- Fix racy locking of mapfile (bz600900)

2010-07-20 22:01:40 +00:00 · 2010-07-20 22:01:40 +00:00 · 7b2267d85b
commit 7b2267d85b
parent e0d3e8aae9
16 changed files with 95 additions and 1366 deletions
--- a/.cvsignore
+++ b/.cvsignore
@ -1,3 +1,4 @@
 mdadm-3.0.3.tar.bz2
 mdadm-3.1.1-gcd9a8b5.tar.bz2
 mdadm-3.1.2.tar.bz2
 mdadm-3.1.3-git07202010.tar.bz2
--- a/mdadm-3.1.1-endian.patch
+++ b/mdadm-3.1.1-endian.patch
@ -1,30 +0,0 @@
 From 86983cce34645efb2b270a737d573e01eeee96e5 Mon Sep 17 00:00:00 2001
 From: NeilBrown <neilb@suse.de>
 Date: Wed, 24 Mar 2010 09:07:02 +1100
 Subject: [PATCH 1/6]
 ---
 util.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)
 diff --git a/util.c b/util.c
 index d292a66..79d2b0f 100644
 --- a/util.c
 +++ b/util.c
@@ -425,7 +425,12 @@ char *__fname_from_uuid(int id[4], int swap, char *buf, char sep)
 char *fname_from_uuid(struct supertype *st, struct mdinfo *info, char *buf, char sep)
 {
 -	return __fname_from_uuid(info->uuid, st->ss->swapuuid, buf, sep);
 +	// dirty hack to work around an issue with super1 superblocks...
 +	// super1 superblocks need swapuuid set in order for assembly to
 +	// work, but can't have it set if we want this printout to match
 +	// all the other uuid printouts in super1.c, so we force swapuuid
 +	// to 1 to make our printout match the rest of super1
 +	return __fname_from_uuid(info->uuid, (st->ss == &super1) ? 1 : st->ss->swapuuid, buf, sep);
 }
 #ifndef MDASSEMBLE
 -- 
 1.6.6.1
--- a/mdadm-3.1.2-container-2.patch
+++ b/mdadm-3.1.2-container-2.patch
@ -1,28 +0,0 @@
 commit e750ad9d6fbd4e6606681a9c81b9a99994255940
 Author: Doug Ledford <dledford@redhat.com>
 Date:   Thu Apr 8 17:34:47 2010 -0400
    Don't even try to activate non-redundant array types unless all disks
    are present.  This keeps arrays from getting marked as broken in
    the superblock and then permanently blocked from being assembled.
    Signed-off-by: Doug Ledford <dledford@redhat.com>
 diff --git a/Assemble.c b/Assemble.c
 index d059155..81178da 100644
 --- a/Assemble.c
 +++ b/Assemble.c
@@ -1337,8 +1337,11 @@ int assemble_container_content(struct supertype *st, int mdfd,
 		case LEVEL_LINEAR:
 		case LEVEL_MULTIPATH:
 		case 0:
 -			err = sysfs_set_str(content, NULL, "array_state",
 -					    "active");
 +			if ((working + preexist) == content->array.raid_disks)
 +				err = sysfs_set_str(content, NULL,
 +						    "array_state", "active");
 +			else
 +				err = 1;
 			break;
 		default:
 			err = sysfs_set_str(content, NULL, "array_state",
--- a/mdadm-3.1.2-container.patch
+++ b/mdadm-3.1.2-container.patch
@ -1,113 +0,0 @@
 commit b775cd39c498b0db4ca9f94bcc8615f59e534c7d
 Author: Doug Ledford <dledford@redhat.com>
 Date:   Thu Apr 8 16:56:02 2010 -0400
    Make Incremental container assembly behave like native array assembly
    Signed-off-by: Doug Ledford <dledford@redhat.com>
 diff --git a/Assemble.c b/Assemble.c
 index 1504f1f..d059155 100644
 --- a/Assemble.c
 +++ b/Assemble.c
@@ -1327,8 +1327,10 @@ int assemble_container_content(struct supertype *st, int mdfd,
 		   content->text_version,
 		   content->uuid, chosen_name);
 -	if (runstop > 0 ||
 -		 (working + preexist) >= content->array.working_disks) {
 +	if ((runstop > 0 &&
 +	     (working + preexist) >= content->array.working_disks) ||
 +	    (runstop == 0 &&
 +	     (working + preexist) == content->array.raid_disks)) {
 		int err;
 		switch(content->array.level) {
 diff --git a/Incremental.c b/Incremental.c
 index d32a8e5..9d77d4d 100644
 --- a/Incremental.c
 +++ b/Incremental.c
@@ -424,20 +424,21 @@ int Incremental(char *devname, int verbose, int runstop,
 		sysfs_uevent(&info, "change");
 		if (verbose >= 0)
 			fprintf(stderr, Name
 -				": container %s now has %d devices\n",
 -				chosen_name, info.array.working_disks);
 +				": container %s now has %d out of %d devices\n",
 +				chosen_name, info.array.working_disks,
 +				info.array.raid_disks);
 		wait_for(chosen_name, mdfd);
 		close(mdfd);
 -		if (runstop < 0)
 -			return 0; /* don't try to assemble */
 -		rv = Incremental(chosen_name, verbose, runstop,
 -				 NULL, homehost, require_homehost, autof);
 -		if (rv == 1)
 -			/* Don't fail the whole -I if a subarray didn't
 -			 * have enough devices to start yet
 +		if (runstop > 0 ||
 +		    info.array.working_disks == info.array.raid_disks)
 +			/* The return value of our container assembly doesn't
 +			 * depend on whether or not subarrays assembled
 +			 * properly, so no need to preserve the return value
 +			 * here.
 			 */
 -			rv = 0;
 -		return rv;
 +			Incremental(chosen_name, verbose, runstop, NULL,
 +				    homehost, require_homehost, autof);
 +		return 0;
 	}
 	avail = NULL;
 	active_disks = count_active(st, mdfd, &avail, &info);
@@ -666,6 +667,17 @@ int IncrementalScan(int verbose)
 		if (mdfd < 0)
 			continue;
 +		if (strcmp("imsm",me->metadata) == 0) {
 +			/*
 +			 * Just do a blind incremental assembly on the
 +			 * container.  If there's anything to be started,
 +			 * we will, if it's already started, we'll silently
 +			 * exit, if there's a problem, incremental will
 +			 * catch it.
 +			 */
 +			Incremental(me->path, verbose, 1, NULL, "<any>", 0, 1);
 +			continue;
 +		}
 		if (ioctl(mdfd, GET_ARRAY_INFO, &array) == 0 ||
 		    errno != ENODEV) {
 			close(mdfd);
 diff --git a/super-intel.c b/super-intel.c
 index 999b970..7bcfcdb 100644
 --- a/super-intel.c
 +++ b/super-intel.c
@@ -1536,12 +1536,27 @@ static void getinfo_super_imsm(struct supertype *st, struct mdinfo *info)
 	if (super->current_vol >= 0) {
 		getinfo_super_imsm_volume(st, info);
 		return;
 +	} else {
 +		struct mdinfo vol_info;
 +
 +		super->current_vol = 0;
 +		getinfo_super_imsm_volume(st, &vol_info);
 +		super->current_vol = -1;
 +		info->array.raid_disks = vol_info.array.raid_disks;
 	}
 	/* Set raid_disks to zero so that Assemble will always pull in valid
 	 * spares
 -	 */
 +	 *
 +	 * Note: my testing with assemble shows that it still works even
 +	 * when we set the raid_disks to a proper setting.  However, without
 +	 * this it is impossible to tell when a container has the right
 +	 * number of disks to start cleanly without violating layering
 +	 * boundaries.  So, not violating layering boundaries trumps spare
 +	 * disk issues.  Doug Ledford
 +	 *
 	info->array.raid_disks    = 0;
 +	 */
 	info->array.level         = LEVEL_CONTAINER;
 	info->array.layout        = 0;
 	info->array.md_minor      = -1;
--- a/mdadm-3.1.2-decremental-2.patch
+++ b/mdadm-3.1.2-decremental-2.patch
@ -1,36 +0,0 @@
 commit ddda49f6daf481471334f5675fbc1fa149d83ad6
 Author: Doug Ledford <dledford@redhat.com>
 Date:   Tue Apr 6 12:17:25 2010 -0400
    Minor bug fix to incremental remove support
    Signed-off-by: Doug Ledford <dledford@redhat.com>
 diff --git a/Manage.c b/Manage.c
 index 6539eda..b15586b 100644
 --- a/Manage.c
 +++ b/Manage.c
@@ -386,6 +386,7 @@ int Manage_subdevs(char *devname, int fd,
 		next = dv->next;
 		jnext = 0;
 +		tfd = -1;
 		if (strcmp(dv->devname, "failed")==0 ||
 		    strcmp(dv->devname, "faulty")==0) {
@@ -406,6 +407,7 @@ int Manage_subdevs(char *devname, int fd,
 				stb.st_rdev = makedev(disc.major, disc.minor);
 				next = dv;
 				jnext = j+1;
 +				tfd = 0;
 				sprintf(dvname,"%d:%d", disc.major, disc.minor);
 				dnprintable = dvname;
 				break;
@@ -440,6 +442,7 @@ int Manage_subdevs(char *devname, int fd,
 				stb.st_rdev = makedev(disc.major, disc.minor);
 				next = dv;
 				jnext = j+1;
 +				tfd = 0;
 				dnprintable = dvname;
 				break;
 			}
--- a/mdadm-3.1.2-decremental-3.patch
+++ b/mdadm-3.1.2-decremental-3.patch
@ -1,35 +0,0 @@
 commit 1934c67fe5bc25a79393ad78e29bf9ef778bc701
 Author: Doug Ledford <dledford@redhat.com>
 Date:   Tue Apr 6 23:02:47 2010 -0400
    Only close lfd if we have an external metadata type since that's the only
    time we'll have it open.
