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 orangefs_show_options(struct seq_file
*m
, struct dentry
*root
)
40 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(root
->d_sb
);
42 if (root
->d_sb
->s_flags
& MS_POSIXACL
)
44 if (orangefs_sb
->flags
& ORANGEFS_OPT_INTR
)
46 if (orangefs_sb
->flags
& ORANGEFS_OPT_LOCAL_LOCK
)
47 seq_puts(m
, ",local_lock");
51 static int parse_mount_options(struct super_block
*sb
, char *options
,
54 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(sb
);
55 substring_t args
[MAX_OPT_ARGS
];
59 * Force any potential flags that might be set from the mount
60 * to zero, ie, initialize to unset.
62 sb
->s_flags
&= ~MS_POSIXACL
;
63 orangefs_sb
->flags
&= ~ORANGEFS_OPT_INTR
;
64 orangefs_sb
->flags
&= ~ORANGEFS_OPT_LOCAL_LOCK
;
66 while ((p
= strsep(&options
, ",")) != NULL
) {
72 token
= match_token(p
, tokens
, args
);
75 sb
->s_flags
|= MS_POSIXACL
;
78 orangefs_sb
->flags
|= ORANGEFS_OPT_INTR
;
81 orangefs_sb
->flags
|= ORANGEFS_OPT_LOCAL_LOCK
;
91 gossip_err("Error: mount option [%s] is not supported.\n", p
);
95 static void orangefs_inode_cache_ctor(void *req
)
97 struct orangefs_inode_s
*orangefs_inode
= req
;
99 inode_init_once(&orangefs_inode
->vfs_inode
);
100 init_rwsem(&orangefs_inode
->xattr_sem
);
102 orangefs_inode
->vfs_inode
.i_version
= 1;
105 static struct inode
*orangefs_alloc_inode(struct super_block
*sb
)
107 struct orangefs_inode_s
*orangefs_inode
;
109 orangefs_inode
= kmem_cache_alloc(orangefs_inode_cache
, GFP_KERNEL
);
110 if (orangefs_inode
== NULL
) {
111 gossip_err("Failed to allocate orangefs_inode\n");
116 * We want to clear everything except for rw_semaphore and the
119 memset(&orangefs_inode
->refn
.khandle
, 0, 16);
120 orangefs_inode
->refn
.fs_id
= ORANGEFS_FS_ID_NULL
;
121 orangefs_inode
->last_failed_block_index_read
= 0;
122 memset(orangefs_inode
->link_target
, 0, sizeof(orangefs_inode
->link_target
));
123 orangefs_inode
->pinode_flags
= 0;
125 gossip_debug(GOSSIP_SUPER_DEBUG
,
126 "orangefs_alloc_inode: allocated %p\n",
127 &orangefs_inode
->vfs_inode
);
128 return &orangefs_inode
->vfs_inode
;
131 static void orangefs_i_callback(struct rcu_head
*head
)
133 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
134 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
135 kmem_cache_free(orangefs_inode_cache
, orangefs_inode
);
138 static void orangefs_destroy_inode(struct inode
*inode
)
140 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
142 gossip_debug(GOSSIP_SUPER_DEBUG
,
143 "%s: deallocated %p destroying inode %pU\n",
144 __func__
, orangefs_inode
, get_khandle_from_ino(inode
));
146 call_rcu(&inode
->i_rcu
, orangefs_i_callback
);
150 * NOTE: information filled in here is typically reflected in the
151 * output of the system command 'df'
153 static int orangefs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
156 struct orangefs_kernel_op_s
*new_op
= NULL
;
158 struct super_block
*sb
= NULL
;
162 gossip_debug(GOSSIP_SUPER_DEBUG
,
163 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
165 (int)(ORANGEFS_SB(sb
)->fs_id
));
167 new_op
= op_alloc(ORANGEFS_VFS_OP_STATFS
);
170 new_op
->upcall
.req
.statfs
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
172 if (ORANGEFS_SB(sb
)->flags
& ORANGEFS_OPT_INTR
)
173 flags
= ORANGEFS_OP_INTERRUPTIBLE
;
175 ret
= service_operation(new_op
, "orangefs_statfs", flags
);
177 if (new_op
->downcall
.status
< 0)
180 gossip_debug(GOSSIP_SUPER_DEBUG
,
181 "%s: got %ld blocks available | "
182 "%ld blocks total | %ld block size | "
183 "%ld files total | %ld files avail\n",
185 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
186 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
187 (long)new_op
->downcall
.resp
.statfs
.block_size
,
188 (long)new_op
->downcall
.resp
.statfs
.files_total
,
189 (long)new_op
->downcall
.resp
.statfs
.files_avail
);
191 buf
->f_type
= sb
->s_magic
;
192 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
193 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
194 buf
->f_namelen
= ORANGEFS_NAME_MAX
;
196 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
197 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
198 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
199 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
200 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
201 buf
->f_frsize
= sb
->s_blocksize
;
205 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_statfs: returning %d\n", ret
);
210 * Remount as initiated by VFS layer. We just need to reparse the mount
211 * options, no need to signal pvfs2-client-core about it.
