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 return inode_setattr(inode
, attr
);
118 spufs_new_file(struct super_block
*sb
, struct dentry
*dentry
,
119 const struct file_operations
*fops
, int mode
,
120 size_t size
, struct spu_context
*ctx
)
122 static struct inode_operations spufs_file_iops
= {
123 .setattr
= spufs_setattr
,
129 inode
= spufs_new_inode(sb
, S_IFREG
| mode
);
134 inode
->i_op
= &spufs_file_iops
;
136 inode
->i_size
= size
;
137 inode
->i_private
= SPUFS_I(inode
)->i_ctx
= get_spu_context(ctx
);
138 d_add(dentry
, inode
);
144 spufs_delete_inode(struct inode
*inode
)
146 struct spufs_inode_info
*ei
= SPUFS_I(inode
);
149 put_spu_context(ei
->i_ctx
);
151 put_spu_gang(ei
->i_gang
);
155 static void spufs_prune_dir(struct dentry
*dir
)
157 struct dentry
*dentry
, *tmp
;
159 mutex_lock(&dir
->d_inode
->i_mutex
);
160 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
161 spin_lock(&dcache_lock
);
162 spin_lock(&dentry
->d_lock
);
163 if (!(d_unhashed(dentry
)) && dentry
->d_inode
) {
166 spin_unlock(&dentry
->d_lock
);
167 simple_unlink(dir
->d_inode
, dentry
);
168 spin_unlock(&dcache_lock
);
171 spin_unlock(&dentry
->d_lock
);
172 spin_unlock(&dcache_lock
);
175 shrink_dcache_parent(dir
);
176 mutex_unlock(&dir
->d_inode
->i_mutex
);
179 /* Caller must hold parent->i_mutex */
180 static int spufs_rmdir(struct inode
*parent
, struct dentry
*dir
)
182 /* remove all entries */
183 spufs_prune_dir(dir
);
186 return simple_rmdir(parent
, dir
);
189 static int spufs_fill_dir(struct dentry
*dir
,
190 const struct spufs_tree_descr
*files
, int mode
,
191 struct spu_context
*ctx
)
193 struct dentry
*dentry
, *tmp
;
196 while (files
->name
&& files
->name
[0]) {
198 dentry
= d_alloc_name(dir
, files
->name
);
201 ret
= spufs_new_file(dir
->d_sb
, dentry
, files
->ops
,
202 files
->mode
& mode
, files
->size
, ctx
);
210 * remove all children from dir. dir->inode is not set so don't
211 * just simply use spufs_prune_dir() and panic afterwards :)
212 * dput() looks like it will do the right thing:
213 * - dec parent's ref counter
214 * - remove child from parent's child list
215 * - free child's inode if possible
218 list_for_each_entry_safe(dentry
, tmp
, &dir
->d_subdirs
, d_u
.d_child
) {
222 shrink_dcache_parent(dir
);
226 static int spufs_dir_close(struct inode
*inode
, struct file
*file
)
228 struct spu_context
*ctx
;
229 struct inode
*parent
;
233 dir
= file
->f_path
.dentry
;
234 parent
= dir
->d_parent
->d_inode
;
235 ctx
= SPUFS_I(dir
->d_inode
)->i_ctx
;
237 mutex_lock_nested(&parent
->i_mutex
, I_MUTEX_PARENT
);
238 ret
= spufs_rmdir(parent
, dir
);
239 mutex_unlock(&parent
->i_mutex
);
242 /* We have to give up the mm_struct */
245 return dcache_dir_close(inode
, file
);
248 const struct file_operations spufs_context_fops
= {
249 .open
= dcache_dir_open
,
250 .release
= spufs_dir_close
,
251 .llseek
= dcache_dir_lseek
,
252 .read
= generic_read_dir
