From: "Darrick J. Wong" <darrick.wong@oracle.com>
To: tytso@mit.edu, darrick.wong@oracle.com
Cc: linux-ext4@vger.kernel.org
Subject: [PATCH 10/47] libext2fs: support BLKZEROOUT/FALLOC_FL_ZERO_RANGE in ext2fs_zero_blocks
Date: Fri, 07 Nov 2014 13:51:52 -0800 [thread overview]
Message-ID: <20141107215152.883.6795.stgit@birch.djwong.org> (raw)
In-Reply-To: <20141107215042.883.49888.stgit@birch.djwong.org>
Plumb a new call into the IO manager to support translating
ext2fs_zero_blocks calls into the equivalent kernel-level BLKZEROOUT
ioctl or FALLOC_FL_ZERO_RANGE fallocate flag primitives when possible.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
lib/ext2fs/ext2_io.h | 7 ++-
lib/ext2fs/io_manager.c | 11 ++++
lib/ext2fs/mkjournal.c | 5 ++
lib/ext2fs/test_io.c | 21 ++++++++
lib/ext2fs/unix_io.c | 125 +++++++++++++++++++++++++++++++++++++++++++++++
5 files changed, 168 insertions(+), 1 deletion(-)
diff --git a/lib/ext2fs/ext2_io.h b/lib/ext2fs/ext2_io.h
index 4c5a5c5..1faa720 100644
--- a/lib/ext2fs/ext2_io.h
+++ b/lib/ext2fs/ext2_io.h
@@ -93,7 +93,9 @@ struct struct_io_manager {
errcode_t (*cache_readahead)(io_channel channel,
unsigned long long block,
unsigned long long count);
- long reserved[15];
+ errcode_t (*zeroout)(io_channel channel, unsigned long long block,
+ unsigned long long count);
+ long reserved[14];
};
#define IO_FLAG_RW 0x0001
@@ -125,6 +127,9 @@ extern errcode_t io_channel_write_blk64(io_channel channel,
extern errcode_t io_channel_discard(io_channel channel,
unsigned long long block,
unsigned long long count);
+extern errcode_t io_channel_zeroout(io_channel channel,
+ unsigned long long block,
+ unsigned long long count);
extern errcode_t io_channel_alloc_buf(io_channel channel,
int count, void *ptr);
extern errcode_t io_channel_cache_readahead(io_channel io,
diff --git a/lib/ext2fs/io_manager.c b/lib/ext2fs/io_manager.c
index dc5888d..c395d61 100644
--- a/lib/ext2fs/io_manager.c
+++ b/lib/ext2fs/io_manager.c
@@ -112,6 +112,17 @@ errcode_t io_channel_discard(io_channel channel, unsigned long long block,
return EXT2_ET_UNIMPLEMENTED;
}
+errcode_t io_channel_zeroout(io_channel channel, unsigned long long block,
+ unsigned long long count)
+{
+ EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+
+ if (channel->manager->zeroout)
+ return (channel->manager->zeroout)(channel, block, count);
+
+ return EXT2_ET_UNIMPLEMENTED;
+}
+
errcode_t io_channel_alloc_buf(io_channel io, int count, void *ptr)
{
size_t size;
diff --git a/lib/ext2fs/mkjournal.c b/lib/ext2fs/mkjournal.c
index fdd788b..2cd7ca5 100644
--- a/lib/ext2fs/mkjournal.c
+++ b/lib/ext2fs/mkjournal.c
@@ -171,6 +171,11 @@ errcode_t ext2fs_zero_blocks3(ext2_filsys fs, blk64_t blk, int num,
if (num <= 0)
return 0;
+ /* Try a zero out command, if supported */
+ retval = io_channel_zeroout(fs->io, blk, num);
+ if (retval == 0)
+ return 0;
+
/* If the user gave us a buffer, write that out */
if (block_buf) {
retval = io_channel_write_blk64(fs->io, blk, num, block_buf);
diff --git a/lib/ext2fs/test_io.c b/lib/ext2fs/test_io.c
index b03a939..f7c50d1 100644
