5 * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
7 * Author: Arnd Bergmann <arndb@de.ibm.com>
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 2, or (at your option)
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License
20 * along with this program; if not, write to the Free Software
21 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24 #include <linux/file.h>
26 #include <linux/fsnotify.h>
27 #include <linux/backing-dev.h>
28 #include <linux/init.h>
29 #include <linux/ioctl.h>
30 #include <linux/module.h>
31 #include <linux/mount.h>
32 #include <linux/namei.h>
33 #include <linux/pagemap.h>
34 #include <linux/poll.h>
35 #include <linux/slab.h>
36 #include <linux/parser.h>
40 #include <asm/spu_priv1.h>
41 #include <asm/uaccess.h>
45 struct spufs_sb_info {
49 static struct kmem_cache *spufs_inode_cache;
50 char *isolated_loader;
51 static int isolated_loader_size;
53 static struct spufs_sb_info *spufs_get_sb_info(struct super_block *sb)
59 spufs_alloc_inode(struct super_block *sb)
61 struct spufs_inode_info *ei;
63 ei = kmem_cache_alloc(spufs_inode_cache, GFP_KERNEL);
71 return &ei->vfs_inode;
74 static void spufs_i_callback(struct rcu_head *head)
76 struct inode *inode = container_of(head, struct inode, i_rcu);
77 kmem_cache_free(spufs_inode_cache, SPUFS_I(inode));
80 static void spufs_destroy_inode(struct inode *inode)
82 call_rcu(&inode->i_rcu, spufs_i_callback);
86 spufs_init_once(void *p)
88 struct spufs_inode_info *ei = p;
90 inode_init_once(&ei->vfs_inode);
94 spufs_new_inode(struct super_block *sb, umode_t mode)
98 inode = new_inode(sb);
102 inode->i_ino = get_next_ino();
103 inode->i_mode = mode;
104 inode->i_uid = current_fsuid();
105 inode->i_gid = current_fsgid();
106 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
112 spufs_setattr(struct dentry *dentry, struct iattr *attr)
114 struct inode *inode = d_inode(dentry);
116 if ((attr->ia_valid & ATTR_SIZE) &&
117 (attr->ia_size != inode->i_size))
119 setattr_copy(inode, attr);
120 mark_inode_dirty(inode);
126 spufs_new_file(struct super_block *sb, struct dentry *dentry,
127 const struct file_operations *fops, umode_t mode,
128 size_t size, struct spu_context *ctx)
130 static const struct inode_operations spufs_file_iops = {
131 .setattr = spufs_setattr,
137 inode = spufs_new_inode(sb, S_IFREG | mode);
142 inode->i_op = &spufs_file_iops;
144 inode->i_size = size;
145 inode->i_private = SPUFS_I(inode)->i_ctx = get_spu_context(ctx);
146 d_add(dentry, inode);
152 spufs_evict_inode(struct inode *inode)
154 struct spufs_inode_info *ei = SPUFS_I(inode);
157 put_spu_context(ei->i_ctx);
159 put_spu_gang(ei->i_gang);
162 static void spufs_prune_dir(struct dentry *dir)
164 struct dentry *dentry, *tmp;
166 inode_lock(d_inode(dir));
167 list_for_each_entry_safe(dentry, tmp, &dir->d_subdirs, d_child) {
168 spin_lock(&dentry->d_lock);
169 if (simple_positive(dentry)) {
172 spin_unlock(&dentry->d_lock);
173 simple_unlink(d_inode(dir), dentry);
174 /* XXX: what was dcache_lock protecting here? Other
175 * filesystems (IB, configfs) release dcache_lock
179 spin_unlock(&dentry->d_lock);
182 shrink_dcache_parent(dir);
183 inode_unlock(d_inode(dir));
186 /* Caller must hold parent->i_mutex */
187 static int spufs_rmdir(struct inode *parent, struct dentry *dir)
189 /* remove all entries */
191 spufs_prune_dir(dir);
193 res = simple_rmdir(parent, dir);
194 /* We have to give up the mm_struct */
195 spu_forget(SPUFS_I(d_inode(dir))->i_ctx);
199 static int spufs_fill_dir(struct dentry *dir,
200 const struct spufs_tree_descr *files, umode_t mode,
