
+        } else {
+            return ovsdb_error("constraint violation",
+                               "%.*g is greater than maximum allowed "
+                               "value %.*g",
+                               DBL_DIG, atom->real,
+                               DBL_DIG, base->u.real.max);
+        }
+        NOT_REACHED();
+
+    case OVSDB_TYPE_BOOLEAN:
+        return NULL;
+
+    case OVSDB_TYPE_STRING:
+        return check_string_constraints(atom->string, &base->u.string);
+
+    case OVSDB_TYPE_UUID:
+        return NULL;
+
+    case OVSDB_N_TYPES:
+    default:
+        NOT_REACHED();
+    }
+}
 
 static union ovsdb_atom *
 alloc_default_atoms(enum ovsdb_atomic_type type, size_t n)
@@ -342,13 +717,43 @@ alloc_default_atoms(enum ovsdb_atomic_type type, size_t n)
     }
 }
 
+void
+ovsdb_datum_init_empty(struct ovsdb_datum *datum)
+{
+    datum->n = 0;
+    datum->keys = NULL;
+    datum->values = NULL;
+}
+
 void
 ovsdb_datum_init_default(struct ovsdb_datum *datum,
                          const struct ovsdb_type *type)
 {
     datum->n = type->n_min;
-    datum->keys = alloc_default_atoms(type->key_type, datum->n);
-    datum->values = alloc_default_atoms(type->value_type, datum->n);
+    datum->keys = alloc_default_atoms(type->key.type, datum->n);
+    datum->values = alloc_default_atoms(type->value.type, datum->n);
+}
+
+bool
+ovsdb_datum_is_default(const struct ovsdb_datum *datum,
+                       const struct ovsdb_type *type)
+{
+    size_t i;
+
+    if (datum->n != type->n_min) {
+        return false;
+    }
+    for (i = 0; i < datum->n; i++) {
+        if (!ovsdb_atom_is_default(&datum->keys[i], type->key.type)) {
+            return false;
+        }
+        if (type->value.type != OVSDB_TYPE_VOID
+            && !ovsdb_atom_is_default(&datum->values[i], type->value.type)) {
+            return false;
+        }
+    }
+
+    return true;
 }
 
 static union ovsdb_atom *
@@ -376,8 +781,8 @@ ovsdb_datum_clone(struct ovsdb_datum *new, const struct ovsdb_datum *old,
 {
     unsigned int n = old->n;
     new->n = n;
-    new->keys = clone_atoms(old->keys, type->key_type, n);
-    new->values = clone_atoms(old->values, type->value_type, n);
+    new->keys = clone_atoms(old->keys, type->key.type, n);
+    new->values = clone_atoms(old->values, type->value.type, n);
 }
 
 static void
@@ -396,8 +801,8 @@ free_data(enum ovsdb_atomic_type type,
 void
 ovsdb_datum_destroy(struct ovsdb_datum *datum, const struct ovsdb_type *type)
 {
-    free_data(type->key_type, datum->keys, datum->n);
-    free_data(type->value_type, datum->values, datum->n);
+    free_data(type->key.type, datum->keys, datum->n);
+    free_data(type->value.type, datum->values, datum->n);
 }
 
 void
@@ -420,7 +825,7 @@ ovsdb_datum_sort_compare_cb(size_t a, size_t b, void *cbdata_)
 
     return ovsdb_atom_compare_3way(&cbdata->datum->keys[a],
                                    &cbdata->datum->keys[b],
-                                   cbdata->type->key_type);
+                                   cbdata->type->key.type);
 }
 
 static void
@@ -429,7 +834,7 @@ ovsdb_datum_sort_swap_cb(size_t a, size_t b, void *cbdata_)
     struct ovsdb_datum_sort_cbdata *cbdata = cbdata_;
 
     ovsdb_atom_swap(&cbdata->datum->keys[a], &cbdata->datum->keys[b]);
-    if (cbdata->type->value_type != OVSDB_TYPE_VOID) {
+    if (cbdata->type->value.type != OVSDB_TYPE_VOID) {
         ovsdb_atom_swap(&cbdata->datum->values[a], &cbdata->datum->values[b]);
     }
 }
@@ -450,7 +855,7 @@ ovsdb_datum_sort(struct ovsdb_datum *datum, const struct ovsdb_type *type)
 
