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 & MS_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 &= ~MS_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 |= MS_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));
120 orangefs_inode->pinode_flags = 0;
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 kmem_cache_free(orangefs_inode_cache, orangefs_inode);
135 static void orangefs_destroy_inode(struct inode *inode)
137 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
139 gossip_debug(GOSSIP_SUPER_DEBUG,
140 "%s: deallocated %p destroying inode %pU\n",
141 __func__, orangefs_inode, get_khandle_from_ino(inode));
143 call_rcu(&inode->i_rcu, orangefs_i_callback);
147 * NOTE: information filled in here is typically reflected in the
148 * output of the system command 'df'
150 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
153 struct orangefs_kernel_op_s *new_op = NULL;
155 struct super_block *sb = NULL;
159 gossip_debug(GOSSIP_SUPER_DEBUG,
160 "orangefs_statfs: called on sb %p (fs_id is %d)\n",
162 (int)(ORANGEFS_SB(sb)->fs_id));
164 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
167 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
169 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
170 flags = ORANGEFS_OP_INTERRUPTIBLE;
172 ret = service_operation(new_op, "orangefs_statfs", flags);
174 if (new_op->downcall.status < 0)
177 gossip_debug(GOSSIP_SUPER_DEBUG,
178 "%s: got %ld blocks available | "
179 "%ld blocks total | %ld block size | "
180 "%ld files total | %ld files avail\n",
182 (long)new_op->downcall.resp.statfs.blocks_avail,
183 (long)new_op->downcall.resp.statfs.blocks_total,
184 (long)new_op->downcall.resp.statfs.block_size,
185 (long)new_op->downcall.resp.statfs.files_total,
186 (long)new_op->downcall.resp.statfs.files_avail);
188 buf->f_type = sb->s_magic;
189 memcpy(&buf->f_fsid, &ORANGEFS_SB(sb)->fs_id, sizeof(buf->f_fsid));
190 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
191 buf->f_namelen = ORANGEFS_NAME_MAX;
193 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
194 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
195 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
196 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
197 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
198 buf->f_frsize = sb->s_blocksize;
202 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_statfs: returning %d\n", ret);
207 * Remount as initiated by VFS layer. We just need to reparse the mount
208 * options, no need to signal pvfs2-client-core about it.
210 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
212 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
213 return parse_mount_options(sb, data, 1);
217 * Remount as initiated by pvfs2-client-core on restart. This is used to
218 * repopulate mount information left from previous pvfs2-client-core.
220 * the idea here is that given a valid superblock, we're
221 * re-initializing the user space client with the initial mount
222 * information specified when the super block was first initialized.
223 * this is very different than the first initialization/creation of a
224 * superblock. we use the special service_priority_operation to make
225 * sure that the mount gets ahead of any other pending operation that
226 * is waiting for servicing. this means that the pvfs2-client won't
227 * fail to start several times for all other pending operations before
228 * the client regains all of the mount information from us.
229 * NOTE: this function assumes that the request_mutex is already acquired!
