X-Git-Url: https://scm.lunaixsky.com/lunaix-os.git/blobdiff_plain/96cf56ea15a22b8b362aac6c3ab202d2608e9151..642855f81fd03b9fd6540ac99c665b57b4b38cc8:/lunaix-os/kernel/fs/vfs.c

diff --git a/lunaix-os/kernel/fs/vfs.c b/lunaix-os/kernel/fs/vfs.c
index 83bd0bd..cbe3c39 100644
--- a/lunaix-os/kernel/fs/vfs.c
+++ b/lunaix-os/kernel/fs/vfs.c
@@ -53,6 +53,7 @@
 #include <lunaix/process.h>
 #include <lunaix/spike.h>
 #include <lunaix/syscall.h>
+#include <lunaix/syscall_utils.h>
 
 #include <lunaix/fs/twifs.h>
 
@@ -109,13 +110,13 @@ vfs_init()
 }
 
 inline struct hbucket*
-__dcache_hash(struct v_dnode* parent, uint32_t* hash)
+__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 +131,7 @@ 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 +164,7 @@ 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;
@@ -225,8 +227,10 @@ 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 +258,35 @@ 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 +295,18 @@ 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 +397,7 @@ 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);
@@ -367,6 +407,7 @@ vfs_d_alloc(struct v_dnode* parent, struct hstr* name)
 
     if (parent) {
         dnode->super_block = parent->super_block;
+        dnode->mnt = parent->mnt;
     }
 
     lru_use_one(dnode_lru, &dnode->lru);
@@ -399,7 +440,7 @@ vfs_d_free(struct v_dnode* 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;
@@ -439,6 +480,7 @@ 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);
 
@@ -459,7 +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);
 }
@@ -486,6 +533,8 @@ __vfs_try_locate_file(const char* path,
     char name_str[VFS_NAME_MAXLEN];
     struct hstr name = HSTR(name_str, 0);
     int errno;
+
+    name_str[0] = 0;
     if ((errno = vfs_walk_proc(path, fdir, &name, VFS_WALK_PARENT))) {
         return errno;
     }
@@ -521,19 +570,20 @@ vfs_do_open(const char* path, int options)
 {
     int errno, fd;
     struct v_dnode *dentry, *file;
-    struct v_file* ofile = 0;
+    struct v_file* ofile = NULL;
 
     errno = __vfs_try_locate_file(
       path, &dentry, &file, (options & FO_CREATE) ? FLOCATE_CREATE_EMPTY : 0);
 
-    if (errno || (errno = vfs_open(file, &ofile))) {
-        return errno;
-    }
+    if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
 
-    struct v_inode* o_inode = ofile->inode;
+        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));
 
-    if (!errno && !(errno = vfs_alloc_fdslot(&fd))) {
-        struct v_fd* fd_s = vzalloc(sizeof(*fd_s));
         ofile->f_pos = ofile->inode->fsize & -((options & FO_APPEND) != 0);
         fd_s->file = ofile;
         fd_s->flags = options;
@@ -558,13 +608,11 @@ __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:
@@ -583,7 +631,7 @@ __vfs_readdir_callback(struct dir_context* dctx,
     dent->d_type = dtype;
 }
 
-__DEFINE_LXSYSCALL2(int, readdir, int, fd, struct dirent*, dent)
+__DEFINE_LXSYSCALL2(int, sys_readdir, int, fd, struct dirent*, dent)
 {
     struct v_fd* fd_s;
     int errno;
@@ -606,9 +654,9 @@ __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) {
@@ -643,13 +691,11 @@ __DEFINE_LXSYSCALL3(int, read, int, fd, void*, buf, size_t, count)
 
     file->inode->atime = clock_unixtime();
 
-    __SYSCALL_INTERRUPTIBLE({
-        if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
-            errno = file->ops->read(file->inode, buf, count, file->f_pos);
-        } else {
-            errno = pcache_read(file->inode, buf, count, file->f_pos);
-        }
-    })
+    if ((file->inode->itype & VFS_IFSEQDEV) || (fd_s->flags & FO_DIRECT)) {
+        errno = file->ops->read(file->inode, buf, count, file->f_pos);
+    } else {
+        errno = pcache_read(file->inode, buf, count, file->f_pos);
+    }
 
     if (errno > 0) {
         file->f_pos += errno;
@@ -686,13 +732,11 @@ __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;
@@ -752,7 +796,7 @@ 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;
     }
 
@@ -760,13 +804,19 @@ 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);
@@ -792,6 +842,19 @@ vfs_readlink(struct v_dnode* dnode, char* buf, size_t size)
     return 0;
 }
 
+int
+vfs_get_dtype(int itype)
+{
+    switch (itype) {
+        case VFS_IFDIR:
+            return DT_DIR;
+        case VFS_IFSYMLINK:
+            return DT_SYMLINK;
+        default:
+            return DT_PIPE;
+    }
+}
+
 __DEFINE_LXSYSCALL3(int, realpathat, int, fd, char*, buf, size_t, size)
 {
     int errno;
@@ -1161,8 +1224,22 @@ done:
     return DO_STATUS(errno);
 }
 
+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;
 
@@ -1173,14 +1250,12 @@ __vfs_do_chdir(struct v_dnode* dnode)
         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);
 
@@ -1197,7 +1272,7 @@ __DEFINE_LXSYSCALL1(int, chdir, const char*, path)
         goto done;
     }
 
-    errno = __vfs_do_chdir(dnode);
+    errno = vfs_do_chdir(__current, dnode);
 
 done:
     return DO_STATUS(errno);
@@ -1212,7 +1287,7 @@ __DEFINE_LXSYSCALL1(int, fchdir, int, fd)
         goto done;
     }
 
-    errno = __vfs_do_chdir(fd_s->file->dnode);
+    errno = vfs_do_chdir(__current, fd_s->file->dnode);
 
 done:
     return DO_STATUS(errno);