features/bitrot: serialize versioning

    Backport of http://review.gluster.org/10832

Current signing interface (fsetxattr()) had couple of issues:

One, a signing request (by bitrot daemon) is denied if the version
against which an object is to be signed is unequal to the current
version of the object (cases where another subsequent modification
increments the version). Such request(s) are rejected with EINVAL
sent back to the signer resulting in a bunch of errors (in logs)
reported by bitrot daemon. Although, the object would be eventaully
signed with the version matching the current version, the "lagging"
request should be correctly handled.

Two, more than one signing request could race against each other
with the object getting signed with a version depending on which
request ended up last in the race. Although harmless to some extent,
such a case could end up marking the object's signature as stale
for infinity (if the object is *never* touched) thereby resulting
in scrubber skipping the object during verification.

This patch fixes these issues by ordering signing request(s) and
fixing version comparison checks at the time of signing.

Change-Id: I9fa83dfa3be664ba4db61d7f2edc408f4bde77dd
BUG: 1224650
Signed-off-by: Venky Shankar <vshankar@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-on: http://review.gluster.org/10900
Tested-by: NetBSD Build System <jenkins@build.gluster.org>

3 files changed, 174 insertions, 32 deletions

diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
index 504b8ab3635..fbb69ce7ea8 100644
--- a/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub-mem-types.h
@@ -30,6 +30,7 @@ enum br_mem_types {
         gf_br_mt_br_fsscan_entry_t,
         gf_br_stub_mt_br_stub_fd_t,
         gf_br_stub_mt_br_scanner_freq_t,
+        gf_br_stub_mt_sigstub_t,
         gf_br_stub_mt_end,
 };
 
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub.c b/xlators/features/bit-rot/src/stub/bit-rot-stub.c
index 524c235b549..1827f48c1c9 100644
--- a/xlators/features/bit-rot/src/stub/bit-rot-stub.c
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub.c
@@ -21,6 +21,7 @@
 #include "logging.h"
 #include "changelog.h"
 #include "compat-errno.h"
+#include "call-stub.h"
 
 #include "bit-rot-stub.h"
 #include "bit-rot-stub-mem-types.h"
@@ -29,6 +30,16 @@
 
 #define BR_STUB_REQUEST_COOKIE  0x1
 
+void *br_stub_signth (void *);
+
+struct br_stub_signentry {
+        unsigned long v;
+
+        call_stub_t *stub;
+
+        struct list_head list;
+};
+
 int32_t
 mem_acct_init (xlator_t *this)
 {
@@ -51,6 +62,7 @@ mem_acct_init (xlator_t *this)
 int32_t
 init (xlator_t *this)
 {
+        int32_t ret = 0;
         char *tmp = NULL;
         struct timeval tv = {0,};
 	br_stub_private_t *priv = NULL;
@@ -79,10 +91,22 @@ init (xlator_t *this)
         priv->boot[0] = htonl (tv.tv_sec);
         priv->boot[1] = htonl (tv.tv_usec);
 
+        pthread_mutex_init (&priv->lock, NULL);
+        pthread_cond_init (&priv->cond, NULL);
+        INIT_LIST_HEAD (&priv->squeue);
+
+        ret = gf_thread_create (&priv->signth, NULL, br_stub_signth, priv);
+        if (ret != 0)
+                goto cleanup_lock;
+
         gf_log (this->name, GF_LOG_DEBUG, "bit-rot stub loaded");
 	this->private = priv;
+
         return 0;
 
+ cleanup_lock:
+        pthread_cond_destroy (&priv->cond);
+        pthread_mutex_destroy (&priv->lock);
  free_mempool:
         mem_pool_destroy (priv->local_pool);
  free_priv:
@@ -94,13 +118,36 @@ init (xlator_t *this)
 void
 fini (xlator_t *this)
 {
+        int32_t ret = 0;
 	br_stub_private_t *priv = this->private;
+        struct br_stub_signentry *sigstub = NULL;
 
         if (!priv)
                 return;
+
+        ret = gf_thread_cleanup_xint (priv->signth);
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR,
+                        "Could not cancel sign serializer thread");
+                goto out;
+        }
+
+        while (!list_empty (&priv->squeue)) {
+                sigstub = list_first_entry (&priv->squeue,
+                                            struct br_stub_signentry, list);
+                list_del_init (&sigstub->list);
+
+                call_stub_destroy (sigstub->stub);
+                GF_FREE (sigstub);
+        }
+
+        pthread_mutex_destroy (&priv->lock);
+        pthread_cond_destroy (&priv->cond);
+
         this->private = NULL;
 	GF_FREE (priv);
 
