This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / fs / fuse / dir.c
diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
new file mode 100644 (file)
index 0000000..8d7546e
--- /dev/null
@@ -0,0 +1,1216 @@
+/*
+  FUSE: Filesystem in Userspace
+  Copyright (C) 2001-2005  Miklos Szeredi <miklos@szeredi.hu>
+
+  This program can be distributed under the terms of the GNU GPL.
+  See the file COPYING.
+*/
+
+#include "fuse_i.h"
+
+#include <linux/pagemap.h>
+#include <linux/file.h>
+#include <linux/gfp.h>
+#include <linux/sched.h>
+#include <linux/namei.h>
+
+/*
+ * FUSE caches dentries and attributes with separate timeout.  The
+ * time in jiffies until the dentry/attributes are valid is stored in
+ * dentry->d_time and fuse_inode->i_time respectively.
+ */
+
+/*
+ * Calculate the time in jiffies until a dentry/attributes are valid
+ */
+static unsigned long time_to_jiffies(unsigned long sec, unsigned long nsec)
+{
+       struct timespec ts = {sec, nsec};
+       return jiffies + timespec_to_jiffies(&ts);
+}
+
+/*
+ * Set dentry and possibly attribute timeouts from the lookup/mk*
+ * replies
+ */
+static void fuse_change_timeout(struct dentry *entry, struct fuse_entry_out *o)
+{
+       entry->d_time = time_to_jiffies(o->entry_valid, o->entry_valid_nsec);
+       if (entry->d_inode)
+               get_fuse_inode(entry->d_inode)->i_time =
+                       time_to_jiffies(o->attr_valid, o->attr_valid_nsec);
+}
+
+/*
+ * Mark the attributes as stale, so that at the next call to
+ * ->getattr() they will be fetched from userspace
+ */
+void fuse_invalidate_attr(struct inode *inode)
+{
+       get_fuse_inode(inode)->i_time = jiffies - 1;
+}
+
+/*
+ * Just mark the entry as stale, so that a next attempt to look it up
+ * will result in a new lookup call to userspace
+ *
+ * This is called when a dentry is about to become negative and the
+ * timeout is unknown (unlink, rmdir, rename and in some cases
+ * lookup)
+ */
+static void fuse_invalidate_entry_cache(struct dentry *entry)
+{
+       entry->d_time = jiffies - 1;
+}
+
+/*
+ * Same as fuse_invalidate_entry_cache(), but also try to remove the
+ * dentry from the hash
+ */
+static void fuse_invalidate_entry(struct dentry *entry)
+{
+       d_invalidate(entry);
+       fuse_invalidate_entry_cache(entry);
+}
+
+static void fuse_lookup_init(struct fuse_req *req, struct inode *dir,
+                            struct dentry *entry,
+                            struct fuse_entry_out *outarg)
+{
+       req->in.h.opcode = FUSE_LOOKUP;
+       req->in.h.nodeid = get_node_id(dir);
+       req->inode = dir;
+       req->in.numargs = 1;
+       req->in.args[0].size = entry->d_name.len + 1;
+       req->in.args[0].value = entry->d_name.name;
+       req->out.numargs = 1;
+       req->out.args[0].size = sizeof(struct fuse_entry_out);
+       req->out.args[0].value = outarg;
+}
+
+/*
+ * Check whether the dentry is still valid
+ *
+ * If the entry validity timeout has expired and the dentry is
+ * positive, try to redo the lookup.  If the lookup results in a
+ * different inode, then let the VFS invalidate the dentry and redo
+ * the lookup once more.  If the lookup results in the same inode,
+ * then refresh the attributes, timeouts and mark the dentry valid.
+ */
+static int fuse_dentry_revalidate(struct dentry *entry, struct nameidata *nd)
+{
+       struct inode *inode = entry->d_inode;
+
+       if (inode && is_bad_inode(inode))
+               return 0;
+       else if (time_after(jiffies, entry->d_time)) {
+               int err;
+               struct fuse_entry_out outarg;
+               struct fuse_conn *fc;
+               struct fuse_req *req;
+
+               /* Doesn't hurt to "reset" the validity timeout */
+               fuse_invalidate_entry_cache(entry);
+
+               /* For negative dentries, always do a fresh lookup */
+               if (!inode)
+                       return 0;
+
+               fc = get_fuse_conn(inode);
+               req = fuse_get_req(fc);
+               if (IS_ERR(req))
+                       return 0;
+
+               fuse_lookup_init(req, entry->d_parent->d_inode, entry, &outarg);
+               request_send(fc, req);
+               err = req->out.h.error;
+               /* Zero nodeid is same as -ENOENT */
+               if (!err && !outarg.nodeid)
+                       err = -ENOENT;
+               if (!err) {
+                       struct fuse_inode *fi = get_fuse_inode(inode);
+                       if (outarg.nodeid != get_node_id(inode)) {
+                               fuse_send_forget(fc, req, outarg.nodeid, 1);
+                               return 0;
+                       }
+                       fi->nlookup ++;
+               }
+               fuse_put_request(fc, req);
+               if (err || (outarg.attr.mode ^ inode->i_mode) & S_IFMT)
+                       return 0;
+
+               fuse_change_attributes(inode, &outarg.attr);
+               fuse_change_timeout(entry, &outarg);
+       }
+       return 1;
+}
+
+/*
+ * Check if there's already a hashed alias of this directory inode.
