2 * (C) 2001 Clemson University and The University of Chicago
4 * See COPYING in top-level directory.
8 #include "orangefs-kernel.h"
9 #include "orangefs-bufmap.h"
11 #include <linux/parser.h>
13 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
14 static struct kmem_cache
*orangefs_inode_cache
;
16 /* list for storing orangefs specific superblocks in use */
17 LIST_HEAD(orangefs_superblocks
);
19 DEFINE_SPINLOCK(orangefs_superblocks_lock
);
29 static const match_table_t tokens
= {
32 { Opt_local_lock
, "local_lock" },
36 uint64_t orangefs_features
;
38 static int parse_mount_options(struct super_block
*sb
, char *options
,
41 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(sb
);
42 substring_t args
[MAX_OPT_ARGS
];
46 * Force any potential flags that might be set from the mount
47 * to zero, ie, initialize to unset.
49 sb
->s_flags
&= ~MS_POSIXACL
;
50 orangefs_sb
->flags
&= ~ORANGEFS_OPT_INTR
;
51 orangefs_sb
->flags
&= ~ORANGEFS_OPT_LOCAL_LOCK
;
53 while ((p
= strsep(&options
, ",")) != NULL
) {
59 token
= match_token(p
, tokens
, args
);
62 sb
->s_flags
|= MS_POSIXACL
;
65 orangefs_sb
->flags
|= ORANGEFS_OPT_INTR
;
68 orangefs_sb
->flags
|= ORANGEFS_OPT_LOCAL_LOCK
;
78 gossip_err("Error: mount option [%s] is not supported.\n", p
);
82 static void orangefs_inode_cache_ctor(void *req
)
84 struct orangefs_inode_s
*orangefs_inode
= req
;
86 inode_init_once(&orangefs_inode
->vfs_inode
);
87 init_rwsem(&orangefs_inode
->xattr_sem
);
89 orangefs_inode
->vfs_inode
.i_version
= 1;
92 static struct inode
*orangefs_alloc_inode(struct super_block
*sb
)
94 struct orangefs_inode_s
*orangefs_inode
;
96 orangefs_inode
= kmem_cache_alloc(orangefs_inode_cache
, GFP_KERNEL
);
97 if (orangefs_inode
== NULL
) {
98 gossip_err("Failed to allocate orangefs_inode\n");
103 * We want to clear everything except for rw_semaphore and the
106 memset(&orangefs_inode
->refn
.khandle
, 0, 16);
107 orangefs_inode
->refn
.fs_id
= ORANGEFS_FS_ID_NULL
;
108 orangefs_inode
->last_failed_block_index_read
= 0;
109 memset(orangefs_inode
->link_target
, 0, sizeof(orangefs_inode
->link_target
));
110 orangefs_inode
->pinode_flags
= 0;
112 gossip_debug(GOSSIP_SUPER_DEBUG
,
113 "orangefs_alloc_inode: allocated %p\n",
114 &orangefs_inode
->vfs_inode
);
115 return &orangefs_inode
->vfs_inode
;
118 static void orangefs_i_callback(struct rcu_head
*head
)
120 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
121 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
122 kmem_cache_free(orangefs_inode_cache
, orangefs_inode
);
125 static void orangefs_destroy_inode(struct inode
*inode
)
127 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
129 gossip_debug(GOSSIP_SUPER_DEBUG
,
130 "%s: deallocated %p destroying inode %pU\n",
131 __func__
, orangefs_inode
, get_khandle_from_ino(inode
));
133 call_rcu(&inode
->i_rcu
, orangefs_i_callback
);
137 * NOTE: information filled in here is typically reflected in the
138 * output of the system command 'df'
140 static int orangefs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
143 struct orangefs_kernel_op_s
*new_op
= NULL
;
145 struct super_block
*sb
= NULL
;
149 gossip_debug(GOSSIP_SUPER_DEBUG
,
150 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
152 (int)(ORANGEFS_SB(sb
)->fs_id
));
154 new_op
= op_alloc(ORANGEFS_VFS_OP_STATFS
);
157 new_op
->upcall
.req
.statfs
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
159 if (ORANGEFS_SB(sb
)->flags
& ORANGEFS_OPT_INTR
)
160 flags
= ORANGEFS_OP_INTERRUPTIBLE
;
162 ret
= service_operation(new_op
, "orangefs_statfs", flags
);
164 if (new_op
->downcall
.status
< 0)
167 gossip_debug(GOSSIP_SUPER_DEBUG
,
168 "%s: got %ld blocks available | "
169 "%ld blocks total | %ld block size | "
170 "%ld files total | %ld files avail\n",
172 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
173 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
174 (long)new_op
->downcall
.resp
.statfs
.block_size
,
175 (long)new_op
->downcall
.resp
.statfs
.files_total
,
176 (long)new_op
->downcall
.resp
.statfs
.files_avail
);
178 buf
->f_type
= sb
->s_magic
;
179 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
180 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
181 buf
->f_namelen
= ORANGEFS_NAME_MAX
;
183 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
184 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
185 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
186 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
187 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
188 buf
->f_frsize
= sb
->s_blocksize
;
192 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_statfs: returning %d\n", ret
);
197 * Remount as initiated by VFS layer. We just need to reparse the mount
198 * options, no need to signal pvfs2-client-core about it.
