]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
ext4: support GETFSMAP ioctls
authorDarrick J. Wong <darrick.wong@oracle.com>
Sun, 30 Apr 2017 04:36:53 +0000 (00:36 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Sun, 30 Apr 2017 04:36:53 +0000 (00:36 -0400)
Support the GETFSMAP ioctls so that we can use the xfs free space
management tools to probe ext4 as well.  Note that this is a partial
implementation -- we only report fixed-location metadata and free space;
everything else is reported as "unknown".

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
fs/ext4/Makefile
fs/ext4/fsmap.c [new file with mode: 0644]
fs/ext4/fsmap.h [new file with mode: 0644]
fs/ext4/ioctl.c
fs/ext4/mballoc.c
fs/ext4/mballoc.h
fs/ext4/super.c
include/trace/events/ext4.h

index 354103f3490c3cba719b64a8d0f297200fde67e6..d9beca1653c597bcc9f48835e766332124f07cc3 100644 (file)
@@ -4,11 +4,11 @@
 
 obj-$(CONFIG_EXT4_FS) += ext4.o
 
-ext4-y := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o page-io.o \
-               ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
-               ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
-               mmp.o indirect.o extents_status.o xattr.o xattr_user.o \
-               xattr_trusted.o inline.o readpage.o sysfs.o
+ext4-y := balloc.o bitmap.o block_validity.o dir.o ext4_jbd2.o extents.o \
+               extents_status.o file.o fsmap.o fsync.o hash.o ialloc.o \
+               indirect.o inline.o inode.o ioctl.o mballoc.o migrate.o \
+               mmp.o move_extent.o namei.o page-io.o readpage.o resize.o \
+               super.o symlink.o sysfs.o xattr.o xattr_trusted.o xattr_user.o
 
 ext4-$(CONFIG_EXT4_FS_POSIX_ACL)       += acl.o
 ext4-$(CONFIG_EXT4_FS_SECURITY)                += xattr_security.o
diff --git a/fs/ext4/fsmap.c b/fs/ext4/fsmap.c
new file mode 100644 (file)
index 0000000..b194360
--- /dev/null
@@ -0,0 +1,722 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "ext4.h"
+#include <linux/fsmap.h>
+#include "fsmap.h"
+#include "mballoc.h"
+#include <linux/sort.h>
+#include <linux/list_sort.h>
+#include <trace/events/ext4.h>
+
+/* Convert an ext4_fsmap to an fsmap. */
+void ext4_fsmap_from_internal(struct super_block *sb, struct fsmap *dest,
+                             struct ext4_fsmap *src)
+{
+       dest->fmr_device = src->fmr_device;
+       dest->fmr_flags = src->fmr_flags;
+       dest->fmr_physical = src->fmr_physical << sb->s_blocksize_bits;
+       dest->fmr_owner = src->fmr_owner;
+       dest->fmr_offset = 0;
+       dest->fmr_length = src->fmr_length << sb->s_blocksize_bits;
+       dest->fmr_reserved[0] = 0;
+       dest->fmr_reserved[1] = 0;
+       dest->fmr_reserved[2] = 0;
+}
+
+/* Convert an fsmap to an ext4_fsmap. */
+void ext4_fsmap_to_internal(struct super_block *sb, struct ext4_fsmap *dest,
+                           struct fsmap *src)
+{
+       dest->fmr_device = src->fmr_device;
+       dest->fmr_flags = src->fmr_flags;
+       dest->fmr_physical = src->fmr_physical >> sb->s_blocksize_bits;
+       dest->fmr_owner = src->fmr_owner;
+       dest->fmr_length = src->fmr_length >> sb->s_blocksize_bits;
+}
+
+/* getfsmap query state */
+struct ext4_getfsmap_info {
+       struct ext4_fsmap_head  *gfi_head;
+       ext4_fsmap_format_t     gfi_formatter;  /* formatting fn */
+       void                    *gfi_format_arg;/* format buffer */
+       ext4_fsblk_t            gfi_next_fsblk; /* next fsblock we expect */
+       u32                     gfi_dev;        /* device id */
+       ext4_group_t            gfi_agno;       /* bg number, if applicable */
+       struct ext4_fsmap       gfi_low;        /* low rmap key */
+       struct ext4_fsmap       gfi_high;       /* high rmap key */
+       struct ext4_fsmap       gfi_lastfree;   /* free ext at end of last bg */
+       struct list_head        gfi_meta_list;  /* fixed metadata list */
+       bool                    gfi_last;       /* last extent? */
+};
+
+/* Associate a device with a getfsmap handler. */
+struct ext4_getfsmap_dev {
+       int                     (*gfd_fn)(struct super_block *sb,
+                                     struct ext4_fsmap *keys,
+                                     struct ext4_getfsmap_info *info);
+       u32                     gfd_dev;
+};
+
+/* Compare two getfsmap device handlers. */
+static int ext4_getfsmap_dev_compare(const void *p1, const void *p2)
+{
+       const struct ext4_getfsmap_dev *d1 = p1;
+       const struct ext4_getfsmap_dev *d2 = p2;
+
+       return d1->gfd_dev - d2->gfd_dev;
+}
+
+/* Compare a record against our starting point */
+static bool ext4_getfsmap_rec_before_low_key(struct ext4_getfsmap_info *info,
+                                            struct ext4_fsmap *rec)
+{
+       return rec->fmr_physical < info->gfi_low.fmr_physical;
+}
+
+/*
+ * Format a reverse mapping for getfsmap, having translated rm_startblock
+ * into the appropriate daddr units.
