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_i_callback(struct rcu_head *head)
130 struct inode *inode = container_of(head, struct inode, i_rcu);
131 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
132 struct orangefs_cached_xattr *cx;
133 struct hlist_node *tmp;
136 hash_for_each_safe(orangefs_inode->xattr_cache, i, tmp, cx, node) {
137 hlist_del(&cx->node);
141 kmem_cache_free(orangefs_inode_cache, orangefs_inode);
144 static void orangefs_destroy_inode(struct inode *inode)
146 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
148 gossip_debug(GOSSIP_SUPER_DEBUG,
149 "%s: deallocated %p destroying inode %pU\n",
150 __func__, orangefs_inode, get_khandle_from_ino(inode));
152 call_rcu(&inode->i_rcu, orangefs_i_callback);
155 static int orangefs_write_inode(struct inode *inode,
156 struct writeback_control *wbc)
158 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_write_inode\n");
159 return orangefs_inode_setattr(inode);
163 * NOTE: information filled in here is typically reflected in the
164 * output of the system command 'df'
166 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
169 struct orangefs_kernel_op_s *new_op = NULL;
171 struct super_block *sb = NULL;
175 gossip_debug(GOSSIP_SUPER_DEBUG,
176 "%s: called on sb %p (fs_id is %d)\n",
179 (int)(ORANGEFS_SB(sb)->fs_id));
181 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
184 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
186 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
187 flags = ORANGEFS_OP_INTERRUPTIBLE;
189 ret = service_operation(new_op, "orangefs_statfs", flags);
191 if (new_op->downcall.status < 0)
194 gossip_debug(GOSSIP_SUPER_DEBUG,
195 "%s: got %ld blocks available | "
196 "%ld blocks total | %ld block size | "
197 "%ld files total | %ld files avail\n",
199 (long)new_op->downcall.resp.statfs.blocks_avail,
200 (long)new_op->downcall.resp.statfs.blocks_total,
201 (long)new_op->downcall.resp.statfs.block_size,
202 (long)new_op->downcall.resp.statfs.files_total,
203 (long)new_op->downcall.resp.statfs.files_avail);
205 buf->f_type = sb->s_magic;
206 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
207 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
208 buf->f_namelen = ORANGEFS_NAME_MAX;
210 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
211 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
212 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
213 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
214 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
215 buf->f_frsize = sb->s_blocksize;
219 gossip_debug(GOSSIP_SUPER_DEBUG, "%s: returning %d\n", __func__, ret);
224 * Remount as initiated by VFS layer. We just need to reparse the mount
225 * options, no need to signal pvfs2-client-core about it.
227 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
229 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
230 return parse_mount_options(sb, data, 1);
234 * Remount as initiated by pvfs2-client-core on restart. This is used to
235 * repopulate mount information left from previous pvfs2-client-core.
237 * the idea here is that given a valid superblock, we're
238 * re-initializing the user space client with the initial mount
239 * information specified when the super block was first initialized.
240 * this is very different than the first initialization/creation of a
241 * superblock. we use the special service_priority_operation to make
242 * sure that the mount gets ahead of any other pending operation that
243 * is waiting for servicing. this means that the pvfs2-client won't
244 * fail to start several times for all other pending operations before
245 * the client regains all of the mount information from us.
246 * NOTE: this function assumes that the request_mutex is already acquired!
