block: use int64_t instead of int in driver discard handlers
authorVladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Fri, 3 Sep 2021 10:28:06 +0000 (13:28 +0300)
committerEric Blake <eblake@redhat.com>
Wed, 29 Sep 2021 18:46:32 +0000 (13:46 -0500)
We are generally moving to int64_t for both offset and bytes parameters
on all io paths.

Main motivation is realization of 64-bit write_zeroes operation for
fast zeroing large disk chunks, up to the whole disk.

We chose signed type, to be consistent with off_t (which is signed) and
with possibility for signed return type (where negative value means
error).

So, convert driver discard handlers bytes parameter to int64_t.

The only caller of all updated function is bdrv_co_pdiscard in
block/io.c. It is already prepared to work with 64bit requests, but
pass at most max(bs->bl.max_pdiscard, INT_MAX) to the driver.

Let's look at all updated functions:

blkdebug: all calculations are still OK, thanks to
  bdrv_check_qiov_request().
  both rule_check and bdrv_co_pdiscard are 64bit

blklogwrites: pass to blk_loc_writes_co_log which is 64bit

blkreplay, copy-on-read, filter-compress: pass to bdrv_co_pdiscard, OK

copy-before-write: pass to bdrv_co_pdiscard which is 64bit and to
  cbw_do_copy_before_write which is 64bit

file-posix: one handler calls raw_account_discard() is 64bit and both
  handlers calls raw_do_pdiscard(). Update raw_do_pdiscard, which pass
  to RawPosixAIOData::aio_nbytes, which is 64bit (and calls
  raw_account_discard())

gluster: somehow, third argument of glfs_discard_async is size_t.
  Let's set max_pdiscard accordingly.

iscsi: iscsi_allocmap_set_invalid is 64bit,
  !is_byte_request_lun_aligned is 64bit.
  list.num is uint32_t. Let's clarify max_pdiscard and
  pdiscard_alignment.

mirror_top: pass to bdrv_mirror_top_do_write() which is
  64bit

nbd: protocol limitation. max_pdiscard is alredy set strict enough,
  keep it as is for now.

nvme: buf.nlb is uint32_t and we do shift. So, add corresponding limits
  to nvme_refresh_limits().

preallocate: pass to bdrv_co_pdiscard() which is 64bit.

rbd: pass to qemu_rbd_start_co() which is 64bit.

qcow2: calculations are still OK, thanks to bdrv_check_qiov_request(),
  qcow2_cluster_discard() is 64bit.

raw-format: raw_adjust_offset() is 64bit, bdrv_co_pdiscard too.

throttle: pass to bdrv_co_pdiscard() which is 64bit and to
  throttle_group_co_io_limits_intercept() which is 64bit as well.

test-block-iothread: bytes argument is unused

Great! Now all drivers are prepared to handle 64bit discard requests,
or else have explicit max_pdiscard limits.

Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
Message-Id: <20210903102807.27127-11-vsementsov@virtuozzo.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Eric Blake <eblake@redhat.com>
20 files changed:
block/blkdebug.c
block/blklogwrites.c
block/blkreplay.c
block/copy-before-write.c
block/copy-on-read.c
block/file-posix.c
block/filter-compress.c
block/gluster.c
block/iscsi.c
block/mirror.c
block/nbd.c
block/nvme.c
block/preallocate.c
block/qcow2.c
block/raw-format.c
block/rbd.c
block/throttle.c
block/trace-events
include/block/block_int.h
tests/unit/test-block-iothread.c

index 742b4a3834d843dffcf2144b6de153d2a7ab4786..bbf2948703083502cd2fa0ad872734e816dac6fa 100644 (file)
@@ -717,7 +717,7 @@ static int coroutine_fn blkdebug_co_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs,
-                                             int64_t offset, int bytes)
+                                             int64_t offset, int64_t bytes)
 {
     uint32_t align = bs->bl.pdiscard_alignment;
     int err;
index d7ae64c22d81c9264b72b3a57cac9cfe7a2fb049..f7a251e91f9eff96c434672711a74c7df0d56084 100644 (file)
@@ -484,9 +484,9 @@ static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
 }
 
