]> git.proxmox.com Git - mirror_spl.git/blobdiff - module/spl/spl-vnode.c
Remove kern_path() wrapper
[mirror_spl.git] / module / spl / spl-vnode.c
index cd0fa2cd1eb83fbc35a5ff607645c232c2891f86..b9f9d7b1fabcdcf2b5551803e01540aa12678a49 100644 (file)
@@ -6,7 +6,7 @@
  *  UCRL-CODE-235197
  *
  *  This file is part of the SPL, Solaris Porting Layer.
- *  For details, see <http://github.com/behlendorf/spl/>.
+ *  For details, see <http://zfsonlinux.org/>.
  *
  *  The SPL is free software; you can redistribute it and/or modify it
  *  under the terms of the GNU General Public License as published by the
@@ -24,7 +24,9 @@
  *  Solaris Porting Layer (SPL) Vnode Implementation.
 \*****************************************************************************/
 
+#include <sys/cred.h>
 #include <sys/vnode.h>
+#include <linux/falloc.h>
 #include <spl-debug.h>
 
 #ifdef SS_DEBUG_SUBSYS
@@ -42,13 +44,6 @@ static spl_kmem_cache_t *vn_file_cache;
 static DEFINE_SPINLOCK(vn_file_lock);
 static LIST_HEAD(vn_file_list);
 
-#ifdef HAVE_KERN_PATH_PARENT_HEADER
-#ifndef HAVE_KERN_PATH_PARENT_SYMBOL
-kern_path_parent_t kern_path_parent_fn = SYMBOL_POISON;
-EXPORT_SYMBOL(kern_path_parent_fn);
-#endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
-#endif /* HAVE_KERN_PATH_PARENT_HEADER */
-
 vtype_t
 vn_mode_to_vtype(mode_t mode)
 {
@@ -170,7 +165,11 @@ vn_open(const char *path, uio_seg_t seg, int flags, int mode,
        if (IS_ERR(fp))
                SRETURN(-PTR_ERR(fp));
 
-       rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&fp->f_path, &stat);
+#else
+       rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
+#endif
        if (rc) {
                filp_close(fp, 0);
                SRETURN(-rc);
@@ -297,164 +296,213 @@ vn_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, void *ct)
 }
 EXPORT_SYMBOL(vn_seek);
 
-static struct dentry *
-vn_lookup_hash(struct nameidata *nd)
+/*
+ * spl_basename() takes a NULL-terminated string s as input containing a path.
+ * It returns a char pointer to a string and a length that describe the
+ * basename of the path. If the basename is not "." or "/", it will be an index
+ * into the string. While the string should be NULL terminated, the section
+ * referring to the basename is not. spl_basename is dual-licensed GPLv2+ and
+ * CC0. Anyone wishing to reuse it in another codebase may pick either license.
+ */
+static void
+spl_basename(const char *s, const char **str, int *len)
 {
-       return lookup_one_len((const char *)nd->last.name,
-                             nd->nd_dentry, nd->last.len);
-} /* lookup_hash() */
+       size_t i, end;
 