201 struct spu_context *ctx)
203 while (files->name && files->name[0]) {
205 struct dentry *dentry = d_alloc_name(dir, files->name);
208 ret = spufs_new_file(dir->d_sb, dentry, files->ops,
209 files->mode & mode, files->size, ctx);
217 static int spufs_dir_close(struct inode *inode, struct file *file)
219 struct spu_context *ctx;
220 struct inode *parent;
224 dir = file->f_path.dentry;
225 parent = d_inode(dir->d_parent);
226 ctx = SPUFS_I(d_inode(dir))->i_ctx;
228 inode_lock_nested(parent, I_MUTEX_PARENT);
229 ret = spufs_rmdir(parent, dir);
230 inode_unlock(parent);
233 return dcache_dir_close(inode, file);
236 const struct file_operations spufs_context_fops = {
237 .open = dcache_dir_open,
238 .release = spufs_dir_close,
239 .llseek = dcache_dir_lseek,
240 .read = generic_read_dir,
241 .iterate_shared = dcache_readdir,
244 EXPORT_SYMBOL_GPL(spufs_context_fops);
247 spufs_mkdir(struct inode *dir, struct dentry *dentry, unsigned int flags,
252 struct spu_context *ctx;
254 inode = spufs_new_inode(dir->i_sb, mode | S_IFDIR);
258 if (dir->i_mode & S_ISGID) {
259 inode->i_gid = dir->i_gid;
260 inode->i_mode &= S_ISGID;
262 ctx = alloc_spu_context(SPUFS_I(dir)->i_gang); /* XXX gang */
263 SPUFS_I(inode)->i_ctx = ctx;
270 inode->i_op = &simple_dir_inode_operations;
271 inode->i_fop = &simple_dir_operations;
279 d_instantiate(dentry, inode);
281 if (flags & SPU_CREATE_NOSCHED)
282 ret = spufs_fill_dir(dentry, spufs_dir_nosched_contents,
285 ret = spufs_fill_dir(dentry, spufs_dir_contents, mode, ctx);
287 if (!ret && spufs_get_sb_info(dir->i_sb)->debug)
288 ret = spufs_fill_dir(dentry, spufs_dir_debug_contents,
292 spufs_rmdir(dir, dentry);
299 static int spufs_context_open(struct path *path)
304 ret = get_unused_fd_flags(0);
308 filp = dentry_open(path, O_RDONLY, current_cred());
311 return PTR_ERR(filp);
314 filp->f_op = &spufs_context_fops;
315 fd_install(ret, filp);
319 static struct spu_context *
320 spufs_assert_affinity(unsigned int flags, struct spu_gang *gang,
323 struct spu_context *tmp, *neighbor, *err;
327 aff_supp = !list_empty(&(list_entry(cbe_spu_info[0].spus.next,
328 struct spu, cbe_list))->aff_list);
331 return ERR_PTR(-EINVAL);
333 if (flags & SPU_CREATE_GANG)
334 return ERR_PTR(-EINVAL);
336 if (flags & SPU_CREATE_AFFINITY_MEM &&
338 gang->aff_ref_ctx->flags & SPU_CREATE_AFFINITY_MEM)
339 return ERR_PTR(-EEXIST);
341 if (gang->aff_flags & AFF_MERGED)
342 return ERR_PTR(-EBUSY);
345 if (flags & SPU_CREATE_AFFINITY_SPU) {
346 if (!filp || filp->f_op != &spufs_context_fops)
347 return ERR_PTR(-EINVAL);
349 neighbor = get_spu_context(
350 SPUFS_I(file_inode(filp))->i_ctx);
352 if (!list_empty(&neighbor->aff_list) && !(neighbor->aff_head) &&
353 !list_is_last(&neighbor->aff_list, &gang->aff_list_head) &&
354 !list_entry(neighbor->aff_list.next, struct spu_context,
355 aff_list)->aff_head) {
356 err = ERR_PTR(-EEXIST);
357 goto out_put_neighbor;
360 if (gang != neighbor->gang) {
361 err = ERR_PTR(-EINVAL);
362 goto out_put_neighbor;
366 list_for_each_entry(tmp, &gang->aff_list_head, aff_list)
368 if (list_empty(&neighbor->aff_list))
371 for (node = 0; node < MAX_NUMNODES; node++) {
372 if ((cbe_spu_info[node].n_spus - atomic_read(
373 &cbe_spu_info[node].reserved_spus)) >= count)
377 if (node == MAX_NUMNODES) {
378 err = ERR_PTR(-EEXIST);
379 goto out_put_neighbor;
386 put_spu_context(neighbor);
391 spufs_set_affinity(unsigned int flags, struct spu_context *ctx,
392 struct spu_context *neighbor)
394 if (flags & SPU_CREATE_AFFINITY_MEM)
395 ctx->gang->aff_ref_ctx = ctx;