+ out:
 	return;
 }
 
@@ -567,13 +614,66 @@ br_stub_perform_incversioning (xlator_t *this,
 
 /* fsetxattr() */
 
+int32_t
+br_stub_perform_objsign (call_frame_t *frame, xlator_t *this,
+                         fd_t *fd, dict_t *dict, int flags, dict_t *xdata)
+{
+        STACK_WIND (frame, default_setxattr_cbk,
+                    FIRST_CHILD (this), FIRST_CHILD (this)->fops->fsetxattr, fd,
+                    dict, flags, xdata);
+
+        dict_unref (xdata);
+        return 0;
+}
+
+void *
+br_stub_signth (void *arg)
+{
+        br_stub_private_t *priv = arg;
+        struct br_stub_signentry *sigstub = NULL;
+
+        while (1) {
+                pthread_mutex_lock (&priv->lock);
+                {
+                        while (list_empty (&priv->squeue))
+                                pthread_cond_wait (&priv->cond, &priv->lock);
+
+                        sigstub = list_first_entry
+                                (&priv->squeue, struct br_stub_signentry, list);
+                        list_del_init (&sigstub->list);
+                }
+                pthread_mutex_unlock (&priv->lock);
+
+                call_resume (sigstub->stub);
+
+                GF_FREE (sigstub);
+        }
+
+        return NULL;
+}
+
+int
+orderq (struct list_head *elem1, struct list_head *elem2)
+{
+        struct br_stub_signentry *s1 = NULL;
+        struct br_stub_signentry *s2 = NULL;
+
+        s1 = list_entry (elem1, struct br_stub_signentry, list);
+        s2 = list_entry (elem2, struct br_stub_signentry, list);
+
+        return (s1->v > s2->v);
+}
+
 static inline int
-br_stub_compare_sign_version (xlator_t *this, inode_t *inode,
-                              br_signature_t *sbuf, dict_t *dict)
+br_stub_compare_sign_version (xlator_t *this,
+                              inode_t *inode,
+                              br_signature_t *sbuf,
+                              dict_t *dict, int *fakesuccess)
 {
-        int32_t ret = -1;
-        br_stub_inode_ctx_t *ctx = NULL;
-        uint64_t tmp_ctx = 0;
+        int32_t              ret     = -1;
+        uint64_t             tmp_ctx = 0;
+        gf_boolean_t         invalid = _gf_false;
+        br_stub_inode_ctx_t *ctx     = NULL;
 
         GF_VALIDATE_OR_GOTO ("bit-rot-stub", this, out);
         GF_VALIDATE_OR_GOTO (this->name, inode, out);
@@ -586,28 +686,36 @@ br_stub_compare_sign_version (xlator_t *this, inode_t *inode,
                 goto out;
         }
 
-        ret = -1;
         ctx = (br_stub_inode_ctx_t *)(long)tmp_ctx;
 
         LOCK (&inode->lock);
         {
-                if (ctx->currentversion == sbuf->signedversion)
-                        ret = 0;
-                else
-                        gf_log (this->name, GF_LOG_WARNING, "current version "
-                                "%lu and version of the signature %lu are not "
-                                "same", ctx->currentversion,
-                                sbuf->signedversion);
+                if (ctx->currentversion < sbuf->signedversion) {
+                        invalid = _gf_true;
+                } else if (ctx->currentversion > sbuf->signedversion) {
+                        gf_log (this->name, GF_LOG_DEBUG, "\"Signing version\" "
+                                "(%lu) lower than \"Current version \" (%lu)",
+                                ctx->currentversion, sbuf->signedversion);
+                        *fakesuccess = 1;
+                }
         }
         UNLOCK (&inode->lock);
 
-out:
+        if (invalid) {
+                ret = -1;
+                gf_log (this->name, GF_LOG_WARNING,
+                        "Signing version exceeds current version [%lu > %lu]",
+                        sbuf->signedversion, ctx->currentversion);
+        }
+
+ out:
         return ret;
 }
 
 static inline int
-br_stub_prepare_signature (xlator_t *this, dict_t *dict,
-                           inode_t *inode, br_isignature_t *sign)
+br_stub_prepare_signature (xlator_t *this,
+                           dict_t *dict, inode_t *inode,
+                           br_isignature_t *sign, int *fakesuccess)
 {
         int32_t ret = 0;
         size_t signaturelen = 0;
@@ -624,7 +732,8 @@ br_stub_prepare_signature (xlator_t *this, dict_t *dict,
         if (ret)
                 goto dealloc_versions;
 
