gfs2-utils/SOURCES/bz1833141-2-gfs2_jadd_error...

609 lines
15 KiB
Diff

commit 206b040657b5125c2f2efe35dddcb7463fb49788
Author: Abhi Das <adas@redhat.com>
Date: Mon May 11 14:41:06 2020 -0500
gfs2_jadd: error handling overhaul
Handle error conditions better and fail gracefully.
Resolves: rhbz#1833141
Signed-off-by: Abhi Das <adas@redhat.com>
diff --git a/gfs2/mkfs/main_jadd.c b/gfs2/mkfs/main_jadd.c
index c5424803..ea89c96b 100644
--- a/gfs2/mkfs/main_jadd.c
+++ b/gfs2/mkfs/main_jadd.c
@@ -42,15 +42,13 @@ struct jadd_opts {
#define JA_FL_SET 0
#define JA_FL_CLEAR 1
-static void set_flags(int fd, int op, uint32_t flags)
+static int set_flags(int fd, int op, uint32_t flags)
{
- int err;
uint32_t val;
- err = ioctl(fd, FS_IOC_GETFLAGS, &val);
- if (err) {
+ if (ioctl(fd, FS_IOC_GETFLAGS, &val)) {
perror("GETFLAGS");
- exit(EXIT_FAILURE);
+ return -1;
}
if (op == JA_FL_SET)
@@ -58,11 +56,11 @@ static void set_flags(int fd, int op, uint32_t flags)
else if (op == JA_FL_CLEAR)
val &= ~flags;
- err = ioctl(fd, FS_IOC_SETFLAGS, &val);
- if (err) {
+ if (ioctl(fd, FS_IOC_SETFLAGS, &val)) {
perror("SETFLAGS");
- exit(EXIT_FAILURE);
+ return -1;
}
+ return 0;
}
static int rename2system(struct jadd_opts *opts, const char *new_dir, const char *new_name)
@@ -243,188 +241,214 @@ static void print_results(struct jadd_opts *opts)
static int create_new_inode(struct jadd_opts *opts, uint64_t *addr)
{
char *name = opts->new_inode;
- int fd;
- int error;
+ int fd, error = 0;
for (;;) {
fd = open(name, O_WRONLY | O_CREAT | O_EXCL | O_NOFOLLOW | O_CLOEXEC, 0600);
if (fd >= 0)
break;
if (errno == EEXIST) {
- error = unlink(name);
- if (error){
+ if (unlink(name)) {
perror("unlink");
- exit(EXIT_FAILURE);
+ return -1;
}
- } else{
- perror("create");
- exit(EXIT_FAILURE);
+ continue;
}
+ perror("create");
+ return -1;
}
+
if (addr != NULL) {
struct stat st;
- fstat(fd, &st);
+ if ((error = fstat(fd, &st))) {
+ perror("fstat");
+ return close(fd);
+ }
*addr = st.st_ino;
}
return fd;
}
-static void add_ir(struct jadd_opts *opts)
+static int add_ir(struct jadd_opts *opts)
{
- int fd;
+ int fd, error = 0;
char new_name[256];
- int error;
+ struct gfs2_inum_range ir;
- fd = create_new_inode(opts, NULL);
+ if ((fd = create_new_inode(opts, NULL)) < 0)
+ return fd;
- {
- struct gfs2_inum_range ir;
+ if ((error = set_flags(fd, JA_FL_SET, FS_JOURNAL_DATA_FL)))
+ goto close_fd;
- set_flags(fd, JA_FL_SET, FS_JOURNAL_DATA_FL);
- memset(&ir, 0, sizeof(struct gfs2_inum_range));
- if (write(fd, (void*)&ir, sizeof(struct gfs2_inum_range)) !=
- sizeof(struct gfs2_inum_range)) {
- perror("add_ir");
- exit(EXIT_FAILURE);
- }
+ memset(&ir, 0, sizeof(struct gfs2_inum_range));
+ if (write(fd, (void*)&ir, sizeof(struct gfs2_inum_range)) !=
+ sizeof(struct gfs2_inum_range)) {
+ perror("add_ir write");
+ error = -1;
+ goto close_fd;
+ }
+
