X-Git-Url: https://scm.lunaixsky.com/lunaix-os.git/blobdiff_plain/bb7ce16533fb6c1384775dea6e1150e74c229daf..6f6da1abb22dff69dbb710bd2fd9d95f083f2b43:/lunaix-os/kernel/fs/vfs.c

diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c
index eaf35e0..cf376d2 100644
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -40,11 +40,10 @@
                     we can do similar thing in Lunaix? A block device emulation
                     above the regular file when we mount it on.
  10. (device) device number (dev_t) allocation
-            [good idea] <class>:<subclass>:<uniq_id> composition
+            [good idea] <class>:<subclass>:<uniq_id> composition (CHECKED)
 */
 
 #include <klibc/string.h>
-#include <lunaix/dirent.h>
 #include <lunaix/foptions.h>
 #include <lunaix/fs.h>
 #include <lunaix/mm/cake.h>
@@ -53,9 +52,12 @@
 #include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/syscall.h>
+#include <lunaix/syscall_utils.h>
 
 #include <lunaix/fs/twifs.h>
 
+#include <usr/lunaix/dirent_defs.h>
+
 static struct cake_pile* dnode_pile;
 static struct cake_pile* inode_pile;
 static struct cake_pile* file_pile;
@@ -108,14 +110,14 @@ vfs_init()
     atomic_fetch_add(&vfs_sysroot->ref_count, 1);
 }
 
-inline struct hbucket*
-__dcache_hash(struct v_dnode* parent, uint32_t* hash)
+static inline struct hbucket*
+__dcache_hash(struct v_dnode* parent, u32_t* hash)
 {
-    uint32_t _hash = *hash;
-    // 与parent的指针值做加法，来减小碰撞的可能性。
-    _hash += (uint32_t)parent;
+    u32_t _hash = *hash;
     // 确保低位更加随机
     _hash = _hash ^ (_hash >> VFS_HASHBITS);
+    // 与parent的指针值做加法，来减小碰撞的可能性。
+    _hash += (u32_t)parent;
     *hash = _hash;
     return &dnode_cache[_hash & VFS_HASH_MASK];
 }
@@ -130,7 +132,7 @@ vfs_dcache_lookup(struct v_dnode* parent, struct hstr* str)
         return parent->parent;
     }
 
-    uint32_t hash = str->hash;
+    u32_t hash = str->hash;
     struct hbucket* slot = __dcache_hash(parent, &hash);
 
     struct v_dnode *pos, *n;
@@ -163,6 +165,7 @@ vfs_dcache_remove(struct v_dnode* dnode)
     assert(dnode->ref_count == 1);
 
     llist_delete(&dnode->siblings);
+    llist_delete(&dnode->aka_list);
     hlist_delete(&dnode->hash_list);
 
     dnode->parent = NULL;
@@ -198,7 +201,7 @@ vfs_open(struct v_dnode* dnode, struct v_file** file)
     vfile->ref_count = ATOMIC_VAR_INIT(1);
     vfile->ops = inode->default_fops;
 
-    if ((inode->itype & VFS_IFFILE) && !inode->pg_cache) {
+    if ((inode->itype & F_MFILE) && !inode->pg_cache) {
         struct pcache* pcache = vzalloc(sizeof(struct pcache));
         pcache_init(pcache);
         pcache->master = inode;
@@ -225,8 +228,10 @@ void
 vfs_assign_inode(struct v_dnode* assign_to, struct v_inode* inode)
 {
     if (assign_to->inode) {
+        llist_delete(&assign_to->aka_list);
         assign_to->inode->link_count--;
     }
+    llist_append(&inode->aka_dnodes, &assign_to->aka_list);
     assign_to->inode = inode;
     inode->link_count++;
 }
@@ -254,12 +259,35 @@ vfs_link(struct v_dnode* to_link, struct v_dnode* name)
 }
 
