Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-stable.git] / fs / orangefs / super.c
blobeba3e357192ee657511918aefb5c803a31833cae
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * (C) 2001 Clemson University and The University of Chicago
5 * See COPYING in top-level directory.
6 */
8 #include "protocol.h"
9 #include "orangefs-kernel.h"
10 #include "orangefs-bufmap.h"
12 #include <linux/parser.h>
13 #include <linux/hashtable.h>
14 #include <linux/seq_file.h>
16 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
17 static struct kmem_cache *orangefs_inode_cache;
19 /* list for storing orangefs specific superblocks in use */
20 LIST_HEAD(orangefs_superblocks);
22 DEFINE_SPINLOCK(orangefs_superblocks_lock);
24 enum {
25 Opt_intr,
26 Opt_acl,
27 Opt_local_lock,
29 Opt_err
32 static const match_table_t tokens = {
33 { Opt_acl, "acl" },
34 { Opt_intr, "intr" },
35 { Opt_local_lock, "local_lock" },
36 { Opt_err, NULL }
39 uint64_t orangefs_features;
41 static int orangefs_show_options(struct seq_file *m, struct dentry *root)
43 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(root->d_sb);
45 if (root->d_sb->s_flags & SB_POSIXACL)
46 seq_puts(m, ",acl");
47 if (orangefs_sb->flags & ORANGEFS_OPT_INTR)
48 seq_puts(m, ",intr");
49 if (orangefs_sb->flags & ORANGEFS_OPT_LOCAL_LOCK)
50 seq_puts(m, ",local_lock");
51 return 0;
54 static int parse_mount_options(struct super_block *sb, char *options,
55 int silent)
57 struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(sb);
58 substring_t args[MAX_OPT_ARGS];
59 char *p;
62 * Force any potential flags that might be set from the mount
63 * to zero, ie, initialize to unset.
65 sb->s_flags &= ~SB_POSIXACL;
66 orangefs_sb->flags &= ~ORANGEFS_OPT_INTR;
67 orangefs_sb->flags &= ~ORANGEFS_OPT_LOCAL_LOCK;
69 while ((p = strsep(&options, ",")) != NULL) {
70 int token;
72 if (!*p)
73 continue;
75 token = match_token(p, tokens, args);
76 switch (token) {
77 case Opt_acl:
78 sb->s_flags |= SB_POSIXACL;
79 break;
80 case Opt_intr:
81 orangefs_sb->flags |= ORANGEFS_OPT_INTR;
82 break;
83 case Opt_local_lock:
84 orangefs_sb->flags |= ORANGEFS_OPT_LOCAL_LOCK;
85 break;
86 default:
87 goto fail;
91 return 0;
92 fail:
93 if (!silent)
94 gossip_err("Error: mount option [%s] is not supported.\n", p);
95 return -EINVAL;
98 static void orangefs_inode_cache_ctor(void *req)
100 struct orangefs_inode_s *orangefs_inode = req;
102 inode_init_once(&orangefs_inode->vfs_inode);
103 init_rwsem(&orangefs_inode->xattr_sem);
106 static struct inode *orangefs_alloc_inode(struct super_block *sb)
108 struct orangefs_inode_s *orangefs_inode;
110 orangefs_inode = alloc_inode_sb(sb, orangefs_inode_cache, GFP_KERNEL);
111 if (!orangefs_inode)
112 return NULL;
115 * We want to clear everything except for rw_semaphore and the
116 * vfs_inode.