+ if ((error = fsync(fd))) {
+ perror("add_ir fsync");
+ goto close_fd;
}
- close(fd);
sprintf(new_name, "inum_range%u", opts->journals);
error = rename2system(opts, opts->per_node, new_name);
- if (error < 0 && errno != EEXIST){
+ if (error < 0 && errno != EEXIST) {
perror("add_ir rename2system");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
+close_fd:
+ return close(fd) || error;
}
-static void add_sc(struct jadd_opts *opts)
+static int add_sc(struct jadd_opts *opts)
{
- int fd;
+ int fd, error = 0;
char new_name[256];
- int error;
+ struct gfs2_statfs_change sc;
- fd = create_new_inode(opts, NULL);
+ if ((fd = create_new_inode(opts, NULL)) < 0)
+ return fd;
- {
- struct gfs2_statfs_change sc;
- set_flags(fd, JA_FL_SET, FS_JOURNAL_DATA_FL);
+ if ((error = set_flags(fd, JA_FL_SET, FS_JOURNAL_DATA_FL)))
+ goto close_fd;
- memset(&sc, 0, sizeof(struct gfs2_statfs_change));
- if (write(fd, (void*)&sc, sizeof(struct gfs2_statfs_change)) !=
- sizeof(struct gfs2_statfs_change)) {
- perror("add_sc");
- exit(EXIT_FAILURE);
- }
+ memset(&sc, 0, sizeof(struct gfs2_statfs_change));
+ if (write(fd, (void*)&sc, sizeof(struct gfs2_statfs_change)) !=
+ sizeof(struct gfs2_statfs_change)) {
+ perror("add_sc write");
+ error = -1;
+ goto close_fd;
}
- close(fd);
+ if ((error = fsync(fd))) {
+ perror("add_sc fsync");
+ goto close_fd;
+ }
sprintf(new_name, "statfs_change%u", opts->journals);
error = rename2system(opts, opts->per_node, new_name);
if (error < 0 && errno != EEXIST){
perror("add_sc rename2system");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
+close_fd:
+ return close(fd) || error;
}
-static void add_qc(struct gfs2_sbd *sdp, struct jadd_opts *opts)
+static int add_qc(struct gfs2_sbd *sdp, struct jadd_opts *opts)
{
- int fd;
- char new_name[256];
- int error;
-
- fd = create_new_inode(opts, NULL);
-
- {
- char buf[sdp->bsize];
- unsigned int blocks =
- sdp->qcsize << (20 - sdp->sd_sb.sb_bsize_shift);
- unsigned int x;
- struct gfs2_meta_header mh;
-
- set_flags(fd, JA_FL_CLEAR, FS_JOURNAL_DATA_FL);
- memset(buf, 0, sdp->bsize);
-
- for (x=0; x<blocks; x++) {
- if (write(fd, buf, sdp->bsize) != sdp->bsize) {
- perror("add_qc");
- exit(EXIT_FAILURE);
- }
+ int fd, error = 0;
+ char new_name[256], buf[sdp->bsize];
+ unsigned int blocks =
+ sdp->qcsize << (20 - sdp->sd_sb.sb_bsize_shift);
+ unsigned int x;
+ struct gfs2_meta_header mh;
+
+ if ((fd = create_new_inode(opts, NULL)) < 0)
+ return fd;
+
+ if ((error = set_flags(fd, JA_FL_CLEAR, FS_JOURNAL_DATA_FL)))
+ goto close_fd;
+
+ memset(buf, 0, sdp->bsize);
+ for (x=0; x<blocks; x++) {
+ if (write(fd, buf, sdp->bsize) != sdp->bsize) {
+ perror("add_qc write");
+ error = -1;
+ goto close_fd;
}
+ }
- lseek(fd, 0, SEEK_SET);
-
- memset(&mh, 0, sizeof(struct gfs2_meta_header));
- mh.mh_magic = GFS2_MAGIC;
- mh.mh_type = GFS2_METATYPE_QC;
