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>
14 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
15 static struct kmem_cache *orangefs_inode_cache;
17 /* list for storing orangefs specific superblocks in use */
18 LIST_HEAD(orangefs_superblocks);
20 DEFINE_SPINLOCK(orangefs_superblocks_lock);
30 static const match_table_t tokens = {
33 { Opt_local_lock, "local_lock" },
37 uint64_t orangefs_features;
39 static int orangefs_show_options(struct seq_file *m, struct dentry *root)
41 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(root->d_sb);
43 if (root->d_sb->s_flags & SB_POSIXACL)
45 if (orangefs_sb->flags & ORANGEFS_OPT_INTR)
47 if (orangefs_sb->flags & ORANGEFS_OPT_LOCAL_LOCK)
48 seq_puts(m, ",local_lock");
52 static int parse_mount_options(struct super_block *sb, char *options,
55 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(sb);
56 substring_t args[MAX_OPT_ARGS];
60 * Force any potential flags that might be set from the mount
61 * to zero, ie, initialize to unset.
63 sb->s_flags &= ~SB_POSIXACL;
64 orangefs_sb->flags &= ~ORANGEFS_OPT_INTR;
65 orangefs_sb->flags &= ~ORANGEFS_OPT_LOCAL_LOCK;
67 while ((p = strsep(&options, ",")) != NULL) {
73 token = match_token(p, tokens, args);
76 sb->s_flags |= SB_POSIXACL;
79 orangefs_sb->flags |= ORANGEFS_OPT_INTR;
82 orangefs_sb->flags |= ORANGEFS_OPT_LOCAL_LOCK;
92 gossip_err("Error: mount option [%s] is not supported.\n", p);
96 static void orangefs_inode_cache_ctor(void *req)
98 struct orangefs_inode_s *orangefs_inode = req;
100 inode_init_once(&orangefs_inode->vfs_inode);
101 init_rwsem(&orangefs_inode->xattr_sem);
104 static struct inode *orangefs_alloc_inode(struct super_block *sb)
106 struct orangefs_inode_s *orangefs_inode;
108 orangefs_inode = kmem_cache_alloc(orangefs_inode_cache, GFP_KERNEL);
113 * We want to clear everything except for rw_semaphore and the
116 memset(&orangefs_inode->refn.khandle, 0, 16);
117 orangefs_inode->refn.fs_id = ORANGEFS_FS_ID_NULL;
118 orangefs_inode->last_failed_block_index_read = 0;
119 memset(orangefs_inode->link_target, 0, sizeof(orangefs_inode->link_target));
121 gossip_debug(GOSSIP_SUPER_DEBUG,
122 "orangefs_alloc_inode: allocated %p\n",
123 &orangefs_inode->vfs_inode);
124 return &orangefs_inode->vfs_inode;
127 static void orangefs_free_inode(struct inode *inode)
129 kmem_cache_free(orangefs_inode_cache, ORANGEFS_I(inode));
132 static void orangefs_destroy_inode(struct inode *inode)
134 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
136 gossip_debug(GOSSIP_SUPER_DEBUG,
137 "%s: deallocated %p destroying inode %pU\n",
138 __func__, orangefs_inode, get_khandle_from_ino(inode));
142 * NOTE: information filled in here is typically reflected in the
143 * output of the system command 'df'
145 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
148 struct orangefs_kernel_op_s *new_op = NULL;
150 struct super_block *sb = NULL;
154 gossip_debug(GOSSIP_SUPER_DEBUG,
155 "%s: called on sb %p (fs_id is %d)\n",
158 (int)(ORANGEFS_SB(sb)->fs_id));
160 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
163 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
165 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
166 flags = ORANGEFS_OP_INTERRUPTIBLE;
168 ret = service_operation(new_op, "orangefs_statfs", flags);
170 if (new_op->downcall.status < 0)
173 gossip_debug(GOSSIP_SUPER_DEBUG,
174 "%s: got %ld blocks available | "
175 "%ld blocks total | %ld block size | "
176 "%ld files total | %ld files avail\n",
178 (long)new_op->downcall.resp.statfs.blocks_avail,
179 (long)new_op->downcall.resp.statfs.blocks_total,
180 (long)new_op->downcall.resp.statfs.block_size,
181 (long)new_op->downcall.resp.statfs.files_total,
182 (long)new_op->downcall.resp.statfs.files_avail);
184 buf->f_type = sb->s_magic;
185 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
186 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
187 buf->f_namelen = ORANGEFS_NAME_MAX;
189 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
190 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
191 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
192 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
193 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
194 buf->f_frsize = sb->s_blocksize;
198 gossip_debug(GOSSIP_SUPER_DEBUG, "%s: returning %d\n", __func__, ret);
203 * Remount as initiated by VFS layer. We just need to reparse the mount
204 * options, no need to signal pvfs2-client-core about it.
206 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
208 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
209 return parse_mount_options(sb, data, 1);
213 * Remount as initiated by pvfs2-client-core on restart. This is used to
214 * repopulate mount information left from previous pvfs2-client-core.
