cluster/ec: Fix incorrect management of NFS requests

Some operations, specially those comming from NFS, do not use a
regular fd and use an anonymous fd (i.e. a previous open call has
not been sent). Any context information created during open or
create will not be present on these fd's, so we simply return NULL
for contexts of those fd.

Also it seems that NFS can send write requests with a very big
buffer (higher that the default value of 128 KB). Some changes
have been made to correctly handle these large buffers.

Change-Id: I281476bd0d2cbaad231822248d6a616fcf5d4003
BUG: 1122417
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
Reviewed-on: http://review.gluster.org/8367
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

3 files changed, 51 insertions, 89 deletions

diff --git a/xlators/cluster/ec/src/ec-helpers.c b/xlators/cluster/ec/src/ec-helpers.c
index cd0ac41ebe7..6eb836a1c62 100644
--- a/xlators/cluster/ec/src/ec-helpers.c
+++ b/xlators/cluster/ec/src/ec-helpers.c
@@ -531,6 +531,11 @@ ec_fd_t * __ec_fd_get(fd_t * fd, xlator_t * xl)
     ec_fd_t * ctx = NULL;
     uint64_t value = 0;
 
+    if (fd->anonymous)
+    {
+        return NULL;
+    }
+
     if ((__fd_ctx_get(fd, xl, &value) != 0) || (value == 0))
     {
         ctx = GF_MALLOC(sizeof(*ctx), ec_mt_ec_fd_t);
@@ -559,11 +564,14 @@ ec_fd_t * ec_fd_get(fd_t * fd, xlator_t * xl)
 {
     ec_fd_t * ctx = NULL;
 
-    LOCK(&fd->lock);
+    if (!fd->anonymous)
+    {
+        LOCK(&fd->lock);
 
-    ctx = __ec_fd_get(fd, xl);
+        ctx = __ec_fd_get(fd, xl);
 
-    UNLOCK(&fd->lock);
+        UNLOCK(&fd->lock);
+    }
 
     return ctx;
 }
diff --git a/xlators/cluster/ec/src/ec-inode-read.c b/xlators/cluster/ec/src/ec-inode-read.c
index 46f904cf321..484c5e40189 100644
--- a/xlators/cluster/ec/src/ec-inode-read.c
+++ b/xlators/cluster/ec/src/ec-inode-read.c
@@ -973,11 +973,10 @@ out:
 int32_t ec_readv_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk)
 {
     ec_cbk_data_t * ans = NULL;
-    struct iobuf_pool * pool = NULL;
     struct iobref * iobref = NULL;
     struct iobuf * iobuf = NULL;
     uint8_t * ptr = NULL, * buff = NULL;
-    size_t fsize = 0, size = 0, max = 0, slice = 0;
+    size_t fsize = 0, size = 0, max = 0;
     int32_t i = 0;
 
     if (cbk->op_ret < 0)
@@ -989,7 +988,7 @@ int32_t ec_readv_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk)
 
     if (cbk->op_ret > 0)
     {
-        struct iovec vector[cbk->int32 * cbk->count];
+        struct iovec vector[1];
         uint8_t * blocks[cbk->count];
         uint32_t values[cbk->count];
 
@@ -1001,50 +1000,33 @@ int32_t ec_readv_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk)
             goto out;
         }
         buff = GF_ALIGN_BUF(ptr, EC_BUFFER_ALIGN_SIZE);
+        for (i = 0, ans = cbk; ans != NULL; i++, ans = ans->next)
+        {
+            values[i] = ans->idx;
+            blocks[i] = buff;
+            buff += ec_iov_copy_to(buff, ans->vector, ans->int32, 0, fsize);
+        }
 
         iobref = iobref_new();
         if (iobref == NULL)
         {
             goto out;
         }
-
-        for (i = 0, ans = cbk; ans != NULL; i++, ans = ans->next)
+        iobuf = iobuf_get2(fop->xl->ctx->iobuf_pool, size);
+        if (iobuf == NULL)
         {
-            values[i] = ans->idx;
-            blocks[i] = buff;
-            buff += ec_iov_copy_to(buff, ans->vector, ans->int32, 0, fsize);
+            goto out;
         }
-
-        pool = fop->xl->ctx->iobuf_pool;
-        max = iobpool_default_pagesize(pool) / ec->stripe_size;
-        max *= ec->fragment_size;
-        i = 0;
-        do
+        if (iobref_add(iobref, iobuf) != 0)
         {
-            iobuf = iobuf_get(pool);
-            if (iobuf == NULL)
-            {
-                goto out;
-            }
-            if (iobref_add(iobref, iobuf) != 0)
-            {
-                goto out;
-            }
-
-            slice = fsize;
-            if (slice > max)
-            {
-                slice = max;
-            }
-            fsize -= slice;
+            goto out;
+        }
 
