#include <linux/kallsyms.h>
#include <linux/module.h>
#include <linux/user_namespace.h>
-#ifdef HAVE_UIDGID_HEADER
-# include <linux/uidgid.h>
-#endif
+#include <linux/uidgid.h>
/* prerequisite for linux/xattr.h */
#include <linux/types.h>
*/
/* LASSERT(current->journal_info == NULL); */
- inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+ inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
if (IS_ERR(inode)) {
CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
id->oii_ino, PTR_ERR(inode));
}
if (fid != NULL)
- CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
- "maximum entry limit\n",
+ /* below message is checked in sanity.sh test_129 */
+ CWARN("%s: directory (inode: %lu, FID: "DFID") %s maximum entry limit\n",
osd_name(osd), parent->i_ino, PFID(fid),
rc == -ENOSPC ? "has reached" : "is approaching");
else
- CWARN("%s: directory (inode: %lu, FID: unknown) %s "
- "maximum entry limit\n",
+ /* below message is checked in sanity.sh test_129 */
+ CWARN("%s: directory (inode: %lu, FID: unknown) %s maximum entry limit\n",
osd_name(osd), parent->i_ino,
rc == -ENOSPC ? "has reached" : "is approaching");
*/
again:
- inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
+ inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
if (IS_ERR(inode)) {
rc = PTR_ERR(inode);
if (!trusted && (rc == -ENOENT || rc == -ESTALE))
}
struct osd_check_lmv_buf {
-#ifdef HAVE_DIR_CONTEXT
/* please keep it as first member */
struct dir_context ctx;
-#endif
struct osd_thread_info *oclb_info;
struct osd_device *oclb_dev;
struct osd_idmap_cache *oclb_oic;
};
/**
- * It is called internally by ->readdir() to filter out the
+ * It is called internally by ->iterate*() to filter out the
* local slave object's FID of the striped directory.
*
* \retval 1 found the local slave's FID
const struct file_operations *fops;
struct lmv_mds_md_v1 *lmv1;
struct osd_check_lmv_buf oclb = {
-#ifdef HAVE_DIR_CONTEXT
.ctx.actor = osd_stripe_dir_filldir,
-#endif
.oclb_info = oti,
.oclb_dev = dev,
.oclb_oic = oic,
dentry->d_sb = inode->i_sb;
filp->f_pos = 0;
filp->f_path.dentry = dentry;
- filp->f_mode = FMODE_64BITHASH;
+ filp->f_flags |= O_NOATIME;
+ filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
filp->f_mapping = inode->i_mapping;
filp->f_op = fops;
filp->private_data = NULL;
+ filp->f_cred = current_cred();
set_file_inode(filp, inode);
+ rc = osd_security_file_alloc(filp);
+ if (rc)
+ goto out;
do {
oclb.oclb_items = 0;
-#ifdef HAVE_DIR_CONTEXT
- oclb.ctx.pos = filp->f_pos;
-#ifdef HAVE_ITERATE_SHARED
- rc = fops->iterate_shared(filp, &oclb.ctx);
-#else
- rc = fops->iterate(filp, &oclb.ctx);
-#endif
- filp->f_pos = oclb.ctx.pos;
-#else
- rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
-#endif
+ rc = iterate_dir(filp, &oclb.ctx);
} while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
fops->release(inode, filp);
goto trigger;
}
+ /* -ESTALE is returned if inode of OST object doesn't exist */
+ if (result == -ESTALE &&
+ fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
+ GOTO(out, result = 0);
+ }
+
if (result)
GOTO(out, result);
LASSERT(!updated);
+ /*
+ * if two OST objects map to the same inode, and inode mode is
+ * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
+ * reserved by precreate, and not written yet, in this case, don't
+ * set inode for the object whose FID mismatch, so that it can create
+ * inode and not block precreate.
+ */
+ if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
+ inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
+ obj->oo_inode = NULL;
+ GOTO(out, result = 0);
+ }
+
result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
/*
* "result == -ENOENT" means the cached OI mapping has been removed
*/
if (last_credits != oh->ot_credits &&
time_after(jiffies, last_printed +
- msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
+ cfs_time_seconds(60)) &&
osd_transaction_size(dev) > 512) {
CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
oh->ot_credits, osd_transaction_size(dev));
struct osd_device *osd = osd_dt_dev(th->th_dev);
struct qsd_instance *qsd = osd_def_qsd(osd);
struct lquota_trans *qtrans;
- struct list_head truncates = LIST_HEAD_INIT(truncates);
+ LIST_HEAD(truncates);
int rc = 0, remove_agents = 0;
ENTRY;
* Concurrency: shouldn't matter.
