LU-11446 e2fsck: check trusted.link when fixing nlink

The inode link count could be higher than what is stored in the
local MDT inode because of remote file links from DNE MDTs.
If we find a mismatched link count, look up the "trusted.link"
xattr.  If it exists, do a sanity check on it, and use the
leh_reccount stored there if larger than the local link count.

If leh_overflow_time is set, then the "trusted.link" xattr may
not hold all of the links, so assume the maximum of available
link counts is valid until LFSCK clears leh_overflow_time.

If the inode is not linked into the namespace, link it into
lost+found before checking trusted.link for the DNE link count.

Update the f_trusted_link test case to include a locally
unreferenced inode that has a trusted.link xattr.  The
inode should be linked into lost+found because of the xattr,
even if it has no blocks/data, and the link cound should be
extracted from the trusted.link xattr.

Signed-off-by: Li Dongyang <dongyangli@ddn.com>
Change-Id: I213d816a92043c348eb55374aaa98e98957ccf23
Reviewed-on: https://review.whamcloud.com/43169
Reviewed-by: Artem Blagodarenko <artem.blagodarenko@hpe.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>

diff --git a/e2fsck/pass4.c b/e2fsck/pass4.c
index 8c2d2f1..f048fff 100644 (file)
--- a/e2fsck/pass4.c
+++ b/e2fsck/pass4.c
@@ -17,6 +17,7 @@
 #include "config.h"
 #include "e2fsck.h"
 #include "problem.h"
+#include "ext2fs/lfsck.h"
 #include <ext2fs/ext2_ext_attr.h>
 
 /*
@@ -26,7 +27,7 @@
  * This subroutine returns 1 then the caller shouldn't bother with the
  * rest of the pass 4 tests.
  */
-static int disconnect_inode(e2fsck_t ctx, ext2_ino_t i,
+static int disconnect_inode(e2fsck_t ctx, ext2_ino_t i, ext2_ino_t *last_ino,
                            struct ext2_inode_large *inode)
 {
        ext2_filsys fs = ctx->fs;
@@ -34,9 +35,12 @@ static int disconnect_inode(e2fsck_t ctx, ext2_ino_t i,
        __u32 eamagic = 0;
        int extra_size = 0;
 
-       e2fsck_read_inode_full(ctx, i, EXT2_INODE(inode),
-                              EXT2_INODE_SIZE(fs->super),
-                              "pass4: disconnect_inode");
+       if (*last_ino != i) {
+               e2fsck_read_inode_full(ctx, i, EXT2_INODE(inode),
+                                      EXT2_INODE_SIZE(fs->super),
+                                      "pass4: disconnect_inode");
+               *last_ino = i;
+       }
        if (EXT2_INODE_SIZE(fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
                extra_size = inode->i_extra_isize;
 
@@ -75,6 +79,7 @@ static int disconnect_inode(e2fsck_t ctx, ext2_ino_t i,
        if (fix_problem(ctx, PR_4_UNATTACHED_INODE, &pctx)) {
                if (e2fsck_reconnect_file(ctx, i))
                        ext2fs_unmark_valid(fs);
+               *last_ino = 0;
        } else {
                /*
                 * If we don't attach the inode, then skip the
@@ -87,20 +92,23 @@ static int disconnect_inode(e2fsck_t ctx, ext2_ino_t i,
        return 0;
 }
 
-static void check_ea_inode(e2fsck_t ctx, ext2_ino_t i,
+/*
+ * This function is called when link_counted is zero. So this may not
+ * be an xattr inode at all. Return immediately if EA_INODE flag is not
+ * set.
+ */
+static void check_ea_inode(e2fsck_t ctx, ext2_ino_t i, ext2_ino_t *last_ino,
                           struct ext2_inode_large *inode, __u16 *link_counted)
 {
        __u64 actual_refs = 0;
        __u64 ref_count;
 