248 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
250 struct orangefs_kernel_op_s *new_op;
253 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
255 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
258 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
259 orangefs_sb->devname,
260 ORANGEFS_MAX_SERVER_ADDR_LEN);
262 gossip_debug(GOSSIP_SUPER_DEBUG,
263 "Attempting ORANGEFS Remount via host %s\n",
264 new_op->upcall.req.fs_mount.orangefs_config_server);
267 * we assume that the calling function has already acquired the
268 * request_mutex to prevent other operations from bypassing
271 ret = service_operation(new_op, "orangefs_remount",
272 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
273 gossip_debug(GOSSIP_SUPER_DEBUG,
274 "orangefs_remount: mount got return value of %d\n",
278 * store the id assigned to this sb -- it's just a
279 * short-lived mapping that the system interface uses
280 * to map this superblock to a particular mount entry
282 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
283 orangefs_sb->mount_pending = 0;
288 if (orangefs_userspace_version >= 20906) {
289 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
292 new_op->upcall.req.features.features = 0;
293 ret = service_operation(new_op, "orangefs_features",
294 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
297 new_op->downcall.resp.features.features;
299 orangefs_features = 0;
302 orangefs_features = 0;
308 int fsid_key_table_initialize(void)
313 void fsid_key_table_finalize(void)
317 static const struct super_operations orangefs_s_ops = {
318 .alloc_inode = orangefs_alloc_inode,
319 .destroy_inode = orangefs_destroy_inode,
320 .write_inode = orangefs_write_inode,
321 .drop_inode = generic_delete_inode,
322 .statfs = orangefs_statfs,
323 .remount_fs = orangefs_remount_fs,
324 .show_options = orangefs_show_options,
327 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
332 struct orangefs_object_kref refn;
334 if (fh_len < 5 || fh_type > 2)
337 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
338 refn.fs_id = (u32) fid->raw[4];
339 gossip_debug(GOSSIP_SUPER_DEBUG,
340 "fh_to_dentry: handle %pU, fs_id %d\n",
344 return d_obtain_alias(orangefs_iget(sb, &refn));
347 static int orangefs_encode_fh(struct inode *inode,
350 struct inode *parent)
352 int len = parent ? 10 : 5;
354 struct orangefs_object_kref refn;
356 if (*max_len < len) {
357 gossip_err("fh buffer is too small for encoding\n");
363 refn = ORANGEFS_I(inode)->refn;
364 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
367 gossip_debug(GOSSIP_SUPER_DEBUG,
368 "Encoding fh: handle %pU, fsid %u\n",
374 refn = ORANGEFS_I(parent)->refn;
375 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
379 gossip_debug(GOSSIP_SUPER_DEBUG,
380 "Encoding parent: handle %pU, fsid %u\n",
390 static const struct export_operations orangefs_export_ops = {
391 .encode_fh = orangefs_encode_fh,
392 .fh_to_dentry = orangefs_fh_to_dentry,
395 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
397 struct orangefs_kernel_op_s *op;
399 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
402 op->upcall.req.fs_umount.id = id;
403 op->upcall.req.fs_umount.fs_id = fs_id;
404 strncpy(op->upcall.req.fs_umount.orangefs_config_server,
405 devname, ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
406 r = service_operation(op, "orangefs_fs_umount", 0);
407 /* Not much to do about an error here. */
409 gossip_err("orangefs_unmount: service_operation %d\n", r);
414 static int orangefs_fill_sb(struct super_block *sb,
415 struct orangefs_fs_mount_response *fs_mount,
416 void *data, int silent)
420 struct dentry *root_dentry;
421 struct orangefs_object_kref root_object;
423 ORANGEFS_SB(sb)->sb = sb;
425 ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle;
426 ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id;
427 ORANGEFS_SB(sb)->id = fs_mount->id;
430 ret = parse_mount_options(sb, data, silent);
435 /* Hang the xattr handlers off the superblock */
436 sb->s_xattr = orangefs_xattr_handlers;
437 sb->s_magic = ORANGEFS_SUPER_MAGIC;
438 sb->s_op = &orangefs_s_ops;
439 sb->s_d_op = &orangefs_dentry_operations;
441 sb->s_blocksize = PAGE_SIZE;
442 sb->s_blocksize_bits = PAGE_SHIFT;
443 sb->s_maxbytes = MAX_LFS_FILESIZE;
445 ret = super_setup_bdi(sb);
449 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
450 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
451 gossip_debug(GOSSIP_SUPER_DEBUG,
452 "get inode %pU, fsid %d\n",
453 &root_object.khandle,
456 root = orangefs_iget(sb, &root_object);
458 return PTR_ERR(root);