-static void
-vn_path_release(struct nameidata *nd)
+       ASSERT(str);
+       ASSERT(len);
+
+       if (!s || !*s) {
+               *str = ".";
+               *len = 1;
+               return;
+       }
+
+       i = strlen(s) - 1;
+
+       while (i && s[i--] == '/');
+
+       if (i == 0) {
+               *str = "/";
+               *len = 1;
+               return;
+       }
+
+       end = i;
+
+       for (end = i; i; i--) {
+               if (s[i] == '/') {
+                       *str = &s[i+1];
+                       *len = end - i + 1;
+                       return;
+               }
+       }
+
+       *str = s;
+       *len = end + 1;
+}
+
+static struct dentry *
+spl_kern_path_locked(const char *name, struct path *path)
 {
-       dput(nd->nd_dentry);
-       mntput(nd->nd_mnt);
+       struct path parent;
+       struct dentry *dentry;
+       const char *basename;
+       int len;
+       int rc;
+
+       ASSERT(name);
+       ASSERT(path);
+
+       spl_basename(name, &basename, &len);
+
+       /* We do not accept "." or ".." */
+       if (len <= 2 && basename[0] == '.')
+               if (len == 1 || basename[1] == '.')
+                       return (ERR_PTR(-EACCES));
+
+       rc = kern_path(name, LOOKUP_PARENT, &parent);
+       if (rc)
+               return (ERR_PTR(rc));
+
+       spl_inode_lock(parent.dentry->d_inode);
+
+       dentry = lookup_one_len(basename, parent.dentry, len);
+       if (IS_ERR(dentry)) {
+               spl_inode_unlock(parent.dentry->d_inode);
+               path_put(&parent);
+       } else {
+               *path = parent;
+       }
+
+       return (dentry);
 }
 
-/* Modified do_unlinkat() from linux/fs/namei.c, only uses exported symbols */
+/* Based on do_unlinkat() from linux/fs/namei.c */
 int
 vn_remove(const char *path, uio_seg_t seg, int flags)
 {
-        struct dentry *dentry;
-        struct nameidata nd;
-        struct inode *inode = NULL;
-        int rc = 0;
-        SENTRY;
+       struct dentry *dentry;
+       struct path parent;
+       struct inode *inode = NULL;
+       int rc = 0;
+       SENTRY;
+
+       ASSERT(seg == UIO_SYSSPACE);
+       ASSERT(flags == RMFILE);
 
-        ASSERT(seg == UIO_SYSSPACE);
-        ASSERT(flags == RMFILE);
+       dentry = spl_kern_path_locked(path, &parent);
+       rc = PTR_ERR(dentry);
+       if (!IS_ERR(dentry)) {
+               if (parent.dentry->d_name.name[parent.dentry->d_name.len])
+                       SGOTO(slashes, rc = 0);
+
+               inode = dentry->d_inode;
+               if (inode)
+                       atomic_inc(&inode->i_count);
+               else
+                       SGOTO(slashes, rc = 0);
 
-       rc = spl_kern_path_parent(path, &nd);
-        if (rc)
-                SGOTO(exit, rc);
-
-        rc = -EISDIR;
-        if (nd.last_type != LAST_NORM)
-                SGOTO(exit1, rc);
-
-        spl_inode_lock_nested(nd.nd_dentry->d_inode, I_MUTEX_PARENT);
-        dentry = vn_lookup_hash(&nd);
-        rc = PTR_ERR(dentry);
-        if (!IS_ERR(dentry)) {
-                /* Why not before? Because we want correct rc value */
-                if (nd.last.name[nd.last.len])
-                        SGOTO(slashes, rc);
-
-                inode = dentry->d_inode;
-                if (inode)
-                        atomic_inc(&inode->i_count);
 #ifdef HAVE_2ARGS_VFS_UNLINK
-                rc = vfs_unlink(nd.nd_dentry->d_inode, dentry);
+               rc = vfs_unlink(parent.dentry->d_inode, dentry);
 #else
-                rc = vfs_unlink(nd.nd_dentry->d_inode, dentry, nd.nd_mnt);
+               rc = vfs_unlink(parent.dentry->d_inode, dentry, NULL);
 #endif /* HAVE_2ARGS_VFS_UNLINK */
-exit2:
-                dput(dentry);
-        }
-
-        spl_inode_unlock(nd.nd_dentry->d_inode);
-        if (inode)
-                iput(inode);    /* truncate the inode here */
 exit1:
-        vn_path_release(&nd);
-exit:
-        SRETURN(-rc);
+               dput(dentry);
+       } else {
+               return (-rc);
+       }
+
+       spl_inode_unlock(parent.dentry->d_inode);
+       if (inode)
+               iput(inode);    /* truncate the inode here */
+
+       path_put(&parent);
+       SRETURN(-rc);
 
 slashes:
-        rc = !dentry->d_inode ? -ENOENT :
-                S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
-        SGOTO(exit2, rc);
+       rc = !dentry->d_inode ? -ENOENT :
+           S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
+       SGOTO(exit1, rc);
 } /* vn_remove() */
 EXPORT_SYMBOL(vn_remove);
 
