]> git.proxmox.com Git - mirror_zfs.git/commitdiff
Handle compressed buffers in __dbuf_hold_impl()
authorTim Chase <tim@chase2k.com>
Wed, 8 Nov 2017 21:32:15 +0000 (15:32 -0600)
committerTony Hutter <hutter2@llnl.gov>
Tue, 21 Nov 2017 19:01:30 +0000 (13:01 -0600)
In __dbuf_hold_impl(), if a buffer is currently syncing and is still
referenced from db_data, a copy is made in case it is dirtied again in
the txg.  Previously, the buffer for the copy was simply allocated with
arc_alloc_buf() which doesn't handle compressed or encrypted buffers
(which are a special case of a compressed buffer).  The result was
typically an invalid memory access because the newly-allocated buffer
was of the uncompressed size.

This commit fixes the problem by handling the 2 compressed cases,
encrypted and unencrypted, respectively, with arc_alloc_raw_buf() and
arc_alloc_compressed_buf().

Although using the proper allocation functions fixes the invalid memory
access by allocating a buffer of the compressed size, another unrelated
issue made it impossible to properly detect compressed buffers in the
first place.  The header's compression flag was set to ZIO_COMPRESS_OFF
in arc_write() when it was possible that an attached buffer was actually
compressed.  This commit adds logic to only set ZIO_COMPRESS_OFF in
the non-ZIO_RAW case which wil handle both cases of compressed buffers
(encrypted or unencrypted).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes #5742
Closes #6797

module/zfs/arc.c
module/zfs/dbuf.c

index 48950245e295fc723e266d0ff7d32a9de52b4595..2b0a78d4be4798a06be0f09cdc6916c34a4ef92a 100644 (file)
@@ -6087,6 +6087,10 @@ arc_write(zio_t *pio, spa_t *spa, uint64_t txg,
                VERIFY3P(buf->b_data, !=, NULL);
                arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF);
        }
+
+       if (!(zio_flags & ZIO_FLAG_RAW))
+               arc_hdr_set_compress(hdr, ZIO_COMPRESS_OFF);
+
        ASSERT(!arc_buf_is_shared(buf));
        ASSERT3P(hdr->b_l1hdr.b_pabd, ==, NULL);
 
index 62340469d3575b089fee6145fedf4201f733ce23..68d6e0160087e276cffcfe68211279a417c676a4 100644 (file)
@@ -63,7 +63,6 @@ struct dbuf_hold_impl_data {
        blkptr_t *dh_bp;
        int dh_err;
        dbuf_dirty_record_t *dh_dr;
-       arc_buf_contents_t dh_type;
        int dh_depth;
 };
 
@@ -2628,6 +2627,36 @@ dbuf_prefetch(dnode_t *dn, int64_t level, uint64_t blkid, zio_priority_t prio,
 
 #define        DBUF_HOLD_IMPL_MAX_DEPTH        20
 
+/*
+ * Helper function for __dbuf_hold_impl() to copy a buffer. Handles
+ * the case of compressed and uncompressed buffers by allocating the
+ * new buffer, respectively, with arc_alloc_raw_buf(),
+ * arc_alloc_compressed_buf() or arc_alloc_buf().*
+ *
+ * NOTE: Declared noinline to avoid stack bloat in __dbuf_hold_impl().
+ */
+noinline static void
+dbuf_hold_copy(struct dbuf_hold_impl_data *dh)
+{
+       dnode_t *dn = dh->dh_dn;
+       dmu_buf_impl_t *db = dh->dh_db;
+       dbuf_dirty_record_t *dr = dh->dh_dr;
+       arc_buf_t *data = dr->dt.dl.dr_data;
+
+       enum zio_compress compress_type = arc_get_compression(data);
+
+       if (compress_type != ZIO_COMPRESS_OFF) {
+               dbuf_set_data(db, arc_alloc_compressed_buf(
+                   dn->dn_objset->os_spa, db, arc_buf_size(data),
+                   arc_buf_lsize(data), compress_type));
+       } else {
+               dbuf_set_data(db, arc_alloc_buf(dn->dn_objset->os_spa, db,
+                   DBUF_GET_BUFC_TYPE(db), db->db.db_size));
+       }
+
+       bcopy(data->b_data, db->db.db_data, arc_buf_size(data));
+}
+
 /*
  * Returns with db_holds incremented, and db_mtx not held.
  * Note: dn_struct_rwlock must be held.
@@ -2693,16 +2722,8 @@ __dbuf_hold_impl(struct dbuf_hold_impl_data *dh)
            dh->dh_dn->dn_object != DMU_META_DNODE_OBJECT &&
            dh->dh_db->db_state == DB_CACHED && dh->dh_db->db_data_pending) {
                dh->dh_dr = dh->dh_db->db_data_pending;
-
-               if (dh->dh_dr->dt.dl.dr_data == dh->dh_db->db_buf) {
-                       dh->dh_type = DBUF_GET_BUFC_TYPE(dh->dh_db);
-
-                       dbuf_set_data(dh->dh_db,
-                           arc_alloc_buf(dh->dh_dn->dn_objset->os_spa,
-                           dh->dh_db, dh->dh_type, dh->dh_db->db.db_size));
-                       bcopy(dh->dh_dr->dt.dl.dr_data->b_data,
-                           dh->dh_db->db.db_data, dh->dh_db->db.db_size);
-               }
+               if (dh->dh_dr->dt.dl.dr_data == dh->dh_db->db_buf)
+                       dbuf_hold_copy(dh);
        }
 
        if (multilist_link_active(&dh->dh_db->db_cache_link)) {
@@ -2775,7 +2796,6 @@ __dbuf_hold_impl_init(struct dbuf_hold_impl_data *dh,
        dh->dh_bp = NULL;
        dh->dh_err = 0;
        dh->dh_dr = NULL;
-       dh->dh_type = 0;
 
        dh->dh_depth = depth;
 }