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);
156 * NOTE: information filled in here is typically reflected in the
157 * output of the system command 'df'
159 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
162 struct orangefs_kernel_op_s *new_op = NULL;
164 struct super_block *sb = NULL;
168 gossip_debug(GOSSIP_SUPER_DEBUG,
169 "%s: called on sb %p (fs_id is %d)\n",
172 (int)(ORANGEFS_SB(sb)->fs_id));
174 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
177 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
179 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
180 flags = ORANGEFS_OP_INTERRUPTIBLE;
182 ret = service_operation(new_op, "orangefs_statfs", flags);
184 if (new_op->downcall.status < 0)
187 gossip_debug(GOSSIP_SUPER_DEBUG,
188 "%s: got %ld blocks available | "
189 "%ld blocks total | %ld block size | "
190 "%ld files total | %ld files avail\n",
192 (long)new_op->downcall.resp.statfs.blocks_avail,
193 (long)new_op->downcall.resp.statfs.blocks_total,
194 (long)new_op->downcall.resp.statfs.block_size,
195 (long)new_op->downcall.resp.statfs.files_total,
196 (long)new_op->downcall.resp.statfs.files_avail);
198 buf->f_type = sb->s_magic;
199 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
200 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
201 buf->f_namelen = ORANGEFS_NAME_MAX;
203 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
204 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
205 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
206 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
207 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
208 buf->f_frsize = sb->s_blocksize;
212 gossip_debug(GOSSIP_SUPER_DEBUG, "%s: returning %d\n", __func__, ret);
217 * Remount as initiated by VFS layer. We just need to reparse the mount
218 * options, no need to signal pvfs2-client-core about it.
220 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
222 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
223 return parse_mount_options(sb, data, 1);
227 * Remount as initiated by pvfs2-client-core on restart. This is used to
228 * repopulate mount information left from previous pvfs2-client-core.
230 * the idea here is that given a valid superblock, we're
231 * re-initializing the user space client with the initial mount
232 * information specified when the super block was first initialized.
233 * this is very different than the first initialization/creation of a
234 * superblock. we use the special service_priority_operation to make
235 * sure that the mount gets ahead of any other pending operation that
236 * is waiting for servicing. this means that the pvfs2-client won't
237 * fail to start several times for all other pending operations before
238 * the client regains all of the mount information from us.
239 * NOTE: this function assumes that the request_mutex is already acquired!
241 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
243 struct orangefs_kernel_op_s *new_op;
246 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
248 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
251 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
252 orangefs_sb->devname,
253 ORANGEFS_MAX_SERVER_ADDR_LEN);
255 gossip_debug(GOSSIP_SUPER_DEBUG,
256 "Attempting ORANGEFS Remount via host %s\n",
257 new_op->upcall.req.fs_mount.orangefs_config_server);
260 * we assume that the calling function has already acquired the
261 * request_mutex to prevent other operations from bypassing
264 ret = service_operation(new_op, "orangefs_remount",
265 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
266 gossip_debug(GOSSIP_SUPER_DEBUG,
267 "orangefs_remount: mount got return value of %d\n",
271 * store the id assigned to this sb -- it's just a
272 * short-lived mapping that the system interface uses
273 * to map this superblock to a particular mount entry
275 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
276 orangefs_sb->mount_pending = 0;
