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_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_MAX
;
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 orangefs_sb_info_s
*orangefs_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
->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 acquired 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_MUTEX
);
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
->id
= new_op
->downcall
.resp
.fs_mount
.id
;
249 orangefs_sb
->mount_pending
= 0;
254 if (orangefs_userspace_version
>= 20906) {
255 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
258 new_op
->upcall
.req
.features
.features
= 0;
259 ret
= service_operation(new_op
, "orangefs_features", 0);
260 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
263 orangefs_features
= 0;
269 int fsid_key_table_initialize(void)
274 void fsid_key_table_finalize(void)
278 /* Called whenever the VFS dirties the inode in response to atime updates */
279 static void orangefs_dirty_inode(struct inode
*inode
, int flags
)
281 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
283 gossip_debug(GOSSIP_SUPER_DEBUG
,
284 "orangefs_dirty_inode: %pU\n",
285 get_khandle_from_ino(inode
));
286 SetAtimeFlag(orangefs_inode
);
289 static const struct super_operations orangefs_s_ops
= {
290 .alloc_inode
= orangefs_alloc_inode
,
291 .destroy_inode
= orangefs_destroy_inode
,
292 .dirty_inode
= orangefs_dirty_inode
,
293 .drop_inode
= generic_delete_inode
,
294 .statfs
= orangefs_statfs
,
295 .remount_fs
= orangefs_remount_fs
,
296 .show_options
= generic_show_options
,
299 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
304 struct orangefs_object_kref refn
;
306 if (fh_len
< 5 || fh_type
> 2)
309 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
310 refn
.fs_id
= (u32
) fid
->raw
[4];
311 gossip_debug(GOSSIP_SUPER_DEBUG
,
312 "fh_to_dentry: handle %pU, fs_id %d\n",
316 return d_obtain_alias(orangefs_iget(sb
, &refn
));
319 static int orangefs_encode_fh(struct inode
*inode
,
322 struct inode
*parent
)
324 int len
= parent
? 10 : 5;
326 struct orangefs_object_kref refn
;
328 if (*max_len
< len
) {
329 gossip_lerr("fh buffer is too small for encoding\n");
335 refn
= ORANGEFS_I(inode
)->refn
;
336 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
339 gossip_debug(GOSSIP_SUPER_DEBUG
,
340 "Encoding fh: handle %pU, fsid %u\n",
346 refn
= ORANGEFS_I(parent
)->refn
;
347 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
351 gossip_debug(GOSSIP_SUPER_DEBUG
,
352 "Encoding parent: handle %pU, fsid %u\n",
362 static const struct export_operations orangefs_export_ops
= {
363 .encode_fh
= orangefs_encode_fh
,
364 .fh_to_dentry
= orangefs_fh_to_dentry
,
367 static int orangefs_fill_sb(struct super_block
*sb
,
368 struct orangefs_fs_mount_response
*fs_mount
,
369 void *data
, int silent
)
372 struct inode
*root
= NULL
;
373 struct dentry
*root_dentry
= NULL
;
374 struct orangefs_object_kref root_object
;
376 /* alloc and init our private orangefs sb info */
377 sb
->s_fs_info
= kzalloc(sizeof(struct orangefs_sb_info_s
), GFP_KERNEL
);
378 if (!ORANGEFS_SB(sb
))
380 ORANGEFS_SB(sb
)->sb
= sb
;
382 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
383 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
384 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
387 ret
= parse_mount_options(sb
, data
, silent
);
392 /* Hang the xattr handlers off the superblock */
393 sb
->s_xattr
= orangefs_xattr_handlers
;
394 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
395 sb
->s_op
= &orangefs_s_ops
;
396 sb
->s_d_op
= &orangefs_dentry_operations
;
398 sb
->s_blocksize
= orangefs_bufmap_size_query();
399 sb
->s_blocksize_bits
= orangefs_bufmap_shift_query();
400 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
402 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
403 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
