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" },
37 static int parse_mount_options(struct super_block
*sb
, char *options
,
40 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(sb
);
41 substring_t args
[MAX_OPT_ARGS
];
45 * Force any potential flags that might be set from the mount
46 * to zero, ie, initialize to unset.
48 sb
->s_flags
&= ~MS_POSIXACL
;
49 orangefs_sb
->flags
&= ~ORANGEFS_OPT_INTR
;
50 orangefs_sb
->flags
&= ~ORANGEFS_OPT_LOCAL_LOCK
;
52 while ((p
= strsep(&options
, ",")) != NULL
) {
58 token
= match_token(p
, tokens
, args
);
61 sb
->s_flags
|= MS_POSIXACL
;
64 orangefs_sb
->flags
|= ORANGEFS_OPT_INTR
;
67 orangefs_sb
->flags
|= ORANGEFS_OPT_LOCAL_LOCK
;
77 gossip_err("Error: mount option [%s] is not supported.\n", p
);
81 static void orangefs_inode_cache_ctor(void *req
)
83 struct orangefs_inode_s
*orangefs_inode
= req
;
85 inode_init_once(&orangefs_inode
->vfs_inode
);
86 init_rwsem(&orangefs_inode
->xattr_sem
);
88 orangefs_inode
->vfs_inode
.i_version
= 1;
91 static struct inode
*orangefs_alloc_inode(struct super_block
*sb
)
93 struct orangefs_inode_s
*orangefs_inode
;
95 orangefs_inode
= kmem_cache_alloc(orangefs_inode_cache
,
96 ORANGEFS_CACHE_ALLOC_FLAGS
);
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_destroy_inode(struct inode
*inode
)
120 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
122 gossip_debug(GOSSIP_SUPER_DEBUG
,
123 "%s: deallocated %p destroying inode %pU\n",
124 __func__
, orangefs_inode
, get_khandle_from_ino(inode
));
126 kmem_cache_free(orangefs_inode_cache
, orangefs_inode
);
130 * NOTE: information filled in here is typically reflected in the
131 * output of the system command 'df'
133 static int orangefs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
136 struct orangefs_kernel_op_s
*new_op
= NULL
;
138 struct super_block
*sb
= NULL
;
142 gossip_debug(GOSSIP_SUPER_DEBUG
,
143 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
145 (int)(ORANGEFS_SB(sb
)->fs_id
));
147 new_op
= op_alloc(ORANGEFS_VFS_OP_STATFS
);
150 new_op
->upcall
.req
.statfs
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
152 if (ORANGEFS_SB(sb
)->flags
& ORANGEFS_OPT_INTR
)
153 flags
= ORANGEFS_OP_INTERRUPTIBLE
;
155 ret
= service_operation(new_op
, "orangefs_statfs", flags
);
157 if (new_op
->downcall
.status
< 0)
160 gossip_debug(GOSSIP_SUPER_DEBUG
,
161 "%s: got %ld blocks available | "
162 "%ld blocks total | %ld block size | "
163 "%ld files total | %ld files avail\n",
165 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
166 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
167 (long)new_op
->downcall
.resp
.statfs
.block_size
,
168 (long)new_op
->downcall
.resp
.statfs
.files_total
,
169 (long)new_op
->downcall
.resp
.statfs
.files_avail
);
171 buf
->f_type
= sb
->s_magic
;
172 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
173 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
174 buf
->f_namelen
= ORANGEFS_NAME_LEN
;
176 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
177 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
178 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
179 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
180 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
181 buf
->f_frsize
= sb
->s_blocksize
;
185 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_statfs: returning %d\n", ret
);
190 * Remount as initiated by VFS layer. We just need to reparse the mount
191 * options, no need to signal pvfs2-client-core about it.
193 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
195 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
196 return parse_mount_options(sb
, data
, 1);
200 * Remount as initiated by pvfs2-client-core on restart. This is used to
201 * repopulate mount information left from previous pvfs2-client-core.