-            vector[i].iov_base = iobuf->ptr;
-            vector[i].iov_len = ec_method_decode(slice, ec->fragments, values,
-                                                 blocks, iobuf->ptr);
-            i++;
+        vector[0].iov_base = iobuf->ptr;
+        vector[0].iov_len = ec_method_decode(fsize, ec->fragments, values,
+                                             blocks, iobuf->ptr);
 
-            iobuf_unref(iobuf);
-        } while (fsize > 0);
+        iobuf_unref(iobuf);
 
         GF_FREE(ptr);
         ptr = NULL;
@@ -1063,27 +1045,20 @@ int32_t ec_readv_rebuild(ec_t * ec, ec_fop_data_t * fop, ec_cbk_data_t * cbk)
             max = fop->user_size;
         }
         size -= fop->head;
-        while (size > max)
+        if (size > max)
         {
-            if (size - max >= vector[i - 1].iov_len)
-            {
-                size -= vector[--i].iov_len;
-            }
-            else
-            {
-                vector[i - 1].iov_len -= size - max;
-                size = max;
-            }
+            vector[0].iov_len -= size - max;
+            size = max;
         }
 
         cbk->op_ret = size;
-        cbk->int32 = i;
+        cbk->int32 = 1;
 
         iobref_unref(cbk->buffers);
         cbk->buffers = iobref;
 
         GF_FREE(cbk->vector);
-        cbk->vector = iov_dup(vector, i);
+        cbk->vector = iov_dup(vector, 1);
         if (cbk->vector == NULL)
         {
             cbk->op_ret = -1;
diff --git a/xlators/cluster/ec/src/ec-inode-write.c b/xlators/cluster/ec/src/ec-inode-write.c
index c0694531151..06a2fef8d17 100644
--- a/xlators/cluster/ec/src/ec-inode-write.c
+++ b/xlators/cluster/ec/src/ec-inode-write.c
@@ -1932,18 +1932,10 @@ void ec_wind_writev(ec_t * ec, ec_fop_data_t * fop, int32_t idx)
 {
     ec_trace("WIND", fop, "idx=%d", idx);
 
-    struct iovec vector[fop->int32];
-    struct iobuf_pool * pool = NULL;
+    struct iovec vector[1];
     struct iobref * iobref = NULL;
     struct iobuf * iobuf = NULL;
-    uint8_t * ptr = NULL;
-    ssize_t size = 0, slice = 0, pagesize = 0, maxsize = 0;
-    int32_t count = 0;
-
-    pool = fop->xl->ctx->iobuf_pool;
-
-    pagesize = iobpool_default_pagesize(pool);
-    maxsize = pagesize * ec->fragments;
+    ssize_t size = 0, bufsize = 0;
 
     iobref = iobref_new();
     if (iobref == NULL)
@@ -1951,43 +1943,30 @@ void ec_wind_writev(ec_t * ec, ec_fop_data_t * fop, int32_t idx)
         goto out;
     }
 
-    ptr = fop->vector[0].iov_base;
     size = fop->vector[0].iov_len;
+    bufsize = size / ec->fragments;
 
-    count = 0;
-    while (size > 0)
+    iobuf = iobuf_get2(fop->xl->ctx->iobuf_pool, bufsize);
+    if (iobuf == NULL)
     {
-        iobuf = iobuf_get(pool);
-        if (iobuf == NULL)
-        {
-            goto out;
-        }
-        if (iobref_add(iobref, iobuf) != 0)
-        {
-            goto out;
-        }
-
-        slice = size;
-        if (slice > maxsize)
-        {
-            slice = maxsize;
-        }
-
-        ec_method_encode(slice, ec->fragments, idx, ptr, iobuf->ptr);
-        ptr += slice;
+        goto out;
+    }
+    if (iobref_add(iobref, iobuf) != 0)
+    {
+        goto out;
+    }
 
-        vector[count].iov_base = iobuf->ptr;
-        vector[count].iov_len = slice / ec->fragments;
-        count++;
+    ec_method_encode(size, ec->fragments, idx, fop->vector[0].iov_base,
+                     iobuf->ptr);
 
-        iobuf_unref(iobuf);
+    vector[0].iov_base = iobuf->ptr;
+    vector[0].iov_len = bufsize;
 
-        size -= slice;
-    }
+    iobuf_unref(iobuf);
 
     STACK_WIND_COOKIE(fop->frame, ec_writev_cbk, (void *)(uintptr_t)idx,
                       ec->xl_list[idx], ec->xl_list[idx]->fops->writev,
-                      fop->fd, vector, count, fop->offset / ec->fragments,
+                      fop->fd, vector, 1, fop->offset / ec->fragments,
                       fop->uint32, iobref, fop->xdata);
 
     iobref_unref(iobref);