]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - fs/ext4/super.c
ext4: Fix fsync error handling after filesystem abort
[mirror_ubuntu-bionic-kernel.git] / fs / ext4 / super.c
index 94cc84db7c9aae349b44be229e993ec906a0a25a..eb52a7be1fc13a174fc963614c35f849d2ef3885 100644 (file)
@@ -69,6 +69,7 @@ static void ext4_mark_recovery_complete(struct super_block *sb,
 static void ext4_clear_journal_err(struct super_block *sb,
                                   struct ext4_super_block *es);
 static int ext4_sync_fs(struct super_block *sb, int wait);
+static int ext4_sync_fs_nojournal(struct super_block *sb, int wait);
 static int ext4_remount(struct super_block *sb, int *flags, char *data);
 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
 static int ext4_unfreeze(struct super_block *sb);
@@ -398,6 +399,11 @@ static void ext4_handle_error(struct super_block *sb)
        }
        if (test_opt(sb, ERRORS_RO)) {
                ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
+               /*
+                * Make sure updated value of ->s_mount_flags will be visible
+                * before ->s_flags update
+                */
+               smp_wmb();
                sb->s_flags |= MS_RDONLY;
        }
        if (test_opt(sb, ERRORS_PANIC))
@@ -570,8 +576,13 @@ void __ext4_abort(struct super_block *sb, const char *function,
 
        if ((sb->s_flags & MS_RDONLY) == 0) {
                ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
-               sb->s_flags |= MS_RDONLY;
                EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
+               /*
+                * Make sure updated value of ->s_mount_flags will be visible
+                * before ->s_flags update
+                */
+               smp_wmb();
+               sb->s_flags |= MS_RDONLY;
                if (EXT4_SB(sb)->s_journal)
                        jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
                save_error_info(sb, function, line);
@@ -750,8 +761,10 @@ static void ext4_put_super(struct super_block *sb)
        ext4_unregister_li_request(sb);
        dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 
-       flush_workqueue(sbi->dio_unwritten_wq);
-       destroy_workqueue(sbi->dio_unwritten_wq);
+       flush_workqueue(sbi->unrsv_conversion_wq);
+       flush_workqueue(sbi->rsv_conversion_wq);
+       destroy_workqueue(sbi->unrsv_conversion_wq);
+       destroy_workqueue(sbi->rsv_conversion_wq);
 
        if (sbi->s_journal) {
                err = jbd2_journal_destroy(sbi->s_journal);
@@ -859,13 +872,15 @@ static struct inode *ext4_alloc_inode(struct super_block *sb)
        ei->i_reserved_quota = 0;
 #endif
        ei->jinode = NULL;
-       INIT_LIST_HEAD(&ei->i_completed_io_list);
+       INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
+       INIT_LIST_HEAD(&ei->i_unrsv_conversion_list);
        spin_lock_init(&ei->i_completed_io_lock);
        ei->i_sync_tid = 0;
        ei->i_datasync_tid = 0;
        atomic_set(&ei->i_ioend_count, 0);
        atomic_set(&ei->i_unwritten, 0);
-       INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
+       INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
+       INIT_WORK(&ei->i_unrsv_conversion_work, ext4_end_io_unrsv_work);
 
        return &ei->vfs_inode;
 }
@@ -1093,6 +1108,7 @@ static const struct super_operations ext4_nojournal_sops = {
        .dirty_inode    = ext4_dirty_inode,
        .drop_inode     = ext4_drop_inode,
        .evict_inode    = ext4_evict_inode,
+       .sync_fs        = ext4_sync_fs_nojournal,
        .put_super      = ext4_put_super,
        .statfs         = ext4_statfs,
        .remount_fs     = ext4_remount,
@@ -2164,19 +2180,22 @@ static void ext4_orphan_cleanup(struct super_block *sb,
                list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
                dquot_initialize(inode);
                if (inode->i_nlink) {
-                       ext4_msg(sb, KERN_DEBUG,
-                               "%s: truncating inode %lu to %lld bytes",
-                               __func__, inode->i_ino, inode->i_size);
+                       if (test_opt(sb, DEBUG))
+                               ext4_msg(sb, KERN_DEBUG,
+                                       "%s: truncating inode %lu to %lld bytes",
+                                       __func__, inode->i_ino, inode->i_size);
                        jbd_debug(2, "truncating inode %lu to %lld bytes\n",
                                  inode->i_ino, inode->i_size);
                        mutex_lock(&inode->i_mutex);
+                       truncate_inode_pages(inode->i_mapping, inode->i_size);
                        ext4_truncate(inode);
                        mutex_unlock(&inode->i_mutex);
                        nr_truncates++;
                } else {
-                       ext4_msg(sb, KERN_DEBUG,
-                               "%s: deleting unreferenced inode %lu",
-                               __func__, inode->i_ino);
+                       if (test_opt(sb, DEBUG))
+                               ext4_msg(sb, KERN_DEBUG,
+                                       "%s: deleting unreferenced inode %lu",
+                                       __func__, inode->i_ino);
                        jbd_debug(2, "deleting unreferenced inode %lu\n",
                                  inode->i_ino);
                        nr_orphans++;
@@ -2377,7 +2396,10 @@ struct ext4_attr {
        ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
        ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
                         const char *, size_t);
-       int offset;
+       union {
+               int offset;
+               int deprecated_val;
+       } u;
 };
 
 static int parse_strtoull(const char *buf,
@@ -2446,7 +2468,7 @@ static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
 static ssize_t sbi_ui_show(struct ext4_attr *a,
                           struct ext4_sb_info *sbi, char *buf)
 {
-       unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
+       unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
 
        return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
 }
@@ -2455,7 +2477,7 @@ static ssize_t sbi_ui_store(struct ext4_attr *a,
                            struct ext4_sb_info *sbi,
                            const char *buf, size_t count)
 {
-       unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
+       unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
        unsigned long t;
        int ret;
 