+ * If yes, then lookup and mkdir must not create a new alias.
+ */
+static int dir_alias(struct inode *inode)
+{
+       if (S_ISDIR(inode->i_mode)) {
+               struct dentry *alias = d_find_alias(inode);
+               if (alias) {
+                       dput(alias);
+                       return 1;
+               }
+       }
+       return 0;
+}
+
+static int invalid_nodeid(u64 nodeid)
+{
+       return !nodeid || nodeid == FUSE_ROOT_ID;
+}
+
+static struct dentry_operations fuse_dentry_operations = {
+       .d_revalidate   = fuse_dentry_revalidate,
+};
+
+static int valid_mode(int m)
+{
+       return S_ISREG(m) || S_ISDIR(m) || S_ISLNK(m) || S_ISCHR(m) ||
+               S_ISBLK(m) || S_ISFIFO(m) || S_ISSOCK(m);
+}
+
+static struct dentry *fuse_lookup(struct inode *dir, struct dentry *entry,
+                                 struct nameidata *nd)
+{
+       int err;
+       struct fuse_entry_out outarg;
+       struct inode *inode = NULL;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req;
+
+       if (entry->d_name.len > FUSE_NAME_MAX)
+               return ERR_PTR(-ENAMETOOLONG);
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return ERR_PTR(PTR_ERR(req));
+
+       fuse_lookup_init(req, dir, entry, &outarg);
+       request_send(fc, req);
+       err = req->out.h.error;
+       /* Zero nodeid is same as -ENOENT, but with valid timeout */
+       if (!err && outarg.nodeid &&
+           (invalid_nodeid(outarg.nodeid) || !valid_mode(outarg.attr.mode)))
+               err = -EIO;
+       if (!err && outarg.nodeid) {
+               inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
+                                 &outarg.attr);
+               if (!inode) {
+                       fuse_send_forget(fc, req, outarg.nodeid, 1);
+                       return ERR_PTR(-ENOMEM);
+               }
+       }
+       fuse_put_request(fc, req);
+       if (err && err != -ENOENT)
+               return ERR_PTR(err);
+
+       if (inode && dir_alias(inode)) {
+               iput(inode);
+               return ERR_PTR(-EIO);
+       }
+       d_add(entry, inode);
+       entry->d_op = &fuse_dentry_operations;
+       if (!err)
+               fuse_change_timeout(entry, &outarg);
+       else
+               fuse_invalidate_entry_cache(entry);
+       return NULL;
+}
+
+/*
+ * Atomic create+open operation
+ *
+ * If the filesystem doesn't support this, then fall back to separate
+ * 'mknod' + 'open' requests.