--- a/lib/ext2fs/test_io.c
+++ b/lib/ext2fs/test_io.c
@@ -86,6 +86,7 @@ void (*test_io_cb_write_byte)
#define TEST_FLAG_SET_OPTION 0x20
#define TEST_FLAG_DISCARD 0x40
#define TEST_FLAG_READAHEAD 0x80
+#define TEST_FLAG_ZEROOUT 0x100
static void test_dump_block(io_channel channel,
struct test_private_data *data,
@@ -507,6 +508,25 @@ static errcode_t test_cache_readahead(io_channel channel,
return retval;
}
+static errcode_t test_zeroout(io_channel channel, unsigned long long block,
+ unsigned long long count)
+{
+ struct test_private_data *data;
+ errcode_t retval = 0;
+
+ EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+ data = (struct test_private_data *) channel->private_data;
+ EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_TEST_IO_CHANNEL);
+
+ if (data->real)
+ retval = io_channel_zeroout(data->real, block, count);
+ if (data->flags & TEST_FLAG_ZEROOUT)
+ fprintf(data->outfile,
+ "Test_io: zeroout(%llu, %llu) returned %s\n",
+ block, count, retval ? error_message(retval) : "OK");
+ return retval;
+}
+
static struct struct_io_manager struct_test_manager = {
.magic = EXT2_ET_MAGIC_IO_MANAGER,
.name = "Test I/O Manager",
@@ -523,6 +543,7 @@ static struct struct_io_manager struct_test_manager = {
.write_blk64 = test_write_blk64,
.discard = test_discard,
.cache_readahead = test_cache_readahead,
+ .zeroout = test_zeroout,
};
io_manager test_io_manager = &struct_test_manager;
diff --git a/lib/ext2fs/unix_io.c b/lib/ext2fs/unix_io.c
index c3a8ea5..6bc543b 100644
--- a/lib/ext2fs/unix_io.c
+++ b/lib/ext2fs/unix_io.c
@@ -987,6 +987,130 @@ unimplemented:
return EXT2_ET_UNIMPLEMENTED;
}
+#if defined(__linux__) && !defined(BLKZEROOUT)
+#define BLKZEROOUT _IO(0x12, 127)
+#endif
+
+#if defined(__linux__) && !defined(BLKFLSBUF)
+#define BLKFLSBUF _IO(0x12, 97)
+#endif
+
+#if defined(__linux__) && !defined(FALLOC_FL_ZERO_RANGE)
+#define FALLOC_FL_ZERO_RANGE 0x10
+#endif
+
+static errcode_t unix_zeroout(io_channel channel, unsigned long long block,
+ unsigned long long count)
+{
+ struct unix_private_data *data;
+ int ret;
+
+ EXT2_CHECK_MAGIC(channel, EXT2_ET_MAGIC_IO_CHANNEL);
+ data = (struct unix_private_data *) channel->private_data;
+ EXT2_CHECK_MAGIC(data, EXT2_ET_MAGIC_UNIX_IO_CHANNEL);
+
+ if (getenv("UNIX_IO_NOZEROOUT"))
+ goto unimplemented;
+
+ if (channel->flags & CHANNEL_FLAGS_BLOCK_DEVICE) {
+ __u64 range[2];
+
+ if (count == 0)
+ return 0;
+ /*
+ * This is a heavyweight operation. Refuse to
+ * zero-out a single block, and hope that the
+ * caller will fall back to io_channel_write().
+ */
+ else if (count == 1)
+ goto unimplemented;
+
+ range[0] = (__u64)(block) * channel->block_size;
+ range[1] = (__u64)(count) * channel->block_size;
+
+ errno = EOPNOTSUPP;
+ ret = -1;
+#if defined(BLKZEROOUT) && (defined(POSIX_FADV_DONTNEED) || defined(BLKFLSBUF))
+ /*
+ * BLKZEROOUT doesn't invalidate the page cache. Therefore,
+ * we need to flush dirty pages to disk prior to calling the
+ * ioctl and invalidate at least those same pages from the page
+ * cache afterwards. If we're using direct IO we can skip all
+ * that wrangling.