,
253 .readdir
= dcache_readdir
,
254 .fsync
= simple_sync_file
,
256 EXPORT_SYMBOL_GPL(spufs_context_fops
);
259 spufs_mkdir(struct inode
*dir
, struct dentry
*dentry
, unsigned int flags
,
264 struct spu_context
*ctx
;
267 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
271 if (dir
->i_mode
& S_ISGID
) {
272 inode
->i_gid
= dir
->i_gid
;
273 inode
->i_mode
&= S_ISGID
;
275 ctx
= alloc_spu_context(SPUFS_I(dir
)->i_gang
); /* XXX gang */
276 SPUFS_I(inode
)->i_ctx
= ctx
;
281 inode
->i_op
= &simple_dir_inode_operations
;
282 inode
->i_fop
= &simple_dir_operations
;
283 if (flags
& SPU_CREATE_NOSCHED
)
284 ret
= spufs_fill_dir(dentry
, spufs_dir_nosched_contents
,
287 ret
= spufs_fill_dir(dentry
, spufs_dir_contents
, mode
, ctx
);
292 if (spufs_get_sb_info(dir
->i_sb
)->debug
)
293 ret
= spufs_fill_dir(dentry
, spufs_dir_debug_contents
,
299 d_instantiate(dentry
, inode
);
302 inc_nlink(dentry
->d_inode
);
307 put_spu_context(ctx
);
314 static int spufs_context_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
319 ret
= get_unused_fd();
326 filp
= dentry_open(dentry
, mnt
, O_RDONLY
, current_cred());
333 filp
->f_op
= &spufs_context_fops
;
334 fd_install(ret
, filp
);
339 static struct spu_context
*
340 spufs_assert_affinity(unsigned int flags
, struct spu_gang
*gang
,
343 struct spu_context
*tmp
, *neighbor
, *err
;
347 aff_supp
= !list_empty(&(list_entry(cbe_spu_info
[0].spus
.next
,
348 struct spu
, cbe_list
))->aff_list
);
351 return ERR_PTR(-EINVAL
);
353 if (flags
& SPU_CREATE_GANG
)
354 return ERR_PTR(-EINVAL
);
356 if (flags
& SPU_CREATE_AFFINITY_MEM
&&
358 gang
->aff_ref_ctx
->flags
& SPU_CREATE_AFFINITY_MEM
)
359 return ERR_PTR(-EEXIST
);
361 if (gang
->aff_flags
& AFF_MERGED
)
362 return ERR_PTR(-EBUSY
);
365 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
366 if (!filp
|| filp
->f_op
!= &spufs_context_fops
)
367 return ERR_PTR(-EINVAL
);
369 neighbor
= get_spu_context(
370 SPUFS_I(filp
->f_dentry
->d_inode
)->i_ctx
);
372 if (!list_empty(&neighbor
->aff_list
) && !(neighbor
->aff_head
) &&
373 !list_is_last(&neighbor
->aff_list
, &gang
->aff_list_head
) &&
374 !list_entry(neighbor
->aff_list
.next
, struct spu_context
,
375 aff_list
)->aff_head
) {
376 err
= ERR_PTR(-EEXIST
);
377 goto out_put_neighbor
;
380 if (gang
!= neighbor
->gang
) {
381 err
= ERR_PTR(-EINVAL
);
382 goto out_put_neighbor
;
386 list_for_each_entry(tmp
, &gang
->aff_list_head
, aff_list
)
388 if (list_empty(&neighbor
->aff_list
))
391 for (node
= 0; node
< MAX_NUMNODES
; node
++) {
392 if ((cbe_spu_info
[node
].n_spus
- atomic_read(
393 &cbe_spu_info
[node
].reserved_spus
)) >= count
)
397 if (node
== MAX_NUMNODES
) {
398 err
= ERR_PTR(-EEXIST
);
399 goto out_put_neighbor
;
406 put_spu_context(neighbor
);
411 spufs_set_affinity(unsigned int flags
, struct spu_context
*ctx
,
412 struct spu_context
*neighbor
)
414 if (flags
& SPU_CREATE_AFFINITY_MEM
)
415 ctx
->gang
->aff_ref_ctx