+ */
+static int fuse_create_open(struct inode *dir, struct dentry *entry, int mode,
+                           struct nameidata *nd)
+{
+       int err;
+       struct inode *inode;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req;
+       struct fuse_open_in inarg;
+       struct fuse_open_out outopen;
+       struct fuse_entry_out outentry;
+       struct fuse_file *ff;
+       struct file *file;
+       int flags = nd->intent.open.flags - 1;
+
+       if (fc->no_create)
+               return -ENOSYS;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       err = -ENOMEM;
+       ff = fuse_file_alloc();
+       if (!ff)
+               goto out_put_request;
+
+       flags &= ~O_NOCTTY;
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.flags = flags;
+       inarg.mode = mode;
+       req->in.h.opcode = FUSE_CREATE;
+       req->in.h.nodeid = get_node_id(dir);
+       req->inode = dir;
+       req->in.numargs = 2;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = entry->d_name.len + 1;
+       req->in.args[1].value = entry->d_name.name;
+       req->out.numargs = 2;
+       req->out.args[0].size = sizeof(outentry);
+       req->out.args[0].value = &outentry;
+       req->out.args[1].size = sizeof(outopen);
+       req->out.args[1].value = &outopen;
+       request_send(fc, req);
+       err = req->out.h.error;
+       if (err) {
+               if (err == -ENOSYS)
+                       fc->no_create = 1;
+               goto out_free_ff;
+       }
+
+       err = -EIO;
+       if (!S_ISREG(outentry.attr.mode) || invalid_nodeid(outentry.nodeid))
+               goto out_free_ff;
+
+       inode = fuse_iget(dir->i_sb, outentry.nodeid, outentry.generation,
+                         &outentry.attr);
+       err = -ENOMEM;
+       if (!inode) {
+               flags &= ~(O_CREAT | O_EXCL | O_TRUNC);
+               ff->fh = outopen.fh;
+               /* Special release, with inode = NULL, this will
+                  trigger a 'forget' request when the release is
+                  complete */
+               fuse_send_release(fc, ff, outentry.nodeid, NULL, flags, 0);
+               goto out_put_request;
+       }
+       fuse_put_request(fc, req);
+       d_instantiate(entry, inode);
+       fuse_change_timeout(entry, &outentry);
+       file = lookup_instantiate_filp(nd, entry, generic_file_open);
+       if (IS_ERR(file)) {
+               ff->fh = outopen.fh;
+               fuse_send_release(fc, ff, outentry.nodeid, inode, flags, 0);
+               return PTR_ERR(file);
+       }
+       fuse_finish_open(inode, file, ff, &outopen);
+       return 0;
+
+ out_free_ff:
+       fuse_file_free(ff);
+ out_put_request:
+       fuse_put_request(fc, req);
+       return err;
+}
+
+/*
+ * Code shared between mknod, mkdir, symlink and link
+ */
+static int create_new_entry(struct fuse_conn *fc, struct fuse_req *req,
+                           struct inode *dir, struct dentry *entry,
+                           int mode)
+{
+       struct fuse_entry_out outarg;
+       struct inode *inode;
+       int err;
+
+       req->in.h.nodeid = get_node_id(dir);
+       req->inode = dir;
+       req->out.numargs = 1;
+       req->out.args[0].size = sizeof(outarg);
+       req->out.args[0].value = &outarg;
+       request_send(fc, req);
+       err = req->out.h.error;
+       if (err) {
+               fuse_put_request(fc, req);
+               return err;
+       }
+       err = -EIO;
+       if (invalid_nodeid(outarg.nodeid))
+               goto out_put_request;
+
+       if ((outarg.attr.mode ^ mode) & S_IFMT)
+               goto out_put_request;
+
+       inode = fuse_iget(dir->i_sb, outarg.nodeid, outarg.generation,
+                         &outarg.attr);
+       if (!inode) {
+               fuse_send_forget(fc, req, outarg.nodeid, 1);
+               return -ENOMEM;
+       }
+       fuse_put_request(fc, req);
+
+       if (dir_alias(inode)) {
+               iput(inode);
+               return -EIO;
+       }
+
+       d_instantiate(entry, inode);
+       fuse_change_timeout(entry, &outarg);
+       fuse_invalidate_attr(dir);
+       return 0;
+
+ out_put_request:
+       fuse_put_request(fc, req);
+       return err;
+}
+
+static int fuse_mknod(struct inode *dir, struct dentry *entry, int mode,
+                     dev_t rdev)
+{
+       struct fuse_mknod_in inarg;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.mode = mode;
+       inarg.rdev = new_encode_dev(rdev);
+       req->in.h.opcode = FUSE_MKNOD;
+       req->in.numargs = 2;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = entry->d_name.len + 1;
+       req->in.args[1].value = entry->d_name.name;
+       return create_new_entry(fc, req, dir, entry, mode);
+}
+
+static int fuse_create(struct inode *dir, struct dentry *entry, int mode,
+                      struct nameidata *nd)
+{
+       if (nd && (nd->flags & LOOKUP_CREATE)) {
+               int err = fuse_create_open(dir, entry, mode, nd);
+               if (err != -ENOSYS)
+                       return err;
+               /* Fall back on mknod */
+       }
+       return fuse_mknod(dir, entry, mode, 0);
+}
+