+ */
+ if (!(data->flags & IO_FLAG_DIRECT_IO)) {
+ ret = fsync(data->dev);
+ if (ret)
+ goto err;
+ }
+ ret = ioctl(data->dev, BLKZEROOUT, range);
+ if (ret)
+ goto err;
+ if (!(data->flags & IO_FLAG_DIRECT_IO)) {
+#ifdef POSIX_FADV_DONTNEED
+ ret = posix_fadvise(data->dev, range[0], range[1],
+ POSIX_FADV_DONTNEED);
+ if (ret == 0)
+ goto err;
+#endif
+#ifdef BLKFLSBUF
+ ret = ioctl(data->dev, BLKFLSBUF, 0);
+#endif
+ }
+#elif defined(BLKZEROOUT)
+ if (data->flags & IO_FLAG_DIRECT_IO)
+ ret = ioctl(data->dev, BLKZEROOUT, range);
+#endif
+ } else {
+ /* Regular file, try to use truncate/punch/zero. */
+#if defined(HAVE_FALLOCATE) && (defined(FALLOC_FL_ZERO_RANGE) || \
+ (defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)))
+ struct stat statbuf;
+
+ if (count == 0)
+ return 0;
+ /*
+ * If we're trying to zero a range past the end of the file,
+ * extend the file size, then punch (or zero_range) everything.
+ */
+ ret = fstat(data->dev, &statbuf);
+ if (ret)
+ goto err;
+ if (statbuf.st_size < (block + count) * channel->block_size) {
+ ret = ftruncate(data->dev,
+ (block + count) * channel->block_size);
+ if (ret)
+ goto err;
+ }
+#if defined(FALLOC_FL_PUNCH_HOLE) && defined(FALLOC_FL_KEEP_SIZE)
+ ret = fallocate(data->dev,
+ FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
+ (off_t)(block) * channel->block_size,
+ (off_t)(count) * channel->block_size);
+ if (ret == 0)
+ goto err;
+#endif
+#ifdef FALLOC_FL_ZERO_RANGE
+ ret = fallocate(data->dev,
+ FALLOC_FL_ZERO_RANGE,
+ (off_t)(block) * channel->block_size,
+ (off_t)(count) * channel->block_size);
+#endif
+#else
+ goto unimplemented;
+#endif /* HAVE_FALLOCATE && (ZERO_RANGE || (PUNCH_HOLE && KEEP_SIZE)) */
+ }
+err:
+ if (ret < 0) {
+ if (errno == EOPNOTSUPP)
+ goto unimplemented;
+ return errno;
+ }
+ return 0;
+unimplemented:
+ return EXT2_ET_UNIMPLEMENTED;
+}
+
static struct struct_io_manager struct_unix_manager = {
.magic = EXT2_ET_MAGIC_IO_MANAGER,
.name = "Unix I/O Manager",
@@ -1003,6 +1127,7 @@ static struct struct_io_manager struct_unix_manager = {
.write_blk64 = unix_write_blk64,
.discard = unix_discard,
.cache_readahead = unix_cache_readahead,
+ .zeroout = unix_zeroout,
};
io_manager unix_io_manager = &struct_unix_manager;
next prev parent reply other threads:[~2014-11-07 21:51 UTC|newest]
Thread overview: 104+ messages / expand[flat|nested] mbox.gz Atom feed top
2014-11-07 21:50 [PATCH 00/47] e2fsprogs November 2014 patchbomb Darrick J. Wong
2014-11-07 21:50 ` [PATCH 01/47] tests: fix test scripts that don't work on non-Linux systems Darrick J. Wong
2014-11-08 1:55 ` Theodore Ts'o
2014-11-07 21:50 ` [PATCH 02/47] misc: fix compiler warnings and minor build errors Darrick J. Wong
2014-11-08 2:25 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 03/47] dumpe2fs: don't crash when the user provides no block device argument Darrick J. Wong
2014-11-08 2:26 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 04/47] libext2fs: fix endian handling error; reduce fragmentation some Darrick J. Wong
2014-11-17 21:26 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 05/47] libext2fs: set BLOCK_UNINIT for non-last blockgroups if all blocks are free Darrick J. Wong
2014-11-17 22:47 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 06/47] libext2fs: don't allow alloc_stats on bad inode/block numbers Darrick J. Wong