    If we weren't able to open the device file, assume the device is gone
    already and skip the sysfs_unique_holder test as it is guaranteed to
    fail.
    Signed-off-by: Doug Ledford <dledford@redhat.com>
 diff --git a/Manage.c b/Manage.c
 index b15586b..a690cfc 100644
 --- a/Manage.c
 +++ b/Manage.c
@@ -811,6 +811,7 @@ int Manage_subdevs(char *devname, int fd,
 				 * rely on the 'detached' checks
 				 */
 				if (strcmp(dv->devname, "detached") == 0 ||
 +				    tfd < 0 ||
 				    sysfs_unique_holder(dnum, stb.st_rdev))
 					/* pass */;
 				else {
@@ -878,8 +879,8 @@ int Manage_subdevs(char *devname, int fd,
 				ping_manager(name);
 				free(name);
 +				close(lfd);
 			}
 -			close(lfd);
 			if (verbose >= 0)
 				fprintf(stderr, Name ": hot removed %s\n",
 					dnprintable);
--- a/mdadm-3.1.2-decremental.patch
+++ b/mdadm-3.1.2-decremental.patch
@ -1,633 +0,0 @@
 From 8c43c776715301ff020639801a8b1b4716fdf745 Mon Sep 17 00:00:00 2001
 From: Doug Ledford <dledford@redhat.com>
 Date: Mon, 5 Apr 2010 12:32:08 -0400
 Subject: [PATCH 6/6] Initial implementation of incremental remove support
 Signed-off-by: Doug Ledford <dledford@redhat.com>
 ---
 Incremental.c      |   36 ++++++++++++++++++
 Manage.c           |  105 +++++++++++++++++++++++++++++++++++++--------------
 ReadMe.c           |   21 +++++++---
 mdadm.8            |   23 ++++++++++-
 mdadm.c            |   13 ++++++
 mdadm.h            |    8 ++++
 mdstat.c           |   90 +++++++++++++++++++++++++++++++++++++++++++-
 sysfs.c            |   13 ++++--
 udev-md-raid.rules |   16 +++++---
 9 files changed, 274 insertions(+), 51 deletions(-)
 diff --git a/Incremental.c b/Incremental.c
 index 7ad648a..d32a8e5 100644
 --- a/Incremental.c
 +++ b/Incremental.c
@@ -843,3 +843,39 @@ int Incremental_container(struct supertype *st, char *devname, int verbose,
 	map_unlock(&map);
 	return 0;
 }
 +
 +/*
 + * IncrementalRemove - Attempt to see if the passed in device belongs to any
 + * raid arrays, and if so first fail (if needed) and then remove the device.
 + *
 + * @devname - The device we want to remove
 + *
 + * Special note: We would like to just use Managedevs to fail/remove the
 + * device, but unfortunately, by the time we are called via udev, the device
 + * special file is already gone, and so we can't stat the device and se we
 + * don't have the right rdev value to use in the ioctls.  So, we use the
 + * sysfs method of device removal instead, but since that's not gauranteed
 + * to work depending on the version of kernel we run on, try to use the
 + * ioctl method first and only fallback if we don't have a valid device
 + * special file.  That way we can support operation manually on older kernels
 + * even if we won't be able to do this automatically via udev on older
 + * kernels.
 + */
 +int IncrementalRemove(char *devname, int verbose)
 +{
 +	char mddev[100] = "/dev/";
 +	int mdfd;
 +	struct mddev_dev_s devlist;
 +
 +	strncpy(mddev + 5, devname, sizeof(mddev) - 5);
 +	if (mdstat_check_active(mddev + 5))
 +		return 1;
 +	if ((mdfd = open_mddev(mddev, 0)) < 0)
 +		return 1;
 +	memset(&devlist, 0, sizeof(devlist));
 +	devlist.devname = devname;
 +	devlist.disposition = 'f';
 +	Manage_subdevs(mddev, mdfd, &devlist, verbose);
 +	devlist.disposition = 'r';
 +	return Manage_subdevs(mddev, mdfd, &devlist, verbose);
 +}
 diff --git a/Manage.c b/Manage.c
 index f848d8b..6539eda 100644
 --- a/Manage.c
 +++ b/Manage.c
@@ -346,6 +346,9 @@ int Manage_subdevs(char *devname, int fd,
 	mdu_disk_info_t disc;
 	unsigned long long array_size;
 	mddev_dev_t dv, next = NULL;
 +	struct mdinfo *mdi = NULL;
 +	struct mdinfo *dev = NULL;
 +	char sys_name[20] = "dev-";
 	struct stat stb;
 	int j, jnext = 0;
 	int tfd;
@@ -443,16 +446,43 @@ int Manage_subdevs(char *devname, int fd,
 			if (jnext == 0)
 				continue;
 		} else {
 +			/*
 +			 * For fail/remove operations, allow the disk
 +			 * to be completely missing, use name matching
 +			 * to a device in our sysfs entries to
 +			 * suffice.  For add we need a valid block device.