         for (i = 0; i < datum->n - 1; i++) {
             if (ovsdb_atom_equals(&datum->keys[i], &datum->keys[i + 1],
-                                  type->key_type)) {
+                                  type->key.type)) {
                 if (ovsdb_type_is_map(type)) {
                     return ovsdb_error(NULL, "map contains duplicate key");
                 } else {
@@ -463,6 +868,40 @@ ovsdb_datum_sort(struct ovsdb_datum *datum, const struct ovsdb_type *type)
     }
 }
 
+/* Checks that each of the atoms in 'datum' conforms to the constraints
+ * specified by its 'type'.  Returns an error if a constraint is violated,
+ * otherwise a null pointer.
+ *
+ * This function is not commonly useful because the most ordinary way to obtain
+ * a datum is ultimately via ovsdb_atom_from_string() or
+ * ovsdb_atom_from_json(), which check constraints themselves. */
+struct ovsdb_error *
+ovsdb_datum_check_constraints(const struct ovsdb_datum *datum,
+                              const struct ovsdb_type *type)
+{
+    struct ovsdb_error *error;
+    unsigned int i;
+
+    for (i = 0; i < datum->n; i++) {
+        error = ovsdb_atom_check_constraints(&datum->keys[i], &type->key);
+        if (error) {
+            return error;
+        }
+    }
+
+    if (type->value.type != OVSDB_TYPE_VOID) {
+        for (i = 0; i < datum->n; i++) {
+            error = ovsdb_atom_check_constraints(&datum->values[i],
+                                                 &type->value);
+            if (error) {
+                return error;
+            }
+        }
+    }
+
+    return NULL;
+}
+
 struct ovsdb_error *
 ovsdb_datum_from_json(struct ovsdb_datum *datum,
                       const struct ovsdb_type *type,
@@ -476,7 +915,7 @@ ovsdb_datum_from_json(struct ovsdb_datum *datum,
         datum->keys = xmalloc(sizeof *datum->keys);
         datum->values = NULL;
 
-        error = ovsdb_atom_from_json(&datum->keys[0], type->key_type,
+        error = ovsdb_atom_from_json(&datum->keys[0], &type->key,
                                      json, symtab);
         if (error) {
             free(datum->keys);
@@ -520,7 +959,7 @@ ovsdb_datum_from_json(struct ovsdb_datum *datum,
                 }
             }
 
-            error = ovsdb_atom_from_json(&datum->keys[i], type->key_type,
+            error = ovsdb_atom_from_json(&datum->keys[i], &type->key,
                                          key, symtab);
             if (error) {
                 goto error;
@@ -528,9 +967,9 @@ ovsdb_datum_from_json(struct ovsdb_datum *datum,
 
             if (is_map) {
                 error = ovsdb_atom_from_json(&datum->values[i],
-                                             type->value_type, value, symtab);
+                                             &type->value, value, symtab);
                 if (error) {
-                    ovsdb_atom_destroy(&datum->keys[i], type->key_type);
+                    ovsdb_atom_destroy(&datum->keys[i], type->key.type);
                     goto error;
                 }
             }
@@ -558,14 +997,14 @@ ovsdb_datum_to_json(const struct ovsdb_datum *datum,
     /* These tests somewhat tolerate a 'datum' that does not exactly match
      * 'type', in particular a datum with 'n' not in the allowed range. */
     if (datum->n == 1 && ovsdb_type_is_scalar(type)) {
-        return ovsdb_atom_to_json(&datum->keys[0], type->key_type);
-    } else if (type->value_type == OVSDB_TYPE_VOID) {
+        return ovsdb_atom_to_json(&datum->keys[0], type->key.type);
+    } else if (type->value.type == OVSDB_TYPE_VOID) {
         struct json **elems;
         size_t i;
 
         elems = xmalloc(datum->n * sizeof *elems);
         for (i = 0; i < datum->n; i++) {
-            elems[i] = ovsdb_atom_to_json(&datum->keys[i], type->key_type);
+            elems[i] = ovsdb_atom_to_json(&datum->keys[i], type->key.type);
         }
 
