Allow resize to backup to a file.

To support resizing an array without a spare, mdadm now understands
  --backup-file=
which should point to a file for storing a backup of critical data.
This can be given to --grow which will create the file, or
--assemble which will restore from the file if needed.

Signed-off-by: Neil Brown <neilb@suse.de>

6 files changed, 104 insertions, 42 deletions

diff --git a/Assemble.c b/Assemble.c
index a8a331e..f38b1a1 100644
--- a/Assemble.c
+++ b/Assemble.c
@@ -31,7 +31,7 @@
 
 int Assemble(struct supertype *st, char *mddev, int mdfd,
 	     mddev_ident_t ident, char *conffile,
-	     mddev_dev_t devlist,
+	     mddev_dev_t devlist, char *backup_file,
 	     int readonly, int runstop,
 	     char *update,
 	     int verbose, int force)
@@ -585,7 +585,7 @@ int Assemble(struct supertype *st, char *mddev, int mdfd,
 				fdlist[i] = -1;
 		}
 		if (!err)
-			err = Grow_restart(st, &info, fdlist, bestcnt);
+			err = Grow_restart(st, &info, fdlist, bestcnt, backup_file);
 		while (i>0) {
 			i--;
 			if (fdlist[i]>=0) close(fdlist[i]);
diff --git a/ChangeLog b/ChangeLog
index 22a97cc..4ee2dff 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -5,6 +5,7 @@ Changes Prior to 2.4 release
     -   Put a 'canary' at each end of the backup so a corruption
 	can be more easily detected.
     -   Remove useless 'ident' arguement from ->getinfo_super method.
+    -   Support --backup-file for backing-up critical section during growth.
 
 Changes Prior to 2.3.1 release
     -   Fixed -O2 compile so I could make and RPM.
diff --git a/Grow.c b/Grow.c
index 04de455..4a225c7 100644
--- a/Grow.c
+++ b/Grow.c
@@ -408,7 +408,7 @@ int bsb_csum(char *buf, int len)
 	return __cpu_to_le32(csum);
 }
 
-int Grow_reshape(char *devname, int fd, int quiet,
+int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
 		 long long size,
 		 int level, int layout, int chunksize, int raid_disks)
 {
@@ -625,8 +625,8 @@ int Grow_reshape(char *devname, int fd, int quiet,
 				devname);
 			return 1;
 		}
-		if (sra->spares == 0) {
-			fprintf(stderr, Name ": %s: Cannot grow - need a spare to backup critical section\n",
+		if (sra->spares == 0 && backup_file == NULL) {
+			fprintf(stderr, Name ": %s: Cannot grow - need a spare or backup-file to backup critical section\n",
 				devname);
 			return 1;
 		}
@@ -634,13 +634,13 @@ int Grow_reshape(char *devname, int fd, int quiet,
 		nrdisks = array.nr_disks + sra->spares;
 		/* Now we need to open all these devices so we can read/write.
 		 */
-		fdlist = malloc(nrdisks * sizeof(int));
-		offsets = malloc(nrdisks * sizeof(offsets[0]));
+		fdlist = malloc((1+nrdisks) * sizeof(int));
+		offsets = malloc((1+nrdisks) * sizeof(offsets[0]));
 		if (!fdlist || !offsets) {
 			fprintf(stderr, Name ": malloc failed: grow aborted\n");
 			return 1;
 		}
-		for (d=0; d< nrdisks; d++)
+		for (d=0; d <= nrdisks; d++)
 			fdlist[d] = -1;
 		d = array.raid_disks;
 		for (sd = sra->devs; sd; sd=sd->next) {
@@ -674,8 +674,20 @@ int Grow_reshape(char *devname, int fd, int quiet,
 					" --grow aborted\n", devname, i);
 				goto abort;
 			}
+		spares = sra->spares;
+		if (backup_file) {
+			fdlist[d] = open(backup_file, O_RDWR|O_CREAT|O_EXCL, 0600);
+			if (fdlist[d] < 0) {
+				fprintf(stderr, Name ": %s: cannot create backup file %s: %s\n",
+					devname, backup_file, strerror(errno));
+				goto abort;
+			}
+			offsets[d] = 8;
+			d++;
+			spares++;
+		}
 		if (fdlist[array.raid_disks] < 0) {
-			fprintf(stderr, Name ": %s: failed to find a spare - --grow aborted\n",
+			fprintf(stderr, Name ": %s: failed to find a spare and no backup-file given - --grow aborted\n",
 				devname);
 			goto abort;
 		}
@@ -687,8 +699,6 @@ int Grow_reshape(char *devname, int fd, int quiet,
 			goto abort;
 		}
 