 +			 * Leave this loop one of three ways:
 +			 * 1) tfd < 0 and dev is set to our device
 +			 * 2) tfd >= 0 and dev is NULL
 +			 * 3) failed to find suitable device and return
 +			 */
 			j = 0;
 			tfd = dev_open(dv->devname, O_RDONLY);
 -			if (tfd < 0 && dv->disposition == 'r' &&
 -			    lstat(dv->devname, &stb) == 0)
 -				/* Be happy, the lstat worked, that is
 -				 * enough for --remove
 -				 */
 -				;
 -			else {
 +			if (tfd < 0 && dv->disposition != 'a') {
 +				strcpy(&sys_name[4],
 +				       strrchr(dv->devname, '/') + 1);
 +				mdi = sysfs_read(fd, 0,
 +						 GET_DEVS | KEEP_GONE_DEVS);
 +				if (!mdi) {
 +					fprintf(stderr, Name ": can't open %s "
 +						"and can't read sysfs info\n",
 +						dv->devname);
 +					return 1;
 +				}
 +				for (dev = mdi->devs; dev; dev = dev->next) {
 +					if (strcmp(sys_name, dev->sys_name))
 +						continue;
 +					break;
 +				}
 +				if (!dev) {
 +					fprintf(stderr, Name ": can't open %s "
 +						"and %s not listed in sysfs\n",
 +						dv->devname, sys_name);
 +					sysfs_free(mdi);
 +					return 1;
 +				}
 +			} else {
 				if (tfd < 0 || fstat(tfd, &stb) != 0) {
 					fprintf(stderr, Name ": cannot find %s: %s\n",
 						dv->devname, strerror(errno));
@@ -461,12 +491,12 @@ int Manage_subdevs(char *devname, int fd,
 					return 1;
 				}
 				close(tfd);
 -			}
 -			if ((stb.st_mode & S_IFMT) != S_IFBLK) {
 -				fprintf(stderr, Name ": %s is not a "
 -					"block device.\n",
 -					dv->devname);
 -				return 1;
 +				if ((stb.st_mode & S_IFMT) != S_IFBLK) {
 +					fprintf(stderr, Name ": %s is not a "
 +						"block device.\n",
 +						dv->devname);
 +					return 1;
 +				}
 			}
 		}
 		switch(dv->disposition){
@@ -790,26 +820,36 @@ int Manage_subdevs(char *devname, int fd,
 					return 1;
 				}
 			}
 -			/* FIXME check that it is a current member */
 -			err = ioctl(fd, HOT_REMOVE_DISK, (unsigned long)stb.st_rdev);
 -			if (err && errno == ENODEV) {
 +			/* stb.st_rdev is only valid if we have a tfd that
 +			 * does not indicate an error on attempt to open
 +			 * the devname
 +			 */
 +			if (tfd >= 0)
 +				err = ioctl(fd, HOT_REMOVE_DISK,
 +					    (unsigned long)stb.st_rdev);
 +			if (tfd < 0 || (err && errno == ENODEV)) {
 				/* Old kernels rejected this if no personality
 				 * registered */
 -				struct mdinfo *sra = sysfs_read(fd, 0, GET_DEVS);
 -				struct mdinfo *dv = NULL;
 -				if (sra)
 -					dv = sra->devs;
 -				for ( ; dv ; dv=dv->next)
 -					if (dv->disk.major == major(stb.st_rdev) &&
 -					    dv->disk.minor == minor(stb.st_rdev))
 +				if (!mdi) {
 +					strcpy(&sys_name[4],
 +					       strrchr(dv->devname, '/') + 1);
 +					mdi = sysfs_read(fd, 0, GET_DEVS |
 +							 KEEP_GONE_DEVS);
 +					if (mdi)
 +						dev = mdi->devs;
 +					for ( ; dev ; dev=dev->next) {
 +						if (strcmp(sys_name, dev->sys_name))
 +							continue;
 						break;
 -				if (dv)
 -					err = sysfs_set_str(sra, dv,
 +					}
 +				}
 +				if (dev)
 +					err = sysfs_set_str(mdi, dev,
 							    "state", "remove");
 				else
 					err = -1;
 -				if (sra)
 -					sysfs_free(sra);
 +				if (mdi)
 +					sysfs_free(mdi);
 			}
 			if (err) {
 				fprintf(stderr, Name ": hot remove failed "
@@ -844,11 +884,18 @@ int Manage_subdevs(char *devname, int fd,
 		case 'f': /* set faulty */
 			/* FIXME check current member */
 -			if (ioctl(fd, SET_DISK_FAULTY, (unsigned long) stb.st_rdev)) {
 +			if ((tfd >= 0 && ioctl(fd, SET_DISK_FAULTY,
 +					       (unsigned long) stb.st_rdev)) ||
 +			    (tfd < 0 && sysfs_set_str(mdi, dev, "state",
 +						      "faulty"))) {
 				fprintf(stderr, Name ": set device faulty failed for %s:  %s\n",
 					dnprintable, strerror(errno));
 +				if (mdi)
 +					sysfs_free(mdi);
 				return 1;
 -			}
 +			} 
 +			if (mdi)
 +				sysfs_free(mdi);
 			if (verbose >= 0)
 				fprintf(stderr, Name ": set %s faulty in %s\n",
 					dnprintable, devname);
 diff --git a/ReadMe.c b/ReadMe.c
 index 9d5a211..fd216ec 100644
 --- a/ReadMe.c
 +++ b/ReadMe.c
@@ -86,11 +86,12 @@ char Version[] = Name " - v3.1.2 - 10th March 2010\n";
  *     At the time if writing, there is only minimal support.
  */
 -char short_options[]="-ABCDEFGIQhVXWZ:vqbc:i:l:p:m:n:x:u:c:d:z:U:N:sarfRSow1tye:";
 +char short_options[]=
 +	"-ABCDEFGIQhVXWZ:vqbc:i:l:p:m:n:x:u:c:d:z:U:N:sarfRSow1tye:";
 char short_bitmap_options[]=
 -                   "-ABCDEFGIQhVXWZ:vqb:c:i:l:p:m:n:x:u:c:d:z:U:N:sarfRSow1tye:";
 +	"-ABCDEFGIQhVXWZ:vqb:c:i:l:p:m:n:x:u:c:d:z:U:N:sarfRSow1tye:";
 char short_bitmap_auto_options[]=
 -                   "-ABCDEFGIQhVXWZ:vqb:c:i:l:p:m:n:x:u:c:d:z:U:N:sa:rfRSow1tye:";
 +	"-ABCDEFGIQhVXWZ:vqb:c:i:l:p:m:n:x:u:c:d:z:U:N:sa:rfRSow1tye:";
 struct option long_options[] = {
     {"manage",    0, 0, '@'},
@@ -213,7 +214,7 @@ char Help[] =
 "       mdadm --grow options device\n"
 "            resize/reshape an active array\n"
 "       mdadm --incremental device\n"
 -"            add a device to an array as appropriate\n"
 +"            add/remove a device to/from an array as appropriate\n"
 "       mdadm --monitor options...\n"
 "            Monitor one or more array for significant changes.\n"
 "       mdadm device options...\n"
@@ -535,20 +536,26 @@ char Help_grow[] =
 ;
 char Help_incr[] =
 -"Usage: mdadm --incremental [-Rqrs] device\n"
 +"Usage: mdadm --incremental [-Rqrsf] device\n"
 "\n"
 "This usage allows for incremental assembly of md arrays.  Devices can be\n"
 "added one at a time as they are discovered.  Once an array has all expected\n"
 "devices, it will be started.\n"
 "\n"
 -"Options that are valid with incremental assembly (-I --incremental) more are:\n"
 -"  --run       -R  : run arrays as soon as a minimal number of devices are\n"
 +"Optionally, the process can be reversed by using the fail option.\n"
 +"When fail mode is invoked, mdadm will see if the device belongs to an array\n"
 +"and then both fail (if needed) and remove the device from that array.\n"
 +"\n"
 +"Options that are valid with incremental assembly (-I --incremental) are:\n"
 +"  --run       -R  : Run arrays as soon as a minimal number of devices are\n"
 "                  : present rather than waiting for all expected.\n"
 "  --quiet     -q  : Don't print any information messages, just errors.\n"
 "  --rebuild   -r  : Rebuild the 'map' file that mdadm uses for tracking\n"
 "                  : partial arrays.\n"
 "  --scan      -s  : Use with -R to start any arrays that have the minimal\n"
 "                  : required number of devices, but are not yet started.\n"
 +"  --fail      -f  : First fail (if needed) and then remove device from\n"
 +"                  : any array that it is a member of.\n"
 ;
 char Help_config[] =
 diff --git a/mdadm.8 b/mdadm.8
 index 4edfc41..eaf9155 100644
 --- a/mdadm.8
 +++ b/mdadm.8
@@ -135,7 +135,11 @@ This provides a convenient interface to a
 .I hot-plug
 system.  As each device is detected,
 .I mdadm
 -has a chance to include it in some array as appropriate.
 +has a chance to include it in some array as appropriate.  Optionally,
 +with the
 +.I \-\-fail
 +flag is passed in then we will remove the device from any active array
 +instead of adding it.
 If a
 .B CONTAINER
@@ -189,7 +193,7 @@ Change the size or shape of an active array.
 .TP
 .BR \-I ", " \-\-incremental
 -Add a single device into an appropriate array, and possibly start the array.
 +Add/remove a single device to/from an appropriate array, and possibly start the array.
 .TP
 .B \-\-auto-detect
@@ -1235,6 +1239,12 @@ in
 .B mdadm.conf
 as requiring an external bitmap, that bitmap will be attached first.
 +.TP
 +.BR \-\-fail ", " \-f
 +This allows the hot-plug system to remove devices that have fully disappeared
 +from the kernel.  It will first fail and then remove the device from any
 +array it belongs to.
 +
 .SH For Monitor mode:
 .TP
 .BR \-m ", " \-\-mail
@@ -2141,6 +2151,10 @@ Usage:
 .I component-device
 .HP 12
 Usage:
 +.B mdadm \-\-incremental \-\-fail
 +.I component-device
 +.HP 12
 +Usage:
 .B mdadm \-\-incremental \-\-rebuild
 .HP 12
 Usage:
@@ -2153,6 +2167,11 @@ passed to
 .B "mdadm \-\-incremental"
 to be conditionally added to an appropriate array.
 +Conversely, it can also be used with the
 +.B \-\-fail
 +flag to do just the opposite and find whatever array a particular device
 +is part of and remove the device from the array.