         return wrap_json("set", json_array_create(elems, datum->n));
@@ -576,14 +1015,201 @@ ovsdb_datum_to_json(const struct ovsdb_datum *datum,
         elems = xmalloc(datum->n * sizeof *elems);
         for (i = 0; i < datum->n; i++) {
             elems[i] = json_array_create_2(
-                ovsdb_atom_to_json(&datum->keys[i], type->key_type),
-                ovsdb_atom_to_json(&datum->values[i], type->value_type));
+                ovsdb_atom_to_json(&datum->keys[i], type->key.type),
+                ovsdb_atom_to_json(&datum->values[i], type->value.type));
         }
 
         return wrap_json("map", json_array_create(elems, datum->n));
     }
 }
 
+static const char *
+skip_spaces(const char *p)
+{
+    while (isspace((unsigned char) *p)) {
+        p++;
+    }
+    return p;
+}
+
+static char *
+parse_atom_token(const char **s, const struct ovsdb_base_type *base,
+                 union ovsdb_atom *atom)
+{
+    char *token, *error;
+
+    error = ovsdb_token_parse(s, &token);
+    if (!error) {
+        error = ovsdb_atom_from_string(atom, base, token);
+        free(token);
+    }
+    return error;
+}
+
+static char *
+parse_key_value(const char **s, const struct ovsdb_type *type,
+                union ovsdb_atom *key, union ovsdb_atom *value)
+{
+    const char *start = *s;
+    char *error;
+
+    error = parse_atom_token(s, &type->key, key);
+    if (!error && type->value.type != OVSDB_TYPE_VOID) {
+        *s = skip_spaces(*s);
+        if (**s == '=') {
+            (*s)++;
+            *s = skip_spaces(*s);
+            error = parse_atom_token(s, &type->value, value);
+        } else {
+            error = xasprintf("%s: syntax error at \"%c\" expecting \"=\"",
+                              start, **s);
+        }
+        if (error) {
+            ovsdb_atom_destroy(key, type->key.type);
+        }
+    }
+    return error;
+}
+
+static void
+free_key_value(const struct ovsdb_type *type,
+               union ovsdb_atom *key, union ovsdb_atom *value)
+{
+    ovsdb_atom_destroy(key, type->key.type);
+    if (type->value.type != OVSDB_TYPE_VOID) {
+        ovsdb_atom_destroy(value, type->value.type);
+    }
+}
+
+/* Initializes 'datum' as a datum of the given 'type', parsing its contents
+ * from 's'.  The format of 's' is a series of space or comma separated atoms
+ * or, for a map, '='-delimited pairs of atoms.  Each atom must in a format
+ * acceptable to ovsdb_atom_from_string().  Optionally, a set may be enclosed
+ * in "[]" or a map in "{}"; for an empty set or map these punctuators are
+ * required. */
+char *
+ovsdb_datum_from_string(struct ovsdb_datum *datum,
+                        const struct ovsdb_type *type, const char *s)
+{
+    bool is_map = ovsdb_type_is_map(type);
+    struct ovsdb_error *dberror;
+    const char *p;
+    int end_delim;
+    char *error;
+
+    ovsdb_datum_init_empty(datum);
+
+    /* Swallow a leading delimiter if there is one. */
+    p = skip_spaces(s);
+    if (*p == (is_map ? '{' : '[')) {
+        end_delim = is_map ? '}' : ']';
+        p = skip_spaces(p + 1);
+    } else if (!*p) {
+        if (is_map) {
+            return xstrdup("use \"{}\" to specify the empty map");
+        } else {
+            return xstrdup("use \"[]\" to specify the empty set");
+        }
+    } else {
+        end_delim = 0;
+    }
+
+    while (*p && *p != end_delim) {
+        union ovsdb_atom key, value;
+
+        if (ovsdb_token_is_delim(*p)) {
+            error = xasprintf("%s: unexpected \"%c\" parsing %s",
+                              s, *p, ovsdb_type_to_english(type));
+            goto error;
+        }
+
+        /* Add to datum. */
+        error = parse_key_value(&p, type, &key, &value);
