summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--c_src/bcachefs.c1
-rw-r--r--c_src/cmd_dump.c82
-rw-r--r--c_src/cmd_migrate.c116
-rw-r--r--c_src/cmd_super.c8
-rw-r--r--c_src/cmds.h2
-rw-r--r--c_src/libbcachefs.c12
-rw-r--r--c_src/qcow2.c51
-rw-r--r--c_src/qcow2.h4
-rw-r--r--c_src/tools-util.c6
-rw-r--r--c_src/tools-util.h7
-rw-r--r--linux/shrinker.c3
-rw-r--r--src/bcachefs.rs1
12 files changed, 246 insertions, 47 deletions
diff --git a/c_src/bcachefs.c b/c_src/bcachefs.c
index 58fa2549..d91e0475 100644
--- a/c_src/bcachefs.c
+++ b/c_src/bcachefs.c
@@ -96,6 +96,7 @@ void bcachefs_usage(void)
"Debug:\n"
"These commands work on offline, unmounted filesystems\n"
" dump Dump filesystem metadata to a qcow2 image\n"
+ " undump Convert qcow2 metadata dumps to sparse raw files\n"
" list List filesystem metadata in textual form\n"
" list_journal List contents of journal\n"
"\n"
diff --git a/c_src/cmd_dump.c b/c_src/cmd_dump.c
index 5f354272..93e7d743 100644
--- a/c_src/cmd_dump.c
+++ b/c_src/cmd_dump.c
@@ -386,3 +386,85 @@ int cmd_dump(int argc, char *argv[])
darray_exit(&dev_names);
return ret;
}
+
+static void undump_usage(void)
+{
+ puts("bcachefs undump - turn qcow2 images from 'bcachefs dump' back into sparse raw images\n"
+ "Usage: bcachefs undump [OPTION]... <files>\n"
+ "\n"
+ "Options:\n"
+ " -f, --force Force; overwrite when needed\n"
+ " -h, --help Display this help and exit\n"
+ "Report bugs to <linux-bcachefs@vger.kernel.org>");
+}
+
+struct undump {
+ char *in, *out;
+ int infd, outfd;
+};
+
+int cmd_undump(int argc, char *argv[])
+{
+ static const struct option longopts[] = {
+ { "force", no_argument, NULL, 'f' },
+ { "help", no_argument, NULL, 'h' },
+ { NULL }
+ };
+ bool force = false;
+ int opt;
+
+ while ((opt = getopt_long(argc, argv, "fh",
+ longopts, NULL)) != -1)
+ switch (opt) {
+ case 'f':
+ force = true;
+ break;
+ case 'h':
+ undump_usage();
+ exit(EXIT_SUCCESS);
+ }
+ args_shift(optind);
+
+ if (!argc) {
+ undump_usage();
+ die("Please supply file(s) to convert");
+ }
+
+ DARRAY(struct undump) files = {};
+
+ for (unsigned i = 0; i < argc; i++) {
+ unsigned len = strlen(argv[i]);
+ const char *suffix = ".qcow2";
+ unsigned suffixlen = strlen(suffix);
+
+ if (len <= suffixlen ||
+ strcmp(suffix, argv[i] + len - suffixlen)) {
+ die("%s not a qcow2 image?", argv[i]);
+ }
+
+ char *out = strdup(argv[i]);
+ out[len - suffixlen] = '\0';
+
+ if (!force && !access(out, F_OK))
+ die("%s already exists", out);
+
+ darray_push(&files, ((struct undump) {
+ .in = argv[i],
+ .out = out,
+ .infd = xopen(argv[i], O_RDONLY),
+ }));
+ }
+
+ darray_for_each(files, i)
+ i->outfd = xopen(i->out, O_WRONLY|O_CREAT|(!force ? O_EXCL : 0), 0600);
+
+ darray_for_each(files, i) {
+ qcow2_to_raw(i->infd, i->outfd);
+ close(i->infd);
+ close(i->outfd);
+ free(i->out);
+ }
+
+ darray_exit(&files);
+ return 0;
+}
diff --git a/c_src/cmd_migrate.c b/c_src/cmd_migrate.c
index aa17a160..4e4fbcba 100644
--- a/c_src/cmd_migrate.c
+++ b/c_src/cmd_migrate.c
@@ -86,6 +86,15 @@ found:
return ret;
}
+static void mark_nouse_range(struct bch_dev *ca, u64 sector_from, u64 sector_to)