-/* Modified do_rename() from linux/fs/namei.c, only uses exported symbols */
+/* Based on do_rename() from linux/fs/namei.c */
 int
 vn_rename(const char *oldname, const char *newname, int x1)
 {
-        struct dentry *old_dir, *new_dir;
-        struct dentry *old_dentry, *new_dentry;
-        struct dentry *trap;
-        struct nameidata oldnd, newnd;
-        int rc = 0;
+       struct dentry *old_dir, *new_dir;
+       struct dentry *old_dentry, *new_dentry;
+       struct dentry *trap;
+       struct path old_parent, new_parent;
+       int rc = 0;
        SENTRY;
 
-        rc = spl_kern_path_parent(oldname, &oldnd);
-        if (rc)
-                SGOTO(exit, rc);
+       old_dentry = spl_kern_path_locked(oldname, &old_parent);
+       if (IS_ERR(old_dentry))
+               SGOTO(exit, rc = PTR_ERR(old_dentry));
 
-        rc = spl_kern_path_parent(newname, &newnd);
-        if (rc)
-                SGOTO(exit1, rc);
-
-        rc = -EXDEV;
-        if (oldnd.nd_mnt != newnd.nd_mnt)
-                SGOTO(exit2, rc);
-
-        old_dir = oldnd.nd_dentry;
-        rc = -EBUSY;
-        if (oldnd.last_type != LAST_NORM)
-                SGOTO(exit2, rc);
-
-        new_dir = newnd.nd_dentry;
-        if (newnd.last_type != LAST_NORM)
-                SGOTO(exit2, rc);
-
-        trap = lock_rename(new_dir, old_dir);
-
-        old_dentry = vn_lookup_hash(&oldnd);
-
-        rc = PTR_ERR(old_dentry);
-        if (IS_ERR(old_dentry))
-                SGOTO(exit3, rc);
-
-        /* source must exist */
-        rc = -ENOENT;
-        if (!old_dentry->d_inode)
-                SGOTO(exit4, rc);
-
-        /* unless the source is a directory trailing slashes give -ENOTDIR */
-        if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
-                rc = -ENOTDIR;
-                if (oldnd.last.name[oldnd.last.len])
-                        SGOTO(exit4, rc);
-                if (newnd.last.name[newnd.last.len])
-                        SGOTO(exit4, rc);
-        }
-
-        /* source should not be ancestor of target */
-        rc = -EINVAL;
-        if (old_dentry == trap)
-                SGOTO(exit4, rc);
-
-        new_dentry = vn_lookup_hash(&newnd);
-        rc = PTR_ERR(new_dentry);
-        if (IS_ERR(new_dentry))
-                SGOTO(exit4, rc);
-
-        /* target should not be an ancestor of source */
-        rc = -ENOTEMPTY;
-        if (new_dentry == trap)