+        if (error) {
+            goto error;
+        }
+        ovsdb_datum_add_unsafe(datum, &key, &value, type);
+        free_key_value(type, &key, &value);
+
+        /* Skip optional white space and comma. */
+        p = skip_spaces(p);
+        if (*p == ',') {
+            p = skip_spaces(p + 1);
+        }
+    }
+
+    if (*p != end_delim) {
+        error = xasprintf("%s: missing \"%c\" at end of data", s, end_delim);
+        goto error;
+    }
+    if (end_delim) {
+        p = skip_spaces(p + 1);
+        if (*p) {
+            error = xasprintf("%s: trailing garbage after \"%c\"",
+                              s, end_delim);
+            goto error;
+        }
+    }
+
+    if (datum->n < type->n_min) {
+        error = xasprintf("%s: %u %s specified but the minimum number is %u",
+                          s, datum->n, is_map ? "pair(s)" : "value(s)",
+                          type->n_min);
+        goto error;
+    } else if (datum->n > type->n_max) {
+        error = xasprintf("%s: %u %s specified but the maximum number is %u",
+                          s, datum->n, is_map ? "pair(s)" : "value(s)",
+            type->n_max);
+        goto error;
+    }
+
+    dberror = ovsdb_datum_sort(datum, type);
+    if (dberror) {
+        ovsdb_error_destroy(dberror);
+        if (ovsdb_type_is_map(type)) {
+            error = xasprintf("%s: map contains duplicate key", s);
+        } else {
+            error = xasprintf("%s: set contains duplicate value", s);
+        }
+        goto error;
+    }
+
+    return NULL;
+
+error:
+    ovsdb_datum_destroy(datum, type);
+    ovsdb_datum_init_empty(datum);
+    return error;
+}
+
+/* Appends to 'out' the 'datum' (with the given 'type') in a format acceptable
+ * to ovsdb_datum_from_string(). */
+void
+ovsdb_datum_to_string(const struct ovsdb_datum *datum,
+                      const struct ovsdb_type *type, struct ds *out)
+{
+    bool is_map = ovsdb_type_is_map(type);
+    size_t i;
+
+    if (type->n_max > 1 || !datum->n) {
+        ds_put_char(out, is_map ? '{' : '[');
+    }
+    for (i = 0; i < datum->n; i++) {
+        if (i > 0) {
+            ds_put_cstr(out, ", ");
+        }
+
+        ovsdb_atom_to_string(&datum->keys[i], type->key.type, out);
+        if (is_map) {
+            ds_put_char(out, '=');
+            ovsdb_atom_to_string(&datum->values[i], type->value.type, out);
+        }
+    }
+    if (type->n_max > 1 || !datum->n) {
+        ds_put_char(out, is_map ? '}' : ']');
+    }
+}
+
 static uint32_t
 hash_atoms(enum ovsdb_atomic_type type, const union ovsdb_atom *atoms,
            unsigned int n, uint32_t basis)
@@ -602,17 +1228,17 @@ uint32_t
 ovsdb_datum_hash(const struct ovsdb_datum *datum,
                  const struct ovsdb_type *type, uint32_t basis)
 {
-    basis = hash_atoms(type->key_type, datum->keys, datum->n, basis);
-    basis ^= (type->key_type << 24) | (type->value_type << 16) | datum->n;
-    basis = hash_atoms(type->value_type, datum->values, datum->n, basis);
+    basis = hash_atoms(type->key.type, datum->keys, datum->n, basis);
+    basis ^= (type->key.type << 24) | (type->value.type << 16) | datum->n;
+    basis = hash_atoms(type->value.type, datum->values, datum->n, basis);
     return basis;
 }
 
 static int
 atom_arrays_compare_3way(const union ovsdb_atom *a,
-                  const union ovsdb_atom *b,
-                  enum ovsdb_atomic_type type,
-                  size_t n)
+                         const union ovsdb_atom *b,
+                         enum ovsdb_atomic_type type,
+                         size_t n)
 {
     unsigned int i;
 
@@ -645,45 +1271,77 @@ ovsdb_datum_compare_3way(const struct ovsdb_datum *a,
         return a->n < b->n ? -1 : 1;
     }
 
