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_i_callback(struct rcu_head *head)
129 struct inode *inode = container_of(head, struct inode, i_rcu);
130 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
131 kmem_cache_free(orangefs_inode_cache, orangefs_inode);
134 static void orangefs_destroy_inode(struct inode *inode)
136 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
138 gossip_debug(GOSSIP_SUPER_DEBUG,
139 "%s: deallocated %p destroying inode %pU\n",
140 __func__, orangefs_inode, get_khandle_from_ino(inode));
142 call_rcu(&inode->i_rcu, orangefs_i_callback);
146 * NOTE: information filled in here is typically reflected in the
147 * output of the system command 'df'
149 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
152 struct orangefs_kernel_op_s *new_op = NULL;
154 struct super_block *sb = NULL;
158 gossip_debug(GOSSIP_SUPER_DEBUG,
159 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
161 (int)(ORANGEFS_SB(sb)->fs_id));
163 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
166 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
168 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
169 flags = ORANGEFS_OP_INTERRUPTIBLE;
171 ret = service_operation(new_op, "orangefs_statfs", flags);
173 if (new_op->downcall.status < 0)
176 gossip_debug(GOSSIP_SUPER_DEBUG,
177 "%s: got %ld blocks available | "
178 "%ld blocks total | %ld block size | "
179 "%ld files total | %ld files avail\n",
181 (long)new_op->downcall.resp.statfs.blocks_avail,
182 (long)new_op->downcall.resp.statfs.blocks_total,
183 (long)new_op->downcall.resp.statfs.block_size,
184 (long)new_op->downcall.resp.statfs.files_total,
185 (long)new_op->downcall.resp.statfs.files_avail);
187 buf->f_type = sb->s_magic;
188 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
189 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
190 buf->f_namelen = ORANGEFS_NAME_MAX;
192 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
193 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
194 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
195 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
196 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
197 buf->f_frsize = sb->s_blocksize;
201 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_statfs: returning %d\n", ret);
206 * Remount as initiated by VFS layer. We just need to reparse the mount
207 * options, no need to signal pvfs2-client-core about it.
209 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
211 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
212 return parse_mount_options(sb, data, 1);
216 * Remount as initiated by pvfs2-client-core on restart. This is used to
217 * repopulate mount information left from previous pvfs2-client-core.
219 * the idea here is that given a valid superblock, we're
220 * re-initializing the user space client with the initial mount
221 * information specified when the super block was first initialized.
222 * this is very different than the first initialization/creation of a
223 * superblock. we use the special service_priority_operation to make
224 * sure that the mount gets ahead of any other pending operation that
225 * is waiting for servicing. this means that the pvfs2-client won't
226 * fail to start several times for all other pending operations before
227 * the client regains all of the mount information from us.
228 * NOTE: this function assumes that the request_mutex is already acquired!
230 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
232 struct orangefs_kernel_op_s *new_op;
235 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
237 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
240 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
241 orangefs_sb->devname,
242 ORANGEFS_MAX_SERVER_ADDR_LEN);
244 gossip_debug(GOSSIP_SUPER_DEBUG,
245 "Attempting ORANGEFS Remount via host %s\n",
246 new_op->upcall.req.fs_mount.orangefs_config_server);
249 * we assume that the calling function has already acquired the
250 * request_mutex to prevent other operations from bypassing
253 ret = service_operation(new_op, "orangefs_remount",