-                SGOTO(exit5, rc);
-
-#ifdef HAVE_4ARGS_VFS_RENAME
-        rc = vfs_rename(old_dir->d_inode, old_dentry,
-                        new_dir->d_inode, new_dentry);
+       spl_inode_unlock(old_parent.dentry->d_inode);
+
+       new_dentry = spl_kern_path_locked(newname, &new_parent);
+       if (IS_ERR(new_dentry))
+               SGOTO(exit2, rc = PTR_ERR(new_dentry));
+
+       spl_inode_unlock(new_parent.dentry->d_inode);
+
+       rc = -EXDEV;
+       if (old_parent.mnt != new_parent.mnt)
+               SGOTO(exit3, rc);
+
+       old_dir = old_parent.dentry;
+       new_dir = new_parent.dentry;
+       trap = lock_rename(new_dir, old_dir);
+
+       /* source should not be ancestor of target */
+       rc = -EINVAL;
+       if (old_dentry == trap)
+               SGOTO(exit4, rc);
+
+       /* target should not be an ancestor of source */
+       rc = -ENOTEMPTY;
+       if (new_dentry == trap)
+               SGOTO(exit4, rc);
+
+       /* source must exist */
+       rc = -ENOENT;
+       if (!old_dentry->d_inode)
+               SGOTO(exit4, rc);
+
+       /* unless the source is a directory trailing slashes give -ENOTDIR */
+       if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
+               rc = -ENOTDIR;
+               if (old_dentry->d_name.name[old_dentry->d_name.len])
+                       SGOTO(exit4, rc);
+               if (new_dentry->d_name.name[new_dentry->d_name.len])
+                       SGOTO(exit4, rc);
+       }
+
+#if defined(HAVE_4ARGS_VFS_RENAME)
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry);
+#elif defined(HAVE_5ARGS_VFS_RENAME)
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry, NULL);
 #else
-        rc = vfs_rename(old_dir->d_inode, old_dentry, oldnd.nd_mnt,
-                        new_dir->d_inode, new_dentry, newnd.nd_mnt);
-#endif /* HAVE_4ARGS_VFS_RENAME */
-exit5:
-        dput(new_dentry);
+       rc = vfs_rename(old_dir->d_inode, old_dentry,
+           new_dir->d_inode, new_dentry, NULL, 0);
+#endif
 exit4:
-        dput(old_dentry);
+       unlock_rename(new_dir, old_dir);
 exit3:
-        unlock_rename(new_dir, old_dir);
+       dput(new_dentry);
+       path_put(&new_parent);
 exit2:
-        vn_path_release(&newnd);
-exit1:
-        vn_path_release(&oldnd);
+       dput(old_dentry);
+       path_put(&old_parent);
 exit:
-        SRETURN(-rc);
+       SRETURN(-rc);
 }
 EXPORT_SYMBOL(vn_rename);
 
