1 // SPDX-License-Identifier: GPL-2.0
3 * (C) 2001 Clemson University and The University of Chicago
5 * See COPYING in top-level directory.
9 #include "orangefs-kernel.h"
10 #include "orangefs-bufmap.h"
12 #include <linux/parser.h>
13 #include <linux/hashtable.h>
15 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
16 static struct kmem_cache
*orangefs_inode_cache
;
18 /* list for storing orangefs specific superblocks in use */
19 LIST_HEAD(orangefs_superblocks
);
21 DEFINE_SPINLOCK(orangefs_superblocks_lock
);
31 static const match_table_t tokens
= {
34 { Opt_local_lock
, "local_lock" },
38 uint64_t orangefs_features
;
40 static int orangefs_show_options(struct seq_file
*m
, struct dentry
*root
)
42 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(root
->d_sb
);
44 if (root
->d_sb
->s_flags
& SB_POSIXACL
)
46 if (orangefs_sb
->flags
& ORANGEFS_OPT_INTR
)
48 if (orangefs_sb
->flags
& ORANGEFS_OPT_LOCAL_LOCK
)
49 seq_puts(m
, ",local_lock");
53 static int parse_mount_options(struct super_block
*sb
, char *options
,
56 struct orangefs_sb_info_s
*orangefs_sb
= ORANGEFS_SB(sb
);
57 substring_t args
[MAX_OPT_ARGS
];
61 * Force any potential flags that might be set from the mount
62 * to zero, ie, initialize to unset.
64 sb
->s_flags
&= ~SB_POSIXACL
;
65 orangefs_sb
->flags
&= ~ORANGEFS_OPT_INTR
;
66 orangefs_sb
->flags
&= ~ORANGEFS_OPT_LOCAL_LOCK
;
68 while ((p
= strsep(&options
, ",")) != NULL
) {
74 token
= match_token(p
, tokens
, args
);
77 sb
->s_flags
|= SB_POSIXACL
;
80 orangefs_sb
->flags
|= ORANGEFS_OPT_INTR
;
83 orangefs_sb
->flags
|= ORANGEFS_OPT_LOCAL_LOCK
;
93 gossip_err("Error: mount option [%s] is not supported.\n", p
);
97 static void orangefs_inode_cache_ctor(void *req
)
99 struct orangefs_inode_s
*orangefs_inode
= req
;
101 inode_init_once(&orangefs_inode
->vfs_inode
);
102 init_rwsem(&orangefs_inode
->xattr_sem
);
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
));
122 gossip_debug(GOSSIP_SUPER_DEBUG
,
123 "orangefs_alloc_inode: allocated %p\n",
124 &orangefs_inode
->vfs_inode
);
125 return &orangefs_inode
->vfs_inode
;
128 static void orangefs_free_inode(struct inode
*inode
)
130 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
131 struct orangefs_cached_xattr
*cx
;
132 struct hlist_node
*tmp
;
135 hash_for_each_safe(orangefs_inode
->xattr_cache
, i
, tmp
, cx
, node
) {
136 hlist_del(&cx
->node
);
140 kmem_cache_free(orangefs_inode_cache
, orangefs_inode
);
143 static void orangefs_destroy_inode(struct inode
*inode
)
145 struct orangefs_inode_s
*orangefs_inode
= ORANGEFS_I(inode
);
147 gossip_debug(GOSSIP_SUPER_DEBUG
,
148 "%s: deallocated %p destroying inode %pU\n",
149 __func__
, orangefs_inode
, get_khandle_from_ino(inode
));
152 static int orangefs_write_inode(struct inode
*inode
,
153 struct writeback_control
*wbc
)
155 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_write_inode\n");
156 return orangefs_inode_setattr(inode
);
160 * NOTE: information filled in here is typically reflected in the