-        ret = br_stub_compare_sign_version (this, inode, sbuf, dict);
+        ret = br_stub_compare_sign_version (this, inode,
+                                            sbuf, dict, fakesuccess);
         if (ret)
                 goto dealloc_versions;
 
@@ -641,15 +750,27 @@ br_stub_handle_object_signature (call_frame_t *frame,
                                  xlator_t *this, fd_t *fd, dict_t *dict,
                                  br_isignature_t *sign, dict_t *xdata)
 {
-        int32_t      ret  = -1;
-        gf_boolean_t xref = _gf_false;
+        int32_t                   ret         = -1;
+        int32_t                   op_ret      = -1;
+        int32_t                   op_errno    = EINVAL;
+        int                       fakesuccess = 0;
+        br_stub_private_t        *priv        = NULL;
+        struct br_stub_signentry *sigstub     = NULL;
+
+        priv = this->private;
 
         if (frame->root->pid != GF_CLIENT_PID_BITD)
                 goto dofop;
 
-        ret = br_stub_prepare_signature (this, dict, fd->inode, sign);
+        ret = br_stub_prepare_signature (this, dict,
+                                         fd->inode, sign, &fakesuccess);
         if (ret)
                 goto dofop;
+        if (fakesuccess) {
+                op_ret = op_errno = 0;
+                goto dofop;
+        }
+
         dict_del (dict, GLUSTERFS_SET_OBJECT_SIGNATURE);
 
         ret = -1;
@@ -661,23 +782,37 @@ br_stub_handle_object_signature (call_frame_t *frame,
                 dict_ref (xdata);
         }
 
-        xref = _gf_true;
         ret = dict_set_int32 (xdata, GLUSTERFS_DURABLE_OP, 0);
-
- dofop:
         if (ret)
-                STACK_UNWIND_STRICT (fsetxattr, frame, -1, EINVAL, NULL);
-        else {
-                gf_log (this->name, GF_LOG_DEBUG, "SIGNED VERSION: %lu",
-                        sign->signedversion);
+                goto unref_dict;
+
+        /* prepare dispatch stub to order object signing */
+        sigstub = GF_CALLOC (1, sizeof (*sigstub), gf_br_stub_mt_sigstub_t);
+        if (!sigstub)
+                goto unref_dict;
+
+        INIT_LIST_HEAD (&sigstub->list);
+        sigstub->v = ntohl (sign->signedversion);
+        sigstub->stub = fop_fsetxattr_stub (frame, br_stub_perform_objsign,
+                                            fd, dict, 0, xdata);
+        if (!sigstub->stub)
+                goto cleanup_stub;
 
-                STACK_WIND (frame, default_setxattr_cbk, FIRST_CHILD (this),
-                            FIRST_CHILD (this)->fops->fsetxattr, fd, dict, 0,
-                            xdata);
+        pthread_mutex_lock (&priv->lock);
+        {
+                list_add_order (&sigstub->list, &priv->squeue, orderq);
+                pthread_cond_signal (&priv->cond);
         }
+        pthread_mutex_unlock (&priv->lock);
 
-        if (xref)
-                dict_unref (xdata);
+        return;
+
+ cleanup_stub:
+        GF_FREE (sigstub);
+ unref_dict:
+        dict_unref (xdata);
+ dofop:
+        STACK_UNWIND_STRICT (fsetxattr, frame, op_ret, op_errno, NULL);
 }
 
 int32_t
diff --git a/xlators/features/bit-rot/src/stub/bit-rot-stub.h b/xlators/features/bit-rot/src/stub/bit-rot-stub.h
index e1e7b383f42..9304ef80c51 100644
--- a/xlators/features/bit-rot/src/stub/bit-rot-stub.h
+++ b/xlators/features/bit-rot/src/stub/bit-rot-stub.h
@@ -75,6 +75,12 @@ typedef struct br_stub_private {
         uint32_t boot[2];
         char export[PATH_MAX];
 
+        pthread_mutex_t lock;
+        pthread_cond_t  cond;
+
+        struct list_head squeue;      /* ordered signing queue */
+        pthread_t signth;
+
         struct mem_pool *local_pool;
 } br_stub_private_t;