Skip to contentSkip to conversation

tools/xenstore: do some cleanup and fixes

PATCH Superseded Juergen Gross 3 participants 44 messages Dec 13, 2022 – Jan 13, 2023

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

v2 cover tools/xenstore: do some cleanup and fixes

This is a first run of post-XSA patches which piled up during the development phase of all the recent Xenstore related XSA patches.

At least the first 5 patches are completely independent from each other. After those the dependencies are starting to be more complex.

This is a mixture of small fixes, enhancements and cleanups.

Changes in V2:
- patches 1+2 of V1 have been applied already
- addressed comments
- new patch 19

Juergen Gross (19):
tools/xenstore: let talloc_free() preserve errno tools/xenstore: let tdb_logger() preserve errno tools/xenstore: preserve errno across corrupt() tools/xenstore: remove all watches when a domain has stopped tools/xenstore: enhance hashtable implementation tools/xenstore: add hashlist for finding struct domain by domid tools/xenstore: introduce dummy nodes for special watch paths tools/xenstore: replace watch->relative_path with a prefix length tools/xenstore: move changed domain handling tools/xenstore: change per-domain node accounting interface tools/xenstore: don't allow creating too many nodes in a transaction tools/xenstore: replace literal domid 0 with dom0_domid tools/xenstore: make domain_is_unprivileged() an inline function tools/xenstore: let chk_domain_generation() return a bool tools/xenstore: switch hashtable to use the talloc framework tools/xenstore: make log macro globally available tools/xenstore: introduce trace classes
tools/xenstore: let check_store() check the accounting data tools/xenstore: make output of "xenstore-control help" more pretty

 docs/misc/xenstore.txt                 |  10 +-
 tools/xenstore/hashtable.c             | 134 +++---
 tools/xenstore/hashtable.h             |  38 +-
 tools/xenstore/talloc.c                |  21 +-
 tools/xenstore/xenstored_control.c     |  36 +-
 tools/xenstore/xenstored_core.c        | 259 +++++++----
 tools/xenstore/xenstored_core.h        |  31 ++
 tools/xenstore/xenstored_domain.c      | 609 +++++++++++++------------
 tools/xenstore/xenstored_domain.h      |  21 +-
 tools/xenstore/xenstored_transaction.c |  76 +--
 tools/xenstore/xenstored_transaction.h |   7 +-
 tools/xenstore/xenstored_watch.c       |  44 +-
 12 files changed, 707 insertions(+), 579 deletions(-)
Signature · 1 line
2.35.3
#1 lore raw

PATCH v2 1/19 tools/xenstore: let talloc_free() preserve errno

Today talloc_free() is not guaranteed to preserve errno, especially in case a custom destructor is being used.

Change that by renaming talloc_free() to _talloc_free() in talloc.c and adding a wrapper to talloc.c.

This allows to remove some errno saving outside of talloc.c.

---
V2:
- drop wrapper (Julien Grall)

---
 tools/xenstore/talloc.c         | 21 +++++++++++++--------
 tools/xenstore/xenstored_core.c |  2 --
 2 files changed, 13 insertions(+), 10 deletions(-)
tools/xenstore/talloc.c +13 -8
index d7edcf3a93..23c3a23b19 100644
@@ -541,38 +541,39 @@ static void talloc_free_children(void *ptr)
 */
 int talloc_free(void *ptr)
 {
+	int saved_errno = errno;
 	struct talloc_chunk *tc;
 
 	if (ptr == NULL) {
-		return -1;
+		goto err;
 	}
 
 	tc = talloc_chunk_from_ptr(ptr);
 
 	if (tc->null_refs) {
 		tc->null_refs--;
-		return -1;
+		goto err;
 	}
 
 	if (tc->refs) {
 		talloc_reference_destructor(tc->refs);
-		return -1;
+		goto err;
 	}
 
 	if (tc->flags & TALLOC_FLAG_LOOP) {
 		/* we have a free loop - stop looping */
-		return 0;
+		goto success;
 	}
 
 	if (tc->destructor) {
 		talloc_destructor_t d = tc->destructor;
 		if (d == (talloc_destructor_t)-1) {
-			return -1;
+			goto err;
 		}
 		tc->destructor = (talloc_destructor_t)-1;
 		if (d(ptr) == -1) {
 			tc->destructor = d;
-			return -1;
+			goto err;
 		}
 		tc->destructor = NULL;
 	}
@@ -594,10 +595,14 @@ int talloc_free(void *ptr)
 	tc->flags |= TALLOC_FLAG_FREE;
 
 	free(tc);
+ success:
+	errno = saved_errno;
 	return 0;
-}
-
 
+ err:
+	errno = saved_errno;
+	return -1;
+}
 
 /*
   A talloc version of realloc. The context argument is only used if
tools/xenstore/xenstored_core.c +0 -2
index 476d5c6d51..5a174b9881 100644
@@ -771,9 +771,7 @@ struct node *read_node(struct connection *conn, const void *ctx,
 	return node;
 
  error:
-	err = errno;
 	talloc_free(node);
-	errno = err;
 	return NULL;
 }
 
Signature · 1 line
2.35.3
#2 lore raw

Re: [PATCH v2 01/19] tools/xenstore: let talloc_free() preserve errno

Quoted 5 lines · Juergen Gross

Today talloc_free() is not guaranteed to preserve errno, especially in case a custom destructor is being used.

Change that by renaming talloc_free() to _talloc_free() in talloc.c and adding a wrapper to talloc.c.

This looks to be stale ...

Quoted 6 lines

This allows to remove some errno saving outside of talloc.c.

---
V2:
- drop wrapper (Julien Grall)

... after this change?

Jan

#3 lore raw

Re: [PATCH v2 01/19] tools/xenstore: let talloc_free() preserve errno

Quoted 9 lines · Jan Beulich
Quoted 5 lines · Juergen Gross

Today talloc_free() is not guaranteed to preserve errno, especially in case a custom destructor is being used.

Change that by renaming talloc_free() to _talloc_free() in talloc.c and adding a wrapper to talloc.c.

This looks to be stale ...

Quoted 8 lines
Quoted 6 lines

This allows to remove some errno saving outside of talloc.c.

---
V2:
- drop wrapper (Julien Grall)

... after this change?

Oh yes, indeed.

It should be replaced by:

"So preserve errno in talloc_free()."

Juergen

#4 lore raw

PATCH v2 2/19 tools/xenstore: let tdb_logger() preserve errno

In reply to Juergen Gross

tdb_logger() is called by TDB for logging errors. As errno is checked often after doing the logging, tdb_logger() should preserve errno.

---
 tools/xenstore/xenstored_core.c | 3 +++
 1 file changed, 3 insertions(+)
tools/xenstore/xenstored_core.c +3 -0
index 5a174b9881..d48208ecfe 100644
@@ -2345,6 +2345,7 @@ static void tdb_logger(TDB_CONTEXT *tdb, int level, const char * fmt, ...)
 {
 	va_list ap;
 	char *s;
+	int saved_errno = errno;
 
 	va_start(ap, fmt);
 	s = talloc_vasprintf(NULL, fmt, ap);
@@ -2360,6 +2361,8 @@ static void tdb_logger(TDB_CONTEXT *tdb, int level, const char * fmt, ...)
 		trace("talloc failure during logging\n");
 		syslog(LOG_ERR, "talloc failure during logging\n");
 	}
+
+	errno = saved_errno;
 }
 
 void setup_structure(bool live_update)
Signature · 1 line
2.35.3
#5 lore raw

PATCH v2 3/19 tools/xenstore: preserve errno across corrupt()

In reply to Juergen Gross

Let corrupt() preserve errno in order to be able to simplify error handling in future.

This is rather easy as the errno value when entering corrupt() is saved already.

---
 tools/xenstore/xenstored_core.c | 2 ++
 1 file changed, 2 insertions(+)
tools/xenstore/xenstored_core.c +2 -0
index d48208ecfe..8c2cca62b7 100644
@@ -2549,6 +2549,8 @@ void corrupt(struct connection *conn, const char *fmt, ...)
 	talloc_free(str);
 
 	check_store();
+
+	errno = saved_errno;
 }
 
 #ifndef NO_SOCKETS
Signature · 1 line
2.35.3
#6 lore raw

PATCH v2 4/19 tools/xenstore: remove all watches when a domain has stopped

In reply to Juergen Gross

When a domain has been released by Xen tools, remove all its registered watches. This avoids sending watch events to the dead domain when all the nodes related to it are being removed by the Xen tools.

---
V2:
- move call to do_release() (Julien Grall)

---
 tools/xenstore/xenstored_domain.c | 3 +++
 1 file changed, 3 insertions(+)
tools/xenstore/xenstored_domain.c +3 -0
index aa86892fed..e669c89e94 100644
@@ -740,6 +740,9 @@ int do_release(const void *ctx, struct connection *conn,
 	if (IS_ERR(domain))
 		return -PTR_ERR(domain);
 
+	/* Avoid triggering watch events when the domain's nodes are deleted. */
+	conn_delete_all_watches(domain->conn);
+
 	talloc_free(domain->conn);
 
 	send_ack(conn, XS_RELEASE);
Signature · 1 line
2.35.3
#7 lore raw

Re: [PATCH v2 04/19] tools/xenstore: remove all watches when a domain has stopped

Hi Juergen,

Quoted 3 lines · Juergen Gross

When a domain has been released by Xen tools, remove all its registered watches. This avoids sending watch events to the dead domain when all the nodes related to it are being removed by the Xen tools.

AFAICT, the only user of the command in the tree is softreset. Would you be able to check this is still working as expected?

Quoted 14 lines

---
V2:
- move call to do_release() (Julien Grall)

---
  tools/xenstore/xenstored_domain.c | 3 +++
  1 file changed, 3 insertions(+)
tools/xenstore/xenstored_domain.c +0 -0
index aa86892fed..e669c89e94 100644

tools/xenstore/xenstored_domain.c int do_release(const void *ctx, struct connection *conn, · line 748

Quoted 9 lines
tools/xenstore/xenstored_domain.c +3 -0
@@ -740,6 +740,9 @@ int do_release(const void *ctx, struct connection *conn,
  	if (IS_ERR(domain))
  		return -PTR_ERR(domain);
 
+	/* Avoid triggering watch events when the domain's nodes are deleted. */
+	conn_delete_all_watches(domain->conn);
+
  	talloc_free(domain->conn);
 
  	send_ack(conn, XS_RELEASE);

Cheers,

Signature · 1 line
Julien Grall
#8 lore raw

Re: [PATCH v2 04/19] tools/xenstore: remove all watches when a domain has stopped

Quoted 8 lines · Julien Grall

Hi Juergen,

Quoted 3 lines · Juergen Gross

When a domain has been released by Xen tools, remove all its registered watches. This avoids sending watch events to the dead domain when all the nodes related to it are being removed by the Xen tools.

AFAICT, the only user of the command in the tree is softreset. Would you be able to check this is still working as expected?

Seems to work fine.

Juergen

Reviewed-by
#9 lore raw

Re: [PATCH v2 04/19] tools/xenstore: remove all watches when a domain has stopped

Hi Juergen,

Quoted 9 lines · Juergen Gross
Quoted 6 lines · Julien Grall
Quoted 3 lines · Juergen Gross

When a domain has been released by Xen tools, remove all its registered watches. This avoids sending watch events to the dead domain when all the nodes related to it are being removed by the Xen tools.

AFAICT, the only user of the command in the tree is softreset. Would you be able to check this is still working as expected?

Seems to work fine.

Thanks for the confirmation! You can add my reviewed-by:

Cheers,

Signature · 1 line
Julien Grall
#10 lore raw

PATCH v2 5/19 tools/xenstore: enhance hashtable implementation

In reply to Juergen Gross

Today it is possible to set a flag when calling hashtable_destroy() in order to specify whether the data associated with the hashtable entries should be freed or not. The keys of the entries will always be freed.

Change that by replacing the flag of hashtable_destroy() by two flags for create_hashtable() which will specify whether the data and/or the key of each entry should be freed or not.

This will enable users to have the key e.g. as part of the data.

Add a new function hashtable_iterate() to call a user specified function for each entry in the hashtable.

Add new primes to the primetable in order to support smaller sizes of the hashtable. The primes are selected according to:

https://planetmath.org/goodhashtableprimes

Update the URL in the source as the old one wasn't correct any longer.

---
V2:
- add const to hashtable_iterate() callback (Julien Grall)

---
 tools/xenstore/hashtable.c      | 66 +++++++++++++++++++++++----------
 tools/xenstore/hashtable.h      | 35 +++++++++++++++--
 tools/xenstore/xenstored_core.c |  7 ++--
 3 files changed, 82 insertions(+), 26 deletions(-)
tools/xenstore/hashtable.c +46 -20
index 6ac336eff1..299549c51e 100644
@@ -16,6 +16,7 @@ struct entry
 
 struct hashtable {
     unsigned int tablelength;
+    unsigned int flags;
     struct entry **table;
     unsigned int entrycount;
     unsigned int loadlimit;
@@ -25,12 +26,11 @@ struct hashtable {
 };
 
 /*
-Credit for primes table: Aaron Krowne
- http://br.endernet.org/~akrowne/
- http://planetmath.org/encyclopedia/GoodHashTablePrimes.html
-*/
+ * Credit for primes table: Aaron Krowne
+ * https://planetmath.org/goodhashtableprimes
+ */
 static const unsigned int primes[] = {
-53, 97, 193, 389,
+11, 23, 53, 97, 193, 389,
 769, 1543, 3079, 6151,
 12289, 24593, 49157, 98317,
 196613, 393241, 786433, 1572869,
@@ -52,7 +52,8 @@ indexFor(unsigned int tablelength, unsigned int hashvalue) {
 struct hashtable *
 create_hashtable(unsigned int minsize,
                  unsigned int (*hashf) (void*),
-                 int (*eqf) (void*,void*))
+                 int (*eqf) (void*,void*),
+                 unsigned int flags)
 {
     struct hashtable *h;
     unsigned int pindex, size = primes[0];
@@ -73,6 +74,7 @@ create_hashtable(unsigned int minsize,
         goto err1;
 
     h->tablelength  = size;
+    h->flags        = flags;
     h->primeindex   = pindex;
     h->entrycount   = 0;
     h->hashfn       = hashf;
@@ -235,7 +237,8 @@ hashtable_remove(struct hashtable *h, void *k)
             *pE = e->next;
             h->entrycount--;
             v = e->v;
-            free(e->k);
+            if (h->flags & HASHTABLE_FREE_KEY)
+                free(e->k);
             free(e);
             return v;
         }
@@ -246,29 +249,52 @@ hashtable_remove(struct hashtable *h, void *k)
 }
 
 /*****************************************************************************/
-/* destroy */
-void
-hashtable_destroy(struct hashtable *h, int free_values)
+int
+hashtable_iterate(struct hashtable *h,
+                  int (*func)(const void *k, void *v, void *arg), void *arg)
 {
+    int ret;
     unsigned int i;
     struct entry *e, *f;
     struct entry **table = h->table;
-    if (free_values)
+
+    for (i = 0; i < h->tablelength; i++)
     {
-        for (i = 0; i < h->tablelength; i++)
+        e = table[i];
+        while (e)
         {
-            e = table[i];
-            while (NULL != e)
-            { f = e; e = e->next; free(f->k); free(f->v); free(f); }
+            f = e;
+            e = e->next;
+            ret = func(f->k, f->v, arg);
+            if (ret)
+                return ret;
         }
     }
-    else
+
+    return 0;
+}
+
+/*****************************************************************************/
+/* destroy */
+void
+hashtable_destroy(struct hashtable *h)
+{
+    unsigned int i;
+    struct entry *e, *f;
+    struct entry **table = h->table;
+
+    for (i = 0; i < h->tablelength; i++)
     {
-        for (i = 0; i < h->tablelength; i++)
+        e = table[i];
+        while (NULL != e)
         {
-            e = table[i];
-            while (NULL != e)
-            { f = e; e = e->next; free(f->k); free(f); }
+            f = e;
+            e = e->next;
+            if (h->flags & HASHTABLE_FREE_KEY)
+                free(f->k);
+            if (h->flags & HASHTABLE_FREE_VALUE)
+                free(f->v);
+            free(f);
         }
     }
     free(h->table);
tools/xenstore/hashtable.h +32 -3
index 62fef6081a..6d65431f96 100644
@@ -12,13 +12,21 @@ struct hashtable;
  * @param   minsize         minimum initial size of hashtable
  * @param   hashfunction    function for hashing keys
  * @param   key_eq_fn       function for determining key equality
+ * @param   flags           flags HASHTABLE_*
  * @return                  newly created hashtable or NULL on failure
  */
 