@@ -472,14 +520,18 @@ vn_getattr(vnode_t *vp, vattr_t *vap, int flags, void *x3, void *x4)
 
        fp = vp->v_file;
 
-        rc = vfs_getattr(fp->f_vfsmnt, fp->f_dentry, &stat);
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&fp->f_path, &stat);
+#else
+       rc = vfs_getattr(fp->f_path.mnt, fp->f_dentry, &stat);
+#endif
        if (rc)
                SRETURN(-rc);
 
        vap->va_type          = vn_mode_to_vtype(stat.mode);
        vap->va_mode          = stat.mode;
-       vap->va_uid           = stat.uid;
-       vap->va_gid           = stat.gid;
+       vap->va_uid           = KUID_TO_SUID(stat.uid);
+       vap->va_gid           = KGID_TO_SGID(stat.gid);
        vap->va_fsid          = 0;
        vap->va_nodeid        = stat.ino;
        vap->va_nlink         = stat.nlink;
@@ -510,6 +562,60 @@ int vn_fsync(vnode_t *vp, int flags, void *x3, void *x4)
 } /* vn_fsync() */
 EXPORT_SYMBOL(vn_fsync);
 
+int vn_space(vnode_t *vp, int cmd, struct flock *bfp, int flag,
+    offset_t offset, void *x6, void *x7)
+{
+       int error = EOPNOTSUPP;
+       SENTRY;
+
+       if (cmd != F_FREESP || bfp->l_whence != 0)
+               SRETURN(EOPNOTSUPP);
+
+       ASSERT(vp);
+       ASSERT(vp->v_file);
+       ASSERT(bfp->l_start >= 0 && bfp->l_len > 0);
+
+#ifdef FALLOC_FL_PUNCH_HOLE
+       /*
+        * When supported by the underlying file system preferentially
+        * use the fallocate() callback to preallocate the space.
+        */
+       error = -spl_filp_fallocate(vp->v_file,
+           FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE,
+           bfp->l_start, bfp->l_len);
+       if (error == 0)
+               SRETURN(0);
+#endif
+
+#ifdef HAVE_INODE_TRUNCATE_RANGE
+       if (vp->v_file->f_dentry && vp->v_file->f_dentry->d_inode &&
+           vp->v_file->f_dentry->d_inode->i_op &&
+           vp->v_file->f_dentry->d_inode->i_op->truncate_range) {
+               off_t end = bfp->l_start + bfp->l_len;
+               /*
+                * Judging from the code in shmem_truncate_range(),
+                * it seems the kernel expects the end offset to be
+                * inclusive and aligned to the end of a page.
+                */
+               if (end % PAGE_SIZE != 0) {
+                       end &= ~(off_t)(PAGE_SIZE - 1);
+                       if (end <= bfp->l_start)
+                               SRETURN(0);
+               }
+               --end;
+
+               vp->v_file->f_dentry->d_inode->i_op->truncate_range(
+                       vp->v_file->f_dentry->d_inode,
+                       bfp->l_start, end
+               );
+               SRETURN(0);
+       }
+#endif
+
+       SRETURN(error);
+}
+EXPORT_SYMBOL(vn_space);
+
 /* Function must be called while holding the vn_file_lock */
 static file_t *
 file_find(int fd)
@@ -570,7 +676,12 @@ vn_getf(int fd)
        if (vp == NULL)
                SGOTO(out_fget, rc);
 
-        if (vfs_getattr(lfp->f_vfsmnt, lfp->f_dentry, &stat))
+#ifdef HAVE_2ARGS_VFS_GETATTR
+       rc = vfs_getattr(&lfp->f_path, &stat);
+#else
+       rc = vfs_getattr(lfp->f_path.mnt, lfp->f_dentry, &stat);
+#endif
+        if (rc)
                SGOTO(out_vnode, rc);
 
        mutex_enter(&vp->v_lock);
@@ -639,62 +750,37 @@ vn_releasef(int fd)
 } /* releasef() */
 EXPORT_SYMBOL(releasef);
 
-#ifndef HAVE_SET_FS_PWD
-# ifdef HAVE_2ARGS_SET_FS_PWD
-/* Used from 2.6.25 - 2.6.31+ */
-void
-set_fs_pwd(struct fs_struct *fs, struct path *path)
+static void
+#ifdef HAVE_SET_FS_PWD_WITH_CONST
+vn_set_fs_pwd(struct fs_struct *fs, const struct path *path)
+#else
+vn_set_fs_pwd(struct fs_struct *fs, struct path *path)
+#endif /* HAVE_SET_FS_PWD_WITH_CONST */
 {
        struct path old_pwd;
 
-#  ifdef HAVE_FS_STRUCT_SPINLOCK
+#ifdef HAVE_FS_STRUCT_SPINLOCK
        spin_lock(&fs->lock);
        old_pwd = fs->pwd;
        fs->pwd = *path;
        path_get(path);
        spin_unlock(&fs->lock);
-#  else
+#else
        write_lock(&fs->lock);
        old_pwd = fs->pwd;
        fs->pwd = *path;
        path_get(path);
        write_unlock(&fs->lock);
-#  endif /* HAVE_FS_STRUCT_SPINLOCK */
+#endif /* HAVE_FS_STRUCT_SPINLOCK */
 