203 * the idea here is that given a valid superblock, we're
204 * re-initializing the user space client with the initial mount
205 * information specified when the super block was first initialized.
206 * this is very different than the first initialization/creation of a
207 * superblock. we use the special service_priority_operation to make
208 * sure that the mount gets ahead of any other pending operation that
209 * is waiting for servicing. this means that the pvfs2-client won't
210 * fail to start several times for all other pending operations before
211 * the client regains all of the mount information from us.
212 * NOTE: this function assumes that the request_mutex is already acquired!
214 int orangefs_remount(struct super_block
*sb
)
216 struct orangefs_kernel_op_s
*new_op
;
219 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
221 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
224 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
225 ORANGEFS_SB(sb
)->devname
,
226 ORANGEFS_MAX_SERVER_ADDR_LEN
);
228 gossip_debug(GOSSIP_SUPER_DEBUG
,
229 "Attempting ORANGEFS Remount via host %s\n",
230 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
233 * we assume that the calling function has already acquire the
234 * request_mutex to prevent other operations from bypassing
237 ret
= service_operation(new_op
, "orangefs_remount",
238 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_SEMAPHORE
);
239 gossip_debug(GOSSIP_SUPER_DEBUG
,
240 "orangefs_remount: mount got return value of %d\n",
244 * store the id assigned to this sb -- it's just a
245 * short-lived mapping that the system interface uses
246 * to map this superblock to a particular mount entry
248 ORANGEFS_SB(sb
)->id
= new_op
->downcall
.resp
.fs_mount
.id
;
249 ORANGEFS_SB(sb
)->mount_pending
= 0;
256 int fsid_key_table_initialize(void)
261 void fsid_key_table_finalize(void)
265 /* Called whenever the VFS dirties the inode in response to atime updates */
266 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
268 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
270 gossip_debug(GOSSIP_SUPER_DEBUG
,
271 "orangefs_dirty_inode: %pU\n",
272 get_khandle_from_ino(inode
));
273 SetAtimeFlag(orangefs_inode
);
276 static const struct super_operations orangefs_s_ops
= {
277 .alloc_inode
= orangefs_alloc_inode
,
278 .destroy_inode
= orangefs_destroy_inode
,
279 .dirty_inode
= orangefs_dirty_inode
,
280 .drop_inode
= generic_delete_inode
,
281 .statfs
= orangefs_statfs
,
282 .remount_fs
= orangefs_remount_fs
,
283 .show_options
= generic_show_options
,
286 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
291 struct orangefs_object_kref refn
;
293 if (fh_len
< 5 || fh_type
> 2)
296 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
297 refn
.fs_id
= (u32
) fid
->raw
[4];
298 gossip_debug(GOSSIP_SUPER_DEBUG
,
299 "fh_to_dentry: handle %pU, fs_id %d\n",
303 return d_obtain_alias(orangefs_iget(sb
, &refn
));
306 static int orangefs_encode_fh(struct inode
*inode
,
309 struct inode
*parent
)
311 int len
= parent
? 10 : 5;
313 struct orangefs_object_kref refn
;
315 if (*max_len
< len
) {
316 gossip_lerr("fh buffer is too small for encoding\n");
322 refn
= ORANGEFS_I(inode
)->refn
;
323 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
326 gossip_debug(GOSSIP_SUPER_DEBUG
,
327 "Encoding fh: handle %pU, fsid %u\n",
333 refn
= ORANGEFS_I(parent
)->refn
;
334 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
338 gossip_debug(GOSSIP_SUPER_DEBUG
,
339 "Encoding parent: handle %pU, fsid %u\n",
349 static const struct export_operations orangefs_export_ops
= {
350 .encode_fh
= orangefs_encode_fh
,
351 .fh_to_dentry
= orangefs_fh_to_dentry
,
354 static int orangefs_fill_sb(struct super_block
*sb
,
355 struct orangefs_fs_mount_response
*fs_mount
,
356 void *data
, int silent
)
359 struct inode
*root
= NULL
;
360 struct dentry
*root_dentry