 +
 If the device passed is a
 .B CONTAINER
 device created by a previous call to
 diff --git a/mdadm.c b/mdadm.c
 index d5e34c0..cd6fd8f 100644
 --- a/mdadm.c
 +++ b/mdadm.c
@@ -124,6 +124,7 @@ int main(int argc, char *argv[])
 	ident.name[0] = 0;
 	ident.container = NULL;
 	ident.member = NULL;
 +	ident.member_index = -1;
 	while ((option_index = -1) ,
 	       (opt=getopt_long(argc, argv,
@@ -774,6 +775,9 @@ int main(int argc, char *argv[])
 			devmode = 'r';
 			continue;
 		case O(MANAGE,'f'): /* set faulty */
 +		case O(INCREMENTAL,'f'): /* r for incremental is taken, use f
 +					  * even though we will both fail and
 +					  * remove the device */
 			devmode = 'f';
 			continue;
 		case O(INCREMENTAL,'R'):
@@ -1517,6 +1521,11 @@ int main(int argc, char *argv[])
 			 ": --incremental --scan meaningless without --run.\n");
 				break;
 			}
 +			if (devmode == 'f') {
 +				fprintf(stderr, Name
 +			 ": --incremental --scan --fail not supported.\n");
 +				break;
 +			}
 			rv = IncrementalScan(verbose);
 		}
 		if (!devlist) {
@@ -1533,6 +1542,10 @@ int main(int argc, char *argv[])
 			rv = 1;
 			break;
 		}
 +		if (devmode == 'f') {
 +			rv = IncrementalRemove(devlist->devname, verbose-quiet);
 +			break;
 +		}
 		rv = Incremental(devlist->devname, verbose-quiet, runstop,
 				 ss, homehost, require_homehost, autof);
 		break;
 diff --git a/mdadm.h b/mdadm.h
 index d8ab85f..c113d0f 100644
 --- a/mdadm.h
 +++ b/mdadm.h
@@ -315,6 +315,7 @@ typedef struct mddev_ident_s {
 				 * of some other entry.
 				 */
 	char	*member;	/* subarray within a container */
 +	int	member_index;	/* subarray index within a container */
 	struct mddev_ident_s *next;
 	union {
@@ -355,6 +356,10 @@ struct mdstat_ent {
 	int		raid_disks;
 	int		chunk_size;
 	char *		metadata_version;
 +	struct dev_member {
 +		char			*name;
 +		struct dev_member	*next;
 +	} 		*members;
 	struct mdstat_ent *next;
 };
@@ -363,6 +368,7 @@ extern void free_mdstat(struct mdstat_ent *ms);
 extern void mdstat_wait(int seconds);
 extern void mdstat_wait_fd(int fd, const sigset_t *sigmask);
 extern int mddev_busy(int devnum);
 +extern int mdstat_check_active(char *devname);
 struct map_ent {
 	struct map_ent *next;
@@ -404,6 +410,7 @@ enum sysfs_read_flags {
 	GET_STATE	= (1 << 13),
 	GET_ERROR	= (1 << 14),
 	SKIP_GONE_DEVS	= (1 << 15),
 +	KEEP_GONE_DEVS	= (1 << 16),
 };
 /* If fd >= 0, get the array it is open on,
@@ -817,6 +824,7 @@ extern int Incremental_container(struct supertype *st, char *devname,
 				 int trustworthy);
 extern void RebuildMap(void);
 extern int IncrementalScan(int verbose);
 +extern int IncrementalRemove(char *devname, int verbose);
 extern int CreateBitmap(char *filename, int force, char uuid[16],
 			unsigned long chunksize, unsigned long daemon_sleep,
 diff --git a/mdstat.c b/mdstat.c
 index 4a9f370..81d2212 100644
 --- a/mdstat.c
 +++ b/mdstat.c
@@ -83,6 +83,45 @@
 #include	<sys/select.h>
 #include	<ctype.h>
 +static void free_member_devnames(struct dev_member **m)
 +{
 +       struct dev_member *t;
 +       if (!*m)
 +               return;
 +       while(*m) {
 +               t = *m;
 +               *m = (*m)->next;
 +               if (t->name)
 +                       free(t->name);
 +               free(t);
 +       }
 +       *m = NULL;
 +}
 +
 +static struct dev_member *add_member_devname(struct dev_member **m, char *name)
 +{
 +       struct dev_member *new;
 +       char *t;
 +
 +       if (!m || !name)
 +               return NULL;
 +
 +       new = malloc(sizeof(*new));
 +       if (!new)
 +               return NULL;
 +       if ((t = strchr(name, '[')) == NULL)
 +       {
 +               /* not a device */
 +               free(new);
 +               return *m;
 +       }
 +       new->name = strndup(name, t - name);
 +       new->next = *m;
 +       *m = new;
 +
 +       return new;
 +}
 +
 void free_mdstat(struct mdstat_ent *ms)
 {
 	while (ms) {
@@ -91,6 +130,7 @@ void free_mdstat(struct mdstat_ent *ms)
 		if (ms->level) free(ms->level);
 		if (ms->pattern) free(ms->pattern);
 		if (ms->metadata_version) free(ms->metadata_version);
 +		if (ms->members) free_member_devnames(&ms->members);
 		t = ms;
 		ms = ms->next;
 		free(t);
@@ -159,6 +199,7 @@ struct mdstat_ent *mdstat_read(int hold, int start)
 		ent->raid_disks = 0;
 		ent->chunk_size = 0;
 		ent->devcnt = 0;
 +		ent->members = NULL;
 		ent->dev = strdup(line);
 		ent->devnum = devnum;
@@ -170,15 +211,23 @@ struct mdstat_ent *mdstat_read(int hold, int start)
 				ent->active = 1;
 			else if (strcmp(w, "inactive")==0)
 				ent->active = 0;
 -			else if (ent->active >=0 &&
 +			else if (ent->active > 0 &&
 				 ent->level == NULL &&
 				 w[0] != '(' /*readonly*/) {
 				ent->level = strdup(w);
 				in_devs = 1;
 			} else if (in_devs && strcmp(w, "blocks")==0)
 				in_devs = 0;
 -			else if (in_devs) {
 +			else if (in_devs || (ent->active == 0 && w[0] != '(' &&
 +					     w[l - 1] == ')')) {
 +				if (isdigit(w[0]))
 +					continue;
 +				in_devs = 1;
 				ent->devcnt++;
 +				if (!add_member_devname(&ent->members, w)) {
 +					free_mdstat(ent);
 +					break;
 +				}
 				if (strncmp(w, "md", 2)==0) {
 					/* This has an md device as a component.
 					 * If that device is already in the
@@ -310,3 +359,40 @@ int mddev_busy(int devnum)
 	free_mdstat(mdstat);
 	return me != NULL;
 }
 +
 +/*
 + * Finds name of the active array holding this device
 + * @param[in]  devname         name of member device
 + * @param[out] devname         name of array
 + *
 + * @return                     found (0), or
 + *                             not found, failure (1)
 + */
 +
 +int mdstat_check_active(char *devname)
 +{
 +       struct mdstat_ent *mdstat = mdstat_read(0, 0);
 +       struct mdstat_ent *ent;
 +       char *name;
 +
 +       if (!devname)
 +               return 1;
 +       name = strrchr(devname, '/');
 +       if (name++ == NULL)
 +               return 1;
 +
 +       for (ent = mdstat; ent; ent = ent->next) {
 +               struct dev_member *m;
 +               if (ent->active && (strstr(ent->metadata_version,"imsm") ||
 +				   strstr(ent->metadata_version,"ddf")))
 +		       /* only return container matches, not subarrays */
 +                       continue;
 +               for (m = ent->members; m; m = m->next) {
 +                       if (!strcmp(m->name, name)) {
 +                               strcpy(devname, ent->dev);
 +                               return 0;
 +                       }
 +               }
 +       }
 +       return 1;
 +}
 diff --git a/sysfs.c b/sysfs.c
 index ebf9d8a..65dd848 100644
 --- a/sysfs.c
 +++ b/sysfs.c
@@ -273,13 +273,16 @@ struct mdinfo *sysfs_read(int fd, int devnum, unsigned long options)
 		strcpy(dbase, "block/dev");
 		if (load_sys(fname, buf)) {
 -			free(dev);
 -			if (options & SKIP_GONE_DEVS)
 +			if (options & SKIP_GONE_DEVS) {
 +				free(dev);
 				continue;
 -			else
 +			} else if (options & KEEP_GONE_DEVS) {
 +				dev->disk.major = dev->disk.minor = -1;
 +			} else
 				goto abort;
 -		}
 -		sscanf(buf, "%d:%d", &dev->disk.major, &dev->disk.minor);
 +		} else
 +			sscanf(buf, "%d:%d", &dev->disk.major,
 +			       &dev->disk.minor);
 		/* special case check for block devices that can go 'offline' */
 		if (options & SKIP_GONE_DEVS) {
 diff --git a/udev-md-raid.rules b/udev-md-raid.rules
 index c9a4f0e..aff14fa 100644
 --- a/udev-md-raid.rules
 +++ b/udev-md-raid.rules
@@ -1,17 +1,16 @@
 # do not edit this file, it will be overwritten on update
 SUBSYSTEM!="block", GOTO="md_end"
 -ACTION!="add|change", GOTO="md_end"
 +ACTION!="add|change|remove", GOTO="md_end"
 +ACTION=="remove", GOTO="md_remove"
 ACTION=="change", GOTO="md_no_incr"
 -# import data from a raid member and activate it
 -#ENV{ID_FS_TYPE}=="linux_raid_member", IMPORT{program}="/sbin/mdadm --examine --export $tempnode", RUN+="/sbin/mdadm --incremental $env{DEVNAME}"
 -# import data from a raid set
 +# we are adding a raid member, activate it
 +ENV{ID_FS_TYPE}=="linux_raid_member", RUN+="/sbin/mdadm -I $env{DEVNAME}"
 LABEL="md_no_incr"
 KERNEL!="md*", GOTO="md_end"
 -# partitions have no md/{array_state,metadata_version}, but should not
 -# for that reason be ignored.