2014-11-17 23:00 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 07/47] libext2fs: refactor extent head creation Darrick J. Wong
2014-12-03 3:55 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 08/47] libext2fs: file IO routines should handle uninit blocks Darrick J. Wong
2014-12-03 3:57 ` Theodore Ts'o
2014-11-07 21:51 ` [PATCH 09/47] libext2fs: use a dynamically sized (or caller-provided) block zeroing buffer Darrick J. Wong
2014-12-12 2:37 ` Theodore Ts'o
2014-12-12 5:02 ` Darrick J. Wong
2014-12-13 16:24 ` Theodore Ts'o
2014-12-13 16:25 ` Theodore Ts'o
2014-12-15 16:09 ` Andreas Dilger
2014-11-07 21:51 ` Darrick J. Wong [this message]
2014-12-13 16:29 ` [PATCH 10/47] libext2fs: support BLKZEROOUT/FALLOC_FL_ZERO_RANGE in ext2fs_zero_blocks Theodore Ts'o
2014-12-14 3:08 ` Darrick J. Wong
2014-11-07 21:51 ` [PATCH 11/47] libext2fs: find inode goal when allocating blocks Darrick J. Wong
2014-12-14 1:13 ` Theodore Ts'o
2014-12-14 21:02 ` Darrick J. Wong
2014-11-07 21:52 ` [PATCH 12/47] libext2fs: set interior tree block goal more intelligently Darrick J. Wong
2014-12-14 1:17 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 13/47] libext2fs: add a way to check the theoretical maximum extent tree depth Darrick J. Wong
2014-12-14 2:23 ` Theodore Ts'o
2014-12-14 3:11 ` Darrick J. Wong
2014-12-15 3:48 ` Theodore Ts'o
2014-12-15 6:38 ` Darrick J. Wong
2014-11-07 21:52 ` [PATCH 14/47] libext2fs: ext2fs_new_block2() should call alloc_block hook Darrick J. Wong
2014-11-07 21:52 ` [PATCH 15/47] misc: don't allow mk_hugefiles unless the fs supports extents Darrick J. Wong
2014-12-14 2:51 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 16/47] dumpe2fs: 80 column outputs, please Darrick J. Wong
2014-12-14 2:52 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 17/47] dumpe2fs: output cleanup Darrick J. Wong
2014-12-14 2:53 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 18/47] e2fsck: only complain about no-checksum directory blocks once Darrick J. Wong
2014-12-14 2:55 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 19/47] e2fsck: don't complain about root dir csum failures when getting lnf Darrick J. Wong
2014-12-14 2:57 ` Theodore Ts'o
2014-11-07 21:52 ` [PATCH 20/47] tune2fs: warn if extents are not enabled when turning on metadata_csum Darrick J. Wong
2014-12-14 2:58 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 21/47] tune2fs: enable uninit_bg when disabling metadata_csum Darrick J. Wong
2014-12-14 2:58 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 22/47] tests: testcases for enabling/disabling metadata_csum via tune2fs Darrick J. Wong
2014-12-14 3:00 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 23/47] mke2fs: don't zero inode table blocks that are already zeroed Darrick J. Wong
2014-12-14 3:01 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 24/47] resize2fs: don't exit if shrinking sparse_super2 fs to one bg Darrick J. Wong
2014-12-14 3:06 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 25/47] resize2fs: quickly rewrite extent blocks when moving an inode w/ metadata_csum Darrick J. Wong
2014-12-14 3:09 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 26/47] resize2fs: use old_fs to detect per-bg metadata blocks to free Darrick J. Wong