-       /*
-        * This function is called when link_counted is zero. So this may not
-        * be an xattr inode at all. Return immediately if EA_INODE flag is not
-        * set.
-        */
-       e2fsck_read_inode_full(ctx, i, EXT2_INODE(inode),
-                              EXT2_INODE_SIZE(ctx->fs->super),
-                              "pass4: check_ea_inode");
+       if (*last_ino != i) {
+               e2fsck_read_inode_full(ctx, i, EXT2_INODE(inode),
+                                      EXT2_INODE_SIZE(ctx->fs->super),
+                                      "pass4: check_ea_inode");
+               *last_ino = i;
+       }
        if (!(inode->i_flags & EXT4_EA_INODE_FL))
                return;
 
@@ -134,6 +142,73 @@ static void check_ea_inode(e2fsck_t ctx, ext2_ino_t i,
        }
 }
 
+static errcode_t check_link_ea(e2fsck_t ctx, ext2_ino_t ino,
+                              ext2_ino_t *last_ino,
+                              struct ext2_inode_large *inode,
+                              __u16 *link_counted)
+{
+       struct ext2_xattr_handle *handle;
+       struct link_ea_header *leh;
+       void *buf;
+       size_t ea_len;
+       errcode_t retval;
+
+       if (*last_ino != ino) {
+               e2fsck_read_inode_full(ctx, ino, EXT2_INODE(inode),
+                                      EXT2_INODE_SIZE(ctx->fs->super),
+                                      "pass4: get link ea count");
+               *last_ino = ino;
+       }
+
+       retval = ext2fs_xattrs_open(ctx->fs, ino, &handle);
+       if (retval)
+               return retval;
+
+       retval = ext2fs_xattrs_read_inode(handle, inode);
+       if (retval)
+               goto err;
+
+       retval = ext2fs_xattr_get(handle, EXT2_ATTR_INDEX_TRUSTED_PREFIX
+                                 LUSTRE_XATTR_MDT_LINK, &buf, &ea_len);
+       if (retval)
+               goto err;
+
+       leh = (struct link_ea_header *)buf;
+       if (leh->leh_magic == ext2fs_swab32(LINK_EA_MAGIC)) {
+               leh->leh_magic = LINK_EA_MAGIC;
+               leh->leh_reccount = ext2fs_swab32(leh->leh_reccount);
+               leh->leh_len = ext2fs_swab64(leh->leh_len);
+       }
+       if (leh->leh_magic != LINK_EA_MAGIC) {
+               retval = EINVAL;
+               goto err_free;
+       }
+       if (leh->leh_reccount == 0 && !leh->leh_overflow_time) {
+               retval = ENODATA;
+               goto err_free;
+       }
+       if (leh->leh_len > ea_len) {
+               retval = EINVAL;
+               goto err_free;
+       }
+
+       /* if linkEA overflowed and does not hold all links, assume *some*
+        * links exist until LFSCK is next run and resets leh_overflow_time */
+       if (leh->leh_overflow_time) {
+               if (inode->i_links_count > *link_counted)
+                       *link_counted = inode->i_links_count;
+               else if (*link_counted == 0)
+                       *link_counted = 1111;
+       }
+       if (leh->leh_reccount > *link_counted)
+               *link_counted = leh->leh_reccount;
+err_free:
+       ext2fs_free_mem(&buf);
+err:
+       ext2fs_xattrs_close(&handle);
+       return retval;
+}
+
 void e2fsck_pass4(e2fsck_t ctx)
 {
        ext2_filsys fs = ctx->fs;
@@ -180,7 +255,8 @@ void e2fsck_pass4(e2fsck_t ctx)
        inode = e2fsck_allocate_memory(ctx, inode_size, "scratch inode");
 
        /* Protect loop from wrap-around if s_inodes_count maxed */
-       for (i=1; i <= fs->super->s_inodes_count && i > 0; i++) {
+       for (i = 1; i <= fs->super->s_inodes_count && i > 0; i++) {
+               ext2_ino_t last_ino = 0;
                int isdir;
 
                if (ctx->flags & E2F_FLAG_SIGNAL_MASK)
@@ -210,7 +286,7 @@ void e2fsck_pass4(e2fsck_t ctx)
                         * check_ea_inode() will update link_counted if
                         * necessary.
                         */
-                       check_ea_inode(ctx, i, inode, &link_counted);
+                       check_ea_inode(ctx, i, &last_ino, inode, &link_counted);
                }
 