+ */
+static int ext4_getfsmap_helper(struct super_block *sb,
+                               struct ext4_getfsmap_info *info,
+                               struct ext4_fsmap *rec)
+{
+       struct ext4_fsmap fmr;
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_fsblk_t rec_fsblk = rec->fmr_physical;
+       ext4_group_t agno;
+       ext4_grpblk_t cno;
+       int error;
+
+       if (fatal_signal_pending(current))
+               return -EINTR;
+
+       /*
+        * Filter out records that start before our startpoint, if the
+        * caller requested that.
+        */
+       if (ext4_getfsmap_rec_before_low_key(info, rec)) {
+               rec_fsblk += rec->fmr_length;
+               if (info->gfi_next_fsblk < rec_fsblk)
+                       info->gfi_next_fsblk = rec_fsblk;
+               return EXT4_QUERY_RANGE_CONTINUE;
+       }
+
+       /* Are we just counting mappings? */
+       if (info->gfi_head->fmh_count == 0) {
+               if (rec_fsblk > info->gfi_next_fsblk)
+                       info->gfi_head->fmh_entries++;
+
+               if (info->gfi_last)
+                       return EXT4_QUERY_RANGE_CONTINUE;
+
+               info->gfi_head->fmh_entries++;
+
+               rec_fsblk += rec->fmr_length;
+               if (info->gfi_next_fsblk < rec_fsblk)
+                       info->gfi_next_fsblk = rec_fsblk;
+               return EXT4_QUERY_RANGE_CONTINUE;
+       }
+
+       /*
+        * If the record starts past the last physical block we saw,
+        * then we've found a gap.  Report the gap as being owned by
+        * whatever the caller specified is the missing owner.
+        */
+       if (rec_fsblk > info->gfi_next_fsblk) {
+               if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
+                       return EXT4_QUERY_RANGE_ABORT;
+
+               ext4_get_group_no_and_offset(sb, info->gfi_next_fsblk,
+                               &agno, &cno);
+               trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno,
+                               EXT4_C2B(sbi, cno),
+                               rec_fsblk - info->gfi_next_fsblk,
+                               EXT4_FMR_OWN_UNKNOWN);
+
+               fmr.fmr_device = info->gfi_dev;
+               fmr.fmr_physical = info->gfi_next_fsblk;
+               fmr.fmr_owner = EXT4_FMR_OWN_UNKNOWN;
+               fmr.fmr_length = rec_fsblk - info->gfi_next_fsblk;
+               fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
+               error = info->gfi_formatter(&fmr, info->gfi_format_arg);
+               if (error)
+                       return error;
+               info->gfi_head->fmh_entries++;
+       }
+
+       if (info->gfi_last)
+               goto out;
+
+       /* Fill out the extent we found */
+       if (info->gfi_head->fmh_entries >= info->gfi_head->fmh_count)
+               return EXT4_QUERY_RANGE_ABORT;
+
+       ext4_get_group_no_and_offset(sb, rec_fsblk, &agno, &cno);
+       trace_ext4_fsmap_mapping(sb, info->gfi_dev, agno, EXT4_C2B(sbi, cno),
+                       rec->fmr_length, rec->fmr_owner);
+
+       fmr.fmr_device = info->gfi_dev;
+       fmr.fmr_physical = rec_fsblk;
+       fmr.fmr_owner = rec->fmr_owner;
+       fmr.fmr_flags = FMR_OF_SPECIAL_OWNER;
+       fmr.fmr_length = rec->fmr_length;
+       error = info->gfi_formatter(&fmr, info->gfi_format_arg);
+       if (error)
+               return error;
+       info->gfi_head->fmh_entries++;
+
+out:
+       rec_fsblk += rec->fmr_length;
+       if (info->gfi_next_fsblk < rec_fsblk)
+               info->gfi_next_fsblk = rec_fsblk;
+       return EXT4_QUERY_RANGE_CONTINUE;
+}
+
+static inline ext4_fsblk_t ext4_fsmap_next_pblk(struct ext4_fsmap *fmr)
+{
+       return fmr->fmr_physical + fmr->fmr_length;
+}
+
+/* Transform a blockgroup's free record into a fsmap */
+static int ext4_getfsmap_datadev_helper(struct super_block *sb,
+                                       ext4_group_t agno, ext4_grpblk_t start,
+                                       ext4_grpblk_t len, void *priv)
+{
+       struct ext4_fsmap irec;
+       struct ext4_getfsmap_info *info = priv;
+       struct ext4_fsmap *p;
+       struct ext4_fsmap *tmp;
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_fsblk_t fsb;
+       ext4_fsblk_t fslen;
+       int error;
+
+       fsb = (EXT4_C2B(sbi, start) + ext4_group_first_block_no(sb, agno));
+       fslen = EXT4_C2B(sbi, len);
+
+       /* If the retained free extent record is set... */
+       if (info->gfi_lastfree.fmr_owner) {
+               /* ...and abuts this one, lengthen it and return. */
+               if (ext4_fsmap_next_pblk(&info->gfi_lastfree) == fsb) {
+                       info->gfi_lastfree.fmr_length += fslen;
+                       return 0;
+               }
+
+               /*
+                * There's a gap between the two free extents; emit the
+                * retained extent prior to merging the meta_list.