161 * output of the system command 'df'
163 static int orangefs_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
166 struct orangefs_kernel_op_s
*new_op
= NULL
;
168 struct super_block
*sb
= NULL
;
172 gossip_debug(GOSSIP_SUPER_DEBUG
,
173 "%s: called on sb %p (fs_id is %d)\n",
176 (int)(ORANGEFS_SB(sb
)->fs_id
));
178 new_op
= op_alloc(ORANGEFS_VFS_OP_STATFS
);
181 new_op
->upcall
.req
.statfs
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
183 if (ORANGEFS_SB(sb
)->flags
& ORANGEFS_OPT_INTR
)
184 flags
= ORANGEFS_OP_INTERRUPTIBLE
;
186 ret
= service_operation(new_op
, "orangefs_statfs", flags
);
188 if (new_op
->downcall
.status
< 0)
191 gossip_debug(GOSSIP_SUPER_DEBUG
,
192 "%s: got %ld blocks available | "
193 "%ld blocks total | %ld block size | "
194 "%ld files total | %ld files avail\n",
196 (long)new_op
->downcall
.resp
.statfs
.blocks_avail
,
197 (long)new_op
->downcall
.resp
.statfs
.blocks_total
,
198 (long)new_op
->downcall
.resp
.statfs
.block_size
,
199 (long)new_op
->downcall
.resp
.statfs
.files_total
,
200 (long)new_op
->downcall
.resp
.statfs
.files_avail
);
202 buf
->f_type
= sb
->s_magic
;
203 memcpy(&buf
->f_fsid
, &ORANGEFS_SB(sb
)->fs_id
, sizeof(buf
->f_fsid
));
204 buf
->f_bsize
= new_op
->downcall
.resp
.statfs
.block_size
;
205 buf
->f_namelen
= ORANGEFS_NAME_MAX
;
207 buf
->f_blocks
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_total
;
208 buf
->f_bfree
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
209 buf
->f_bavail
= (sector_t
) new_op
->downcall
.resp
.statfs
.blocks_avail
;
210 buf
->f_files
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_total
;
211 buf
->f_ffree
= (sector_t
) new_op
->downcall
.resp
.statfs
.files_avail
;
212 buf
->f_frsize
= sb
->s_blocksize
;
216 gossip_debug(GOSSIP_SUPER_DEBUG
, "%s: returning %d\n", __func__
, ret
);
221 * Remount as initiated by VFS layer. We just need to reparse the mount
222 * options, no need to signal pvfs2-client-core about it.
224 static int orangefs_remount_fs(struct super_block
*sb
, int *flags
, char *data
)
226 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount_fs: called\n");
227 return parse_mount_options(sb
, data
, 1);
231 * Remount as initiated by pvfs2-client-core on restart. This is used to
232 * repopulate mount information left from previous pvfs2-client-core.
234 * the idea here is that given a valid superblock, we're
235 * re-initializing the user space client with the initial mount
236 * information specified when the super block was first initialized.
237 * this is very different than the first initialization/creation of a
238 * superblock. we use the special service_priority_operation to make
239 * sure that the mount gets ahead of any other pending operation that
240 * is waiting for servicing. this means that the pvfs2-client won't
241 * fail to start several times for all other pending operations before
242 * the client regains all of the mount information from us.
243 * NOTE: this function assumes that the request_mutex is already acquired!
