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_evict_inode(struct inode
*inode
)
148 struct spufs_inode_info
*ei
= SPUFS_I(inode
);
149 end_writeback(inode
);
151 put_spu_context(ei
->i_ctx
);
153 put_spu_gang(ei
->i_gang
);
156 static void spufs_prune_dir(struct dentry
*dir
)
158 struct dentry
*dentry
, *tmp
;
160 mutex_lock(&dir
->d_inode
->i_mutex
);
161 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
162 spin_lock(&dentry
->d_lock
);
163 if (!(d_unhashed(dentry
)) && dentry
->d_inode
) {
164 dget_locked_dlock(dentry
);
166 spin_unlock(&dentry
->d_lock
);
167 simple_unlink(dir
->d_inode
, dentry
);
168 /* XXX: what was dcache_lock protecting here? Other
169 * filesystems (IB, configfs) release dcache_lock
173 spin_unlock(&dentry
->d_lock
);
176 shrink_dcache_parent(dir
);
177 mutex_unlock(&dir
->d_inode
->i_mutex
);
180 /* Caller must hold parent->i_mutex */
181 static int spufs_rmdir(struct inode
*parent
, struct dentry
*dir
)
183 /* remove all entries */
184 spufs_prune_dir(dir
);
187 return simple_rmdir(parent
, dir
);
190 static int spufs_fill_dir(struct dentry
*dir
,
191 const struct spufs_tree_descr
*files
, int mode
,
192 struct spu_context
*ctx
)
194 struct dentry
*dentry
, *tmp
;
197 while (files
->name
&& files
->name
[0]) {
199 dentry
= d_alloc_name(dir
, files
->name
);
202 ret
= spufs_new_file(dir
->d_sb
, dentry
, files
->ops
,
203 files
->mode
& mode
, files
->size
, ctx
);
211 * remove all children from dir. dir->inode is not set so don't
212 * just simply use spufs_prune_dir() and panic afterwards :)
213 * dput() looks like it will do the right thing:
214 * - dec parent's ref counter
215 * - remove child from parent's child list
216 * - free child's inode if possible
219 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
223 shrink_dcache_parent(dir
);
227 static int spufs_dir_close(struct inode
*inode
, struct file
*file
)
229 struct spu_context
*ctx
;
230 struct inode
*parent
;
234 dir
= file
->f_path
.dentry
;
235 parent
= dir
->d_parent
->d_inode
;
236 ctx
= SPUFS_I(dir
->d_inode
)->i_ctx
;
238 mutex_lock_nested(&parent
->i_mutex
, I_MUTEX_PARENT
);
239 ret
= spufs_rmdir(parent
, dir
);
240 mutex_unlock(&parent
->i_mutex
);
243 /* We have to give up the mm_struct */
246 return dcache_dir_close(inode
, file
);
249 const struct file_operations spufs_context_fops
= {
250 .open
= dcache_dir_open
,
251 .release
= spufs_dir_close
,
252 .llseek
= dcache_dir_lseek
,
253 .read
= generic_read_dir
,
254 .readdir
= dcache_readdir
,
257 EXPORT_SYMBOL_GPL(spufs_context_fops
);
260 spufs_mkdir(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
,
265 struct spu_context
*ctx
;
268 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
272 if (dir
->i_mode
& S_ISGID
) {
273 inode
->i_gid
= dir
->i_gid
;
274 inode
->i_mode
&= S_ISGID
;
276 ctx
= alloc_spu_context(SPUFS_I(dir
)->i_gang
); /* XXX gang */
277 SPUFS_I(inode
)->i_ctx
= ctx
;
282 inode
->i_op
= &simple_dir_inode_operations
;
283 inode
->i_fop
= &simple_dir_operations
;
284 if (flags
& SPU_CREATE_NOSCHED
)
285 ret
= spufs_fill_dir(dentry
, spufs_dir_nosched_contents
,
288 ret
= spufs_fill_dir(dentry
, spufs_dir_contents
, mode
, ctx
);
293 if (spufs_get_sb_info(dir
->i_sb
)->debug
)
294 ret
= spufs_fill_dir(dentry
, spufs_dir_debug_contents
,
300 d_instantiate(dentry
, inode
);
303 inc_nlink(dentry
->d_inode
);
308 put_spu_context(ctx
);
315 static int spufs_context_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
320 ret