200 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
202 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
203 return parse_mount_options(sb
, data
, 1);
207 * Remount as initiated by pvfs2-client-core on restart. This is used to
208 * repopulate mount information left from previous pvfs2-client-core.
210 * the idea here is that given a valid superblock, we're
211 * re-initializing the user space client with the initial mount
212 * information specified when the super block was first initialized.
213 * this is very different than the first initialization/creation of a
214 * superblock. we use the special service_priority_operation to make
215 * sure that the mount gets ahead of any other pending operation that
216 * is waiting for servicing. this means that the pvfs2-client won't
217 * fail to start several times for all other pending operations before
218 * the client regains all of the mount information from us.
219 * NOTE: this function assumes that the request_mutex is already acquired!
221 int orangefs_remount(struct orangefs_sb_info_s
*orangefs_sb
)
223 struct orangefs_kernel_op_s
*new_op
;
226 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
228 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
231 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
232 orangefs_sb
->devname
,
233 ORANGEFS_MAX_SERVER_ADDR_LEN
);
235 gossip_debug(GOSSIP_SUPER_DEBUG
,
236 "Attempting ORANGEFS Remount via host %s\n",
237 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
240 * we assume that the calling function has already acquired the
241 * request_mutex to prevent other operations from bypassing
244 ret
= service_operation(new_op
, "orangefs_remount",
245 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
246 gossip_debug(GOSSIP_SUPER_DEBUG
,
247 "orangefs_remount: mount got return value of %d\n",
251 * store the id assigned to this sb -- it's just a
252 * short-lived mapping that the system interface uses
253 * to map this superblock to a particular mount entry
255 orangefs_sb
->id
= new_op
->downcall
.resp
.fs_mount
.id
;
256 orangefs_sb
->mount_pending
= 0;
261 if (orangefs_userspace_version
>= 20906) {
262 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
265 new_op
->upcall
.req
.features
.features
= 0;
266 ret
= service_operation(new_op
, "orangefs_features",
267 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
270 new_op
->downcall
.resp
.features
.features
;
272 orangefs_features
= 0;
275 orangefs_features
= 0;
281 int fsid_key_table_initialize(void)
286 void fsid_key_table_finalize(void)
290 /* Called whenever the VFS dirties the inode in response to atime updates */
291 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
293 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
295 gossip_debug(GOSSIP_SUPER_DEBUG
,
296 "orangefs_dirty_inode: %pU\n",
297 get_khandle_from_ino(inode
));
298 SetAtimeFlag(orangefs_inode
);
301 static const struct super_operations orangefs_s_ops
= {
302 .alloc_inode
= orangefs_alloc_inode
,
303 .destroy_inode
= orangefs_destroy_inode
,
304 .dirty_inode
= orangefs_dirty_inode
,
305 .drop_inode
= generic_delete_inode
,
306 .statfs
= orangefs_statfs
,
307 .remount_fs
= orangefs_remount_fs
,
308 .show_options
= generic_show_options
,
311 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
316 struct orangefs_object_kref refn
;
318 if (fh_len
< 5 || fh_type
> 2)
321 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
322 refn
.fs_id
= (u32
) fid
->raw
[4];