281 if (orangefs_userspace_version >= 20906) {
282 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
285 new_op->upcall.req.features.features = 0;
286 ret = service_operation(new_op, "orangefs_features",
287 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
290 new_op->downcall.resp.features.features;
292 orangefs_features = 0;
295 orangefs_features = 0;
301 int fsid_key_table_initialize(void)
306 void fsid_key_table_finalize(void)
310 static const struct super_operations orangefs_s_ops = {
311 .alloc_inode = orangefs_alloc_inode,
312 .destroy_inode = orangefs_destroy_inode,
313 .drop_inode = generic_delete_inode,
314 .statfs = orangefs_statfs,
315 .remount_fs = orangefs_remount_fs,
316 .show_options = orangefs_show_options,
319 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
324 struct orangefs_object_kref refn;
326 if (fh_len < 5 || fh_type > 2)
329 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
330 refn.fs_id = (u32) fid->raw[4];
331 gossip_debug(GOSSIP_SUPER_DEBUG,
332 "fh_to_dentry: handle %pU, fs_id %d\n",
336 return d_obtain_alias(orangefs_iget(sb, &refn));
339 static int orangefs_encode_fh(struct inode *inode,
342 struct inode *parent)
344 int len = parent ? 10 : 5;
346 struct orangefs_object_kref refn;
348 if (*max_len < len) {
349 gossip_err("fh buffer is too small for encoding\n");
355 refn = ORANGEFS_I(inode)->refn;
356 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
359 gossip_debug(GOSSIP_SUPER_DEBUG,
360 "Encoding fh: handle %pU, fsid %u\n",
366 refn = ORANGEFS_I(parent)->refn;
367 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
371 gossip_debug(GOSSIP_SUPER_DEBUG,
372 "Encoding parent: handle %pU, fsid %u\n",
382 static const struct export_operations orangefs_export_ops = {
383 .encode_fh = orangefs_encode_fh,
384 .fh_to_dentry = orangefs_fh_to_dentry,
387 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
389 struct orangefs_kernel_op_s *op;
391 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
394 op->upcall.req.fs_umount.id = id;
395 op->upcall.req.fs_umount.fs_id = fs_id;
396 strncpy(op->upcall.req.fs_umount.orangefs_config_server,
397 devname, ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
398 r = service_operation(op, "orangefs_fs_umount", 0);
399 /* Not much to do about an error here. */
401 gossip_err("orangefs_unmount: service_operation %d\n", r);
406 static int orangefs_fill_sb(struct super_block *sb,
407 struct orangefs_fs_mount_response *fs_mount,
408 void *data, int silent)
411 struct inode *root = NULL;
412 struct dentry *root_dentry = NULL;
413 struct orangefs_object_kref root_object;
415 /* alloc and init our private orangefs sb info */
416 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
417 if (!ORANGEFS_SB(sb))
419 ORANGEFS_SB(sb)->sb = sb;
421 ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle;
422 ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id;
423 ORANGEFS_SB(sb)->id = fs_mount->id;
426 ret = parse_mount_options(sb, data, silent);
431 /* Hang the xattr handlers off the superblock */
432 sb->s_xattr = orangefs_xattr_handlers;
433 sb->s_magic = ORANGEFS_SUPER_MAGIC;
434 sb->s_op = &orangefs_s_ops;
435 sb->s_d_op = &orangefs_dentry_operations;
437 sb->s_blocksize = PAGE_SIZE;
438 sb->s_blocksize_bits = PAGE_SHIFT;
439 sb->s_maxbytes = MAX_LFS_FILESIZE;
441 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
442 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
443 gossip_debug(GOSSIP_SUPER_DEBUG,
444 "get inode %pU, fsid %d\n",
445 &root_object.khandle,
448 root = orangefs_iget(sb, &root_object);
450 return PTR_ERR(root);
452 gossip_debug(GOSSIP_SUPER_DEBUG,
453 "Allocated root inode [%p] with mode %x\n",
457 /* allocates and places root dentry in dcache */
458 root_dentry = d_make_root(root);
462 sb->s_export_op = &orangefs_export_ops;
463 sb->s_root = root_dentry;
467 struct dentry *orangefs_mount(struct file_system_type *fst,
473 struct super_block *sb = ERR_PTR(-EINVAL);
474 struct orangefs_kernel_op_s *new_op;
