From: Allison Henderson Subject: Re: [XFS Punch Hole 1/1] XFS Add Punch Hole Testing to FSX Date: Mon, 02 May 2011 15:39:58 -0700 Message-ID: <4DBF32BE.70009@linux.vnet.ibm.com> References: <4DBF02FF.608@linux.vnet.ibm.com> <4DBF0498.6070905@redhat.com> <4DBF0789.3090808@redhat.com> <61E784AC-2E07-41DC-A65C-0C1B766A4A6F@dilger.ca> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Eric Sandeen , linux-fsdevel , Ext4 Developers List , xfs-oss To: Andreas Dilger Return-path: Received: from e7.ny.us.ibm.com ([32.97.182.137]:56117 "EHLO e7.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751646Ab1EBWkJ (ORCPT ); Mon, 2 May 2011 18:40:09 -0400 In-Reply-To: <61E784AC-2E07-41DC-A65C-0C1B766A4A6F@dilger.ca> Sender: linux-ext4-owner@vger.kernel.org List-ID: On 5/2/2011 1:29 PM, Andreas Dilger wrote: > On 5/2/11 2:16 PM, Allison Henderson wrote: >> This patch adds punch hole tests to the fsx >> stress test. The test is performed through >> the fallocate call by randomly choosing to >> use the punch hole flag when running the >> fallocate test. Regions that have >> been punched out should contain zeros, so >> the expected file contents buffer is updated >> to contain zeros when a hole is punched out. >> >> Signed-off-by: Allison Henderson >> --- >> :100644 100644 32cd380... d424941... M ltp/Makefile >> :100644 100644 fe072d3... 4f54ef6... M ltp/fsx.c >> ltp/Makefile | 2 +- >> ltp/fsx.c | 73 ++++++++++++++++++++++++++++++++++++++++++++++++--------- >> 2 files changed, 62 insertions(+), 13 deletions(-) >> >> diff --git a/ltp/Makefile b/ltp/Makefile >> index 32cd380..d424941 100644 >> --- a/ltp/Makefile >> +++ b/ltp/Makefile >> @@ -27,7 +27,7 @@ LCFLAGS += -DAIO >> LLDLIBS += -laio -lpthread >> endif >> >> -ifeq ($(HAVE_FALLOCATE), true) >> +ifeq ($(HAVE_FALLOCATE), yes) >> > >> LCFLAGS += -DFALLOCATE >> endif >> >> diff --git a/ltp/fsx.c b/ltp/fsx.c >> index fe072d3..4f54ef6 100644 >> --- a/ltp/fsx.c >> +++ b/ltp/fsx.c >> @@ -207,7 +207,8 @@ logdump(void) >> { >> int i, count, down; >> struct log_entry *lp; >> - char *falloc_type[3] = {"PAST_EOF", "EXTENDING", "INTERIOR"}; >> + char *falloc_type[4] = {"PAST_EOF", "EXTENDING", "INTERIOR", >> + "PUNCH_HOLE"}; >> >> prt("LOG DUMP (%d total operations):\n", logcount); >> if (logcount< LOGSIZE) { >> @@ -791,7 +792,11 @@ dofallocate(unsigned offset, unsigned length) >> { >> unsigned end_offset; >> int keep_size; >> - >> + int max_offset = 0; >> + int max_len = 0; >> + int punch_hole = 0; >> + int mode = 0; >> + char *op_name; >> if (length == 0) { >> if (!quiet&& testcalls> simulatedopcount) >> prt("skipping zero length fallocate\n"); >> @@ -799,11 +804,31 @@ dofallocate(unsigned offset, unsigned length) >> return; >> } >> >> +#ifdef FALLOC_FL_PUNCH_HOLE >> + punch_hole = random() % 2; >> + /* Keep size must be set for punch hole */ >> + if (punch_hole) { >> + keep_size = 1; >> + mode = FALLOC_FL_PUNCH_HOLE; >> + } else >> + keep_size = random() % 2; >> +#else >> keep_size = random() % 2; >> +#endif >> + >> + if (keep_size) >> + mode |= FALLOC_FL_KEEP_SIZE; >> + >> + if (punch_hole&& file_size<= (loff_t)offset) { >> + if (!quiet&& testcalls> simulatedopcount) >> + prt("skipping hole punch off the end of the file\n"); >> + log4(OP_SKIPPED, OP_FALLOCATE, offset, length); >> + return; >> + } > > Isn't a hole punch off the end of the file is just a truncate? I