245 int orangefs_remount(struct orangefs_sb_info_s
*orangefs_sb
)
247 struct orangefs_kernel_op_s
*new_op
;
250 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_remount: called\n");
252 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
255 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
256 orangefs_sb
->devname
,
257 ORANGEFS_MAX_SERVER_ADDR_LEN
);
259 gossip_debug(GOSSIP_SUPER_DEBUG
,
260 "Attempting ORANGEFS Remount via host %s\n",
261 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
264 * we assume that the calling function has already acquired the
265 * request_mutex to prevent other operations from bypassing
268 ret
= service_operation(new_op
, "orangefs_remount",
269 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
270 gossip_debug(GOSSIP_SUPER_DEBUG
,
271 "orangefs_remount: mount got return value of %d\n",
275 * store the id assigned to this sb -- it's just a
276 * short-lived mapping that the system interface uses
277 * to map this superblock to a particular mount entry
279 orangefs_sb
->id
= new_op
->downcall
.resp
.fs_mount
.id
;
280 orangefs_sb
->mount_pending
= 0;
285 if (orangefs_userspace_version
>= 20906) {
286 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
289 new_op
->upcall
.req
.features
.features
= 0;
290 ret
= service_operation(new_op
, "orangefs_features",
291 ORANGEFS_OP_PRIORITY
| ORANGEFS_OP_NO_MUTEX
);
294 new_op
->downcall
.resp
.features
.features
;
296 orangefs_features
= 0;
299 orangefs_features
= 0;
305 int fsid_key_table_initialize(void)
310 void fsid_key_table_finalize(void)
314 static const struct super_operations orangefs_s_ops
= {
315 .alloc_inode
= orangefs_alloc_inode
,
316 .free_inode
= orangefs_free_inode
,
317 .destroy_inode
= orangefs_destroy_inode
,
318 .write_inode
= orangefs_write_inode
,
319 .drop_inode
= generic_delete_inode
,
320 .statfs
= orangefs_statfs
,
321 .remount_fs
= orangefs_remount_fs
,
322 .show_options
= orangefs_show_options
,
325 static struct dentry
*orangefs_fh_to_dentry(struct super_block
*sb
,
330 struct orangefs_object_kref refn
;
332 if (fh_len
< 5 || fh_type
> 2)
335 ORANGEFS_khandle_from(&(refn
.khandle
), fid
->raw
, 16);
336 refn
.fs_id
= (u32
) fid
->raw
[4];
337 gossip_debug(GOSSIP_SUPER_DEBUG
,
338 "fh_to_dentry: handle %pU, fs_id %d\n",
342 return d_obtain_alias(orangefs_iget(sb
, &refn
));
345 static int orangefs_encode_fh(struct inode
*inode
,
348 struct inode
*parent
)
350 int len
= parent
? 10 : 5;
352 struct orangefs_object_kref refn
;
354 if (*max_len
< len
) {
355 gossip_err("fh buffer is too small for encoding\n");
361 refn
= ORANGEFS_I(inode
)->refn
;
362 ORANGEFS_khandle_to(&refn
.khandle
, fh
, 16);
365 gossip_debug(GOSSIP_SUPER_DEBUG
,
366 "Encoding fh: handle %pU, fsid %u\n",
372 refn
= ORANGEFS_I(parent
)->refn
;
373 ORANGEFS_khandle_to(&refn
.khandle
, (char *) fh
+ 20, 16);
377 gossip_debug(GOSSIP_SUPER_DEBUG
,
378 "Encoding parent: handle %pU, fsid %u\n",
388 static const struct export_operations orangefs_export_ops
= {
389 .encode_fh
= orangefs_encode_fh
,
390 .fh_to_dentry
= orangefs_fh_to_dentry
,
393 static int orangefs_unmount(int id
, __s32 fs_id
, const char *devname
