]> git.karo-electronics.de Git - linux-beck.git/commitdiff
Btrfs: make filesystem read-only when submitting barrier fails
authorStefan Behrens <sbehrens@giantdisaster.de>
Wed, 1 Aug 2012 16:56:49 +0000 (18:56 +0200)
committerChris Mason <chris.mason@fusionio.com>
Tue, 9 Oct 2012 13:20:19 +0000 (09:20 -0400)
So far the return code of barrier_all_devices() is ignored, which
means that errors are ignored. The result can be a corrupt
filesystem which is not consistent.
This commit adds code to evaluate the return code of
barrier_all_devices(). The normal btrfs_error() mechanism is used to
switch the filesystem into read-only mode when errors are detected.

In order to decide whether barrier_all_devices() should return
error or success, the number of disks that are allowed to fail the
barrier submission is calculated. This calculation accounts for the
worst RAID level of metadata, system and data. If single, dup or
RAID0 is in use, a single disk error is already considered to be
fatal. Otherwise a single disk error is tolerated.

The calculation of the number of disks that are tolerated to fail
the barrier operation is performed when the filesystem gets mounted,
when a balance operation is started and finished, and when devices
are added or removed.

Signed-off-by: Stefan Behrens <sbehrens@giantdisaster.de>
fs/btrfs/ctree.h
fs/btrfs/disk-io.c
fs/btrfs/disk-io.h
fs/btrfs/ioctl.c
fs/btrfs/tree-log.c
fs/btrfs/volumes.c

index 50dcd0fbae111eae757ac9076a01a615c51ff815..1630be831210c9c9a0bb57eb7d5e98ad36a3b235 100644 (file)
@@ -1468,6 +1468,8 @@ struct btrfs_fs_info {
 
        /* next backup root to be overwritten */
        int backup_root_index;
+
+       int num_tolerated_disk_barrier_failures;
 };
 
 /*
@@ -3361,6 +3363,9 @@ void btrfs_inherit_iflags(struct inode *inode, struct inode *dir);
 int btrfs_defrag_file(struct inode *inode, struct file *file,
                      struct btrfs_ioctl_defrag_range_args *range,
                      u64 newer_than, unsigned long max_pages);
+void btrfs_get_block_group_info(struct list_head *groups_list,
+                               struct btrfs_ioctl_space_info *space);
+
 /* file.c */
 int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans,
                           struct inode *inode);
index c69995556f61f331728245607433b5c90e09064e..83552368770796471f21f5af7218813139d3c9e2 100644 (file)
@@ -2505,6 +2505,8 @@ retry_root_backup:
                printk(KERN_ERR "Failed to read block groups: %d\n", ret);
                goto fail_block_groups;
        }
+       fs_info->num_tolerated_disk_barrier_failures =
+               btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
 
        fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
                                               "btrfs-cleaner");
@@ -2888,12 +2890,10 @@ static int write_dev_flush(struct btrfs_device *device, int wait)
                        printk_in_rcu("btrfs: disabling barriers on dev %s\n",
                                      rcu_str_deref(device->name));
                        device->nobarriers = 1;
-               }
-               if (!bio_flagged(bio, BIO_UPTODATE)) {
+               } else if (!bio_flagged(bio, BIO_UPTODATE)) {
                        ret = -EIO;
-                       if (!bio_flagged(bio, BIO_EOPNOTSUPP))
-                               btrfs_dev_stat_inc_and_print(device,
-                                       BTRFS_DEV_STAT_FLUSH_ERRS);
+                       btrfs_dev_stat_inc_and_print(device,
+                               BTRFS_DEV_STAT_FLUSH_ERRS);
                }
 
                /* drop the reference from the wait == 0 run */
@@ -2932,14 +2932,15 @@ static int barrier_all_devices(struct btrfs_fs_info *info)
 {
        struct list_head *head;
        struct btrfs_device *dev;
-       int errors = 0;
+       int errors_send = 0;
+       int errors_wait = 0;
        int ret;
 
        /* send down all the barriers */
        head = &info->fs_devices->devices;
        list_for_each_entry_rcu(dev, head, dev_list) {
                if (!dev->bdev) {
-                       errors++;
+                       errors_send++;
                        continue;
                }
                if (!dev->in_fs_metadata || !dev->writeable)
@@ -2947,13 +2948,13 @@ static int barrier_all_devices(struct btrfs_fs_info *info)
 
                ret = write_dev_flush(dev, 0);
                if (ret)
-                       errors++;
+                       errors_send++;
        }
 