+/* Let hashtable_destroy() free the entries' values. */
+#define HASHTABLE_FREE_VALUE (1U << 0)
+/* Let hashtable_remove() and hashtable_destroy() free the entries' keys. */
+#define HASHTABLE_FREE_KEY   (1U << 1)
+
 struct hashtable *
 create_hashtable(unsigned int minsize,
                  unsigned int (*hashfunction) (void*),
-                 int (*key_eq_fn) (void*,void*));
+                 int (*key_eq_fn) (void*,void*),
+                 unsigned int flags
+);
 
 /*****************************************************************************
  * hashtable_insert
@@ -76,16 +84,37 @@ hashtable_remove(struct hashtable *h, void *k);
 unsigned int
 hashtable_count(struct hashtable *h);
 
+/*****************************************************************************
+ * hashtable_iterate
+
+ * @name           hashtable_iterate
+ * @param   h      the hashtable
+ * @param   func   function to call for each entry
+ * @param   arg    user supplied parameter for func
+ * @return         0 if okay, non-zero return value of func (and iteration
+ *                 was aborted)
+ *
+ * Iterates over all entries in the hashtable and calls func with the
+ * key, value, and the user supplied parameter.
+ * func returning a non-zero value will abort the iteration. In case func is
+ * removing an entry other than itself from the hashtable, it must return a
+ * non-zero value in order to abort the iteration. Inserting entries is
+ * allowed, but it is undefined whether func will be called for those new
+ * entries during this iteration.
+ */
+int
+hashtable_iterate(struct hashtable *h,
+                  int (*func)(const void *k, void *v, void *arg), void *arg);
+
 /*****************************************************************************
  * hashtable_destroy
    
  * @name        hashtable_destroy
  * @param   h   the hashtable
- * @param       free_values     whether to call 'free' on the remaining values
  */
 
 void
-hashtable_destroy(struct hashtable *h, int free_values);
+hashtable_destroy(struct hashtable *h);
 
 #endif /* __HASHTABLE_CWC22_H__ */
 
tools/xenstore/xenstored_core.c +4 -3
index 8c2cca62b7..1650821922 100644
@@ -2512,7 +2512,9 @@ void check_store(void)
 		.enoent = check_store_enoent,
 	};
 
-	reachable = create_hashtable(16, hash_from_key_fn, keys_equal_fn);
+	/* Don't free values (they are all void *1) */
+	reachable = create_hashtable(16, hash_from_key_fn, keys_equal_fn,
+				     HASHTABLE_FREE_KEY);
 	if (!reachable) {
 		log("check_store: ENOMEM");
 		return;
@@ -2526,8 +2528,7 @@ void check_store(void)
 		clean_store(reachable);
 	log("Checking store complete.");
 
-	hashtable_destroy(reachable, 0 /* Don't free values (they are all
-					  (void *)1) */);
+	hashtable_destroy(reachable);
 }
 
 
Signature · 1 line
2.35.3
#11 lore raw

PATCH v2 6/19 tools/xenstore: add hashlist for finding struct domain by domid

In reply to Juergen Gross

Today finding a struct domain by its domain id requires to scan the list of domains until finding the correct domid.

Add a hashlist for being able to speed this up. This allows to remove the linking of struct domain in a list. Note that the list of changed domains per transaction is kept as a list, as there are no known use cases with more than 4 domains being touched in a single transaction (this would be a device handled by a driver domain and being assigned to a HVM domain with device model in a stubdom, plus the control domain).

Some simple performance tests comparing the scanning and hashlist have shown that the hashlist will win as soon as more than 6 entries need to be scanned.

---
V2:
- add comment, fix return value of check_domain() (Julien Grall)

---
 tools/xenstore/xenstored_domain.c | 102 ++++++++++++++++++------------
 1 file changed, 60 insertions(+), 42 deletions(-)
tools/xenstore/xenstored_domain.c +60 -42
index e669c89e94..3ad1028edb 100644
@@ -48,8 +48,6 @@ static struct node_perms dom_introduce_perms;
 
 struct domain
 {
-	struct list_head list;
-
 	/* The id of this domain */
 	unsigned int domid;
 
@@ -96,7 +94,7 @@ struct domain
 	bool wrl_delay_logged;
 };
 
-static LIST_HEAD(domains);
+static struct hashtable *domhash;
 
 static bool check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
 {
@@ -309,7 +307,7 @@ static int destroy_domain(void *_domain)
 
 	domain_tree_remove(domain);
 
-	list_del(&domain->list);
+	hashtable_remove(domhash, &domain->domid);
 
 	if (!domain->introduced)
 		return 0;
@@ -341,43 +339,50 @@ static bool get_domain_info(unsigned int domid, xc_dominfo_t *dominfo)
 	       dominfo->domid == domid;
 }
 
-void check_domains(void)
+static int check_domain(const void *k, void *v, void *arg)
 {
 	xc_dominfo_t dominfo;
-	struct domain *domain;
 	struct connection *conn;
-	int notify = 0;
 	bool dom_valid;
+	struct domain *domain = v;
+	bool *notify = arg;
 
- again:
-	list_for_each_entry(domain, &domains, list) {
-		dom_valid = get_domain_info(domain->domid, &dominfo);
-		if (!domain->introduced) {
-			if (!dom_valid) {
-				talloc_free(domain);
-				goto again;
-			}
-			continue;
-		}
-		if (dom_valid) {
-			if ((dominfo.crashed || dominfo.shutdown)
-			    && !domain->shutdown) {
-				domain->shutdown = true;
-				notify = 1;
-			}
-			if (!dominfo.dying)
-				continue;
-		}
-		if (domain->conn) {
-			/* domain is a talloc child of domain->conn. */
-			conn = domain->conn;
-			domain->conn = NULL;
-			talloc_unlink(talloc_autofree_context(), conn);
-			notify = 0; /* destroy_domain() fires the watch */
-			goto again;
+	dom_valid = get_domain_info(domain->domid, &dominfo);
+	if (!domain->introduced) {
+		if (!dom_valid)
+			talloc_free(domain);
+		return 0;
+	}
+	if (dom_valid) {
+		if ((dominfo.crashed || dominfo.shutdown)
+		    && !domain->shutdown) {
+			domain->shutdown = true;
+			*notify = true;
 		}
+		if (!dominfo.dying)
+			return 0;
+	}
+	if (domain->conn) {
+		/* domain is a talloc child of domain->conn. */
+		conn = domain->conn;
+		domain->conn = NULL;
+		talloc_unlink(talloc_autofree_context(), conn);
+		*notify = false; /* destroy_domain() fires the watch */
+
+		/* Above unlink might result in 2 domains being freed! */
+		return 1;
 	}
 
+	return 0;
+}
+
+void check_domains(void)
+{
+	bool notify = false;
+
+	while (hashtable_iterate(domhash, check_domain, &notify))
+		;
+
 	if (notify)
 		fire_watches(NULL, NULL, "@releaseDomain", NULL, true, NULL);
 }
@@ -415,13 +420,7 @@ static char *talloc_domain_path(const void *context, unsigned int domid)
 
 static struct domain *find_domain_struct(unsigned int domid)
 {
-	struct domain *i;
-
-	list_for_each_entry(i, &domains, list) {
-		if (i->domid == domid)
-			return i;
-	}
-	return NULL;
+	return hashtable_search(domhash, &domid);
 }
 
 int domain_get_quota(const void *ctx, struct connection *conn,
@@ -470,9 +469,13 @@ static struct domain *alloc_domain(const void *context, unsigned int domid)
 	domain->generation = generation;
 	domain->introduced = false;
 
-	talloc_set_destructor(domain, destroy_domain);
+	if (!hashtable_insert(domhash, &domain->domid, domain)) {
+		talloc_free(domain);
+		errno = ENOMEM;
+		return NULL;
+	}
 
-	list_add(&domain->list, &domains);
+	talloc_set_destructor(domain, destroy_domain);
 
 	return domain;
 }
@@ -906,10 +909,25 @@ void dom0_init(void)
 	xenevtchn_notify(xce_handle, dom0->port);
 }
 
+static unsigned int domhash_fn(void *k)
+{
+	return *(unsigned int *)k;
+}
+
+static int domeq_fn(void *key1, void *key2)
+{
+	return *(unsigned int *)key1 == *(unsigned int *)key2;
+}
+
 void domain_init(int evtfd)
 {
 	int rc;
 
+	/* Start with a random rather low domain count for the hashtable. */
+	domhash = create_hashtable(8, domhash_fn, domeq_fn, 0);
+	if (!domhash)
+		barf_perror("Failed to allocate domain hashtable");
+
 	xc_handle = talloc(talloc_autofree_context(), xc_interface*);
 	if (!xc_handle)
 		barf_perror("Failed to allocate domain handle");
Signature · 1 line
2.35.3
Reviewed-by on v2 6/19
#12 lore raw

Re: [PATCH v2 06/19] tools/xenstore: add hashlist for finding struct domain by domid

Hi Juergen,

Quoted 11 lines · Juergen Gross

Today finding a struct domain by its domain id requires to scan the list of domains until finding the correct domid.

Add a hashlist for being able to speed this up. This allows to remove the linking of struct domain in a list. Note that the list of changed domains per transaction is kept as a list, as there are no known use cases with more than 4 domains being touched in a single transaction (this would be a device handled by a driver domain and being assigned to a HVM domain with device model in a stubdom, plus the control domain).

Quoted 5 lines

Some simple performance tests comparing the scanning and hashlist have shown that the hashlist will win as soon as more than 6 entries need to be scanned.

Cheers,

Signature · 1 line
Julien Grall
#13 lore raw

PATCH v2 7/19 tools/xenstore: introduce dummy nodes for special watch paths

In reply to Juergen Gross

Instead of special casing the permission handling and watch event firing for the special watch paths "@introduceDomain" and "@releaseDomain", use static dummy nodes added to the data base when starting Xenstore.

The node accounting needs to reflect that change by adding the special nodes in the domain_entry_fix() call in setup_structure().

Note that this requires to rework the calls of fire_watches() for the special events in order to avoid leaking memory.

Move the check for a valid node name from get_node() to get_node_canonicalized(), as it allows to use get_node() for the special nodes, too.

In order to avoid read and write accesses to the special nodes use a special variant for obtaining the current node data for the permission handling.

This allows to simplify quite some code. In future sub-nodes of the special nodes will be possible due to this change, allowing more fine grained permission control of special events for specific domains.

---
V2:
- add get_spec_node()
- expand commit message (Julien Grall)

---
 tools/xenstore/xenstored_control.c |   3 -
 tools/xenstore/xenstored_core.c    |  92 +++++++++-------
 tools/xenstore/xenstored_domain.c  | 162 ++++-------------------------
 tools/xenstore/xenstored_domain.h  |   6 --
 tools/xenstore/xenstored_watch.c   |  17 +--
 5 files changed, 80 insertions(+), 200 deletions(-)
tools/xenstore/xenstored_control.c +0 -3
index d1aaa00bf4..41e6992591 100644
@@ -676,9 +676,6 @@ static const char *lu_dump_state(const void *ctx, struct connection *conn)
 	if (ret)
 		goto out;
 	ret = dump_state_connections(fp);
-	if (ret)
-		goto out;
-	ret = dump_state_special_nodes(fp);
 	if (ret)
 		goto out;
 	ret = dump_state_nodes(fp, ctx);
tools/xenstore/xenstored_core.c +57 -35

152 lines not on this page · raw message

tools/xenstore/xenstored_domain.c +21 -141

216 lines not on this page · raw message

tools/xenstore/xenstored_domain.h +0 -6
index b38c82991d..630641d620 100644
@@ -99,11 +99,6 @@ void domain_outstanding_domid_dec(unsigned int domid);
 int domain_get_quota(const void *ctx, struct connection *conn,
 		     unsigned int domid);
 
-/* Special node permission handling. */
-int set_perms_special(struct connection *conn, const char *name,
-		      struct node_perms *perms);
-bool check_perms_special(const char *name, struct connection *conn);
-
 /* Write rate limiting */
 
 #define WRL_FACTOR   1000 /* for fixed-point arithmetic */
@@ -132,7 +127,6 @@ void wrl_apply_debit_direct(struct connection *conn);
 void wrl_apply_debit_trans_commit(struct connection *conn);
 
 const char *dump_state_connections(FILE *fp);
-const char *dump_state_special_nodes(FILE *fp);
 
 void read_state_connection(const void *ctx, const void *state);
 
tools/xenstore/xenstored_watch.c +2 -15
index 316c08b7f7..75748ac109 100644
@@ -46,13 +46,6 @@ struct watch
 	char *node;
 };
 
-static bool check_special_event(const char *name)
-{
-	assert(name);
-
-	return strstarts(name, "@");
-}
-
 /* Is child a subnode of parent, or equal? */
 static bool is_child(const char *child, const char *parent)
 {
@@ -153,14 +146,8 @@ void fire_watches(struct connection *conn, const void *ctx, const char *name,
 
 	/* Create an event for each watch. */
 	list_for_each_entry(i, &connections, list) {
-		/* introduce/release domain watches */
-		if (check_special_event(name)) {
-			if (!check_perms_special(name, i))
-				continue;
-		} else {
-			if (!watch_permitted(i, ctx, name, node, perms))
-				continue;
-		}
+		if (!watch_permitted(i, ctx, name, node, perms))
+			continue;
 
 		list_for_each_entry(watch, &i->watches, list) {
 			if (exact) {
Signature · 1 line
2.35.3
#14 lore raw

Re: [PATCH v2 07/19] tools/xenstore: introduce dummy nodes for special watch paths

Hi Juergen,

Quoted 55 lines · Juergen Gross

Instead of special casing the permission handling and watch event firing for the special watch paths "@introduceDomain" and "@releaseDomain", use static dummy nodes added to the data base when starting Xenstore.

The node accounting needs to reflect that change by adding the special nodes in the domain_entry_fix() call in setup_structure().

Note that this requires to rework the calls of fire_watches() for the special events in order to avoid leaking memory.

Move the check for a valid node name from get_node() to get_node_canonicalized(), as it allows to use get_node() for the special nodes, too.

In order to avoid read and write accesses to the special nodes use a special variant for obtaining the current node data for the permission handling.

This allows to simplify quite some code. In future sub-nodes of the special nodes will be possible due to this change, allowing more fine grained permission control of special events for specific domains.

---
V2:
- add get_spec_node()
- expand commit message (Julien Grall)

---
  tools/xenstore/xenstored_control.c |   3 -
  tools/xenstore/xenstored_core.c    |  92 +++++++++-------
  tools/xenstore/xenstored_domain.c  | 162 ++++-------------------------
  tools/xenstore/xenstored_domain.h  |   6 --
  tools/xenstore/xenstored_watch.c   |  17 +--
  5 files changed, 80 insertions(+), 200 deletions(-)
tools/xenstore/xenstored_control.c +0 -3
index d1aaa00bf4..41e6992591 100644
@@ -676,9 +676,6 @@ static const char *lu_dump_state(const void *ctx, struct connection *conn)
  	if (ret)
  		goto out;
  	ret = dump_state_connections(fp);
-	if (ret)
-		goto out;
-	ret = dump_state_special_nodes(fp);
  	if (ret)
  		goto out;
  	ret = dump_state_nodes(fp, ctx);
tools/xenstore/xenstored_core.c +0 -0
index 1650821922..f96714e1b8 100644

tools/xenstore/xenstored_core.c static void get_acc_data(TDB_DATA *key, struct node_account_data *acc) · line 619

Quoted 5 lines
tools/xenstore/xenstored_core.c +1 -1
@@ -616,7 +616,8 @@ static void get_acc_data(TDB_DATA *key, struct node_account_data *acc)
  static unsigned int get_acc_domid(struct connection *conn, TDB_DATA *key,
  				  unsigned int domid)
  {
-	return (!conn || key->dptr[0] == '/') ? domid : conn->id;
+	return (!conn || key->dptr[0] == '/' || key->dptr[0] == '@')

The comment on top of get_acc_domid() needs to be updated.