*/
int osd_statfs(const struct lu_env *env, struct dt_device *d,
- struct obd_statfs *sfs)
+ struct obd_statfs *sfs, struct obd_statfs_info *info)
{
struct osd_device *osd = osd_dt_dev(d);
struct super_block *sb = osd_sb(osd);
goto out;
statfs_pack(sfs, ksfs);
- if (unlikely(sb->s_flags & MS_RDONLY))
+ if (unlikely(sb->s_flags & SB_RDONLY))
sfs->os_state |= OS_STATE_READONLY;
sfs->os_state |= osd->od_nonrotational ? OS_STATE_NONROT : 0;
*
* Reserve 0.78% of total space, at least 8MB for small filesystems.
*/
- CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
+ BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
#endif
param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
- if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE - ea_overhead)
- param->ddp_max_ea_size = OBD_MAX_EA_SIZE - ea_overhead;
+ if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
+ param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
/*
* Preferred RPC size for efficient disk IO. 4MB shows good
static int (*priv_dev_set_rdonly)(struct block_device *bdev);
static int (*priv_dev_check_rdonly)(struct block_device *bdev);
/* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
+static int (*priv_security_file_alloc)(struct file *file);
+
+int osd_security_file_alloc(struct file *file)
+{
+ if (priv_security_file_alloc)
+ return priv_security_file_alloc(file);
+ return 0;
+}
/*
* Concurrency: shouldn't matter.
return obj->oo_owner == env;
}
-static struct timespec *osd_inode_time(const struct lu_env *env,
- struct inode *inode, __u64 seconds)
-{
- struct osd_thread_info *oti = osd_oti_get(env);
- struct timespec *t = &oti->oti_time;
-
- t->tv_sec = seconds;
- t->tv_nsec = 0;
- *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
- return t;
-}
-
static void osd_inode_getattr(const struct lu_env *env,
struct inode *inode, struct lu_attr *attr)
{
RETURN(rc);
gid = i_gid_read(obj->oo_inode);
+ CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
+ attr->la_uid, gid, attr->la_gid);
enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
- rc = osd_declare_attr_qid(env, obj, oh, bspace,
- i_gid_read(obj->oo_inode),
+ rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
attr->la_gid, enforce, GRPQUOTA,
ignore_edquot);
if (rc)
return 0;
if (bits & LA_ATIME)
- inode->i_atime = *osd_inode_time(env, inode, attr->la_atime);
+ inode->i_atime = osd_inode_time(inode, attr->la_atime);
if (bits & LA_CTIME)
- inode->i_ctime = *osd_inode_time(env, inode, attr->la_ctime);
+ inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
if (bits & LA_MTIME)
- inode->i_mtime = *osd_inode_time(env, inode, attr->la_mtime);
+ inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
if (bits & LA_SIZE) {
spin_lock(&inode->i_lock);
LDISKFS_I(inode)->i_disksize = attr->la_size;
(attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
struct iattr iattr;
- ll_vfs_dq_init(inode);
+ CDEBUG(D_QUOTA,
+ "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
+ inode->i_ino, i_uid_read(inode), attr->la_uid,
+ i_gid_read(inode), attr->la_gid);
+
+ dquot_initialize(inode);
iattr.ia_valid = 0;
if (attr->la_valid & LA_UID)
iattr.ia_valid |= ATTR_UID;
iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
- rc = ll_vfs_dq_transfer(inode, &iattr);
+ rc = dquot_transfer(inode, &iattr);
if (rc) {
CERROR("%s: quota transfer failed: rc = %d. Is quota "
"enforcement enabled on the ldiskfs "
ll_dirty_inode(inode, I_DIRTY_DATASYNC);
+ osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
+
if (!(attr->la_valid & LA_FLAGS))
GOTO(out, rc);
lma->lma_incompat |=
lustre_to_lma_flags(attr->la_flags);
lustre_lma_swab(lma);
+
+ osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
+
rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
lma, sizeof(*lma), XATTR_REPLACE);
if (rc != 0) {
osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
}
out:
- osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
return rc;
}
osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
OI_CHECK_FLD, NULL);
+ if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP) && osd->od_is_ost) {
+ struct lu_fid next_fid = *fid;
+
+ /* insert next object in advance, and map to the same inode */
+ next_fid.f_oid++;
+ if (next_fid.f_oid != 0) {
+ osd_trans_exec_op(env, th, OSD_OT_INSERT);
+ osd_oi_insert(info, osd, &next_fid, id, oh->ot_handle,
+ OI_CHECK_FLD, NULL);
+ osd_trans_exec_check(env, th, OSD_OT_INSERT);
+ }
+ }
+
osd_trans_exec_check(env, th, OSD_OT_INSERT);
return rc;
*/
osd_trans_declare_op(env, oh, OSD_OT_INSERT,
osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
+ if (CFS_FAIL_CHECK(OBD_FAIL_OSD_DUPLICATE_MAP))
+ osd_trans_declare_op(env, oh, OSD_OT_INSERT,
+ osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
/* will help to find FID->ino mapping at dt_insert() */
rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
* debugging if we need to determine where this symlink came from.
