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;
75 spufs_destroy_inode(struct inode *inode)
77 kmem_cache_free(spufs_inode_cache, SPUFS_I(inode));
81 spufs_init_once(void *p)
83 struct spufs_inode_info *ei = p;
85 inode_init_once(&ei->vfs_inode);
89 spufs_new_inode(struct super_block *sb, int mode)
93 inode = new_inode(sb);
98 inode->i_uid = current_fsuid();
99 inode->i_gid = current_fsgid();
100 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
106 spufs_setattr(struct dentry *dentry, struct iattr *attr)
108 struct inode *inode = dentry->d_inode;
110 if ((attr->ia_valid & ATTR_SIZE) &&
111 (attr->ia_size != inode->i_size))
113 setattr_copy(inode, attr);
114 mark_inode_dirty(inode);
120 spufs_new_file(struct super_block *sb, struct dentry *dentry,
121 const struct file_operations *fops, int mode,
122 size_t size, struct spu_context *ctx)
124 static const struct inode_operations spufs_file_iops = {
125 .setattr = spufs_setattr,
131 inode = spufs_new_inode(sb, S_IFREG | mode);
136 inode->i_op = &spufs_file_iops;
138 inode->i_size = size;
139 inode->i_private = SPUFS_I(inode)->i_ctx = get_spu_context(ctx);
140 d_add(dentry, inode);
146 spufs_delete_inode(struct inode *inode)
148 struct spufs_inode_info *ei = SPUFS_I(inode);
151 put_spu_context(ei->i_ctx);
153 put_spu_gang(ei->i_gang);
157 static void spufs_prune_dir(struct dentry *dir)
159 struct dentry *dentry, *tmp;
161 mutex_lock(&dir->d_inode->i_mutex);
162 list_for_each_entry_safe(dentry, tmp, &dir->d_subdirs, d_u.d_child) {
163 spin_lock(&dcache_lock);
164 spin_lock(&dentry->d_lock);
165 if (!(d_unhashed(dentry)) && dentry->d_inode) {
168 spin_unlock(&dentry->d_lock);
169 simple_unlink(dir->d_inode, dentry);
170 spin_unlock(&dcache_lock);
173 spin_unlock(&dentry->d_lock);
174 spin_unlock(&dcache_lock);
177 shrink_dcache_parent(dir);
178 mutex_unlock(&dir->d_inode->i_mutex);
181 /* Caller must hold parent->i_mutex */
182 static int spufs_rmdir(struct inode *parent, struct dentry *dir)
184 /* remove all entries */
185 spufs_prune_dir(dir);
188 return simple_rmdir(parent, dir);
191 static int spufs_fill_dir(struct dentry *dir,
192 const struct spufs_tree_descr *files, int mode,
193 struct spu_context *ctx)
195 struct dentry *dentry, *tmp;
198 while (files->name && files->name[0]) {
200 dentry = d_alloc_name(dir, files->name);
203 ret = spufs_new_file(dir->d_sb, dentry, files->ops,
204 files->mode & mode, files->size, ctx);
212 * remove all children from dir. dir->inode is not set so don't
213 * just simply use spufs_prune_dir() and panic afterwards :)
214 * dput() looks like it will do the right thing:
215 * - dec parent's ref counter
216 * - remove child from parent's child list
217 * - free child's inode if possible
220 list_for_each_entry_safe(dentry, tmp, &dir->d_subdirs, d_u.d_child) {
224 shrink_dcache_parent(dir);
228 static int spufs_dir_close(struct inode *inode, struct file *file)
230 struct spu_context *ctx;
231 struct inode *parent;
235 dir = file->f_path.dentry;