Quoted 3 lines
tools/xenstore/xenstored_core.c +1 -0
+	       ? domid : conn->id;
  }
 

[...]

tools/xenstore/xenstored_core.c

Quoted 12 lines
tools/xenstore/xenstored_core.c +12 -0
+static void fire_special_watches(const char *name)
+{
+	void *ctx = talloc_new(NULL);
+	struct node *node;
+
+	if (!ctx)
+		return;
+
+	node = read_node(NULL, ctx, name);
+
+	if (node)
+		fire_watches(NULL, ctx, name, node, true, NULL);

NIT: I would consider to log an error (maybe only once) if 'node' is NULL. The purpose is to help debugging Xenstored.

The rest of the code LGTM.

Cheers,

Signature · 1 line
Julien Grall
#15 lore raw

Re: [PATCH v2 07/19] tools/xenstore: introduce dummy nodes for special watch paths

Quoted 59 lines · Julien Grall

Hi Juergen,

Quoted 56 lines · Juergen Gross

Instead of special casing the permission handling and watch event firing for the special watch paths "@introduceDomain" and "@releaseDomain", use static dummy nodes added to the data base when starting Xenstore.

The node accounting needs to reflect that change by adding the special nodes in the domain_entry_fix() call in setup_structure().

Note that this requires to rework the calls of fire_watches() for the special events in order to avoid leaking memory.

Move the check for a valid node name from get_node() to get_node_canonicalized(), as it allows to use get_node() for the special nodes, too.

In order to avoid read and write accesses to the special nodes use a special variant for obtaining the current node data for the permission handling.

This allows to simplify quite some code. In future sub-nodes of the special nodes will be possible due to this change, allowing more fine grained permission control of special events for specific domains.

---
V2:
- add get_spec_node()
- expand commit message (Julien Grall)
---
  tools/xenstore/xenstored_control.c |   3 - tools/xenstore/xenstored_core.c    |  92 +++++++++------- tools/xenstore/xenstored_domain.c  | 162 ++++------------------------- tools/xenstore/xenstored_domain.h  |   6 -- tools/xenstore/xenstored_watch.c   |  17 +--

  5 files changed, 80 insertions(+), 200 deletions(-)
tools/xenstore/xenstored_control.c +0 -3
index d1aaa00bf4..41e6992591 100644
@@ -676,9 +676,6 @@ static const char *lu_dump_state(const void *ctx, struct 
 connection *conn)
       if (ret)
           goto out;
       ret = dump_state_connections(fp);
-    if (ret)
-        goto out;
-    ret = dump_state_special_nodes(fp);
       if (ret)
           goto out;

      ret = dump_state_nodes(fp, ctx);

tools/xenstore/xenstored_core.c +0 -0
index 1650821922..f96714e1b8 100644
@@ -616,7 +616,8 @@ static void get_acc_data(TDB_DATA *key, struct 
Quoted 8 lines

tools/xenstore/xenstored_core.c static void get_acc_data(TDB_DATA *key, struct node_account_data *acc) · line 619

Quoted 6 lines
tools/xenstore/xenstored_core.c +1 -1
 node_account_data *acc)
   static unsigned int get_acc_domid(struct connection *conn, TDB_DATA *key,
                     unsigned int domid)
   {
-    return (!conn || key->dptr[0] == '/') ? domid : conn->id;
+    return (!conn || key->dptr[0] == '/' || key->dptr[0] == '@')

The comment on top of get_acc_domid() needs to be updated.

Oh, yes.

Quoted 6 lines
Quoted 3 lines
tools/xenstore/xenstored_core.c +1 -0
+           ? domid : conn->id;
   }

[...]

Quoted 14 lines

tools/xenstore/xenstored_core.c

Quoted 12 lines
tools/xenstore/xenstored_core.c +12 -0
+static void fire_special_watches(const char *name)
+{
+    void *ctx = talloc_new(NULL);
+    struct node *node;
+
+    if (!ctx)
+        return;
+
+    node = read_node(NULL, ctx, name);
+
+    if (node)
+        fire_watches(NULL, ctx, name, node, true, NULL);

NIT: I would consider to log an error (maybe only once) if 'node' is NULL. The purpose is to help debugging Xenstored.

I think we can log it always, as this is really a bad problem.

Juergen

#16 lore raw

Re: [PATCH v2 07/19] tools/xenstore: introduce dummy nodes for special watch paths

Hi Juergen,

Quoted 16 lines · Juergen Gross
Quoted 6 lines · Julien Grall
Quoted 6 lines
tools/xenstore/xenstored_domain.c +6 -0
+static void fire_special_watches(const char *name)
+{
+    void *ctx = talloc_new(NULL);
+    struct node *node;
+
+    if (!ctx)
Quoted 8 lines

tools/xenstore/xenstored_domain.c static void domain_tree_remove(struct domain *domain) · line 291

Quoted 6 lines

+        return;

tools/xenstore/xenstored_domain.c +5 -0
+
+    node = read_node(NULL, ctx, name);
+
+    if (node)
+        fire_watches(NULL, ctx, name, node, true, NULL);

NIT: I would consider to log an error (maybe only once) if 'node' is NULL. The purpose is to help debugging Xenstored.

Quoted 1 line

I think we can log it always, as this is really a bad problem.

That works for me. If it is always logged, then I guess this would need to be a syslog message as well.

Cheers,

Signature · 1 line
Julien Grall
#17 lore raw

PATCH v2 8/19 tools/xenstore: replace watch->relative_path with a prefix length

In reply to Juergen Gross

Instead of storing a pointer to the path which is prepended to relative paths in struct watch, just use the length of the prepended path.

It should be noted that the now removed special case of the relative path being "" in get_watch_path() can't happen at all.

---
V2:
- don't open code get_watch_path() (Julien Grall)

---
 tools/xenstore/xenstored_watch.c | 27 ++++++++-------------------
 1 file changed, 8 insertions(+), 19 deletions(-)
tools/xenstore/xenstored_watch.c +8 -19
index 75748ac109..5c0f764781 100644
@@ -39,8 +39,8 @@ struct watch
 	/* Current outstanding events applying to this watch. */
 	struct list_head events;
 
-	/* Is this relative to connnection's implicit path? */
-	const char *relative_path;
+	/* Offset into path for skipping prefix (used for relative paths). */
+	unsigned int prefix_len;
 
 	char *token;
 	char *node;
@@ -66,15 +66,7 @@ static bool is_child(const char *child, const char *parent)
 
 static const char *get_watch_path(const struct watch *watch, const char *name)
 {
-	const char *path = name;
-
-	if (watch->relative_path) {
-		path += strlen(watch->relative_path);
-		if (*path == '/') /* Could be "" */
-			path++;
-	}
-
-	return path;
+	return name + watch->prefix_len;
 }
 
 /*
@@ -211,10 +203,7 @@ static struct watch *add_watch(struct connection *conn, char *path, char *token,
 			      no_quota_check))
 		goto nomem;
 
-	if (relative)
-		watch->relative_path = get_implicit_path(conn);
-	else
-		watch->relative_path = NULL;
+	watch->prefix_len = relative ? strlen(get_implicit_path(conn)) + 1 : 0;
 
 	INIT_LIST_HEAD(&watch->events);
 
@@ -313,19 +302,19 @@ const char *dump_state_watches(FILE *fp, struct connection *conn,
 			       unsigned int conn_id)
 {
 	const char *ret = NULL;
+	const char *watch_path;
 	struct watch *watch;
 	struct xs_state_watch sw;
 	struct xs_state_record_header head;
-	const char *path;
 
 	head.type = XS_STATE_TYPE_WATCH;
 
 	list_for_each_entry(watch, &conn->watches, list) {
 		head.length = sizeof(sw);
 
+		watch_path = get_watch_path(watch, watch->node);
 		sw.conn_id = conn_id;
-		path = get_watch_path(watch, watch->node);
-		sw.path_length = strlen(path) + 1;
+		sw.path_length = strlen(watch_path) + 1;
 		sw.token_length = strlen(watch->token) + 1;
 		head.length += sw.path_length + sw.token_length;
 		head.length = ROUNDUP(head.length, 3);
@@ -334,7 +323,7 @@ const char *dump_state_watches(FILE *fp, struct connection *conn,
 		if (fwrite(&sw, sizeof(sw), 1, fp) != 1)
 			return "Dump watch state error";
 
-		if (fwrite(path, sw.path_length, 1, fp) != 1)
+		if (fwrite(watch_path, sw.path_length, 1, fp) != 1)
 			return "Dump watch path error";
 		if (fwrite(watch->token, sw.token_length, 1, fp) != 1)
 			return "Dump watch token error";
Signature · 1 line
2.35.3
#18 lore raw

Re: [PATCH v2 08/19] tools/xenstore: replace watch->relative_path with a prefix length

Hi Juergen,

Quoted 6 lines · Juergen Gross
tools/xenstore/xenstored_watch.c +1 -0
@@ -313,19 +302,19 @@ const char *dump_state_watches(FILE *fp, struct connection *conn,
  			       unsigned int conn_id)
  {
  	const char *ret = NULL;
+	const char *watch_path;
  	struct watch *watch;

tools/xenstore/xenstored_watch.c const char *dump_state_watches(FILE *fp, struct connection *conn, · line 315

Quoted 10 lines
tools/xenstore/xenstored_watch.c +1 -1
  	struct xs_state_watch sw;
  	struct xs_state_record_header head;
-	const char *path;
 
  	head.type = XS_STATE_TYPE_WATCH;
 
  	list_for_each_entry(watch, &conn->watches, list) {
  		head.length = sizeof(sw);
 
+		watch_path = get_watch_path(watch, watch->node);

It is not clear to me why you call get_watch_path() earlier and also rename the variable.

I don't mind the new name, but it doesn't feel like it belongs to this patch as the code in duymp_state_watches() would not be changed otherwise.

Cheers,

Quoted 8 lines
tools/xenstore/xenstored_watch.c +1 -2
  		sw.conn_id = conn_id;
-		path = get_watch_path(watch, watch->node);
-		sw.path_length = strlen(path) + 1;
+		sw.path_length = strlen(watch_path) + 1;
  		sw.token_length = strlen(watch->token) + 1;
  		head.length += sw.path_length + sw.token_length;
  		head.length = ROUNDUP(head.length, 3);

tools/xenstore/xenstored_watch.c const char *dump_state_watches(FILE *fp, struct connection *conn, · line 329

Quoted 8 lines
tools/xenstore/xenstored_watch.c +1 -1
@@ -334,7 +323,7 @@ const char *dump_state_watches(FILE *fp, struct connection *conn,
  		if (fwrite(&sw, sizeof(sw), 1, fp) != 1)
  			return "Dump watch state error";
 
-		if (fwrite(path, sw.path_length, 1, fp) != 1)
+		if (fwrite(watch_path, sw.path_length, 1, fp) != 1)
  			return "Dump watch path error";
  		if (fwrite(watch->token, sw.token_length, 1, fp) != 1)
  			return "Dump watch token error";

Cheers,

Signature · 1 line
Julien Grall
#19 lore raw

Re: [PATCH v2 08/19] tools/xenstore: replace watch->relative_path with a prefix length

Quoted 20 lines · Julien Grall

Hi Juergen,

Quoted 4 lines · Juergen Gross
tools/xenstore/xenstored_watch.c +0 -0
@@ -313,19 +302,19 @@ const char *dump_state_watches(FILE *fp, struct 
 connection *conn,
                      unsigned int conn_id)
   {

tools/xenstore/xenstored_watch.c const char *dump_state_watches(FILE *fp, struct · line 313

Quoted 10 lines
tools/xenstore/xenstored_watch.c +2 -1
       const char *ret = NULL;
+    const char *watch_path;
       struct watch *watch;
       struct xs_state_watch sw;
       struct xs_state_record_header head;
-    const char *path;
       head.type = XS_STATE_TYPE_WATCH;
       list_for_each_entry(watch, &conn->watches, list) {
           head.length = sizeof(sw);
+        watch_path = get_watch_path(watch, watch->node);

It is not clear to me why you call get_watch_path() earlier and also rename the variable.

Quoted 1 line

I don't mind the new name, but it doesn't feel like it belongs to this patch as the code in duymp_state_watches() would not be changed otherwise.

Both changes are the result of V1 of the series. Will undo them.

Juergen

#20 lore raw

PATCH v2 9/19 tools/xenstore: move changed domain handling

In reply to Juergen Gross

Move all code related to struct changed_domain from xenstored_transaction.c to xenstored_domain.c.

This will be needed later in order to simplify the accounting data updates in cases of errors during a request.

Split the code to have a more generic base framework.

---
V2:
- remove unrelated change (Julien Grall)

---
 tools/xenstore/xenstored_domain.c      | 77 ++++++++++++++++++++++++++
 tools/xenstore/xenstored_domain.h      |  3 +
 tools/xenstore/xenstored_transaction.c | 64 ++-------------------
 3 files changed, 84 insertions(+), 60 deletions(-)
tools/xenstore/xenstored_domain.c +77 -0
index c881c93d1d..3216119e83 100644
@@ -91,6 +91,18 @@ struct domain
 	bool wrl_delay_logged;
 };
 
+struct changed_domain
+{
+	/* List of all changed domains. */
+	struct list_head list;
+
+	/* Identifier of the changed domain. */
+	unsigned int domid;
+
+	/* Amount by which this domain's nbentry field has changed. */
+	int nbentry;
+};
+
 static struct hashtable *domhash;
 
 static bool check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
@@ -541,6 +553,71 @@ static struct domain *find_domain_by_domid(unsigned int domid)
 	return (d && d->introduced) ? d : NULL;
 }
 
+int acc_fix_domains(struct list_head *head, bool update)
+{
+	struct changed_domain *cd;
+	int cnt;
+
+	list_for_each_entry(cd, head, list) {
+		cnt = domain_entry_fix(cd->domid, cd->nbentry, update);
+		if (!update) {
+			if (cnt >= quota_nb_entry_per_domain)
+				return ENOSPC;
+			if (cnt < 0)
+				return ENOMEM;
+		}
+	}
+
+	return 0;
+}
+
+static struct changed_domain *acc_find_changed_domain(struct list_head *head,
+						      unsigned int domid)
+{
+	struct changed_domain *cd;
+
+	list_for_each_entry(cd, head, list) {
+		if (cd->domid == domid)
+			return cd;
+	}
+
+	return NULL;
+}
+
+static struct changed_domain *acc_get_changed_domain(const void *ctx,
+						     struct list_head *head,
+						     unsigned int domid)
+{
+	struct changed_domain *cd;
+
+	cd = acc_find_changed_domain(head, domid);
+	if (cd)
+		return cd;
+
+	cd = talloc_zero(ctx, struct changed_domain);
+	if (!cd)
+		return NULL;
+
+	cd->domid = domid;
+	list_add_tail(&cd->list, head);
+
+	return cd;
+}
+
+int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
+			unsigned int domid)
+{
+	struct changed_domain *cd;
+
+	cd = acc_get_changed_domain(ctx, head, domid);
+	if (!cd)
+		return errno;
+
+	cd->nbentry += val;
+
+	return 0;
+}
+
 static void domain_conn_reset(struct domain *domain)
 {
 	struct connection *conn = domain->conn;
tools/xenstore/xenstored_domain.h +3 -0
index 630641d620..9e20d2b17d 100644
@@ -98,6 +98,9 @@ void domain_outstanding_dec(struct connection *conn);
 void domain_outstanding_domid_dec(unsigned int domid);
 int domain_get_quota(const void *ctx, struct connection *conn,
 		     unsigned int domid);
+int acc_fix_domains(struct list_head *head, bool update);
+int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
+			unsigned int domid);
 
 /* Write rate limiting */
 
tools/xenstore/xenstored_transaction.c +4 -60

104 lines not on this page · raw message

Signature · 1 line
2.35.3
Reviewed-by on v2 9/19
#21 lore raw

Re: [PATCH v2 09/19] tools/xenstore: move changed domain handling

Hi Juergen,

Quoted 9 lines · Juergen Gross

Move all code related to struct changed_domain from xenstored_transaction.c to xenstored_domain.c.

This will be needed later in order to simplify the accounting data updates in cases of errors during a request.

Split the code to have a more generic base framework.

Cheers,

Signature · 1 line
Julien Grall
#22 lore raw

PATCH v2 10/19 tools/xenstore: change per-domain node accounting interface

In reply to Juergen Gross

Rework the interface and the internals of the per-domain node accounting:

- rename the functions to domain_nbentry_*() in order to better match the related counter name

- switch from node pointer to domid as interface, as all nodes have the owner filled in

- use a common internal function for adding a value to the counter

For the transaction case add a helper function to get the list head of the per-transaction changed domains, enabling to eliminate the transaction_entry_*() functions.