- mh.mh_format = GFS2_FORMAT_QC;
- gfs2_meta_header_out(&mh, buf);
+ if ((error = lseek(fd, 0, SEEK_SET)) < 0) {
+ perror("add_qc lseek");
+ goto close_fd;
+ }
- for (x=0; x<blocks; x++) {
- if (write(fd, buf, sdp->bsize) != sdp->bsize) {
- perror("add_qc");
- exit(EXIT_FAILURE);
- }
+ memset(&mh, 0, sizeof(struct gfs2_meta_header));
+ mh.mh_magic = GFS2_MAGIC;
+ mh.mh_type = GFS2_METATYPE_QC;
+ mh.mh_format = GFS2_FORMAT_QC;
+ gfs2_meta_header_out(&mh, buf);
+
+ for (x=0; x<blocks; x++) {
+ if (write(fd, buf, sdp->bsize) != sdp->bsize) {
+ perror("add_qc write");
+ error = 1;
+ goto close_fd;
}
-
- error = fsync(fd);
- if (error){
+ if ((error = fsync(fd))) {
perror("add_qc fsync");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
}
- close(fd);
-
sprintf(new_name, "quota_change%u", opts->journals);
error = rename2system(opts, opts->per_node, new_name);
if (error < 0 && errno != EEXIST){
perror("add_qc rename2system");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
+close_fd:
+ return close(fd) || error;
}
-static void gather_info(struct gfs2_sbd *sdp, struct jadd_opts *opts)
+static int gather_info(struct gfs2_sbd *sdp, struct jadd_opts *opts)
{
struct statfs statbuf;
+
if (statfs(opts->path, &statbuf) < 0) {
perror(opts->path);
- exit(EXIT_FAILURE);
+ return -1;
}
+
sdp->bsize = statbuf.f_bsize;
sdp->blks_total = statbuf.f_blocks;
sdp->blks_alloced = sdp->blks_total - statbuf.f_bfree;
+
+ return 0;
}
-static void find_current_journals(struct jadd_opts *opts)
+static int find_current_journals(struct jadd_opts *opts)
{
struct dirent *dp;
DIR *dirp;
unsigned existing_journals = 0;
+ int ret = 0;
dirp = opendir(opts->jindex);
if (!dirp) {
perror("jindex");
- exit(EXIT_FAILURE);
+ ret = -1;
+ goto out;
}
while (dirp) {
if ((dp = readdir(dirp)) != NULL) {
if (strncmp(dp->d_name, "journal", 7) == 0)
existing_journals++;
} else
- goto close;
+ goto close_fd;
}
-close:
- closedir(dirp);
+close_fd:
+ if ((ret = closedir(dirp)))
+ goto out;
+
if (existing_journals == 0) {
- die( _("No journals found. Did you run mkfs.gfs2 correctly?\n"));
+ errno = EINVAL;
+ perror("No journals found. Did you run mkfs.gfs2 correctly?\n");
+ ret = -1;
+ goto out;
}
opts->orig_journals = existing_journals;
+out:
+ return ret;
}
#ifdef GFS2_HAS_LH_V2
@@ -450,83 +474,88 @@ static uint64_t find_block_address(int fd, off_t offset, unsigned bsize)
}
#endif
-static void add_j(struct gfs2_sbd *sdp, struct jadd_opts *opts)
+static int add_j(struct gfs2_sbd *sdp, struct jadd_opts *opts)
{
- int fd;
- char new_name[256];
- int error;
- uint64_t addr;
-
- fd = create_new_inode(opts, &addr);
-
- {
- char buf[sdp->bsize];
- unsigned int blocks =
- sdp->jsize << (20 - sdp->sd_sb.sb_bsize_shift);
- unsigned int x;
- struct gfs2_log_header lh;
- uint64_t seq = RANDOM(blocks);
- off_t off = 0;
-
- set_flags(fd, JA_FL_CLEAR, FS_JOURNAL_DATA_FL);