-		spares = sra->spares;
-
 
 		memcpy(bsb.magic, "md_backup_data-1", 16);
 		st->ss->uuid_from_super((int*)&bsb.set_uuid, super);
@@ -701,7 +711,11 @@ int Grow_reshape(char *devname, int fd, int quiet,
 		 */
 		for (i=array.raid_disks; i<d; i++) {
 			char buf[4096];
-			offsets[i] += sra->component_size - last_block - 8;
+			if (i==d-1 && backup_file) {
+				/* This is the backup file */
+				offsets[i] = 8;
+			} else
+				offsets[i] += sra->component_size - last_block - 8;
 			if (lseek64(fdlist[i], (offsets[i]<<9) - 4096, 0)
 			    != (offsets[i]<<9) - 4096) {
 				fprintf(stderr, Name ": could not seek...\n");
@@ -739,7 +753,7 @@ int Grow_reshape(char *devname, int fd, int quiet,
 		err = save_stripes(fdlist, offsets,
 				   odisks, ochunk, olevel, olayout,
 				   spares, fdlist+odisks,
-				   0ULL, nstripe*512);
+				   0ULL, last_block*512);
 
 		/* abort if there was an error */
 		if (err < 0) {
@@ -752,7 +766,8 @@ int Grow_reshape(char *devname, int fd, int quiet,
 			bsb.devstart = __cpu_to_le64(offsets[i]);
 			bsb.sb_csum = bsb_csum((char*)&bsb, ((char*)&bsb.sb_csum)-((char*)&bsb));
 			if (lseek64(fdlist[i], (offsets[i]+last_block)<<9, 0) < 0 ||
-			    write(fdlist[i], &bsb, sizeof(bsb)) != sizeof(bsb)) {
+			    write(fdlist[i], &bsb, sizeof(bsb)) != sizeof(bsb) ||
+			    fsync(fdlist[i]) != 0) {
 				fprintf(stderr, Name ": %s: fail to save metadata for critical region backups.\n",
 					devname);
 				goto abort_resume;
@@ -792,6 +807,8 @@ int Grow_reshape(char *devname, int fd, int quiet,
 				close(fdlist[i]);
 		free(fdlist);
 		free(offsets);
+		if (backup_file)
+			unlink(backup_file);
 
 		printf(Name ": ... critical section passed.\n");
 		break;
@@ -807,6 +824,8 @@ int Grow_reshape(char *devname, int fd, int quiet,
 			close(fdlist[i]);
 	free(fdlist);
 	free(offsets);
+	if (backup_file)
+		unlink(backup_file);
 	return 1;
 
 }
@@ -816,7 +835,7 @@ int Grow_reshape(char *devname, int fd, int quiet,
  * write that data into the array and update the super blocks with
  * the new reshape_progress
  */
-int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt)
+int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt, char *backup_file)
 {
 	int i, j;
 	int old_disks;
@@ -832,11 +851,12 @@ int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt
 
 	old_disks = info->array.raid_disks - info->delta_disks;
 
-	for (i=old_disks; i<cnt; i++) {
+	for (i=old_disks-(backup_file?1:0); i<cnt; i++) {
 		void *super = NULL;
 		struct mdinfo dinfo;
 		struct mdp_backup_super bsb;
 		char buf[4096];
+		int fd;
 
 		/* This was a spare and may have some saved data on it.
 		 * Load the superblock, find and load the
@@ -845,18 +865,27 @@ int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt
 		 * If the backup contains no new info, just return
 		 * else restore data and update all superblocks
 		 */
-		if (fdlist[i] < 0)
-			continue;
-		if (st->ss->load_super(st, fdlist[i], &super, NULL))
-			continue;
+		if (i == old_disks-1) {
+			fd = open(backup_file, O_RDONLY);
+			if (fd<0)
+				continue;
+			if (lseek(fd, 4096, 0) != 4096)
+				continue;
+		} else {
+			fd = fdlist[i];
+			if (fd < 0)
+				continue;
+			if (st->ss->load_super(st, fd, &super, NULL))
+				continue;
 