2014-12-15 2:09 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 27/47] resize2fs: don't interpret bitmap shift while crossing flexbg as raid stride Darrick J. Wong
2014-12-15 2:11 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 28/47] resize2fs: set bg flags and unused inode count when resizing Darrick J. Wong
2014-12-15 2:12 ` Theodore Ts'o
2014-11-07 21:53 ` [PATCH 29/47] resize2fs: don't mark unallocated bg metadata blocks when fixing bg flags Darrick J. Wong
2014-12-15 2:17 ` Theodore Ts'o
2014-12-15 7:09 ` Darrick J. Wong
2014-12-19 19:15 ` Darrick J. Wong
2014-11-07 21:54 ` [PATCH 30/47] resize2fs: don't play stupid games with the block count Darrick J. Wong
2014-12-15 3:13 ` Theodore Ts'o
2014-11-07 21:54 ` [PATCH 31/47] libext2fs/e2fsck: provide routines to read-ahead metadata Darrick J. Wong
2014-12-15 17:36 ` Theodore Ts'o
2014-12-15 19:19 ` Darrick J. Wong
2014-11-07 21:54 ` [PATCH 32/47] e2fsck: read-ahead metadata during passes 1, 2, and 4 Darrick J. Wong
2014-12-10 20:27 ` Darrick J. Wong
2014-12-10 22:00 ` Theodore Ts'o
2014-11-07 21:54 ` [PATCH 33/47] e2fsck: rebuild sparse extent trees/convert non-extent ext3 files Darrick J. Wong
2014-12-04 23:20 ` Andreas Dilger
2014-12-04 23:45 ` Darrick J. Wong
2014-12-11 22:05 ` [PATCH v2 " Darrick J. Wong
2014-11-07 21:54 ` [PATCH 34/47] tests: verify proper rebuilding of sparse extent trees and block map file conversion Darrick J. Wong
2014-11-07 21:54 ` [PATCH 35/47] resize2fs: convert fs to and from 64bit mode Darrick J. Wong
2014-12-15 3:32 ` Theodore Ts'o
2014-12-15 6:41 ` Darrick J. Wong
2014-12-15 17:46 ` Theodore Ts'o
2014-11-07 21:54 ` [PATCH 36/47] tests: test resize2fs 32->64 and 64->32bit conversion code Darrick J. Wong
2014-12-15 17:46 ` Theodore Ts'o
2014-11-07 21:54 ` [PATCH 37/47] libext2fs: support allocating uninit blocks in bmap2() Darrick J. Wong
2014-11-07 21:54 ` [PATCH 38/47] libext2fs: find/alloc a range of empty blocks Darrick J. Wong
2014-11-07 21:54 ` [PATCH 39/47] libext2fs: add new hooks to support large allocations Darrick J. Wong
2014-11-07 21:55 ` [PATCH 40/47] libext2fs: implement fallocate Darrick J. Wong
2014-11-07 21:55 ` [PATCH 41/47] libext2fs: use fallocate for creating journals and hugefiles Darrick J. Wong
2014-11-07 21:55 ` [PATCH 42/47] debugfs: implement fallocate Darrick J. Wong
2014-11-07 21:55 ` [PATCH 43/47] tests: test debugfs punch command Darrick J. Wong
2014-11-07 21:55 ` [PATCH 45/47] fuse2fs: translate ACL structures Darrick J. Wong
2014-11-07 21:55 ` [PATCH 46/47] fuse2fs: handle 64-bit dates correctly Darrick J. Wong
2014-11-07 21:55 ` [PATCH 47/47] fuse2fs: implement fallocate Darrick J. Wong
2014-11-12 22:43 ` [PATCH 48/47] misc: fix infinite loop when finding the start of the hugefile start range Darrick J. Wong
2014-12-03 3:06 ` Theodore Ts'o
2014-11-27 0:01 ` [PATCH 49/47] libext2fs: don't report garbage inodes with really large inodes Darrick J. Wong
2014-12-03 3:18 ` Theodore Ts'o
2014-12-04 20:39 ` [PATCH 51/47] e2fsck: force-reread of inode from disk when re-checking a checksum error Darrick J. Wong
2014-12-11 22:49 ` Theodore Ts'o
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20141107215152.883.6795.stgit@birch.djwong.org \
--to=darrick.wong@oracle.com \
--cc=linux-ext4@vger.kernel.org \
--cc=tytso@mit.edu \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).