 +# partitions have no md/{array_state,metadata_version}
 ENV{DEVTYPE}=="partition", GOTO="md_ignore_state"
 # container devices have a metadata version of e.g. 'external:ddf' and
@@ -32,7 +31,12 @@ ENV{DEVTYPE}=="partition", ENV{MD_DEVNAME}=="*[0-9]", SYMLINK+="md/$env{MD_DEVNA
 IMPORT{program}="/sbin/blkid -o udev -p $tempnode"
 OPTIONS+="link_priority=100"
 +OPTIONS+="watch"
 ENV{ID_FS_USAGE}=="filesystem|other|crypto", ENV{ID_FS_UUID_ENC}=="?*", SYMLINK+="disk/by-uuid/$env{ID_FS_UUID_ENC}"
 ENV{ID_FS_USAGE}=="filesystem|other", ENV{ID_FS_LABEL_ENC}=="?*", SYMLINK+="disk/by-label/$env{ID_FS_LABEL_ENC}"
 +GOTO="md_end"
 +
 +LABEL="md_remove"
 +ENV{ID_FS_TYPE}=="linux_raid_member", RUN+="/sbin/mdadm -If $env{DEVNAME}"
 LABEL="md_end"
 -- 
 1.6.6.1
--- a/mdadm-3.1.2-directory.patch
+++ b/mdadm-3.1.2-directory.patch
@ -1,265 +0,0 @@
 From df3e08eaa7ee07528968bb7152dfaaeebd3c97cb Mon Sep 17 00:00:00 2001
 From: Doug Ledford <dledford@redhat.com>
 Date: Fri, 2 Apr 2010 10:53:47 -0400
 Subject: [PATCH 5/6] Fix all the confusion over directories once and for all.  We now have
 3 directory definitions: mdmon directory for its pid and sock files,
 mdmonitor directory which is for the mdadm monitor mode pid file and
 can only be passed in via command line, and the directory for the
 mdadm map file.  Only the mdadm map file still hunts multiple locations,
 and the number of locations has been reduced to /var/run and the
 compile time specified location.  I could see lowering that to just
 1 location, but I didn't do that with this patch.
 Signed-off-by: Doug Ledford <dledford@redhat.com>
 ---
 Makefile  |   18 +++++++++---------
 mapfile.c |   18 ++++++++----------
 mdadm.h   |   41 +++++++++++++++++++++--------------------
 mdmon.c   |   28 ++++++----------------------
 msg.c     |    2 +-
 util.c    |    4 +---
 6 files changed, 46 insertions(+), 65 deletions(-)
 diff --git a/Makefile b/Makefile
 index 2aafad0..30e52a7 100644
 --- a/Makefile
 +++ b/Makefile
@@ -58,16 +58,16 @@ CONFFILE = $(SYSCONFDIR)/mdadm.conf
 CONFFILE2 = $(SYSCONFDIR)/mdadm/mdadm.conf
 MAILCMD =/usr/sbin/sendmail -t
 CONFFILEFLAGS = -DCONFFILE=\"$(CONFFILE)\" -DCONFFILE2=\"$(CONFFILE2)\"
 -# ALT_RUN should be somewhere that persists across the pivotroot
 -# from early boot to late boot.
 +# Both MAP_DIR and MDMON_DIR should be somewhere that persists across the
 +# pivotroot from early boot to late boot.
 # If you don't have /lib/init/rw you might want to use /dev/.something
 -#  e.g. make ALT_RUN=/dev/.mdadm
 -ALT_RUN = /lib/init/rw
 -ALT_MAPFILE = map
 -VAR_RUN = /var/run
 -ALTFLAGS = -DALT_RUN=\"$(ALT_RUN)\" -DALT_MAPFILE=\"$(ALT_MAPFILE)\"
 -VARFLAGS = -DVAR_RUN=\"$(VAR_RUN)\"
 -CFLAGS = $(CWFLAGS) $(CXFLAGS) -DSendmail=\""$(MAILCMD)"\" $(CONFFILEFLAGS) $(ALTFLAGS) $(VARFLAGS)
 +#  e.g. make MAP_DIR=/dev/.mdadm
 +MAP_DIR = /lib/init/rw/mdadm
 +MAP_FILE = map
 +MDMON_DIR = /lib/init/rw/mdmon
 +DIRFLAGS = -DMAP_DIR=\"$(MAP_DIR)\" -DMAP_FILE=\"$(MAP_FILE)\"
 +DIRFLAGS += -DMDMON_DIR=\"$(MDMON_DIR)\"
 +CFLAGS = $(CWFLAGS) $(CXFLAGS) -DSendmail=\""$(MAILCMD)"\" $(CONFFILEFLAGS) $(DIRFLAGS)
 # If you want a static binary, you might uncomment these
 # LDFLAGS = -static
 diff --git a/mapfile.c b/mapfile.c
 index d47fde1..f4339db 100644
 --- a/mapfile.c
 +++ b/mapfile.c
@@ -52,28 +52,26 @@
 #include	<ctype.h>
 #define mapnames(base) { base, base ".new", base ".lock"}
 -char *mapname[3][3] = {
 -	mapnames(VAR_RUN "/map"),
 -	mapnames("/var/run/mdadm.map"),
 -	mapnames(ALT_RUN "/" ALT_MAPFILE)
 +char *mapname[2][3] = {
 +	mapnames(MAP_DIR "/" MAP_FILE),
 +	mapnames("/var/run/mdadm.map")
 };
 -char *mapdir[3] = { VAR_RUN, NULL, ALT_RUN };
 +char *mapdir[2] = { MAP_DIR, NULL };
 -int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT | O_TRUNC };
 +int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT|O_TRUNC };
 char *mapsmode[3] = { "r", "w", "w"};
 FILE *open_map(int modenum, int *choice)
 {
 	int i;
 -	for (i = 0 ; i < 3 ; i++) {
 +	for (i = 0 ; i < 2 ; i++) {
 		int fd;
 -		if ((mapmode[modenum] & O_CREAT) &&
 -		    mapdir[modenum])
 +		if ((mapmode[modenum] & O_CREAT) && mapdir[i])
 			/* Attempt to create directory, don't worry about
 			 * failure.
 			 */
 -			mkdir(mapdir[modenum], 0755);
 +			mkdir(mapdir[i], 0755);
 		fd = open(mapname[i][modenum], mapmode[modenum], 0600);
 		if (fd >= 0) {
 			*choice = i;
 diff --git a/mdadm.h b/mdadm.h
 index 0386129..d8ab85f 100644
 --- a/mdadm.h
 +++ b/mdadm.h
@@ -68,28 +68,30 @@ extern __off64_t lseek64 __P ((int __fd, __off64_t __offset, int __whence));
 #define DEFAULT_BITMAP_DELAY 5
 #define DEFAULT_MAX_WRITE_BEHIND 256
 -/* VAR_RUN is where pid and socket files used for communicating
 - * with mdmon normally live.  It should be /var/run, but if
 - * it is too hard to remount /var/run as read-only rather than
 - * unmounting it at shutdown time, then it should be
 - * redefined to some place that comfortably persists until
 - * final shutdown, possibly in /dev if that is a tmpfs.