236 parent = dir->d_parent->d_inode;
237 ctx = SPUFS_I(dir->d_inode)->i_ctx;
239 mutex_lock_nested(&parent->i_mutex, I_MUTEX_PARENT);
240 ret = spufs_rmdir(parent, dir);
241 mutex_unlock(&parent->i_mutex);
244 /* We have to give up the mm_struct */
247 return dcache_dir_close(inode, file);
250 const struct file_operations spufs_context_fops = {
251 .open = dcache_dir_open,
252 .release = spufs_dir_close,
253 .llseek = dcache_dir_lseek,
254 .read = generic_read_dir,
255 .readdir = dcache_readdir,
258 EXPORT_SYMBOL_GPL(spufs_context_fops);
261 spufs_mkdir(struct inode *dir, struct dentry *dentry, unsigned int flags,
266 struct spu_context *ctx;
269 inode = spufs_new_inode(dir->i_sb, mode | S_IFDIR);
273 if (dir->i_mode & S_ISGID) {
274 inode->i_gid = dir->i_gid;
275 inode->i_mode &= S_ISGID;
277 ctx = alloc_spu_context(SPUFS_I(dir)->i_gang); /* XXX gang */
278 SPUFS_I(inode)->i_ctx = ctx;
283 inode->i_op = &simple_dir_inode_operations;
284 inode->i_fop = &simple_dir_operations;
285 if (flags & SPU_CREATE_NOSCHED)
286 ret = spufs_fill_dir(dentry, spufs_dir_nosched_contents,
289 ret = spufs_fill_dir(dentry, spufs_dir_contents, mode, ctx);
294 if (spufs_get_sb_info(dir->i_sb)->debug)
295 ret = spufs_fill_dir(dentry, spufs_dir_debug_contents,
301 d_instantiate(dentry, inode);
304 inc_nlink(dentry->d_inode);
309 put_spu_context(ctx);
316 static int spufs_context_open(struct dentry *dentry, struct vfsmount *mnt)
321 ret = get_unused_fd();
328 filp = dentry_open(dentry, mnt, O_RDONLY, current_cred());
335 filp->f_op = &spufs_context_fops;
336 fd_install(ret, filp);
341 static struct spu_context *
342 spufs_assert_affinity(unsigned int flags, struct spu_gang *gang,
345 struct spu_context *tmp, *neighbor, *err;
349 aff_supp = !list_empty(&(list_entry(cbe_spu_info[0].spus.next,
350 struct spu, cbe_list))->aff_list);
353 return ERR_PTR(-EINVAL);
355 if (flags & SPU_CREATE_GANG)
356 return ERR_PTR(-EINVAL);
358 if (flags & SPU_CREATE_AFFINITY_MEM &&
360 gang->aff_ref_ctx->flags & SPU_CREATE_AFFINITY_MEM)
361 return ERR_PTR(-EEXIST);
363 if (gang->aff_flags & AFF_MERGED)
364 return ERR_PTR(-EBUSY);
367 if (flags & SPU_CREATE_AFFINITY_SPU) {
368 if (!filp || filp->f_op != &spufs_context_fops)
369 return ERR_PTR(-EINVAL);
371 neighbor = get_spu_context(
372 SPUFS_I(filp->f_dentry->d_inode)->i_ctx);
374 if (!list_empty(&neighbor->aff_list) && !(neighbor->aff_head) &&
375 !list_is_last(&neighbor->aff_list, &gang->aff_list_head) &&
376 !list_entry(neighbor->aff_list.next, struct spu_context,
377 aff_list)->aff_head) {
378 err = ERR_PTR(-EEXIST);
379 goto out_put_neighbor;
382 if (gang != neighbor->gang) {
383 err = ERR_PTR(-EINVAL);
384 goto out_put_neighbor;
388 list_for_each_entry(tmp, &gang->aff_list_head, aff_list)
390 if (list_empty(&neighbor->aff_list))
393 for (node = 0; node < MAX_NUMNODES; node++) {
394 if ((cbe_spu_info[node].n_spus - atomic_read(
395 &cbe_spu_info[node].reserved_spus)) >= count)
399 if (node == MAX_NUMNODES) {
400 err = ERR_PTR(-EEXIST);
401 goto out_put_neighbor;
