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/backing-dev.h>
27 #include <linux/init.h>
28 #include <linux/ioctl.h>
29 #include <linux/module.h>
30 #include <linux/mount.h>
31 #include <linux/namei.h>
32 #include <linux/pagemap.h>
33 #include <linux/poll.h>
34 #include <linux/slab.h>
35 #include <linux/parser.h>
38 #include <asm/semaphore.h>
40 #include <asm/spu_priv1.h>
41 #include <asm/uaccess.h>
45 static struct kmem_cache
*spufs_inode_cache
;
46 char *isolated_loader
;
47 static int isolated_loader_size
;
50 spufs_alloc_inode(struct super_block
*sb
)
52 struct spufs_inode_info
*ei
;
54 ei
= kmem_cache_alloc(spufs_inode_cache
, GFP_KERNEL
);
62 return &ei
->vfs_inode
;
66 spufs_destroy_inode(struct inode
*inode
)
68 kmem_cache_free(spufs_inode_cache
, SPUFS_I(inode
));
72 spufs_init_once(struct kmem_cache
*cachep
, void *p
)
74 struct spufs_inode_info
*ei
= p
;
76 inode_init_once(&ei
->vfs_inode
);
80 spufs_new_inode(struct super_block
*sb
, int mode
)
84 inode
= new_inode(sb
);
89 inode
->i_uid
= current
->fsuid
;
90 inode
->i_gid
= current
->fsgid
;
92 inode
->i_atime
= inode
->i_mtime
= inode
->i_ctime
= CURRENT_TIME
;
98 spufs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
100 struct inode
*inode
= dentry
->d_inode
;
102 if ((attr
->ia_valid
& ATTR_SIZE
) &&
103 (attr
->ia_size
!= inode
->i_size
))
105 return inode_setattr(inode
, attr
);
110 spufs_new_file(struct super_block
*sb
, struct dentry
*dentry
,
111 const struct file_operations
*fops
, int mode
,
112 struct spu_context
*ctx
)
114 static struct inode_operations spufs_file_iops
= {
115 .setattr
= spufs_setattr
,
121 inode
= spufs_new_inode(sb
, S_IFREG
| mode
);
126 inode
->i_op
= &spufs_file_iops
;
128 inode
->i_private
= SPUFS_I(inode
)->i_ctx
= get_spu_context(ctx
);
129 d_add(dentry
, inode
);
135 spufs_delete_inode(struct inode
*inode
)
137 struct spufs_inode_info
*ei
= SPUFS_I(inode
);
140 put_spu_context(ei
->i_ctx
);
142 put_spu_gang(ei
->i_gang
);
146 static void spufs_prune_dir(struct dentry
*dir
)
148 struct dentry
*dentry
, *tmp
;
150 mutex_lock(&dir
->d_inode
->i_mutex
);
151 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
152 spin_lock(&dcache_lock
);
153 spin_lock(&dentry
->d_lock
);
154 if (!(d_unhashed(dentry
)) && dentry
->d_inode
) {
157 spin_unlock(&dentry
->d_lock
);
158 simple_unlink(dir
->d_inode
, dentry
);
159 spin_unlock(&dcache_lock
);
162 spin_unlock(&dentry
->d_lock
);
163 spin_unlock(&dcache_lock
);
166 shrink_dcache_parent(dir
);
167 mutex_unlock(&dir
->d_inode
->i_mutex
);
170 /* Caller must hold parent->i_mutex */
171 static int spufs_rmdir(struct inode
*parent
, struct dentry
*dir
)
173 /* remove all entries */
174 spufs_prune_dir(dir
);
177 return simple_rmdir(parent
, dir
);
180 static int spufs_fill_dir(struct dentry
*dir
, struct tree_descr
*files
,
181 int mode
, struct spu_context
*ctx
)
183 struct dentry
*dentry
, *tmp
;
186 while (files
->name
&& files
->name
[0]) {
188 dentry
= d_alloc_name(dir
, files
->name
);
191 ret
= spufs_new_file(dir
->d_sb
, dentry
, files
->ops
,
192 files
->mode
& mode
, ctx
);
200 * remove all children from dir. dir->inode is not set so don't
201 * just simply use spufs_prune_dir() and panic afterwards :)
202 * dput() looks like it will do the right thing:
203 * - dec parent's ref counter
204 * - remove child from parent's child list
205 * - free child's inode if possible
208 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
212 shrink_dcache_parent(dir
);
216 static int spufs_dir_close(struct inode
*inode
, struct file
*file
)
218 struct spu_context
*ctx
;
219 struct inode
*parent
;
223 dir
= file
->f_path
.dentry
;
224 parent
= dir
->d_parent
->d_inode
;
225 ctx
= SPUFS_I(dir
->d_inode
)->i_ctx
;
227 mutex_lock(&parent
->i_mutex
);
228 ret
= spufs_rmdir(parent
, dir
);
229 mutex_unlock(&parent
->i_mutex
);
232 /* We have to give up the mm_struct */
235 return dcache_dir_close(inode
, file
);
238 const struct file_operations spufs_context_fops
= {
239 .open
= dcache_dir_open
,
240 .release
= spufs_dir_close
,
241 .llseek
= dcache_dir_lseek
,
242 .read
= generic_read_dir
,
243 .readdir
= dcache_readdir
,
244 .fsync
= simple_sync_file
,
246 EXPORT_SYMBOL_GPL(spufs_context_fops
);
249 spufs_mkdir(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
,
254 struct spu_context
*ctx
;
257 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
261 if (dir
->i_mode
& S_ISGID
) {
262 inode
->i_gid
= dir
->i_gid
;
263 inode
->i_mode
&= S_ISGID
;
265 ctx
= alloc_spu_context(SPUFS_I(dir
)->i_gang
); /* XXX gang */
266 SPUFS_I(inode
)->i_ctx
= ctx
;
271 inode
->i_op
= &simple_dir_inode_operations
;
272 inode
->i_fop
= &simple_dir_operations
;
273 if (flags
& SPU_CREATE_NOSCHED
)
274 ret
= spufs_fill_dir(dentry
, spufs_dir_nosched_contents
,
277 ret
= spufs_fill_dir(dentry
, spufs_dir_contents
, mode
, ctx
);
282 d_instantiate(dentry
, inode
);
285 dentry
->d_inode
->i_nlink
++;
290 put_spu_context(ctx
);
297 static int spufs_context_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
302 ret
= get_unused_fd();
309 filp
= dentry_open(dentry
, mnt
, O_RDONLY
);
316 filp
->f_op
= &spufs_context_fops
;
317 fd_install(ret
, filp
);
322 static struct spu_context
*
323 spufs_assert_affinity(unsigned int flags
, struct spu_gang
*gang
,
326 struct spu_context
*tmp
, *neighbor
, *err
;
330 aff_supp
= !list_empty(&(list_entry(cbe_spu_info
[0].spus
.next
,
331 struct spu
, cbe_list
))->aff_list
);
334 return ERR_PTR(-EINVAL
);
336 if (flags
& SPU_CREATE_GANG
)
337 return ERR_PTR(-EINVAL
);
339 if (flags
& SPU_CREATE_AFFINITY_MEM
&&
341 gang
->aff_ref_ctx
->flags
& SPU_CREATE_AFFINITY_MEM
)
342 return ERR_PTR(-EEXIST
);
344 if (gang
->aff_flags
& AFF_MERGED
)
345 return ERR_PTR(-EBUSY
);
348 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
349 if (!filp
|| filp
->f_op
!= &spufs_context_fops
)
350 return ERR_PTR(-EINVAL
);
352 neighbor
= get_spu_context(
353 SPUFS_I(filp
->f_dentry
->d_inode
)->i_ctx
);
355 if (!list_empty(&neighbor
->aff_list
) && !(neighbor
->aff_head
) &&
356 !list_is_last(&neighbor
->aff_list
, &gang
->aff_list_head
) &&
357 !list_entry(neighbor
->aff_list
.next
, struct spu_context
,
358 aff_list
)->aff_head
) {
359 err
= ERR_PTR(-EEXIST
);
360 goto out_put_neighbor
;
363 if (gang
!= neighbor
->gang
) {
364 err
= ERR_PTR(-EINVAL
);
365 goto out_put_neighbor
;
369 list_for_each_entry(tmp
, &gang
->aff_list_head
, aff_list
)
371 if (list_empty(&neighbor
->aff_list
))
374 for (node
= 0; node
< MAX_NUMNODES
; node
++) {
375 if ((cbe_spu_info
[node