 static int coroutine_fn
-blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
+blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
 {
-    return blk_log_writes_co_log(bs, offset, count, NULL, 0,
+    return blk_log_writes_co_log(bs, offset, bytes, NULL, 0,
                                  blk_log_writes_co_do_file_pdiscard,
                                  LOG_DISCARD_FLAG, false);
 }
index 89d74a3cca29561fb54b4252cb83ae2b5bef842e..dcbe780ddbd36b3c2071fb350ee7fdc8b5349cb7 100644 (file)
@@ -105,7 +105,7 @@ static int coroutine_fn blkreplay_co_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static int coroutine_fn blkreplay_co_pdiscard(BlockDriverState *bs,
-                                              int64_t offset, int bytes)
+                                              int64_t offset, int64_t bytes)
 {
     uint64_t reqid = blkreplay_next_id();
     int ret = bdrv_co_pdiscard(bs->file, offset, bytes);
index d210e87a4511c7ae364a0394afb419951ad29ac6..c30a5ff8dea933fce5b1913012d184c19aae6e83 100644 (file)
@@ -64,7 +64,7 @@ static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs,
 }
 
 static int coroutine_fn cbw_co_pdiscard(BlockDriverState *bs,
-                                        int64_t offset, int bytes)
+                                        int64_t offset, int64_t bytes)
 {
     int ret = cbw_do_copy_before_write(bs, offset, bytes, 0);
     if (ret < 0) {
index f83dd83f14edb4132fe238c22f927ad632d9b2fa..1fc7fb3333b1ef0e070eebdc873d55a24d3d5d45 100644 (file)
@@ -201,7 +201,7 @@ static int coroutine_fn cor_co_pwrite_zeroes(BlockDriverState *bs,
 
 
 static int coroutine_fn cor_co_pdiscard(BlockDriverState *bs,
-                                        int64_t offset, int bytes)
+                                        int64_t offset, int64_t bytes)
 {
     return bdrv_co_pdiscard(bs->file, offset, bytes);
 }
index f375070f25bbba41185807d98f2dfbb5d2db74f5..c62e42743de1ef7dc8713d71da2598ba9140dcce 100644 (file)
@@ -2942,7 +2942,8 @@ static void raw_account_discard(BDRVRawState *s, uint64_t nbytes, int ret)
 }
 
 static coroutine_fn int
-raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
+raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes,
+                bool blkdev)
 {
     BDRVRawState *s = bs->opaque;
     RawPosixAIOData acb;
@@ -2966,7 +2967,7 @@ raw_do_pdiscard(BlockDriverState *bs, int64_t offset, int bytes, bool blkdev)
 }
 
 static coroutine_fn int
-raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
+raw_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
 {
     return raw_do_pdiscard(bs, offset, bytes, false);
 }
@@ -3591,7 +3592,7 @@ hdev_co_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
 #endif /* linux */
 
 static coroutine_fn int
-hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
+hdev_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
 {
     BDRVRawState *s = bs->opaque;
     int ret;
index fb85686b69308d36182c8acc6138bde53808d942..d5be538619ae74d4f13ce6cc65c29144a88b4784 100644 (file)
@@ -94,7 +94,7 @@ static int coroutine_fn compress_co_pwrite_zeroes(BlockDriverState *bs,
 
 
 static int coroutine_fn compress_co_pdiscard(BlockDriverState *bs,
-                                             int64_t offset, int bytes)
+                                             int64_t offset, int64_t bytes)
 {
     return bdrv_co_pdiscard(bs->file, offset, bytes);
 }
index 4e3c9cd14fd9697c7f307b4feb5f365bba51c7e4..398976bc66d21208a09639d12b0288ca93bd7cf7 100644 (file)
@@ -891,6 +891,7 @@ out:
 static void qemu_gluster_refresh_limits(BlockDriverState *bs, Error **errp)
 {
     bs->bl.max_transfer = GLUSTER_MAX_TRANSFER;
+    bs->bl.max_pdiscard = SIZE_MAX;
 }
 
 static int qemu_gluster_reopen_prepare(BDRVReopenState *state,
@@ -1297,18 +1298,20 @@ error:
 