---
 tools/xenstore/xenstored_core.c        |  22 ++---
 tools/xenstore/xenstored_domain.c      | 122 +++++++++++--------------
 tools/xenstore/xenstored_domain.h      |  10 +-
 tools/xenstore/xenstored_transaction.c |  15 +--
 tools/xenstore/xenstored_transaction.h |   7 +-
 5 files changed, 72 insertions(+), 104 deletions(-)
tools/xenstore/xenstored_core.c +11 -11
index f96714e1b8..61569cecbb 100644
@@ -1459,7 +1459,7 @@ static void destroy_node_rm(struct connection *conn, struct node *node)
 static int destroy_node(struct connection *conn, struct node *node)
 {
 	destroy_node_rm(conn, node);
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
 
 	/*
 	 * It is not possible to easily revert the changes in a transaction.
@@ -1498,7 +1498,7 @@ static struct node *create_node(struct connection *conn, const void *ctx,
 	for (i = node; i; i = i->parent) {
 		/* i->parent is set for each new node, so check quota. */
 		if (i->parent &&
-		    domain_entry(conn) >= quota_nb_entry_per_domain) {
+		    domain_nbentry(conn) >= quota_nb_entry_per_domain) {
 			ret = ENOSPC;
 			goto err;
 		}
@@ -1509,7 +1509,7 @@ static struct node *create_node(struct connection *conn, const void *ctx,
 
 		/* Account for new node */
 		if (i->parent) {
-			if (domain_entry_inc(conn, i)) {
+			if (domain_nbentry_inc(conn, i->perms.p[0].id)) {
 				destroy_node_rm(conn, i);
 				return NULL;
 			}
@@ -1662,7 +1662,7 @@ static int delnode_sub(const void *ctx, struct connection *conn,
 	watch_exact = strcmp(root, node->name);
 	fire_watches(conn, ctx, node->name, node, watch_exact, NULL);
 
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
 
 	return WALK_TREE_RM_CHILDENTRY;
 }
@@ -1802,25 +1802,25 @@ static int do_set_perms(const void *ctx, struct connection *conn,
 		return EPERM;
 
 	old_perms = node->perms;
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
 	node->perms = perms;
-	if (domain_entry_inc(conn, node)) {
+	if (domain_nbentry_inc(conn, node->perms.p[0].id)) {
 		node->perms = old_perms;
 		/*
 		 * This should never fail because we had a reference on the
 		 * domain before and Xenstored is single-threaded.
 		 */
-		domain_entry_inc(conn, node);
+		domain_nbentry_inc(conn, node->perms.p[0].id);
 		return ENOMEM;
 	}
 
 	if (write_node(conn, node, false)) {
 		int saved_errno = errno;
 
-		domain_entry_dec(conn, node);
+		domain_nbentry_dec(conn, node->perms.p[0].id);
 		node->perms = old_perms;
 		/* No failure possible as above. */
-		domain_entry_inc(conn, node);
+		domain_nbentry_inc(conn, node->perms.p[0].id);
 
 		errno = saved_errno;
 		return errno;
@@ -2392,7 +2392,7 @@ void setup_structure(bool live_update)
 		manual_node("/tool/xenstored", NULL);
 		manual_node("@releaseDomain", NULL);
 		manual_node("@introduceDomain", NULL);
-		domain_entry_fix(dom0_domid, 5, true);
+		domain_nbentry_fix(dom0_domid, 5, true);
 	}
 
 	check_store();
@@ -3400,7 +3400,7 @@ void read_state_node(const void *ctx, const void *state)
 	if (write_node_raw(NULL, &key, node, true))
 		barf("write node error restoring node");
 
-	if (domain_entry_inc(&conn, node))
+	if (domain_nbentry_inc(&conn, node->perms.p[0].id))
 		barf("node accounting error restoring node");
 
 	talloc_free(node);
tools/xenstore/xenstored_domain.c +52 -70

176 lines not on this page · raw message

tools/xenstore/xenstored_domain.h +4 -6
index 9e20d2b17d..1e402f2609 100644
@@ -66,10 +66,10 @@ int domain_adjust_node_perms(struct node *node);
 int domain_alloc_permrefs(struct node_perms *perms);
 
 /* Quota manipulation */
-int domain_entry_inc(struct connection *conn, struct node *);
-void domain_entry_dec(struct connection *conn, struct node *);
-int domain_entry_fix(unsigned int domid, int num, bool update);
-int domain_entry(struct connection *conn);
+int domain_nbentry_inc(struct connection *conn, unsigned int domid);
+int domain_nbentry_dec(struct connection *conn, unsigned int domid);
+int domain_nbentry_fix(unsigned int domid, int num, bool update);
+int domain_nbentry(struct connection *conn);
 int domain_memory_add(unsigned int domid, int mem, bool no_quota_check);
 
 /*
@@ -99,8 +99,6 @@ void domain_outstanding_domid_dec(unsigned int domid);
 int domain_get_quota(const void *ctx, struct connection *conn,
 		     unsigned int domid);
 int acc_fix_domains(struct list_head *head, bool update);
-int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
-			unsigned int domid);
 
 /* Write rate limiting */
 
tools/xenstore/xenstored_transaction.c +2 -13
index 89b92f0baf..82e5e66c18 100644
@@ -548,20 +548,9 @@ int do_transaction_end(const void *ctx, struct connection *conn,
 	return 0;
 }
 
-void transaction_entry_inc(struct transaction *trans, unsigned int domid)
+struct list_head *transaction_get_changed_domains(struct transaction *trans)
 {
-	if (acc_add_dom_nbentry(trans, &trans->changed_domains, 1, domid)) {
-		/* Let the transaction fail. */
-		trans->fail = true;
-	}
-}
-
-void transaction_entry_dec(struct transaction *trans, unsigned int domid)
-{
-	if (acc_add_dom_nbentry(trans, &trans->changed_domains, -1, domid)) {
-		/* Let the transaction fail. */
-		trans->fail = true;
-	}
+	return &trans->changed_domains;
 }
 
 void fail_transaction(struct transaction *trans)
tools/xenstore/xenstored_transaction.h +3 -4
index 3417303f94..b6f8cb7d0a 100644
@@ -36,10 +36,6 @@ int do_transaction_end(const void *ctx, struct connection *conn,
 
 struct transaction *transaction_lookup(struct connection *conn, uint32_t id);
 
-/* inc/dec entry number local to trans while changing a node */
-void transaction_entry_inc(struct transaction *trans, unsigned int domid);
-void transaction_entry_dec(struct transaction *trans, unsigned int domid);
-
 /* This node was accessed. */
 int __must_check access_node(struct connection *conn, struct node *node,
                              enum node_access_type type, TDB_DATA *key);
@@ -54,6 +50,9 @@ void transaction_prepend(struct connection *conn, const char *name,
 /* Mark the transaction as failed. This will prevent it to be committed. */
 void fail_transaction(struct transaction *trans);
 
+/* Get the list head of the changed domains. */
+struct list_head *transaction_get_changed_domains(struct transaction *trans);
+
 void conn_delete_all_transactions(struct connection *conn);
 int check_transactions(struct hashtable *hash);
 
Signature · 1 line
2.35.3
#23 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Hi,

Quoted 8 lines · Juergen Gross

Rework the interface and the internals of the per-domain node accounting:

- rename the functions to domain_nbentry_*() in order to better match the related counter name

- switch from node pointer to domid as interface, as all nodes have the owner filled in

The downside is now you have may place open-coding "...->perms->p[0].id". IHMO this is making the code more complicated. So can you introduce a few wrappers that would take a node and then convert to the owner?

Quoted 138 lines

- use a common internal function for adding a value to the counter

For the transaction case add a helper function to get the list head of the per-transaction changed domains, enabling to eliminate the transaction_entry_*() functions.

---
  tools/xenstore/xenstored_core.c        |  22 ++---
  tools/xenstore/xenstored_domain.c      | 122 +++++++++++--------------
  tools/xenstore/xenstored_domain.h      |  10 +-
  tools/xenstore/xenstored_transaction.c |  15 +--
  tools/xenstore/xenstored_transaction.h |   7 +-
  5 files changed, 72 insertions(+), 104 deletions(-)
tools/xenstore/xenstored_core.c +11 -11
index f96714e1b8..61569cecbb 100644
@@ -1459,7 +1459,7 @@ static void destroy_node_rm(struct connection *conn, struct node *node)
  static int destroy_node(struct connection *conn, struct node *node)
  {
  	destroy_node_rm(conn, node);
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
 
  	/*
  	 * It is not possible to easily revert the changes in a transaction.
@@ -1498,7 +1498,7 @@ static struct node *create_node(struct connection *conn, const void *ctx,
  	for (i = node; i; i = i->parent) {
  		/* i->parent is set for each new node, so check quota. */
  		if (i->parent &&
-		    domain_entry(conn) >= quota_nb_entry_per_domain) {
+		    domain_nbentry(conn) >= quota_nb_entry_per_domain) {
  			ret = ENOSPC;
  			goto err;
  		}
@@ -1509,7 +1509,7 @@ static struct node *create_node(struct connection *conn, const void *ctx,
 
  		/* Account for new node */
  		if (i->parent) {
-			if (domain_entry_inc(conn, i)) {
+			if (domain_nbentry_inc(conn, i->perms.p[0].id)) {
  				destroy_node_rm(conn, i);
  				return NULL;
  			}
@@ -1662,7 +1662,7 @@ static int delnode_sub(const void *ctx, struct connection *conn,
  	watch_exact = strcmp(root, node->name);
  	fire_watches(conn, ctx, node->name, node, watch_exact, NULL);
 
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
 
  	return WALK_TREE_RM_CHILDENTRY;
  }
@@ -1802,25 +1802,25 @@ static int do_set_perms(const void *ctx, struct connection *conn,
  		return EPERM;
 
  	old_perms = node->perms;
-	domain_entry_dec(conn, node);
+	domain_nbentry_dec(conn, node->perms.p[0].id);
  	node->perms = perms;
-	if (domain_entry_inc(conn, node)) {
+	if (domain_nbentry_inc(conn, node->perms.p[0].id)) {
  		node->perms = old_perms;
  		/*
  		 * This should never fail because we had a reference on the
  		 * domain before and Xenstored is single-threaded.
  		 */
-		domain_entry_inc(conn, node);
+		domain_nbentry_inc(conn, node->perms.p[0].id);
  		return ENOMEM;
  	}
 
  	if (write_node(conn, node, false)) {
  		int saved_errno = errno;
 
-		domain_entry_dec(conn, node);
+		domain_nbentry_dec(conn, node->perms.p[0].id);
  		node->perms = old_perms;
  		/* No failure possible as above. */
-		domain_entry_inc(conn, node);
+		domain_nbentry_inc(conn, node->perms.p[0].id);
 
  		errno = saved_errno;
  		return errno;
@@ -2392,7 +2392,7 @@ void setup_structure(bool live_update)
  		manual_node("/tool/xenstored", NULL);
  		manual_node("@releaseDomain", NULL);
  		manual_node("@introduceDomain", NULL);
-		domain_entry_fix(dom0_domid, 5, true);
+		domain_nbentry_fix(dom0_domid, 5, true);
  	}
 
  	check_store();
@@ -3400,7 +3400,7 @@ void read_state_node(const void *ctx, const void *state)
  	if (write_node_raw(NULL, &key, node, true))
  		barf("write node error restoring node");
 
-	if (domain_entry_inc(&conn, node))
+	if (domain_nbentry_inc(&conn, node->perms.p[0].id))
  		barf("node accounting error restoring node");
 
  	talloc_free(node);
tools/xenstore/xenstored_domain.c +4 -4
index 3216119e83..40b24056c5 100644
@@ -249,7 +249,7 @@ static int domain_tree_remove_sub(const void *ctx, struct connection *conn,
  		domain->nbentry--;
  		node->perms.p[0].id = priv_domid;
  		node->acc.memory = 0;
-		domain_entry_inc(NULL, node);
+		domain_nbentry_inc(NULL, priv_domid);
  		if (write_node_raw(NULL, &key, node, true)) {
  			/* That's unfortunate. We only can try to continue. */
  			syslog(LOG_ERR,
@@ -559,7 +559,7 @@ int acc_fix_domains(struct list_head *head, bool update)
  	int cnt;
 
  	list_for_each_entry(cd, head, list) {
-		cnt = domain_entry_fix(cd->domid, cd->nbentry, update);
+		cnt = domain_nbentry_fix(cd->domid, cd->nbentry, update);
  		if (!update) {
  			if (cnt >= quota_nb_entry_per_domain)
  				return ENOSPC;
@@ -604,18 +604,19 @@ static struct changed_domain *acc_get_changed_domain(const void *ctx,
  	return cd;
  }
 
-int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
-			unsigned int domid)
+static int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
+			       unsigned int domid)
  {
  	struct changed_domain *cd;
 

tools/xenstore/xenstored_domain.c static struct changed_domain *acc_get_changed_domain(const void *ctx, · line 619

Quoted 10 lines
tools/xenstore/xenstored_domain.c +3 -2
  	cd = acc_get_changed_domain(ctx, head, domid);
  	if (!cd)
-		return errno;
+		return 0;
 
+	errno = 0;
  	cd->nbentry += val;
 
-	return 0;
+	return cd->nbentry;

You just introduced this helper in the previous patch (i.e. #9). So can you get the interface correct from the start? This will make easier to review the series.

I don't mind too much if you add the static here. Although, it would have been nice if we avoid changing code just introduced.

Quoted 35 lines
tools/xenstore/xenstored_domain.c +0 -24
  }
 
  static void domain_conn_reset(struct domain *domain)
@@ -988,30 +989,6 @@ void domain_deinit(void)
  		xenevtchn_unbind(xce_handle, virq_port);
  }
 
-int domain_entry_inc(struct connection *conn, struct node *node)
-{
-	struct domain *d;
-	unsigned int domid;
-
-	if (!node->perms.p)
-		return 0;
-
-	domid = node->perms.p[0].id;
-
-	if (conn && conn->transaction) {
-		transaction_entry_inc(conn->transaction, domid);
-	} else {
-		d = (conn && domid == conn->id && conn->domain) ? conn->domain
-		    : find_or_alloc_existing_domain(domid);
-		if (d)
-			d->nbentry++;
-		else
-			return ENOMEM;
-	}
-
-	return 0;
-}
-
  /*
   * Check whether a domain was created before or after a specific generation
   * count (used for testing whether a node permission is older than a domain).

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1060

Quoted 6 lines
tools/xenstore/xenstored_domain.c +2 -1
@@ -1079,62 +1056,67 @@ int domain_adjust_node_perms(struct node *node)
  	return 0;
  }
 
-void domain_entry_dec(struct connection *conn, struct node *node)
+static int domain_nbentry_add(struct connection *conn, unsigned int domid,
+			      int add, bool dom_exists)

The name of the variable suggests that that if it is false then it doesn't exists. However, looking at how you use it, it is more a "Can struct domain be allocated?". So I would rename it to "dom_alloc_allowed" or similar.

Quoted 36 lines
tools/xenstore/xenstored_domain.c +18 -13
  {
  	struct domain *d;
-	unsigned int domid;
-
-	if (!node->perms.p)
-		return;
+	struct list_head *head;
+	int ret;
 
-	domid = node->perms.p ? node->perms.p[0].id : conn->id;
+	if (conn && domid == conn->id && conn->domain)
+		d = conn->domain;
+	else if (dom_exists) {
+		d = find_domain_struct(domid);
+		if (!d) {
+			errno = ENOENT;
+			corrupt(conn, "Missing domain %u\n", domid);
+			return -1;
+		}
+	} else {
+		d = find_or_alloc_existing_domain(domid);
+		if (!d) {
+			errno = ENOMEM;
+			return -1;
+		}
+	}
 
  	if (conn && conn->transaction) {
-		transaction_entry_dec(conn->transaction, domid);
-	} else {
-		d = (conn && domid == conn->id && conn->domain) ? conn->domain
-		    : find_domain_struct(domid);
-		if (d) {
-			d->nbentry--;
-		} else {
-			errno = ENOENT;

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1090

Quoted 10 lines
tools/xenstore/xenstored_domain.c +6 -3
-			corrupt(conn,
-				"Node \"%s\" owned by non-existing domain %u\n",
-				node->name, domid);
+		head = transaction_get_changed_domains(conn->transaction);
+		ret = acc_add_dom_nbentry(conn->transaction, head, add, domid);
+		if (errno) {
+			fail_transaction(conn->transaction);
+			return -1;
  		}
+		return d->nbentry + ret;

It is not entirely clear why you are return "d->nbentry + ret" here. If it is ...