+static int fuse_mkdir(struct inode *dir, struct dentry *entry, int mode)
+{
+       struct fuse_mkdir_in inarg;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.mode = mode;
+       req->in.h.opcode = FUSE_MKDIR;
+       req->in.numargs = 2;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = entry->d_name.len + 1;
+       req->in.args[1].value = entry->d_name.name;
+       return create_new_entry(fc, req, dir, entry, S_IFDIR);
+}
+
+static int fuse_symlink(struct inode *dir, struct dentry *entry,
+                       const char *link)
+{
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       unsigned len = strlen(link) + 1;
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       req->in.h.opcode = FUSE_SYMLINK;
+       req->in.numargs = 2;
+       req->in.args[0].size = entry->d_name.len + 1;
+       req->in.args[0].value = entry->d_name.name;
+       req->in.args[1].size = len;
+       req->in.args[1].value = link;
+       return create_new_entry(fc, req, dir, entry, S_IFLNK);
+}
+
+static int fuse_unlink(struct inode *dir, struct dentry *entry)
+{
+       int err;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       req->in.h.opcode = FUSE_UNLINK;
+       req->in.h.nodeid = get_node_id(dir);
+       req->inode = dir;
+       req->in.numargs = 1;
+       req->in.args[0].size = entry->d_name.len + 1;
+       req->in.args[0].value = entry->d_name.name;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err) {
+               struct inode *inode = entry->d_inode;
+
+               /* Set nlink to zero so the inode can be cleared, if
+                   the inode does have more links this will be
+                   discovered at the next lookup/getattr */
+               inode->i_nlink = 0;
+               fuse_invalidate_attr(inode);
+               fuse_invalidate_attr(dir);
+               fuse_invalidate_entry_cache(entry);
+       } else if (err == -EINTR)
+               fuse_invalidate_entry(entry);
+       return err;
+}
+
+static int fuse_rmdir(struct inode *dir, struct dentry *entry)
+{
+       int err;
+       struct fuse_conn *fc = get_fuse_conn(dir);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       req->in.h.opcode = FUSE_RMDIR;
+       req->in.h.nodeid = get_node_id(dir);
+       req->inode = dir;
+       req->in.numargs = 1;
+       req->in.args[0].size = entry->d_name.len + 1;
+       req->in.args[0].value = entry->d_name.name;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err) {
+               entry->d_inode->i_nlink = 0;
+               fuse_invalidate_attr(dir);
+               fuse_invalidate_entry_cache(entry);
+       } else if (err == -EINTR)
+               fuse_invalidate_entry(entry);
+       return err;
+}
+
+static int fuse_rename(struct inode *olddir, struct dentry *oldent,
+                      struct inode *newdir, struct dentry *newent)
+{
+       int err;
+       struct fuse_rename_in inarg;
+       struct fuse_conn *fc = get_fuse_conn(olddir);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.newdir = get_node_id(newdir);
+       req->in.h.opcode = FUSE_RENAME;
+       req->in.h.nodeid = get_node_id(olddir);
+       req->inode = olddir;
+       req->inode2 = newdir;
+       req->in.numargs = 3;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = oldent->d_name.len + 1;
+       req->in.args[1].value = oldent->d_name.name;
+       req->in.args[2].size = newent->d_name.len + 1;
+       req->in.args[2].value = newent->d_name.name;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err) {
+               fuse_invalidate_attr(olddir);
+               if (olddir != newdir)
+                       fuse_invalidate_attr(newdir);
+
+               /* newent will end up negative */
+               if (newent->d_inode)
+                       fuse_invalidate_entry_cache(newent);
+       } else if (err == -EINTR) {
+               /* If request was interrupted, DEITY only knows if the
+                  rename actually took place.  If the invalidation
+                  fails (e.g. some process has CWD under the renamed
+                  directory), then there can be inconsistency between
+                  the dcache and the real filesystem.  Tough luck. */
+               fuse_invalidate_entry(oldent);
+               if (newent->d_inode)
+                       fuse_invalidate_entry(newent);
+       }
+
+       return err;
+}
+
+static int fuse_link(struct dentry *entry, struct inode *newdir,
+                    struct dentry *newent)
+{
+       int err;
+       struct fuse_link_in inarg;
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.oldnodeid = get_node_id(inode);
+       req->in.h.opcode = FUSE_LINK;
+       req->inode2 = inode;
+       req->in.numargs = 2;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = newent->d_name.len + 1;
+       req->in.args[1].value = newent->d_name.name;
+       err = create_new_entry(fc, req, newdir, newent, inode->i_mode);
+       /* Contrary to "normal" filesystems it can happen that link
+          makes two "logical" inodes point to the same "physical"
+          inode.  We invalidate the attributes of the old one, so it
+          will reflect changes in the backing inode (link count,
+          etc.)