 #ifdef CONFIG_GLUSTERFS_DISCARD
 static coroutine_fn int qemu_gluster_co_pdiscard(BlockDriverState *bs,
-                                                 int64_t offset, int size)
+                                                 int64_t offset, int64_t bytes)
 {
     int ret;
     GlusterAIOCB acb;
     BDRVGlusterState *s = bs->opaque;
 
+    assert(bytes <= SIZE_MAX); /* rely on max_pdiscard */
+
     acb.size = 0;
     acb.ret = 0;
     acb.coroutine = qemu_coroutine_self();
     acb.aio_context = bdrv_get_aio_context(bs);
 
-    ret = glfs_discard_async(s->fd, offset, size, gluster_finish_aiocb, &acb);
+    ret = glfs_discard_async(s->fd, offset, bytes, gluster_finish_aiocb, &acb);
     if (ret < 0) {
         return -errno;
     }
index 74ff7e307ece02e867ab0bb9934f5ac5b3039462..57aa07a40d7f4cafa26197afca05a862681c9b4d 100644 (file)
@@ -1138,7 +1138,8 @@ iscsi_getlength(BlockDriverState *bs)
 }
 
 static int
-coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
+coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset,
+                               int64_t bytes)
 {
     IscsiLun *iscsilun = bs->opaque;
     struct IscsiTask iTask;
@@ -1154,6 +1155,12 @@ coroutine_fn iscsi_co_pdiscard(BlockDriverState *bs, int64_t offset, int bytes)
         return 0;
     }
 
+    /*
+     * We don't want to overflow list.num which is uint32_t.
+     * We rely on our max_pdiscard.
+     */
+    assert(bytes / iscsilun->block_size <= UINT32_MAX);
+
     list.lba = offset / iscsilun->block_size;
     list.num = bytes / iscsilun->block_size;
 
@@ -2071,10 +2078,9 @@ static void iscsi_refresh_limits(BlockDriverState *bs, Error **errp)
     }
 
     if (iscsilun->lbp.lbpu) {
-        if (iscsilun->bl.max_unmap < 0xffffffff / block_size) {
-            bs->bl.max_pdiscard =
-                iscsilun->bl.max_unmap * iscsilun->block_size;
-        }
+        bs->bl.max_pdiscard =
+            MIN_NON_ZERO(iscsilun->bl.max_unmap * iscsilun->block_size,
+                         (uint64_t)UINT32_MAX * iscsilun->block_size);
         bs->bl.pdiscard_alignment =
             iscsilun->bl.opt_unmap_gran * iscsilun->block_size;
     } else {
index fab75087b0bad2b457dec334a98658d5305ad070..c962e8b4712851645362c73ff2b55311983d2cff 100644 (file)
@@ -1508,7 +1508,7 @@ static int coroutine_fn bdrv_mirror_top_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static int coroutine_fn bdrv_mirror_top_pdiscard(BlockDriverState *bs,
-    int64_t offset, int bytes)
+    int64_t offset, int64_t bytes)
 {
     return bdrv_mirror_top_do_write(bs, MIRROR_METHOD_DISCARD, offset, bytes,
                                     NULL, 0);
index c0c479abe9ce464f5b55e9d94dacf3a678c6e53c..a66b2c282dc3d2e3748eba1cda4da7229880fc06 100644 (file)
@@ -1457,15 +1457,17 @@ static int nbd_client_co_flush(BlockDriverState *bs)
 }
 
 static int nbd_client_co_pdiscard(BlockDriverState *bs, int64_t offset,
-                                  int bytes)
+                                  int64_t bytes)
 {
     BDRVNBDState *s = (BDRVNBDState *)bs->opaque;
     NBDRequest request = {
         .type = NBD_CMD_TRIM,
         .from = offset,
-        .len = bytes,
+        .len = bytes, /* len is uint32_t */
     };
 