Quoted 22 lines
tools/xenstore/xenstored_domain.c +7 -10
  	}
+
+	d->nbentry += add;
+
+	return d->nbentry;
  }
 
-int domain_entry_fix(unsigned int domid, int num, bool update)
+int domain_nbentry_inc(struct connection *conn, unsigned int domid)
  {
-	struct domain *d;
-	int cnt;
+	return (domain_nbentry_add(conn, domid, 1, false) < 0) ? errno : 0;
+}
 
-	if (update) {
-		d = find_domain_struct(domid);
-		assert(d);
-	} else {
-		/*
-		 * We are called first with update == false in order to catch
-		 * any error. So do a possible allocation and check for error

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1105

Quoted 10 lines
tools/xenstore/xenstored_domain.c +3 -7
-		 * only in this case, as in the case of update == true nothing
-		 * can go wrong anymore as the allocation already happened.
-		 */
-		d = find_or_alloc_existing_domain(domid);
-		if (!d)
-			return -1;
-	}
+int domain_nbentry_dec(struct connection *conn, unsigned int domid)
+{
+	return (domain_nbentry_add(conn, domid, -1, true) < 0) ? errno : 0;

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Such change of behavior should pointed in the commit message. But then I am not convinced this should be part of this commit which is mainly reworking an interface (e.g. no functional change is expected).

Cheers,

Signature · 1 line
Julien Grall
#24 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Quoted 6 lines · Julien Grall

Hi,

Quoted 3 lines · Juergen Gross

Rework the interface and the internals of the per-domain node accounting:

Quoted 7 lines
Quoted 5 lines

- rename the functions to domain_nbentry_*() in order to better match the related counter name

- switch from node pointer to domid as interface, as all nodes have the owner filled in

The downside is now you have may place open-coding "...->perms->p[0].id". IHMO this is making the code more complicated. So can you introduce a few wrappers that would take a node and then convert to the owner?

Okay.

Quoted 143 lines
Quoted 130 lines

- use a common internal function for adding a value to the counter

For the transaction case add a helper function to get the list head of the per-transaction changed domains, enabling to eliminate the transaction_entry_*() functions.

---
  tools/xenstore/xenstored_core.c        |  22 ++--- tools/xenstore/xenstored_domain.c      | 122 +++++++++++-------------- tools/xenstore/xenstored_domain.h      |  10 +- tools/xenstore/xenstored_transaction.c |  15 +-- tools/xenstore/xenstored_transaction.h |   7 +-

  5 files changed, 72 insertions(+), 104 deletions(-)
tools/xenstore/xenstored_core.c +1 -1
index f96714e1b8..61569cecbb 100644
@@ -1459,7 +1459,7 @@ static void destroy_node_rm(struct connection *conn, 
 struct node *node)
   static int destroy_node(struct connection *conn, struct node *node)
   {
       destroy_node_rm(conn, node);
-    domain_entry_dec(conn, node);
+    domain_nbentry_dec(conn, node->perms.p[0].id);
       /*
        * It is not possible to easily revert the changes in a transaction.
@@ -1498,7 +1498,7 @@ static struct node *create_node(struct connection *conn, 
 const void *ctx,
       for (i = node; i; i = i->parent) {
           /* i->parent is set for each new node, so check quota. */

          if (i->parent &&
-            domain_entry(conn) >= quota_nb_entry_per_domain) { +            domain_nbentry(conn) >= quota_nb_entry_per_domain) { ret = ENOSPC;
              goto err;
          }

tools/xenstore/xenstored_core.c +9 -9
@@ -1509,7 +1509,7 @@ static struct node *create_node(struct connection *conn, 
 const void *ctx,
           /* Account for new node */
           if (i->parent) {
-            if (domain_entry_inc(conn, i)) {
+            if (domain_nbentry_inc(conn, i->perms.p[0].id)) {
                   destroy_node_rm(conn, i);
                   return NULL;
               }
@@ -1662,7 +1662,7 @@ static int delnode_sub(const void *ctx, struct 
 connection *conn,
       watch_exact = strcmp(root, node->name);
       fire_watches(conn, ctx, node->name, node, watch_exact, NULL);
-    domain_entry_dec(conn, node);
+    domain_nbentry_dec(conn, node->perms.p[0].id);
       return WALK_TREE_RM_CHILDENTRY;
   }
@@ -1802,25 +1802,25 @@ static int do_set_perms(const void *ctx, struct 
 connection *conn,
           return EPERM;
       old_perms = node->perms;
-    domain_entry_dec(conn, node);
+    domain_nbentry_dec(conn, node->perms.p[0].id);
       node->perms = perms;
-    if (domain_entry_inc(conn, node)) {
+    if (domain_nbentry_inc(conn, node->perms.p[0].id)) {
           node->perms = old_perms;
           /*
            * This should never fail because we had a reference on the
            * domain before and Xenstored is single-threaded.
            */
-        domain_entry_inc(conn, node);
+        domain_nbentry_inc(conn, node->perms.p[0].id);
           return ENOMEM;
       }
       if (write_node(conn, node, false)) {
           int saved_errno = errno;
-        domain_entry_dec(conn, node);
+        domain_nbentry_dec(conn, node->perms.p[0].id);
           node->perms = old_perms;
           /* No failure possible as above. */
-        domain_entry_inc(conn, node);
+        domain_nbentry_inc(conn, node->perms.p[0].id);
           errno = saved_errno;
           return errno;
@@ -2392,7 +2392,7 @@ void setup_structure(bool live_update)
           manual_node("/tool/xenstored", NULL);
           manual_node("@releaseDomain", NULL);
           manual_node("@introduceDomain", NULL);
-        domain_entry_fix(dom0_domid, 5, true);
+        domain_nbentry_fix(dom0_domid, 5, true);
       }
       check_store();
@@ -3400,7 +3400,7 @@ void read_state_node(const void *ctx, const void *state)
       if (write_node_raw(NULL, &key, node, true))
           barf("write node error restoring node");
-    if (domain_entry_inc(&conn, node))
+    if (domain_nbentry_inc(&conn, node->perms.p[0].id))
           barf("node accounting error restoring node");
       talloc_free(node);
tools/xenstore/xenstored_domain.c +1 -1
index 3216119e83..40b24056c5 100644
@@ -249,7 +249,7 @@ static int domain_tree_remove_sub(const void *ctx, struct 
 connection *conn,
           domain->nbentry--;
           node->perms.p[0].id = priv_domid;
           node->acc.memory = 0;
-        domain_entry_inc(NULL, node);
+        domain_nbentry_inc(NULL, priv_domid);
           if (write_node_raw(NULL, &key, node, true)) {
               /* That's unfortunate. We only can try to continue. */

              syslog(LOG_ERR,

tools/xenstore/xenstored_domain.c +3 -3
@@ -559,7 +559,7 @@ int acc_fix_domains(struct list_head *head, bool update)
       int cnt;
       list_for_each_entry(cd, head, list) {
-        cnt = domain_entry_fix(cd->domid, cd->nbentry, update);
+        cnt = domain_nbentry_fix(cd->domid, cd->nbentry, update);
           if (!update) {
               if (cnt >= quota_nb_entry_per_domain)
                   return ENOSPC;
@@ -604,18 +604,19 @@ static struct changed_domain 
 *acc_get_changed_domain(const void *ctx,
       return cd;
   }
-int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
-            unsigned int domid)
+static int acc_add_dom_nbentry(const void *ctx, struct list_head *head, int val,
+                   unsigned int domid)

tools/xenstore/xenstored_domain.c static struct changed_domain · line 616

Quoted 10 lines
tools/xenstore/xenstored_domain.c +3 -2
   {
       struct changed_domain *cd;
       cd = acc_get_changed_domain(ctx, head, domid);
       if (!cd)
-        return errno;
+        return 0;
+    errno = 0;
       cd->nbentry += val;
-    return 0;
+    return cd->nbentry;

You just introduced this helper in the previous patch (i.e. #9). So can you get the interface correct from the start? This will make easier to review the series.

Quoted 1 line

I don't mind too much if you add the static here. Although, it would have been nice if we avoid changing code just introduced.

Fine with me.

Quoted 33 lines
Quoted 33 lines
tools/xenstore/xenstored_domain.c +0 -24
   }
   static void domain_conn_reset(struct domain *domain)
@@ -988,30 +989,6 @@ void domain_deinit(void)
           xenevtchn_unbind(xce_handle, virq_port);
   }
-int domain_entry_inc(struct connection *conn, struct node *node)
-{
-    struct domain *d;
-    unsigned int domid;
-
-    if (!node->perms.p)
-        return 0;
-
-    domid = node->perms.p[0].id;
-
-    if (conn && conn->transaction) {
-        transaction_entry_inc(conn->transaction, domid);
-    } else {
-        d = (conn && domid == conn->id && conn->domain) ? conn->domain
-            : find_or_alloc_existing_domain(domid);
-        if (d)
-            d->nbentry++;
-        else
-            return ENOMEM;
-    }
-
-    return 0;
-}
-
   /*
    * Check whether a domain was created before or after a specific generation
    * count (used for testing whether a node permission is older than a domain).
Quoted 8 lines

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1059

Quoted 6 lines
tools/xenstore/xenstored_domain.c +2 -1
@@ -1079,62 +1056,67 @@ int domain_adjust_node_perms(struct node *node)
       return 0;
   }
-void domain_entry_dec(struct connection *conn, struct node *node)
+static int domain_nbentry_add(struct connection *conn, unsigned int domid,
+                  int add, bool dom_exists)

The name of the variable suggests that that if it is false then it doesn't exists. However, looking at how you use it, it is more a "Can struct domain be allocated?". So I would rename it to "dom_alloc_allowed" or similar.

I'll name it "no_dom_alloc".

Quoted 81 lines
Quoted 35 lines
tools/xenstore/xenstored_domain.c +18 -13
   {
       struct domain *d;
-    unsigned int domid;
-
-    if (!node->perms.p)
-        return;
+    struct list_head *head;
+    int ret;
-    domid = node->perms.p ? node->perms.p[0].id : conn->id;
+    if (conn && domid == conn->id && conn->domain)
+        d = conn->domain;
+    else if (dom_exists) {
+        d = find_domain_struct(domid);
+        if (!d) {
+            errno = ENOENT;
+            corrupt(conn, "Missing domain %u\n", domid);
+            return -1;
+        }
+    } else {
+        d = find_or_alloc_existing_domain(domid);
+        if (!d) {
+            errno = ENOMEM;
+            return -1;
+        }
+    }
       if (conn && conn->transaction) {
-        transaction_entry_dec(conn->transaction, domid);
-    } else {
-        d = (conn && domid == conn->id && conn->domain) ? conn->domain
-            : find_domain_struct(domid);
-        if (d) {
-            d->nbentry--;
-        } else {
-            errno = ENOENT;

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1090

Quoted 10 lines
tools/xenstore/xenstored_domain.c +6 -3
-            corrupt(conn,
-                "Node \"%s\" owned by non-existing domain %u\n",
-                node->name, domid);
+        head = transaction_get_changed_domains(conn->transaction);
+        ret = acc_add_dom_nbentry(conn->transaction, head, add, domid);
+        if (errno) {
+            fail_transaction(conn->transaction);
+            return -1;
           }
+        return d->nbentry + ret;

It is not entirely clear why you are return "d->nbentry + ret" here. If it is ...

Quoted 20 lines
tools/xenstore/xenstored_domain.c +7 -10
       }
+
+    d->nbentry += add;
+
+    return d->nbentry;
   }
-int domain_entry_fix(unsigned int domid, int num, bool update)
+int domain_nbentry_inc(struct connection *conn, unsigned int domid)
   {
-    struct domain *d;
-    int cnt;
+    return (domain_nbentry_add(conn, domid, 1, false) < 0) ? errno : 0;
+}
-    if (update) {
-        d = find_domain_struct(domid);
-        assert(d);
-    } else {
-        /*
-         * We are called first with update == false in order to catch
-         * any error. So do a possible allocation and check for error

tools/xenstore/xenstored_domain.c int domain_adjust_node_perms(struct node *node) · line 1105

Quoted 10 lines
tools/xenstore/xenstored_domain.c +3 -7
-         * only in this case, as in the case of update == true nothing
-         * can go wrong anymore as the allocation already happened.
-         */
-        d = find_or_alloc_existing_domain(domid);
-        if (!d)
-            return -1;
-    }
+int domain_nbentry_dec(struct connection *conn, unsigned int domid)
+{
+    return (domain_nbentry_add(conn, domid, -1, true) < 0) ? errno : 0;

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

Quoted 3 lines

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Thanks for catching this.

I think the correct way to handle this is to return max(d->nbentry + ret, 0) in domain_nbentry_add(). The value might be imprecise, but always >= 0 and never wrong outside of a transaction collision.

Quoted 2 lines

Such change of behavior should pointed in the commit message. But then I am not convinced this should be part of this commit which is mainly reworking an interface (e.g. no functional change is expected).

Juergen

#25 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Hi Juergen,

Quoted 11 lines · Juergen Gross
Quoted 5 lines

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Thanks for catching this.

I think the correct way to handle this is to return max(d->nbentry + ret, 0) in domain_nbentry_add(). The value might be imprecise, but always >= 0 and never wrong outside of a transaction collision.

I am bit confused with your proposal. If the return value is imprecise, then what's the point of returning max(...) instead of simply 0?

Cheers,

Signature · 1 line
Julien Grall
#26 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Quoted 9 lines · Julien Grall

Hi Juergen,

Quoted 6 lines · Juergen Gross
Quoted 5 lines

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Quoted 7 lines
Quoted 5 lines

Thanks for catching this.

I think the correct way to handle this is to return max(d->nbentry + ret, 0) in domain_nbentry_add(). The value might be imprecise, but always >= 0 and never wrong outside of a transaction collision.

I am bit confused with your proposal. If the return value is imprecise, then what's the point of returning max(...) instead of simply 0?

Please have a look at the use case especially in domain_nbentry(). Returning always 0 would clearly break quota checks.

Juergen

#27 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Hi Juergen,

Quoted 12 lines · Juergen Gross
Quoted 11 lines · Julien Grall

Hi Juergen,

Quoted 8 lines · Juergen Gross
Quoted 5 lines

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Thanks for catching this.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

I think the correct way to handle this is to return max(d->nbentry + ret, 0) in domain_nbentry_add(). The value might be imprecise, but always >= 0 and never wrong outside of a transaction collision.

I am bit confused with your proposal. If the return value is imprecise, then what's the point of returning max(...) instead of simply 0?

Please have a look at the use case especially in domain_nbentry(). Returning always 0 would clearly break quota checks.

I am a bit concerned that we would have a code checking the quota based on an imprecise value.

At the moment, I don't have a better suggestion. But we should at least document in the code when we think the value is imprecise and explain why bypassing the quota check is OK (IOW who will check it?).

Cheers,

Signature · 1 line
Julien Grall
#28 lore raw

Re: [PATCH v2 10/19] tools/xenstore: change per-domain node accounting interface

Quoted 19 lines · Julien Grall

Hi Juergen,

Quoted 16 lines · Juergen Gross
Quoted 15 lines · Julien Grall

Hi Juergen,

Quoted 11 lines · Juergen Gross
Quoted 5 lines

... to make sure domain_nbentry_add() is not returning a negative value. Then it would not work.

A good example imagine you have a transaction removing nodes from tree but not adding any. Then the "ret" would be negative.

Meanwhile the nodes are also removed outside of the transaction. So the sum of "d->nbentry + ret" would be negative resulting to a failure here.

Thanks for catching this.

I think the correct way to handle this is to return max(d->nbentry + ret, 0) in domain_nbentry_add(). The value might be imprecise, but always >= 0 and never wrong outside of a transaction collision.

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

I am bit confused with your proposal. If the return value is imprecise, then what's the point of returning max(...) instead of simply 0?

Please have a look at the use case especially in domain_nbentry(). Returning always 0 would clearly break quota checks.

I am a bit concerned that we would have a code checking the quota based on an imprecise value.

At the moment, I don't have a better suggestion. But we should at least document in the code when we think the value is imprecise and explain why bypassing the quota check is OK (IOW who will check it?).

The imprecise value will never be too low, it can only be too high (i.e. 0 instead of negative), and that will only happen in a transaction which can't succeed.