 int
-vfs_close(struct v_file* file)
+vfs_pclose(struct v_file* file, pid_t pid)
 {
     int errno = 0;
-    if (!(errno = file->ops->close(file))) {
+    if (file->ref_count > 1) {
+        atomic_fetch_sub(&file->ref_count, 1);
+    } else if (!(errno = file->ops->close(file))) {
         atomic_fetch_sub(&file->dnode->ref_count, 1);
         file->inode->open_count--;
+
+        /*
+         * Prevent dead lock.
+         * This happened when process is terminated while blocking on read.
+         * In that case, the process is still holding the inode lock and it
+             will never get released.
+         * The unlocking should also include ownership check.
+         *
+         * To see why, consider two process both open the same file both with
+         * fd=x.
+         *      Process A: busy on reading x
+         *      Process B: do nothing with x
+         * Assuming that, after a very short time, process B get terminated
+         * while process A is still busy in it's reading business. By this
+         * design, the inode lock of this file x is get released by B rather
+         * than A. And this will cause a probable race condition on A if other
+         * process is writing to this file later after B exit.
+         */
+        if (mutex_on_hold(&file->inode->lock)) {
+            mutex_unlock_for(&file->inode->lock, pid);
+        }
         mnt_chillax(file->dnode->mnt);
 
         pcache_commit_all(file->inode);
@@ -268,6 +296,18 @@ vfs_close(struct v_file* file)
     return errno;
 }
 
+int
+vfs_close(struct v_file* file)
+{
+    return vfs_pclose(file, __current->pid);
+}
+
+void
+vfs_free_fd(struct v_fd* fd)
+{
+    cake_release(fd_pile, fd);
+}
+
 int
 vfs_fsync(struct v_file* file)
 {
@@ -358,6 +398,7 @@ vfs_d_alloc(struct v_dnode* parent, struct hstr* name)
     memset(dnode, 0, sizeof(*dnode));
     llist_init_head(&dnode->children);
     llist_init_head(&dnode->siblings);
+    llist_init_head(&dnode->aka_list);
     mutex_init(&dnode->lock);
 
     dnode->ref_count = ATOMIC_VAR_INIT(0);
@@ -395,12 +436,12 @@ vfs_d_free(struct v_dnode* dnode)
         vfs_dcache_remove(pos);
     }
 
-    vfree(dnode->name.value);
+    vfree((void*)dnode->name.value);
     cake_release(dnode_pile, dnode);
 }
 
 struct v_inode*
-vfs_i_find(struct v_superblock* sb, uint32_t i_id)
+vfs_i_find(struct v_superblock* sb, u32_t i_id)
 {
     struct hbucket* slot = &sb->i_cache[i_id & VFS_HASH_MASK];
     struct v_inode *pos, *n;
@@ -440,6 +481,7 @@ vfs_i_alloc(struct v_superblock* sb)
     memset(inode, 0, sizeof(*inode));
     mutex_init(&inode->lock);
     llist_init_head(&inode->xattrs);
+    llist_init_head(&inode->aka_dnodes);
 
     sb->ops.init_inode(sb, inode);
 
@@ -448,7 +490,6 @@ vfs_i_alloc(struct v_superblock* sb)
     inode->atime = inode->ctime;
     inode->mtime = inode->ctime;
 
-done:
     lru_use_one(inode_lru, &inode->lru);
     return inode;
 }
@@ -460,7 +501,12 @@ vfs_i_free(struct v_inode* inode)
         pcache_release(inode->pg_cache);
         vfree(inode->pg_cache);
     }
-    inode->ops->sync(inode);
+    // we don't need to sync inode.
+    // If an inode can be free, then it must be properly closed.
+    // Hence it must be synced already!
+    if (inode->destruct) {
+        inode->destruct(inode);
+    }
     hlist_delete(&inode->hash_list);
     cake_release(inode_pile, inode);
 }
@@ -468,6 +514,8 @@ vfs_i_free(struct v_inode* inode)
 /* ---- System call definition and support ---- */
 