*/
if (S_ISLNK(type)) {
- CLASSERT(LDISKFS_N_BLOCKS * 4 >= FID_LEN + 1);
+ BUILD_BUG_ON(LDISKFS_N_BLOCKS * 4 < FID_LEN + 1);
rc = snprintf((char *)LDISKFS_I(local)->i_data,
LDISKFS_N_BLOCKS * 4, DFID, PFID(fid));
struct osd_thread_info *info = osd_oti_get(env);
struct osd_obj_orphan *oor, *tmp;
struct osd_inode_id id;
- struct list_head list;
+ LIST_HEAD(list);
struct inode *inode;
struct lu_fid fid;
handle_t *jh;
__u32 ino;
- INIT_LIST_HEAD(&list);
-
spin_lock(&osd->od_osfs_lock);
list_for_each_entry_safe(oor, tmp, &osd->od_orphan_list, oor_list) {
- if (oor->oor_env == env) {
- list_del(&oor->oor_list);
- list_add(&oor->oor_list, &list);
- }
+ if (oor->oor_env == env)
+ list_move(&oor->oor_list, &list);
}
spin_unlock(&osd->od_osfs_lock);
RETURN(fl);
/* Remove old PFID EA entry firstly. */
- ll_vfs_dq_init(inode);
+ dquot_initialize(inode);
rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
if (rc == -ENODATA) {
if ((fl & LU_XATTR_REPLACE) && !(fl & LU_XATTR_CREATE))
obj->oo_pfid_in_lma = 0;
}
} else {
- ll_vfs_dq_init(inode);
+ dquot_initialize(inode);
dentry->d_inode = inode;
dentry->d_sb = inode->i_sb;
rc = osd_removexattr(dentry, inode, name);
file->f_op = inode->i_fop;
set_file_inode(file, inode);
- rc = ll_vfs_fsync_range(file, start, end, 0);
+ rc = vfs_fsync_range(file, start, end, 0);
RETURN(rc);
}
LASSERT(oh->ot_handle != NULL);
LASSERT(oh->ot_handle->h_transaction != NULL);
- ll_vfs_dq_init(dir);
+ dquot_initialize(dir);
dentry = osd_child_dentry_get(env, obj,
(char *)key, strlen((char *)key));
osd_get_ldiskfs_dirent_param(ldp, fid);
child = osd_child_dentry_get(info->oti_env, pobj, name, strlen(name));
child->d_fsdata = (void *)ldp;
- ll_vfs_dq_init(pobj->oo_inode);
+ dquot_initialize(pobj->oo_inode);
rc = osd_ldiskfs_add_entry(info, osd_obj2dev(pobj), oth->ot_handle,
child, cinode, hlock);
if (rc == 0 && OBD_FAIL_CHECK(OBD_FAIL_LFSCK_BAD_TYPE)) {
}
struct osd_filldir_cbs {
-#ifdef HAVE_DIR_CONTEXT
struct dir_context ctx;
-#endif
struct osd_it_ea *it;
};
/**
- * It is called internally by ->readdir(). It fills the
+ * It is called internally by ->iterate*(). It fills the
* iterator's in-memory data structure with required
* information i.e. name, namelen, rec_size etc.
*
}
/**
- * Calls ->readdir() to load a directory entry at a time
+ * Calls ->iterate*() to load a directory entry at a time
* and stored it in iterator's in-memory data structure.