-    cmp = atom_arrays_compare_3way(a->keys, b->keys, type->key_type, a->n);
+    cmp = atom_arrays_compare_3way(a->keys, b->keys, type->key.type, a->n);
     if (cmp) {
         return cmp;
     }
 
-    return (type->value_type == OVSDB_TYPE_VOID ? 0
-            : atom_arrays_compare_3way(a->values, b->values, type->value_type,
+    return (type->value.type == OVSDB_TYPE_VOID ? 0
+            : atom_arrays_compare_3way(a->values, b->values, type->value.type,
                                        a->n));
 }
 
-/* If atom 'i' in 'a' is also in 'b', returns its index in 'b', otherwise
- * UINT_MAX.  'type' must be the type of 'a' and 'b', except that
- * type->value_type may be set to OVSDB_TYPE_VOID to compare keys but not
- * values. */
-static unsigned int
-ovsdb_datum_find(const struct ovsdb_datum *a, int i,
-                 const struct ovsdb_datum *b,
-                 const struct ovsdb_type *type)
+/* If 'key' is one of the keys in 'datum', returns its index within 'datum',
+ * otherwise UINT_MAX.  'key.type' must be the type of the atoms stored in the
+ * 'keys' array in 'datum'.
+ */
+unsigned int
+ovsdb_datum_find_key(const struct ovsdb_datum *datum,
+                     const union ovsdb_atom *key,
+                     enum ovsdb_atomic_type key_type)
 {
-    int low = 0;
-    int high = b->n;
+    unsigned int low = 0;
+    unsigned int high = datum->n;
     while (low < high) {
-        int j = (low + high) / 2;
-        int cmp = ovsdb_atom_compare_3way(&a->keys[i], &b->keys[j],
-                                          type->key_type);
+        unsigned int idx = (low + high) / 2;
+        int cmp = ovsdb_atom_compare_3way(key, &datum->keys[idx], key_type);
         if (cmp < 0) {
-            high = j;
+            high = idx;
         } else if (cmp > 0) {
-            low = j + 1;
+            low = idx + 1;
         } else {
-            bool eq_value = (type->value_type == OVSDB_TYPE_VOID
-                             || ovsdb_atom_equals(&a->values[i], &b->values[j],
-                                                  type->value_type));
-            return eq_value ? j : UINT_MAX;
+            return idx;
         }
     }
     return UINT_MAX;
 }
 
+/* If 'key' and 'value' is one of the key-value pairs in 'datum', returns its
+ * index within 'datum', otherwise UINT_MAX.  'key.type' must be the type of
+ * the atoms stored in the 'keys' array in 'datum'.  'value_type' may be the
+ * type of the 'values' atoms or OVSDB_TYPE_VOID to compare only keys.
+ */
+unsigned int
+ovsdb_datum_find_key_value(const struct ovsdb_datum *datum,
+                           const union ovsdb_atom *key,
+                           enum ovsdb_atomic_type key_type,
+                           const union ovsdb_atom *value,
+                           enum ovsdb_atomic_type value_type)
+{
+    unsigned int idx = ovsdb_datum_find_key(datum, key, key_type);
+    if (idx != UINT_MAX
+        && value_type != OVSDB_TYPE_VOID
+        && !ovsdb_atom_equals(&datum->values[idx], value, value_type)) {
+        idx = UINT_MAX;
+    }
+    return idx;
+}
+
+/* If atom 'i' in 'a' is also in 'b', returns its index in 'b', otherwise
+ * UINT_MAX.  'type' must be the type of 'a' and 'b', except that
+ * type->value.type may be set to OVSDB_TYPE_VOID to compare keys but not
+ * values. */
+static unsigned int
+ovsdb_datum_find(const struct ovsdb_datum *a, int i,
+                 const struct ovsdb_datum *b,
+                 const struct ovsdb_type *type)
+{
+    return ovsdb_datum_find_key_value(b,
+                                      &a->keys[i], type->key.type,
+                                      a->values ? &a->values[i] : NULL,
+                                      type->value.type);
+}
+
 /* Returns true if every element in 'a' is also in 'b', false otherwise. */
 bool
 ovsdb_datum_includes_all(const struct ovsdb_datum *a,
@@ -721,46 +1379,81 @@ ovsdb_datum_reallocate(struct ovsdb_datum *a, const struct ovsdb_type *type,
                        unsigned int capacity)
 {
     a->keys = xrealloc(a->keys, capacity * sizeof *a->keys);
-    if (type->value_type != OVSDB_TYPE_VOID) {
+    if (type->value.type != OVSDB_TYPE_VOID) {
         a->values = xrealloc(a->values, capacity * sizeof *a->values);
     }
 }
 