323 gossip_debug(GOSSIP_SUPER_DEBUG
,
324 "fh_to_dentry: handle %pU, fs_id %d\n",
328 return d_obtain_alias(orangefs_iget(sb
, &refn
));
331 static int orangefs_encode_fh(struct inode
*inode
,
334 struct inode
*parent
)
336 int len
= parent
? 10 : 5;
338 struct orangefs_object_kref refn
;
340 if (*max_len
< len
) {
341 gossip_lerr("fh buffer is too small for encoding\n");
347 refn
= ORANGEFS_I(inode
)->refn
;
348 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
351 gossip_debug(GOSSIP_SUPER_DEBUG
,
352 "Encoding fh: handle %pU, fsid %u\n",
358 refn
= ORANGEFS_I(parent
)->refn
;
359 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
363 gossip_debug(GOSSIP_SUPER_DEBUG
,
364 "Encoding parent: handle %pU, fsid %u\n",
374 static const struct export_operations orangefs_export_ops
= {
375 .encode_fh
= orangefs_encode_fh
,
376 .fh_to_dentry
= orangefs_fh_to_dentry
,
379 static int orangefs_unmount(int id
, __s32 fs_id
, const char *devname
)
381 struct orangefs_kernel_op_s
*op
;
383 op
= op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT
);
386 op
->upcall
.req
.fs_umount
.id
= id
;
387 op
->upcall
.req
.fs_umount
.fs_id
= fs_id
;
388 strncpy(op
->upcall
.req
.fs_umount
.orangefs_config_server
,
389 devname
, ORANGEFS_MAX_SERVER_ADDR_LEN
);
390 r
= service_operation(op
, "orangefs_fs_umount", 0);
391 /* Not much to do about an error here. */
393 gossip_err("orangefs_unmount: service_operation %d\n", r
);
398 static int orangefs_fill_sb(struct super_block
*sb
,
399 struct orangefs_fs_mount_response
*fs_mount
,
400 void *data
, int silent
)
403 struct inode
*root
= NULL
;
404 struct dentry
*root_dentry
= NULL
;
405 struct orangefs_object_kref root_object
;
407 /* alloc and init our private orangefs sb info */
408 sb
->s_fs_info
= kzalloc(sizeof(struct orangefs_sb_info_s
), GFP_KERNEL
);
409 if (!ORANGEFS_SB(sb
))
411 ORANGEFS_SB(sb
)->sb
= sb
;
413 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
414 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
415 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
418 ret
= parse_mount_options(sb
, data
, silent
);
423 /* Hang the xattr handlers off the superblock */
424 sb
->s_xattr
= orangefs_xattr_handlers
;
425 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
426 sb
->s_op
= &orangefs_s_ops
;
427 sb
->s_d_op
= &orangefs_dentry_operations
;
429 sb
->s_blocksize
= orangefs_bufmap_size_query();
430 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
431 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
433 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
434 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
435 gossip_debug(GOSSIP_SUPER_DEBUG
,
436 "get inode %pU, fsid %d\n",
437 &root_object
.khandle
,
440 root
= orangefs_iget(sb
, &root_object
);
442 return PTR_ERR(root
);
444 gossip_debug(GOSSIP_SUPER_DEBUG
,
445 "Allocated root inode [%p] with mode %x\n",
449 /* allocates and places root dentry in dcache */
450 root_dentry
= d_make_root(root
);
454 sb
->s_export_op
= &orangefs_export_ops
;
455 sb
->s_root
= root_dentry
;
459 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
465 struct super_block
*sb
= ERR_PTR(-EINVAL
);
466 struct orangefs_kernel_op_s
*new_op
;
467 struct dentry
*d
= ERR_PTR(-EINVAL
);
469 gossip_debug(GOSSIP_SUPER_DEBUG
,
470 "orangefs_mount: called with devname %s\n",
474 gossip_err("ERROR: device name not specified.\n");
475 return ERR_PTR(-EINVAL
);
478 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
480 return ERR_PTR(-ENOMEM
);
