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
);
114 * We want to clear everything except for rw_semaphore and the
117 memset(&orangefs_inode
->refn
.khandle
, 0, 16);
118 orangefs_inode
->refn
.fs_id
= ORANGEFS_FS_ID_NULL
;
119 orangefs_inode
->last_failed_block_index_read
= 0;
120 memset(orangefs_inode
->link_target
, 0, sizeof(orangefs_inode
->link_target
));
121 orangefs_inode
->pinode_flags
= 0;
123 gossip_debug(GOSSIP_SUPER_DEBUG
,
124 "orangefs_alloc_inode: allocated %p\n",
125 &orangefs_inode
->vfs_inode
);
126 return &orangefs_inode
->vfs_inode
;
129 static void orangefs_i_callback(struct rcu_head
*head
)
131 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
132 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
133 kmem_cache_free(orangefs_inode_cache
, orangefs_inode
);
136 static void orangefs_destroy_inode(struct inode
*inode
)
138 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
140 gossip_debug(GOSSIP_SUPER_DEBUG
,
141 "%s: deallocated %p destroying inode %pU\n",
142 __func__
, orangefs_inode
, get_khandle_from_ino(inode
));
144 call_rcu(&inode
->i_rcu
, orangefs_i_callback
);
148 * NOTE: information filled in here is typically reflected in the
149 * output of the system command 'df'
151 static int orangefs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
154 struct orangefs_kernel_op_s
*new_op
= NULL
;
156 struct super_block
*sb
= NULL
;
160 gossip_debug(GOSSIP_SUPER_DEBUG
,
161 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
163 (int)(ORANGEFS_SB(sb
)->fs_id
));
165 new_op
= op_alloc(ORANGEFS_VFS_OP_STATFS
);
168 new_op
->upcall
.req
.statfs
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
170 if (ORANGEFS_SB(sb
)->flags
& ORANGEFS_OPT_INTR
)
171 flags
= ORANGEFS_OP_INTERRUPTIBLE
;
173 ret
= service_operation(new_op
, "orangefs_statfs", flags
);
175 if (new_op
->downcall
.status
< 0)
178 gossip_debug(GOSSIP_SUPER_DEBUG
,
179 "%s: got %ld blocks available | "
180 "%ld blocks total | %ld block size | "
181 "%ld files total | %ld files avail\n",
183 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
184 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
185 (long)new_op
->downcall
.resp
.statfs
.block_size
,
186 (long)new_op
->downcall
.resp
.statfs
.files_total
,
187 (long)new_op
->downcall
.resp
.statfs
.files_avail
);
189 buf
->f_type
= sb
->s_magic
;
190 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
191 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
192 buf
->f_namelen
= ORANGEFS_NAME_MAX
;
194 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
195 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
196 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
197 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
198 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
199 buf
->f_frsize
= sb
->s_blocksize
;
203 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_statfs: returning %d\n", ret
);
208 * Remount as initiated by VFS layer. We just need to reparse the mount
209 * options, no need to signal pvfs2-client-core about it.
211 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
213 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
214 return parse_mount_options(sb
, data
, 1);
218 * Remount as initiated by pvfs2-client-core on restart. This is used to
219 * repopulate mount information left from previous pvfs2-client-core.
221 * the idea here is that given a valid superblock, we're
222 * re-initializing the user space client with the initial mount
223 * information specified when the super block was first initialized.
224 * this is very different than the first initialization/creation of a
225 * superblock. we use the special service_priority_operation to make
226 * sure that the mount gets ahead of any other pending operation that
227 * is waiting for servicing. this means that the pvfs2-client won't
228 * fail to start several times for all other pending operations before
229 * the client regains all of the mount information from us.
230 * NOTE: this function assumes that the request_mutex is already acquired!