= ctx
;
417 if (flags
& SPU_CREATE_AFFINITY_SPU
) {
418 if (list_empty(&neighbor
->aff_list
)) {
419 list_add_tail(&neighbor
->aff_list
,
420 &ctx
->gang
->aff_list_head
);
421 neighbor
->aff_head
= 1;
424 if (list_is_last(&neighbor
->aff_list
, &ctx
->gang
->aff_list_head
)
425 || list_entry(neighbor
->aff_list
.next
, struct spu_context
,
426 aff_list
)->aff_head
) {
427 list_add(&ctx
->aff_list
, &neighbor
->aff_list
);
429 list_add_tail(&ctx
->aff_list
, &neighbor
->aff_list
);
430 if (neighbor
->aff_head
) {
431 neighbor
->aff_head
= 0;
436 if (!ctx
->gang
->aff_ref_ctx
)
437 ctx
->gang
->aff_ref_ctx
= ctx
;
442 spufs_create_context(struct inode
*inode
, struct dentry
*dentry
,
443 struct vfsmount
*mnt
, int flags
, int mode
,
444 struct file
*aff_filp
)
448 struct spu_gang
*gang
;
449 struct spu_context
*neighbor
;
452 if ((flags
& SPU_CREATE_NOSCHED
) &&
453 !capable(CAP_SYS_NICE
))
457 if ((flags
& (SPU_CREATE_NOSCHED
| SPU_CREATE_ISOLATE
))
458 == SPU_CREATE_ISOLATE
)
462 if ((flags
& SPU_CREATE_ISOLATE
) && !isolated_loader
)
467 affinity
= flags
& (SPU_CREATE_AFFINITY_MEM
| SPU_CREATE_AFFINITY_SPU
);
469 gang
= SPUFS_I(inode
)->i_gang
;
473 mutex_lock(&gang
->aff_mutex
);
474 neighbor
= spufs_assert_affinity(flags
, gang
, aff_filp
);
475 if (IS_ERR(neighbor
)) {
476 ret
= PTR_ERR(neighbor
);
481 ret
= spufs_mkdir(inode
, dentry
, flags
, mode
& S_IRWXUGO
);
486 spufs_set_affinity(flags
, SPUFS_I(dentry
->d_inode
)->i_ctx
,
489 put_spu_context(neighbor
);
493 * get references for dget and mntget, will be released
494 * in error path of *_open().
496 ret
= spufs_context_open(dget(dentry
), mntget(mnt
));
498 WARN_ON(spufs_rmdir(inode
, dentry
));
500 mutex_unlock(&gang
->aff_mutex
);
501 mutex_unlock(&inode
->i_mutex
);
502 spu_forget(SPUFS_I(dentry
->d_inode
)->i_ctx
);
508 mutex_unlock(&gang
->aff_mutex
);
510 mutex_unlock(&inode
->i_mutex
);
517 spufs_mkgang(struct inode
*dir
, struct dentry
*dentry
, int mode
)
521 struct spu_gang
*gang
;
524 inode
= spufs_new_inode(dir
->i_sb
, mode
| S_IFDIR
);
529 if (dir
->i_mode
& S_ISGID
) {
530 inode
->i_gid
= dir
->i_gid
;
531 inode
->i_mode
&= S_ISGID
;
533 gang
= alloc_spu_gang();
534 SPUFS_I(inode
)->i_ctx
= NULL
;
535 SPUFS_I(inode
)->i_gang
= gang
;
539 inode
->i_op
= &simple_dir_inode_operations
;
540 inode
->i_fop
= &simple_dir_operations
;
542 d_instantiate(dentry
, inode
);
544 inc_nlink(dentry
->d_inode
);
553 static int spufs_gang_open(struct dentry
*dentry
, struct vfsmount
*mnt
)
558 ret
= get_unused_fd();
565 filp
= dentry_open(dentry
, mnt
, O_RDONLY
, current_cred());
572 filp
->f_op
= &simple_dir_operations
;
573 fd_install(ret
, filp
);
578 static int spufs_create_gang(struct inode
*inode
,
579 struct dentry
*dentry
,
580 struct vfsmount
*mnt
, int mode
)
584 ret
= spufs_mkgang(inode
, dentry
, mode
& S_IRWXUGO
);
589 * get references for dget and mntget, will be released
590 * in error path of *_open().