                if (link_counted == 0) {
@@ -219,12 +295,13 @@ void e2fsck_pass4(e2fsck_t ctx)
                                     fs->blocksize, "bad_inode buffer");
                        if (e2fsck_process_bad_inode(ctx, 0, i, buf))
                                continue;
-                       if (disconnect_inode(ctx, i, inode))
+                       if (disconnect_inode(ctx, i, &last_ino, inode))
                                continue;
                        ext2fs_icount_fetch(ctx->inode_link_info, i,
                                            &link_count);
                        ext2fs_icount_fetch(ctx->inode_count, i,
                                            &link_counted);
+                       check_link_ea(ctx, i, &last_ino, inode, &link_counted);
                }
                isdir = ext2fs_test_inode_bitmap2(ctx->inode_dir_map, i);
                if (isdir && (link_counted > EXT2_LINK_MAX)) {
@@ -236,11 +313,18 @@ void e2fsck_pass4(e2fsck_t ctx)
                        }
                        link_counted = 1;
                }
+               if (link_counted != link_count)
+                       check_link_ea(ctx, i, &last_ino, inode, &link_counted);
+
                if (link_counted != link_count) {
                        int fix_nlink = 0;
 
-                       e2fsck_read_inode_full(ctx, i, EXT2_INODE(inode),
-                                              inode_size, "pass4");
+                       if (last_ino != i) {
+                               e2fsck_read_inode_full(ctx, i,
+                                                      EXT2_INODE(inode),
+                                                      inode_size, "pass4");
+                               last_ino = i;
+                       }
                        pctx.ino = i;
                        pctx.inode = EXT2_INODE(inode);
                        if ((link_count != inode->i_links_count) && !isdir &&

diff --git a/lib/ext2fs/ext2fs.h b/lib/ext2fs/ext2fs.h
index 53fc1b1..b7fd692 100644 (file)
--- a/lib/ext2fs/ext2fs.h
+++ b/lib/ext2fs/ext2fs.h
@@ -1318,6 +1318,8 @@ extern errcode_t ext2fs_adjust_ea_refcount3(ext2_filsys fs, blk64_t blk,
                                           ext2_ino_t inum);
 errcode_t ext2fs_xattrs_write(struct ext2_xattr_handle *handle);
 errcode_t ext2fs_xattrs_read(struct ext2_xattr_handle *handle);
+errcode_t ext2fs_xattrs_read_inode(struct ext2_xattr_handle *handle,
+                                  struct ext2_inode_large *inode);
 errcode_t ext2fs_xattrs_iterate(struct ext2_xattr_handle *h,
                                int (*func)(char *name, char *value,
                                            size_t value_len,

diff --git a/lib/ext2fs/ext_attr.c b/lib/ext2fs/ext_attr.c
index c718ef3..00b3bbf 100644 (file)
--- a/lib/ext2fs/ext_attr.c
+++ b/lib/ext2fs/ext_attr.c
@@ -1024,9 +1024,11 @@ static void xattrs_free_keys(struct ext2_xattr_handle *h)
        h->ibody_count = 0;
 }
 
-errcode_t ext2fs_xattrs_read(struct ext2_xattr_handle *handle)
+/* fetch xattrs from an already-loaded inode */
+errcode_t ext2fs_xattrs_read_inode(struct ext2_xattr_handle *handle,
+                                  struct ext2_inode_large *inode)
 {
-       struct ext2_inode_large *inode;
+
        struct ext2_ext_attr_header *header;
        __u32 ea_inode_magic;
        unsigned int storage_size;
@@ -1036,18 +1038,6 @@ errcode_t ext2fs_xattrs_read(struct ext2_xattr_handle *handle)
        errcode_t err;
 
        EXT2_CHECK_MAGIC(handle, EXT2_ET_MAGIC_EA_HANDLE);
-       i = EXT2_INODE_SIZE(handle->fs->super);
-       if (i < sizeof(*inode))
-               i = sizeof(*inode);
-       err = ext2fs_get_memzero(i, &inode);
-       if (err)
-               return err;
-
-       err = ext2fs_read_inode_full(handle->fs, handle->ino,
-                                    (struct ext2_inode *)inode,
-                                    EXT2_INODE_SIZE(handle->fs->super));
-       if (err)
-               goto out;
 
        xattrs_free_keys(handle);
 