Adding a comment is good idea, though.

Juergen

#29 lore raw

PATCH v2 11/19 tools/xenstore: don't allow creating too many nodes in a transaction

In reply to Juergen Gross

The accounting for the number of nodes of a domain in an active transaction is not working correctly, as it allows to create arbitrary number of nodes. The transaction will finally fail due to exceeding the number of nodes quota, but before closing the transaction an unprivileged guest could cause Xenstore to use a lot of memory.

---
 tools/xenstore/xenstored_domain.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
tools/xenstore/xenstored_domain.c +2 -3
index 40b24056c5..1ae79b5b54 100644
@@ -1118,9 +1118,8 @@ int domain_nbentry_fix(unsigned int domid, int num, bool update)
 
 int domain_nbentry(struct connection *conn)
 {
-	return (domain_is_unprivileged(conn))
-		? conn->domain->nbentry
-		: 0;
+	return domain_is_unprivileged(conn)
+	       ? domain_nbentry_add(conn, conn->id, 0, true) : 0;
 }
 
 static bool domain_chk_quota(struct domain *domain, int mem)
Signature · 1 line
2.35.3
#30 lore raw

Re: [PATCH v2 11/19] tools/xenstore: don't allow creating too many nodes in a transaction

Hi,

Quoted 5 lines · Juergen Gross

The accounting for the number of nodes of a domain in an active transaction is not working correctly, as it allows to create arbitrary number of nodes. The transaction will finally fail due to exceeding the number of nodes quota, but before closing the transaction an unprivileged guest could cause Xenstore to use a lot of memory.

As per the discussion in v1, the commit message needs to be reworded.

I will look at this patch in more details once I have reached the 2nd series.

Cheers,

Signature · 1 line
Julien Grall
#31 lore raw

Re: [PATCH v2 11/19] tools/xenstore: don't allow creating too many nodes in a transaction

Quoted 9 lines · Julien Grall

Hi,

Quoted 5 lines · Juergen Gross

The accounting for the number of nodes of a domain in an active transaction is not working correctly, as it allows to create arbitrary number of nodes. The transaction will finally fail due to exceeding the number of nodes quota, but before closing the transaction an unprivileged guest could cause Xenstore to use a lot of memory.

Quoted 3 lines

As per the discussion in v1, the commit message needs to be reworded.

I will look at this patch in more details once I have reached the 2nd series.

I'll wait with the rewording until then.

Juergen

#32 lore raw

PATCH v2 12/19 tools/xenstore: replace literal domid 0 with dom0_domid

In reply to Juergen Gross

There are some places left where dom0 is associated with domid 0.

Use dom0_domid instead.

---
 tools/xenstore/xenstored_core.c   | 5 +++--
 tools/xenstore/xenstored_domain.c | 8 ++++----
 2 files changed, 7 insertions(+), 6 deletions(-)
tools/xenstore/xenstored_core.c +3 -2
index 61569cecbb..6c9d22b8a2 100644
@@ -2311,9 +2311,10 @@ static void accept_connection(int sock)
 		return;
 
 	conn = new_connection(&socket_funcs);
-	if (conn)
+	if (conn) {
 		conn->fd = fd;
-	else
+		conn->id = dom0_domid;
+	} else
 		close(fd);
 }
 #endif
tools/xenstore/xenstored_domain.c +4 -4
index 1ae79b5b54..1e9d7545b7 100644
@@ -324,7 +324,7 @@ static int destroy_domain(void *_domain)
 	if (domain->interface) {
 		/* Domain 0 was mapped by dom0_init, so it must be unmapped
 		   using munmap() and not the grant unmap call. */
-		if (domain->domid == 0)
+		if (domain->domid == dom0_domid)
 			unmap_xenbus(domain->interface);
 		else
 			unmap_interface(domain->interface);
@@ -408,7 +408,7 @@ void handle_event(void)
 
 static bool domid_is_unprivileged(unsigned int domid)
 {
-	return domid != 0 && domid != priv_domid;
+	return domid != dom0_domid && domid != priv_domid;
 }
 
 bool domain_is_unprivileged(struct connection *conn)
@@ -796,7 +796,7 @@ static struct domain *onearg_domain(struct connection *conn,
 		return ERR_PTR(-EINVAL);
 
 	domid = atoi(domid_str);
-	if (!domid)
+	if (domid == dom0_domid)
 		return ERR_PTR(-EINVAL);
 
 	return find_connected_domain(domid);
@@ -1002,7 +1002,7 @@ static int chk_domain_generation(unsigned int domid, uint64_t gen)
 {
 	struct domain *d;
 
-	if (!xc_handle && domid == 0)
+	if (!xc_handle && domid == dom0_domid)
 		return 1;
 
 	d = find_domain_struct(domid);
Signature · 1 line
2.35.3
#33 lore raw

PATCH v2 13/19 tools/xenstore: make domain_is_unprivileged() an inline function

In reply to Juergen Gross

clang 14 is complaining about a NULL dereference for constructs like:

domain_is_unprivileged(conn) ? conn->in : NULL

as it can't know that domain_is_unprivileged(conn) will return false if conn is NULL.

Fix that by making domain_is_unprivileged() an inline function (and related to that domid_is_unprivileged(), too).

In order not having to make struct domain public, use conn->id instead of conn->domain->domid for the test.

---
 tools/xenstore/xenstored_core.h   | 10 ++++++++++
 tools/xenstore/xenstored_domain.c | 11 -----------
 tools/xenstore/xenstored_domain.h |  2 --
 3 files changed, 10 insertions(+), 13 deletions(-)
tools/xenstore/xenstored_core.h +10 -0
index 37006d508d..3c4e27d0dd 100644
@@ -297,6 +297,16 @@ void unmap_xenbus(void *interface);
 
 static inline int xenbus_master_domid(void) { return dom0_domid; }
 
+static inline bool domid_is_unprivileged(unsigned int domid)
+{
+	return domid != dom0_domid && domid != priv_domid;
+}
+
+static inline bool domain_is_unprivileged(const struct connection *conn)
+{
+	return conn && domid_is_unprivileged(conn->id);
+}
+
 /* Return the event channel used by xenbus. */
 evtchn_port_t xenbus_evtchn(void);
 
tools/xenstore/xenstored_domain.c +0 -11
index 1e9d7545b7..a3ecdb382f 100644
@@ -406,17 +406,6 @@ void handle_event(void)
 		barf_perror("Failed to write to event fd");
 }
 
-static bool domid_is_unprivileged(unsigned int domid)
-{
-	return domid != dom0_domid && domid != priv_domid;
-}
-
-bool domain_is_unprivileged(struct connection *conn)
-{
-	return conn && conn->domain &&
-	       domid_is_unprivileged(conn->domain->domid);
-}
-
 static char *talloc_domain_path(const void *context, unsigned int domid)
 {
 	return talloc_asprintf(context, "/local/domain/%u", domid);
tools/xenstore/xenstored_domain.h +0 -2
index 1e402f2609..22996e2576 100644
@@ -59,8 +59,6 @@ void ignore_connection(struct connection *conn, unsigned int err);
 /* Returns the implicit path of a connection (only domains have this) */
 const char *get_implicit_path(const struct connection *conn);
 
-bool domain_is_unprivileged(struct connection *conn);
-
 /* Remove node permissions for no longer existing domains. */
 int domain_adjust_node_perms(struct node *node);
 int domain_alloc_permrefs(struct node_perms *perms);
Signature · 1 line
2.35.3
#34 lore raw

PATCH v2 14/19 tools/xenstore: let chk_domain_generation() return a bool

In reply to Juergen Gross

Instead of returning 0 or 1 let chk_domain_generation() return a boolean value.

Simplify the only caller by removing the ret variable.

---
 tools/xenstore/xenstored_domain.c | 18 ++++++++----------
 1 file changed, 8 insertions(+), 10 deletions(-)
tools/xenstore/xenstored_domain.c +8 -10
index a3ecdb382f..65e94e3822 100644
@@ -983,20 +983,20 @@ void domain_deinit(void)
  * count (used for testing whether a node permission is older than a domain).
  *
  * Return values:
- *  0: domain has higher generation count (it is younger than a node with the
- *     given count), or domain isn't existing any longer
- *  1: domain is older than the node
+ *  false: domain has higher generation count (it is younger than a node with
+ *     the given count), or domain isn't existing any longer
+ *  true: domain is older than the node
  */
-static int chk_domain_generation(unsigned int domid, uint64_t gen)
+static bool chk_domain_generation(unsigned int domid, uint64_t gen)
 {
 	struct domain *d;
 
 	if (!xc_handle && domid == dom0_domid)
-		return 1;
+		return true;
 
 	d = find_domain_struct(domid);
 
-	return (d && d->generation <= gen) ? 1 : 0;
+	return d && d->generation <= gen;
 }
 
 /*
@@ -1031,14 +1031,12 @@ int domain_alloc_permrefs(struct node_perms *perms)
 int domain_adjust_node_perms(struct node *node)
 {
 	unsigned int i;
-	int ret;
 
 	for (i = 1; i < node->perms.num; i++) {
 		if (node->perms.p[i].perms & XS_PERM_IGNORE)
 			continue;
-		ret = chk_domain_generation(node->perms.p[i].id,
-					    node->generation);
-		if (!ret)
+		if (!chk_domain_generation(node->perms.p[i].id,
+					   node->generation))
 			node->perms.p[i].perms |= XS_PERM_IGNORE;
 	}
 
Signature · 1 line
2.35.3
#35 lore raw

PATCH v2 15/19 tools/xenstore: switch hashtable to use the talloc framework

In reply to Juergen Gross

Instead of using malloc() and friends, let the hashtable implementation use the talloc framework.

This is more consistent with the rest of xenstored and it allows to track memory usage via "xenstore-control memreport".

---
 tools/xenstore/hashtable.c        | 86 ++++++++++++-------------------
 tools/xenstore/hashtable.h        |  3 +-
 tools/xenstore/xenstored_core.c   |  5 +-
 tools/xenstore/xenstored_domain.c |  2 +-
 4 files changed, 39 insertions(+), 57 deletions(-)
tools/xenstore/hashtable.c +34 -52
index 299549c51e..3b6745b692 100644
@@ -6,6 +6,8 @@
 #include <string.h>
 #include <math.h>
 #include <stdint.h>
+#include <stdarg.h>
+#include "talloc.h"
 
 struct entry
 {
@@ -50,7 +52,7 @@ indexFor(unsigned int tablelength, unsigned int hashvalue) {
 
 /*****************************************************************************/
 struct hashtable *
-create_hashtable(unsigned int minsize,
+create_hashtable(const void *ctx, unsigned int minsize,
                  unsigned int (*hashf) (void*),
                  int (*eqf) (void*,void*),
                  unsigned int flags)
@@ -66,10 +68,10 @@ create_hashtable(unsigned int minsize,
         if (primes[pindex] > minsize) { size = primes[pindex]; break; }
     }
 
-    h = (struct hashtable *)calloc(1, sizeof(struct hashtable));
+    h = talloc_zero(ctx, struct hashtable);
     if (NULL == h)
         goto err0;
-    h->table = (struct entry **)calloc(size, sizeof(struct entry *));
+    h->table = talloc_zero_array(h, struct entry *, size);
     if (NULL == h->table)
         goto err1;
 
@@ -83,7 +85,7 @@ create_hashtable(unsigned int minsize,
     return h;
 
 err1:
-   free(h);
+   talloc_free(h);
 err0:
    return NULL;
 }
@@ -115,47 +117,32 @@ hashtable_expand(struct hashtable *h)
     if (h->primeindex == (prime_table_length - 1)) return 0;
     newsize = primes[++(h->primeindex)];
 
-    newtable = (struct entry **)calloc(newsize, sizeof(struct entry*));
-    if (NULL != newtable)
+    newtable = talloc_realloc(h, h->table, struct entry *, newsize);
+    if (!newtable)
     {
-        /* This algorithm is not 'stable'. ie. it reverses the list
-         * when it transfers entries between the tables */
-        for (i = 0; i < h->tablelength; i++) {
-            while (NULL != (e = h->table[i])) {
-                h->table[i] = e->next;
-                index = indexFor(newsize,e->h);
+        h->primeindex--;
+        return 0;
+    }
+
+    h->table = newtable;
+    memset(newtable + h->tablelength, 0,
+           (newsize - h->tablelength) * sizeof(*newtable));
+    for (i = 0; i < h->tablelength; i++) {
+        for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
+            index = indexFor(newsize,e->h);
+            if (index == i)
+            {
+                pE = &(e->next);
+            }
+            else
+            {
+                *pE = e->next;
                 e->next = newtable[index];
                 newtable[index] = e;
             }
         }
-        free(h->table);
-        h->table = newtable;
-    }
-    /* Plan B: realloc instead */
-    else 
-    {
-        newtable = (struct entry **)
-                   realloc(h->table, newsize * sizeof(struct entry *));
-        if (NULL == newtable) { (h->primeindex)--; return 0; }
-        h->table = newtable;
-        memset(newtable + h->tablelength, 0,
-               (newsize - h->tablelength) * sizeof(*newtable));
-        for (i = 0; i < h->tablelength; i++) {
-            for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
-                index = indexFor(newsize,e->h);
-                if (index == i)
-                {
-                    pE = &(e->next);
-                }
-                else
-                {
-                    *pE = e->next;
-                    e->next = newtable[index];
-                    newtable[index] = e;
-                }
-            }
-        }
     }
+
     h->tablelength = newsize;
     h->loadlimit   = (unsigned int)
         (((uint64_t)newsize * max_load_factor) / 100);
@@ -184,7 +171,7 @@ hashtable_insert(struct hashtable *h, void *k, void *v)
          * element may be ok. Next time we insert, we'll try expanding again.*/
         hashtable_expand(h);
     }
-    e = (struct entry *)calloc(1, sizeof(struct entry));
+    e = talloc_zero(h, struct entry);
     if (NULL == e) { --(h->entrycount); return 0; } /*oom*/
     e->h = hash(h,k);
     index = indexFor(h->tablelength,e->h);
@@ -238,8 +225,8 @@ hashtable_remove(struct hashtable *h, void *k)
             h->entrycount--;
             v = e->v;
             if (h->flags & HASHTABLE_FREE_KEY)
-                free(e->k);
-            free(e);
+                talloc_free(e->k);
+            talloc_free(e);
             return v;
         }
         pE = &(e->next);
@@ -280,25 +267,20 @@ void
 hashtable_destroy(struct hashtable *h)
 {
     unsigned int i;
-    struct entry *e, *f;
+    struct entry *e;
     struct entry **table = h->table;
 
     for (i = 0; i < h->tablelength; i++)
     {
-        e = table[i];
-        while (NULL != e)
+        for (e = table[i]; e; e = e->next)
         {
-            f = e;
-            e = e->next;
             if (h->flags & HASHTABLE_FREE_KEY)
-                free(f->k);
+                talloc_free(e->k);
             if (h->flags & HASHTABLE_FREE_VALUE)
-                free(f->v);
-            free(f);
+                talloc_free(e->v);
         }
     }
-    free(h->table);
-    free(h);
+    talloc_free(h);
 }
 
 /*
tools/xenstore/hashtable.h +2 -1
index 6d65431f96..becec73092 100644
@@ -9,6 +9,7 @@ struct hashtable;
  * create_hashtable
    
  * @name                    create_hashtable
+ * @param   ctx             talloc context to use for allocations
  * @param   minsize         minimum initial size of hashtable
  * @param   hashfunction    function for hashing keys
  * @param   key_eq_fn       function for determining key equality
@@ -22,7 +23,7 @@ struct hashtable;
 #define HASHTABLE_FREE_KEY   (1U << 1)
 
 struct hashtable *
-create_hashtable(unsigned int minsize,
+create_hashtable(const void *ctx, unsigned int minsize,
                  unsigned int (*hashfunction) (void*),
                  int (*key_eq_fn) (void*,void*),
                  unsigned int flags
tools/xenstore/xenstored_core.c +2 -3
index 6c9d22b8a2..60433265ed 100644
@@ -2419,11 +2419,10 @@ static int keys_equal_fn(void *key1, void *key2)
 
 int remember_string(struct hashtable *hash, const char *str)
 {
-	char *k = malloc(strlen(str) + 1);
+	char *k = talloc_strdup(NULL, str);
 
 	if (!k)
 		return 0;
-	strcpy(k, str);
 	return hashtable_insert(hash, k, (void *)1);
 }
 