= NULL
;
361 struct orangefs_object_kref root_object
;
363 /* alloc and init our private orangefs sb info */
365 kzalloc(sizeof(struct orangefs_sb_info_s
), ORANGEFS_GFP_FLAGS
);
366 if (!ORANGEFS_SB(sb
))
368 ORANGEFS_SB(sb
)->sb
= sb
;
370 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
371 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
372 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
375 ret
= parse_mount_options(sb
, data
, silent
);
380 /* Hang the xattr handlers off the superblock */
381 sb
->s_xattr
= orangefs_xattr_handlers
;
382 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
383 sb
->s_op
= &orangefs_s_ops
;
384 sb
->s_d_op
= &orangefs_dentry_operations
;
386 sb
->s_blocksize
= orangefs_bufmap_size_query();
387 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
388 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
390 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
391 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
392 gossip_debug(GOSSIP_SUPER_DEBUG
,
393 "get inode %pU, fsid %d\n",
394 &root_object
.khandle
,
397 root
= orangefs_iget(sb
, &root_object
);
399 return PTR_ERR(root
);
401 gossip_debug(GOSSIP_SUPER_DEBUG
,
402 "Allocated root inode [%p] with mode %x\n",
406 /* allocates and places root dentry in dcache */
407 root_dentry
= d_make_root(root
);
411 sb
->s_export_op
= &orangefs_export_ops
;
412 sb
->s_root
= root_dentry
;
416 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
422 struct super_block
*sb
= ERR_PTR(-EINVAL
);
423 struct orangefs_kernel_op_s
*new_op
;
424 struct dentry
*d
= ERR_PTR(-EINVAL
);
426 gossip_debug(GOSSIP_SUPER_DEBUG
,
427 "orangefs_mount: called with devname %s\n",
431 gossip_err("ERROR: device name not specified.\n");
432 return ERR_PTR(-EINVAL
);
435 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
437 return ERR_PTR(-ENOMEM
);
439 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
441 ORANGEFS_MAX_SERVER_ADDR_LEN
);
443 gossip_debug(GOSSIP_SUPER_DEBUG
,
444 "Attempting ORANGEFS Mount via host %s\n",
445 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
447 ret
= service_operation(new_op
, "orangefs_mount", 0);
448 gossip_debug(GOSSIP_SUPER_DEBUG
,
449 "orangefs_mount: mount got return value of %d\n", ret
);
453 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
454 gossip_err("ERROR: Retrieved null fs_id\n");
459 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
466 ret
= orangefs_fill_sb(sb
,
467 &new_op
->downcall
.resp
.fs_mount
, data
,
468 flags
& MS_SILENT
? 1 : 0);
476 * on successful mount, store the devname and data
479 strncpy(ORANGEFS_SB(sb
)->devname
,
481 ORANGEFS_MAX_SERVER_ADDR_LEN
);
483 /* mount_pending must be cleared */
484 ORANGEFS_SB(sb
)->mount_pending
= 0;
487 * finally, add this sb to our list of known orangefs
492 return dget(sb
->s_root
);
495 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
496 if (ret
== -EINVAL
) {
497 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
498 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
506 void orangefs_kill_sb(struct super_block
*sb
)
508 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
511 * issue the unmount to userspace to tell it to remove the
512 * dynamic mount info it has for this superblock
514 orangefs_unmount_sb(sb
);
516 /* remove the sb from our list of orangefs specific sb's */
517 remove_orangefs_sb(sb
);
519 /* provided sb cleanup */
522 /* free the orangefs superblock private data */
523 kfree(ORANGEFS_SB(sb
));
526 int orangefs_inode_cache_initialize(void)
528 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
529 sizeof(struct orangefs_inode_s
),
531 ORANGEFS_CACHE_CREATE_FLAGS
,
532 orangefs_inode_cache_ctor
);
534 if (!orangefs_inode_cache
) {
535 gossip_err("Cannot create orangefs_inode_cache\n");
541 int orangefs_inode_cache_finalize(void)
543 kmem_cache_destroy(orangefs_inode_cache
);