 #define FLOCATE_CREATE_EMPTY 1
+#define FLOCATE_CREATE_ONLY 2
+#define FLOCATE_NOFOLLOW 4
 
 int
 vfs_getfd(int fd, struct v_fd** fd_s)
@@ -486,15 +534,25 @@ __vfs_try_locate_file(const char* path,
 {
     char name_str[VFS_NAME_MAXLEN];
     struct hstr name = HSTR(name_str, 0);
-    int errno;
+    int errno, woption = 0;
+
+    if ((options & FLOCATE_NOFOLLOW)) {
+        woption |= VFS_WALK_NOFOLLOW;
+    }
 
     name_str[0] = 0;
-    if ((errno = vfs_walk_proc(path, fdir, &name, VFS_WALK_PARENT))) {
+    if ((errno = vfs_walk_proc(path, fdir, &name, woption | VFS_WALK_PARENT))) {
         return errno;
     }
 
-    errno = vfs_walk(*fdir, name.value, file, NULL, 0);
-    if (errno != ENOENT || !(options & FLOCATE_CREATE_EMPTY)) {
+    errno = vfs_walk(*fdir, name.value, file, NULL, woption);
+
+    if (errno != ENOENT && (options & FLOCATE_CREATE_ONLY)) {
+        return EEXIST;
+    }
+
+    if (errno != ENOENT ||
+        !(options & (FLOCATE_CREATE_EMPTY | FLOCATE_CREATE_ONLY))) {
         return errno;
     }
 
@@ -522,21 +580,27 @@ __vfs_try_locate_file(const char* path,
 int
 vfs_do_open(const char* path, int options)
 {
-    int errno, fd;
+    int errno, fd, loptions = 0;
     struct v_dnode *dentry, *file;
-    struct v_file* ofile = 0;
-
-    errno = __vfs_try_locate_file(
-      path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0);
+    struct v_file* ofile = NULL;
 
-    if (errno || (errno = vfs_open(file, &ofile))) {
-        return errno;
+    if ((options & FO_CREATE)) {
+        loptions |= FLOCATE_CREATE_EMPTY;
+    } else if ((options & FO_NOFOLLOW)) {
+        loptions |= FLOCATE_NOFOLLOW;
     }
 
-    struct v_inode* o_inode = ofile->inode;
+    errno = __vfs_try_locate_file(path, &dentry, &file, loptions);
 
     if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
-        struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
+
+        if (errno || (errno = vfs_open(file, &ofile))) {
+            return errno;
+        }
+
+        struct v_fd* fd_s = cake_grab(fd_pile);
+        memset(fd_s, 0, sizeof(*fd_s));
+
         ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0);
         fd_s->file = ofile;
         fd_s->flags = options;
@@ -561,13 +625,11 @@ __DEFINE_LXSYSCALL1(int, close, int, fd)
         goto done_err;
     }
 
-    if (fd_s->file->ref_count > 1) {
-        fd_s->file->ref_count--;
-    } else if ((errno = vfs_close(fd_s->file))) {
+    if ((errno = vfs_close(fd_s->file))) {
         goto done_err;
     }
 
-    vfree(fd_s);
+    cake_release(fd_pile, fd_s);
     __current->fdtable->fds[fd] = 0;
 
 done_err:
@@ -580,13 +642,13 @@ __vfs_readdir_callback(struct dir_context* dctx,
                        const int len,
                        const int dtype)
 {
-    struct dirent* dent = (struct dirent*)dctx->cb_data;
+    struct lx_dirent* dent = (struct lx_dirent*)dctx->cb_data;
     strncpy(dent->d_name, name, DIRENT_NAME_MAX_LEN);
     dent->d_nlen = len;
     dent->d_type = dtype;
 }
 
-__DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
+__DEFINE_LXSYSCALL2(int, sys_readdir, int, fd, struct lx_dirent*, dent)
 {
     struct v_fd* fd_s;
     int errno;
@@ -599,7 +661,7 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
 
     lock_inode(inode);
 