475 struct dentry *d = ERR_PTR(-EINVAL);
477 gossip_debug(GOSSIP_SUPER_DEBUG,
478 "orangefs_mount: called with devname %s\n",
482 gossip_err("ERROR: device name not specified.\n");
483 return ERR_PTR(-EINVAL);
486 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
488 return ERR_PTR(-ENOMEM);
490 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
492 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
494 gossip_debug(GOSSIP_SUPER_DEBUG,
495 "Attempting ORANGEFS Mount via host %s\n",
496 new_op->upcall.req.fs_mount.orangefs_config_server);
498 ret = service_operation(new_op, "orangefs_mount", 0);
499 gossip_debug(GOSSIP_SUPER_DEBUG,
500 "orangefs_mount: mount got return value of %d\n", ret);
504 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
505 gossip_err("ERROR: Retrieved null fs_id\n");
510 sb = sget(fst, NULL, set_anon_super, flags, NULL);
514 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
515 new_op->downcall.resp.fs_mount.fs_id, devname);
519 ret = orangefs_fill_sb(sb,
520 &new_op->downcall.resp.fs_mount, data,
521 flags & SB_SILENT ? 1 : 0);
529 * on successful mount, store the devname and data
532 strncpy(ORANGEFS_SB(sb)->devname,
534 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
536 /* mount_pending must be cleared */
537 ORANGEFS_SB(sb)->mount_pending = 0;
540 * finally, add this sb to our list of known orangefs
543 gossip_debug(GOSSIP_SUPER_DEBUG,
544 "Adding SB %p to orangefs superblocks\n",
546 spin_lock(&orangefs_superblocks_lock);
547 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
548 spin_unlock(&orangefs_superblocks_lock);
551 /* Must be removed from the list now. */
552 ORANGEFS_SB(sb)->no_list = 0;
554 if (orangefs_userspace_version >= 20906) {
555 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
557 return ERR_PTR(-ENOMEM);
558 new_op->upcall.req.features.features = 0;
559 ret = service_operation(new_op, "orangefs_features", 0);
560 orangefs_features = new_op->downcall.resp.features.features;
563 orangefs_features = 0;
566 return dget(sb->s_root);
569 /* Will call orangefs_kill_sb with sb not in list. */
570 ORANGEFS_SB(sb)->no_list = 1;
571 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
572 deactivate_locked_super(sb);
574 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
575 if (ret == -EINVAL) {
576 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
577 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
585 void orangefs_kill_sb(struct super_block *sb)
588 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
590 /* provided sb cleanup */
593 if (!ORANGEFS_SB(sb)) {
594 mutex_lock(&orangefs_request_mutex);
595 mutex_unlock(&orangefs_request_mutex);
599 * issue the unmount to userspace to tell it to remove the
600 * dynamic mount info it has for this superblock
602 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
603 ORANGEFS_SB(sb)->devname);
605 ORANGEFS_SB(sb)->mount_pending = 1;
607 if (!ORANGEFS_SB(sb)->no_list) {
608 /* remove the sb from our list of orangefs specific sb's */
609 spin_lock(&orangefs_superblocks_lock);
610 /* not list_del_init */
611 __list_del_entry(&ORANGEFS_SB(sb)->list);
612 ORANGEFS_SB(sb)->list.prev = NULL;
613 spin_unlock(&orangefs_superblocks_lock);
617 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
618 * gets completed before we free the dang thing.
620 mutex_lock(&orangefs_request_mutex);
621 mutex_unlock(&orangefs_request_mutex);
623 /* free the orangefs superblock private data */
624 kfree(ORANGEFS_SB(sb));
627 int orangefs_inode_cache_initialize(void)
629 orangefs_inode_cache = kmem_cache_create_usercopy(
630 "orangefs_inode_cache",
631 sizeof(struct orangefs_inode_s),
633 ORANGEFS_CACHE_CREATE_FLAGS,
634 offsetof(struct orangefs_inode_s,
636 sizeof_field(struct orangefs_inode_s,
638 orangefs_inode_cache_ctor);
640 if (!orangefs_inode_cache) {
641 gossip_err("Cannot create orangefs_inode_cache\n");
647 int orangefs_inode_cache_finalize(void)
649 kmem_cache_destroy(orangefs_inode_cache);