+    assert(bytes <= UINT32_MAX); /* rely on max_pdiscard */
+
     assert(!(s->info.flags & NBD_FLAG_READ_ONLY));
     if (!(s->info.flags & NBD_FLAG_SEND_TRIM) || !bytes) {
         return 0;
index 2e0fd9e76a2dacd1de682b440b559edc5ada764e..1cc7b62bb4b6073d1890cc7eb491475de2f51669 100644 (file)
@@ -1360,7 +1360,7 @@ static coroutine_fn int nvme_co_pwrite_zeroes(BlockDriverState *bs,
 
 static int coroutine_fn nvme_co_pdiscard(BlockDriverState *bs,
                                          int64_t offset,
-                                         int bytes)
+                                         int64_t bytes)
 {
     BDRVNVMeState *s = bs->opaque;
     NVMeQueuePair *ioq = s->queues[INDEX_IO(0)];
@@ -1387,6 +1387,14 @@ static int coroutine_fn nvme_co_pdiscard(BlockDriverState *bs,
 
     assert(s->queue_count > 1);
 
+    /*
+     * Filling the @buf requires @offset and @bytes to satisfy restrictions
+     * defined in nvme_refresh_limits().
+     */
+    assert(QEMU_IS_ALIGNED(bytes, 1UL << s->blkshift));
+    assert(QEMU_IS_ALIGNED(offset, 1UL << s->blkshift));
+    assert((bytes >> s->blkshift) <= UINT32_MAX);
+
     buf = qemu_try_memalign(s->page_size, s->page_size);
     if (!buf) {
         return -ENOMEM;
@@ -1490,6 +1498,10 @@ static void nvme_refresh_limits(BlockDriverState *bs, Error **errp)
     bs->bl.max_pwrite_zeroes = 1ULL << (s->blkshift + 16);
     bs->bl.pwrite_zeroes_alignment = MAX(bs->bl.request_alignment,
                                          1UL << s->blkshift);
+
+    bs->bl.max_pdiscard = (uint64_t)UINT32_MAX << s->blkshift;
+    bs->bl.pdiscard_alignment = MAX(bs->bl.request_alignment,
+                                    1UL << s->blkshift);
 }
 
 static void nvme_detach_aio_context(BlockDriverState *bs)
index 99e28d9f086ecf2200f796f2953285244fdfd349..1d4233f730082ae61f13b8bb55350ed8abf447d3 100644 (file)
@@ -235,7 +235,7 @@ static coroutine_fn int preallocate_co_preadv_part(
 }
 
 static int coroutine_fn preallocate_co_pdiscard(BlockDriverState *bs,
-                                               int64_t offset, int bytes)
+                                               int64_t offset, int64_t bytes)
 {
     return bdrv_co_pdiscard(bs->file, offset, bytes);
 }
index 4b2e8694952f0fb1d642b4b20c4d1bae2e9e6788..d50901675699ef3806bacb780ee0cf74e946cce5 100644 (file)
@@ -3996,7 +3996,7 @@ static coroutine_fn int qcow2_co_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static coroutine_fn int qcow2_co_pdiscard(BlockDriverState *bs,
-                                          int64_t offset, int bytes)
+                                          int64_t offset, int64_t bytes)
 {
     int ret;
     BDRVQcow2State *s = bs->opaque;
index a2485926b8cd93ab0f98c353c124a3e69168eaf8..bda757fd1954764c0b5e31127d4716b67302d8d1 100644 (file)
@@ -302,7 +302,7 @@ static int coroutine_fn raw_co_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static int coroutine_fn raw_co_pdiscard(BlockDriverState *bs,
-                                        int64_t offset, int bytes)
+                                        int64_t offset, int64_t bytes)
 {
     int ret;
 
index 053eb8e48f4427cba723cbf3a41e69cbeefce7cf..701fbf2b0cf035864c8d0a24c876713ba48e44c3 100644 (file)
@@ -1197,9 +1197,9 @@ static int coroutine_fn qemu_rbd_co_flush(BlockDriverState *bs)
 }
 
 static int coroutine_fn qemu_rbd_co_pdiscard(BlockDriverState *bs,
-                                             int64_t offset, int count)
+                                             int64_t offset, int64_t bytes)
 {
-    return qemu_rbd_start_co(bs, offset, count, NULL, 0, RBD_AIO_DISCARD);
+    return qemu_rbd_start_co(bs, offset, bytes, NULL, 0, RBD_AIO_DISCARD);
 }
 