482 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
484 ORANGEFS_MAX_SERVER_ADDR_LEN
);
486 gossip_debug(GOSSIP_SUPER_DEBUG
,
487 "Attempting ORANGEFS Mount via host %s\n",
488 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
490 ret
= service_operation(new_op
, "orangefs_mount", 0);
491 gossip_debug(GOSSIP_SUPER_DEBUG
,
492 "orangefs_mount: mount got return value of %d\n", ret
);
496 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
497 gossip_err("ERROR: Retrieved null fs_id\n");
502 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
506 orangefs_unmount(new_op
->downcall
.resp
.fs_mount
.id
,
507 new_op
->downcall
.resp
.fs_mount
.fs_id
, devname
);
511 ret
= orangefs_fill_sb(sb
,
512 &new_op
->downcall
.resp
.fs_mount
, data
,
513 flags
& MS_SILENT
? 1 : 0);
521 * on successful mount, store the devname and data
524 strncpy(ORANGEFS_SB(sb
)->devname
,
526 ORANGEFS_MAX_SERVER_ADDR_LEN
);
528 /* mount_pending must be cleared */
529 ORANGEFS_SB(sb
)->mount_pending
= 0;
532 * finally, add this sb to our list of known orangefs
535 gossip_debug(GOSSIP_SUPER_DEBUG
,
536 "Adding SB %p to orangefs superblocks\n",
538 spin_lock(&orangefs_superblocks_lock
);
539 list_add_tail(&ORANGEFS_SB(sb
)->list
, &orangefs_superblocks
);
540 spin_unlock(&orangefs_superblocks_lock
);
543 /* Must be removed from the list now. */
544 ORANGEFS_SB(sb
)->no_list
= 0;
546 if (orangefs_userspace_version
>= 20906) {
547 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
549 return ERR_PTR(-ENOMEM
);
550 new_op
->upcall
.req
.features
.features
= 0;
551 ret
= service_operation(new_op
, "orangefs_features", 0);
552 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
555 orangefs_features
= 0;
558 return dget(sb
->s_root
);
561 /* Will call orangefs_kill_sb with sb not in list. */
562 ORANGEFS_SB(sb
)->no_list
= 1;
563 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
564 deactivate_locked_super(sb
);
566 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
567 if (ret
== -EINVAL
) {
568 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
569 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
577 void orangefs_kill_sb(struct super_block
*sb
)
580 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
582 /* provided sb cleanup */
586 * issue the unmount to userspace to tell it to remove the
587 * dynamic mount info it has for this superblock
589 r
= orangefs_unmount(ORANGEFS_SB(sb
)->id
, ORANGEFS_SB(sb
)->fs_id
,
590 ORANGEFS_SB(sb
)->devname
);
592 ORANGEFS_SB(sb
)->mount_pending
= 1;
594 if (!ORANGEFS_SB(sb
)->no_list
) {
595 /* remove the sb from our list of orangefs specific sb's */
596 spin_lock(&orangefs_superblocks_lock
);
597 /* not list_del_init */
598 __list_del_entry(&ORANGEFS_SB(sb
)->list
);
599 ORANGEFS_SB(sb
)->list
.prev
= NULL
;
600 spin_unlock(&orangefs_superblocks_lock
);
604 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
605 * gets completed before we free the dang thing.
607 mutex_lock(&orangefs_request_mutex
);
608 mutex_unlock(&orangefs_request_mutex
);
610 /* free the orangefs superblock private data */
611 kfree(ORANGEFS_SB(sb
));
614 int orangefs_inode_cache_initialize(void)
616 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
617 sizeof(struct orangefs_inode_s
),
619 ORANGEFS_CACHE_CREATE_FLAGS
,
620 orangefs_inode_cache_ctor
);
622 if (!orangefs_inode_cache
) {
623 gossip_err("Cannot create orangefs_inode_cache\n");
629 int orangefs_inode_cache_finalize(void)
631 kmem_cache_destroy(orangefs_inode_cache
);