-static void
-ovsdb_datum_remove(struct ovsdb_datum *a, size_t i,
-                   const struct ovsdb_type *type)
+/* Removes the element with index 'idx' from 'datum', which has type 'type'.
+ * If 'idx' is not the last element in 'datum', then the removed element is
+ * replaced by the (former) last element.
+ *
+ * This function does not maintain ovsdb_datum invariants.  Use
+ * ovsdb_datum_sort() to check and restore these invariants. */
+void
+ovsdb_datum_remove_unsafe(struct ovsdb_datum *datum, size_t idx,
+                          const struct ovsdb_type *type)
 {
-    ovsdb_atom_destroy(&a->keys[i], type->key_type);
-    a->keys[i] = a->keys[a->n - 1];
-    if (type->value_type != OVSDB_TYPE_VOID) {
-        ovsdb_atom_destroy(&a->values[i], type->value_type);
-        a->values[i] = a->values[a->n - 1];
+    ovsdb_atom_destroy(&datum->keys[idx], type->key.type);
+    datum->keys[idx] = datum->keys[datum->n - 1];
+    if (type->value.type != OVSDB_TYPE_VOID) {
+        ovsdb_atom_destroy(&datum->values[idx], type->value.type);
+        datum->values[idx] = datum->values[datum->n - 1];
     }
-    a->n--;
+    datum->n--;
 }
 
+/* Adds the element with the given 'key' and 'value' to 'datum', which must
+ * have the specified 'type'.
+ *
+ * This function always allocates memory, so it is not an efficient way to add
+ * a number of elements to a datum.
+ *
+ * This function does not maintain ovsdb_datum invariants.  Use
+ * ovsdb_datum_sort() to check and restore these invariants.  (But a datum with
+ * 0 or 1 elements cannot violate the invariants anyhow.) */
 void
-ovsdb_datum_union(struct ovsdb_datum *a,
-                  const struct ovsdb_datum *b, const struct ovsdb_type *type)
+ovsdb_datum_add_unsafe(struct ovsdb_datum *datum,
+                       const union ovsdb_atom *key,
+                       const union ovsdb_atom *value,
+                       const struct ovsdb_type *type)
+{
+    size_t idx = datum->n++;
+    datum->keys = xrealloc(datum->keys, datum->n * sizeof *datum->keys);
+    ovsdb_atom_clone(&datum->keys[idx], key, type->key.type);
+    if (type->value.type != OVSDB_TYPE_VOID) {
+        datum->values = xrealloc(datum->values,
+                                 datum->n * sizeof *datum->values);
+        ovsdb_atom_clone(&datum->values[idx], value, type->value.type);
+    }
+}
+
+void
+ovsdb_datum_union(struct ovsdb_datum *a, const struct ovsdb_datum *b,
+                  const struct ovsdb_type *type, bool replace)
 {
-    struct ovsdb_type type_without_value;
     unsigned int n;
-    size_t i;
+    size_t bi;
 