+       */
+       if (!err || err == -EINTR)
+               fuse_invalidate_attr(inode);
+       return err;
+}
+
+int fuse_do_getattr(struct inode *inode)
+{
+       int err;
+       struct fuse_attr_out arg;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       req->in.h.opcode = FUSE_GETATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->out.numargs = 1;
+       req->out.args[0].size = sizeof(arg);
+       req->out.args[0].value = &arg;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err) {
+               if ((inode->i_mode ^ arg.attr.mode) & S_IFMT) {
+                       make_bad_inode(inode);
+                       err = -EIO;
+               } else {
+                       struct fuse_inode *fi = get_fuse_inode(inode);
+                       fuse_change_attributes(inode, &arg.attr);
+                       fi->i_time = time_to_jiffies(arg.attr_valid,
+                                                    arg.attr_valid_nsec);
+               }
+       }
+       return err;
+}
+
+/*
+ * Calling into a user-controlled filesystem gives the filesystem
+ * daemon ptrace-like capabilities over the requester process.  This
+ * means, that the filesystem daemon is able to record the exact
+ * filesystem operations performed, and can also control the behavior
+ * of the requester process in otherwise impossible ways.  For example
+ * it can delay the operation for arbitrary length of time allowing
+ * DoS against the requester.
+ *
+ * For this reason only those processes can call into the filesystem,
+ * for which the owner of the mount has ptrace privilege.  This
+ * excludes processes started by other users, suid or sgid processes.
+ */
+static int fuse_allow_task(struct fuse_conn *fc, struct task_struct *task)
+{
+       if (fc->flags & FUSE_ALLOW_OTHER)
+               return 1;
+
+       if (task->euid == fc->user_id &&
+           task->suid == fc->user_id &&
+           task->uid == fc->user_id &&
+           task->egid == fc->group_id &&
+           task->sgid == fc->group_id &&
+           task->gid == fc->group_id)
+               return 1;
+
+       return 0;
+}
+
+/*
+ * Check whether the inode attributes are still valid
+ *
+ * If the attribute validity timeout has expired, then fetch the fresh
+ * attributes with a 'getattr' request
+ *
+ * I'm not sure why cached attributes are never returned for the root
+ * inode, this is probably being too cautious.
+ */
+static int fuse_revalidate(struct dentry *entry)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_inode *fi = get_fuse_inode(inode);
+       struct fuse_conn *fc = get_fuse_conn(inode);
+
+       if (!fuse_allow_task(fc, current))
+               return -EACCES;
+       if (get_node_id(inode) != FUSE_ROOT_ID &&
+           time_before_eq(jiffies, fi->i_time))
+               return 0;
+
+       return fuse_do_getattr(inode);
+}
+
+static int fuse_access(struct inode *inode, int mask)
+{
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+       struct fuse_access_in inarg;
+       int err;
+
+       if (fc->no_access)
+               return 0;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.mask = mask;
+       req->in.h.opcode = FUSE_ACCESS;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 1;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (err == -ENOSYS) {
+               fc->no_access = 1;
+               err = 0;
+       }
+       return err;
+}
+
+/*
+ * Check permission.  The two basic access models of FUSE are:
+ *
+ * 1) Local access checking ('default_permissions' mount option) based
+ * on file mode.  This is the plain old disk filesystem permission
+ * modell.
+ *
+ * 2) "Remote" access checking, where server is responsible for
+ * checking permission in each inode operation.  An exception to this
+ * is if ->permission() was invoked from sys_access() in which case an
+ * access request is sent.  Execute permission is still checked
+ * locally based on file mode.