408 put_spu_context(neighbor);
413 spufs_set_affinity(unsigned int flags, struct spu_context *ctx,
414 struct spu_context *neighbor)
416 if (flags & SPU_CREATE_AFFINITY_MEM)
417 ctx->gang->aff_ref_ctx = ctx;
419 if (flags & SPU_CREATE_AFFINITY_SPU) {
420 if (list_empty(&neighbor->aff_list)) {
421 list_add_tail(&neighbor->aff_list,
422 &ctx->gang->aff_list_head);
423 neighbor->aff_head = 1;
426 if (list_is_last(&neighbor->aff_list, &ctx->gang->aff_list_head)
427 || list_entry(neighbor->aff_list.next, struct spu_context,
428 aff_list)->aff_head) {
429 list_add(&ctx->aff_list, &neighbor->aff_list);
431 list_add_tail(&ctx->aff_list, &neighbor->aff_list);
432 if (neighbor->aff_head) {
433 neighbor->aff_head = 0;
438 if (!ctx->gang->aff_ref_ctx)
439 ctx->gang->aff_ref_ctx = ctx;
444 spufs_create_context(struct inode *inode, struct dentry *dentry,
445 struct vfsmount *mnt, int flags, int mode,
446 struct file *aff_filp)
450 struct spu_gang *gang;
451 struct spu_context *neighbor;
454 if ((flags & SPU_CREATE_NOSCHED) &&
455 !capable(CAP_SYS_NICE))
459 if ((flags & (SPU_CREATE_NOSCHED | SPU_CREATE_ISOLATE))
460 == SPU_CREATE_ISOLATE)
464 if ((flags & SPU_CREATE_ISOLATE) && !isolated_loader)
469 affinity = flags & (SPU_CREATE_AFFINITY_MEM | SPU_CREATE_AFFINITY_SPU);
471 gang = SPUFS_I(inode)->i_gang;
475 mutex_lock(&gang->aff_mutex);
476 neighbor = spufs_assert_affinity(flags, gang, aff_filp);
477 if (IS_ERR(neighbor)) {
478 ret = PTR_ERR(neighbor);
483 ret = spufs_mkdir(inode, dentry, flags, mode & S_IRWXUGO);
488 spufs_set_affinity(flags, SPUFS_I(dentry->d_inode)->i_ctx,
491 put_spu_context(neighbor);
495 * get references for dget and mntget, will be released
496 * in error path of *_open().
498 ret = spufs_context_open(dget(dentry), mntget(mnt));
500 WARN_ON(spufs_rmdir(inode, dentry));
502 mutex_unlock(&gang->aff_mutex);
503 mutex_unlock(&inode->i_mutex);
504 spu_forget(SPUFS_I(dentry->d_inode)->i_ctx);
510 mutex_unlock(&gang->aff_mutex);
512 mutex_unlock(&inode->i_mutex);
519 spufs_mkgang(struct inode *dir, struct dentry *dentry, int mode)
523 struct spu_gang *gang;
526 inode = spufs_new_inode(dir->i_sb, mode | S_IFDIR);
531 if (dir->i_mode & S_ISGID) {
532 inode->i_gid = dir->i_gid;
533 inode->i_mode &= S_ISGID;
535 gang = alloc_spu_gang();
536 SPUFS_I(inode)->i_ctx = NULL;
537 SPUFS_I(inode)->i_gang = gang;
541 inode->i_op = &simple_dir_inode_operations;
542 inode->i_fop = &simple_dir_operations;
544 d_instantiate(dentry, inode);
546 inc_nlink(dentry->d_inode);
555 static int spufs_gang_open(struct dentry *dentry, struct vfsmount *mnt)
560 ret = get_unused_fd();
567 filp = dentry_open(dentry, mnt, O_RDONLY, current_cred());
574 filp->f_op = &simple_dir_operations;
575 fd_install(ret, filp);
580 static int spufs_create_gang(struct inode *inode,
581 struct dentry *dentry,
582 struct vfsmount *mnt, int mode)
586 ret = spufs_mkgang(inode, dentry, mode & S_IRWXUGO);
591 * get references for dget and mntget, will be released
592 * in error path of *_open().
