2 * (C) 2001 Clemson University and The University of Chicago
4 * See COPYING in top-level directory.
8 #include "pvfs2-kernel.h"
9 #include "pvfs2-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 "orangefs_statfs: got %ld blocks available | "
162 "%ld blocks total | %ld block size\n",
163 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
164 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
165 (long)new_op
->downcall
.resp
.statfs
.block_size
);
167 buf
->f_type
= sb
->s_magic
;
168 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
169 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
170 buf
->f_namelen
= ORANGEFS_NAME_LEN
;
172 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
173 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
174 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
175 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
176 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
177 buf
->f_frsize
= sb
->s_blocksize
;
181 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_statfs: returning %d\n", ret
);
186 * Remount as initiated by VFS layer. We just need to reparse the mount
187 * options, no need to signal pvfs2-client-core about it.
189 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
191 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
192 return parse_mount_options(sb
, data
, 1);
196 * Remount as initiated by pvfs2-client-core on restart. This is used to
197 * repopulate mount information left from previous pvfs2-client-core.
199 * the idea here is that given a valid superblock, we're
200 * re-initializing the user space client with the initial mount
201 * information specified when the super block was first initialized.
202 * this is very different than the first initialization/creation of a
203 * superblock. we use the special service_priority_operation to make
204 * sure that the mount gets ahead of any other pending operation that
205 * is waiting for servicing. this means that the pvfs2-client won't
206 * fail to start several times for all other pending operations before
207 * the client regains all of the mount information from us.
208 * NOTE: this function assumes that the request_mutex is already acquired!
210 int orangefs_remount(struct super_block
*sb
)
212 struct orangefs_kernel_op_s
*new_op
;
215 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
217 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
220 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
221 ORANGEFS_SB(sb
)->devname
,
222 ORANGEFS_MAX_SERVER_ADDR_LEN
);
224 gossip_debug(GOSSIP_SUPER_DEBUG
,
225 "Attempting ORANGEFS Remount via host %s\n",
226 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
229 * we assume that the calling function has already acquire the
230 * request_mutex to prevent other operations from bypassing
233 ret
= service_operation(new_op
, "orangefs_remount",
234 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_SEMAPHORE
);
235 gossip_debug(GOSSIP_SUPER_DEBUG
,
236 "orangefs_remount: mount got return value of %d\n",
240 * store the id assigned to this sb -- it's just a
241 * short-lived mapping that the system interface uses
242 * to map this superblock to a particular mount entry
244 ORANGEFS_SB(sb
)->id
= new_op
->downcall
.resp
.fs_mount
.id
;
245 ORANGEFS_SB(sb
)->mount_pending
= 0;
252 int fsid_key_table_initialize(void)
257 void fsid_key_table_finalize(void)
261 /* Called whenever the VFS dirties the inode in response to atime updates */
262 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
264 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
266 gossip_debug(GOSSIP_SUPER_DEBUG
,
267 "orangefs_dirty_inode: %pU\n",
268 get_khandle_from_ino(inode
));
269 SetAtimeFlag(orangefs_inode
);
272 static const struct super_operations orangefs_s_ops
= {
273 .alloc_inode
= orangefs_alloc_inode
,
274 .destroy_inode
= orangefs_destroy_inode
,
275 .dirty_inode
= orangefs_dirty_inode
,
276 .drop_inode
= generic_delete_inode
,
277 .statfs
= orangefs_statfs
,
278 .remount_fs
= orangefs_remount_fs
,
279 .show_options
= generic_show_options
,
282 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
287 struct orangefs_object_kref refn
;
289 if (fh_len
< 5 || fh_type
> 2)
292 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
293 refn
.fs_id
= (u32
) fid
->raw
[4];
294 gossip_debug(GOSSIP_SUPER_DEBUG
,
295 "fh_to_dentry: handle %pU, fs_id %d\n",
299 return d_obtain_alias(orangefs_iget(sb
, &refn
));
302 static int orangefs_encode_fh(struct inode
*inode
,
305 struct inode
*parent
)
307 int len
= parent
? 10 : 5;
309 struct orangefs_object_kref refn
;
311 if (*max_len
< len