118 memset(&orangefs_inode->refn.khandle, 0, 16);
119 orangefs_inode->refn.fs_id = ORANGEFS_FS_ID_NULL;
120 orangefs_inode->last_failed_block_index_read = 0;
121 memset(orangefs_inode->link_target, 0, sizeof(orangefs_inode->link_target));
123 gossip_debug(GOSSIP_SUPER_DEBUG,
124 "orangefs_alloc_inode: allocated %p\n",
125 &orangefs_inode->vfs_inode);
126 return &orangefs_inode->vfs_inode;
129 static void orangefs_free_inode(struct inode *inode)
131 struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
132 struct orangefs_cached_xattr *cx;
133 struct hlist_node *tmp;
134 int i;
136 hash_for_each_safe(orangefs_inode->xattr_cache, i, tmp, cx, node) {
137 hlist_del(&cx->node);
138 kfree(cx);
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));
153 static int orangefs_write_inode(struct inode *inode,
154 struct writeback_control *wbc)
156 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_write_inode\n");
157 return orangefs_inode_setattr(inode);
161 * NOTE: information filled in here is typically reflected in the
162 * output of the system command 'df'
164 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
166 int ret = -ENOMEM;
167 struct orangefs_kernel_op_s *new_op = NULL;
168 int flags = 0;
169 struct super_block *sb = NULL;
171 sb = dentry->d_sb;
173 gossip_debug(GOSSIP_SUPER_DEBUG,
174 "%s: called on sb %p (fs_id is %d)\n",
175 __func__,
177 (int)(ORANGEFS_SB(sb)->fs_id));
179 new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
180 if (!new_op)
181 return ret;
182 new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
184 if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
185 flags = ORANGEFS_OP_INTERRUPTIBLE;
187 ret = service_operation(new_op, "orangefs_statfs", flags);
189 if (new_op->downcall.status < 0)
190 goto out_op_release;
192 gossip_debug(GOSSIP_SUPER_DEBUG,
193 "%s: got %ld blocks available | "
194 "%ld blocks total | %ld block size | "
195 "%ld files total | %ld files avail\n",
196 __func__,
197 (long)new_op->downcall.resp.statfs.blocks_avail,
198 (long)new_op->downcall.resp.statfs.blocks_total,
199 (long)new_op->downcall.resp.statfs.block_size,
200 (long)new_op->downcall.resp.statfs.files_total,
201 (long)new_op->downcall.resp.statfs.files_avail);
203 buf->f_type = sb->s_magic;
204 buf->f_fsid.val[0] = ORANGEFS_SB(sb)->fs_id;
205 buf->f_fsid.val[1] = ORANGEFS_SB(sb)->id;
206 buf->f_bsize = new_op->downcall.resp.statfs.block_size;
207 buf->f_namelen = ORANGEFS_NAME_MAX;
209 buf->f_blocks = (sector_t) new_op->downcall.resp.statfs.blocks_total;
210 buf->f_bfree = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
211 buf->f_bavail = (sector_t) new_op->downcall.resp.statfs.blocks_avail;
212 buf->f_files = (sector_t) new_op->downcall.resp.statfs.files_total;
213 buf->f_ffree = (sector_t) new_op->downcall.resp.statfs.files_avail;
214 buf->f_frsize = 0;
216 out_op_release:
217 op_release(new_op);
218 gossip_debug(GOSSIP_SUPER_DEBUG, "%s: returning %d\n", __func__, ret);
219 return ret;
223 * Remount as initiated by VFS layer. We just need to reparse the mount
224 * options, no need to signal pvfs2-client-core about it.
226 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
228 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
229 return parse_mount_options(sb, data, 1);
233 * Remount as initiated by pvfs2-client-core on restart. This is used to
234 * repopulate mount information left from previous pvfs2-client-core.
236 * the idea here is that given a valid superblock, we're
237 * re-initializing the user space client with the initial mount
238 * information specified when the super block was first initialized.
239 * this is very different than the first initialization/creation of a
240 * superblock. we use the special service_priority_operation to make
241 * sure that the mount gets ahead of any other pending operation that
242 * is waiting for servicing. this means that the pvfs2-client won't
243 * fail to start several times for all other pending operations before
244 * the client regains all of the mount information from us.
245 * NOTE: this function assumes that the request_mutex is already acquired!