-    if (!(inode->itype & VFS_IFDIR)) {
+    if ((inode->itype & F_FILE)) {
         errno = ENOTDIR;
     } else {
         struct dir_context dctx =
@@ -609,9 +671,9 @@ __DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
                                   __vfs_readdir_callback };
         errno = 1;
         if (dent->d_offset == 0) {
-            __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, 0);
+            __vfs_readdir_callback(&dctx, vfs_dot.value, vfs_dot.len, DT_DIR);
         } else if (dent->d_offset == 1) {
-            __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, 0);
+            __vfs_readdir_callback(&dctx, vfs_ddot.value, vfs_ddot.len, DT_DIR);
         } else {
             dctx.index -= 2;
             if ((errno = fd_s->file->ops->readdir(fd_s->file, &dctx)) != 1) {
@@ -637,7 +699,7 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
     }
 
     struct v_file* file = fd_s->file;
-    if ((file->inode->itype & VFS_IFDIR)) {
+    if (!(file->inode->itype & F_FILE)) {
         errno = EISDIR;
         goto done;
     }
@@ -678,7 +740,7 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
         goto done;
     }
 
-    if ((file->inode->itype & VFS_IFDIR)) {
+    if (!(file->inode->itype & F_FILE)) {
         errno = EISDIR;
         goto done;
     }
@@ -687,13 +749,11 @@ __DEFINE_LXSYSCALL3(int, write, int, fd, void*, buf, size_t, count)
 
     file->inode->mtime = clock_unixtime();
 
-    __SYSCALL_INTERRUPTIBLE({
-        if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
-            errno = file->ops->write(file->inode, buf, count, file->f_pos);
-        } else {
-            errno = pcache_write(file->inode, buf, count, file->f_pos);
-        }
-    })
+    if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
+        errno = file->ops->write(file->inode, buf, count, file->f_pos);
+    } else {
+        errno = pcache_write(file->inode, buf, count, file->f_pos);
+    }
 
     if (errno > 0) {
         file->f_pos += errno;
@@ -753,7 +813,7 @@ done:
 int
 vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth)
 {
-    if (!dnode || dnode->parent == dnode) {
+    if (!dnode) {
         return 0;
     }
 
@@ -761,13 +821,19 @@ vfs_get_path(struct v_dnode* dnode, char* buf, size_t size, int depth)
         return ENAMETOOLONG;
     }
 
-    size_t len = vfs_get_path(dnode->parent, buf, size, depth + 1);
+    size_t len = 0;
+
+    if (dnode->parent != dnode) {
+        len = vfs_get_path(dnode->parent, buf, size, depth + 1);
+    }
 
     if (len >= size) {
         return len;
     }
 
-    buf[len++] = VFS_PATH_DELIM;
+    if (!len || buf[len - 1] != VFS_PATH_DELIM) {
+        buf[len++] = VFS_PATH_DELIM;
+    }
 
     size_t cpy_size = MIN(dnode->name.len, size - len);
     strncpy(buf + len, dnode->name.value, cpy_size);
@@ -793,6 +859,18 @@ vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
     return 0;
 }
 
+int
+vfs_get_dtype(int itype)
+{
+    if ((itype & VFS_IFSYMLINK)) {
+        return DT_SYMLINK;
+    } else if (!(itype & VFS_IFFILE)) {
+        return DT_DIR;
+    } else {
+        return DT_FILE;
+    }
+}
+
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 {
     int errno;
@@ -844,6 +922,8 @@ __DEFINE_LXSYSCALL4(int,
         goto done;
     }
 
+    pathname = pathname ? pathname : "";
+
     struct v_dnode* dnode;
     if (!(errno = vfs_walk(
             fd_s->file->dnode, pathname, &dnode, NULL, VFS_WALK_NOFOLLOW))) {
@@ -905,7 +985,7 @@ __DEFINE_LXSYSCALL1(int, rmdir, const char*, pathname)
     lock_dnode(parent);
     lock_inode(parent->inode);
 