) {
312 gossip_lerr("fh buffer is too small for encoding\n");
318 refn
= ORANGEFS_I(inode
)->refn
;
319 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
322 gossip_debug(GOSSIP_SUPER_DEBUG
,
323 "Encoding fh: handle %pU, fsid %u\n",
329 refn
= ORANGEFS_I(parent
)->refn
;
330 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
334 gossip_debug(GOSSIP_SUPER_DEBUG
,
335 "Encoding parent: handle %pU, fsid %u\n",
345 static struct export_operations orangefs_export_ops
= {
346 .encode_fh
= orangefs_encode_fh
,
347 .fh_to_dentry
= orangefs_fh_to_dentry
,
350 static int orangefs_fill_sb(struct super_block
*sb
,
351 struct orangefs_fs_mount_response
*fs_mount
,
352 void *data
, int silent
)
355 struct inode
*root
= NULL
;
356 struct dentry
*root_dentry
= NULL
;
357 struct orangefs_object_kref root_object
;
359 /* alloc and init our private orangefs sb info */
361 kzalloc(sizeof(struct orangefs_sb_info_s
), ORANGEFS_GFP_FLAGS
);
362 if (!ORANGEFS_SB(sb
))
364 ORANGEFS_SB(sb
)->sb
= sb
;
366 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
367 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
368 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
371 ret
= parse_mount_options(sb
, data
, silent
);
376 /* Hang the xattr handlers off the superblock */
377 sb
->s_xattr
= orangefs_xattr_handlers
;
378 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
379 sb
->s_op
= &orangefs_s_ops
;
380 sb
->s_d_op
= &orangefs_dentry_operations
;
382 sb
->s_blocksize
= orangefs_bufmap_size_query();
383 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
384 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
386 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
387 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
388 gossip_debug(GOSSIP_SUPER_DEBUG
,
389 "get inode %pU, fsid %d\n",
390 &root_object
.khandle
,
393 root
= orangefs_iget(sb
, &root_object
);
395 return PTR_ERR(root
);
397 gossip_debug(GOSSIP_SUPER_DEBUG
,
398 "Allocated root inode [%p] with mode %x\n",
402 /* allocates and places root dentry in dcache */
403 root_dentry
= d_make_root(root
);
407 sb
->s_export_op
= &orangefs_export_ops
;
408 sb
->s_root
= root_dentry
;
412 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
418 struct super_block
*sb
= ERR_PTR(-EINVAL
);
419 struct orangefs_kernel_op_s
*new_op
;
420 struct dentry
*d
= ERR_PTR(-EINVAL
);
422 gossip_debug(GOSSIP_SUPER_DEBUG
,
423 "orangefs_mount: called with devname %s\n",
427 gossip_err("ERROR: device name not specified.\n");
428 return ERR_PTR(-EINVAL
);
431 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
433 return ERR_PTR(-ENOMEM
);
435 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
437 ORANGEFS_MAX_SERVER_ADDR_LEN
);
439 gossip_debug(GOSSIP_SUPER_DEBUG
,
440 "Attempting ORANGEFS Mount via host %s\n",
441 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
443 ret
= service_operation(new_op
, "orangefs_mount", 0);
444 gossip_debug(GOSSIP_SUPER_DEBUG
,
445 "orangefs_mount: mount got return value of %d\n", ret
);
449 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
450 gossip_err("ERROR: Retrieved null fs_id\n");
455 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
462 ret
= orangefs_fill_sb(sb
,
463 &new_op
->downcall
.resp
.fs_mount
, data
,
464 flags
& MS_SILENT
? 1 : 0);
472 * on successful mount, store the devname and data
475 strncpy(ORANGEFS_SB(sb
)->devname
,
477 ORANGEFS_MAX_SERVER_ADDR_LEN
);
479 /* mount_pending must be cleared */
480 ORANGEFS_SB(sb
)->mount_pending
= 0;
483 * finally, add this sb to our list of known orangefs
488 return dget(sb
->s_root
);
491 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
492 if (ret
== -EINVAL
) {
493 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
494 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
502 void orangefs_kill_sb(struct super_block
*sb
)
504 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
507 * issue the unmount to userspace to tell it to remove the
508 * dynamic mount info it has for this superblock
510 orangefs_unmount_sb(sb
);
512 /* remove the sb from our list of orangefs specific sb's */
513 remove_orangefs_sb(sb
);
515 /* provided sb cleanup */
518 /* free the orangefs superblock private data */
519 kfree(ORANGEFS_SB(sb
));
522 int orangefs_inode_cache_initialize(void)
524 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
525 sizeof(struct orangefs_inode_s
),
527 ORANGEFS_CACHE_CREATE_FLAGS
,
528 orangefs_inode_cache_ctor
);
530 if (!orangefs_inode_cache
) {
531 gossip_err("Cannot create orangefs_inode_cache\n");
537 int orangefs_inode_cache_finalize(void)
539 kmem_cache_destroy(orangefs_inode_cache
);