+{
+ u64 b = sector_to_bucket(ca, sector_from);
+ do {
+ set_bit(b, ca->buckets_nouse);
+ b++;
+ } while (bucket_to_sector(ca, b) < sector_to);
+}
+
static void mark_unreserved_space(struct bch_fs *c, ranges extents)
{
struct bch_dev *ca = c->devs[0];
@@ -93,17 +102,16 @@ static void mark_unreserved_space(struct bch_fs *c, ranges extents)
struct range i;
for_each_hole(iter, extents, bucket_to_sector(ca, ca->mi.nbuckets) << 9, i) {
- u64 b;
-
if (i.start == i.end)
return;
- b = sector_to_bucket(ca, i.start >> 9);
- do {
- set_bit(b, ca->buckets_nouse);
- b++;
- } while (bucket_to_sector(ca, b) << 9 < i.end);
+ mark_nouse_range(ca, i.start >> 9,
+ round_up(i.end, 1 << 9) >> 9);
}
+
+ /* Also be sure to mark the space for the default sb layout */
+ unsigned sb_size = 1U << ca->disk_sb.sb->layout.sb_max_size_bits;
+ mark_nouse_range(ca, 0, BCH_SB_SECTOR + sb_size * 2);
}
static ranges reserve_new_fs_space(const char *file_path, unsigned block_size,
@@ -279,7 +287,7 @@ static int migrate_fs(const char *fs_path,
.dev = stat.st_dev,
.extents = extents,
.type = BCH_MIGRATE_migrate,
- .reserve_start = roundup((format_opts.superblock_size * 2 + 8) << 9,
+ .reserve_start = roundup((format_opts.superblock_size * 2 + BCH_SB_SECTOR) << 9,
bucket_bytes(c->devs[0])),
};
@@ -381,6 +389,28 @@ static void migrate_superblock_usage(void)
"Report bugs to <linux-bcachefs@vger.kernel.org>");
}
+static void add_default_sb_layout(struct bch_sb* sb, unsigned *out_sb_size)
+{
+ unsigned sb_size = 1U << sb->layout.sb_max_size_bits;
+ if (out_sb_size)
+ *out_sb_size = sb_size;
+
+ if (sb->layout.nr_superblocks >= ARRAY_SIZE(sb->layout.sb_offset))
+ die("Can't add superblock: no space left in superblock layout");
+
+ for (unsigned i = 0; i < sb->layout.nr_superblocks; i++)
+ if (le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR ||
+ le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR + sb_size)
+ die("Superblock layout already has default superblocks");
+
+ memmove(&sb->layout.sb_offset[2],
+ &sb->layout.sb_offset[0],
+ sb->layout.nr_superblocks * sizeof(u64));
+ sb->layout.nr_superblocks += 2;
+ sb->layout.sb_offset[0] = cpu_to_le64(BCH_SB_SECTOR);
+ sb->layout.sb_offset[1] = cpu_to_le64(BCH_SB_SECTOR + sb_size);
+}
+
int cmd_migrate_superblock(int argc, char *argv[])
{
static const struct option longopts[] = {
@@ -414,34 +444,23 @@ int cmd_migrate_superblock(int argc, char *argv[])
if (!sb_offset)
die("Please specify offset of existing superblock");
- int fd = xopen(devs.data[0], O_RDWR);
+ int fd = xopen(devs.data[0], O_RDWR | O_EXCL);
struct bch_sb *sb = __bch2_super_read(fd, sb_offset);
- unsigned sb_size = 1U << sb->layout.sb_max_size_bits;
+ unsigned sb_size;
+ /* Check for invocation errors early */
+ add_default_sb_layout(sb, &sb_size);
- if (sb->layout.nr_superblocks >= ARRAY_SIZE(sb->layout.sb_offset))
- die("Can't add superblock: no space left in superblock layout");
-
- for (unsigned i = 0; i < sb->layout.nr_superblocks; i++)
- if (le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR ||