397 if (flags & SPU_CREATE_AFFINITY_SPU) {
398 if (list_empty(&neighbor->aff_list)) {
399 list_add_tail(&neighbor->aff_list,
400 &ctx->gang->aff_list_head);
401 neighbor->aff_head = 1;
404 if (list_is_last(&neighbor->aff_list, &ctx->gang->aff_list_head)
405 || list_entry(neighbor->aff_list.next, struct spu_context,
406 aff_list)->aff_head) {
407 list_add(&ctx->aff_list, &neighbor->aff_list);
409 list_add_tail(&ctx->aff_list, &neighbor->aff_list);
410 if (neighbor->aff_head) {
411 neighbor->aff_head = 0;
416 if (!ctx->gang->aff_ref_ctx)
417 ctx->gang->aff_ref_ctx = ctx;
422 spufs_create_context(struct inode *inode, struct dentry *dentry,
423 struct vfsmount *mnt, int flags, umode_t mode,
424 struct file *aff_filp)
428 struct spu_gang *gang;
429 struct spu_context *neighbor;
430 struct path path = {.mnt = mnt, .dentry = dentry};
432 if ((flags & SPU_CREATE_NOSCHED) &&
433 !capable(CAP_SYS_NICE))
436 if ((flags & (SPU_CREATE_NOSCHED | SPU_CREATE_ISOLATE))
437 == SPU_CREATE_ISOLATE)
440 if ((flags & SPU_CREATE_ISOLATE) && !isolated_loader)
445 affinity = flags & (SPU_CREATE_AFFINITY_MEM | SPU_CREATE_AFFINITY_SPU);
447 gang = SPUFS_I(inode)->i_gang;
450 mutex_lock(&gang->aff_mutex);
451 neighbor = spufs_assert_affinity(flags, gang, aff_filp);
452 if (IS_ERR(neighbor)) {
453 ret = PTR_ERR(neighbor);
458 ret = spufs_mkdir(inode, dentry, flags, mode & S_IRWXUGO);
463 spufs_set_affinity(flags, SPUFS_I(d_inode(dentry))->i_ctx,
466 put_spu_context(neighbor);
469 ret = spufs_context_open(&path);
471 WARN_ON(spufs_rmdir(inode, dentry));
475 mutex_unlock(&gang->aff_mutex);
480 spufs_mkgang(struct inode *dir, struct dentry *dentry, umode_t mode)
484 struct spu_gang *gang;
487 inode = spufs_new_inode(dir->i_sb, mode | S_IFDIR);
492 if (dir->i_mode & S_ISGID) {
493 inode->i_gid = dir->i_gid;
494 inode->i_mode &= S_ISGID;
496 gang = alloc_spu_gang();
497 SPUFS_I(inode)->i_ctx = NULL;
498 SPUFS_I(inode)->i_gang = gang;
504 inode->i_op = &simple_dir_inode_operations;
505 inode->i_fop = &simple_dir_operations;
507 d_instantiate(dentry, inode);
509 inc_nlink(d_inode(dentry));
518 static int spufs_gang_open(struct path *path)
523 ret = get_unused_fd_flags(0);
528 * get references for dget and mntget, will be released
529 * in error path of *_open().
531 filp = dentry_open(path, O_RDONLY, current_cred());
534 return PTR_ERR(filp);
537 filp->f_op = &simple_dir_operations;
538 fd_install(ret, filp);
542 static int spufs_create_gang(struct inode *inode,
543 struct dentry *dentry,
544 struct vfsmount *mnt, umode_t mode)
546 struct path path = {.mnt = mnt, .dentry = dentry};
549 ret = spufs_mkgang(inode, dentry, mode & S_IRWXUGO);
551 ret = spufs_gang_open(&path);
553 int err = simple_rmdir(inode, dentry);
561 static struct file_system_type spufs_type;
563 long spufs_create(struct path *path, struct dentry *dentry,
564 unsigned int flags, umode_t mode, struct file *filp)
566 struct inode *dir = d_inode(path->dentry);
569 /* check if we are on spufs */
570 if (path->dentry->d_sb->s_type != &spufs_type)
573 /* don't accept undefined flags */
574 if (flags & (~SPU_CREATE_FLAG_ALL))
577 /* only threads can be underneath a gang */
578 if (path->dentry != path->dentry->d_sb->s_root)
579 if ((flags & SPU_CREATE_GANG) || !SPUFS_I(dir)->i_gang)
582 mode &= ~current_umask();
584 if (flags & SPU_CREATE_GANG)
585 ret = spufs_create_gang(dir, dentry, path->mnt, mode);
587 ret = spufs_create_context(dir, dentry, path->mnt, flags, mode,
590 fsnotify_mkdir(dir, dentry);
595 /* File system initialization */