231 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
233 struct orangefs_kernel_op_s *new_op;
236 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
238 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
241 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
242 orangefs_sb->devname,
243 ORANGEFS_MAX_SERVER_ADDR_LEN);
245 gossip_debug(GOSSIP_SUPER_DEBUG,
246 "Attempting ORANGEFS Remount via host %s\n",
247 new_op->upcall.req.fs_mount.orangefs_config_server);
250 * we assume that the calling function has already acquired the
251 * request_mutex to prevent other operations from bypassing
254 ret = service_operation(new_op, "orangefs_remount",
255 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
256 gossip_debug(GOSSIP_SUPER_DEBUG,
257 "orangefs_remount: mount got return value of %d\n",
261 * store the id assigned to this sb -- it's just a
262 * short-lived mapping that the system interface uses
263 * to map this superblock to a particular mount entry
265 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
266 orangefs_sb->mount_pending = 0;
271 if (orangefs_userspace_version >= 20906) {
272 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
275 new_op->upcall.req.features.features = 0;
276 ret = service_operation(new_op, "orangefs_features",
277 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
280 new_op->downcall.resp.features.features;
282 orangefs_features = 0;
285 orangefs_features = 0;
291 int fsid_key_table_initialize(void)
296 void fsid_key_table_finalize(void)
300 /* Called whenever the VFS dirties the inode in response to atime updates */
301 static void orangefs_dirty_inode(struct inode *inode, int flags)
303 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
305 gossip_debug(GOSSIP_SUPER_DEBUG,
306 "orangefs_dirty_inode: %pU\n",
307 get_khandle_from_ino(inode));
308 SetAtimeFlag(orangefs_inode);
311 static const struct super_operations orangefs_s_ops = {
312 .alloc_inode = orangefs_alloc_inode,
313 .destroy_inode = orangefs_destroy_inode,
314 .dirty_inode = orangefs_dirty_inode,
315 .drop_inode = generic_delete_inode,
316 .statfs = orangefs_statfs,
317 .remount_fs = orangefs_remount_fs,
318 .show_options = orangefs_show_options,
321 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
326 struct orangefs_object_kref refn;
328 if (fh_len < 5 || fh_type > 2)
331 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
332 refn.fs_id = (u32) fid->raw[4];
333 gossip_debug(GOSSIP_SUPER_DEBUG,
334 "fh_to_dentry: handle %pU, fs_id %d\n",
338 return d_obtain_alias(orangefs_iget(sb, &refn));
341 static int orangefs_encode_fh(struct inode *inode,
344 struct inode *parent)
346 int len = parent ? 10 : 5;
348 struct orangefs_object_kref refn;
350 if (*max_len < len) {
351 gossip_lerr("fh buffer is too small for encoding\n");
357 refn = ORANGEFS_I(inode)->refn;
358 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
361 gossip_debug(GOSSIP_SUPER_DEBUG,
362 "Encoding fh: handle %pU, fsid %u\n",
368 refn = ORANGEFS_I(parent)->refn;
369 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
373 gossip_debug(GOSSIP_SUPER_DEBUG,
374 "Encoding parent: handle %pU, fsid %u\n",
384 static const struct export_operations orangefs_export_ops = {
385 .encode_fh = orangefs_encode_fh,
386 .fh_to_dentry = orangefs_fh_to_dentry,
389 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
391 struct orangefs_kernel_op_s *op;
393 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
396 op->upcall.req.fs_umount.id = id;
397 op->upcall.req.fs_umount.fs_id = fs_id;
398 strncpy(op->upcall.req.fs_umount.orangefs_config_server,
399 devname, ORANGEFS_MAX_SERVER_ADDR_LEN);
400 r = service_operation(op, "orangefs_fs_umount", 0);
401 /* Not much to do about an error here. */
403 gossip_err("orangefs_unmount: service_operation %d\n", r);
408 static int orangefs_fill_sb(struct super_block *sb,
409 struct orangefs_fs_mount_response *fs_mount,
410 void *data, int silent)
413 struct inode *root = NULL;
414 struct dentry *root_dentry = NULL;
415 struct orangefs_object_kref root_object;
417 /* alloc and init our private orangefs sb info */
418 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
419 if (!ORANGEFS_SB(sb))
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 = orangefs_bufmap_size_query();
440 sb->s_blocksize_bits = orangefs_bufmap_shift_query();
441 sb->s_maxbytes = MAX_LFS_FILESIZE;