@@ -2504,12 +2526,20 @@ static ssize_t trigger_test_error(struct ext4_attr *a,
        return count;
 }
 
+static ssize_t sbi_deprecated_show(struct ext4_attr *a,
+                                  struct ext4_sb_info *sbi, char *buf)
+{
+       return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
+}
+
 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
 static struct ext4_attr ext4_attr_##_name = {                  \
        .attr = {.name = __stringify(_name), .mode = _mode },   \
        .show   = _show,                                        \
        .store  = _store,                                       \
-       .offset = offsetof(struct ext4_sb_info, _elname),       \
+       .u = {                                                  \
+               .offset = offsetof(struct ext4_sb_info, _elname),\
+       },                                                      \
 }
 #define EXT4_ATTR(name, mode, show, store) \
 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
@@ -2520,6 +2550,14 @@ static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
 #define EXT4_RW_ATTR_SBI_UI(name, elname)      \
        EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
 #define ATTR_LIST(name) &ext4_attr_##name.attr
+#define EXT4_DEPRECATED_ATTR(_name, _val)      \
+static struct ext4_attr ext4_attr_##_name = {                  \
+       .attr = {.name = __stringify(_name), .mode = 0444 },    \
+       .show   = sbi_deprecated_show,                          \
+       .u = {                                                  \
+               .deprecated_val = _val,                         \
+       },                                                      \
+}
 
 EXT4_RO_ATTR(delayed_allocation_blocks);
 EXT4_RO_ATTR(session_write_kbytes);
@@ -2534,7 +2572,7 @@ EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
-EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
+EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
 
@@ -3787,7 +3825,6 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
        }
 
        sbi->s_stripe = ext4_get_stripe_size(sbi);
-       sbi->s_max_writeback_mb_bump = 128;
        sbi->s_extent_max_zeroout_kb = 32;
 
        /*
@@ -3915,12 +3952,20 @@ no_journal:
         * The maximum number of concurrent works can be high and
         * concurrency isn't really necessary.  Limit it to 1.
         */
-       EXT4_SB(sb)->dio_unwritten_wq =
-               alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
-       if (!EXT4_SB(sb)->dio_unwritten_wq) {
-               printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
+       EXT4_SB(sb)->rsv_conversion_wq =
+               alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
+       if (!EXT4_SB(sb)->rsv_conversion_wq) {
+               printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
                ret = -ENOMEM;
-               goto failed_mount_wq;
+               goto failed_mount4;
+       }
+
+       EXT4_SB(sb)->unrsv_conversion_wq =
+               alloc_workqueue("ext4-unrsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
+       if (!EXT4_SB(sb)->unrsv_conversion_wq) {
+               printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
+               ret = -ENOMEM;
+               goto failed_mount4;
        }
 
        /*
@@ -4074,7 +4119,10 @@ failed_mount4a:
        sb->s_root = NULL;
 failed_mount4:
        ext4_msg(sb, KERN_ERR, "mount failed");
-       destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
+       if (EXT4_SB(sb)->rsv_conversion_wq)
+               destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
+       if (EXT4_SB(sb)->unrsv_conversion_wq)
+               destroy_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
 failed_mount_wq:
        if (sbi->s_journal) {
                jbd2_journal_destroy(sbi->s_journal);
@@ -4517,19 +4565,52 @@ static int ext4_sync_fs(struct super_block *sb, int wait)
 {
        int ret = 0;
        tid_t target;
+       bool needs_barrier = false;
        struct ext4_sb_info *sbi = EXT4_SB(sb);
 
        trace_ext4_sync_fs(sb, wait);
-       flush_workqueue(sbi->dio_unwritten_wq);
+       flush_workqueue(sbi->rsv_conversion_wq);
+       flush_workqueue(sbi->unrsv_conversion_wq);
        /*
         * Writeback quota in non-journalled quota case - journalled quota has
         * no dirty dquots
         */
        dquot_writeback_dquots(sb, -1);
+       /*
+        * Data writeback is possible w/o journal transaction, so barrier must
+        * being sent at the end of the function. But we can skip it if
+        * transaction_commit will do it for us.
+        */
+       target = jbd2_get_latest_transaction(sbi->s_journal);
+       if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
+           !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
+               needs_barrier = true;
+
        if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
                if (wait)
-                       jbd2_log_wait_commit(sbi->s_journal, target);
+                       ret = jbd2_log_wait_commit(sbi->s_journal, target);
+       }
+       if (needs_barrier) {
+               int err;
+               err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
+               if (!ret)
+                       ret = err;
        }
+
+       return ret;
+}
+
+static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
+{
+       int ret = 0;
+
+       trace_ext4_sync_fs(sb, wait);
+       flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
+       flush_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
+       dquot_writeback_dquots(sb, -1);
+       if (wait && test_opt(sb, BARRIER))
+               ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
+
        return ret;
 }