@@ -1083,7 +1073,7 @@ errcode_t ext2fs_xattrs_read(struct ext2_xattr_handle *handle)
 
 read_ea_block:
        /* Look for EA in a separate EA block */
-       blk = ext2fs_file_acl_block(handle->fs, (struct ext2_inode *)inode);
+       blk = ext2fs_file_acl_block(handle->fs, EXT2_INODE(inode));
        if (blk != 0) {
                if ((blk < handle->fs->super->s_first_data_block) ||
                    (blk >= ext2fs_blocks_count(handle->fs->super))) {
@@ -1114,20 +1104,39 @@ read_ea_block:
                err = read_xattrs_from_buffer(handle, inode,
                                        (struct ext2_ext_attr_entry *) start,
                                        storage_size, block_buf);
-               if (err)
-                       goto out3;
+       }
 
+out3:
+       if (block_buf)
                ext2fs_free_mem(&block_buf);
-       }
+out:
+       return err;
+}
 
-       ext2fs_free_mem(&block_buf);
-       ext2fs_free_mem(&inode);
-       return 0;
+errcode_t ext2fs_xattrs_read(struct ext2_xattr_handle *handle)
+{
+       struct ext2_inode_large *inode;
+       size_t inode_size = EXT2_INODE_SIZE(handle->fs->super);
+       errcode_t err;
+
+       EXT2_CHECK_MAGIC(handle, EXT2_ET_MAGIC_EA_HANDLE);
+
+       if (inode_size < sizeof(*inode))
+               inode_size = sizeof(*inode);
+       err = ext2fs_get_memzero(inode_size, &inode);
+       if (err)
+               return err;
+
+       err = ext2fs_read_inode_full(handle->fs, handle->ino, EXT2_INODE(inode),
+                                    EXT2_INODE_SIZE(handle->fs->super));
+       if (err)
+               goto out;
+
+       err = ext2fs_xattrs_read_inode(handle, inode);
 
-out3:
-       ext2fs_free_mem(&block_buf);
 out:
        ext2fs_free_mem(&inode);
+
        return err;
 }
 

diff --git a/tests/f_trusted_link/expect.1 b/tests/f_trusted_link/expect.1
new file mode 100644 (file)
index 0000000..1063d22
--- /dev/null
+++ b/tests/f_trusted_link/expect.1
@@ -0,0 +1,16 @@
+Pass 1: Checking inodes, blocks, and sizes
+Pass 2: Checking directory structure
+Pass 3: Checking directory connectivity
+Pass 4: Checking reference counts
+Inode 15 ref count is 7, should be 6.  Fix? yes
+
+Inode 16 ref count is 3, should be 6.  Fix? yes
+
+Unattached inode 18
+Connect to /lost+found? yes
+
+Pass 5: Checking group summary information
+
+test_filesys: ***** FILE SYSTEM WAS MODIFIED *****
+test_filesys: 18/4096 files (0.0% non-contiguous), 6424/16384 blocks
+Exit status is 1

diff --git a/tests/f_trusted_link/expect.2 b/tests/f_trusted_link/expect.2
new file mode 100644 (file)
index 0000000..a8ff12a
--- /dev/null
+++ b/tests/f_trusted_link/expect.2
@@ -0,0 +1,7 @@
+Pass 1: Checking inodes, blocks, and sizes
+Pass 2: Checking directory structure
+Pass 3: Checking directory connectivity
+Pass 4: Checking reference counts
+Pass 5: Checking group summary information
+test_filesys: 18/4096 files (0.0% non-contiguous), 6424/16384 blocks
+Exit status is 0

diff --git a/tests/f_trusted_link/image.gz b/tests/f_trusted_link/image.gz
new file mode 100644 (file)
index 0000000..8e25873
Binary files /dev/null and b/tests/f_trusted_link/image.gz differ

diff --git a/tests/f_trusted_link/name b/tests/f_trusted_link/name
new file mode 100644 (file)
index 0000000..a6e424b
--- /dev/null
+++ b/tests/f_trusted_link/name
@@ -0,0 +1 @@
+inode nlink according to trusted.link xattr