-    if ((dnode->inode->itype & VFS_IFDIR)) {
+    if (!(dnode->inode->itype & F_MFILE)) {
         errno = parent->inode->ops->rmdir(parent->inode, dnode);
         if (!errno) {
             vfs_dcache_remove(dnode);
@@ -949,7 +1029,7 @@ __DEFINE_LXSYSCALL1(int, mkdir, const char*, path)
         errno = ENOTSUP;
     } else if (!parent->inode->ops->mkdir) {
         errno = ENOTSUP;
-    } else if (!(parent->inode->itype & VFS_IFDIR)) {
+    } else if ((parent->inode->itype & F_FILE)) {
         errno = ENOTDIR;
     } else if (!(errno = parent->inode->ops->mkdir(parent->inode, dir))) {
         vfs_dcache_add(parent, dir);
@@ -983,9 +1063,7 @@ __vfs_do_unlink(struct v_dnode* dnode)
 
     if (inode->open_count) {
         errno = EBUSY;
-    } else if (!(inode->itype & VFS_IFDIR)) {
-        // The underlying unlink implementation should handle
-        //  symlink case
+    } else if ((inode->itype & F_MFILE)) {
         errno = inode->ops->unlink(inode);
         if (!errno) {
             vfs_d_free(dnode);
@@ -1038,10 +1116,8 @@ __DEFINE_LXSYSCALL2(int, link, const char*, oldpath, const char*, newpath)
     errno = __vfs_try_locate_file(oldpath, &dentry, &to_link, 0);
     if (!errno) {
         errno = __vfs_try_locate_file(
-          newpath, &name_dentry, &name_file, FLOCATE_CREATE_EMPTY);
+          newpath, &name_dentry, &name_file, FLOCATE_CREATE_ONLY);
         if (!errno) {
-            errno = EEXIST;
-        } else if (name_file) {
             errno = vfs_link(to_link, name_file);
         }
     }
@@ -1138,50 +1214,69 @@ __DEFINE_LXSYSCALL2(int,
                     link_target)
 {
     int errno;
-    struct v_dnode* dnode;
-    if ((errno = vfs_walk_proc(pathname, &dnode, NULL, 0))) {
+    struct v_dnode *dnode, *file;
+    if ((errno = __vfs_try_locate_file(
+           pathname, &dnode, &file, FLOCATE_CREATE_ONLY))) {
         goto done;
     }
 
-    if (errno = vfs_check_writable(dnode)) {
+    if ((errno = vfs_check_writable(file))) {
         goto done;
     }
 
-    if (!dnode->inode->ops->set_symlink) {
+    if (!file->inode->ops->set_symlink) {
         errno = ENOTSUP;
         goto done;
     }
 
-    lock_inode(dnode->inode);
+    lock_inode(file->inode);
 
-    errno = dnode->inode->ops->set_symlink(dnode->inode, link_target);
+    errno = file->inode->ops->set_symlink(file->inode, link_target);
 
-    unlock_inode(dnode->inode);
+    unlock_inode(file->inode);
 
 done:
     return DO_STATUS(errno);
 }
 
+void
+vfs_ref_file(struct v_file* file)
+{
+    atomic_fetch_add(&file->ref_count, 1);
+}
+
+void
+vfs_ref_dnode(struct v_dnode* dnode)
+{
+    atomic_fetch_add(&dnode->ref_count, 1);
+    mnt_mkbusy(dnode->mnt);
+}
+
+void
+vfs_unref_dnode(struct v_dnode* dnode)
+{
+    atomic_fetch_sub(&dnode->ref_count, 1);
+    mnt_chillax(dnode->mnt);
+}
+
 int
-__vfs_do_chdir(struct v_dnode* dnode)
+vfs_do_chdir(struct proc_info* proc, struct v_dnode* dnode)
 {
     int errno = 0;
 
     lock_dnode(dnode);
 