216 * the idea here is that given a valid superblock, we're
217 * re-initializing the user space client with the initial mount
218 * information specified when the super block was first initialized.
219 * this is very different than the first initialization/creation of a
220 * superblock. we use the special service_priority_operation to make
221 * sure that the mount gets ahead of any other pending operation that
222 * is waiting for servicing. this means that the pvfs2-client won't
223 * fail to start several times for all other pending operations before
224 * the client regains all of the mount information from us.
225 * NOTE: this function assumes that the request_mutex is already acquired!
227 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
229 struct orangefs_kernel_op_s *new_op;
232 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
234 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
237 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
238 orangefs_sb->devname,
239 ORANGEFS_MAX_SERVER_ADDR_LEN);
241 gossip_debug(GOSSIP_SUPER_DEBUG,
242 "Attempting ORANGEFS Remount via host %s\n",
243 new_op->upcall.req.fs_mount.orangefs_config_server);
246 * we assume that the calling function has already acquired the
247 * request_mutex to prevent other operations from bypassing
250 ret = service_operation(new_op, "orangefs_remount",
251 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
252 gossip_debug(GOSSIP_SUPER_DEBUG,
253 "orangefs_remount: mount got return value of %d\n",
257 * store the id assigned to this sb -- it's just a
258 * short-lived mapping that the system interface uses
259 * to map this superblock to a particular mount entry
261 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
262 orangefs_sb->mount_pending = 0;
267 if (orangefs_userspace_version >= 20906) {
268 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
271 new_op->upcall.req.features.features = 0;
272 ret = service_operation(new_op, "orangefs_features",
273 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
276 new_op->downcall.resp.features.features;
278 orangefs_features = 0;
281 orangefs_features = 0;
287 int fsid_key_table_initialize(void)
292 void fsid_key_table_finalize(void)
296 static const struct super_operations orangefs_s_ops = {
297 .alloc_inode = orangefs_alloc_inode,
298 .free_inode = orangefs_free_inode,
299 .destroy_inode = orangefs_destroy_inode,
300 .drop_inode = generic_delete_inode,
301 .statfs = orangefs_statfs,
302 .remount_fs = orangefs_remount_fs,
303 .show_options = orangefs_show_options,
306 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
311 struct orangefs_object_kref refn;
313 if (fh_len < 5 || fh_type > 2)
316 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
317 refn.fs_id = (u32) fid->raw[4];
318 gossip_debug(GOSSIP_SUPER_DEBUG,
319 "fh_to_dentry: handle %pU, fs_id %d\n",
323 return d_obtain_alias(orangefs_iget(sb, &refn));
326 static int orangefs_encode_fh(struct inode *inode,
329 struct inode *parent)
331 int len = parent ? 10 : 5;
333 struct orangefs_object_kref refn;
335 if (*max_len < len) {
336 gossip_err("fh buffer is too small for encoding\n");
342 refn = ORANGEFS_I(inode)->refn;
343 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
346 gossip_debug(GOSSIP_SUPER_DEBUG,
347 "Encoding fh: handle %pU, fsid %u\n",
353 refn = ORANGEFS_I(parent)->refn;
354 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
358 gossip_debug(GOSSIP_SUPER_DEBUG,
359 "Encoding parent: handle %pU, fsid %u\n",
369 static const struct export_operations orangefs_export_ops = {
370 .encode_fh = orangefs_encode_fh,
371 .fh_to_dentry = orangefs_fh_to_dentry,
374 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
376 struct orangefs_kernel_op_s *op;
378 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
381 op->upcall.req.fs_umount.id = id;
382 op->upcall.req.fs_umount.fs_id = fs_id;
383 strncpy(op->upcall.req.fs_umount.orangefs_config_server,
384 devname, ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
385 r = service_operation(op, "orangefs_fs_umount", 0);
386 /* Not much to do about an error here. */
388 gossip_err("orangefs_unmount: service_operation %d\n", r);
393 static int orangefs_fill_sb(struct super_block *sb,
394 struct orangefs_fs_mount_response *fs_mount,
395 void *data, int silent)
398 struct inode *root = NULL;
399 struct dentry *root_dentry = NULL;
400 struct orangefs_object_kref root_object;
402 /* alloc and init our private orangefs sb info */
403 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
404 if (!ORANGEFS_SB(sb))
406 ORANGEFS_SB(sb)->sb = sb;
408 ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle;
409 ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id;
410 ORANGEFS_SB(sb)->id = fs_mount->id;
413 ret = parse_mount_options(sb, data, silent);
418 /* Hang the xattr handlers off the superblock */
419 sb->s_xattr = orangefs_xattr_handlers;