443 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
444 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
445 gossip_debug(GOSSIP_SUPER_DEBUG,
446 "get inode %pU, fsid %d\n",
447 &root_object.khandle,
450 root = orangefs_iget(sb, &root_object);
452 return PTR_ERR(root);
454 gossip_debug(GOSSIP_SUPER_DEBUG,
455 "Allocated root inode [%p] with mode %x\n",
459 /* allocates and places root dentry in dcache */
460 root_dentry = d_make_root(root);
464 sb->s_export_op = &orangefs_export_ops;
465 sb->s_root = root_dentry;
469 struct dentry *orangefs_mount(struct file_system_type *fst,
475 struct super_block *sb = ERR_PTR(-EINVAL);
476 struct orangefs_kernel_op_s *new_op;
477 struct dentry *d = ERR_PTR(-EINVAL);
479 gossip_debug(GOSSIP_SUPER_DEBUG,
480 "orangefs_mount: called with devname %s\n",
484 gossip_err("ERROR: device name not specified.\n");
485 return ERR_PTR(-EINVAL);
488 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
490 return ERR_PTR(-ENOMEM);
492 strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
494 ORANGEFS_MAX_SERVER_ADDR_LEN);
496 gossip_debug(GOSSIP_SUPER_DEBUG,
497 "Attempting ORANGEFS Mount via host %s\n",
498 new_op->upcall.req.fs_mount.orangefs_config_server);
500 ret = service_operation(new_op, "orangefs_mount", 0);
501 gossip_debug(GOSSIP_SUPER_DEBUG,
502 "orangefs_mount: mount got return value of %d\n", ret);
506 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
507 gossip_err("ERROR: Retrieved null fs_id\n");
512 sb = sget(fst, NULL, set_anon_super, flags, NULL);
516 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
517 new_op->downcall.resp.fs_mount.fs_id, devname);
521 ret = orangefs_fill_sb(sb,
522 &new_op->downcall.resp.fs_mount, data,
523 flags & MS_SILENT ? 1 : 0);
531 * on successful mount, store the devname and data
534 strncpy(ORANGEFS_SB(sb)->devname,
536 ORANGEFS_MAX_SERVER_ADDR_LEN);
538 /* mount_pending must be cleared */
539 ORANGEFS_SB(sb)->mount_pending = 0;
542 * finally, add this sb to our list of known orangefs
545 gossip_debug(GOSSIP_SUPER_DEBUG,
546 "Adding SB %p to orangefs superblocks\n",
548 spin_lock(&orangefs_superblocks_lock);
549 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
550 spin_unlock(&orangefs_superblocks_lock);
553 /* Must be removed from the list now. */
554 ORANGEFS_SB(sb)->no_list = 0;
556 if (orangefs_userspace_version >= 20906) {
557 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
559 return ERR_PTR(-ENOMEM);
560 new_op->upcall.req.features.features = 0;
561 ret = service_operation(new_op, "orangefs_features", 0);
562 orangefs_features = new_op->downcall.resp.features.features;
565 orangefs_features = 0;
568 return dget(sb->s_root);
571 /* Will call orangefs_kill_sb with sb not in list. */
572 ORANGEFS_SB(sb)->no_list = 1;
573 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
574 deactivate_locked_super(sb);
576 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
577 if (ret == -EINVAL) {
578 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
579 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
587 void orangefs_kill_sb(struct super_block *sb)
590 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
592 /* provided sb cleanup */
596 * issue the unmount to userspace to tell it to remove the
597 * dynamic mount info it has for this superblock
599 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
600 ORANGEFS_SB(sb)->devname);
602 ORANGEFS_SB(sb)->mount_pending = 1;
604 if (!ORANGEFS_SB(sb)->no_list) {
605 /* remove the sb from our list of orangefs specific sb's */
606 spin_lock(&orangefs_superblocks_lock);
607 /* not list_del_init */
608 __list_del_entry(&ORANGEFS_SB(sb)->list);
609 ORANGEFS_SB(sb)->list.prev = NULL;
610 spin_unlock(&orangefs_superblocks_lock);
614 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
615 * gets completed before we free the dang thing.
617 mutex_lock(&orangefs_request_mutex);
618 mutex_unlock(&orangefs_request_mutex);
620 /* free the orangefs superblock private data */
621 kfree(ORANGEFS_SB(sb));
624 int orangefs_inode_cache_initialize(void)
626 orangefs_inode_cache = kmem_cache_create("orangefs_inode_cache",
627 sizeof(struct orangefs_inode_s),
629 ORANGEFS_CACHE_CREATE_FLAGS,
630 orangefs_inode_cache_ctor);
632 if (!orangefs_inode_cache) {
633 gossip_err("Cannot create orangefs_inode_cache\n");
639 int orangefs_inode_cache_finalize(void)
641 kmem_cache_destroy(orangefs_inode_cache);