597 Opt_uid, Opt_gid, Opt_mode, Opt_debug, Opt_err,
600 static const match_table_t spufs_tokens = {
601 { Opt_uid, "uid=%d" },
602 { Opt_gid, "gid=%d" },
603 { Opt_mode, "mode=%o" },
604 { Opt_debug, "debug" },
609 spufs_parse_options(struct super_block *sb, char *options, struct inode *root)
612 substring_t args[MAX_OPT_ARGS];
614 while ((p = strsep(&options, ",")) != NULL) {
620 token = match_token(p, spufs_tokens, args);
623 if (match_int(&args[0], &option))
625 root->i_uid = make_kuid(current_user_ns(), option);
626 if (!uid_valid(root->i_uid))
630 if (match_int(&args[0], &option))
632 root->i_gid = make_kgid(current_user_ns(), option);
633 if (!gid_valid(root->i_gid))
637 if (match_octal(&args[0], &option))
639 root->i_mode = option | S_IFDIR;
642 spufs_get_sb_info(sb)->debug = 1;
651 static void spufs_exit_isolated_loader(void)
653 free_pages((unsigned long) isolated_loader,
654 get_order(isolated_loader_size));
658 spufs_init_isolated_loader(void)
660 struct device_node *dn;
664 dn = of_find_node_by_path("/spu-isolation");
668 loader = of_get_property(dn, "loader", &size);
672 /* the loader must be align on a 16 byte boundary */
673 isolated_loader = (char *)__get_free_pages(GFP_KERNEL, get_order(size));
674 if (!isolated_loader)
677 isolated_loader_size = size;
678 memcpy(isolated_loader, loader, size);
679 printk(KERN_INFO "spufs: SPU isolation mode enabled\n");
683 spufs_create_root(struct super_block *sb, void *data)
689 if (!spu_management_ops)
693 inode = spufs_new_inode(sb, S_IFDIR | 0775);
697 inode->i_op = &simple_dir_inode_operations;
698 inode->i_fop = &simple_dir_operations;
699 SPUFS_I(inode)->i_ctx = NULL;
703 if (!spufs_parse_options(sb, data, inode))
707 sb->s_root = d_make_root(inode);
719 spufs_fill_super(struct super_block *sb, void *data, int silent)
721 struct spufs_sb_info *info;
722 static const struct super_operations s_ops = {
723 .alloc_inode = spufs_alloc_inode,
724 .destroy_inode = spufs_destroy_inode,
725 .statfs = simple_statfs,
726 .evict_inode = spufs_evict_inode,
727 .show_options = generic_show_options,
730 save_mount_options(sb, data);
732 info = kzalloc(sizeof(*info), GFP_KERNEL);
736 sb->s_maxbytes = MAX_LFS_FILESIZE;
737 sb->s_blocksize = PAGE_SIZE;
738 sb->s_blocksize_bits = PAGE_SHIFT;
739 sb->s_magic = SPUFS_MAGIC;
741 sb->s_fs_info = info;
743 return spufs_create_root(sb, data);
746 static struct dentry *
747 spufs_mount(struct file_system_type *fstype, int flags,
748 const char *name, void *data)
750 return mount_single(fstype, flags, data, spufs_fill_super);
753 static struct file_system_type spufs_type = {
754 .owner = THIS_MODULE,
756 .mount = spufs_mount,
757 .kill_sb = kill_litter_super,
759 MODULE_ALIAS_FS("spufs");
761 static int __init spufs_init(void)
766 if (!spu_management_ops)
770 spufs_inode_cache = kmem_cache_create("spufs_inode_cache",
771 sizeof(struct spufs_inode_info), 0,
772 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, spufs_init_once);
774 if (!spufs_inode_cache)
776 ret = spu_sched_init();
779 ret = register_spu_syscalls(&spufs_calls);
782 ret = register_filesystem(&spufs_type);
786 spufs_init_isolated_loader();
791 unregister_spu_syscalls(&spufs_calls);
795 kmem_cache_destroy(spufs_inode_cache);
799 module_init(spufs_init);
801 static void __exit spufs_exit(void)
804 spufs_exit_isolated_loader();
805 unregister_spu_syscalls(&spufs_calls);
806 unregister_filesystem(&spufs_type);
807 kmem_cache_destroy(spufs_inode_cache);
809 module_exit(spufs_exit);
811 MODULE_LICENSE("GPL");
812 MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");