        if (old_pwd.dentry)
                path_put(&old_pwd);
 }
-# else
-/* Used from 2.6.11 - 2.6.24 */
-void
-set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, struct dentry *dentry)
-{
-        struct dentry *old_pwd;
-        struct vfsmount *old_pwdmnt;
-
-        write_lock(&fs->lock);
-        old_pwd = fs->pwd;
-        old_pwdmnt = fs->pwdmnt;
-        fs->pwdmnt = mntget(mnt);
-        fs->pwd = dget(dentry);
-        write_unlock(&fs->lock);
-
-        if (old_pwd) {
-                dput(old_pwd);
-                mntput(old_pwdmnt);
-        }
-}
-# endif /* HAVE_2ARGS_SET_FS_PWD */
-#endif /* HAVE_SET_FS_PWD */
 
 int
 vn_set_pwd(const char *filename)
 {
-#if defined(HAVE_2ARGS_SET_FS_PWD) && defined(HAVE_USER_PATH_DIR)
         struct path path;
-#else
-        struct nameidata nd;
-#endif /* HAVE_2ARGS_SET_FS_PWD */
         mm_segment_t saved_fs;
         int rc;
         SENTRY;
@@ -707,8 +793,6 @@ vn_set_pwd(const char *filename)
         saved_fs = get_fs();
         set_fs(get_ds());
 
-#ifdef HAVE_2ARGS_SET_FS_PWD
-# ifdef HAVE_USER_PATH_DIR
         rc = user_path_dir(filename, &path);
         if (rc)
                 SGOTO(out, rc);
@@ -717,40 +801,10 @@ vn_set_pwd(const char *filename)
         if (rc)
                 SGOTO(dput_and_out, rc);
 
-        set_fs_pwd(current->fs, &path);
+        vn_set_fs_pwd(current->fs, &path);
 
 dput_and_out:
         path_put(&path);
-# else
-        rc = __user_walk(filename,
-                         LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
-        if (rc)
-                SGOTO(out, rc);
-
-        rc = vfs_permission(&nd, MAY_EXEC);
-        if (rc)
-                SGOTO(dput_and_out, rc);
-
-        set_fs_pwd(current->fs, &nd.path);
-
-dput_and_out:
-        path_put(&nd.path);
-# endif /* HAVE_USER_PATH_DIR */
-#else
-        rc = __user_walk(filename,
-                         LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_CHDIR, &nd);
-        if (rc)
-                SGOTO(out, rc);
-
-        rc = vfs_permission(&nd, MAY_EXEC);
-        if (rc)
-                SGOTO(dput_and_out, rc);
-
-        set_fs_pwd(current->fs, nd.nd_mnt, nd.nd_dentry);
-
-dput_and_out:
-        vn_path_release(&nd);
-#endif /* HAVE_2ARGS_SET_FS_PWD */
 out:
        set_fs(saved_fs);
 
@@ -796,22 +850,6 @@ vn_file_cache_destructor(void *buf, void *cdrarg)
        mutex_destroy(&fp->f_lock);
 } /* vn_file_cache_destructor() */
 
-int spl_vn_init_kallsyms_lookup(void)
-{
-#ifdef HAVE_KERN_PATH_PARENT_HEADER
-#ifndef HAVE_KERN_PATH_PARENT_SYMBOL
-       kern_path_parent_fn = (kern_path_parent_t)
-               spl_kallsyms_lookup_name("kern_path_parent");
-       if (!kern_path_parent_fn) {
-               printk(KERN_ERR "Error: Unknown symbol kern_path_parent\n");
-               return -EFAULT;
-       }
-#endif /* HAVE_KERN_PATH_PARENT_SYMBOL */
-#endif /* HAVE_KERN_PATH_PARENT_HEADER */
-
-       return (0);
-}
-
 int
 spl_vn_init(void)
 {
@@ -845,13 +883,12 @@ spl_vn_fini(void)
                leaked++;
        }
 
-       kmem_cache_destroy(vn_file_cache);
-       vn_file_cache = NULL;
        spin_unlock(&vn_file_lock);
 
        if (leaked > 0)
                SWARN("Warning %d files leaked\n", leaked);
 
+       kmem_cache_destroy(vn_file_cache);
        kmem_cache_destroy(vn_cache);
 
        SEXIT;