213 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
215 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
216 return parse_mount_options(sb
, data
, 1);
220 * Remount as initiated by pvfs2-client-core on restart. This is used to
221 * repopulate mount information left from previous pvfs2-client-core.
223 * the idea here is that given a valid superblock, we're
224 * re-initializing the user space client with the initial mount
225 * information specified when the super block was first initialized.
226 * this is very different than the first initialization/creation of a
227 * superblock. we use the special service_priority_operation to make
228 * sure that the mount gets ahead of any other pending operation that
229 * is waiting for servicing. this means that the pvfs2-client won't
230 * fail to start several times for all other pending operations before
231 * the client regains all of the mount information from us.
232 * NOTE: this function assumes that the request_mutex is already acquired!
234 int orangefs_remount(struct orangefs_sb_info_s
*orangefs_sb
)
236 struct orangefs_kernel_op_s
*new_op
;
239 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
241 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
244 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
245 orangefs_sb
->devname
,
246 ORANGEFS_MAX_SERVER_ADDR_LEN
);
248 gossip_debug(GOSSIP_SUPER_DEBUG
,
249 "Attempting ORANGEFS Remount via host %s\n",
250 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
253 * we assume that the calling function has already acquired the
254 * request_mutex to prevent other operations from bypassing
257 ret
= service_operation(new_op
, "orangefs_remount",
258 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
259 gossip_debug(GOSSIP_SUPER_DEBUG
,
260 "orangefs_remount: mount got return value of %d\n",
264 * store the id assigned to this sb -- it's just a
265 * short-lived mapping that the system interface uses
266 * to map this superblock to a particular mount entry
268 orangefs_sb
->id
= new_op
->downcall
.resp
.fs_mount
.id
;
269 orangefs_sb
->mount_pending
= 0;
274 if (orangefs_userspace_version
>= 20906) {
275 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
278 new_op
->upcall
.req
.features
.features
= 0;
279 ret
= service_operation(new_op
, "orangefs_features",
280 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
283 new_op
->downcall
.resp
.features
.features
;
285 orangefs_features
= 0;
288 orangefs_features
= 0;
294 int fsid_key_table_initialize(void)
299 void fsid_key_table_finalize(void)
303 /* Called whenever the VFS dirties the inode in response to atime updates */
304 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
306 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
308 gossip_debug(GOSSIP_SUPER_DEBUG
,
309 "orangefs_dirty_inode: %pU\n",
310 get_khandle_from_ino(inode
));
311 SetAtimeFlag(orangefs_inode
);
314 static const struct super_operations orangefs_s_ops
= {
315 .alloc_inode
= orangefs_alloc_inode
,
316 .destroy_inode
= orangefs_destroy_inode
,
317 .dirty_inode
= orangefs_dirty_inode
,
318 .drop_inode
= generic_delete_inode
,
319 .statfs
= orangefs_statfs
,
320 .remount_fs
= orangefs_remount_fs
,
321 .show_options
= orangefs_show_options
,
324 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
329 struct orangefs_object_kref refn
;
331 if (fh_len
< 5 || fh_type
> 2)
334 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
335 refn
.fs_id
= (u32
) fid
->raw
[4];