)
395 struct orangefs_kernel_op_s
*op
;
397 op
= op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT
);
400 op
->upcall
.req
.fs_umount
.id
= id
;
401 op
->upcall
.req
.fs_umount
.fs_id
= fs_id
;
402 strncpy(op
->upcall
.req
.fs_umount
.orangefs_config_server
,
403 devname
, ORANGEFS_MAX_SERVER_ADDR_LEN
- 1);
404 r
= service_operation(op
, "orangefs_fs_umount", 0);
405 /* Not much to do about an error here. */
407 gossip_err("orangefs_unmount: service_operation %d\n", r
);
412 static int orangefs_fill_sb(struct super_block
*sb
,
413 struct orangefs_fs_mount_response
*fs_mount
,
414 void *data
, int silent
)
418 struct dentry
*root_dentry
;
419 struct orangefs_object_kref root_object
;
421 ORANGEFS_SB(sb
)->sb
= sb
;
423 ORANGEFS_SB(sb
)->root_khandle
= fs_mount
->root_khandle
;
424 ORANGEFS_SB(sb
)->fs_id
= fs_mount
->fs_id
;
425 ORANGEFS_SB(sb
)->id
= fs_mount
->id
;
428 ret
= parse_mount_options(sb
, data
, silent
);
433 /* Hang the xattr handlers off the superblock */
434 sb
->s_xattr
= orangefs_xattr_handlers
;
435 sb
->s_magic
= ORANGEFS_SUPER_MAGIC
;
436 sb
->s_op
= &orangefs_s_ops
;
437 sb
->s_d_op
= &orangefs_dentry_operations
;
439 sb
->s_blocksize
= PAGE_SIZE
;
440 sb
->s_blocksize_bits
= PAGE_SHIFT
;
441 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
443 ret
= super_setup_bdi(sb
);
447 root_object
.khandle
= ORANGEFS_SB(sb
)->root_khandle
;
448 root_object
.fs_id
= ORANGEFS_SB(sb
)->fs_id
;
449 gossip_debug(GOSSIP_SUPER_DEBUG
,
450 "get inode %pU, fsid %d\n",
451 &root_object
.khandle
,
454 root
= orangefs_iget(sb
, &root_object
);
456 return PTR_ERR(root
);
458 gossip_debug(GOSSIP_SUPER_DEBUG
,
459 "Allocated root inode [%p] with mode %x\n",
463 /* allocates and places root dentry in dcache */
464 root_dentry
= d_make_root(root
);
468 sb
->s_export_op
= &orangefs_export_ops
;
469 sb
->s_root
= root_dentry
;
473 struct dentry
*orangefs_mount(struct file_system_type
*fst
,
479 struct super_block
*sb
= ERR_PTR(-EINVAL
);
480 struct orangefs_kernel_op_s
*new_op
;
481 struct dentry
*d
= ERR_PTR(-EINVAL
);
483 gossip_debug(GOSSIP_SUPER_DEBUG
,
484 "orangefs_mount: called with devname %s\n",
488 gossip_err("ERROR: device name not specified.\n");
489 return ERR_PTR(-EINVAL
);
492 new_op
= op_alloc(ORANGEFS_VFS_OP_FS_MOUNT
);
494 return ERR_PTR(-ENOMEM
);
496 strncpy(new_op
->upcall
.req
.fs_mount
.orangefs_config_server
,
498 ORANGEFS_MAX_SERVER_ADDR_LEN
- 1);
500 gossip_debug(GOSSIP_SUPER_DEBUG
,
501 "Attempting ORANGEFS Mount via host %s\n",
502 new_op
->upcall
.req
.fs_mount
.orangefs_config_server
);
504 ret
= service_operation(new_op
, "orangefs_mount", 0);
505 gossip_debug(GOSSIP_SUPER_DEBUG
,
506 "orangefs_mount: mount got return value of %d\n", ret
);
510 if (new_op
->downcall
.resp
.fs_mount
.fs_id
== ORANGEFS_FS_ID_NULL
) {
511 gossip_err("ERROR: Retrieved null fs_id\n");
516 sb
= sget(fst
, NULL
, set_anon_super
, flags
, NULL
);
520 orangefs_unmount(new_op
->downcall
.resp
.fs_mount
.id
,
521 new_op
->downcall
.resp
.fs_mount
.fs_id
, devname
);
525 /* alloc and init our private orangefs sb info */
526 sb
->s_fs_info
= kzalloc(sizeof(struct orangefs_sb_info_s