254 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
255 gossip_debug(GOSSIP_SUPER_DEBUG,
256 "orangefs_remount: mount got return value of %d\n",
260 * store the id assigned to this sb -- it's just a
261 * short-lived mapping that the system interface uses
262 * to map this superblock to a particular mount entry
264 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
265 orangefs_sb->mount_pending = 0;
270 if (orangefs_userspace_version >= 20906) {
271 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
274 new_op->upcall.req.features.features = 0;
275 ret = service_operation(new_op, "orangefs_features",
276 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
279 new_op->downcall.resp.features.features;
281 orangefs_features = 0;
284 orangefs_features = 0;
290 int fsid_key_table_initialize(void)
295 void fsid_key_table_finalize(void)
299 static const struct super_operations orangefs_s_ops = {
300 .alloc_inode = orangefs_alloc_inode,
301 .destroy_inode = orangefs_destroy_inode,
302 .drop_inode = generic_delete_inode,
303 .statfs = orangefs_statfs,
304 .remount_fs = orangefs_remount_fs,
305 .show_options = orangefs_show_options,
308 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
313 struct orangefs_object_kref refn;
315 if (fh_len < 5 || fh_type > 2)
318 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
319 refn.fs_id = (u32) fid->raw[4];
320 gossip_debug(GOSSIP_SUPER_DEBUG,
321 "fh_to_dentry: handle %pU, fs_id %d\n",
325 return d_obtain_alias(orangefs_iget(sb, &refn));
328 static int orangefs_encode_fh(struct inode *inode,
331 struct inode *parent)
333 int len = parent ? 10 : 5;
335 struct orangefs_object_kref refn;
337 if (*max_len < len) {
338 gossip_err("fh buffer is too small for encoding\n");
344 refn = ORANGEFS_I(inode)->refn;
345 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
348 gossip_debug(GOSSIP_SUPER_DEBUG,
349 "Encoding fh: handle %pU, fsid %u\n",
355 refn = ORANGEFS_I(parent)->refn;
356 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
360 gossip_debug(GOSSIP_SUPER_DEBUG,
361 "Encoding parent: handle %pU, fsid %u\n",
371 static const struct export_operations orangefs_export_ops = {
372 .encode_fh = orangefs_encode_fh,
373 .fh_to_dentry = orangefs_fh_to_dentry,
376 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
378 struct orangefs_kernel_op_s *op;
380 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
383 op->upcall.req.fs_umount.id = id;
384 op->upcall.req.fs_umount.fs_id = fs_id;
385 strncpy(op->upcall.req.fs_umount.orangefs_config_server,
386 devname, ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
387 r = service_operation(op, "orangefs_fs_umount", 0);
388 /* Not much to do about an error here. */
390 gossip_err("orangefs_unmount: service_operation %d\n", r);
395 static int orangefs_fill_sb(struct super_block *sb,
396 struct orangefs_fs_mount_response *fs_mount,
397 void *data, int silent)
400 struct inode *root = NULL;
401 struct dentry *root_dentry = NULL;
402 struct orangefs_object_kref root_object;
404 /* alloc and init our private orangefs sb info */
405 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
406 if (!ORANGEFS_SB(sb))
408 ORANGEFS_SB(sb)->sb = sb;
410 ORANGEFS_SB(sb)->root_khandle = fs_mount->root_khandle;
411 ORANGEFS_SB(sb)->fs_id = fs_mount->fs_id;
412 ORANGEFS_SB(sb)->id = fs_mount->id;
415 ret = parse_mount_options(sb, data, silent);
420 /* Hang the xattr handlers off the superblock */
421 sb->s_xattr = orangefs_xattr_handlers;
422 sb->s_magic = ORANGEFS_SUPER_MAGIC;
423 sb->s_op = &orangefs_s_ops;
424 sb->s_d_op = &orangefs_dentry_operations;
426 sb->s_blocksize = orangefs_bufmap_size_query();
427 sb->s_blocksize_bits = orangefs_bufmap_shift_query();
428 sb->s_maxbytes = MAX_LFS_FILESIZE;
430 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
431 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
432 gossip_debug(GOSSIP_SUPER_DEBUG,
433 "get inode %pU, fsid %d\n",
434 &root_object.khandle,
437 root = orangefs_iget(sb, &root_object);
439 return PTR_ERR(root);
441 gossip_debug(GOSSIP_SUPER_DEBUG,
442 "Allocated root inode [%p] with mode %x\n",
446 /* allocates and places root dentry in dcache */