420 sb->s_magic = ORANGEFS_SUPER_MAGIC;
421 sb->s_op = &orangefs_s_ops;
422 sb->s_d_op = &orangefs_dentry_operations;
424 sb->s_blocksize = PAGE_SIZE;
425 sb->s_blocksize_bits = PAGE_SHIFT;
426 sb->s_maxbytes = MAX_LFS_FILESIZE;
428 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
429 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
430 gossip_debug(GOSSIP_SUPER_DEBUG,
431 "get inode %pU, fsid %d\n",
432 &root_object.khandle,
435 root = orangefs_iget(sb, &root_object);
437 return PTR_ERR(root);
439 gossip_debug(GOSSIP_SUPER_DEBUG,
440 "Allocated root inode [%p] with mode %x\n",
444 /* allocates and places root dentry in dcache */
445 root_dentry = d_make_root(root);
449 sb->s_export_op = &orangefs_export_ops;
450 sb->s_root = root_dentry;
454 struct dentry *orangefs_mount(struct file_system_type *fst,
460 struct super_block *sb = ERR_PTR(-EINVAL);
461 struct orangefs_kernel_op_s *new_op;
462 struct dentry *d = ERR_PTR(-EINVAL);
464 gossip_debug(GOSSIP_SUPER_DEBUG,
465 "orangefs_mount: called with devname %s\n",
469 gossip_err("ERROR: device name not specified.\n");
470 return ERR_PTR(-EINVAL);
473 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
475 return ERR_PTR(-ENOMEM);
477 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
479 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
481 gossip_debug(GOSSIP_SUPER_DEBUG,
482 "Attempting ORANGEFS Mount via host %s\n",
483 new_op->upcall.req.fs_mount.orangefs_config_server);
485 ret = service_operation(new_op, "orangefs_mount", 0);
486 gossip_debug(GOSSIP_SUPER_DEBUG,
487 "orangefs_mount: mount got return value of %d\n", ret);
491 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
492 gossip_err("ERROR: Retrieved null fs_id\n");
497 sb = sget(fst, NULL, set_anon_super, flags, NULL);
501 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
502 new_op->downcall.resp.fs_mount.fs_id, devname);
506 ret = orangefs_fill_sb(sb,
507 &new_op->downcall.resp.fs_mount, data,
508 flags & SB_SILENT ? 1 : 0);
516 * on successful mount, store the devname and data
519 strncpy(ORANGEFS_SB(sb)->devname,
521 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
523 /* mount_pending must be cleared */
524 ORANGEFS_SB(sb)->mount_pending = 0;
527 * finally, add this sb to our list of known orangefs
530 gossip_debug(GOSSIP_SUPER_DEBUG,
531 "Adding SB %p to orangefs superblocks\n",
533 spin_lock(&orangefs_superblocks_lock);
534 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
535 spin_unlock(&orangefs_superblocks_lock);
538 /* Must be removed from the list now. */
539 ORANGEFS_SB(sb)->no_list = 0;
541 if (orangefs_userspace_version >= 20906) {
542 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
544 return ERR_PTR(-ENOMEM);
545 new_op->upcall.req.features.features = 0;
546 ret = service_operation(new_op, "orangefs_features", 0);
547 orangefs_features = new_op->downcall.resp.features.features;
550 orangefs_features = 0;
553 return dget(sb->s_root);
556 /* Will call orangefs_kill_sb with sb not in list. */
557 ORANGEFS_SB(sb)->no_list = 1;
558 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
559 deactivate_locked_super(sb);
561 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
562 if (ret == -EINVAL) {
563 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
564 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
572 void orangefs_kill_sb(struct super_block *sb)
575 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
577 /* provided sb cleanup */
580 if (!ORANGEFS_SB(sb)) {
581 mutex_lock(&orangefs_request_mutex);
582 mutex_unlock(&orangefs_request_mutex);
586 * issue the unmount to userspace to tell it to remove the
587 * dynamic mount info it has for this superblock
589 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
590 ORANGEFS_SB(sb)->devname);
592 ORANGEFS_SB(sb)->mount_pending = 1;
594 if (!ORANGEFS_SB(sb)->no_list) {
595 /* remove the sb from our list of orangefs specific sb's */
596 spin_lock(&orangefs_superblocks_lock);
597 /* not list_del_init */
598 __list_del_entry(&ORANGEFS_SB(sb)->list);
599 ORANGEFS_SB(sb)->list.prev = NULL;
600 spin_unlock(&orangefs_superblocks_lock);
604 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
605 * gets completed before we free the dang thing.
607 mutex_lock(&orangefs_request_mutex);
608 mutex_unlock(&orangefs_request_mutex);
610 /* free the orangefs superblock private data */
611 kfree(ORANGEFS_SB(sb));
614 int orangefs_inode_cache_initialize(void)
616 orangefs_inode_cache = kmem_cache_create_usercopy(
617 "orangefs_inode_cache",
618 sizeof(struct orangefs_inode_s),
620 ORANGEFS_CACHE_CREATE_FLAGS,
621 offsetof(struct orangefs_inode_s,
623 sizeof_field(struct orangefs_inode_s,
625 orangefs_inode_cache_ctor);
627 if (!orangefs_inode_cache) {
628 gossip_err("Cannot create orangefs_inode_cache\n");
634 int orangefs_inode_cache_finalize(void)
636 kmem_cache_destroy(orangefs_inode_cache);