= get_unused_fd();
327 filp
= dentry_open(dentry
, mnt
, O_RDONLY
, current_cred());
334 filp
->f_op
= &spufs_context_fops
;
335 fd_install(ret
, filp
);
340 static struct spu_context
*
341 spufs_assert_affinity(unsigned int flags
, struct spu_gang
*gang
,
344 struct spu_context
*tmp
, *neighbor
, *err
;
348 aff_supp
= !list_empty(&(list_entry(cbe_spu_info
[0].spus
.next
,
349 struct spu
, cbe_list
))->aff_list
);
352 return ERR_PTR(-EINVAL
);
354 if (flags
& SPU_CREATE_GANG
)
355 return ERR_PTR(-EINVAL
);
357 if (flags
& SPU_CREATE_AFFINITY_MEM
&&
359 gang
->aff_ref_ctx
->flags
& SPU_CREATE_AFFINITY_MEM
)
360 return ERR_PTR(-EEXIST
);
362 if (gang
->aff_flags
& AFF_MERGED
)
363 return ERR_PTR(-EBUSY
);
366 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
367 if (!filp
|| filp
->f_op
!= &spufs_context_fops
)
368 return ERR_PTR(-EINVAL
);
370 neighbor
= get_spu_context(
371 SPUFS_I(filp
->f_dentry
->d_inode
)->i_ctx
);
373 if (!list_empty(&neighbor
->aff_list
) && !(neighbor
->aff_head
) &&
374 !list_is_last(&neighbor
->aff_list
, &gang
->aff_list_head
) &&
375 !list_entry(neighbor
->aff_list
.next
, struct spu_context
,
376 aff_list
)->aff_head
) {
377 err
= ERR_PTR(-EEXIST
);
378 goto out_put_neighbor
;
381 if (gang
!= neighbor
->gang
) {
382 err
= ERR_PTR(-EINVAL
);
383 goto out_put_neighbor
;
387 list_for_each_entry(tmp
, &gang
->aff_list_head
, aff_list
)
389 if (list_empty(&neighbor
->aff_list
))
392 for (node
= 0; node
< MAX_NUMNODES
; node
++) {
393 if ((cbe_spu_info
[node
].n_spus
- atomic_read(
394 &cbe_spu_info
[node
].reserved_spus
)) >= count
)
398 if (node
== MAX_NUMNODES
) {
399 err
= ERR_PTR(-EEXIST
);
400 goto out_put_neighbor
;
407 put_spu_context(neighbor
);
412 spufs_set_affinity(unsigned int flags
, struct spu_context
*ctx
,
413 struct spu_context
*neighbor
)
415 if (flags
& SPU_CREATE_AFFINITY_MEM
)
416 ctx
->gang
->aff_ref_ctx
= ctx
;
418 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
419 if (list_empty(&neighbor
->aff_list
)) {
420 list_add_tail(&neighbor
->aff_list
,
421 &ctx
->gang
->aff_list_head
);
422 neighbor
->aff_head
= 1;
425 if (list_is_last(&neighbor
->aff_list
, &ctx
->gang
->aff_list_head
)
426 || list_entry(neighbor
->aff_list
.next
, struct spu_context
,
427 aff_list
)->aff_head
) {
428 list_add(&ctx
->aff_list
, &neighbor
->aff_list
);
430 list_add_tail(&ctx
->aff_list
, &neighbor
->aff_list
);
431 if (neighbor
->aff_head
) {
432 neighbor
->aff_head
= 0;
437 if (!ctx
->gang
->aff_ref_ctx
)
438 ctx
->gang
->aff_ref_ctx
= ctx
;
443 spufs_create_context(struct inode
*inode
, struct dentry
*dentry
,
444 struct vfsmount
*mnt
, int flags
, int mode
,
445 struct file
*aff_filp
)
449 struct spu_gang
*gang
;
450 struct spu_context
*neighbor
;
453 if ((flags
& SPU_CREATE_NOSCHED
) &&
454 !capable(CAP_SYS_NICE
))
458 if ((flags
& (SPU_CREATE_NOSCHED
| SPU_CREATE_ISOLATE
))
459 == SPU_CREATE_ISOLATE
)
463 if ((flags
& SPU_CREATE_ISOLATE
) && !isolated_loader
)
468 affinity
= flags
& (SPU_CREATE_AFFINITY_MEM
| SPU_CREATE_AFFINITY_SPU
);
470 gang
= SPUFS_I(inode
)->i_gang
;
474 mutex_lock(&gang
->aff_mutex
);
475 neighbor
= spufs_assert_affinity(flags
, gang
, aff_filp
);
476 if (IS_ERR(neighbor
)) {
477 ret
= PTR_ERR(neighbor
);
482 ret
= spufs_mkdir(inode
, dentry
, flags
, mode
& S_IRWXUGO
);
487 spufs_set_affinity(flags
, SPUFS_I(dentry
->d_inode
)->i_ctx
,
490 put_spu_context(neighbor
);
494 * get references for dget and mntget, will be released
495 * in error path of *_open().