447 root_dentry = d_make_root(root);
451 sb->s_export_op = &orangefs_export_ops;
452 sb->s_root = root_dentry;
456 struct dentry *orangefs_mount(struct file_system_type *fst,
462 struct super_block *sb = ERR_PTR(-EINVAL);
463 struct orangefs_kernel_op_s *new_op;
464 struct dentry *d = ERR_PTR(-EINVAL);
466 gossip_debug(GOSSIP_SUPER_DEBUG,
467 "orangefs_mount: called with devname %s\n",
471 gossip_err("ERROR: device name not specified.\n");
472 return ERR_PTR(-EINVAL);
475 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
477 return ERR_PTR(-ENOMEM);
479 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
481 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
483 gossip_debug(GOSSIP_SUPER_DEBUG,
484 "Attempting ORANGEFS Mount via host %s\n",
485 new_op->upcall.req.fs_mount.orangefs_config_server);
487 ret = service_operation(new_op, "orangefs_mount", 0);
488 gossip_debug(GOSSIP_SUPER_DEBUG,
489 "orangefs_mount: mount got return value of %d\n", ret);
493 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
494 gossip_err("ERROR: Retrieved null fs_id\n");
499 sb = sget(fst, NULL, set_anon_super, flags, NULL);
503 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
504 new_op->downcall.resp.fs_mount.fs_id, devname);
508 ret = orangefs_fill_sb(sb,
509 &new_op->downcall.resp.fs_mount, data,
510 flags & SB_SILENT ? 1 : 0);
518 * on successful mount, store the devname and data
521 strncpy(ORANGEFS_SB(sb)->devname,
523 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
525 /* mount_pending must be cleared */
526 ORANGEFS_SB(sb)->mount_pending = 0;
529 * finally, add this sb to our list of known orangefs
532 gossip_debug(GOSSIP_SUPER_DEBUG,
533 "Adding SB %p to orangefs superblocks\n",
535 spin_lock(&orangefs_superblocks_lock);
536 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
537 spin_unlock(&orangefs_superblocks_lock);
540 /* Must be removed from the list now. */
541 ORANGEFS_SB(sb)->no_list = 0;
543 if (orangefs_userspace_version >= 20906) {
544 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
546 return ERR_PTR(-ENOMEM);
547 new_op->upcall.req.features.features = 0;
548 ret = service_operation(new_op, "orangefs_features", 0);
549 orangefs_features = new_op->downcall.resp.features.features;
552 orangefs_features = 0;
555 return dget(sb->s_root);
558 /* Will call orangefs_kill_sb with sb not in list. */
559 ORANGEFS_SB(sb)->no_list = 1;
560 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
561 deactivate_locked_super(sb);
563 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
564 if (ret == -EINVAL) {
565 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
566 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
574 void orangefs_kill_sb(struct super_block *sb)
577 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
579 /* provided sb cleanup */
583 * issue the unmount to userspace to tell it to remove the
584 * dynamic mount info it has for this superblock
586 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
587 ORANGEFS_SB(sb)->devname);
589 ORANGEFS_SB(sb)->mount_pending = 1;
591 if (!ORANGEFS_SB(sb)->no_list) {
592 /* remove the sb from our list of orangefs specific sb's */
593 spin_lock(&orangefs_superblocks_lock);
594 /* not list_del_init */
595 __list_del_entry(&ORANGEFS_SB(sb)->list);
596 ORANGEFS_SB(sb)->list.prev = NULL;
597 spin_unlock(&orangefs_superblocks_lock);
601 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
602 * gets completed before we free the dang thing.
604 mutex_lock(&orangefs_request_mutex);
605 mutex_unlock(&orangefs_request_mutex);
607 /* free the orangefs superblock private data */
608 kfree(ORANGEFS_SB(sb));
611 int orangefs_inode_cache_initialize(void)
613 orangefs_inode_cache = kmem_cache_create_usercopy(
614 "orangefs_inode_cache",
615 sizeof(struct orangefs_inode_s),
617 ORANGEFS_CACHE_CREATE_FLAGS,
618 offsetof(struct orangefs_inode_s,
620 sizeof_field(struct orangefs_inode_s,
622 orangefs_inode_cache_ctor);
624 if (!orangefs_inode_cache) {
625 gossip_err("Cannot create orangefs_inode_cache\n");
631 int orangefs_inode_cache_finalize(void)
633 kmem_cache_destroy(orangefs_inode_cache);