@@ -2518,7 +2517,7 @@ void check_store(void)
 	};
 
 	/* Don't free values (they are all void *1) */
-	reachable = create_hashtable(16, hash_from_key_fn, keys_equal_fn,
+	reachable = create_hashtable(NULL, 16, hash_from_key_fn, keys_equal_fn,
 				     HASHTABLE_FREE_KEY);
 	if (!reachable) {
 		log("check_store: ENOMEM");
tools/xenstore/xenstored_domain.c +1 -1
index 65e94e3822..4d735a5951 100644
@@ -929,7 +929,7 @@ void domain_init(int evtfd)
 	int rc;
 
 	/* Start with a random rather low domain count for the hashtable. */
-	domhash = create_hashtable(8, domhash_fn, domeq_fn, 0);
+	domhash = create_hashtable(NULL, 8, domhash_fn, domeq_fn, 0);
 	if (!domhash)
 		barf_perror("Failed to allocate domain hashtable");
 
Signature · 1 line
2.35.3
#36 lore raw

Re: [PATCH v2 15/19] tools/xenstore: switch hashtable to use the talloc framework

Hi Juergen,

Quoted 13 lines · Juergen Gross
tools/xenstore/hashtable.c +2 -6
@@ -115,47 +117,32 @@ hashtable_expand(struct hashtable *h)
      if (h->primeindex == (prime_table_length - 1)) return 0;
      newsize = primes[++(h->primeindex)];
 
-    newtable = (struct entry **)calloc(newsize, sizeof(struct entry*));
-    if (NULL != newtable)
+    newtable = talloc_realloc(h, h->table, struct entry *, newsize);
+    if (!newtable)
      {
-        /* This algorithm is not 'stable'. ie. it reverses the list
-         * when it transfers entries between the tables */
-        for (i = 0; i < h->tablelength; i++) {
-            while (NULL != (e = h->table[i])) {

tools/xenstore/hashtable.c hashtable_expand(struct hashtable *h) · line 130

Quoted 10 lines
tools/xenstore/hashtable.c +8 -2
-                h->table[i] = e->next;
-                index = indexFor(newsize,e->h);
+        h->primeindex--;
+        return 0;
+    }
+
+    h->table = newtable;
+    memset(newtable + h->tablelength, 0,
+           (newsize - h->tablelength) * sizeof(*newtable));
+    for (i = 0; i < h->tablelength; i++) {

I understand this code is taken from the realloc path. However, isn't this algorithm also not stable? If so, then I think we move the comment here.

tools/xenstore/hashtable.c hashtable_expand(struct hashtable *h) · line 132

Quoted 2 lines
tools/xenstore/hashtable.c +2 -0
+        for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
+            index = indexFor(newsize,e->h);

Missing space after ",".

Quoted 75 lines
tools/xenstore/hashtable.c +12 -33
+            if (index == i)
+            {
+                pE = &(e->next);
+            }
+            else
+            {
+                *pE = e->next;
                  e->next = newtable[index];
                  newtable[index] = e;
              }
          }
-        free(h->table);
-        h->table = newtable;
-    }
-    /* Plan B: realloc instead */
-    else
-    {
-        newtable = (struct entry **)
-                   realloc(h->table, newsize * sizeof(struct entry *));
-        if (NULL == newtable) { (h->primeindex)--; return 0; }
-        h->table = newtable;
-        memset(newtable + h->tablelength, 0,
-               (newsize - h->tablelength) * sizeof(*newtable));
-        for (i = 0; i < h->tablelength; i++) {
-            for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
-                index = indexFor(newsize,e->h);
-                if (index == i)
-                {
-                    pE = &(e->next);
-                }
-                else
-                {
-                    *pE = e->next;
-                    e->next = newtable[index];
-                    newtable[index] = e;
-                }
-            }
-        }
      }
+
      h->tablelength = newsize;
      h->loadlimit   = (unsigned int)
          (((uint64_t)newsize * max_load_factor) / 100);
@@ -184,7 +171,7 @@ hashtable_insert(struct hashtable *h, void *k, void *v)
           * element may be ok. Next time we insert, we'll try expanding again.*/
          hashtable_expand(h);
      }
-    e = (struct entry *)calloc(1, sizeof(struct entry));
+    e = talloc_zero(h, struct entry);
      if (NULL == e) { --(h->entrycount); return 0; } /*oom*/
      e->h = hash(h,k);
      index = indexFor(h->tablelength,e->h);
@@ -238,8 +225,8 @@ hashtable_remove(struct hashtable *h, void *k)
              h->entrycount--;
              v = e->v;
              if (h->flags & HASHTABLE_FREE_KEY)
-                free(e->k);
-            free(e);
+                talloc_free(e->k);
+            talloc_free(e);
              return v;
          }
          pE = &(e->next);
@@ -280,25 +267,20 @@ void
  hashtable_destroy(struct hashtable *h)
  {
      unsigned int i;
-    struct entry *e, *f;
+    struct entry *e;
      struct entry **table = h->table;
 
      for (i = 0; i < h->tablelength; i++)
      {
-        e = table[i];
-        while (NULL != e)

tools/xenstore/hashtable.c void · line 297

Quoted 10 lines
tools/xenstore/hashtable.c +2 -5
+        for (e = table[i]; e; e = e->next)
          {
-            f = e;
-            e = e->next;
              if (h->flags & HASHTABLE_FREE_KEY)
-                free(f->k);
+                talloc_free(e->k);
              if (h->flags & HASHTABLE_FREE_VALUE)
-                free(f->v);
-            free(f);

AFAIU, the loop is reworked so you let talloc to free each element with the parent. Using a while loop is definitely cleaner, but now you will end up to have two separate loop for the elements.

There is a risk that the overall performance of hashtable_destroy() will be worse as the data accessed in one loop may not fit in the cache. So you will have to reload it on the second loop.

Therefore, I think it would be better to keep the loop as-is.

Cheers,

Signature · 1 line
Julien Grall
#37 lore raw

Re: [PATCH v2 15/19] tools/xenstore: switch hashtable to use the talloc framework

Quoted 19 lines · Julien Grall

Hi Juergen,

Quoted 16 lines · Juergen Gross
tools/xenstore/hashtable.c +4 -8
@@ -115,47 +117,32 @@ hashtable_expand(struct hashtable *h)
       if (h->primeindex == (prime_table_length - 1)) return 0;
       newsize = primes[++(h->primeindex)];
-    newtable = (struct entry **)calloc(newsize, sizeof(struct entry*));
-    if (NULL != newtable)
+    newtable = talloc_realloc(h, h->table, struct entry *, newsize);
+    if (!newtable)
       {
-        /* This algorithm is not 'stable'. ie. it reverses the list
-         * when it transfers entries between the tables */
-        for (i = 0; i < h->tablelength; i++) {
-            while (NULL != (e = h->table[i])) {
-                h->table[i] = e->next;
-                index = indexFor(newsize,e->h);
+        h->primeindex--;
+        return 0;
Quoted 8 lines

tools/xenstore/hashtable.c hashtable_expand(struct hashtable *h) · line 130

Quoted 6 lines
tools/xenstore/hashtable.c +6 -0
+    }
+
+    h->table = newtable;
+    memset(newtable + h->tablelength, 0,
+           (newsize - h->tablelength) * sizeof(*newtable));
+    for (i = 0; i < h->tablelength; i++) {

I understand this code is taken from the realloc path. However, isn't this algorithm also not stable? If so, then I think we move the comment here.

I'm fine with that, even if I don't see how it would matter. There is no guarantee regarding the order of entries for a given index.

Quoted 5 lines

tools/xenstore/hashtable.c hashtable_expand(struct hashtable *h) · line 132

Quoted 3 lines
tools/xenstore/hashtable.c +2 -0
+        for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
+            index = indexFor(newsize,e->h);

Missing space after ",".

Will fix.

Quoted 88 lines
Quoted 75 lines
tools/xenstore/hashtable.c +9 -28
+            if (index == i)
+            {
+                pE = &(e->next);
+            }
+            else
+            {
+                *pE = e->next;
                   e->next = newtable[index];
                   newtable[index] = e;
               }
           }
-        free(h->table);
-        h->table = newtable;
-    }
-    /* Plan B: realloc instead */
-    else
-    {
-        newtable = (struct entry **)
-                   realloc(h->table, newsize * sizeof(struct entry *));
-        if (NULL == newtable) { (h->primeindex)--; return 0; }
-        h->table = newtable;
-        memset(newtable + h->tablelength, 0,
-               (newsize - h->tablelength) * sizeof(*newtable));
-        for (i = 0; i < h->tablelength; i++) {
-            for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
-                index = indexFor(newsize,e->h);
-                if (index == i)
-                {
-                    pE = &(e->next);
-                }
-                else
-                {
-                    *pE = e->next;
-                    e->next = newtable[index];
-                    newtable[index] = e;
-                }
-            }
-        }
       }
+
       h->tablelength = newsize;
       h->loadlimit   = (unsigned int)
           (((uint64_t)newsize * max_load_factor) / 100);
@@ -184,7 +171,7 @@ hashtable_insert(struct hashtable *h, void *k, void *v)
            * element may be ok. Next time we insert, we'll try expanding again.*/
           hashtable_expand(h);
       }
-    e = (struct entry *)calloc(1, sizeof(struct entry));
+    e = talloc_zero(h, struct entry);

      if (NULL == e) { --(h->entrycount); return 0; } /*oom*/ e->h = hash(h,k);
      index = indexFor(h->tablelength,e->h);

tools/xenstore/hashtable.c +3 -5
@@ -238,8 +225,8 @@ hashtable_remove(struct hashtable *h, void *k)
               h->entrycount--;
               v = e->v;
               if (h->flags & HASHTABLE_FREE_KEY)
-                free(e->k);
-            free(e);
+                talloc_free(e->k);
+            talloc_free(e);
               return v;
           }
           pE = &(e->next);
@@ -280,25 +267,20 @@ void
   hashtable_destroy(struct hashtable *h)
   {
       unsigned int i;
-    struct entry *e, *f;
+    struct entry *e;
       struct entry **table = h->table;
       for (i = 0; i < h->tablelength; i++)
       {
-        e = table[i];
-        while (NULL != e)

tools/xenstore/hashtable.c void · line 296

Quoted 10 lines
tools/xenstore/hashtable.c +2 -5
+        for (e = table[i]; e; e = e->next)
           {
-            f = e;
-            e = e->next;
               if (h->flags & HASHTABLE_FREE_KEY)
-                free(f->k);
+                talloc_free(e->k);
               if (h->flags & HASHTABLE_FREE_VALUE)
-                free(f->v);
-            free(f);

AFAIU, the loop is reworked so you let talloc to free each element with the parent. Using a while loop is definitely cleaner, but now you will end up to have two separate loop for the elements.

Quoted 3 lines

There is a risk that the overall performance of hashtable_destroy() will be worse as the data accessed in one loop may not fit in the cache. So you will have to reload it on the second loop.

Therefore, I think it would be better to keep the loop as-is.

What about a completely different approach? I could make the key and value talloc children of e when _inserting_ the element and the related flag is set. This would reduce hashtable_destroy to a single talloc_free().

Juergen

#38 lore raw

Re: [PATCH v2 15/19] tools/xenstore: switch hashtable to use the talloc framework

Hi,

Quoted 29 lines · Juergen Gross
Quoted 19 lines · Julien Grall

Hi Juergen,

Quoted 16 lines · Juergen Gross
tools/xenstore/hashtable.c +4 -8
@@ -115,47 +117,32 @@ hashtable_expand(struct hashtable *h)
       if (h->primeindex == (prime_table_length - 1)) return 0;
       newsize = primes[++(h->primeindex)];
-    newtable = (struct entry **)calloc(newsize, sizeof(struct entry*));
-    if (NULL != newtable)
+    newtable = talloc_realloc(h, h->table, struct entry *, newsize);
+    if (!newtable)
       {
-        /* This algorithm is not 'stable'. ie. it reverses the list
-         * when it transfers entries between the tables */
-        for (i = 0; i < h->tablelength; i++) {
-            while (NULL != (e = h->table[i])) {
-                h->table[i] = e->next;
-                index = indexFor(newsize,e->h);
+        h->primeindex--;
+        return 0;
Quoted 8 lines

tools/xenstore/hashtable.c hashtable_expand(struct hashtable *h) · line 130

Quoted 6 lines
tools/xenstore/hashtable.c +6 -0
+    }
+
+    h->table = newtable;
+    memset(newtable + h->tablelength, 0,
+           (newsize - h->tablelength) * sizeof(*newtable));
+    for (i = 0; i < h->tablelength; i++) {

I understand this code is taken from the realloc path. However, isn't this algorithm also not stable? If so, then I think we move the comment here.

Quoted 2 lines

I'm fine with that, even if I don't see how it would matter. There is no guarantee regarding the order of entries for a given index.

That's a fair point. Feel free to ignore my comment then :).

Quoted 88 lines
Quoted 88 lines
Quoted 75 lines
tools/xenstore/hashtable.c +9 -28
+            if (index == i)
+            {
+                pE = &(e->next);
+            }
+            else
+            {
+                *pE = e->next;
                   e->next = newtable[index];
                   newtable[index] = e;
               }
           }
-        free(h->table);
-        h->table = newtable;
-    }
-    /* Plan B: realloc instead */
-    else
-    {
-        newtable = (struct entry **)
-                   realloc(h->table, newsize * sizeof(struct entry *));
-        if (NULL == newtable) { (h->primeindex)--; return 0; }
-        h->table = newtable;
-        memset(newtable + h->tablelength, 0,
-               (newsize - h->tablelength) * sizeof(*newtable));
-        for (i = 0; i < h->tablelength; i++) {
-            for (pE = &(newtable[i]), e = *pE; e != NULL; e = *pE) {
-                index = indexFor(newsize,e->h);
-                if (index == i)
-                {
-                    pE = &(e->next);
-                }
-                else
-                {
-                    *pE = e->next;
-                    e->next = newtable[index];
-                    newtable[index] = e;
-                }
-            }
-        }
       }
+
       h->tablelength = newsize;
       h->loadlimit   = (unsigned int)
           (((uint64_t)newsize * max_load_factor) / 100);
@@ -184,7 +171,7 @@ hashtable_insert(struct hashtable *h, void *k, 
 void *v)
            * element may be ok. Next time we insert, we'll try expanding again.*/
           hashtable_expand(h);
       }
-    e = (struct entry *)calloc(1, sizeof(struct entry));
+    e = talloc_zero(h, struct entry);

      if (NULL == e) { --(h->entrycount); return 0; } /*oom*/ e->h = hash(h,k);
      index = indexFor(h->tablelength,e->h);

tools/xenstore/hashtable.c +3 -5
@@ -238,8 +225,8 @@ hashtable_remove(struct hashtable *h, void *k)
               h->entrycount--;
               v = e->v;
               if (h->flags & HASHTABLE_FREE_KEY)
-                free(e->k);
-            free(e);
+                talloc_free(e->k);
+            talloc_free(e);
               return v;
           }
           pE = &(e->next);
@@ -280,25 +267,20 @@ void
   hashtable_destroy(struct hashtable *h)
   {
       unsigned int i;
-    struct entry *e, *f;
+    struct entry *e;
       struct entry **table = h->table;
       for (i = 0; i < h->tablelength; i++)
       {
-        e = table[i];
-        while (NULL != e)

tools/xenstore/hashtable.c void · line 296

Quoted 10 lines
tools/xenstore/hashtable.c +2 -5
+        for (e = table[i]; e; e = e->next)
           {
-            f = e;
-            e = e->next;
               if (h->flags & HASHTABLE_FREE_KEY)
-                free(f->k);
+                talloc_free(e->k);
               if (h->flags & HASHTABLE_FREE_VALUE)
-                free(f->v);
-            free(f);

AFAIU, the loop is reworked so you let talloc to free each element with the parent. Using a while loop is definitely cleaner, but now you will end up to have two separate loop for the elements.

Quoted 7 lines
Quoted 3 lines

There is a risk that the overall performance of hashtable_destroy() will be worse as the data accessed in one loop may not fit in the cache. So you will have to reload it on the second loop.

Therefore, I think it would be better to keep the loop as-is.

What about a completely different approach? I could make the key and value talloc children of e when _inserting_ the element and the related flag is set. This would reduce hashtable_destroy to a single talloc_free().

I am fine with that.