- le64_to_cpu(sb->layout.sb_offset[i]) == BCH_SB_SECTOR + sb_size)
- die("Superblock layout already has default superblocks");
-
- memmove(&sb->layout.sb_offset[2],
- &sb->layout.sb_offset[0],
- sb->layout.nr_superblocks * sizeof(u64));
- sb->layout.nr_superblocks += 2;
- sb->layout.sb_offset[0] = cpu_to_le64(BCH_SB_SECTOR);
- sb->layout.sb_offset[1] = cpu_to_le64(BCH_SB_SECTOR + sb_size);
+ /* Rewrite first 0-3.5k bytes with zeroes, ensuring we blow away
+ * the old superblock */
+ // TODO: fix the "Superblock write was silently dropped" warning properly
+ static const char zeroes[(BCH_SB_SECTOR << 9) + sizeof(struct bch_sb)];
+ xpwrite(fd, zeroes, ARRAY_SIZE(zeroes), 0, "zeroing start of disk");
- /* also write first 0-3.5k bytes with zeroes, ensure we blow away old
- * superblock */
- static const char zeroes[BCH_SB_SECTOR << 9];
- xpwrite(fd, zeroes, BCH_SB_SECTOR << 9, 0, "zeroing start of disk");
-
- bch2_super_write(fd, sb);
xclose(fd);
- /* mark new superblocks */
+ /* We start a normal FS instance with the sb buckets temporarily
+ * prohibited from allocation, performing any recovery/upgrade/downgrade
+ * as needed, and only then change the superblock layout */
struct bch_opts opts = bch2_opts_empty();
opt_set(opts, nostart, true);
@@ -454,29 +473,46 @@ int cmd_migrate_superblock(int argc, char *argv[])
die("error opening filesystem: %s", bch2_err_str(ret));
struct bch_dev *ca = c->devs[0];
- for (u64 b = 0; bucket_to_sector(ca, b) < BCH_SB_SECTOR + sb_size * 2; b++)
- set_bit(b, ca->buckets_nouse);
+ mark_nouse_range(ca, 0, BCH_SB_SECTOR + sb_size * 2);
ret = bch2_fs_start(c);
if (ret)
die("Error starting filesystem: %s", bch2_err_str(ret));
+ BUG_ON(1U << ca->disk_sb.sb->layout.sb_max_size_bits != sb_size);
+
+ /* Here the FS is already RW.
+ * Apply the superblock layout changes first, everything else can be
+ * repaired on a subsequent recovery */
+ add_default_sb_layout(ca->disk_sb.sb, NULL);
+ ret = bch2_write_super(c);
+ if (ret)
+ die("Error writing superblock: %s", bch2_err_str(ret));
+
+ /* Now explicitly mark the new sb buckets in FS metadata */
+ ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
+ if (ret)
+ die("Error marking superblock buckets: %s", bch2_err_str(ret));
+
bch2_fs_stop(c);
+#if CONFIG_BCACHEFS_DEBUG
+ /* Verify that filesystem is clean and consistent */
+
opts = bch2_opts_empty();
opt_set(opts, fsck, true);
opt_set(opts, fix_errors, true);
-
- /*
- * Hack: the free space counters are coming out wrong after marking the
- * new superblock, but it's just the device counters so it's
- * inconsequential:
- */
+ opt_set(opts, nochanges, true);
c = bch2_fs_open(&devs, &opts);
ret = PTR_ERR_OR_ZERO(c);
if (ret)
- die("error opening filesystem: %s", bch2_err_str(ret));
+ die("error checking filesystem: %s", bch2_err_str(ret));
+
+ if (test_bit(BCH_FS_errors, &c->flags) || test_bit(BCH_FS_errors_fixed, &c->flags))
+ die("Filesystem has errors after migration");
+
bch2_fs_stop(c);
+#endif
return 0;
}
diff --git a/c_src/cmd_super.c b/c_src/cmd_super.c
index 18e4df80..1f93ed25 100644
--- a/c_src/cmd_super.c