247 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
249 struct orangefs_kernel_op_s *new_op;
250 int ret = -EINVAL;
252 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
254 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
255 if (!new_op)
256 return -ENOMEM;
257 strscpy(new_op->upcall.req.fs_mount.orangefs_config_server,
258 orangefs_sb->devname);
260 gossip_debug(GOSSIP_SUPER_DEBUG,
261 "Attempting ORANGEFS Remount via host %s\n",
262 new_op->upcall.req.fs_mount.orangefs_config_server);
265 * we assume that the calling function has already acquired the
266 * request_mutex to prevent other operations from bypassing
267 * this one
269 ret = service_operation(new_op, "orangefs_remount",
270 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
271 gossip_debug(GOSSIP_SUPER_DEBUG,
272 "orangefs_remount: mount got return value of %d\n",
273 ret);
274 if (ret == 0) {
276 * store the id assigned to this sb -- it's just a
277 * short-lived mapping that the system interface uses
278 * to map this superblock to a particular mount entry
280 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
281 orangefs_sb->mount_pending = 0;
284 op_release(new_op);
286 if (orangefs_userspace_version >= 20906) {
287 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
288 if (!new_op)
289 return -ENOMEM;
290 new_op->upcall.req.features.features = 0;
291 ret = service_operation(new_op, "orangefs_features",
292 ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
293 if (!ret)
294 orangefs_features =
295 new_op->downcall.resp.features.features;
296 else
297 orangefs_features = 0;
298 op_release(new_op);
299 } else {
300 orangefs_features = 0;
303 return ret;
306 int fsid_key_table_initialize(void)
308 return 0;
311 void fsid_key_table_finalize(void)
315 static const struct super_operations orangefs_s_ops = {
316 .alloc_inode = orangefs_alloc_inode,
317 .free_inode = orangefs_free_inode,
318 .destroy_inode = orangefs_destroy_inode,
319 .write_inode = orangefs_write_inode,
320 .drop_inode = generic_delete_inode,
321 .statfs = orangefs_statfs,
322 .remount_fs = orangefs_remount_fs,
323 .show_options = orangefs_show_options,
326 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
327 struct fid *fid,
328 int fh_len,
329 int fh_type)
331 struct orangefs_object_kref refn;
333 if (fh_len < 5 || fh_type > 2)
334 return NULL;
336 ORANGEFS_khandle_from(&(refn.khandle), fid->raw, 16);
337 refn.fs_id = (u32) fid->raw[4];
338 gossip_debug(GOSSIP_SUPER_DEBUG,
339 "fh_to_dentry: handle %pU, fs_id %d\n",
340 &refn.khandle,
341 refn.fs_id);
343 return d_obtain_alias(orangefs_iget(sb, &refn));
346 static int orangefs_encode_fh(struct inode *inode,
347 __u32 *fh,
348 int *max_len,
349 struct inode *parent)
351 int len = parent ? 10 : 5;
352 int type = 1;
353 struct orangefs_object_kref refn;
355 if (*max_len < len) {
356 gossip_err("fh buffer is too small for encoding\n");
357 *max_len = len;
358 type = 255;
359 goto out;
362 refn = ORANGEFS_I(inode)->refn;
363 ORANGEFS_khandle_to(&refn.khandle, fh, 16);
364 fh[4] = refn.fs_id;
366 gossip_debug(GOSSIP_SUPER_DEBUG,
367 "Encoding fh: handle %pU, fsid %u\n",
368 &refn.khandle,
369 refn.fs_id);
372 if (parent) {
373 refn = ORANGEFS_I(parent)->refn;
374 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
375 fh[9] = refn.fs_id;
377 type = 2;
378 gossip_debug(GOSSIP_SUPER_DEBUG,
379 "Encoding parent: handle %pU, fsid %u\n",
380 &refn.khandle,
381 refn.fs_id);
383 *max_len = len;
385 out:
386 return type;
389 static const struct export_operations orangefs_export_ops = {
390 .encode_fh = orangefs_encode_fh,
391 .fh_to_dentry = orangefs_fh_to_dentry,
394 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
396 struct orangefs_kernel_op_s *op;
397 int r;
398 op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
399 if (!op)
400 return -ENOMEM;
401 op->upcall.req.fs_umount.id = id;
402 op->upcall.req.fs_umount.fs_id = fs_id;
403 strscpy(op->upcall.req.fs_umount.orangefs_config_server, devname);
404 r = service_operation(op, "orangefs_fs_umount", 0);
405 /* Not much to do about an error here. */
406 if (r)
407 gossip_err("orangefs_unmount: service_operation %d\n", r);
408 op_release(op);
409 return r;
412 static int orangefs_fill_sb(struct super_block *sb,
413 struct orangefs_fs_mount_response *fs_mount,
414 void *data, int silent)
416 int ret;
417 struct inode *root;
418 struct dentry *root_dentry;
419 struct orangefs_object_kref root_object;
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;
427 if (data) {
428 ret = parse_mount_options(sb, data, silent);
429 if (ret)
430 return ret;
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 = PAGE_SIZE;
440 sb->s_blocksize_bits = PAGE_SHIFT;
441 sb->s_maxbytes = MAX_LFS_FILESIZE;
443 ret = super_setup_bdi(sb);
444 if (ret)
445 return ret;
447 root_object.khandle = ORANGEFS_SB(sb)->root_khandle;
448 root_object.fs_id = ORANGEFS_SB(sb)->fs_id;
449 gossip_debug(GOSSIP_SUPER_DEBUG,
450 "get inode %pU, fsid %d\n",
451 &root_object.khandle,
452 root_object.fs_id);
454 root = orangefs_iget(sb, &root_object);
455 if (IS_ERR(root))
456 return PTR_ERR(root);
458 gossip_debug(GOSSIP_SUPER_DEBUG,
459 "Allocated root inode [%p] with mode %x\n",
460 root,
461 root->i_mode);
463 /* allocates and places root dentry in dcache */
464 root_dentry = d_make_root(root);
465 if (!root_dentry)
466 return -ENOMEM;
468 sb->s_export_op = &orangefs_export_ops;
469 sb->s_root = root_dentry;
470 return 0;
473 struct dentry *orangefs_mount(struct file_system_type *fst,
474 int flags,
475 const char *devname,
476 void *data)
478 int ret;
479 struct super_block *sb = ERR_PTR(-EINVAL);
480 struct orangefs_kernel_op_s *new_op;
481 struct dentry *d = ERR_PTR(-EINVAL);
483 gossip_debug(GOSSIP_SUPER_DEBUG,
484 "orangefs_mount: called with devname %s\n",
485 devname);
487 if (!devname) {
488 gossip_err("ERROR: device name not specified.\n");
489 return ERR_PTR(-EINVAL);
492 new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
493 if (!new_op)
494 return ERR_PTR(-ENOMEM);
496 strscpy(new_op->upcall.req.fs_mount.orangefs_config_server, devname);
498 gossip_debug(GOSSIP_SUPER_DEBUG,
499 "Attempting ORANGEFS Mount via host %s\n",
500 new_op->upcall.req.fs_mount.orangefs_config_server);
502 ret = service_operation(new_op, "orangefs_mount", 0);
503 gossip_debug(GOSSIP_SUPER_DEBUG,
504 "orangefs_mount: mount got return value of %d\n", ret);
505 if (ret)
506 goto free_op;
508 if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
509 gossip_err("ERROR: Retrieved null fs_id\n");
510 ret = -EINVAL;
511 goto free_op;
514 sb = sget(fst, NULL, set_anon_super, flags, NULL);
516 if (IS_ERR(sb)) {
517 d = ERR_CAST(sb);
518 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
519 new_op->downcall.resp.fs_mount.fs_id, devname);
520 goto free_op;
523 /* alloc and init our private orangefs sb info */
524 sb->s_fs_info = kzalloc(sizeof(struct orangefs_sb_info_s), GFP_KERNEL);
525 if (!ORANGEFS_SB(sb)) {