336 gossip_debug(GOSSIP_SUPER_DEBUG
,
337 "fh_to_dentry: handle %pU, fs_id %d\n",
341 return d_obtain_alias(orangefs_iget(sb
, &refn
));
344 static int orangefs_encode_fh(struct inode
*inode
,
347 struct inode
*parent
)
349 int len
= parent
? 10 : 5;
351 struct orangefs_object_kref refn
;
353 if (*max_len
< len
) {
354 gossip_lerr("fh buffer is too small for encoding\n");
360 refn
= ORANGEFS_I(inode
)->refn
;
361 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
364 gossip_debug(GOSSIP_SUPER_DEBUG
,
365 "Encoding fh: handle %pU, fsid %u\n",
371 refn
= ORANGEFS_I(parent
)->refn
;
372 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
376 gossip_debug(GOSSIP_SUPER_DEBUG
,
377 "Encoding parent: handle %pU, fsid %u\n",
387 static const struct export_operations orangefs_export_ops
= {
388 .encode_fh
= orangefs_encode_fh
,
389 .fh_to_dentry
= orangefs_fh_to_dentry
,
392 static int orangefs_unmount(int id
, __s32 fs_id
, const char *devname
)
394 struct orangefs_kernel_op_s
*op
;
396 op
= op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT
);
399 op
->upcall
.req
.fs_umount
.id
= id
;
400 op
->upcall
.req
.fs_umount
.fs_id
= fs_id
;
401 strncpy(op
->upcall
.req
.fs_umount
.orangefs_config_server
,
402 devname
, ORANGEFS_MAX_SERVER_ADDR_LEN
);
403 r
= service_operation(op
, "orangefs_fs_umount", 0);
404 /* Not much to do about an error here. */
406 gossip_err("orangefs_unmount: service_operation %d\n", r
);
411 static int orangefs_fill_sb(struct super_block
*sb
,
412 struct orangefs_fs_mount_response
*fs_mount
,
413 void *data
, int silent
)
416 struct inode
*root
= NULL
;
417 struct dentry
*root_dentry
= NULL
;
418 struct orangefs_object_kref root_object
;
420 /* alloc and init our private orangefs sb info */
421 sb
->s_fs_info
= kzalloc(sizeof(struct orangefs_sb_info_s
), GFP_KERNEL
);
422 if (!ORANGEFS_SB(sb
))
424 ORANGEFS_SB(sb
)->sb
= sb
;
426 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
427 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
428 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
431 ret
= parse_mount_options(sb
, data
, silent
);
436 /* Hang the xattr handlers off the superblock */
437 sb
->s_xattr
= orangefs_xattr_handlers
;
438 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
439 sb
->s_op
= &orangefs_s_ops
;
440 sb
->s_d_op
= &orangefs_dentry_operations
;
442 sb
->s_blocksize
= orangefs_bufmap_size_query();
443 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
444 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
446 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
447 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
448 gossip_debug(GOSSIP_SUPER_DEBUG
,
449 "get inode %pU, fsid %d\n",
450 &root_object
.khandle
,
453 root
= orangefs_iget(sb
, &root_object
);
455 return PTR_ERR(root
);
457 gossip_debug(GOSSIP_SUPER_DEBUG
,
458 "Allocated root inode [%p] with mode %x\n",
462 /* allocates and places root dentry in dcache */
463 root_dentry
= d_make_root(root
);
467 sb
->s_export_op
= &orangefs_export_ops
;
468 sb
->s_root
= root_dentry
;
472 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
478 struct super_block
*sb
= ERR_PTR(-EINVAL
);
479 struct orangefs_kernel_op_s
*new_op
;
480 struct dentry
*d
= ERR_PTR(-EINVAL
);
482 gossip_debug(GOSSIP_SUPER_DEBUG
,
483 "orangefs_mount: called with devname %s\n",
487 gossip_err("ERROR: device name not specified.\n");
488 return ERR_PTR(-EINVAL
);
491 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
493 return ERR_PTR(-ENOMEM
);