think that is a valid test case, since punch(newsize, ~0ULL) should be identical to truncate(newsize). In that case, "keep_size" would affect whether the file size is left alone, or it is now "newsize", so I don't think it should always be set to run with keep_size = 1 for FALLOC_FL_PUNCH_HOLE operations. Hi there, Well actually punch hole requires that the keep size flag be set when using the punch hole flag, so it's not quite the same as truncate. The punch hole operation does not modify the length of the file, it only changes whether or not the specified blocks are allocated. So a hole that extends off the end of the file would cause the last few blocks of the file to be released, but the length of the file would not shrink. The above code snippet though is checking to see if the entire hole is off the edge of the file. For example, the file is 3 blocks in length, but we are trying to punch out blocks 5 though 7. This operation would have no effect, so we skip it here. Allison Henderson > >> end_offset = keep_size ? 0 : offset + length; >> >> - if (end_offset> biggest) { >> + if ((end_offset> biggest)&& !punch_hole) { >> biggest = end_offset; >> if (!quiet&& testcalls> simulatedopcount) >> prt("fallocating to largest ever: 0x%x\n", end_offset); >> @@ -811,13 +836,15 @@ dofallocate(unsigned offset, unsigned length) >> >> /* >> * last arg: >> - * 1: allocate past EOF >> - * 2: extending prealloc >> - * 3: interior prealloc >> + * 0: allocate past EOF >> + * 1: extending prealloc >> + * 2: interior prealloc >> + * 3: punch hole >> */ >> - log4(OP_FALLOCATE, offset, length, (end_offset> file_size) ? (keep_size ? 1 : 2) : 3); >> + log4(OP_FALLOCATE, offset, length, punch_hole ? 3 : >> + (end_offset> file_size) ? (keep_size ? 0 : 1) : 2); >> >> - if (end_offset> file_size) { >> + if (((loff_t)end_offset> file_size)&& !punch_hole) { >> memset(good_buf + file_size, '\0', end_offset - file_size); >> file_size = end_offset; >> } >> @@ -827,13 +854,35 @@ dofallocate(unsigned offset, unsigned length) >> >> if ((progressinterval&& testcalls % progressinterval == 0) || >> (debug&& (monitorstart == -1 || monitorend == -1 || >> - end_offset<= monitorend))) >> - prt("%lu falloc\tfrom 0x%x to 0x%x\n", testcalls, offset, length); >> - if (fallocate(fd, keep_size ? FALLOC_FL_KEEP_SIZE : 0, (loff_t)offset, (loff_t)length) == -1) { >> - prt("fallocate: %x to %x\n", offset, length); >> + end_offset<= monitorend))) { >> +#ifdef FALLOC_FL_PUNCH_HOLE >> + op_name = (mode& FALLOC_FL_PUNCH_HOLE) ? >> + "punch hole" : "falloc"; >> +#else >> + op_name = "falloc"; >> +#endif >> + prt("%lu %s\tfrom 0x%x to 0x%x, (0x%x bytes)\n", testcalls, >> + op_name, offset, offset+length, length); >> + } >> + if (fallocate(fd, mode, (loff_t)offset, (loff_t)length) == -1) { >> +#ifdef FALLOC_FL_PUNCH_HOLE >> + op_name = (mode& FALLOC_FL_PUNCH_HOLE) ? >> + "punch hole" : "fallocate"; >> +#else >> + op_name = "fallocate"; >> +#endif >> + >> + prt("%s: %x to %x\n", op_name, offset, length); >> prterr("dofallocate: fallocate"); >> report_failure(161); >> } >> + >> + if (punch_hole) { >> + max_offset = offset< file_size ? offset : file_size; >> + max_len = max_offset + length<= file_size ? length : >> + file_size - max_offset; >> + memset(good_buf + max_offset, '\0', max_len); >> + } >> } >> #else >> void > > > Cheers, Andreas > > > > >