        /* wait for all the barriers */
        list_for_each_entry_rcu(dev, head, dev_list) {
                if (!dev->bdev) {
-                       errors++;
+                       errors_wait++;
                        continue;
                }
                if (!dev->in_fs_metadata || !dev->writeable)
@@ -2961,13 +2962,87 @@ static int barrier_all_devices(struct btrfs_fs_info *info)
 
                ret = write_dev_flush(dev, 1);
                if (ret)
-                       errors++;
+                       errors_wait++;
        }
-       if (errors)
+       if (errors_send > info->num_tolerated_disk_barrier_failures ||
+           errors_wait > info->num_tolerated_disk_barrier_failures)
                return -EIO;
        return 0;
 }
 
+int btrfs_calc_num_tolerated_disk_barrier_failures(
+       struct btrfs_fs_info *fs_info)
+{
+       struct btrfs_ioctl_space_info space;
+       struct btrfs_space_info *sinfo;
+       u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
+                      BTRFS_BLOCK_GROUP_SYSTEM,
+                      BTRFS_BLOCK_GROUP_METADATA,
+                      BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
+       int num_types = 4;
+       int i;
+       int c;
+       int num_tolerated_disk_barrier_failures =
+               (int)fs_info->fs_devices->num_devices;
+
+       for (i = 0; i < num_types; i++) {
+               struct btrfs_space_info *tmp;
+
+               sinfo = NULL;
+               rcu_read_lock();
+               list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
+                       if (tmp->flags == types[i]) {
+                               sinfo = tmp;
+                               break;
+                       }
+               }
+               rcu_read_unlock();
+
+               if (!sinfo)
+                       continue;
+
+               down_read(&sinfo->groups_sem);
+               for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
+                       if (!list_empty(&sinfo->block_groups[c])) {
+                               u64 flags;
+
+                               btrfs_get_block_group_info(
+                                       &sinfo->block_groups[c], &space);
+                               if (space.total_bytes == 0 ||
+                                   space.used_bytes == 0)
+                                       continue;
+                               flags = space.flags;
+                               /*
+                                * return
+                                * 0: if dup, single or RAID0 is configured for
+                                *    any of metadata, system or data, else
+                                * 1: if RAID5 is configured, or if RAID1 or
+                                *    RAID10 is configured and only two mirrors
+                                *    are used, else
+                                * 2: if RAID6 is configured, else
+                                * num_mirrors - 1: if RAID1 or RAID10 is
+                                *                  configured and more than
+                                *                  2 mirrors are used.
+                                */
+                               if (num_tolerated_disk_barrier_failures > 0 &&
+                                   ((flags & (BTRFS_BLOCK_GROUP_DUP |
+                                              BTRFS_BLOCK_GROUP_RAID0)) ||
+                                    ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
+                                     == 0)))
+                                       num_tolerated_disk_barrier_failures = 0;
+                               else if (num_tolerated_disk_barrier_failures > 1
+                                        &&
+                                        (flags & (BTRFS_BLOCK_GROUP_RAID1 |
+                                                  BTRFS_BLOCK_GROUP_RAID10)))
+                                       num_tolerated_disk_barrier_failures = 1;
+                       }
+               }
+               up_read(&sinfo->groups_sem);
+       }
+
+       return num_tolerated_disk_barrier_failures;
+}
+
 int write_all_supers(struct btrfs_root *root, int max_mirrors)
 {
        struct list_head *head;
@@ -2990,8 +3065,16 @@ int write_all_supers(struct btrfs_root *root, int max_mirrors)
        mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
        head = &root->fs_info->fs_devices->devices;
 