*
* \param di iterator's in memory structure
struct file *filp = &it->oie_file;
int rc = 0;
struct osd_filldir_cbs buf = {
-#ifdef HAVE_DIR_CONTEXT
.ctx.actor = osd_ldiskfs_filldir,
-#endif
.it = it
};
down_read(&obj->oo_ext_idx_sem);
}
-#ifdef HAVE_DIR_CONTEXT
- buf.ctx.pos = filp->f_pos;
-#ifdef HAVE_ITERATE_SHARED
- rc = inode->i_fop->iterate_shared(filp, &buf.ctx);
-#else
- rc = inode->i_fop->iterate(filp, &buf.ctx);
-#endif
- filp->f_pos = buf.ctx.pos;
-#else
- rc = inode->i_fop->readdir(filp, &buf, osd_ldiskfs_filldir);
-#endif
+ filp->f_cred = current_cred();
+ rc = osd_security_file_alloc(filp);
+ if (rc)
+ RETURN(rc);
+
+ filp->f_flags |= O_NOATIME;
+ filp->f_mode |= FMODE_NONOTIFY;
+ rc = iterate_dir(filp, &buf.ctx);
+ if (rc)
+ RETURN(rc);
if (hlock != NULL)
ldiskfs_htree_unlock(hlock);
}
/**
- * It calls osd_ldiskfs_it_fill() which will use ->readdir()
+ * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
* to load a directory entry at a time and stored it in
* iterator's in-memory data structure.
*
ldp = (struct ldiskfs_dentry_param *)osd_oti_get(env)->oti_ldp;
osd_get_ldiskfs_dirent_param(ldp, fid);
dentry->d_fsdata = (void *)ldp;
- ll_vfs_dq_init(dir);
+ dquot_initialize(dir);
rc = osd_ldiskfs_add_entry(info, dev, jh, dentry, inode, hlock);
/*
* It is too bad, we cannot reinsert the name entry back.
}
/**
- * It calls osd_ldiskfs_it_fill() which will use ->readdir()
+ * It calls osd_ldiskfs_it_fill() which will use ->iterate*()
* to load a directory entry at a time and stored it i inn,
* in iterator's in-memory data structure.
*
o->od_read_cache = 1;
o->od_writethrough_cache = 1;
o->od_readcache_max_filesize = OSD_MAX_CACHE_SIZE;
-
+ o->od_readcache_max_iosize = OSD_READCACHE_MAX_IO_MB << 20;
+ o->od_writethrough_max_iosize = OSD_WRITECACHE_MAX_IO_MB << 20;
o->od_auto_scrub_interval = AS_DEFAULT;
cplen = strlcpy(o->od_svname, lustre_cfg_string(cfg, 4),
struct osd_device *osd = osd_dev(obd->obd_lu_dev);
struct super_block *sb = osd_sb(osd);
- return (osd->od_mnt == NULL || sb->s_flags & MS_RDONLY);
+ return (osd->od_mnt == NULL || sb->s_flags & SB_RDONLY);
}
/*
* lprocfs legacy support.
*/
-static struct obd_ops osd_obd_device_ops = {
+static const struct obd_ops osd_obd_device_ops = {
.o_owner = THIS_MODULE,
.o_connect = osd_obd_connect,
.o_disconnect = osd_obd_disconnect,
struct kobject *kobj;
int rc;
- CLASSERT(BH_DXLock < sizeof(((struct buffer_head *)0)->b_state) * 8);
+ BUILD_BUG_ON(BH_DXLock >=
+ sizeof(((struct buffer_head *)0)->b_state) * 8);
#if !defined(CONFIG_DEBUG_MUTEXES) && !defined(CONFIG_DEBUG_SPINLOCK)
/* please, try to keep osd_thread_info smaller than a page */
- CLASSERT(sizeof(struct osd_thread_info) <= PAGE_SIZE);
+ BUILD_BUG_ON(sizeof(struct osd_thread_info) > PAGE_SIZE);
#endif
osd_oi_mod_init();
return rc;
#ifdef CONFIG_KALLSYMS
+ priv_security_file_alloc =
+ (void *)kallsyms_lookup_name("security_file_alloc");
priv_dev_set_rdonly = (void *)kallsyms_lookup_name("dev_set_rdonly");
priv_dev_check_rdonly =
(void *)kallsyms_lookup_name("dev_check_rdonly");