-    type_without_value = *type;
-    type_without_value.value_type = OVSDB_TYPE_VOID;
     n = a->n;
-    for (i = 0; i < b->n; i++) {
-        if (ovsdb_datum_find(b, i, a, &type_without_value) == UINT_MAX) {
+    for (bi = 0; bi < b->n; bi++) {
+        unsigned int ai;
+
+        ai = ovsdb_datum_find_key(a, &b->keys[bi], type->key.type);
+        if (ai == UINT_MAX) {
             if (n == a->n) {
-                ovsdb_datum_reallocate(a, type, a->n + (b->n - i));
+                ovsdb_datum_reallocate(a, type, a->n + (b->n - bi));
             }
-            ovsdb_atom_clone(&a->keys[n], &b->keys[i], type->key_type);
-            if (type->value_type != OVSDB_TYPE_VOID) {
-                ovsdb_atom_clone(&a->values[n], &b->values[i],
-                                 type->value_type);
+            ovsdb_atom_clone(&a->keys[n], &b->keys[bi], type->key.type);
+            if (type->value.type != OVSDB_TYPE_VOID) {
+                ovsdb_atom_clone(&a->values[n], &b->values[bi],
+                                 type->value.type);
             }
             n++;
+        } else if (replace && type->value.type != OVSDB_TYPE_VOID) {
+            ovsdb_atom_destroy(&a->values[ai], type->value.type);
+            ovsdb_atom_clone(&a->values[ai], &b->values[bi],
+                             type->value.type);
         }
     }
     if (n != a->n) {
@@ -779,16 +1472,16 @@ ovsdb_datum_subtract(struct ovsdb_datum *a, const struct ovsdb_type *a_type,
     bool changed = false;
     size_t i;
 
-    assert(a_type->key_type == b_type->key_type);
-    assert(a_type->value_type == b_type->value_type
-           || b_type->value_type == OVSDB_TYPE_VOID);
+    assert(a_type->key.type == b_type->key.type);
+    assert(a_type->value.type == b_type->value.type
+           || b_type->value.type == OVSDB_TYPE_VOID);
 
     /* XXX The big-O of this could easily be improved. */
     for (i = 0; i < a->n; ) {
         unsigned int idx = ovsdb_datum_find(a, i, b, b_type);
         if (idx != UINT_MAX) {
             changed = true;
-            ovsdb_datum_remove(a, i, a_type);
+            ovsdb_datum_remove_unsafe(a, i, a_type);
         } else {
             i++;
         }
@@ -846,3 +1539,69 @@ ovsdb_symbol_table_put(struct ovsdb_symbol_table *symtab, const char *name,
     symbol->used = used;
     shash_add(&symtab->sh, name, symbol);
 }
+
+/* Extracts a token from the beginning of 's' and returns a pointer just after
+ * the token.  Stores the token itself into '*outp', which the caller is
+ * responsible for freeing (with free()).
+ *
+ * If 's[0]' is a delimiter, the returned token is the empty string.
+ *
+ * A token extends from 's' to the first delimiter, as defined by
+ * ovsdb_token_is_delim(), or until the end of the string.  A delimiter can be
+ * escaped with a backslash, in which case the backslash does not appear in the
+ * output.  Double quotes also cause delimiters to be ignored, but the double
+ * quotes are retained in the output.  (Backslashes inside double quotes are
+ * not removed, either.)
+ */
+char *
+ovsdb_token_parse(const char **s, char **outp)
+{
+    const char *p;
+    struct ds out;
+    bool in_quotes;
+    char *error;
+
+    ds_init(&out);
+    in_quotes = false;
+    for (p = *s; *p != '\0'; ) {
+        int c = *p++;
+        if (c == '\\') {
+            if (in_quotes) {
+                ds_put_char(&out, '\\');
+            }
+            if (!*p) {
+                error = xasprintf("%s: backslash at end of argument", *s);
+                goto error;
+            }
+            ds_put_char(&out, *p++);
+        } else if (!in_quotes && ovsdb_token_is_delim(c)) {
+            p--;
+            break;
+        } else {
+            ds_put_char(&out, c);
+            if (c == '"') {
+                in_quotes = !in_quotes;
+            }
+        }
+    }
+    if (in_quotes) {
+        error = xasprintf("%s: quoted string extends past end of argument",
+                          *s);
+        goto error;
+    }
+    *outp = ds_cstr(&out);
+    *s = p;
+    return NULL;
+
+error:
+    ds_destroy(&out);
+    *outp = NULL;
+    return error;
+}
+
+/* Returns true if 'c' delimits tokens, or if 'c' is 0, and false otherwise. */
+bool
+ovsdb_token_is_delim(unsigned char c)
+{
+    return strchr(":=, []{}", c) != NULL;
+}