460 gossip_debug(GOSSIP_SUPER_DEBUG,
461 "Allocated root inode [%p] with mode %x\n",
465 /* allocates and places root dentry in dcache */
466 root_dentry = d_make_root(root);
470 sb->s_export_op = &orangefs_export_ops;
471 sb->s_root = root_dentry;
475 struct dentry *orangefs_mount(struct file_system_type *fst,
481 struct super_block *sb = ERR_PTR(-EINVAL);
482 struct orangefs_kernel_op_s *new_op;
483 struct dentry *d = ERR_PTR(-EINVAL);
485 gossip_debug(GOSSIP_SUPER_DEBUG,
486 "orangefs_mount: called with devname %s\n",
490 gossip_err("ERROR: device name not specified.\n");
491 return ERR_PTR(-EINVAL);
494 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
496 return ERR_PTR(-ENOMEM);
498 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
500 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
502 gossip_debug(GOSSIP_SUPER_DEBUG,
503 "Attempting ORANGEFS Mount via host %s\n",
504 new_op->upcall.req.fs_mount.orangefs_config_server);
506 ret = service_operation(new_op, "orangefs_mount", 0);
507 gossip_debug(GOSSIP_SUPER_DEBUG,
508 "orangefs_mount: mount got return value of %d\n", ret);
512 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
513 gossip_err("ERROR: Retrieved null fs_id\n");
518 sb = sget(fst, NULL, set_anon_super, flags, NULL);
522 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
523 new_op->downcall.resp.fs_mount.fs_id, devname);
527 /* alloc and init our private orangefs sb info */
528 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
529 if (!ORANGEFS_SB(sb)) {
530 d = ERR_PTR(-ENOMEM);
534 ret = orangefs_fill_sb(sb,
535 &new_op->downcall.resp.fs_mount, data,
536 flags & SB_SILENT ? 1 : 0);
544 * on successful mount, store the devname and data
547 strncpy(ORANGEFS_SB(sb)->devname,
549 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
551 /* mount_pending must be cleared */
552 ORANGEFS_SB(sb)->mount_pending = 0;
555 * finally, add this sb to our list of known orangefs
558 gossip_debug(GOSSIP_SUPER_DEBUG,
559 "Adding SB %p to orangefs superblocks\n",
561 spin_lock(&orangefs_superblocks_lock);
562 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
563 spin_unlock(&orangefs_superblocks_lock);
566 /* Must be removed from the list now. */
567 ORANGEFS_SB(sb)->no_list = 0;
569 if (orangefs_userspace_version >= 20906) {
570 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
572 return ERR_PTR(-ENOMEM);
573 new_op->upcall.req.features.features = 0;
574 ret = service_operation(new_op, "orangefs_features", 0);
575 orangefs_features = new_op->downcall.resp.features.features;
578 orangefs_features = 0;
581 return dget(sb->s_root);
584 /* Will call orangefs_kill_sb with sb not in list. */
585 ORANGEFS_SB(sb)->no_list = 1;
586 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
587 deactivate_locked_super(sb);
589 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
590 if (ret == -EINVAL) {
591 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
592 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
600 void orangefs_kill_sb(struct super_block *sb)
603 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
605 /* provided sb cleanup */
608 if (!ORANGEFS_SB(sb)) {
609 mutex_lock(&orangefs_request_mutex);
610 mutex_unlock(&orangefs_request_mutex);
614 * issue the unmount to userspace to tell it to remove the
615 * dynamic mount info it has for this superblock
617 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
618 ORANGEFS_SB(sb)->devname);
620 ORANGEFS_SB(sb)->mount_pending = 1;
622 if (!ORANGEFS_SB(sb)->no_list) {
623 /* remove the sb from our list of orangefs specific sb's */
624 spin_lock(&orangefs_superblocks_lock);
625 /* not list_del_init */
626 __list_del_entry(&ORANGEFS_SB(sb)->list);
627 ORANGEFS_SB(sb)->list.prev = NULL;
628 spin_unlock(&orangefs_superblocks_lock);
632 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
633 * gets completed before we free the dang thing.
635 mutex_lock(&orangefs_request_mutex);
636 mutex_unlock(&orangefs_request_mutex);
638 /* free the orangefs superblock private data */
639 kfree(ORANGEFS_SB(sb));
642 int orangefs_inode_cache_initialize(void)
644 orangefs_inode_cache = kmem_cache_create_usercopy(
645 "orangefs_inode_cache",
646 sizeof(struct orangefs_inode_s),
648 ORANGEFS_CACHE_CREATE_FLAGS,
649 offsetof(struct orangefs_inode_s,
651 sizeof_field(struct orangefs_inode_s,
653 orangefs_inode_cache_ctor);
655 if (!orangefs_inode_cache) {
656 gossip_err("Cannot create orangefs_inode_cache\n");
662 int orangefs_inode_cache_finalize(void)
664 kmem_cache_destroy(orangefs_inode_cache);