497 ret
= spufs_context_open(dget(dentry
), mntget(mnt
));
499 WARN_ON(spufs_rmdir(inode
, dentry
));
501 mutex_unlock(&gang
->aff_mutex
);
502 mutex_unlock(&inode
->i_mutex
);
503 spu_forget(SPUFS_I(dentry
->d_inode
)->i_ctx
);
509 mutex_unlock(&gang
->aff_mutex
);
511 mutex_unlock(&inode
->i_mutex
);
518 spufs_mkgang(struct inode
*dir
, struct dentry
*dentry
, int mode
)
522 struct spu_gang
*gang
;
525 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
530 if (dir
->i_mode
& S_ISGID
) {
531 inode
->i_gid
= dir
->i_gid
;
532 inode
->i_mode
&= S_ISGID
;
534 gang
= alloc_spu_gang();
535 SPUFS_I(inode
)->i_ctx
= NULL
;
536 SPUFS_I(inode
)->i_gang
= gang
;
540 inode
->i_op
= &simple_dir_inode_operations
;
541 inode
->i_fop
= &simple_dir_operations
;
543 d_instantiate(dentry
, inode
);
545 inc_nlink(dentry
->d_inode
);
554 static int spufs_gang_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
559 ret
= get_unused_fd();
566 filp
= dentry_open(dentry
, mnt
, O_RDONLY
, current_cred());
573 filp
->f_op
= &simple_dir_operations
;
574 fd_install(ret
, filp
);
579 static int spufs_create_gang(struct inode
*inode
,
580 struct dentry
*dentry
,
581 struct vfsmount
*mnt
, int mode
)
585 ret
= spufs_mkgang(inode
, dentry
, mode
& S_IRWXUGO
);
590 * get references for dget and mntget, will be released
591 * in error path of *_open().
593 ret
= spufs_gang_open(dget(dentry
), mntget(mnt
));
595 int err
= simple_rmdir(inode
, dentry
);
600 mutex_unlock(&inode
->i_mutex
);
606 static struct file_system_type spufs_type
;
608 long spufs_create(struct nameidata
*nd
, unsigned int flags
, mode_t mode
,
611 struct dentry
*dentry
;
615 /* check if we are on spufs */
616 if (nd
->path
.dentry
->d_sb
->s_type
!= &spufs_type
)
619 /* don't accept undefined flags */
620 if (flags
& (~SPU_CREATE_FLAG_ALL
))
623 /* only threads can be underneath a gang */
624 if (nd
->path
.dentry
!= nd
->path
.dentry
->d_sb
->s_root
) {
625 if ((flags
& SPU_CREATE_GANG
) ||
626 !SPUFS_I(nd
->path
.dentry
->d_inode
)->i_gang
)
630 dentry
= lookup_create(nd
, 1);
631 ret
= PTR_ERR(dentry
);
635 mode
&= ~current_umask();
637 if (flags
& SPU_CREATE_GANG
)
638 ret
= spufs_create_gang(nd
->path
.dentry
->d_inode
,
639 dentry
, nd
->path
.mnt
, mode
);
641 ret
= spufs_create_context(nd
->path
.dentry
->d_inode
,
642 dentry
, nd
->path
.mnt
, flags
, mode
,
645 fsnotify_mkdir(nd
->path
.dentry
->d_inode
, dentry
);
649 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
654 /* File system initialization */
656 Opt_uid
, Opt_gid
, Opt_mode
, Opt_debug
, Opt_err
,
659 static const match_table_t spufs_tokens
= {
660 { Opt_uid
, "uid=%d" },
661 { Opt_gid
, "gid=%d" },
662 { Opt_mode
, "mode=%o" },
663 { Opt_debug
, "debug" },
668 spufs_parse_options(struct super_block
*sb
, char *options
, struct inode
*root
)
671 substring_t args
[MAX_OPT_ARGS
];
673 while ((p
= strsep(&options
, ",")) != NULL
) {
679 token
= match_token(p
, spufs_tokens
, args
);
682 if (match_int(&args
[0], &option
))
684 root
->i_uid