+                */
+               error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
+               if (error)
+                       return error;
+               info->gfi_lastfree.fmr_owner = 0;
+       }
+
+       /* Merge in any relevant extents from the meta_list */
+       list_for_each_entry_safe(p, tmp, &info->gfi_meta_list, fmr_list) {
+               if (p->fmr_physical + p->fmr_length <= info->gfi_next_fsblk) {
+                       list_del(&p->fmr_list);
+                       kfree(p);
+               } else if (p->fmr_physical < fsb) {
+                       error = ext4_getfsmap_helper(sb, info, p);
+                       if (error)
+                               return error;
+
+                       list_del(&p->fmr_list);
+                       kfree(p);
+               }
+       }
+
+       irec.fmr_device = 0;
+       irec.fmr_physical = fsb;
+       irec.fmr_length = fslen;
+       irec.fmr_owner = EXT4_FMR_OWN_FREE;
+       irec.fmr_flags = 0;
+
+       /* If this is a free extent at the end of a bg, buffer it. */
+       if (ext4_fsmap_next_pblk(&irec) ==
+                       ext4_group_first_block_no(sb, agno + 1)) {
+               info->gfi_lastfree = irec;
+               return 0;
+       }
+
+       /* Otherwise, emit it */
+       return ext4_getfsmap_helper(sb, info, &irec);
+}
+
+/* Execute a getfsmap query against the log device. */
+static int ext4_getfsmap_logdev(struct super_block *sb, struct ext4_fsmap *keys,
+                               struct ext4_getfsmap_info *info)
+{
+       journal_t *journal = EXT4_SB(sb)->s_journal;
+       struct ext4_fsmap irec;
+
+       /* Set up search keys */
+       info->gfi_low = keys[0];
+       info->gfi_low.fmr_length = 0;
+
+       memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
+
+       trace_ext4_fsmap_low_key(sb, info->gfi_dev, 0,
+                       info->gfi_low.fmr_physical,
+                       info->gfi_low.fmr_length,
+                       info->gfi_low.fmr_owner);
+
+       trace_ext4_fsmap_high_key(sb, info->gfi_dev, 0,
+                       info->gfi_high.fmr_physical,
+                       info->gfi_high.fmr_length,
+                       info->gfi_high.fmr_owner);
+
+       if (keys[0].fmr_physical > 0)
+               return 0;
+
+       /* Fabricate an rmap entry for the external log device. */
+       irec.fmr_physical = journal->j_blk_offset;
+       irec.fmr_length = journal->j_maxlen;
+       irec.fmr_owner = EXT4_FMR_OWN_LOG;
+       irec.fmr_flags = 0;
+
+       return ext4_getfsmap_helper(sb, info, &irec);
+}
+
+/* Helper to fill out an ext4_fsmap. */
+static inline int ext4_getfsmap_fill(struct list_head *meta_list,
+                                    ext4_fsblk_t fsb, ext4_fsblk_t len,
+                                    uint64_t owner)
+{
+       struct ext4_fsmap *fsm;
+
+       fsm = kmalloc(sizeof(*fsm), GFP_NOFS);
+       if (!fsm)
+               return -ENOMEM;
+       fsm->fmr_device = 0;
+       fsm->fmr_flags = 0;
+       fsm->fmr_physical = fsb;
+       fsm->fmr_owner = owner;
+       fsm->fmr_length = len;
+       list_add_tail(&fsm->fmr_list, meta_list);
+
+       return 0;
+}
+
+/*
+ * This function returns the number of file system metadata blocks at
+ * the beginning of a block group, including the reserved gdt blocks.
+ */
+static unsigned int ext4_getfsmap_find_sb(struct super_block *sb,
+                                         ext4_group_t agno,
+                                         struct list_head *meta_list)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_fsblk_t fsb = ext4_group_first_block_no(sb, agno);
+       ext4_fsblk_t len;
+       unsigned long first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
+       unsigned long metagroup = agno / EXT4_DESC_PER_BLOCK(sb);
+       int error;
+
+       /* Record the superblock. */
+       if (ext4_bg_has_super(sb, agno)) {
+               error = ext4_getfsmap_fill(meta_list, fsb, 1, EXT4_FMR_OWN_FS);
+               if (error)
+                       return error;
+               fsb++;
+       }
+
+       /* Record the group descriptors. */
+       len = ext4_bg_num_gdb(sb, agno);
+       if (!len)
+               return 0;
+       error = ext4_getfsmap_fill(meta_list, fsb, len,
+                                  EXT4_FMR_OWN_GDT);
+       if (error)
+               return error;
+       fsb += len;
+
+       /* Reserved GDT blocks */
+       if (!ext4_has_feature_meta_bg(sb) || metagroup < first_meta_bg) {
+               len = le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks);
+               error = ext4_getfsmap_fill(meta_list, fsb, len,
+                                          EXT4_FMR_OWN_RESV_GDT);
+               if (error)
+                       return error;
+       }
+
+       return 0;
+}
+
+/* Compare two fsmap items. */
+static int ext4_getfsmap_compare(void *priv,
+                                struct list_head *a,
+                                struct list_head *b)
+{
+       struct ext4_fsmap *fa;
+       struct ext4_fsmap *fb;
+
+       fa = container_of(a, struct ext4_fsmap, fmr_list);
+       fb = container_of(b, struct ext4_fsmap, fmr_list);
+       if (fa->fmr_physical < fb->fmr_physical)
+               return -1;
+       else if (fa->fmr_physical > fb->fmr_physical)
+               return 1;
+       return 0;
+}
+
+/* Merge adjacent extents of fixed metadata. */
+static void ext4_getfsmap_merge_fixed_metadata(struct list_head *meta_list)
+{
+       struct ext4_fsmap *p;
+       struct ext4_fsmap *prev = NULL;
+       struct ext4_fsmap *tmp;
+
+       list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
+               if (!prev) {
+                       prev = p;
+                       continue;
+               }
+
+               if (prev->fmr_owner == p->fmr_owner &&