232 int orangefs_remount(struct orangefs_sb_info_s
*orangefs_sb
)
234 struct orangefs_kernel_op_s
*new_op
;
237 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
239 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
242 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
243 orangefs_sb
->devname
,
244 ORANGEFS_MAX_SERVER_ADDR_LEN
);
246 gossip_debug(GOSSIP_SUPER_DEBUG
,
247 "Attempting ORANGEFS Remount via host %s\n",
248 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
251 * we assume that the calling function has already acquired the
252 * request_mutex to prevent other operations from bypassing
255 ret
= service_operation(new_op
, "orangefs_remount",
256 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
257 gossip_debug(GOSSIP_SUPER_DEBUG
,
258 "orangefs_remount: mount got return value of %d\n",
262 * store the id assigned to this sb -- it's just a
263 * short-lived mapping that the system interface uses
264 * to map this superblock to a particular mount entry
266 orangefs_sb
->id
= new_op
->downcall
.resp
.fs_mount
.id
;
267 orangefs_sb
->mount_pending
= 0;
272 if (orangefs_userspace_version
>= 20906) {
273 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
276 new_op
->upcall
.req
.features
.features
= 0;
277 ret
= service_operation(new_op
, "orangefs_features",
278 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
281 new_op
->downcall
.resp
.features
.features
;
283 orangefs_features
= 0;
286 orangefs_features
= 0;
292 int fsid_key_table_initialize(void)
297 void fsid_key_table_finalize(void)
301 /* Called whenever the VFS dirties the inode in response to atime updates */
302 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
304 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
306 gossip_debug(GOSSIP_SUPER_DEBUG
,
307 "orangefs_dirty_inode: %pU\n",
308 get_khandle_from_ino(inode
));
309 SetAtimeFlag(orangefs_inode
);
312 static const struct super_operations orangefs_s_ops
= {
313 .alloc_inode
= orangefs_alloc_inode
,
314 .destroy_inode
= orangefs_destroy_inode
,
315 .dirty_inode
= orangefs_dirty_inode
,
316 .drop_inode
= generic_delete_inode
,
317 .statfs
= orangefs_statfs
,
318 .remount_fs
= orangefs_remount_fs
,
319 .show_options
= orangefs_show_options
,
322 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
327 struct orangefs_object_kref refn
;
329 if (fh_len
< 5 || fh_type
> 2)
332 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
333 refn
.fs_id
= (u32
) fid
->raw
[4];
334 gossip_debug(GOSSIP_SUPER_DEBUG
,
335 "fh_to_dentry: handle %pU, fs_id %d\n",
339 return d_obtain_alias(orangefs_iget(sb
, &refn
));
342 static int orangefs_encode_fh(struct inode
*inode
,
345 struct inode
*parent
)
347 int len
= parent
? 10 : 5;
349 struct orangefs_object_kref refn
;
351 if (*max_len
< len
) {
352 gossip_lerr("fh buffer is too small for encoding\n");
358 refn
= ORANGEFS_I(inode
)->refn
;
359 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
362 gossip_debug(GOSSIP_SUPER_DEBUG
,
363 "Encoding fh: handle %pU, fsid %u\n",
369 refn
= ORANGEFS_I(parent
)->refn
;
370 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
374 gossip_debug(GOSSIP_SUPER_DEBUG
,
375 "Encoding parent: handle %pU, fsid %u\n",
385 static const struct export_operations orangefs_export_ops
= {
386 .encode_fh
= orangefs_encode_fh
,
387 .fh_to_dentry
= orangefs_fh_to_dentry
,
390 static int orangefs_unmount(int id
, __s32 fs_id
, const char *devname
)
392 struct orangefs_kernel_op_s
*op
;
394 op
= op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT
);
397 op
->upcall
.req
.fs_umount
.id
= id
;
398 op
->upcall
.req
.fs_umount
.fs_id
= fs_id
;
399 strncpy(op
->upcall
.req
.fs_umount
.orangefs_config_server
,
400 devname
, ORANGEFS_MAX_SERVER_ADDR_LEN
);
401 r
= service_operation(op
, "orangefs_fs_umount", 0);
402 /* Not much to do about an error here. */
404 gossip_err("orangefs_unmount: service_operation %d\n", r
);
409 static int orangefs_fill_sb(struct super_block
*sb
,
410 struct orangefs_fs_mount_response
*fs_mount
,
411 void *data
, int silent
)
414 struct inode
*root
= NULL
;
415 struct dentry
*root_dentry
= NULL
;
416 struct orangefs_object_kref root_object
;
418 /* alloc and init our private orangefs sb info */
419 sb
->s_fs_info
= kzalloc(sizeof(struct orangefs_sb_info_s
), GFP_KERNEL
);
420 if (!ORANGEFS_SB(sb
))
422 ORANGEFS_SB(sb
)->sb
= sb
;
424 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
425 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
426 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
429 ret
= parse_mount_options(sb
, data
, silent
);
434 /* Hang the xattr handlers off the superblock */
435 sb
->s_xattr
= orangefs_xattr_handlers
;
436 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
437 sb