 - * Note: VAR_RUN does not need to be writable at shutdown,
 - * only during boot when "mdmon --takeover" is run.
 - */
 -#ifndef VAR_RUN
 -#define VAR_RUN "/var/run/mdadm"
 -#endif /* VAR_RUN */
 -/* ALT_RUN should be somewhere that persists across the pivotroot
 +/* MAP_DIR should be somewhere that persists across the pivotroot
  * from early boot to late boot.
  * If you don't have /lib/init/rw you might want to use /dev/.something
  */
 -#ifndef ALT_RUN
 -#define ALT_RUN "/lib/init/rw/mdadm"
 -#endif /* ALT_RUN */
 -#ifndef ALT_MAPFILE
 -#define ALT_MAPFILE "map"
 -#endif /* ALT_MAPFILE */
 +#ifndef MAP_DIR
 +#define MAP_DIR "/lib/init/rw/mdadm"
 +#endif /* MAP_DIR */
 +/* MAP_FILE is what we name the map file we put in MAP_DIR, in case you
 + * want something other than the default of "map"
 + */
 +#ifndef MAP_FILE
 +#define MAP_FILE "map"
 +#endif /* MAP_FILE */
 +/* MDMON_DIR is where pid and socket files used for communicating
 + * with mdmon normally live.  It *should* be /var/run, but when
 + * mdmon is needed at early boot then it needs to write there prior
 + * to /var/run being mounted read/write, and it also then needs to
 + * persist beyond when /var/run is mounter read-only.  So, to be
 + * safe, the default is somewhere that is read/write early in the
 + * boot process and stays up as long as possible during shutdown.
 + */
 +#ifndef MDMON_DIR
 +#define MDMON_DIR "/lib/init/rw/mdmon"
 +#endif /* MDMON_DIR */
 #include	"md_u.h"
 #include	"md_p.h"
@@ -910,7 +912,6 @@ extern int create_mddev(char *dev, char *name, int autof, int trustworthy,
 extern int open_mddev(char *dev, int report_errors);
 extern int open_container(int fd);
 -extern char *pid_dir;
 extern int mdmon_running(int devnum);
 extern int mdmon_pid(int devnum);
 extern int check_env(char *name);
 diff --git a/mdmon.c b/mdmon.c
 index 69c320e..cbb72cc 100644
 --- a/mdmon.c
 +++ b/mdmon.c
@@ -120,10 +120,10 @@ static int make_pidfile(char *devname)
 	int fd;
 	int n;
 -	if (mkdir(pid_dir, 0700) < 0 &&
 +	if (mkdir(MDMON_DIR, 0755) < 0 &&
 	    errno != EEXIST)
 		return -errno;
 -	sprintf(path, "%s/%s.pid", pid_dir, devname);
 +	sprintf(path, "%s/%s.pid", MDMON_DIR, devname);
 	fd = open(path, O_RDWR|O_CREAT|O_EXCL, 0600);
 	if (fd < 0)
@@ -189,13 +189,10 @@ void remove_pidfile(char *devname)
 {
 	char buf[100];
 -	sprintf(buf, "%s/%s.pid", pid_dir, devname);
 +	sprintf(buf, "%s/%s.pid", MDMON_DIR, devname);
 	unlink(buf);
 -	sprintf(buf, "%s/%s.sock", pid_dir, devname);
 +	sprintf(buf, "%s/%s.sock", MDMON_DIR, devname);
 	unlink(buf);
 -	if (strcmp(pid_dir, ALT_RUN) == 0)
 -		/* try to clean up when we are finished with this dir */
 -		rmdir(pid_dir);
 }
 static int make_control_sock(char *devname)
@@ -208,7 +205,7 @@ static int make_control_sock(char *devname)
 	if (sigterm)
 		return -1;
 -	sprintf(path, "%s/%s.sock", pid_dir, devname);
 +	sprintf(path, "%s/%s.sock", MDMON_DIR, devname);
 	unlink(path);
 	sfd = socket(PF_LOCAL, SOCK_STREAM, 0);
 	if (sfd < 0)
@@ -445,12 +442,7 @@ static int mdmon(char *devname, int devnum, int must_fork, int takeover)
 	act.sa_handler = SIG_IGN;
 	sigaction(SIGPIPE, &act, NULL);
 -	pid_dir = VAR_RUN;
 	victim = mdmon_pid(container->devnum);
 -	if (victim < 0) {
 -		pid_dir = ALT_RUN;
 -		victim = mdmon_pid(container->devnum);
 -	}
 	if (victim >= 0)
 		victim_sock = connect_monitor(container->devname);
@@ -474,16 +466,8 @@ static int mdmon(char *devname, int devnum, int must_fork, int takeover)
 	 */
 	if (victim > 0)
 		remove_pidfile(devname);
 -	pid_dir = VAR_RUN;
 	if (make_pidfile(devname) < 0) {
 -		/* Try the alternate */
 -		pid_dir = ALT_RUN;
 -		if (make_pidfile(devname) < 0) {
 -			fprintf(stderr, "mdmon: Neither %s nor %s are writable\n"
 -				"       cannot create .pid or .sock files.  Aborting\n",
 -				VAR_RUN, ALT_RUN);
 -			exit(3);
 -		}
 +		exit(3);
 	}
 	container->sock = make_control_sock(devname);
 diff --git a/msg.c b/msg.c
 index d2d8445..aabfa8f 100644
 --- a/msg.c
 +++ b/msg.c
@@ -147,7 +147,7 @@ int connect_monitor(char *devname)
 	int pos;
 	char *c;
 -	pos = sprintf(path, "%s/", pid_dir);
 +	pos = sprintf(path, "%s/", MDMON_DIR);
 	if (is_subarray(devname)) {
 		devname++;
 		c = strchr(devname, '/');
 diff --git a/util.c b/util.c
 index 79d2b0f..bdb19f9 100644
 --- a/util.c
 +++ b/util.c
@@ -1499,15 +1499,13 @@ int fd2devnum(int fd)
 	return NoMdDev;
 }
 -char *pid_dir = VAR_RUN;
 -
 int mdmon_pid(int devnum)
 {
 	char path[100];
 	char pid[10];
 	int fd;
 	int n;
 -	sprintf(path, "%s/%s.pid", pid_dir, devnum2devname(devnum));
 +	sprintf(path, "%s/%s.pid", MDMON_DIR, devnum2devname(devnum));
 	fd = open(path, O_RDONLY | O_NOATIME, 0);
 	if (fd < 0)
 -- 
 1.6.6.1
--- a/mdadm-3.1.2-lock.patch
+++ b/mdadm-3.1.2-lock.patch
@ -0,0 +1,65 @@
 --- mdadm-3.1.2/Incremental.c.lock	2010-07-20 17:21:41.879217630 -0400
 +++ mdadm-3.1.2/Incremental.c	2010-07-20 17:21:41.916342431 -0400
@@ -285,7 +285,9 @@ int Incremental(char *devname, int verbo
 	/* 4/ Check if array exists.