+                   prev->fmr_physical + prev->fmr_length == p->fmr_physical) {
+                       prev->fmr_length += p->fmr_length;
+                       list_del(&p->fmr_list);
+                       kfree(p);
+               } else
+                       prev = p;
+       }
+}
+
+/* Free a list of fixed metadata. */
+static void ext4_getfsmap_free_fixed_metadata(struct list_head *meta_list)
+{
+       struct ext4_fsmap *p;
+       struct ext4_fsmap *tmp;
+
+       list_for_each_entry_safe(p, tmp, meta_list, fmr_list) {
+               list_del(&p->fmr_list);
+               kfree(p);
+       }
+}
+
+/* Find all the fixed metadata in the filesystem. */
+int ext4_getfsmap_find_fixed_metadata(struct super_block *sb,
+                                     struct list_head *meta_list)
+{
+       struct ext4_group_desc *gdp;
+       ext4_group_t agno;
+       int error;
+
+       INIT_LIST_HEAD(meta_list);
+
+       /* Collect everything. */
+       for (agno = 0; agno < EXT4_SB(sb)->s_groups_count; agno++) {
+               gdp = ext4_get_group_desc(sb, agno, NULL);
+               if (!gdp) {
+                       error = -EFSCORRUPTED;
+                       goto err;
+               }
+
+               /* Superblock & GDT */
+               error = ext4_getfsmap_find_sb(sb, agno, meta_list);
+               if (error)
+                       goto err;
+
+               /* Block bitmap */
+               error = ext4_getfsmap_fill(meta_list,
+                                          ext4_block_bitmap(sb, gdp), 1,
+                                          EXT4_FMR_OWN_BLKBM);
+               if (error)
+                       goto err;
+
+               /* Inode bitmap */
+               error = ext4_getfsmap_fill(meta_list,
+                                          ext4_inode_bitmap(sb, gdp), 1,
+                                          EXT4_FMR_OWN_INOBM);
+               if (error)
+                       goto err;
+
+               /* Inodes */
+               error = ext4_getfsmap_fill(meta_list,
+                                          ext4_inode_table(sb, gdp),
+                                          EXT4_SB(sb)->s_itb_per_group,
+                                          EXT4_FMR_OWN_INODES);
+               if (error)
+                       goto err;
+       }
+
+       /* Sort the list */
+       list_sort(NULL, meta_list, ext4_getfsmap_compare);
+
+       /* Merge adjacent extents */
+       ext4_getfsmap_merge_fixed_metadata(meta_list);
+
+       return 0;
+err:
+       ext4_getfsmap_free_fixed_metadata(meta_list);
+       return error;
+}
+
+/* Execute a getfsmap query against the buddy bitmaps */
+static int ext4_getfsmap_datadev(struct super_block *sb,
+                                struct ext4_fsmap *keys,
+                                struct ext4_getfsmap_info *info)
+{
+       struct ext4_sb_info *sbi = EXT4_SB(sb);
+       ext4_fsblk_t start_fsb;
+       ext4_fsblk_t end_fsb;
+       ext4_fsblk_t eofs;
+       ext4_group_t start_ag;
+       ext4_group_t end_ag;
+       ext4_grpblk_t first_cluster;
+       ext4_grpblk_t last_cluster;
+       int error = 0;
+
+       eofs = ext4_blocks_count(sbi->s_es);
+       if (keys[0].fmr_physical >= eofs)
+               return 0;
+       if (keys[1].fmr_physical >= eofs)
+               keys[1].fmr_physical = eofs - 1;
+       start_fsb = keys[0].fmr_physical;
+       end_fsb = keys[1].fmr_physical;
+
+       /* Determine first and last group to examine based on start and end */
+       ext4_get_group_no_and_offset(sb, start_fsb, &start_ag, &first_cluster);
+       ext4_get_group_no_and_offset(sb, end_fsb, &end_ag, &last_cluster);
+
+       /*
+        * Convert the fsmap low/high keys to bg based keys.  Initialize
+        * low to the fsmap low key and max out the high key to the end
+        * of the bg.
+        */
+       info->gfi_low = keys[0];
+       info->gfi_low.fmr_physical = EXT4_C2B(sbi, first_cluster);
+       info->gfi_low.fmr_length = 0;
+
+       memset(&info->gfi_high, 0xFF, sizeof(info->gfi_high));
+
+       /* Assemble a list of all the fixed-location metadata. */
+       error = ext4_getfsmap_find_fixed_metadata(sb, &info->gfi_meta_list);
+       if (error)
+               goto err;
+
+       /* Query each bg */
+       for (info->gfi_agno = start_ag;
+            info->gfi_agno <= end_ag;
+            info->gfi_agno++) {
+               /*
+                * Set the bg high key from the fsmap high key if this
+                * is the last bg that we're querying.
+                */
+               if (info->gfi_agno == end_ag) {
+                       info->gfi_high = keys[1];
+                       info->gfi_high.fmr_physical = EXT4_C2B(sbi,
+                                       last_cluster);
+                       info->gfi_high.fmr_length = 0;
+               }
+
+               trace_ext4_fsmap_low_key(sb, info->gfi_dev, info->gfi_agno,
+                               info->gfi_low.fmr_physical,
+                               info->gfi_low.fmr_length,
+                               info->gfi_low.fmr_owner);
+
+               trace_ext4_fsmap_high_key(sb, info->gfi_dev, info->gfi_agno,
+                               info->gfi_high.fmr_physical,
+                               info->gfi_high.fmr_length,
+                               info->gfi_high.fmr_owner);
+
+               error = ext4_mballoc_query_range(sb, info->gfi_agno,
+                               EXT4_B2C(sbi, info->gfi_low.fmr_physical),
+                               EXT4_B2C(sbi, info->gfi_high.fmr_physical),
+                               ext4_getfsmap_datadev_helper, info);
+               if (error)
+                       goto err;
+
+               /*
+                * Set the bg low key to the start of the bg prior to
+                * moving on to the next bg.