-		st->ss->getinfo_super(&dinfo, super);
-		free(super); super = NULL;
-		if (lseek64(fdlist[i],
-			(dinfo.data_offset + dinfo.component_size - 8) <<9,
-			    0) < 0)
-			continue; /* Cannot seek */
-		if (read(fdlist[i], &bsb, sizeof(bsb)) != sizeof(bsb))
+			st->ss->getinfo_super(&dinfo, super);
+			free(super); super = NULL;
+			if (lseek64(fd,
+				    (dinfo.data_offset + dinfo.component_size - 8) <<9,
+				    0) < 0)
+				continue; /* Cannot seek */
+		}
+		if (read(fd, &bsb, sizeof(bsb)) != sizeof(bsb))
 			continue; /* Cannot read */
 		if (memcmp(bsb.magic, "md_backup_data-1", 16) != 0)
 			continue;
@@ -875,15 +904,14 @@ int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt
 		    info->reshape_progress)
 			continue; /* No new data here */
 
-		if (lseek64(fdlist[i], __le64_to_cpu(bsb.devstart)*512, 0)< 0)
+		if (lseek64(fd, __le64_to_cpu(bsb.devstart)*512, 0)< 0)
 			continue; /* Cannot seek */
 		/* There should be a duplicate backup superblock 4k before here */
-		if (lseek64(fdlist[i], -4096, 1) < 0 ||
-		    read(fdlist[i], buf, 4096) != 4096 ||
+		if (lseek64(fd, -4096, 1) < 0 ||
+		    read(fd, buf, 4096) != 4096 ||
 		    memcmp(buf, &bsb, sizeof(buf)) != 0)
 			continue; /* Cannot find leading superblock */
 
-
 		/* Now need the data offsets for all devices. */
 		offsets = malloc(sizeof(*offsets)*info->array.raid_disks);
 		for(j=0; j<info->array.raid_disks; j++) {
@@ -903,7 +931,7 @@ int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt
 				    info->new_chunk,
 				    info->new_level,
 				    info->new_layout,
-				    fdlist[i], __le64_to_cpu(bsb.devstart)*512,
+				    fd, __le64_to_cpu(bsb.devstart)*512,
 				    0, __le64_to_cpu(bsb.length)*512)) {
 			/* didn't succeed, so giveup */
 			return -1;
diff --git a/ReadMe.c b/ReadMe.c
index c508a36..448e966 100644
--- a/ReadMe.c
+++ b/ReadMe.c
@@ -172,6 +172,8 @@ struct option long_options[] = {
     {"daemonize", 0, 0, 'f'},
     {"oneshot",   0, 0, '1'},
     {"pid-file",  1, 0, 'i'},
+    /* For Grow */
+    {"backup-file", 1,0, 7},
     
     {0, 0, 0, 0}
 };
diff --git a/mdadm.c b/mdadm.c
index 5347999..6e57c5b 100644
--- a/mdadm.c
+++ b/mdadm.c
@@ -62,6 +62,7 @@ int main(int argc, char *argv[])
 	int write_behind = 0;
 	int bitmap_fd = -1;
 	char *bitmap_file = NULL;
+	char *backup_file = NULL;
 	int bitmap_chunk = UnSet;
 	int SparcAdjust = 0;
 	mddev_dev_t devlist = NULL;
@@ -769,6 +770,18 @@ int main(int argc, char *argv[])
 			ident.bitmap_fd = bitmap_fd; /* for Assemble */
 			continue;
 
+		case O(ASSEMBLE, 7):
+		case O(GROW, 7):
+			/* Specify a file into which grow might place a backup,
+			 * or from which assemble might recover a backup
+			 */
+			if (backup_file) {
+				fprintf(stderr, Name ": backup file already specified, rejecting %s\n", optarg);
+				exit(2);
+			}
+			backup_file = optarg;
+			continue;
+
 		case O(GROW,'b'):
 		case O(BUILD,'b'):
 		case O(CREATE,'b'): /* here we create the bitmap */
@@ -812,8 +825,13 @@ int main(int argc, char *argv[])
 		/* We have now processed all the valid options. Anything else is
 		 * an error
 		 */
-		fprintf(stderr, Name ": option %c not valid in %s mode\n",
-			opt, map_num(modes, mode));
+		if (option_index > 0)
+			fprintf(stderr, Name ":option --%s not valid in %s mode\n",
+				long_options[option_index].name,
+				map_num(modes, mode));
+		else
+			fprintf(stderr, Name ": option -%c not valid in %s mode\n",
+				opt, map_num(modes, mode));
 		exit(2);
 