495 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
497 ORANGEFS_MAX_SERVER_ADDR_LEN
);
499 gossip_debug(GOSSIP_SUPER_DEBUG
,
500 "Attempting ORANGEFS Mount via host %s\n",
501 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
503 ret
= service_operation(new_op
, "orangefs_mount", 0);
504 gossip_debug(GOSSIP_SUPER_DEBUG
,
505 "orangefs_mount: mount got return value of %d\n", ret
);
509 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
510 gossip_err("ERROR: Retrieved null fs_id\n");
515 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
519 orangefs_unmount(new_op
->downcall
.resp
.fs_mount
.id
,
520 new_op
->downcall
.resp
.fs_mount
.fs_id
, devname
);
524 ret
= orangefs_fill_sb(sb
,
525 &new_op
->downcall
.resp
.fs_mount
, data
,
526 flags
& MS_SILENT
? 1 : 0);
534 * on successful mount, store the devname and data
537 strncpy(ORANGEFS_SB(sb
)->devname
,
539 ORANGEFS_MAX_SERVER_ADDR_LEN
);
541 /* mount_pending must be cleared */
542 ORANGEFS_SB(sb
)->mount_pending
= 0;
545 * finally, add this sb to our list of known orangefs
548 gossip_debug(GOSSIP_SUPER_DEBUG
,
549 "Adding SB %p to orangefs superblocks\n",
551 spin_lock(&orangefs_superblocks_lock
);
552 list_add_tail(&ORANGEFS_SB(sb
)->list
, &orangefs_superblocks
);
553 spin_unlock(&orangefs_superblocks_lock
);
556 /* Must be removed from the list now. */
557 ORANGEFS_SB(sb
)->no_list
= 0;
559 if (orangefs_userspace_version
>= 20906) {
560 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
562 return ERR_PTR(-ENOMEM
);
563 new_op
->upcall
.req
.features
.features
= 0;
564 ret
= service_operation(new_op
, "orangefs_features", 0);
565 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
568 orangefs_features
= 0;
571 return dget(sb
->s_root
);
574 /* Will call orangefs_kill_sb with sb not in list. */
575 ORANGEFS_SB(sb
)->no_list
= 1;
576 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
577 deactivate_locked_super(sb
);
579 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
580 if (ret
== -EINVAL
) {
581 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
582 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
590 void orangefs_kill_sb(struct super_block
*sb
)
593 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
595 /* provided sb cleanup */
599 * issue the unmount to userspace to tell it to remove the
600 * dynamic mount info it has for this superblock
602 r
= orangefs_unmount(ORANGEFS_SB(sb
)->id
, ORANGEFS_SB(sb
)->fs_id
,
603 ORANGEFS_SB(sb
)->devname
);
605 ORANGEFS_SB(sb
)->mount_pending
= 1;
607 if (!ORANGEFS_SB(sb
)->no_list
) {
608 /* remove the sb from our list of orangefs specific sb's */
609 spin_lock(&orangefs_superblocks_lock
);
610 /* not list_del_init */
611 __list_del_entry(&ORANGEFS_SB(sb
)->list
);
612 ORANGEFS_SB(sb
)->list
.prev
= NULL
;
613 spin_unlock(&orangefs_superblocks_lock
);
617 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
618 * gets completed before we free the dang thing.
620 mutex_lock(&orangefs_request_mutex
);
621 mutex_unlock(&orangefs_request_mutex
);
623 /* free the orangefs superblock private data */
624 kfree(ORANGEFS_SB(sb
));
627 int orangefs_inode_cache_initialize(void)
629 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
630 sizeof(struct orangefs_inode_s
),
632 ORANGEFS_CACHE_CREATE_FLAGS
,
633 orangefs_inode_cache_ctor
);
635 if (!orangefs_inode_cache
) {
636 gossip_err("Cannot create orangefs_inode_cache\n");
642 int orangefs_inode_cache_finalize(void)
644 kmem_cache_destroy(orangefs_inode_cache
);