+                */
+               if (info->gfi_agno == start_ag)
+                       memset(&info->gfi_low, 0, sizeof(info->gfi_low));
+       }
+
+       /* Do we have a retained free extent? */
+       if (info->gfi_lastfree.fmr_owner) {
+               error = ext4_getfsmap_helper(sb, info, &info->gfi_lastfree);
+               if (error)
+                       goto err;
+       }
+
+       /* Report any gaps at the end of the bg */
+       info->gfi_last = true;
+       error = ext4_getfsmap_datadev_helper(sb, end_ag, last_cluster, 0, info);
+       if (error)
+               goto err;
+
+err:
+       ext4_getfsmap_free_fixed_metadata(&info->gfi_meta_list);
+       return error;
+}
+
+/* Do we recognize the device? */
+static bool ext4_getfsmap_is_valid_device(struct super_block *sb,
+                                         struct ext4_fsmap *fm)
+{
+       if (fm->fmr_device == 0 || fm->fmr_device == UINT_MAX ||
+           fm->fmr_device == new_encode_dev(sb->s_bdev->bd_dev))
+               return true;
+       if (EXT4_SB(sb)->journal_bdev &&
+           fm->fmr_device == new_encode_dev(EXT4_SB(sb)->journal_bdev->bd_dev))
+               return true;
+       return false;
+}
+
+/* Ensure that the low key is less than the high key. */
+static bool ext4_getfsmap_check_keys(struct ext4_fsmap *low_key,
+                                    struct ext4_fsmap *high_key)
+{
+       if (low_key->fmr_device > high_key->fmr_device)
+               return false;
+       if (low_key->fmr_device < high_key->fmr_device)
+               return true;
+
+       if (low_key->fmr_physical > high_key->fmr_physical)
+               return false;
+       if (low_key->fmr_physical < high_key->fmr_physical)
+               return true;
+
+       if (low_key->fmr_owner > high_key->fmr_owner)
+               return false;
+       if (low_key->fmr_owner < high_key->fmr_owner)
+               return true;
+
+       return false;
+}
+
+#define EXT4_GETFSMAP_DEVS     2
+/*
+ * Get filesystem's extents as described in head, and format for
+ * output.  Calls formatter to fill the user's buffer until all
+ * extents are mapped, until the passed-in head->fmh_count slots have
+ * been filled, or until the formatter short-circuits the loop, if it
+ * is tracking filled-in extents on its own.
+ *
+ * Key to Confusion
+ * ----------------
+ * There are multiple levels of keys and counters at work here:
+ * _fsmap_head.fmh_keys                -- low and high fsmap keys passed in;
+ *                                these reflect fs-wide block addrs.
+ * dkeys                       -- fmh_keys used to query each device;
+ *                                these are fmh_keys but w/ the low key
+ *                                bumped up by fmr_length.
+ * _getfsmap_info.gfi_next_fsblk-- next fs block we expect to see; this
+ *                                is how we detect gaps in the fsmap
+ *                                records and report them.
+ * _getfsmap_info.gfi_low/high -- per-bg low/high keys computed from
+ *                                dkeys; used to query the free space.
+ */
+int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
+                 ext4_fsmap_format_t formatter, void *arg)
+{
+       struct ext4_fsmap dkeys[2];     /* per-dev keys */
+       struct ext4_getfsmap_dev handlers[EXT4_GETFSMAP_DEVS];
+       struct ext4_getfsmap_info info = {0};
+       int i;
+       int error = 0;
+
+       if (head->fmh_iflags & ~FMH_IF_VALID)
+               return -EINVAL;
+       if (!ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[0]) ||
+           !ext4_getfsmap_is_valid_device(sb, &head->fmh_keys[1]))
+               return -EINVAL;
+
+       head->fmh_entries = 0;
+
+       /* Set up our device handlers. */
+       memset(handlers, 0, sizeof(handlers));
+       handlers[0].gfd_dev = new_encode_dev(sb->s_bdev->bd_dev);
+       handlers[0].gfd_fn = ext4_getfsmap_datadev;
+       if (EXT4_SB(sb)->journal_bdev) {
+               handlers[1].gfd_dev = new_encode_dev(
+                               EXT4_SB(sb)->journal_bdev->bd_dev);
+               handlers[1].gfd_fn = ext4_getfsmap_logdev;
+       }
+
+       sort(handlers, EXT4_GETFSMAP_DEVS, sizeof(struct ext4_getfsmap_dev),
+                       ext4_getfsmap_dev_compare, NULL);
+
+       /*
+        * To continue where we left off, we allow userspace to use the
+        * last mapping from a previous call as the low key of the next.
+        * This is identified by a non-zero length in the low key. We
+        * have to increment the low key in this scenario to ensure we
+        * don't return the same mapping again, and instead return the
+        * very next mapping.
+        *
+        * Bump the physical offset as there can be no other mapping for
+        * the same physical block range.
+        */
+       dkeys[0] = head->fmh_keys[0];
+       dkeys[0].fmr_physical += dkeys[0].fmr_length;
+       dkeys[0].fmr_owner = 0;
+       dkeys[0].fmr_length = 0;
+       memset(&dkeys[1], 0xFF, sizeof(struct ext4_fsmap));
+
+       if (!ext4_getfsmap_check_keys(dkeys, &head->fmh_keys[1]))
+               return -EINVAL;
+
+       info.gfi_next_fsblk = head->fmh_keys[0].fmr_physical +
+                         head->fmh_keys[0].fmr_length;
+       info.gfi_formatter = formatter;
+       info.gfi_format_arg = arg;
+       info.gfi_head = head;
+
+       /* For each device we support... */
+       for (i = 0; i < EXT4_GETFSMAP_DEVS; i++) {
+               /* Is this device within the range the user asked for? */
+               if (!handlers[i].gfd_fn)
+                       continue;
+               if (head->fmh_keys[0].fmr_device > handlers[i].gfd_dev)
+                       continue;
+               if (head->fmh_keys[1].fmr_device < handlers[i].gfd_dev)
+                       break;
+
+               /*
+                * If this device number matches the high key, we have
+                * to pass the high key to the handler to limit the
+                * query results.  If the device number exceeds the
+                * low key, zero out the low key so that we get
+                * everything from the beginning.