594 ret = spufs_gang_open(dget(dentry), mntget(mnt));
596 int err = simple_rmdir(inode, dentry);
601 mutex_unlock(&inode->i_mutex);
607 static struct file_system_type spufs_type;
609 long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode,
612 struct dentry *dentry;
616 /* check if we are on spufs */
617 if (nd->path.dentry->d_sb->s_type != &spufs_type)
620 /* don't accept undefined flags */
621 if (flags & (~SPU_CREATE_FLAG_ALL))
624 /* only threads can be underneath a gang */
625 if (nd->path.dentry != nd->path.dentry->d_sb->s_root) {
626 if ((flags & SPU_CREATE_GANG) ||
627 !SPUFS_I(nd->path.dentry->d_inode)->i_gang)
631 dentry = lookup_create(nd, 1);
632 ret = PTR_ERR(dentry);
636 mode &= ~current_umask();
638 if (flags & SPU_CREATE_GANG)
639 ret = spufs_create_gang(nd->path.dentry->d_inode,
640 dentry, nd->path.mnt, mode);
642 ret = spufs_create_context(nd->path.dentry->d_inode,
643 dentry, nd->path.mnt, flags, mode,
646 fsnotify_mkdir(nd->path.dentry->d_inode, dentry);
650 mutex_unlock(&nd->path.dentry->d_inode->i_mutex);
655 /* File system initialization */
657 Opt_uid, Opt_gid, Opt_mode, Opt_debug, Opt_err,
660 static const match_table_t spufs_tokens = {
661 { Opt_uid, "uid=%d" },
662 { Opt_gid, "gid=%d" },
663 { Opt_mode, "mode=%o" },
664 { Opt_debug, "debug" },
669 spufs_parse_options(struct super_block *sb, char *options, struct inode *root)
672 substring_t args[MAX_OPT_ARGS];
674 while ((p = strsep(&options, ",")) != NULL) {
680 token = match_token(p, spufs_tokens, args);
683 if (match_int(&args[0], &option))
685 root->i_uid = option;
688 if (match_int(&args[0], &option))
690 root->i_gid = option;
693 if (match_octal(&args[0], &option))
695 root->i_mode = option | S_IFDIR;
698 spufs_get_sb_info(sb)->debug = 1;
707 static void spufs_exit_isolated_loader(void)
709 free_pages((unsigned long) isolated_loader,
710 get_order(isolated_loader_size));
714 spufs_init_isolated_loader(void)
716 struct device_node *dn;
720 dn = of_find_node_by_path("/spu-isolation");
724 loader = of_get_property(dn, "loader", &size);
728 /* the loader must be align on a 16 byte boundary */
729 isolated_loader = (char *)__get_free_pages(GFP_KERNEL, get_order(size));
730 if (!isolated_loader)
733 isolated_loader_size = size;
734 memcpy(isolated_loader, loader, size);
735 printk(KERN_INFO "spufs: SPU isolation mode enabled\n");
739 spufs_create_root(struct super_block *sb, void *data)
745 if (!spu_management_ops)
749 inode = spufs_new_inode(sb, S_IFDIR | 0775);
753 inode->i_op = &simple_dir_inode_operations;
754 inode->i_fop = &simple_dir_operations;
755 SPUFS_I(inode)->i_ctx = NULL;
759 if (!spufs_parse_options(sb, data, inode))
763 sb->s_root = d_alloc_root(inode);
775 spufs_fill_super(struct super_block *sb, void *data, int silent)
777 struct spufs_sb_info *info;
778 static const struct super_operations s_ops = {
779 .alloc_inode = spufs_alloc_inode,
780 .destroy_inode = spufs_destroy_inode,
781 .statfs = simple_statfs,
782 .delete_inode = spufs_delete_inode,
783 .drop_inode = generic_delete_inode,
784 .show_options = generic_show_options,
787 save_mount_options(sb, data);
789 info = kzalloc(sizeof(*info), GFP_KERNEL);
793 sb->s_maxbytes = MAX_LFS_FILESIZE;
794 sb->s_blocksize = PAGE_CACHE_SIZE;
795 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
796 sb->s_magic = SPUFS_MAGIC;
798 sb->s_fs_info = info;
800 return spufs_create_root(sb, data);
804 spufs_get_sb(struct file_system_type *fstype, int flags,
805 const char *name, void *data, struct vfsmount *mnt)
807 return get_sb_single(fstype, flags, data, spufs_fill_super, mnt);
810 static struct file_system_type spufs_type = {
811 .owner = THIS_MODULE,
813 .get_sb = spufs_get_sb,
814 .kill_sb = kill_litter_super,
817 static int __init spufs_init(void)
822 if (!spu_management_ops)
826 spufs_inode_cache = kmem_cache_create("spufs_inode_cache",
827 sizeof(struct spufs_inode_info), 0,
828 SLAB_HWCACHE_ALIGN, spufs_init_once);
830 if (!spufs_inode_cache)
832 ret = spu_sched_init();
835 ret = register_filesystem(&spufs_type);
838 ret = register_spu_syscalls(&spufs_calls);
842 spufs_init_isolated_loader();
847 unregister_filesystem(&spufs_type);
851 kmem_cache_destroy(spufs_inode_cache);
855 module_init(spufs_init);
857 static void __exit spufs_exit(void)
860 spufs_exit_isolated_loader();
861 unregister_spu_syscalls(&spufs_calls);
862 unregister_filesystem(&spufs_type);
863 kmem_cache_destroy(spufs_inode_cache);
865 module_exit(spufs_exit);
867 MODULE_LICENSE("GPL");
868 MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");