].n_spus
- atomic_read(
376 &cbe_spu_info
[node
].reserved_spus
)) >= count
)
380 if (node
== MAX_NUMNODES
) {
381 err
= ERR_PTR(-EEXIST
);
382 goto out_put_neighbor
;
389 put_spu_context(neighbor
);
394 spufs_set_affinity(unsigned int flags
, struct spu_context
*ctx
,
395 struct spu_context
*neighbor
)
397 if (flags
& SPU_CREATE_AFFINITY_MEM
)
398 ctx
->gang
->aff_ref_ctx
= ctx
;
400 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
401 if (list_empty(&neighbor
->aff_list
)) {
402 list_add_tail(&neighbor
->aff_list
,
403 &ctx
->gang
->aff_list_head
);
404 neighbor
->aff_head
= 1;
407 if (list_is_last(&neighbor
->aff_list
, &ctx
->gang
->aff_list_head
)
408 || list_entry(neighbor
->aff_list
.next
, struct spu_context
,
409 aff_list
)->aff_head
) {
410 list_add(&ctx
->aff_list
, &neighbor
->aff_list
);
412 list_add_tail(&ctx
->aff_list
, &neighbor
->aff_list
);
413 if (neighbor
->aff_head
) {
414 neighbor
->aff_head
= 0;
419 if (!ctx
->gang
->aff_ref_ctx
)
420 ctx
->gang
->aff_ref_ctx
= ctx
;
425 spufs_create_context(struct inode
*inode
, struct dentry
*dentry
,
426 struct vfsmount
*mnt
, int flags
, int mode
,
427 struct file
*aff_filp
)
431 struct spu_gang
*gang
;
432 struct spu_context
*neighbor
;
435 if ((flags
& SPU_CREATE_NOSCHED
) &&
436 !capable(CAP_SYS_NICE
))
440 if ((flags
& (SPU_CREATE_NOSCHED
| SPU_CREATE_ISOLATE
))
441 == SPU_CREATE_ISOLATE
)
445 if ((flags
& SPU_CREATE_ISOLATE
) && !isolated_loader
)
450 affinity
= flags
& (SPU_CREATE_AFFINITY_MEM
| SPU_CREATE_AFFINITY_SPU
);
452 gang
= SPUFS_I(inode
)->i_gang
;
456 mutex_lock(&gang
->aff_mutex
);
457 neighbor
= spufs_assert_affinity(flags
, gang
, aff_filp
);
458 if (IS_ERR(neighbor
)) {
459 ret
= PTR_ERR(neighbor
);
464 ret
= spufs_mkdir(inode
, dentry
, flags
, mode
& S_IRWXUGO
);
469 spufs_set_affinity(flags
, SPUFS_I(dentry
->d_inode
)->i_ctx
,
472 put_spu_context(neighbor
);
476 * get references for dget and mntget, will be released
477 * in error path of *_open().
479 ret
= spufs_context_open(dget(dentry
), mntget(mnt
));
481 WARN_ON(spufs_rmdir(inode
, dentry
));
482 mutex_unlock(&inode
->i_mutex
);
483 spu_forget(SPUFS_I(dentry
->d_inode
)->i_ctx
);
489 mutex_unlock(&gang
->aff_mutex
);
491 mutex_unlock(&inode
->i_mutex
);
498 spufs_mkgang(struct inode
*dir
, struct dentry
*dentry
, int mode
)
502 struct spu_gang
*gang
;
505 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
510 if (dir
->i_mode
& S_ISGID
) {
511 inode
->i_gid
= dir
->i_gid
;
512 inode
->i_mode
&= S_ISGID
;
514 gang
= alloc_spu_gang();
515 SPUFS_I(inode
)->i_ctx
= NULL
;
516 SPUFS_I(inode
)->i_gang
= gang
;
520 inode
->i_op
= &simple_dir_inode_operations
;
521 inode
->i_fop
= &simple_dir_operations
;
523 d_instantiate(dentry
, inode
);
525 dentry
->d_inode
->i_nlink
++;
534 static int spufs_gang_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
539 ret
= get_unused_fd();
546 filp
= dentry_open(dentry
, mnt
, O_RDONLY
);
553 filp
->f_op
= &simple_dir_operations
;
554 fd_install(ret
, filp
);
559 static int spufs_create_gang(struct inode
*inode
,
560 struct dentry
*dentry
,
561 struct vfsmount
*mnt
, int mode
)
565 ret
= spufs_mkgang(inode
, dentry
, mode
& S_IRWXUGO
);
570 * get references for dget and mntget, will be released
571 * in error path of *_open().