-       if (do_barriers)
-               barrier_all_devices(root->fs_info);
+       if (do_barriers) {
+               ret = barrier_all_devices(root->fs_info);
+               if (ret) {
+                       mutex_unlock(
+                               &root->fs_info->fs_devices->device_list_mutex);
+                       btrfs_error(root->fs_info, ret,
+                                   "errors while submitting device barriers.");
+                       return ret;
+               }
+       }
 
        list_for_each_entry_rcu(dev, head, dev_list) {
                if (!dev->bdev) {
index c5b00a735fefac258b7914223139d902efacbf99..2025a9132c16119c5795b2614b7d114ada047643 100644 (file)
@@ -95,6 +95,8 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
                                     u64 objectid);
 int btree_lock_page_hook(struct page *page, void *data,
                                void (*flush_fn)(void *));
+int btrfs_calc_num_tolerated_disk_barrier_failures(
+       struct btrfs_fs_info *fs_info);
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 void btrfs_init_lockdep(void);
index d6836af6d60f5a23768eb6467c611a417621832f..f5a2e6c4320ae9e4283800ff2b0e6c4bbe60687a 100644 (file)
@@ -2875,8 +2875,8 @@ static long btrfs_ioctl_default_subvol(struct file *file, void __user *argp)
        return 0;
 }
 
-static void get_block_group_info(struct list_head *groups_list,
-                                struct btrfs_ioctl_space_info *space)
+void btrfs_get_block_group_info(struct list_head *groups_list,
+                               struct btrfs_ioctl_space_info *space)
 {
        struct btrfs_block_group_cache *block_group;
 
@@ -2984,8 +2984,8 @@ long btrfs_ioctl_space_info(struct btrfs_root *root, void __user *arg)
                down_read(&info->groups_sem);
                for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
                        if (!list_empty(&info->block_groups[c])) {
-                               get_block_group_info(&info->block_groups[c],
-                                                    &space);
+                               btrfs_get_block_group_info(
+                                       &info->block_groups[c], &space);
                                memcpy(dest, &space, sizeof(space));
                                dest++;
                                space_args.total_spaces++;
index 47911fd18310a58964a6f07fd12c29c887ab4dba..67eab2d4d8a9bc125279454ed10577cff095b5de 100644 (file)
@@ -2425,9 +2425,12 @@ int btrfs_sync_log(struct btrfs_trans_handle *trans,
         * in and cause problems either.
         */
        btrfs_scrub_pause_super(root);
-       write_ctree_super(trans, root->fs_info->tree_root, 1);
+       ret = write_ctree_super(trans, root->fs_info->tree_root, 1);
        btrfs_scrub_continue_super(root);
-       ret = 0;
+       if (ret) {
+               btrfs_abort_transaction(trans, root, ret);
+               goto out_wake_log_root;
+       }
 
        mutex_lock(&root->log_mutex);
        if (root->last_log_commit < log_transid)
index dfe5e3a22f5517b4251e5565bdc34340e39a7bae..029b903a4ae3797322e05090790b86c9e8596c43 100644 (file)
@@ -1475,6 +1475,9 @@ int btrfs_rm_device(struct btrfs_root *root, char *device_path)
                free_fs_devices(cur_devices);
        }
 
+       root->fs_info->num_tolerated_disk_barrier_failures =
+               btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
+
        /*
         * at this point, the device is zero sized.  We want to
         * remove it from the devices list and zero out the old super
@@ -1799,6 +1802,8 @@ int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
        btrfs_clear_space_info_full(root->fs_info);
 
        unlock_chunks(root);
+       root->fs_info->num_tolerated_disk_barrier_failures =
+               btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
        ret = btrfs_commit_transaction(trans, root);
 
        if (seeding_dev) {
@@ -2809,6 +2814,26 @@ int btrfs_balance(struct btrfs_balance_control *bctl,
                }
        }
 
+       if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
+               int num_tolerated_disk_barrier_failures;
+               u64 target = bctl->sys.target;
+
+               num_tolerated_disk_barrier_failures =
+                       btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
+               if (num_tolerated_disk_barrier_failures > 0 &&
+                   (target &
+                    (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
+                     BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
+                       num_tolerated_disk_barrier_failures = 0;
+               else if (num_tolerated_disk_barrier_failures > 1 &&
+                        (target &
+                         (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
+                       num_tolerated_disk_barrier_failures = 1;
+
+               fs_info->num_tolerated_disk_barrier_failures =
+                       num_tolerated_disk_barrier_failures;
+       }
+
        ret = insert_balance_item(fs_info->tree_root, bctl);
        if (ret && ret != -EEXIST)
                goto out;
@@ -2841,6 +2866,11 @@ int btrfs_balance(struct btrfs_balance_control *bctl,
                __cancel_balance(fs_info);
        }
 
+       if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
+               fs_info->num_tolerated_disk_barrier_failures =
+                       btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
+       }
+
        wake_up(&fs_info->balance_wait_q);
 
        return ret;