404 gossip_debug(GOSSIP_SUPER_DEBUG
,
405 "get inode %pU, fsid %d\n",
406 &root_object
.khandle
,
409 root
= orangefs_iget(sb
, &root_object
);
411 return PTR_ERR(root
);
413 gossip_debug(GOSSIP_SUPER_DEBUG
,
414 "Allocated root inode [%p] with mode %x\n",
418 /* allocates and places root dentry in dcache */
419 root_dentry
= d_make_root(root
);
423 sb
->s_export_op
= &orangefs_export_ops
;
424 sb
->s_root
= root_dentry
;
428 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
434 struct super_block
*sb
= ERR_PTR(-EINVAL
);
435 struct orangefs_kernel_op_s
*new_op
;
436 struct dentry
*d
= ERR_PTR(-EINVAL
);
438 gossip_debug(GOSSIP_SUPER_DEBUG
,
439 "orangefs_mount: called with devname %s\n",
443 gossip_err("ERROR: device name not specified.\n");
444 return ERR_PTR(-EINVAL
);
447 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
449 return ERR_PTR(-ENOMEM
);
451 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
453 ORANGEFS_MAX_SERVER_ADDR_LEN
);
455 gossip_debug(GOSSIP_SUPER_DEBUG
,
456 "Attempting ORANGEFS Mount via host %s\n",
457 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
459 ret
= service_operation(new_op
, "orangefs_mount", 0);
460 gossip_debug(GOSSIP_SUPER_DEBUG
,
461 "orangefs_mount: mount got return value of %d\n", ret
);
465 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
466 gossip_err("ERROR: Retrieved null fs_id\n");
471 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
478 ret
= orangefs_fill_sb(sb
,
479 &new_op
->downcall
.resp
.fs_mount
, data
,
480 flags
& MS_SILENT
? 1 : 0);
488 * on successful mount, store the devname and data
491 strncpy(ORANGEFS_SB(sb
)->devname
,
493 ORANGEFS_MAX_SERVER_ADDR_LEN
);
495 /* mount_pending must be cleared */
496 ORANGEFS_SB(sb
)->mount_pending
= 0;
499 * finally, add this sb to our list of known orangefs
502 gossip_debug(GOSSIP_SUPER_DEBUG
,
503 "Adding SB %p to orangefs superblocks\n",
505 spin_lock(&orangefs_superblocks_lock
);
506 list_add_tail(&ORANGEFS_SB(sb
)->list
, &orangefs_superblocks
);
507 spin_unlock(&orangefs_superblocks_lock
);
510 if (orangefs_userspace_version
>= 20906) {
511 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
513 return ERR_PTR(-ENOMEM
);
514 new_op
->upcall
.req
.features
.features
= 0;
515 ret
= service_operation(new_op
, "orangefs_features", 0);
516 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
519 orangefs_features
= 0;
522 return dget(sb
->s_root
);
525 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
526 if (ret
== -EINVAL
) {
527 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
528 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
536 void orangefs_kill_sb(struct super_block
*sb
)
538 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
540 /* provided sb cleanup */
544 * issue the unmount to userspace to tell it to remove the
545 * dynamic mount info it has for this superblock
547 orangefs_unmount_sb(sb
);
549 /* remove the sb from our list of orangefs specific sb's */
551 spin_lock(&orangefs_superblocks_lock
);
552 __list_del_entry(&ORANGEFS_SB(sb
)->list
); /* not list_del_init */
553 ORANGEFS_SB(sb
)->list
.prev
= NULL
;
554 spin_unlock(&orangefs_superblocks_lock
);
557 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
558 * gets completed before we free the dang thing.
560 mutex_lock(&orangefs_request_mutex
);
561 mutex_unlock(&orangefs_request_mutex
);
563 /* free the orangefs superblock private data */
564 kfree(ORANGEFS_SB(sb
));
567 int orangefs_inode_cache_initialize(void)
569 orangefs_inode_cache
= kmem_cache_create("orangefs_inode_cache",
570 sizeof(struct orangefs_inode_s
),
572 ORANGEFS_CACHE_CREATE_FLAGS
,
573 orangefs_inode_cache_ctor
);
575 if (!orangefs_inode_cache
) {
576 gossip_err("Cannot create orangefs_inode_cache\n");
582 int orangefs_inode_cache_finalize(void)
584 kmem_cache_destroy(orangefs_inode_cache
);