+++ b/c_src/cmd_super.c
@@ -480,6 +480,14 @@ int cmd_recover_super(int argc, char *argv[])
if (args.yes || ask_yn())
bch2_super_write(dev_fd, sb);
+ /*
+ * Ensure that 'bcachefs mount' sees the newly formatted devices when
+ * scanning by UUID in the udev database:
+ */
+ CLASS(printbuf, udevadm_cmd)();
+ prt_printf(&udevadm_cmd, "udevadm trigger --settle %s", dev_path);
+ system(udevadm_cmd.buf);
+
if (args.src_device)
printf("Recovered device will no longer have a journal, please run fsck\n");
diff --git a/c_src/cmds.h b/c_src/cmds.h
index 30b47286..fa4b87a9 100644
--- a/c_src/cmds.h
+++ b/c_src/cmds.h
@@ -34,6 +34,8 @@ int cmd_fsck(int argc, char *argv[]);
int cmd_recovery_pass(int argc, char *argv[]);
int cmd_dump(int argc, char *argv[]);
+int cmd_undump(int argc, char *argv[]);
+
int cmd_list_journal(int argc, char *argv[]);
int cmd_kill_btree_node(int argc, char *argv[]);
diff --git a/c_src/libbcachefs.c b/c_src/libbcachefs.c
index ed84be1b..656c9132 100644
--- a/c_src/libbcachefs.c
+++ b/c_src/libbcachefs.c
@@ -205,7 +205,7 @@ struct bch_sb *bch2_format(struct bch_opt_strs fs_opt_strs,
/* calculate block size: */
if (!opt_defined(fs_opts, block_size)) {
unsigned max_dev_block_size = 0;
-
+
darray_for_each(devs, i)
max_dev_block_size = max(max_dev_block_size, get_blocksize(i->bdev->bd_fd));
@@ -357,6 +357,16 @@ struct bch_sb *bch2_format(struct bch_opt_strs fs_opt_strs,
xclose(i->bdev->bd_fd);
}
+ /*
+ * Ensure that 'bcachefs mount' sees the newly formatted devices when
+ * scanning by UUID in the udev database:
+ */
+ CLASS(printbuf, udevadm_cmd)();
+ prt_str(&udevadm_cmd, "udevadm trigger --settle");
+ darray_for_each(devs, i)
+ prt_printf(&udevadm_cmd, " %s", i->path);
+ system(udevadm_cmd.buf);
+
return sb.sb;
}
diff --git a/c_src/qcow2.c b/c_src/qcow2.c
index 53959a00..7afb6812 100644
--- a/c_src/qcow2.c
+++ b/c_src/qcow2.c
@@ -143,8 +143,7 @@ void qcow2_image_finish(struct qcow2_image *img)
memset(buf, 0, img->block_size);
memcpy(buf, &hdr, sizeof(hdr));
- xpwrite(img->outfd, buf, img->block_size, 0,
- "qcow2 header");
+ xpwrite(img->outfd, buf, img->block_size, 0, "qcow2 header");
free(img->l2_table);
free(img->l1_table);
@@ -160,3 +159,51 @@ void qcow2_write_image(int infd, int outfd, ranges *data,
qcow2_write_ranges(&img, data);
qcow2_image_finish(&img);
}
+
+void qcow2_to_raw(int infd, int outfd)
+{
+ struct qcow2_hdr hdr;
+
+ xpread(infd, &hdr, sizeof(hdr), 0);
+
+ if (hdr.magic != cpu_to_be32(QCOW_MAGIC))
+ die("not a qcow2 image");
+
+ if (hdr.version != cpu_to_be32(QCOW_VERSION))
+ die("incorrect qcow2 version");
+
+ ftruncate(outfd, be64_to_cpu(hdr.size));
+
+ unsigned block_size = 1U << be32_to_cpu(hdr.block_bits);
+
+ unsigned l1_size = be32_to_cpu(hdr.l1_size);
+ unsigned l2_size = block_size / sizeof(u64);
+
+ __be64 *l1_table = xcalloc(l1_size, sizeof(u64));
+ __be64 *l2_table = xmalloc(block_size);
+ void *data_buf = xmalloc(block_size);
+
+ xpread(infd, l1_table, l1_size * sizeof(u64), be64_to_cpu(hdr.l1_table_offset));
+
+ for (u64 i = 0; i < l1_size; i++) {
+ if (!l1_table[i])
+ continue;