+                */
+               if (handlers[i].gfd_dev == head->fmh_keys[1].fmr_device)
+                       dkeys[1] = head->fmh_keys[1];
+               if (handlers[i].gfd_dev > head->fmh_keys[0].fmr_device)
+                       memset(&dkeys[0], 0, sizeof(struct ext4_fsmap));
+
+               info.gfi_dev = handlers[i].gfd_dev;
+               info.gfi_last = false;
+               info.gfi_agno = -1;
+               error = handlers[i].gfd_fn(sb, dkeys, &info);
+               if (error)
+                       break;
+               info.gfi_next_fsblk = 0;
+       }
+
+       head->fmh_oflags = FMH_OF_DEV_T;
+       return error;
+}
diff --git a/fs/ext4/fsmap.h b/fs/ext4/fsmap.h
new file mode 100644 (file)
index 0000000..9a2cd36
--- /dev/null
@@ -0,0 +1,69 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __EXT4_FSMAP_H__
+#define        __EXT4_FSMAP_H__
+
+struct fsmap;
+
+/* internal fsmap representation */
+struct ext4_fsmap {
+       struct list_head        fmr_list;
+       dev_t           fmr_device;     /* device id */
+       uint32_t        fmr_flags;      /* mapping flags */
+       uint64_t        fmr_physical;   /* device offset of segment */
+       uint64_t        fmr_owner;      /* owner id */
+       uint64_t        fmr_length;     /* length of segment, blocks */
+};
+
+struct ext4_fsmap_head {
+       uint32_t        fmh_iflags;     /* control flags */
+       uint32_t        fmh_oflags;     /* output flags */
+       unsigned int    fmh_count;      /* # of entries in array incl. input */
+       unsigned int    fmh_entries;    /* # of entries filled in (output). */
+
+       struct ext4_fsmap fmh_keys[2];  /* low and high keys */
+};
+
+void ext4_fsmap_from_internal(struct super_block *sb, struct fsmap *dest,
+               struct ext4_fsmap *src);
+void ext4_fsmap_to_internal(struct super_block *sb, struct ext4_fsmap *dest,
+               struct fsmap *src);
+
+/* fsmap to userspace formatter - copy to user & advance pointer */
+typedef int (*ext4_fsmap_format_t)(struct ext4_fsmap *, void *);
+
+int ext4_getfsmap(struct super_block *sb, struct ext4_fsmap_head *head,
+               ext4_fsmap_format_t formatter, void *arg);
+
+#define EXT4_QUERY_RANGE_ABORT         1
+#define EXT4_QUERY_RANGE_CONTINUE      0
+
+/*     fmr_owner special values for FS_IOC_GETFSMAP; some share w/ XFS */
+#define EXT4_FMR_OWN_FREE      FMR_OWN_FREE      /* free space */
+#define EXT4_FMR_OWN_UNKNOWN   FMR_OWN_UNKNOWN   /* unknown owner */
+#define EXT4_FMR_OWN_FS                FMR_OWNER('X', 1) /* static fs metadata */
+#define EXT4_FMR_OWN_LOG       FMR_OWNER('X', 2) /* journalling log */
+#define EXT4_FMR_OWN_INODES    FMR_OWNER('X', 5) /* inodes */
+#define EXT4_FMR_OWN_GDT       FMR_OWNER('f', 1) /* group descriptors */
+#define EXT4_FMR_OWN_RESV_GDT  FMR_OWNER('f', 2) /* reserved gdt blocks */
+#define EXT4_FMR_OWN_BLKBM     FMR_OWNER('f', 3) /* inode bitmap */
+#define EXT4_FMR_OWN_INOBM     FMR_OWNER('f', 4) /* block bitmap */
+
+#endif /* __EXT4_FSMAP_H__ */
index a4273ddb992232e1587e33fd7f3a23ea926aaa7e..dfcb815275f1d05e92084e7b535054f2d7bdee75 100644 (file)
@@ -19,6 +19,9 @@
 #include <linux/delay.h>
 #include "ext4_jbd2.h"
 #include "ext4.h"
+#include <linux/fsmap.h>
+#include "fsmap.h"
+#include <trace/events/ext4.h>
 
 /**
  * Swap memory between @a and @b for @len bytes.