+ */
+static int fuse_permission(struct inode *inode, int mask, struct nameidata *nd)
+{
+       struct fuse_conn *fc = get_fuse_conn(inode);
+
+       if (!fuse_allow_task(fc, current))
+               return -EACCES;
+       else if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
+               int err = generic_permission(inode, mask, NULL);
+
+               /* If permission is denied, try to refresh file
+                  attributes.  This is also needed, because the root
+                  node will at first have no permissions */
+               if (err == -EACCES) {
+                       err = fuse_do_getattr(inode);
+                       if (!err)
+                               err = generic_permission(inode, mask, NULL);
+               }
+
+               /* Note: the opposite of the above test does not
+                  exist.  So if permissions are revoked this won't be
+                  noticed immediately, only after the attribute
+                  timeout has expired */
+
+               return err;
+       } else {
+               int mode = inode->i_mode;
+               if ((mask & MAY_EXEC) && !S_ISDIR(mode) && !(mode & S_IXUGO))
+                       return -EACCES;
+
+               if (nd && (nd->flags & LOOKUP_ACCESS))
+                       return fuse_access(inode, mask);
+               return 0;
+       }
+}
+
+static int parse_dirfile(char *buf, size_t nbytes, struct file *file,
+                        void *dstbuf, filldir_t filldir)
+{
+       while (nbytes >= FUSE_NAME_OFFSET) {
+               struct fuse_dirent *dirent = (struct fuse_dirent *) buf;
+               size_t reclen = FUSE_DIRENT_SIZE(dirent);
+               int over;
+               if (!dirent->namelen || dirent->namelen > FUSE_NAME_MAX)
+                       return -EIO;
+               if (reclen > nbytes)
+                       break;
+
+               over = filldir(dstbuf, dirent->name, dirent->namelen,
+                              file->f_pos, dirent->ino, dirent->type);
+               if (over)
+                       break;
+
+               buf += reclen;
+               nbytes -= reclen;
+               file->f_pos = dirent->off;
+       }
+
+       return 0;
+}
+
+static int fuse_readdir(struct file *file, void *dstbuf, filldir_t filldir)
+{
+       int err;
+       size_t nbytes;
+       struct page *page;
+       struct inode *inode = file->f_dentry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+
+       if (is_bad_inode(inode))
+               return -EIO;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       page = alloc_page(GFP_KERNEL);
+       if (!page) {
+               fuse_put_request(fc, req);
+               return -ENOMEM;
+       }
+       req->num_pages = 1;
+       req->pages[0] = page;
+       fuse_read_fill(req, file, inode, file->f_pos, PAGE_SIZE, FUSE_READDIR);
+       request_send(fc, req);
+       nbytes = req->out.args[0].size;
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err)
+               err = parse_dirfile(page_address(page), nbytes, file, dstbuf,
+                                   filldir);
+
+       __free_page(page);
+       fuse_invalidate_attr(inode); /* atime changed */
+       return err;
+}
+
+static char *read_link(struct dentry *dentry)
+{
+       struct inode *inode = dentry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req = fuse_get_req(fc);
+       char *link;
+
+       if (IS_ERR(req))
+               return ERR_PTR(PTR_ERR(req));
+
+       link = (char *) __get_free_page(GFP_KERNEL);
+       if (!link) {
+               link = ERR_PTR(-ENOMEM);
+               goto out;
+       }
+       req->in.h.opcode = FUSE_READLINK;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->out.argvar = 1;
+       req->out.numargs = 1;
+       req->out.args[0].size = PAGE_SIZE - 1;
+       req->out.args[0].value = link;
+       request_send(fc, req);
+       if (req->out.h.error) {
+               free_page((unsigned long) link);
+               link = ERR_PTR(req->out.h.error);
+       } else
+               link[req->out.args[0].size] = '\0';
+ out:
+       fuse_put_request(fc, req);
+       fuse_invalidate_attr(inode); /* atime changed */
+       return link;
+}
+
+static void free_link(char *link)
+{
+       if (!IS_ERR(link))
+               free_page((unsigned long) link);
+}
+
+static void *fuse_follow_link(struct dentry *dentry, struct nameidata *nd)
+{
+       nd_set_link(nd, read_link(dentry));
+       return NULL;
+}
+
+static void fuse_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
+{
+       free_link(nd_get_link(nd));
+}
+
+static int fuse_dir_open(struct inode *inode, struct file *file)
+{
+       return fuse_open_common(inode, file, 1);
+}
+
+static int fuse_dir_release(struct inode *inode, struct file *file)
+{
+       return fuse_release_common(inode, file, 1);
+}
+
+static int fuse_dir_fsync(struct file *file, struct dentry *de, int datasync)
+{
+       /* nfsd can call this with no file */
+       return file ? fuse_fsync_common(file, de, datasync, 1) : 0;
+}
+
+static void iattr_to_fattr(struct iattr *iattr, struct fuse_setattr_in *arg)
+{
+       unsigned ivalid = iattr->ia_valid;
+
+       if (ivalid & ATTR_MODE)
+               arg->valid |= FATTR_MODE,   arg->mode = iattr->ia_mode;
+       if (ivalid & ATTR_UID)
+               arg->valid |= FATTR_UID,    arg->uid = iattr->ia_uid;
+       if (ivalid & ATTR_GID)
+               arg->valid |= FATTR_GID,    arg->gid = iattr->ia_gid;
+       if (ivalid & ATTR_SIZE)
+               arg->valid |= FATTR_SIZE,   arg->size = iattr->ia_size;
+       /* You can only _set_ these together (they may change by themselves) */
+       if ((ivalid & (ATTR_ATIME | ATTR_MTIME)) == (ATTR_ATIME | ATTR_MTIME)) {
+               arg->valid |= FATTR_ATIME | FATTR_MTIME;
+               arg->atime = iattr->ia_atime.tv_sec;
+               arg->mtime = iattr->ia_mtime.tv_sec;
+       }
+       if (ivalid & ATTR_FILE) {
+               struct fuse_file *ff = iattr->ia_file->private_data;
+               arg->valid |= FATTR_FH;
+               arg->fh = ff->fh;
+       }
+}
+
+/*
+ * Set attributes, and at the same time refresh them.