+
+ xpread(infd, l2_table, block_size, be64_to_cpu(l1_table[i]) & ~QCOW_OFLAG_COPIED);
+
+ for (unsigned j = 0; j < l2_size; j++) {
+ u64 src_offset = be64_to_cpu(l2_table[j]) & ~QCOW_OFLAG_COPIED;
+ if (!src_offset)
+ continue;
+
+ u64 dst_offset = (i * l2_size + j) * block_size;
+
+ xpread(infd, data_buf, block_size, src_offset);
+ xpwrite(outfd, data_buf, block_size, dst_offset, "qcow2 data");
+ }
+ }
+
+ free(data_buf);
+ free(l2_table);
+ free(l1_table);
+}
diff --git a/c_src/qcow2.h b/c_src/qcow2.h
index c7b35627..7ccb2773 100644
--- a/c_src/qcow2.h
+++ b/c_src/qcow2.h
@@ -4,6 +4,8 @@
#include <linux/types.h>
#include "tools-util.h"
+#define QCOW2_L1_MAX (4ULL << 20)
+
struct qcow2_image {
int infd;
int outfd;
@@ -25,4 +27,6 @@ void qcow2_image_finish(struct qcow2_image *);
void qcow2_write_image(int, int, ranges *, unsigned);
+void qcow2_to_raw(int, int);
+
#endif /* _QCOW2_H */
diff --git a/c_src/tools-util.c b/c_src/tools-util.c
index f48d3f21..7733b9e9 100644
--- a/c_src/tools-util.c
+++ b/c_src/tools-util.c
@@ -52,15 +52,15 @@ char *mprintf(const char *fmt, ...)
return str;
}
-void xpread(int fd, void *buf, size_t count, off_t offset)
+void __xpread(int fd, void *buf, size_t count, off_t offset, const char *file, unsigned line)
{
while (count) {
ssize_t r = pread(fd, buf, count, offset);
if (r < 0)
- die("read error: %m");
+ die("read error: %m at %s:%u", file, line);
if (!r)
- die("pread error: unexpected eof");
+ die("pread error: unexpected eof at %s:%u", file, line);
count -= r;
offset += r;
}
diff --git a/c_src/tools-util.h b/c_src/tools-util.h
index b8104002..988e2d62 100644
--- a/c_src/tools-util.h
+++ b/c_src/tools-util.h
@@ -28,8 +28,13 @@ void die(const char *, ...)
__attribute__ ((format (printf, 1, 2))) noreturn;
char *mprintf(const char *, ...)
__attribute__ ((format (printf, 1, 2)));
-void xpread(int, void *, size_t, off_t);
+
+void __xpread(int, void *, size_t, off_t, const char *, unsigned);
+#define xpread(_fd, _buf, _count, _offset) \
+ __xpread(_fd, _buf, _count, _offset, __FILE__, __LINE__)
+
void xpwrite(int, const void *, size_t, off_t, const char *);
+
struct stat xfstatat(int, const char *, int);
struct stat xfstat(int);
struct stat xstat(const char *);
diff --git a/linux/shrinker.c b/linux/shrinker.c
index c0098cad..7a8a9a6a 100644
--- a/linux/shrinker.c
+++ b/linux/shrinker.c
@@ -15,6 +15,9 @@ static DEFINE_MUTEX(shrinker_lock);
void shrinker_free(struct shrinker *s)
{
+ if (!s)
+ return;
+
if (s->list.next) {
mutex_lock(&shrinker_lock);
list_del(&s->list);
diff --git a/src/bcachefs.rs b/src/bcachefs.rs
index 4fb0b14c..d0c1d6e8 100644
--- a/src/bcachefs.rs
+++ b/src/bcachefs.rs
@@ -48,6 +48,7 @@ fn handle_c_command(mut argv: Vec<String>, symlink_cmd: Option<&str>) -> i32 {
"data" => c::data_cmds(argc, argv),
"device" => c::device_cmds(argc, argv),
"dump" => c::cmd_dump(argc, argv),
+ "undump" => c::cmd_undump(argc, argv),
"format" => c::cmd_format(argc, argv),
"fs" => c::fs_cmds(argc, argv),
"fsck" => c::cmd_fsck(argc, argv),
generated by cgit v1.2.3 (git 2.51.0) at 2025-09-20 08:07:01 +0000