 	}
@@ -927,20 +945,24 @@ int main(int argc, char *argv[])
 					rv |= 1;
 				else {
 					rv |= Assemble(ss, devlist->devname, mdfd, array_ident, configfile,
-						       NULL,
+						       NULL, backup_file,
 						       readonly, runstop, update, verbose-quiet, force);
 					close(mdfd);
 				}
 			}
 		} else if (!scan)
 			rv = Assemble(ss, devlist->devname, mdfd, &ident, configfile,
-				      devlist->next,
+				      devlist->next, backup_file,
 				      readonly, runstop, update, verbose-quiet, force);
 		else if (devs_found>0) {
 			if (update && devs_found > 1) {
 				fprintf(stderr, Name ": can only update a single array at a time\n");
 				exit(1);
 			}
+			if (backup_file && devs_found > 1) {
+				fprintf(stderr, Name ": can only assemble a single array when providing a backup file.\n");
+				exit(1);
+			}
 			for (dv = devlist ; dv ; dv=dv->next) {
 				mddev_ident_t array_ident = conf_get_ident(configfile, dv->devname);
 				if (array_ident == NULL) {
@@ -956,7 +978,7 @@ int main(int argc, char *argv[])
 					continue;
 				}
 				rv |= Assemble(ss, dv->devname, mdfd, array_ident, configfile,
-					       NULL,
+					       NULL, backup_file,
 					       readonly, runstop, update, verbose-quiet, force);
 				close(mdfd);
 			}
@@ -966,6 +988,14 @@ int main(int argc, char *argv[])
 				fprintf(stderr, Name ": No arrays found in config file\n");
 				rv = 1;
 			} else
+				if (update) {
+					fprintf(stderr, Name ": --update not meaningful with a --scan assembly.\n");
+					exit(1);
+				}
+				if (backup_file) {
+					fprintf(stderr, Name ": --backup_file not meaningful with a --scan assembly.\n");
+					exit(1);
+				}
 				for (; array_list; array_list = array_list->next) {
 					mdu_array_info_t array;
 					mdfd = open_mddev(array_list->devname, 
@@ -980,7 +1010,7 @@ int main(int argc, char *argv[])
 					else
 						rv |= Assemble(ss, array_list->devname, mdfd,
 							       array_list, configfile,
-							       NULL,
+							       NULL, NULL,
 							       readonly, runstop, NULL, verbose-quiet, force);
 					close(mdfd);
 				}
@@ -1155,7 +1185,7 @@ int main(int argc, char *argv[])
 		} else if (layout != UnSet)
 			rv = Manage_reconfig(devlist->devname, mdfd, layout);
 		else if (size >= 0 || raiddisks)
-			rv = Grow_reshape(devlist->devname, mdfd, quiet,
+			rv = Grow_reshape(devlist->devname, mdfd, quiet, backup_file,
 					  size, level, layout, chunk, raiddisks);
 		else if (bitmap_file) {
 			if (delay == 0) delay = DEFAULT_BITMAP_DELAY;
diff --git a/mdadm.h b/mdadm.h
index c929be9..1752623 100644
--- a/mdadm.h
+++ b/mdadm.h
@@ -309,16 +309,17 @@ extern int Manage_subdevs(char *devname, int fd,
 			  mddev_dev_t devlist, int verbose);
 extern int Grow_Add_device(char *devname, int fd, char *newdev);
 extern int Grow_addbitmap(char *devname, int fd, char *file, int chunk, int delay, int write_behind, int force);
-extern int Grow_reshape(char *devname, int fd, int quiet,
+extern int Grow_reshape(char *devname, int fd, int quiet, char *backup_file,
 			long long size,
 			int level, int layout, int chunksize, int raid_disks);
-extern int Grow_restart(struct supertype *st, struct mdinfo *info, int *fdlist, int cnt);
+extern int Grow_restart(struct supertype *st, struct mdinfo *info,
+			int *fdlist, int cnt, char *backup_file);
 
 
 extern int Assemble(struct supertype *st, char *mddev, int mdfd,
 		    mddev_ident_t ident,
 		    char *conffile,
-		    mddev_dev_t devlist,
+		    mddev_dev_t devlist, char *backup_file,
 		    int readonly, int runstop,
 		    char *update,
 		    int verbose, int force);