->s_op
= &orangefs_s_ops
;
438 sb
->s_d_op
= &orangefs_dentry_operations
;
440 sb
->s_blocksize
= orangefs_bufmap_size_query();
441 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
442 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
444 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
445 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
446 gossip_debug(GOSSIP_SUPER_DEBUG
,
447 "get inode %pU, fsid %d\n",
448 &root_object
.khandle
,
451 root
= orangefs_iget(sb
, &root_object
);
453 return PTR_ERR(root
);
455 gossip_debug(GOSSIP_SUPER_DEBUG
,
456 "Allocated root inode [%p] with mode %x\n",
460 /* allocates and places root dentry in dcache */
461 root_dentry
= d_make_root(root
);
465 sb
->s_export_op
= &orangefs_export_ops
;
466 sb
->s_root
= root_dentry
;
470 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
476 struct super_block
*sb
= ERR_PTR(-EINVAL
);
477 struct orangefs_kernel_op_s
*new_op
;
478 struct dentry
*d
= ERR_PTR(-EINVAL
);
480 gossip_debug(GOSSIP_SUPER_DEBUG
,
481 "orangefs_mount: called with devname %s\n",
485 gossip_err("ERROR: device name not specified.\n");
486 return ERR_PTR(-EINVAL
);
489 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
491 return ERR_PTR(-ENOMEM
);
493 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
495 ORANGEFS_MAX_SERVER_ADDR_LEN
);
497 gossip_debug(GOSSIP_SUPER_DEBUG
,
498 "Attempting ORANGEFS Mount via host %s\n",
499 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
501 ret
= service_operation(new_op
, "orangefs_mount", 0);
502 gossip_debug(GOSSIP_SUPER_DEBUG
,
503 "orangefs_mount: mount got return value of %d\n", ret
);
507 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
508 gossip_err("ERROR: Retrieved null fs_id\n");
513 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
517 orangefs_unmount(new_op
->downcall
.resp
.fs_mount
.id
,
518 new_op
->downcall
.resp
.fs_mount
.fs_id
, devname
);
522 ret
= orangefs_fill_sb(sb
,
523 &new_op
->downcall
.resp
.fs_mount
, data
,
524 flags
& MS_SILENT
? 1 : 0);
532 * on successful mount, store the devname and data
535 strncpy(ORANGEFS_SB(sb
)->devname
,
537 ORANGEFS_MAX_SERVER_ADDR_LEN
);
539 /* mount_pending must be cleared */
540 ORANGEFS_SB(sb
)->mount_pending
= 0;
543 * finally, add this sb to our list of known orangefs
546 gossip_debug(GOSSIP_SUPER_DEBUG
,
547 "Adding SB %p to orangefs superblocks\n",
549 spin_lock(&orangefs_superblocks_lock
);
550 list_add_tail(&ORANGEFS_SB(sb
)->list
, &orangefs_superblocks
);
551 spin_unlock(&orangefs_superblocks_lock
);
554 /* Must be removed from the list now. */
555 ORANGEFS_SB(sb
)->no_list
= 0;
557 if (orangefs_userspace_version
>= 20906) {
558 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
560 return ERR_PTR(-ENOMEM
);
561 new_op
->upcall
.req
.features
.features
= 0;
562 ret
= service_operation(new_op
, "orangefs_features", 0);
563 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
566 orangefs_features
= 0;
569 return dget(sb
->s_root
);
572 /* Will call orangefs_kill_sb with sb not in list. */
573 ORANGEFS_SB(sb
)->no_list
= 1;
574 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
575 deactivate_locked_super(sb
);
577 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
578 if (ret
== -EINVAL
) {
579 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
580 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
588 void orangefs_kill_sb(struct super_block
*sb
)
591 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
593 /* provided sb cleanup */
597 * issue the unmount to userspace to tell it to remove the
598 * dynamic mount info it has for this superblock
600 r
= orangefs_unmount(ORANGEFS_SB(sb
)->id
, ORANGEFS_SB(sb
)->fs_id
,
601 ORANGEFS_SB(sb
)->devname
);
603 ORANGEFS_SB(sb
)->mount_pending
= 1;
605 if (!ORANGEFS_SB(sb
)->no_list
) {
606 /* remove the sb from our list of orangefs specific sb's */
607 spin_lock(&orangefs_superblocks_lock
);
608 /* not list_del_init */
609 __list_del_entry(&ORANGEFS_SB(sb
)->list
);
610 ORANGEFS_SB(sb
)->list
.prev
= NULL
;
611 spin_unlock(&orangefs_superblocks_lock
);
615 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
616 * gets completed before we free the dang thing.
618 mutex_lock(&orangefs_request_mutex
);
619 mutex_unlock(&orangefs_request_mutex
);
621 /* free the orangefs superblock private data */
622 kfree(ORANGEFS_SB(sb
));
625 int orangefs_inode_cache_initialize(void)
627 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
628 sizeof(struct orangefs_inode_s
),
630 ORANGEFS_CACHE_CREATE_FLAGS
,
631 orangefs_inode_cache_ctor
);
633 if (!orangefs_inode_cache
) {
634 gossip_err("Cannot create orangefs_inode_cache\n");
640 int orangefs_inode_cache_finalize(void)
642 kmem_cache_destroy(orangefs_inode_cache
);