@@ -489,6 +492,90 @@ int ext4_shutdown(struct super_block *sb, unsigned long arg)
        return 0;
 }
 
+struct getfsmap_info {
+       struct super_block      *gi_sb;
+       struct fsmap_head __user *gi_data;
+       unsigned int            gi_idx;
+       __u32                   gi_last_flags;
+};
+
+static int ext4_getfsmap_format(struct ext4_fsmap *xfm, void *priv)
+{
+       struct getfsmap_info *info = priv;
+       struct fsmap fm;
+
+       trace_ext4_getfsmap_mapping(info->gi_sb, xfm);
+
+       info->gi_last_flags = xfm->fmr_flags;
+       ext4_fsmap_from_internal(info->gi_sb, &fm, xfm);
+       if (copy_to_user(&info->gi_data->fmh_recs[info->gi_idx++], &fm,
+                       sizeof(struct fsmap)))
+               return -EFAULT;
+
+       return 0;
+}
+
+static int ext4_ioc_getfsmap(struct super_block *sb,
+                            struct fsmap_head __user *arg)
+{
+       struct getfsmap_info info = {0};
+       struct ext4_fsmap_head xhead = {0};
+       struct fsmap_head head;
+       bool aborted = false;
+       int error;
+
+       if (copy_from_user(&head, arg, sizeof(struct fsmap_head)))
+               return -EFAULT;
+       if (memchr_inv(head.fmh_reserved, 0, sizeof(head.fmh_reserved)) ||
+           memchr_inv(head.fmh_keys[0].fmr_reserved, 0,
+                      sizeof(head.fmh_keys[0].fmr_reserved)) ||
+           memchr_inv(head.fmh_keys[1].fmr_reserved, 0,
+                      sizeof(head.fmh_keys[1].fmr_reserved)))
+               return -EINVAL;
+       /*
+        * ext4 doesn't report file extents at all, so the only valid
+        * file offsets are the magic ones (all zeroes or all ones).
+        */
+       if (head.fmh_keys[0].fmr_offset ||
+           (head.fmh_keys[1].fmr_offset != 0 &&
+            head.fmh_keys[1].fmr_offset != -1ULL))
+               return -EINVAL;
+
+       xhead.fmh_iflags = head.fmh_iflags;
+       xhead.fmh_count = head.fmh_count;
+       ext4_fsmap_to_internal(sb, &xhead.fmh_keys[0], &head.fmh_keys[0]);
+       ext4_fsmap_to_internal(sb, &xhead.fmh_keys[1], &head.fmh_keys[1]);
+
+       trace_ext4_getfsmap_low_key(sb, &xhead.fmh_keys[0]);
+       trace_ext4_getfsmap_high_key(sb, &xhead.fmh_keys[1]);
+
+       info.gi_sb = sb;
+       info.gi_data = arg;
+       error = ext4_getfsmap(sb, &xhead, ext4_getfsmap_format, &info);
+       if (error == EXT4_QUERY_RANGE_ABORT) {
+               error = 0;
+               aborted = true;
+       } else if (error)
+               return error;
+
+       /* If we didn't abort, set the "last" flag in the last fmx */
+       if (!aborted && info.gi_idx) {
+               info.gi_last_flags |= FMR_OF_LAST;
+               if (copy_to_user(&info.gi_data->fmh_recs[info.gi_idx - 1].fmr_flags,
+                                &info.gi_last_flags,
+                                sizeof(info.gi_last_flags)))
+                       return -EFAULT;
+       }
+
+       /* copy back header */
+       head.fmh_entries = xhead.fmh_entries;
+       head.fmh_oflags = xhead.fmh_oflags;
+       if (copy_to_user(arg, &head, sizeof(struct fsmap_head)))
+               return -EFAULT;
+
+       return 0;
+}
+
 long ext4_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
 {
        struct inode *inode = file_inode(filp);
@@ -499,6 +586,8 @@ long ext4_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
        ext4_debug("cmd = %u, arg = %lu\n", cmd, arg);
 
        switch (cmd) {
+       case FS_IOC_GETFSMAP:
+               return ext4_ioc_getfsmap(sb, (void __user *)arg);
        case EXT4_IOC_GETFLAGS:
                ext4_get_inode_flags(ei);
                flags = ei->i_flags & EXT4_FL_USER_VISIBLE;
@@ -1009,6 +1098,7 @@ long ext4_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
        case EXT4_IOC_GET_ENCRYPTION_PWSALT:
        case EXT4_IOC_GET_ENCRYPTION_POLICY:
        case EXT4_IOC_SHUTDOWN:
+       case FS_IOC_GETFSMAP:
                break;
        default:
                return -ENOIOCTLCMD;
index dbe51301c2bbe34b8635e12c0e6e34d3b04f1b65..36de58a37653c021793f85ae28a0c4e9066f9601 100644 (file)
@@ -5277,3 +5277,52 @@ out:
        range->len = EXT4_C2B(EXT4_SB(sb), trimmed) << sb->s_blocksize_bits;
        return ret;
 }
+
+/* Iterate all the free extents in the group. */
+int
+ext4_mballoc_query_range(
+       struct super_block              *sb,
+       ext4_group_t                    group,
+       ext4_grpblk_t                   start,
+       ext4_grpblk_t                   end,
+       ext4_mballoc_query_range_fn     formatter,
+       void                            *priv)
+{
+       void                            *bitmap;
+       ext4_grpblk_t                   next;
+       struct ext4_buddy               e4b;
+       int                             error;
+
+       error = ext4_mb_load_buddy(sb, group, &e4b);
+       if (error)
+               return error;
+       bitmap = e4b.bd_bitmap;
+
+       ext4_lock_group(sb, group);
+
+       start = (e4b.bd_info->bb_first_free > start) ?