 #ifdef LIBRBD_SUPPORTS_WRITE_ZEROES
index c13fe9067f4e9eedbcf86a579a3e0cf37d39370d..6e8d52fa24511d6f115803f9ef7ae9cc2697633a 100644 (file)
@@ -145,7 +145,7 @@ static int coroutine_fn throttle_co_pwrite_zeroes(BlockDriverState *bs,
 }
 
 static int coroutine_fn throttle_co_pdiscard(BlockDriverState *bs,
-                                             int64_t offset, int bytes)
+                                             int64_t offset, int64_t bytes)
 {
     ThrottleGroupMember *tgm = bs->opaque;
     throttle_group_co_io_limits_intercept(tgm, bytes, true);
index d8a08563f1caf3f74e4e7ce98de872c33f3e9157..f2d0a9b62a7eb272f8e59d58cd0ff0c73ccfed63 100644 (file)
@@ -152,8 +152,8 @@ nvme_write_zeroes(void *s, uint64_t offset, uint64_t bytes, int flags) "s %p off
 nvme_qiov_unaligned(const void *qiov, int n, void *base, size_t size, int align) "qiov %p n %d base %p size 0x%zx align 0x%x"
 nvme_prw_buffered(void *s, uint64_t offset, uint64_t bytes, int niov, int is_write) "s %p offset 0x%"PRIx64" bytes %"PRId64" niov %d is_write %d"
 nvme_rw_done(void *s, int is_write, uint64_t offset, uint64_t bytes, int ret) "s %p is_write %d offset 0x%"PRIx64" bytes %"PRId64" ret %d"
-nvme_dsm(void *s, uint64_t offset, uint64_t bytes) "s %p offset 0x%"PRIx64" bytes %"PRId64""
-nvme_dsm_done(void *s, uint64_t offset, uint64_t bytes, int ret) "s %p offset 0x%"PRIx64" bytes %"PRId64" ret %d"
+nvme_dsm(void *s, int64_t offset, int64_t bytes) "s %p offset 0x%"PRIx64" bytes %"PRId64""
+nvme_dsm_done(void *s, int64_t offset, int64_t bytes, int ret) "s %p offset 0x%"PRIx64" bytes %"PRId64" ret %d"
 nvme_dma_map_flush(void *s) "s %p"
 nvme_free_req_queue_wait(void *s, unsigned q_index) "s %p q #%u"
 nvme_create_queue_pair(unsigned q_index, void *q, size_t size, void *aio_context, int fd) "index %u q %p size %zu aioctx %p fd %d"
index 9b4e0748bc86ea33b141720dd4f485d7c99e460d..ffe86068d4d5a1ecb3afbb77ee5ee0ed37a6b301 100644 (file)
@@ -303,7 +303,7 @@ struct BlockDriver {
     int coroutine_fn (*bdrv_co_pwrite_zeroes)(BlockDriverState *bs,
         int64_t offset, int64_t bytes, BdrvRequestFlags flags);
     int coroutine_fn (*bdrv_co_pdiscard)(BlockDriverState *bs,
-        int64_t offset, int bytes);
+        int64_t offset, int64_t bytes);
 
     /* Map [offset, offset + nbytes) range onto a child of @bs to copy from,
      * and invoke bdrv_co_copy_range_from(child, ...), or invoke
index c86954c7baa1af0df1bd6f7c1de8ac824e92fb6f..aea660aeed8601f48cf5e73c8ec35c2e362c7258 100644 (file)
@@ -48,7 +48,7 @@ static int coroutine_fn bdrv_test_co_pwritev(BlockDriverState *bs,
 }
 
 static int coroutine_fn bdrv_test_co_pdiscard(BlockDriverState *bs,
-                                              int64_t offset, int bytes)
+                                              int64_t offset, int64_t bytes)
 {
     return 0;
 }