= option
;
687 if (match_int(&args
[0], &option
))
689 root
->i_gid
= option
;
692 if (match_octal(&args
[0], &option
))
694 root
->i_mode
= option
| S_IFDIR
;
697 spufs_get_sb_info(sb
)->debug
= 1;
706 static void spufs_exit_isolated_loader(void)
708 free_pages((unsigned long) isolated_loader
,
709 get_order(isolated_loader_size
));
713 spufs_init_isolated_loader(void)
715 struct device_node
*dn
;
719 dn
= of_find_node_by_path("/spu-isolation");
723 loader
= of_get_property(dn
, "loader", &size
);
727 /* the loader must be align on a 16 byte boundary */
728 isolated_loader
= (char *)__get_free_pages(GFP_KERNEL
, get_order(size
));
729 if (!isolated_loader
)
732 isolated_loader_size
= size
;
733 memcpy(isolated_loader
, loader
, size
);
734 printk(KERN_INFO
"spufs: SPU isolation mode enabled\n");
738 spufs_create_root(struct super_block
*sb
, void *data
)
744 if (!spu_management_ops
)
748 inode
= spufs_new_inode(sb
, S_IFDIR
| 0775);
752 inode
->i_op
= &simple_dir_inode_operations
;
753 inode
->i_fop
= &simple_dir_operations
;
754 SPUFS_I(inode
)->i_ctx
= NULL
;
758 if (!spufs_parse_options(sb
, data
, inode
))
762 sb
->s_root
= d_alloc_root(inode
);
774 spufs_fill_super(struct super_block
*sb
, void *data
, int silent
)
776 struct spufs_sb_info
*info
;
777 static const struct super_operations s_ops
= {
778 .alloc_inode
= spufs_alloc_inode
,
779 .destroy_inode
= spufs_destroy_inode
,
780 .statfs
= simple_statfs
,
781 .evict_inode
= spufs_evict_inode
,
782 .show_options
= generic_show_options
,
785 save_mount_options(sb
, data
);
787 info
= kzalloc(sizeof(*info
), GFP_KERNEL
);
791 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
792 sb
->s_blocksize
= PAGE_CACHE_SIZE
;
793 sb
->s_blocksize_bits
= PAGE_CACHE_SHIFT
;
794 sb
->s_magic
= SPUFS_MAGIC
;
796 sb
->s_fs_info
= info
;
798 return spufs_create_root(sb
, data
);
801 static struct dentry
*
802 spufs_mount(struct file_system_type
*fstype
, int flags
,
803 const char *name
, void *data
)
805 return mount_single(fstype
, flags
, data
, spufs_fill_super
);
808 static struct file_system_type spufs_type
= {
809 .owner
= THIS_MODULE
,
811 .mount
= spufs_mount
,
812 .kill_sb
= kill_litter_super
,
815 static int __init
spufs_init(void)
820 if (!spu_management_ops
)
824 spufs_inode_cache
= kmem_cache_create("spufs_inode_cache",
825 sizeof(struct spufs_inode_info
), 0,
826 SLAB_HWCACHE_ALIGN
, spufs_init_once
);
828 if (!spufs_inode_cache
)
830 ret
= spu_sched_init();
833 ret
= register_filesystem(&spufs_type
);
836 ret
= register_spu_syscalls(&spufs_calls
);
840 spufs_init_isolated_loader();
845 unregister_filesystem(&spufs_type
);
849 kmem_cache_destroy(spufs_inode_cache
);
853 module_init(spufs_init
);
855 static void __exit
spufs_exit(void)
858 spufs_exit_isolated_loader();
859 unregister_spu_syscalls(&spufs_calls
);
860 unregister_filesystem(&spufs_type
);
861 kmem_cache_destroy(spufs_inode_cache
);
863 module_exit(spufs_exit
);
865 MODULE_LICENSE("GPL");
866 MODULE_AUTHOR("Arnd Bergmann <arndb@de.ibm.com>");