573 ret
= spufs_gang_open(dget(dentry
), mntget(mnt
));
575 int err
= simple_rmdir(inode
, dentry
);
580 mutex_unlock(&inode
->i_mutex
);
586 static struct file_system_type spufs_type
;
588 long spufs_create(struct nameidata
*nd
, unsigned int flags
, mode_t mode
,
591 struct dentry
*dentry
;
595 /* check if we are on spufs */
596 if (nd
->path
.dentry
->d_sb
->s_type
!= &spufs_type
)
599 /* don't accept undefined flags */
600 if (flags
& (~SPU_CREATE_FLAG_ALL
))
603 /* only threads can be underneath a gang */
604 if (nd
->path
.dentry
!= nd
->path
.dentry
->d_sb
->s_root
) {
605 if ((flags
& SPU_CREATE_GANG
) ||
606 !SPUFS_I(nd
->path
.dentry
->d_inode
)->i_gang
)
610 dentry
= lookup_create(nd
, 1);
611 ret
= PTR_ERR(dentry
);
619 mode
&= ~current
->fs
->umask
;
621 if (flags
& SPU_CREATE_GANG
)
622 return spufs_create_gang(nd
->path
.dentry
->d_inode
,
623 dentry
, nd
->path
.mnt
, mode
);
625 return spufs_create_context(nd
->path
.dentry
->d_inode
,
626 dentry
, nd
->path
.mnt
, flags
, mode
,
632 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
637 /* File system initialization */
639 Opt_uid
, Opt_gid
, Opt_mode
, Opt_err
,
642 static match_table_t spufs_tokens
= {
643 { Opt_uid
, "uid=%d" },
644 { Opt_gid
, "gid=%d" },
645 { Opt_mode
, "mode=%o" },
650 spufs_parse_options(char *options
, struct inode
*root
)
653 substring_t args
[MAX_OPT_ARGS
];
655 while ((p
= strsep(&options
, ",")) != NULL
) {
661 token
= match_token(p
, spufs_tokens
, args
);
664 if (match_int(&args
[0], &option
))
666 root
->i_uid
= option
;
669 if (match_int(&args
[0], &option
))
671 root
->i_gid
= option
;
674 if (match_octal(&args
[0], &option
))
676 root
->i_mode
= option
| S_IFDIR
;
685 static void spufs_exit_isolated_loader(void)
687 free_pages((unsigned long) isolated_loader
,
688 get_order(isolated_loader_size
));
692 spufs_init_isolated_loader(void)
694 struct device_node
*dn
;
698 dn
= of_find_node_by_path("/spu-isolation");
702 loader
= of_get_property(dn
, "loader", &size
);
706 /* the loader must be align on a 16 byte boundary */
707 isolated_loader
= (char *)__get_free_pages(GFP_KERNEL
, get_order(size
));
708 if (!isolated_loader
)
711 isolated_loader_size
= size
;
712 memcpy(isolated_loader
, loader
, size
);
713 printk(KERN_INFO
"spufs: SPU isolation mode enabled\n");
717 spufs_create_root(struct super_block
*sb
, void *data
)
723 if (!spu_management_ops
)
727 inode
= spufs_new_inode(sb
, S_IFDIR
| 0775);
731 inode
->i_op
= &simple_dir_inode_operations
;
732 inode
->i_fop
= &simple_dir_operations
;
733 SPUFS_I(inode
)->i_ctx
= NULL
;
736 if (!spufs_parse_options(data
, inode
))
740 sb
->s_root
= d_alloc_root(inode
);
752 spufs_fill_super(struct super_block
*sb
, void *data
, int silent
)
754 static struct super_operations s_ops
= {
755 .alloc_inode
= spufs_alloc_inode
,
756 .destroy_inode
= spufs_destroy_inode
,
757 .statfs
= simple_statfs
,
758 .delete_inode
= spufs_delete_inode
,
759 .drop_inode
= generic_delete_inode
,
760 .show_options
= generic_show_options
,
763 save_mount_options(sb
, data
);
765 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
766 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
767 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
768 sb
->s_magic
= SPUFS_MAGIC
;
771 return spufs_create_root(sb
, data
);
775 spufs_get_sb(struct file_system_type
*fstype
, int flags
,
776 const char *name
, void *data
, struct vfsmount
*mnt
)
778 return get_sb_single(fstype
, flags
, data
, spufs_fill_super
, mnt
);
781 static struct file_system_type spufs_type
= {
782 .owner
= THIS_MODULE
,
784 .get_sb
= spufs_get_sb
,
785 .kill_sb
= kill_litter_super
,
788 static int __init
spufs_init(void)
793 if (!spu_management_ops
)
797 spufs_inode_cache
= kmem_cache_create("spufs_inode_cache",
798 sizeof(struct spufs_inode_info
), 0,
799 SLAB_HWCACHE_ALIGN
, spufs_init_once
);
801 if (!spufs_inode_cache
)
803 ret
= spu_sched_init();
806 ret
= register_filesystem(&spufs_type
);
809 ret
= register_spu_syscalls(&spufs_calls
);
813 spufs_init_isolated_loader();
818 unregister_filesystem(&spufs_type
);
822 kmem_cache_destroy(spufs_inode_cache
);
826 module_init(spufs_init
);
828 static void __exit
spufs_exit(void)
831 spufs_exit_isolated_loader();
832 unregister_spu_syscalls(&spufs_calls
);
833 unregister_filesystem(&spufs_type
);
834 kmem_cache_destroy(spufs_inode_cache
);
836 module_exit(spufs_exit
);
838 MODULE_LICENSE("GPL");
839 MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");