+ *
+ * Truncation is slightly complicated, because the 'truncate' request
+ * may fail, in which case we don't want to touch the mapping.
+ * vmtruncate() doesn't allow for this case.  So do the rlimit
+ * checking by hand and call vmtruncate() only after the file has
+ * actually been truncated.
+ */
+static int fuse_setattr(struct dentry *entry, struct iattr *attr)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_inode *fi = get_fuse_inode(inode);
+       struct fuse_req *req;
+       struct fuse_setattr_in inarg;
+       struct fuse_attr_out outarg;
+       int err;
+       int is_truncate = 0;
+
+       if (fc->flags & FUSE_DEFAULT_PERMISSIONS) {
+               err = inode_change_ok(inode, attr);
+               if (err)
+                       return err;
+       }
+
+       if (attr->ia_valid & ATTR_SIZE) {
+               unsigned long limit;
+               is_truncate = 1;
+               limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
+               if (limit != RLIM_INFINITY && attr->ia_size > (loff_t) limit) {
+                       send_sig(SIGXFSZ, current, 0);
+                       return -EFBIG;
+               }
+       }
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       iattr_to_fattr(attr, &inarg);
+       req->in.h.opcode = FUSE_SETATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 1;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->out.numargs = 1;
+       req->out.args[0].size = sizeof(outarg);
+       req->out.args[0].value = &outarg;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (!err) {
+               if ((inode->i_mode ^ outarg.attr.mode) & S_IFMT) {
+                       make_bad_inode(inode);
+                       err = -EIO;
+               } else {
+                       if (is_truncate) {
+                               loff_t origsize = i_size_read(inode);
+                               i_size_write(inode, outarg.attr.size);
+                               if (origsize > outarg.attr.size)
+                                       vmtruncate(inode, outarg.attr.size);
+                       }
+                       fuse_change_attributes(inode, &outarg.attr);
+                       fi->i_time = time_to_jiffies(outarg.attr_valid,
+                                                    outarg.attr_valid_nsec);
+               }
+       } else if (err == -EINTR)
+               fuse_invalidate_attr(inode);
+
+       return err;
+}
+
+static int fuse_getattr(struct vfsmount *mnt, struct dentry *entry,
+                       struct kstat *stat)
+{
+       struct inode *inode = entry->d_inode;
+       int err = fuse_revalidate(entry);
+       if (!err)
+               generic_fillattr(inode, stat);
+
+       return err;
+}
+
+static int fuse_setxattr(struct dentry *entry, const char *name,
+                        const void *value, size_t size, int flags)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+       struct fuse_setxattr_in inarg;
+       int err;
+
+       if (fc->no_setxattr)
+               return -EOPNOTSUPP;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.size = size;
+       inarg.flags = flags;
+       req->in.h.opcode = FUSE_SETXATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 3;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = strlen(name) + 1;
+       req->in.args[1].value = name;
+       req->in.args[2].size = size;
+       req->in.args[2].value = value;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (err == -ENOSYS) {
+               fc->no_setxattr = 1;
+               err = -EOPNOTSUPP;
+       }
+       return err;
+}
+
+static ssize_t fuse_getxattr(struct dentry *entry, const char *name,
+                            void *value, size_t size)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+       struct fuse_getxattr_in inarg;
+       struct fuse_getxattr_out outarg;
+       ssize_t ret;
+
+       if (fc->no_getxattr)
+               return -EOPNOTSUPP;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.size = size;
+       req->in.h.opcode = FUSE_GETXATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 2;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       req->in.args[1].size = strlen(name) + 1;
+       req->in.args[1].value = name;
+       /* This is really two different operations rolled into one */
+       req->out.numargs = 1;
+       if (size) {
+               req->out.argvar = 1;
+               req->out.args[0].size = size;
+               req->out.args[0].value = value;
+       } else {
+               req->out.args[0].size = sizeof(outarg);
+               req->out.args[0].value = &outarg;
+       }
+       request_send(fc, req);
+       ret = req->out.h.error;