526 d = ERR_PTR(-ENOMEM);
527 goto free_op;
530 ret = orangefs_fill_sb(sb,
531 &new_op->downcall.resp.fs_mount, data,
532 flags & SB_SILENT ? 1 : 0);
534 if (ret) {
535 d = ERR_PTR(ret);
536 goto free_sb_and_op;
540 * on successful mount, store the devname and data
541 * used
543 strscpy(ORANGEFS_SB(sb)->devname, devname);
546 /* mount_pending must be cleared */
547 ORANGEFS_SB(sb)->mount_pending = 0;
550 * finally, add this sb to our list of known orangefs
551 * sb's
553 gossip_debug(GOSSIP_SUPER_DEBUG,
554 "Adding SB %p to orangefs superblocks\n",
555 ORANGEFS_SB(sb));
556 spin_lock(&orangefs_superblocks_lock);
557 list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
558 spin_unlock(&orangefs_superblocks_lock);
559 op_release(new_op);
561 /* Must be removed from the list now. */
562 ORANGEFS_SB(sb)->no_list = 0;
564 if (orangefs_userspace_version >= 20906) {
565 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
566 if (!new_op)
567 return ERR_PTR(-ENOMEM);
568 new_op->upcall.req.features.features = 0;
569 ret = service_operation(new_op, "orangefs_features", 0);
570 orangefs_features = new_op->downcall.resp.features.features;
571 op_release(new_op);
572 } else {
573 orangefs_features = 0;
576 return dget(sb->s_root);
578 free_sb_and_op:
579 /* Will call orangefs_kill_sb with sb not in list. */
580 ORANGEFS_SB(sb)->no_list = 1;
581 /* ORANGEFS_VFS_OP_FS_UMOUNT is done by orangefs_kill_sb. */
582 deactivate_locked_super(sb);
583 free_op:
584 gossip_err("orangefs_mount: mount request failed with %d\n", ret);
585 if (ret == -EINVAL) {
586 gossip_err("Ensure that all orangefs-servers have the same FS configuration files\n");
587 gossip_err("Look at pvfs2-client-core log file (typically /tmp/pvfs2-client.log) for more details\n");
590 op_release(new_op);
592 return d;
595 void orangefs_kill_sb(struct super_block *sb)
597 int r;
598 gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
600 /* provided sb cleanup */
601 kill_anon_super(sb);
603 if (!ORANGEFS_SB(sb)) {
604 mutex_lock(&orangefs_request_mutex);
605 mutex_unlock(&orangefs_request_mutex);
606 return;
609 * issue the unmount to userspace to tell it to remove the
610 * dynamic mount info it has for this superblock
612 r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
613 ORANGEFS_SB(sb)->devname);
614 if (!r)
615 ORANGEFS_SB(sb)->mount_pending = 1;
617 if (!ORANGEFS_SB(sb)->no_list) {
618 /* remove the sb from our list of orangefs specific sb's */
619 spin_lock(&orangefs_superblocks_lock);
620 /* not list_del_init */
621 __list_del_entry(&ORANGEFS_SB(sb)->list);
622 ORANGEFS_SB(sb)->list.prev = NULL;
623 spin_unlock(&orangefs_superblocks_lock);
627 * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
628 * gets completed before we free the dang thing.
630 mutex_lock(&orangefs_request_mutex);
631 mutex_unlock(&orangefs_request_mutex);
633 /* free the orangefs superblock private data */
634 kfree(ORANGEFS_SB(sb));
637 int orangefs_inode_cache_initialize(void)
639 orangefs_inode_cache = kmem_cache_create_usercopy(
640 "orangefs_inode_cache",
641 sizeof(struct orangefs_inode_s),
644 offsetof(struct orangefs_inode_s,
645 link_target),
646 sizeof_field(struct orangefs_inode_s,
647 link_target),
648 orangefs_inode_cache_ctor);
650 if (!orangefs_inode_cache) {
651 gossip_err("Cannot create orangefs_inode_cache\n");
652 return -ENOMEM;
654 return 0;
657 int orangefs_inode_cache_finalize(void)
659 kmem_cache_destroy(orangefs_inode_cache);
660 return 0;