+               e4b.bd_info->bb_first_free : start;
+       if (end >= EXT4_CLUSTERS_PER_GROUP(sb))
+               end = EXT4_CLUSTERS_PER_GROUP(sb) - 1;
+
+       while (start <= end) {
+               start = mb_find_next_zero_bit(bitmap, end + 1, start);
+               if (start > end)
+                       break;
+               next = mb_find_next_bit(bitmap, end + 1, start);
+
+               ext4_unlock_group(sb, group);
+               error = formatter(sb, group, start, next - start, priv);
+               if (error)
+                       goto out_unload;
+               ext4_lock_group(sb, group);
+
+               start = next + 1;
+       }
+
+       ext4_unlock_group(sb, group);
+out_unload:
+       ext4_mb_unload_buddy(&e4b);
+
+       return error;
+}
index 1aba469f82209fe40d602579a17964346749e024..2bed62084a8c1592e638f311a1a9574057a7e6a6 100644 (file)
@@ -199,4 +199,21 @@ static inline ext4_fsblk_t ext4_grp_offs_to_block(struct super_block *sb,
        return ext4_group_first_block_no(sb, fex->fe_group) +
                (fex->fe_start << EXT4_SB(sb)->s_cluster_bits);
 }
+
+typedef int (*ext4_mballoc_query_range_fn)(
+       struct super_block              *sb,
+       ext4_group_t                    agno,
+       ext4_grpblk_t                   start,
+       ext4_grpblk_t                   len,
+       void                            *priv);
+
+int
+ext4_mballoc_query_range(
+       struct super_block              *sb,
+       ext4_group_t                    agno,
+       ext4_grpblk_t                   start,
+       ext4_grpblk_t                   end,
+       ext4_mballoc_query_range_fn     formatter,
+       void                            *priv);
+
 #endif
index 7b82487401d254212cbb676f5c9a7c826fa5b5d3..9ec8963ba6049a20009b77465f1d607b0238e8f2 100644 (file)
@@ -49,6 +49,7 @@
 #include "xattr.h"
 #include "acl.h"
 #include "mballoc.h"
+#include "fsmap.h"
 
 #define CREATE_TRACE_POINTS
 #include <trace/events/ext4.h>
index 09c71e9aaebf7d7bf83857294b9ec37a13d0920e..dfae175ddebc2340c68e9145d588f57f2f45ebce 100644 (file)
@@ -15,6 +15,7 @@ struct ext4_inode_info;
 struct mpage_da_data;
 struct ext4_map_blocks;
 struct extent_status;
+struct ext4_fsmap;
 
 #define EXT4_I(inode) (container_of(inode, struct ext4_inode_info, vfs_inode))
 
@@ -2529,6 +2530,79 @@ TRACE_EVENT(ext4_es_shrink,
                  __entry->scan_time, __entry->nr_skipped, __entry->retried)
 );
 
+/* fsmap traces */
+DECLARE_EVENT_CLASS(ext4_fsmap_class,
+       TP_PROTO(struct super_block *sb, u32 keydev, u32 agno, u64 bno, u64 len,
+                u64 owner),
+       TP_ARGS(sb, keydev, agno, bno, len, owner),
+       TP_STRUCT__entry(
+               __field(dev_t, dev)
+               __field(dev_t, keydev)
+               __field(u32, agno)
+               __field(u64, bno)
+               __field(u64, len)
+               __field(u64, owner)
+       ),
+       TP_fast_assign(
+               __entry->dev = sb->s_bdev->bd_dev;
+               __entry->keydev = new_decode_dev(keydev);
+               __entry->agno = agno;
+               __entry->bno = bno;
+               __entry->len = len;
+               __entry->owner = owner;
+       ),
+       TP_printk("dev %d:%d keydev %d:%d agno %u bno %llu len %llu owner %lld\n",
+                 MAJOR(__entry->dev), MINOR(__entry->dev),
+                 MAJOR(__entry->keydev), MINOR(__entry->keydev),
+                 __entry->agno,
+                 __entry->bno,
+                 __entry->len,
+                 __entry->owner)
+)
+#define DEFINE_FSMAP_EVENT(name) \
+DEFINE_EVENT(ext4_fsmap_class, name, \
+       TP_PROTO(struct super_block *sb, u32 keydev, u32 agno, u64 bno, u64 len, \
+                u64 owner), \
+       TP_ARGS(sb, keydev, agno, bno, len, owner))
+DEFINE_FSMAP_EVENT(ext4_fsmap_low_key);
+DEFINE_FSMAP_EVENT(ext4_fsmap_high_key);
+DEFINE_FSMAP_EVENT(ext4_fsmap_mapping);
+
+DECLARE_EVENT_CLASS(ext4_getfsmap_class,
+       TP_PROTO(struct super_block *sb, struct ext4_fsmap *fsmap),
+       TP_ARGS(sb, fsmap),
+       TP_STRUCT__entry(
+               __field(dev_t, dev)
+               __field(dev_t, keydev)
+               __field(u64, block)
+               __field(u64, len)
+               __field(u64, owner)
+               __field(u64, flags)
+       ),
+       TP_fast_assign(
+               __entry->dev = sb->s_bdev->bd_dev;
+               __entry->keydev = new_decode_dev(fsmap->fmr_device);
+               __entry->block = fsmap->fmr_physical;
+               __entry->len = fsmap->fmr_length;
+               __entry->owner = fsmap->fmr_owner;
+               __entry->flags = fsmap->fmr_flags;
+       ),
+       TP_printk("dev %d:%d keydev %d:%d block %llu len %llu owner %lld flags 0x%llx\n",
+                 MAJOR(__entry->dev), MINOR(__entry->dev),
+                 MAJOR(__entry->keydev), MINOR(__entry->keydev),
+                 __entry->block,
+                 __entry->len,
+                 __entry->owner,
+                 __entry->flags)
+)
+#define DEFINE_GETFSMAP_EVENT(name) \
+DEFINE_EVENT(ext4_getfsmap_class, name, \
+       TP_PROTO(struct super_block *sb, struct ext4_fsmap *fsmap), \
+       TP_ARGS(sb, fsmap))
+DEFINE_GETFSMAP_EVENT(ext4_getfsmap_low_key);
+DEFINE_GETFSMAP_EVENT(ext4_getfsmap_high_key);
+DEFINE_GETFSMAP_EVENT(ext4_getfsmap_mapping);
+
 #endif /* _TRACE_EXT4_H */
 
 /* This part must be outside protection */