+       if (!ret)
+               ret = size ? req->out.args[0].size : outarg.size;
+       else {
+               if (ret == -ENOSYS) {
+                       fc->no_getxattr = 1;
+                       ret = -EOPNOTSUPP;
+               }
+       }
+       fuse_put_request(fc, req);
+       return ret;
+}
+
+static ssize_t fuse_listxattr(struct dentry *entry, char *list, size_t size)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+       struct fuse_getxattr_in inarg;
+       struct fuse_getxattr_out outarg;
+       ssize_t ret;
+
+       if (fc->no_listxattr)
+               return -EOPNOTSUPP;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       memset(&inarg, 0, sizeof(inarg));
+       inarg.size = size;
+       req->in.h.opcode = FUSE_LISTXATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 1;
+       req->in.args[0].size = sizeof(inarg);
+       req->in.args[0].value = &inarg;
+       /* This is really two different operations rolled into one */
+       req->out.numargs = 1;
+       if (size) {
+               req->out.argvar = 1;
+               req->out.args[0].size = size;
+               req->out.args[0].value = list;
+       } else {
+               req->out.args[0].size = sizeof(outarg);
+               req->out.args[0].value = &outarg;
+       }
+       request_send(fc, req);
+       ret = req->out.h.error;
+       if (!ret)
+               ret = size ? req->out.args[0].size : outarg.size;
+       else {
+               if (ret == -ENOSYS) {
+                       fc->no_listxattr = 1;
+                       ret = -EOPNOTSUPP;
+               }
+       }
+       fuse_put_request(fc, req);
+       return ret;
+}
+
+static int fuse_removexattr(struct dentry *entry, const char *name)
+{
+       struct inode *inode = entry->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(inode);
+       struct fuse_req *req;
+       int err;
+
+       if (fc->no_removexattr)
+               return -EOPNOTSUPP;
+
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
+
+       req->in.h.opcode = FUSE_REMOVEXATTR;
+       req->in.h.nodeid = get_node_id(inode);
+       req->inode = inode;
+       req->in.numargs = 1;
+       req->in.args[0].size = strlen(name) + 1;
+       req->in.args[0].value = name;
+       request_send(fc, req);
+       err = req->out.h.error;
+       fuse_put_request(fc, req);
+       if (err == -ENOSYS) {
+               fc->no_removexattr = 1;
+               err = -EOPNOTSUPP;
+       }
+       return err;
+}
+
+static struct inode_operations fuse_dir_inode_operations = {
+       .lookup         = fuse_lookup,
+       .mkdir          = fuse_mkdir,
+       .symlink        = fuse_symlink,
+       .unlink         = fuse_unlink,
+       .rmdir          = fuse_rmdir,
+       .rename         = fuse_rename,
+       .link           = fuse_link,
+       .setattr        = fuse_setattr,
+       .create         = fuse_create,
+       .mknod          = fuse_mknod,
+       .permission     = fuse_permission,
+       .getattr        = fuse_getattr,
+       .setxattr       = fuse_setxattr,
+       .getxattr       = fuse_getxattr,
+       .listxattr      = fuse_listxattr,
+       .removexattr    = fuse_removexattr,
+};
+
+static const struct file_operations fuse_dir_operations = {
+       .llseek         = generic_file_llseek,
+       .read           = generic_read_dir,
+       .readdir        = fuse_readdir,
+       .open           = fuse_dir_open,
+       .release        = fuse_dir_release,
+       .fsync          = fuse_dir_fsync,
+};
+
+static struct inode_operations fuse_common_inode_operations = {
+       .setattr        = fuse_setattr,
+       .permission     = fuse_permission,
+       .getattr        = fuse_getattr,
+       .setxattr       = fuse_setxattr,
+       .getxattr       = fuse_getxattr,
+       .listxattr      = fuse_listxattr,
+       .removexattr    = fuse_removexattr,
+};
+
+static struct inode_operations fuse_symlink_inode_operations = {
+       .setattr        = fuse_setattr,
+       .follow_link    = fuse_follow_link,
+       .put_link       = fuse_put_link,
+       .readlink       = generic_readlink,
+       .getattr        = fuse_getattr,
+       .setxattr       = fuse_setxattr,
+       .getxattr       = fuse_getxattr,
+       .listxattr      = fuse_listxattr,
+       .removexattr    = fuse_removexattr,
+};
+
+void fuse_init_common(struct inode *inode)
+{
+       inode->i_op = &fuse_common_inode_operations;
+}
+
+void fuse_init_dir(struct inode *inode)
+{
+       inode->i_op = &fuse_dir_inode_operations;
+       inode->i_fop = &fuse_dir_operations;
+}
+
+void fuse_init_symlink(struct inode *inode)
+{
+       inode->i_op = &fuse_symlink_inode_operations;
+}