- memset(buf, 0, sdp->bsize);
- for (x=0; x<blocks; x++) {
- if (write(fd, buf, sdp->bsize) != sdp->bsize) {
- perror("add_j");
- exit(EXIT_FAILURE);
- }
+ int fd, error = 0;
+ char new_name[256], buf[sdp->bsize];
+ unsigned int x, blocks =
+ sdp->jsize << (20 - sdp->sd_sb.sb_bsize_shift);
+ struct gfs2_log_header lh;
+ uint64_t seq = RANDOM(blocks), addr;
+ off_t off = 0;
+
+ if ((fd = create_new_inode(opts, &addr)) < 0)
+ return fd;
+
+ if ((error = set_flags(fd, JA_FL_CLEAR, FS_JOURNAL_DATA_FL)))
+ goto close_fd;
+
+ memset(buf, 0, sdp->bsize);
+ for (x=0; x<blocks; x++) {
+ if (write(fd, buf, sdp->bsize) != sdp->bsize) {
+ perror("add_j write");
+ error = -1;
+ goto close_fd;
}
+ }
- lseek(fd, 0, SEEK_SET);
+ if ((error = lseek(fd, 0, SEEK_SET)) < 0) {
+ perror("add_j lseek");
+ goto close_fd;
+ }
- memset(&lh, 0, sizeof(struct gfs2_log_header));
- lh.lh_header.mh_magic = GFS2_MAGIC;
- lh.lh_header.mh_type = GFS2_METATYPE_LH;
- lh.lh_header.mh_format = GFS2_FORMAT_LH;
- lh.lh_flags = GFS2_LOG_HEAD_UNMOUNT;
+ memset(&lh, 0, sizeof(struct gfs2_log_header));
+ lh.lh_header.mh_magic = GFS2_MAGIC;
+ lh.lh_header.mh_type = GFS2_METATYPE_LH;
+ lh.lh_header.mh_format = GFS2_FORMAT_LH;
+ lh.lh_flags = GFS2_LOG_HEAD_UNMOUNT;
#ifdef GFS2_HAS_LH_V2
- lh.lh_flags |= GFS2_LOG_HEAD_USERSPACE;
- lh.lh_jinode = addr;
+ lh.lh_flags |= GFS2_LOG_HEAD_USERSPACE;
+ lh.lh_jinode = addr;
#endif
- for (x=0; x<blocks; x++) {
- uint32_t hash;
-
- lh.lh_sequence = seq;
- lh.lh_blkno = x;
- gfs2_log_header_out(&lh, buf);
- hash = lgfs2_log_header_hash(buf);
- ((struct gfs2_log_header *)buf)->lh_hash = cpu_to_be32(hash);
+ for (x=0; x<blocks; x++) {
+ uint32_t hash;
#ifdef GFS2_HAS_LH_V2
- ((struct gfs2_log_header *)buf)->lh_addr = cpu_to_be64(
- find_block_address(fd, off, sdp->bsize));
- hash = lgfs2_log_header_crc(buf, sdp->bsize);
- ((struct gfs2_log_header *)buf)->lh_crc = cpu_to_be32(hash);
+ uint64_t blk_addr = 0;
#endif
- if (write(fd, buf, sdp->bsize) != sdp->bsize) {
- perror("add_j");
- exit(EXIT_FAILURE);
- }
-
- if (++seq == blocks)
- seq = 0;
- off += sdp->bsize;
+ lh.lh_sequence = seq;
+ lh.lh_blkno = x;
+ gfs2_log_header_out(&lh, buf);
+ hash = lgfs2_log_header_hash(buf);
+ ((struct gfs2_log_header *)buf)->lh_hash = cpu_to_be32(hash);
+#ifdef GFS2_HAS_LH_V2
+ if (!(blk_addr = find_block_address(fd, off, sdp->bsize))) {
+ error = -1;
+ goto close_fd;
+ }
+ ((struct gfs2_log_header *)buf)->lh_addr = cpu_to_be64(blk_addr);
+ hash = lgfs2_log_header_crc(buf, sdp->bsize);
+ ((struct gfs2_log_header *)buf)->lh_crc = cpu_to_be32(hash);
+#endif
+ if (write(fd, buf, sdp->bsize) != sdp->bsize) {
+ perror("add_j write");
+ error = -1;
+ goto close_fd;
}
- error = fsync(fd);
- if (error){
+ if (++seq == blocks)
+ seq = 0;
+ off += sdp->bsize;
+
+ if ((error = fsync(fd))) {
perror("add_j fsync");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