-    if (!(dnode->inode->itype & VFS_IFDIR)) {
+    if ((dnode->inode->itype & F_FILE)) {
         errno = ENOTDIR;
         goto done;
     }
 
-    if (__current->cwd) {
-        atomic_fetch_sub(&__current->cwd->ref_count, 1);
-        mnt_chillax(__current->cwd->mnt);
+    if (proc->cwd) {
+        vfs_unref_dnode(proc->cwd);
     }
 
-    atomic_fetch_add(&dnode->ref_count, 1);
-    mnt_mkbusy(dnode->mnt);
-    __current->cwd = dnode;
+    vfs_ref_dnode(dnode);
+    proc->cwd = dnode;
 
     unlock_dnode(dnode);
 
@@ -1198,7 +1293,7 @@ __DEFINE_LXSYSCALL1(int, chdir, const char*, path)
         goto done;
     }
 
-    errno = __vfs_do_chdir(dnode);
+    errno = vfs_do_chdir((struct proc_info*)__current, dnode);
 
 done:
     return DO_STATUS(errno);
@@ -1213,7 +1308,7 @@ __DEFINE_LXSYSCALL1(int, fchdir, int, fd)
         goto done;
     }
 
-    errno = __vfs_do_chdir(fd_s->file->dnode);
+    errno = vfs_do_chdir((struct proc_info*)__current, fd_s->file->dnode);
 
 done:
     return DO_STATUS(errno);
@@ -1241,7 +1336,7 @@ __DEFINE_LXSYSCALL2(char*, getcwd, char*, buf, size_t, size)
         }
     }
 
-    buf[len + 1] = '\0';
+    buf[len] = '\0';
 
     ret_ptr = buf;
 
@@ -1259,7 +1354,7 @@ vfs_do_rename(struct v_dnode* current, struct v_dnode* target)
         return 0;
     }
 
-    if (errno = vfs_check_writable(current)) {
+    if ((errno = vfs_check_writable(current))) {
         return errno;
     }
 
@@ -1343,6 +1438,48 @@ __DEFINE_LXSYSCALL2(int, rename, const char*, oldpath, const char*, newpath)
     errno = vfs_do_rename(cur, target);
 
 done:
-    vfree(name.value);
+    vfree((void*)name.value);
+    return DO_STATUS(errno);
+}
+
+__DEFINE_LXSYSCALL2(int, fstat, int, fd, struct file_stat*, stat)
+{
+    int errno = 0;
+    struct v_fd* fds;
+
+    if ((errno = vfs_getfd(fd, &fds))) {
+        goto done;
+    }
+
+    struct v_inode* vino = fds->file->inode;
+    struct device* fdev = vino->sb->dev;
+
+    *stat = (struct file_stat){ .st_ino = vino->id,
+                                .st_blocks = vino->lb_usage,
+                                .st_size = vino->fsize,
+                                .mode = vino->itype,
+                                .st_ioblksize = PG_SIZE,
+                                .st_blksize = vino->sb->blksize };
+
+    if (VFS_DEVFILE(vino->itype)) {
+        struct device* rdev = (struct device*)vino->data;
+        if (!rdev || rdev->magic != DEV_STRUCT_MAGIC) {
+            errno = EINVAL;
+            goto done;
+        }
+
+        stat->st_rdev = (dev_t){ .meta = rdev->class->meta,
+                                 .devident = device_id_from_class(rdev->class),
+                                 .dev_uid = rdev->dev_uid };
+    }
+
+    if (fdev) {
+        u32_t devident = device_id_from_class(fdev->class);
+        stat->st_dev = (dev_t){ .meta = fdev->class->meta,
+                                .devident = devident,
+                                .dev_uid = fdev->dev_uid };
+    }
+
+done:
     return DO_STATUS(errno);
 }
\ No newline at end of file