), GFP_KERNEL
);
527 if (!ORANGEFS_SB(sb
)) {
528 d
= ERR_PTR(-ENOMEM
);
532 ret
= orangefs_fill_sb(sb
,
533 &new_op
->downcall
.resp
.fs_mount
, data
,
534 flags
& SB_SILENT
? 1 : 0);
542 * on successful mount, store the devname and data
545 strncpy(ORANGEFS_SB(sb
)->devname
,
547 ORANGEFS_MAX_SERVER_ADDR_LEN
- 1);
549 /* mount_pending must be cleared */
550 ORANGEFS_SB(sb
)->mount_pending
= 0;
553 * finally, add this sb to our list of known orangefs
556 gossip_debug(GOSSIP_SUPER_DEBUG
,
557 "Adding SB %p to orangefs superblocks\n",
559 spin_lock(&orangefs_superblocks_lock
);
560 list_add_tail(&ORANGEFS_SB(sb
)->list
, &orangefs_superblocks
);
561 spin_unlock(&orangefs_superblocks_lock
);
564 /* Must be removed from the list now. */
565 ORANGEFS_SB(sb
)->no_list
= 0;
567 if (orangefs_userspace_version
>= 20906) {
568 new_op
= op_alloc(ORANGEFS_VFS_OP_FEATURES
);
570 return ERR_PTR(-ENOMEM
);
571 new_op
->upcall
.req
.features
.features
= 0;
572 ret
= service_operation(new_op
, "orangefs_features", 0);
573 orangefs_features
= new_op
->downcall
.resp
.features
.features
;
576 orangefs_features
= 0;
579 return dget(sb
->s_root
);
582 /* Will call orangefs_kill_sb with sb not in list. */
583 ORANGEFS_SB(sb
)->no_list
= 1;
584 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
585 deactivate_locked_super(sb
);
587 gossip_err("orangefs_mount: mount request failed with %d\n", ret
);
588 if (ret
== -EINVAL
) {
589 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
590 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
598 void orangefs_kill_sb(struct super_block
*sb
)
601 gossip_debug(GOSSIP_SUPER_DEBUG
, "orangefs_kill_sb: called\n");
603 /* provided sb cleanup */
606 if (!ORANGEFS_SB(sb
)) {
607 mutex_lock(&orangefs_request_mutex
);
608 mutex_unlock(&orangefs_request_mutex
);
612 * issue the unmount to userspace to tell it to remove the
613 * dynamic mount info it has for this superblock
615 r
= orangefs_unmount(ORANGEFS_SB(sb
)->id
, ORANGEFS_SB(sb
)->fs_id
,
616 ORANGEFS_SB(sb
)->devname
);
618 ORANGEFS_SB(sb
)->mount_pending
= 1;
620 if (!ORANGEFS_SB(sb
)->no_list
) {
621 /* remove the sb from our list of orangefs specific sb's */
622 spin_lock(&orangefs_superblocks_lock
);
623 /* not list_del_init */
624 __list_del_entry(&ORANGEFS_SB(sb
)->list
);
625 ORANGEFS_SB(sb
)->list
.prev
= NULL
;
626 spin_unlock(&orangefs_superblocks_lock
);
630 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
631 * gets completed before we free the dang thing.
633 mutex_lock(&orangefs_request_mutex
);
634 mutex_unlock(&orangefs_request_mutex
);
636 /* free the orangefs superblock private data */
637 kfree(ORANGEFS_SB(sb
));
640 int orangefs_inode_cache_initialize(void)
642 orangefs_inode_cache
= kmem_cache_create_usercopy(
643 "orangefs_inode_cache",
644 sizeof(struct orangefs_inode_s
),
646 ORANGEFS_CACHE_CREATE_FLAGS
,
647 offsetof(struct orangefs_inode_s
,
649 sizeof_field(struct orangefs_inode_s
,
651 orangefs_inode_cache_ctor
);
653 if (!orangefs_inode_cache
) {
654 gossip_err("Cannot create orangefs_inode_cache\n");
660 int orangefs_inode_cache_finalize(void)
662 kmem_cache_destroy(orangefs_inode_cache
);