}
- close(fd);
-
sprintf(new_name, "journal%u", opts->journals);
error = rename2system(opts, opts->jindex, new_name);
if (error < 0 && errno != EEXIST){
perror("add_j rename2system");
- exit(EXIT_FAILURE);
+ goto close_fd;
}
+close_fd:
+ return close(fd) || error;
}
static int check_fit(struct gfs2_sbd *sdp, struct jadd_opts *opts)
@@ -554,7 +583,7 @@ static int check_fit(struct gfs2_sbd *sdp, struct jadd_opts *opts)
printf( _("Available space : %*lu blks\n\n"), 10,
sdp->blks_total - sdp->blks_alloced);
errno = ENOSPC;
- return 1;
+ return -1;
}
return 0;
}
@@ -581,35 +610,42 @@ int main(int argc, char *argv[])
sbd.path_fd = lgfs2_open_mnt_dir(opts.path, O_RDONLY|O_CLOEXEC, &mnt);
if (sbd.path_fd < 0) {
- fprintf(stderr, _("Error looking up mount '%s': %s\n"), opts.path, strerror(errno));
- exit(EXIT_FAILURE);
+ fprintf(stderr, "Error looking up mount '%s': %s\n",
+ opts.path, strerror(errno));
+ ret = -1;
+ goto out;
}
if (mnt == NULL) {
- fprintf(stderr, _("%s: not a mounted gfs2 file system\n"), opts.path);
- exit(EXIT_FAILURE);
+ fprintf(stderr, "%s: not a mounted gfs2 file system\n", opts.path);
+ ret = -1;
+ goto close_sb;
}
- gather_info(sdp, &opts);
+
+ if ((ret = gather_info(sdp, &opts)))
+ goto close_sb;
+
mfs.context = copy_context_opt(mnt);
- if (mount_gfs2_meta(&mfs, mnt->mnt_dir, opts.debug)) {
+ if ((ret = mount_gfs2_meta(&mfs, mnt->mnt_dir, opts.debug))) {
perror("GFS2 metafs");
- exit(EXIT_FAILURE);
+ goto close_sb;
}
- if (build_paths(mfs.path, &opts)) {
+ if ((ret = build_paths(mfs.path, &opts))) {
perror(_("Failed to build paths"));
- exit(EXIT_FAILURE);
+ goto umount_meta;
}
- if (compute_constants(sdp)) {
+ if ((ret = compute_constants(sdp))) {
perror(_("Failed to compute file system constants"));
- exit(EXIT_FAILURE);
+ goto free_paths;
}
- find_current_journals(&opts);
- ret = check_fit(sdp, &opts);
- if (ret) {
+ if ((ret = find_current_journals(&opts)))
+ goto free_paths;
+
+ if ((ret = check_fit(sdp, &opts))) {
perror(_("Failed to add journals"));
- goto out;
+ goto free_paths;
}
total = opts.orig_journals + opts.journals;
@@ -617,23 +653,29 @@ int main(int argc, char *argv[])
opts.journals < total;
opts.journals++) {
if (metafs_interrupted) {
- cleanup_metafs(&mfs);
- exit(130);
+ errno = 130;
+ goto free_paths;
}
- add_ir(&opts);
- add_sc(&opts);
- add_qc(sdp, &opts);
- add_j(sdp, &opts);
+ if ((ret = add_ir(&opts)))
+ goto free_paths;
+ if ((ret = add_sc(&opts)))
+ goto free_paths;
+ if ((ret = add_qc(sdp, &opts)))
+ goto free_paths;
+ if ((ret = add_j(sdp, &opts)))
+ goto free_paths;
}
-out:
+free_paths:
free(opts.new_inode);
free(opts.per_node);
free(opts.jindex);
- close(sdp->path_fd);
- cleanup_metafs(&mfs);
+umount_meta:
sync();
-
+ cleanup_metafs(&mfs);
+close_sb:
+ close(sdp->path_fd);
+out:
if (!ret)
print_results(&opts);