 	 */
 -	map_lock(&map);
 +	if (map_lock(&map))
 +		fprintf(stderr, Name ": failed to get exclusive lock on "
 +			"mapfile\n");
 	mp = map_by_uuid(&map, info.uuid);
 	if (mp)
 		mdfd = open_dev(mp->devnum);
@@ -769,7 +771,9 @@ int Incremental_container(struct superty
 	struct mdinfo *ra;
 	struct map_ent *map = NULL;
 -	map_lock(&map);
 +	if (map_lock(&map))
 +		fprintf(stderr, Name ": failed to get exclusive lock on "
 +			"mapfile\n");
 	for (ra = list ; ra ; ra = ra->next) {
 		int mdfd;
 --- mdadm-3.1.2/mapfile.c.lock	2010-07-20 17:21:41.687217302 -0400
 +++ mdadm-3.1.2/mapfile.c	2010-07-20 17:23:23.038096346 -0400
@@ -58,7 +58,7 @@ char *mapname[2][3] = {
 };
 char *mapdir[2] = { MAP_DIR, NULL };
 -int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT|O_TRUNC };
 +int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT|O_EXCL };
 char *mapsmode[3] = { "r", "w", "w"};
 FILE *open_map(int modenum, int *choice)
@@ -120,14 +120,11 @@ static int lwhich = 0;
 int map_lock(struct map_ent **melp)
 {
 	if (lf == NULL) {
 -		lf = open_map(2, &lwhich);
 +		do {
 +			lf = open_map(2, &lwhich);
 +		} while (lf == NULL && errno == EEXIST);
 		if (lf == NULL)
 			return -1;
 -		if (flock(fileno(lf), LOCK_EX) != 0) {
 -			fclose(lf);
 -			lf = NULL;
 -			return -1;
 -		}
 	}
 	if (*melp)
 		map_free(*melp);
@@ -137,10 +134,7 @@ int map_lock(struct map_ent **melp)
 void map_unlock(struct map_ent **melp)
 {
 -	if (lf) {
 -		flock(fileno(lf), LOCK_UN);
 -		fclose(lf);
 -	}
 +	fclose(lf);
 	unlink(mapname[lwhich][2]);
 	lf = NULL;
 }
--- a/mdadm-3.1.2-mapfile.patch
+++ b/mdadm-3.1.2-mapfile.patch
@ -1,80 +0,0 @@
 From 435b90e7d41a17f35a984112cace471c975906f4 Mon Sep 17 00:00:00 2001
 From: Doug Ledford <dledford@redhat.com>
 Date: Tue, 16 Mar 2010 23:00:11 -0400
 Subject: [PATCH 2/6] Create directory to contain mapfile (Assuming parent exists and
 filesystem is writable).
 This particularly keeps udev happy if VAR_RUN is set to /dev/md.
 Signed-off-by: Doug Ledford <dledford@redhat.com>
 Signed-off-by: NeilBrown <neilb@suse.de>
 ---
 mapfile.c |   17 +++++++++++++----
 mdmon.c   |    2 +-
 2 files changed, 14 insertions(+), 5 deletions(-)
 diff --git a/mapfile.c b/mapfile.c
 index 366ebe3..89187ac 100644
 --- a/mapfile.c
 +++ b/mapfile.c
@@ -29,7 +29,7 @@
  */
 /* /var/run/mdadm.map is used to track arrays being created in --incremental
 - * more.  It particularly allows lookup from UUID to array device, but
 + * mode.  It particularly allows lookup from UUID to array device, but
  * also allows the array device name to be easily found.
  *
  * The map file is line based with space separated fields.  The fields are:
@@ -42,8 +42,8 @@
  * However /var/run may not exist or be writable in early boot.  And if
  * no-one has created /var/run/mdadm, we still want to survive.
  * So possible locations are:
 - *   /var/run/mdadm/map  /var/run/mdadm.map  /dev/.mdadm.map
 - * the last, because udev requires a writable /dev very early.
 + *   /var/run/mdadm/map  /var/run/mdadm.map  /lib/initrw/madam/map
 + * The last can easily be change at compile to e.g. somewhere in /dev.
  * We read from the first one that exists and write to the first
  * one that we can.
  */
@@ -57,6 +57,7 @@ char *mapname[3][3] = {
 	mapnames("/var/run/mdadm.map"),
 	mapnames(ALT_RUN "/map")
 };
 +char *mapdir[3] = { VAR_RUN, NULL, ALT_RUN };
 int mapmode[3] = { O_RDONLY, O_RDWR|O_CREAT, O_RDWR|O_CREAT | O_TRUNC };
 char *mapsmode[3] = { "r", "w", "w"};
@@ -64,8 +65,16 @@ char *mapsmode[3] = { "r", "w", "w"};
 FILE *open_map(int modenum, int *choice)
 {
 	int i;
 +
 	for (i = 0 ; i < 3 ; i++) {
 -		int fd = open(mapname[i][modenum], mapmode[modenum], 0600);
 +		int fd;
 +		if ((mapmode[modenum] & O_CREAT) &&
 +		    mapdir[modenum])
 +			/* Attempt to create directory, don't worry about
 +			 * failure.
 +			 */
 +			mkdir(mapdir[modenum], 0755);
 +		fd = open(mapname[i][modenum], mapmode[modenum], 0600);
 		if (fd >= 0) {
 			*choice = i;
 			return fdopen(fd, mapsmode[modenum]);
 diff --git a/mdmon.c b/mdmon.c
 index 961aa77..69c320e 100644
 --- a/mdmon.c
 +++ b/mdmon.c
@@ -120,7 +120,7 @@ static int make_pidfile(char *devname)
 	int fd;
 	int n;
 -	if (mkdir(pid_dir, 0600) < 0 &&
 +	if (mkdir(pid_dir, 0700) < 0 &&
 	    errno != EEXIST)
 		return -errno;
 	sprintf(path, "%s/%s.pid", pid_dir, devname);
 -- 
 1.6.6.1
--- a/mdadm-3.1.2-mapname.patch
+++ b/mdadm-3.1.2-mapname.patch
@ -1,58 +0,0 @@
 From e259df4e63f553c1271fa7d7612c110d2518e572 Mon Sep 17 00:00:00 2001
 From: Doug Ledford <dledford@redhat.com>
 Date: Wed, 17 Mar 2010 10:52:22 -0400
 Subject: [PATCH 3/6] mapfile: if we putting the mapfile in a custom location via ALT_RUN, allow
 a custom filename too.
 Signed-off-by: Doug Ledford <dledford@redhat.com>
 ---
 Makefile  |    3 ++-
 mapfile.c |    2 +-
 mdadm.h   |    3 +++
 3 files changed, 6 insertions(+), 2 deletions(-)
 diff --git a/Makefile b/Makefile
 index 1035ea8..2aafad0 100644
 --- a/Makefile
 +++ b/Makefile
@@ -63,8 +63,9 @@ CONFFILEFLAGS = -DCONFFILE=\"$(CONFFILE)\" -DCONFFILE2=\"$(CONFFILE2)\"
 # If you don't have /lib/init/rw you might want to use /dev/.something
 #  e.g. make ALT_RUN=/dev/.mdadm
 ALT_RUN = /lib/init/rw
 +ALT_MAPFILE = map
 VAR_RUN = /var/run
 -ALTFLAGS = -DALT_RUN=\"$(ALT_RUN)\"
 +ALTFLAGS = -DALT_RUN=\"$(ALT_RUN)\" -DALT_MAPFILE=\"$(ALT_MAPFILE)\"
 VARFLAGS = -DVAR_RUN=\"$(VAR_RUN)\"
 CFLAGS = $(CWFLAGS) $(CXFLAGS) -DSendmail=\""$(MAILCMD)"\" $(CONFFILEFLAGS) $(ALTFLAGS) $(VARFLAGS)
 diff --git a/mapfile.c b/mapfile.c
 index 89187ac..74f7256 100644
 --- a/mapfile.c
 +++ b/mapfile.c
@@ -55,7 +55,7 @@
 char *mapname[3][3] = {
 	mapnames(VAR_RUN "/map"),
 	mapnames("/var/run/mdadm.map"),
 -	mapnames(ALT_RUN "/map")
 +	mapnames(ALT_RUN "/" ALT_MAPFILE)
 };
 char *mapdir[3] = { VAR_RUN, NULL, ALT_RUN };
 diff --git a/mdadm.h b/mdadm.h
 index 362b66b..0386129 100644
 --- a/mdadm.h
 +++ b/mdadm.h
@@ -87,6 +87,9 @@ extern __off64_t lseek64 __P ((int __fd, __off64_t __offset, int __whence));
 #ifndef ALT_RUN
 #define ALT_RUN "/lib/init/rw/mdadm"
 #endif /* ALT_RUN */
 +#ifndef ALT_MAPFILE
 +#define ALT_MAPFILE "map"
 +#endif /* ALT_MAPFILE */
 #include	"md_u.h"
