From: Richard W.M. Jones Date: Tue, 16 Aug 2011 11:39:28 +0000 (+0100) Subject: Improve zeroing and detection of zeroes. X-Git-Tag: 1.13.5~5 X-Git-Url: http://git.annexia.org/?a=commitdiff_plain;h=5a50c04906828f6e99db6a9be420c84114476d39;p=libguestfs.git Improve zeroing and detection of zeroes. This code modifies zero, zero-device, is-zero, is-zero-device. zero and zero-device are modified so that if the blocks of the device already contain zeroes, then we don't write zeroes. The reason for this is to avoid unnecessarily making the underlying storage non-sparse or (in the qcow2 case) growing it. is-zero and is-zero-device are modified so that zero detection is faster. This is a nice side effect of making the first change. Since avoiding unnecessary zeroing involves reading the blocks before writing them, whereas before we just blindly wrote, this can be slower. As you can see from the tests below, in the case where the disk is sparse, it actually turns out to be faster, because we avoid allocating the underlying blocks. However in the case where the disk is non-sparse and full of existing data, it is much slower. There might be a case for an API flag to adjust whether or not we perform the zero check. I did not add this flag because it is unlikely that the caller would have enough information to be able to set the flag correctly. (Elapsed time in seconds) Format Test case Before After Raw Sparse 16.4 5.3 Preallocated zero 17.0 18.8 Preallocated random 16.0 41.3 Qcow2 preallocation=off 18.7 5.6 preallocation=metadata 17.4 5.8 The current code uses a fixed block size of 4K for reading and writing. I also tried the same tests with a block size of 64K but it didn't make any significant difference. (Thanks to Federico Simoncelli for suggesting this change) --- diff --git a/daemon/zero.c b/daemon/zero.c index c8e79ae..c9f6bf7 100644 --- a/daemon/zero.c +++ b/daemon/zero.c @@ -28,26 +28,57 @@ #include "daemon.h" #include "actions.h" +/* Return true iff the buffer is all zero bytes. + * + * Note that gcc is smart enough to optimize this properly: + * http://stackoverflow.com/questions/1493936/faster-means-of-checking-for-an-empty-buffer-in-c/1493989#1493989 + */ +static int +is_zero (const char *buffer, size_t size) +{ + size_t i; + + for (i = 0; i < size; ++i) { + if (buffer[i] != 0) + return 0; + } + + return 1; +} + +static const char zero_buf[4096]; + int do_zero (const char *device) { - int fd, i; - char buf[4096]; + char buf[sizeof zero_buf]; + int fd; + size_t i, offset; - fd = open (device, O_WRONLY); + fd = open (device, O_RDWR); if (fd == -1) { reply_with_perror ("%s", device); return -1; } - memset (buf, 0, sizeof buf); - for (i = 0; i < 32; ++i) { - if (write (fd, buf, sizeof buf) != sizeof buf) { - reply_with_perror ("write: %s", device); + offset = i * sizeof zero_buf; + + /* Check if the block is already zero before overwriting it. */ + if (pread (fd, buf, sizeof buf, offset) != sizeof buf) { + reply_with_perror ("pread: %s", device); close (fd); return -1; } + + if (!is_zero (buf, sizeof buf)) { + if (pwrite (fd, zero_buf, sizeof zero_buf, offset) != sizeof zero_buf) { + reply_with_perror ("pwrite: %s", device); + close (fd); + return -1; + } + } + notify_progress ((uint64_t) i, 32); } @@ -67,14 +98,13 @@ do_zero_device (const char *device) return -1; uint64_t size = (uint64_t) ssize; - int fd = open (device, O_WRONLY); + int fd = open (device, O_RDWR); if (fd == -1) { reply_with_perror ("%s", device); return -1; } - char buf[1024*1024]; - memset (buf, 0, sizeof buf); + char buf[sizeof zero_buf]; uint64_t pos = 0; @@ -86,15 +116,28 @@ do_zero_device (const char *device) else n = (size_t) n64; /* safe because of if condition */ - ssize_t r = write (fd, buf, n); + /* Check if the block is already zero before overwriting it. */ + ssize_t r; + r = pread (fd, buf, n, pos); if (r == -1) { - reply_with_perror ("write: %s (with %" PRId64 " bytes left to write)", - device, size); + reply_with_perror ("pread: %s at offset %" PRIu64, device, pos); close (fd); return -1; } - pos += r; + if (!is_zero (buf, sizeof buf)) { + r = pwrite (fd, zero_buf, n, pos); + if (r == -1) { + reply_with_perror ("pwrite: %s (with %" PRId64 " bytes left to write)", + device, size); + close (fd); + return -1; + } + pos += r; + } + else + pos += n; + notify_progress (pos, size); } @@ -106,13 +149,11 @@ do_zero_device (const char *device) return 0; } -static char zero[BUFSIZ]; - int do_is_zero (const char *path) { int fd; - char buf[BUFSIZ]; + char buf[1024*1024]; ssize_t r; CHROOT_IN; @@ -125,7 +166,7 @@ do_is_zero (const char *path) } while ((r = read (fd, buf, sizeof buf)) > 0) { - if (memcmp (buf, zero, r) != 0) { + if (!is_zero (buf, r)) { close (fd); return 0; } @@ -149,7 +190,7 @@ int do_is_zero_device (const char *device) { int fd; - char buf[BUFSIZ]; + char buf[1024*1024]; ssize_t r; fd = open (device, O_RDONLY); @@ -159,7 +200,7 @@ do_is_zero_device (const char *device) } while ((r = read (fd, buf, sizeof buf)) > 0) { - if (memcmp (buf, zero, r) != 0) { + if (!is_zero (buf, r)) { close (fd); return 0; } diff --git a/generator/generator_actions.ml b/generator/generator_actions.ml index c684e35..a924d87 100644 --- a/generator/generator_actions.ml +++ b/generator/generator_actions.ml @@ -3006,6 +3006,10 @@ How many blocks are zeroed isn't specified (but it's I enough to securely wipe the device). It should be sufficient to remove any partition tables, filesystem superblocks and so on. +If blocks are already zero, then this command avoids writing +zeroes. This prevents the underlying device from becoming non-sparse +or growing unnecessarily. + See also: C, C, C"); @@ -5229,7 +5233,11 @@ is not large enough."); "\ This command writes zeroes over the entire C. Compare with C which just zeroes the first few blocks of -a device."); +a device. + +If blocks are already zero, then this command avoids writing +zeroes. This prevents the underlying device from becoming non-sparse +or growing unnecessarily."); ("txz_in", (RErr, [FileIn "tarball"; Pathname "directory"], []), 229, [Optional "xz"], [InitScratchFS, Always, TestOutput (