592 ret
= spufs_gang_open(dget(dentry
), mntget(mnt
));
594 int err
= simple_rmdir(inode
, dentry
);
599 mutex_unlock(&inode
->i_mutex
);
605 static struct file_system_type spufs_type
;
607 long spufs_create(struct nameidata
*nd
, unsigned int flags
, mode_t mode
,
610 struct dentry
*dentry
;
614 /* check if we are on spufs */
615 if (nd
->path
.dentry
->d_sb
->s_type
!= &spufs_type
)
618 /* don't accept undefined flags */
619 if (flags
& (~SPU_CREATE_FLAG_ALL
))
622 /* only threads can be underneath a gang */
623 if (nd
->path
.dentry
!= nd
->path
.dentry
->d_sb
->s_root
) {
624 if ((flags
& SPU_CREATE_GANG
) ||
625 !SPUFS_I(nd
->path
.dentry
->d_inode
)->i_gang
)
629 dentry
= lookup_create(nd
, 1);
630 ret
= PTR_ERR(dentry
);
634 mode
&= ~current_umask();
636 if (flags
& SPU_CREATE_GANG
)
637 ret
= spufs_create_gang(nd
->path
.dentry
->d_inode
,
638 dentry
, nd
->path
.mnt
, mode
);
640 ret
= spufs_create_context(nd
->path
.dentry
->d_inode
,
641 dentry
, nd
->path
.mnt
, flags
, mode
,
644 fsnotify_mkdir(nd
->path
.dentry
->d_inode
, dentry
);
648 mutex_unlock(&nd
->path
.dentry
->d_inode
->i_mutex
);
653 /* File system initialization */
655 Opt_uid
, Opt_gid
, Opt_mode
, Opt_debug
, Opt_err
,
658 static const match_table_t spufs_tokens
= {
659 { Opt_uid
, "uid=%d" },
660 { Opt_gid
, "gid=%d" },
661 { Opt_mode
, "mode=%o" },
662 { Opt_debug
, "debug" },
667 spufs_parse_options(struct super_block
*sb
, char *options
, struct inode
*root
)
670 substring_t args
[MAX_OPT_ARGS
];
672 while ((p
= strsep(&options
, ",")) != NULL
) {
678 token
= match_token(p
, spufs_tokens
, args
);
681 if (match_int(&args
[0], &option
))
683 root
->i_uid
= option
;
686 if (match_int(&args
[0], &option
))
688 root
->i_gid
= option
;
691 if (match_octal(&args
[0], &option
))
693 root
->i_mode
= option
| S_IFDIR
;
696 spufs_get_sb_info(sb
)->debug
= 1;
705 static void spufs_exit_isolated_loader(void)
707 free_pages((unsigned long) isolated_loader
,
708 get_order(isolated_loader_size
));
712 spufs_init_isolated_loader(void)
714 struct device_node
*dn
;
718 dn
= of_find_node_by_path("/spu-isolation");
722 loader
= of_get_property(dn
, "loader", &size
);
726 /* the loader must be align on a 16 byte boundary */
727 isolated_loader
= (char *)__get_free_pages(GFP_KERNEL
, get_order(size
));
728 if (!isolated_loader
)
731 isolated_loader_size
= size
;
732 memcpy(isolated_loader
, loader
, size
);
733 printk(KERN_INFO
"spufs: SPU isolation mode enabled\n");
737 spufs_create_root(struct super_block
*sb
, void *data
)
743 if (!spu_management_ops
)
747 inode
= spufs_new_inode(sb
, S_IFDIR
| 0775);
751 inode
->i_op
= &simple_dir_inode_operations
;
752 inode
->i_fop
= &simple_dir_operations
;
753 SPUFS_I(inode
)->i_ctx
= NULL
;
757 if (!spufs_parse_options(sb
, data
, inode
))
761 sb
->s_root
= d_alloc_root(inode
);
773 spufs_fill_super(struct super_block
*sb
, void *data
, int silent
)
775 struct spufs_sb_info
*info
;
776 static struct super_operations s_ops
= {
777 .alloc_inode
= spufs_alloc_inode
,
778 .destroy_inode
= spufs_destroy_inode
,
779 .statfs
= simple_statfs
,
780 .delete_inode
= spufs_delete_inode
,
781 .drop_inode
= generic_delete_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
);
802 spufs_get_sb(struct file_system_type
*fstype
, int flags
,
803 const char *name
, void *data
, struct vfsmount
*mnt
)
805 return get_sb_single(fstype
, flags
, data
, spufs_fill_super
, mnt
);
808 static struct file_system_type spufs_type
= {
809 .owner
= THIS_MODULE
,
811 .get_sb
= spufs_get_sb
,
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>");