 #include	"md_p.h"
 -- 
 1.6.6.1
--- a/mdadm-3.1.2-powerpc-compile.patch
+++ b/mdadm-3.1.2-powerpc-compile.patch
@ -1,21 +0,0 @@
 commit 995511f26e8d661d32d1c3fc42a08960989d1e6d
 Author: Doug Ledford <dledford@redhat.com>
 Date:   Tue Apr 6 14:04:30 2010 -0400
    powerpc compile fix
    Signed-off-by: Doug Ledford <dledford@redhat.com>
 diff --git a/super-intel.c b/super-intel.c
 index a196ca3..999b970 100644
 --- a/super-intel.c
 +++ b/super-intel.c
@@ -697,7 +697,7 @@ static void print_imsm_dev(struct imsm_dev *dev, char *uuid, int disk_idx)
 		printf(" <-- %s", map_state_str[map->map_state]);
 		printf("\n     Checkpoint : %u (%llu)",
 		       __le32_to_cpu(dev->vol.curr_migr_unit),
 -		       blocks_per_migr_unit(dev));
 +		       (unsigned long long)blocks_per_migr_unit(dev));
 	}
 	printf("\n");
 	printf("    Dirty State : %s\n", dev->vol.dirty ? "dirty" : "clean");
--- a/mdadm-3.1.2-rebuild.patch
+++ b/mdadm-3.1.2-rebuild.patch
@ -1,38 +0,0 @@
 From 7bf59f5c16d928d3826fdf0c406d1aac5775e78b Mon Sep 17 00:00:00 2001
 From: Doug Ledford <dledford@redhat.com>
 Date: Wed, 17 Mar 2010 09:28:07 -0400
 Subject: [PATCH 4/6] Only signal a udev change event if we actually write a mapfile in RebuildMap
 Signed-off-by: Doug Ledford <dledford@redhat.com>
 ---
 mapfile.c |   14 ++++++++------
 1 files changed, 8 insertions(+), 6 deletions(-)
 diff --git a/mapfile.c b/mapfile.c
 index 74f7256..d47fde1 100644
 --- a/mapfile.c
 +++ b/mapfile.c
@@ -471,12 +471,14 @@ void RebuildMap(void)
 		}
 		sysfs_free(sra);
 	}
 -	map_write(map);
 +	/* Only trigger a change if we wrote a new map file */
 +	if (map_write(map))
 +		for (md = mdstat ; md ; md = md->next) {
 +			struct mdinfo *sra = sysfs_read(-1, md->devnum,
 +							GET_VERSION);
 +			sysfs_uevent(sra, "change");
 +			sysfs_free(sra);
 +		}
 	map_free(map);
 -	for (md = mdstat ; md ; md = md->next) {
 -		struct mdinfo *sra = sysfs_read(-1, md->devnum, GET_VERSION);
 -		sysfs_uevent(sra, "change");
 -		sysfs_free(sra);
 -	}
 	free_mdstat(mdstat);
 }
 -- 
 1.6.6.1
--- a/mdadm-3.1.3-read.patch
+++ b/mdadm-3.1.3-read.patch
@ -0,0 +1,11 @@
 --- mdadm-3.1.3-git07202010/Grow.c.read	2010-07-20 12:18:12.000000000 -0400
 +++ mdadm-3.1.3-git07202010/Grow.c	2010-07-20 12:55:26.816805170 -0400
@@ -1820,7 +1820,7 @@ int Grow_restart(struct supertype *st, s
 		}
 		/* There should be a duplicate backup superblock 4k before here */
 		if (lseek64(fd, -4096, 1) < 0 ||
 -		    read(fd, &bsb2, 4096) != 4096)
 +		    read(fd, &bsb2, sizeof(bsb2)) != sizeof(bsb2))
 			goto second_fail; /* Cannot find leading superblock */
 		if (bsb.magic[15] == '1')
 			bsbsize = offsetof(struct mdp_backup_super, pad1);
--- a/mdadm.spec
+++ b/mdadm.spec
@ -1,23 +1,14 @@
 Summary:     The mdadm program controls Linux md devices (software RAID arrays)
 Name:        mdadm
-Version:     3.1.2
+Version:     3.1.3
-Release:     10%{?dist}
+Release:     0.git07202010.2%{?dist}
-Source:      http://www.kernel.org/pub/linux/utils/raid/mdadm/mdadm-%{version}.tar.bz2
+Source:      http://www.kernel.org/pub/linux/utils/raid/mdadm/mdadm-%{version}-git07202010.tar.bz2
 Source1:     mdmonitor.init
 Source2:     raid-check
 Source3:     mdadm.rules
 Source4:     mdadm-raid-check-sysconfig
-Patch1:      mdadm-3.1.2-mapname.patch
+Patch0:      mdadm-3.1.3-read.patch
-Patch2:      mdadm-3.1.1-endian.patch
+PAtch1:      mdadm-3.1.2-lock.patch
 Patch3:      mdadm-3.1.2-mapfile.patch
 Patch4:      mdadm-3.1.2-rebuild.patch
 Patch5:      mdadm-3.1.2-directory.patch
 Patch6:      mdadm-3.1.2-decremental.patch
 Patch7:      mdadm-3.1.2-decremental-2.patch
 Patch8:      mdadm-3.1.2-decremental-3.patch
 Patch9:      mdadm-3.1.2-powerpc-compile.patch
 Patch10:     mdadm-3.1.2-container.patch
 Patch11:     mdadm-3.1.2-container-2.patch
 Patch20:     mdadm-2.5.2-static.patch
 URL:         http://www.kernel.org/pub/linux/utils/raid/mdadm/
 License:     GPLv2+
@ -38,18 +29,9 @@ almost all functions without a configuration file, though a configuration
 file can be used to help with some common tasks.
 %prep
-%setup -q
+%setup -q -n mdadm-3.1.3-git07202010
-%patch3 -p1 -b .mapfile
+%patch0 -p1 -b .read
-%patch1 -p1 -b .mapname
+%patch1 -p1 -b .lock
 %patch2 -p1 -b .endian
 %patch4 -p1 -b .rebuild
 %patch5 -p1 -b .directory
 %patch6 -p1 -b .decremental
 %patch7 -p1 -b .decremental-2
 %patch8 -p1 -b .decremental-3
 %patch9 -p1 -b .powerpc
 %patch10 -p1 -b .container
 %patch11 -p1 -b .container-2
 %patch20 -p1 -b .static
 %build
@ -57,14 +39,13 @@ make %{?_smp_mflags} CXFLAGS="$RPM_OPT_FLAGS -fno-strict-aliasing" SYSCONFDIR="%
 %install
 rm -rf %{buildroot}
-make DESTDIR=%{buildroot} MANDIR=%{_mandir} BINDIR=/sbin install install-static
+make DESTDIR=%{buildroot} MANDIR=%{_mandir} BINDIR=/sbin install
-rm -f %{buildroot}/lib/udev/rules.d/*
+rm -f %{buildroot}/lib/udev/rules.d/64*
 install -Dp -m 755 %{SOURCE1} %{buildroot}%{_initrddir}/mdmonitor
 install -Dp -m 755 %{SOURCE2} %{buildroot}%{_sysconfdir}/cron.weekly/99-raid-check
 install -Dp -m 644 %{SOURCE3} %{buildroot}/lib/udev/rules.d/65-md-incremental.rules
 install -Dp -m 644 %{SOURCE4} %{buildroot}%{_sysconfdir}/sysconfig/raid-check
 mkdir -p -m 700 %{buildroot}/var/run/mdadm
 ln -s mdadm.8.gz %{buildroot}%{_mandir}/man8/mdadm.static.8.gz
 %clean
 rm -rf %{buildroot}
@ -97,6 +78,13 @@ fi
 %attr(0700,root,root) %dir /var/run/mdadm
 %changelog
 * Tue Jul 20 2010 Doug Ledford <dledford@redhat.com> - 3.1.3-0.git07202010.2
 - Fix racy locking of mapfile (bz600900)
 * Tue Jul 20 2010 Doug Ledford <dledford@redhat.com> - 3.1.3-0.git07202010.1
 - Update to latest git repo (3.1.2 plus pending changes, fixes bz602457)
 - Remove mdadm.static as its no longer used in initrd creation
 * Tue Apr 13 2010 Doug Ledford <dledford@redhat.com> - 3.1.2-10
 - Minor update to mdadm.rules to make anaconda happy
--- a/1
+++ b/1
@ -1 +1,2 @@
 c5a39f38c465229767a5af2a4eb81bef  mdadm-3.1.2.tar.bz2
 3c1df7259e17ed909ddd4b28e434f10a  mdadm-3.1.3-git07202010.tar.bz2
`@ -1 +1,2 @@`
	`c5a39f38c465229767a5af2a4eb81bef mdadm-3.1.2.tar.bz2`	`c5a39f38c465229767a5af2a4eb81bef mdadm-3.1.2.tar.bz2`
		`3c1df7259e17ed909ddd4b28e434f10a mdadm-3.1.3-git07202010.tar.bz2`