Cheers,

Signature · 1 line
Julien Grall
#39 lore raw

PATCH v2 16/19 tools/xenstore: make log macro globally available

In reply to Juergen Gross

Move the definition of the log() macro to xenstored_core.h in order to make it usable from other source files, too.

While at it preserve errno from being modified.

---
 tools/xenstore/xenstored_core.c | 14 --------------
 tools/xenstore/xenstored_core.h | 15 +++++++++++++++
 2 files changed, 15 insertions(+), 14 deletions(-)
tools/xenstore/xenstored_core.c +0 -14
index 60433265ed..48c2793aeb 100644
@@ -88,20 +88,6 @@ TDB_CONTEXT *tdb_ctx = NULL;
 
 static const char *sockmsg_string(enum xsd_sockmsg_type type);
 
-#define log(...)							\
-	do {								\
-		char *s = talloc_asprintf(NULL, __VA_ARGS__);		\
-		if (s) {						\
-			trace("%s\n", s);				\
-			syslog(LOG_ERR, "%s\n",  s);			\
-			talloc_free(s);					\
-		} else {						\
-			trace("talloc failure during logging\n");	\
-			syslog(LOG_ERR, "talloc failure during logging\n"); \
-		}							\
-	} while (0)
-
-
 int quota_nb_entry_per_domain = 1000;
 int quota_nb_watch_per_domain = 128;
 int quota_max_entry_size = 2048; /* 2K */
tools/xenstore/xenstored_core.h +15 -0
index 3c4e27d0dd..3b96ecd018 100644
@@ -267,6 +267,21 @@ void trace(const char *fmt, ...) __attribute__ ((format (printf, 1, 2)));
 void reopen_log(void);
 void close_log(void);
 
+#define log(...)							\
+	do {								\
+		int _saved_errno = errno;				\
+		char *s = talloc_asprintf(NULL, __VA_ARGS__);		\
+		if (s) {						\
+			trace("%s\n", s);				\
+			syslog(LOG_ERR, "%s\n",	s);			\
+			talloc_free(s);					\
+		} else {						\
+			trace("talloc failure during logging\n");	\
+			syslog(LOG_ERR, "talloc failure during logging\n"); \
+		}							\
+		errno = _saved_errno;					\
+	} while (0)
+
 extern int orig_argc;
 extern char **orig_argv;
 
Signature · 1 line
2.35.3
#40 lore raw

PATCH v2 17/19 tools/xenstore: introduce trace classes

In reply to Juergen Gross

Make the xenstored internal trace configurable by adding classes which can be switched on and off independently from each other.

Define the following classes:

- obj: Creation and deletion of interesting "objects" (watch, transaction, connection)
- io: incoming requests and outgoing responses
- wrl: write limiting

Per default "obj" and "io" are switched on.

Entries written via trace() will always be printed (if tracing is on at all).

Add the capability to control the trace settings via the "log" command and via a new "--log-control" command line option.

Add a missing trace_create() call for creating a transaction.

---
V2:
- keep "log" and "logfile" command names (Julien Grall)

---
 docs/misc/xenstore.txt                 | 10 +++--
 tools/xenstore/xenstored_control.c     | 31 +++++++++++++--
 tools/xenstore/xenstored_core.c        | 55 ++++++++++++++++++++++++--
 tools/xenstore/xenstored_core.h        |  6 +++
 tools/xenstore/xenstored_domain.c      | 27 +++++++------
 tools/xenstore/xenstored_transaction.c |  1 +
 6 files changed, 106 insertions(+), 24 deletions(-)
docs/misc/xenstore.txt +6 -4
index 44428ae3a7..8887e7df88 100644
@@ -409,10 +409,12 @@ CONTROL			<command>|[<parameters>|]
 		error string in case of failure. -s can return "BUSY" in case
 		of an active transaction, a retry of -s can be done in that
 		case.
-	log|on
-		turn xenstore logging on
-	log|off
-		turn xenstore logging off
+	log|[on|off|+<switch>|-<switch>]
+		without parameters: show possible log switches
+		on: turn xenstore logging on
+		off: turn xenstore logging off
+		+<switch>: activates log entries for <switch>,
+		-<switch>: deactivates log entries for <switch>
 	logfile|<file-name>
 		log to specified file
 	memreport|[<file-name>]
tools/xenstore/xenstored_control.c +28 -3
index 41e6992591..000b2bb8c7 100644
@@ -182,6 +182,28 @@ static int do_control_check(const void *ctx, struct connection *conn,
 static int do_control_log(const void *ctx, struct connection *conn,
 			  char **vec, int num)
 {
+	int ret;
+
+	if (num == 0) {
+		char *resp = talloc_asprintf(ctx, "Log switch settings:\n");
+		unsigned int idx;
+		bool on;
+
+		if (!resp)
+			return ENOMEM;
+		for (idx = 0; trace_switches[idx]; idx++) {
+			on = trace_flags & (1u << idx);
+			resp = talloc_asprintf_append(resp, "%-8s: %s\n",
+						      trace_switches[idx],
+						      on ? "on" : "off");
+			if (!resp)
+				return ENOMEM;
+		}
+
+		send_reply(conn, XS_CONTROL, resp, strlen(resp) + 1);
+		return 0;
+	}
+
 	if (num != 1)
 		return EINVAL;
 
@@ -189,8 +211,11 @@ static int do_control_log(const void *ctx, struct connection *conn,
 		reopen_log();
 	else if (!strcmp(vec[0], "off"))
 		close_log();
-	else
-		return EINVAL;
+	else {
+		ret = set_trace_switch(vec[0]);
+		if (ret)
+			return ret;
+	}
 
 	send_ack(conn, XS_CONTROL);
 	return 0;
@@ -923,7 +948,7 @@ static int do_control_help(const void *, struct connection *, char **, int);
 
 static struct cmd_s cmds[] = {
 	{ "check", do_control_check, "" },
-	{ "log", do_control_log, "on|off" },
+	{ "log", do_control_log, "[on|off|+<switch>|-<switch>]" },
 
 #ifndef NO_LIVE_UPDATE
 	/*
tools/xenstore/xenstored_core.c +51 -4

106 lines not on this page · raw message

tools/xenstore/xenstored_core.h +6 -0
index 3b96ecd018..c85b15515c 100644
@@ -287,6 +287,12 @@ extern char **orig_argv;
 
 extern char *tracefile;
 extern int tracefd;
+extern unsigned int trace_flags;
+#define TRACE_OBJ	0x00000001
+#define TRACE_IO	0x00000002
+#define TRACE_WRL	0x00000004
+extern const char *trace_switches[];
+int set_trace_switch(const char *arg);
 
 extern TDB_CONTEXT *tdb_ctx;
 extern int dom0_domid;
tools/xenstore/xenstored_domain.c +14 -13
index 4d735a5951..28c58b5c6d 100644
@@ -1249,6 +1249,12 @@ static long wrl_ndomains;
 static wrl_creditt wrl_reserve; /* [-wrl_config_newdoms_dburst, +_gburst ] */
 static time_t wrl_log_last_warning; /* 0: no previous warning */
 
+#define trace_wrl(...)				\
+do {						\
+	if (trace_flags & TRACE_WRL)		\
+		trace("wrl: " __VA_ARGS__);	\
+} while (0)
+
 void wrl_gettime_now(struct wrl_timestampt *now_wt)
 {
 	struct timespec now_ts;
@@ -1354,12 +1360,9 @@ void wrl_credit_update(struct domain *domain, struct wrl_timestampt now)
 
 	domain->wrl_timestamp = now;
 
-	trace("wrl: dom %4d %6ld  msec  %9ld credit   %9ld reserve"
-	      "  %9ld discard\n",
-	      domain->domid,
-	      msec,
-	      (long)domain->wrl_credit, (long)wrl_reserve,
-	      (long)surplus);
+	trace_wrl("dom %4d %6ld msec %9ld credit %9ld reserve %9ld discard\n",
+		  domain->domid, msec, (long)domain->wrl_credit,
+		  (long)wrl_reserve, (long)surplus);
 }
 
 void wrl_check_timeout(struct domain *domain,
@@ -1391,10 +1394,9 @@ void wrl_check_timeout(struct domain *domain,
 	if (*ptimeout==-1 || wakeup < *ptimeout)
 		*ptimeout = wakeup;
 
-	trace("wrl: domain %u credit=%ld (reserve=%ld) SLEEPING for %d\n",
-	      domain->domid,
-	      (long)domain->wrl_credit, (long)wrl_reserve,
-	      wakeup);
+	trace_wrl("domain %u credit=%ld (reserve=%ld) SLEEPING for %d\n",
+		  domain->domid, (long)domain->wrl_credit, (long)wrl_reserve,
+		  wakeup);
 }
 
 #define WRL_LOG(now, ...) \
@@ -1412,9 +1414,8 @@ void wrl_apply_debit_actual(struct domain *domain)
 	wrl_credit_update(domain, now);
 
 	domain->wrl_credit -= wrl_config_writecost;
-	trace("wrl: domain %u credit=%ld (reserve=%ld)\n",
-	      domain->domid,
-	      (long)domain->wrl_credit, (long)wrl_reserve);
+	trace_wrl("domain %u credit=%ld (reserve=%ld)\n", domain->domid,
+		  (long)domain->wrl_credit, (long)wrl_reserve);
 
 	if (domain->wrl_credit < 0) {
 		if (!domain->wrl_delay_logged) {
tools/xenstore/xenstored_transaction.c +1 -0
index 82e5e66c18..1aa9d3cb3d 100644
@@ -475,6 +475,7 @@ int do_transaction_start(const void *ctx, struct connection *conn,
 	if (!trans)
 		return ENOMEM;
 
+	trace_create(trans, "transaction");
 	INIT_LIST_HEAD(&trans->accessed);
 	INIT_LIST_HEAD(&trans->changed_domains);
 	trans->conn = conn;
Signature · 1 line
2.35.3
#41 lore raw

PATCH v2 18/19 tools/xenstore: let check_store() check the accounting data

In reply to Juergen Gross

Today check_store() is only testing the correctness of the node tree.

Add verification of the accounting data (number of nodes) and correct the data if it is wrong.

Do the initial check_store() call only after Xenstore entries of a live update have been read.

---
 tools/xenstore/xenstored_core.c   | 62 ++++++++++++++++------
 tools/xenstore/xenstored_domain.c | 86 +++++++++++++++++++++++++++++++
 tools/xenstore/xenstored_domain.h |  4 ++
 3 files changed, 137 insertions(+), 15 deletions(-)
tools/xenstore/xenstored_core.c +47 -15
index 0d5b5a0d82..8fac7c9477 100644
@@ -2384,8 +2384,6 @@ void setup_structure(bool live_update)
 		manual_node("@introduceDomain", NULL);
 		domain_nbentry_fix(dom0_domid, 5, true);
 	}
-
-	check_store();
 }
 
 static unsigned int hash_from_key_fn(void *k)
@@ -2428,20 +2426,28 @@ int remember_string(struct hashtable *hash, const char *str)
  * As we go, we record each node in the given reachable hashtable.  These
  * entries will be used later in clean_store.
  */
+
+struct check_store_data {
+	struct hashtable *reachable;
+	struct hashtable *domains;
+};
+
 static int check_store_step(const void *ctx, struct connection *conn,
 			    struct node *node, void *arg)
 {
-	struct hashtable *reachable = arg;
+	struct check_store_data *data = arg;
 
-	if (hashtable_search(reachable, (void *)node->name)) {
+	if (hashtable_search(data->reachable, (void *)node->name)) {
 		log("check_store: '%s' is duplicated!", node->name);
 		return recovery ? WALK_TREE_RM_CHILDENTRY
 				: WALK_TREE_SKIP_CHILDREN;
 	}
 
-	if (!remember_string(reachable, node->name))
+	if (!remember_string(data->reachable, node->name))
 		return WALK_TREE_ERROR_STOP;
 
+	domain_check_acc_add(node, data->domains);
+
 	return WALK_TREE_OK;
 }
 
@@ -2491,37 +2497,61 @@ static int clean_store_(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA val,
  * Given the list of reachable nodes, iterate over the whole store, and
  * remove any that were not reached.
  */
-static void clean_store(struct hashtable *reachable)
+static void clean_store(struct check_store_data *data)
 {
-	tdb_traverse(tdb_ctx, &clean_store_, reachable);
+	tdb_traverse(tdb_ctx, &clean_store_, data->reachable);
+	domain_check_acc(data->domains);
 }
 
+int check_store_path(const char *name, struct check_store_data *data)
+{
+	struct node *node;
+
+	node = read_node(NULL, NULL, name);
+	if (!node) {
+		log("check_store: error %d reading special node '%s'", errno,
+		    name);
+		return errno;
+	}
+
+	return check_store_step(NULL, NULL, node, data);
+}
 
 void check_store(void)
 {
-	struct hashtable *reachable;
 	struct walk_funcs walkfuncs = {
 		.enter = check_store_step,
 		.enoent = check_store_enoent,
 	};
+	struct check_store_data data;
 
 	/* Don't free values (they are all void *1) */
-	reachable = create_hashtable(NULL, 16, hash_from_key_fn, keys_equal_fn,
-				     HASHTABLE_FREE_KEY);
-	if (!reachable) {
+	data.reachable = create_hashtable(NULL, 16, hash_from_key_fn,
+					  keys_equal_fn, HASHTABLE_FREE_KEY);
+	if (!data.reachable) {
 		log("check_store: ENOMEM");
 		return;
 	}
 
+	data.domains = domain_check_acc_init();
+	if (!data.domains) {
+		log("check_store: ENOMEM");
+		goto out_hash;
+	}
+
 	log("Checking store ...");
-	if (walk_node_tree(NULL, NULL, "/", &walkfuncs, reachable)) {
+	if (walk_node_tree(NULL, NULL, "/", &walkfuncs, &data)) {
 		if (errno == ENOMEM)
 			log("check_store: ENOMEM");
-	} else if (!check_transactions(reachable))
-		clean_store(reachable);
+	} else if (!check_store_path("@introduceDomain", &data) &&
+		   !check_store_path("@releaseDomain", &data) &&
+		   !check_transactions(data.reachable))
+		clean_store(&data);
 	log("Checking store complete.");
 
-	hashtable_destroy(reachable);
+	hashtable_destroy(data.domains);
+ out_hash:
+	hashtable_destroy(data.reachable);
 }
 
 
@@ -2920,6 +2950,8 @@ int main(int argc, char *argv[])
 		lu_read_state();
 #endif
 
+	check_store();
+
 	/* Get ready to listen to the tools. */
 	initialize_fds(&sock_pollfd_idx, &timeout);
 
tools/xenstore/xenstored_domain.c +86 -0

92 lines not on this page · raw message

tools/xenstore/xenstored_domain.h +4 -0
index 22996e2576..dc4660861e 100644
@@ -129,4 +129,8 @@ const char *dump_state_connections(FILE *fp);
 
 void read_state_connection(const void *ctx, const void *state);
 
+struct hashtable *domain_check_acc_init(void);
+void domain_check_acc_add(const struct node *node, struct hashtable *domains);
+void domain_check_acc(struct hashtable *domains);
+
 #endif /* _XENSTORED_DOMAIN_H */
Signature · 1 line
2.35.3
#42 lore raw

PATCH v2 19/19 tools/xenstore: make output of "xenstore-control help" more pretty

In reply to Juergen Gross

Using a tab for separating the command from the options in the output of "xenstore-control help" results in a rather ugly list.

Use a fixed size for the command instead.

---
V2:
- new patch

---
 tools/xenstore/xenstored_control.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/xenstore/xenstored_control.c +1 -1
index 000b2bb8c7..cbd62556c3 100644
@@ -996,7 +996,7 @@ static int do_control_help(const void *ctx, struct connection *conn,
 	if (!resp)
 		return ENOMEM;
 	for (cmd = 0; cmd < ARRAY_SIZE(cmds); cmd++) {
-		resp = talloc_asprintf_append(resp, "%s\t%s\n",
+		resp = talloc_asprintf_append(resp, "%-15s %s\n",
 					      cmds[cmd].cmd, cmds[cmd].pars);
 		if (!resp)
 			return ENOMEM;
Signature · 1 line
2.35.3
#43 lore raw
In reply to Juergen Gross
Quoted 5 lines · Juergen Gross

This is a first run of post-XSA patches which piled up during the development phase of all the recent Xenstore related XSA patches.

At least the first 5 patches are completely independent from each other. After those the dependencies are starting to be more complex.

I've applied the three ones of these which were fully ready.

Jan