Merge tag 'kvm-4.16-1' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-2.6-block.git] / fs / orangefs / super.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * (C) 2001 Clemson University and The University of Chicago
4  *
5  * See COPYING in top-level directory.
6  */
7
8 #include "protocol.h"
9 #include "orangefs-kernel.h"
10 #include "orangefs-bufmap.h"
11
12 #include <linux/parser.h>
13
14 /* a cache for orangefs-inode objects (i.e. orangefs inode private data) */
15 static struct kmem_cache *orangefs_inode_cache;
16
17 /* list for storing orangefs specific superblocks in use */
18 LIST_HEAD(orangefs_superblocks);
19
20 DEFINE_SPINLOCK(orangefs_superblocks_lock);
21
22 enum {
23         Opt_intr,
24         Opt_acl,
25         Opt_local_lock,
26
27         Opt_err
28 };
29
30 static const match_table_t tokens = {
31         { Opt_acl,              "acl" },
32         { Opt_intr,             "intr" },
33         { Opt_local_lock,       "local_lock" },
34         { Opt_err,      NULL }
35 };
36
37 uint64_t orangefs_features;
38
39 static int orangefs_show_options(struct seq_file *m, struct dentry *root)
40 {
41         struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(root->d_sb);
42
43         if (root->d_sb->s_flags & SB_POSIXACL)
44                 seq_puts(m, ",acl");
45         if (orangefs_sb->flags & ORANGEFS_OPT_INTR)
46                 seq_puts(m, ",intr");
47         if (orangefs_sb->flags & ORANGEFS_OPT_LOCAL_LOCK)
48                 seq_puts(m, ",local_lock");
49         return 0;
50 }
51
52 static int parse_mount_options(struct super_block *sb, char *options,
53                 int silent)
54 {
55         struct orangefs_sb_info_s *orangefs_sb = ORANGEFS_SB(sb);
56         substring_t args[MAX_OPT_ARGS];
57         char *p;
58
59         /*
60          * Force any potential flags that might be set from the mount
61          * to zero, ie, initialize to unset.
62          */
63         sb->s_flags &= ~SB_POSIXACL;
64         orangefs_sb->flags &= ~ORANGEFS_OPT_INTR;
65         orangefs_sb->flags &= ~ORANGEFS_OPT_LOCAL_LOCK;
66
67         while ((p = strsep(&options, ",")) != NULL) {
68                 int token;
69
70                 if (!*p)
71                         continue;
72
73                 token = match_token(p, tokens, args);
74                 switch (token) {
75                 case Opt_acl:
76                         sb->s_flags |= SB_POSIXACL;
77                         break;
78                 case Opt_intr:
79                         orangefs_sb->flags |= ORANGEFS_OPT_INTR;
80                         break;
81                 case Opt_local_lock:
82                         orangefs_sb->flags |= ORANGEFS_OPT_LOCAL_LOCK;
83                         break;
84                 default:
85                         goto fail;
86                 }
87         }
88
89         return 0;
90 fail:
91         if (!silent)
92                 gossip_err("Error: mount option [%s] is not supported.\n", p);
93         return -EINVAL;
94 }
95
96 static void orangefs_inode_cache_ctor(void *req)
97 {
98         struct orangefs_inode_s *orangefs_inode = req;
99
100         inode_init_once(&orangefs_inode->vfs_inode);
101         init_rwsem(&orangefs_inode->xattr_sem);
102 }
103
104 static struct inode *orangefs_alloc_inode(struct super_block *sb)
105 {
106         struct orangefs_inode_s *orangefs_inode;
107
108         orangefs_inode = kmem_cache_alloc(orangefs_inode_cache, GFP_KERNEL);
109         if (!orangefs_inode)
110                 return NULL;
111
112         /*
113          * We want to clear everything except for rw_semaphore and the
114          * vfs_inode.
115          */
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
121         gossip_debug(GOSSIP_SUPER_DEBUG,
122                      "orangefs_alloc_inode: allocated %p\n",
123                      &orangefs_inode->vfs_inode);
124         return &orangefs_inode->vfs_inode;
125 }
126
127 static void orangefs_i_callback(struct rcu_head *head)
128 {
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);
132 }
133
134 static void orangefs_destroy_inode(struct inode *inode)
135 {
136         struct orangefs_inode_s *orangefs_inode = ORANGEFS_I(inode);
137
138         gossip_debug(GOSSIP_SUPER_DEBUG,
139                         "%s: deallocated %p destroying inode %pU\n",
140                         __func__, orangefs_inode, get_khandle_from_ino(inode));
141
142         call_rcu(&inode->i_rcu, orangefs_i_callback);
143 }
144
145 /*
146  * NOTE: information filled in here is typically reflected in the
147  * output of the system command 'df'
148 */
149 static int orangefs_statfs(struct dentry *dentry, struct kstatfs *buf)
150 {
151         int ret = -ENOMEM;
152         struct orangefs_kernel_op_s *new_op = NULL;
153         int flags = 0;
154         struct super_block *sb = NULL;
155
156         sb = dentry->d_sb;
157
158         gossip_debug(GOSSIP_SUPER_DEBUG,
159                      "orangefs_statfs: called on sb %p (fs_id is %d)\n",
160                      sb,
161                      (int)(ORANGEFS_SB(sb)->fs_id));
162
163         new_op = op_alloc(ORANGEFS_VFS_OP_STATFS);
164         if (!new_op)
165                 return ret;
166         new_op->upcall.req.statfs.fs_id = ORANGEFS_SB(sb)->fs_id;
167
168         if (ORANGEFS_SB(sb)->flags & ORANGEFS_OPT_INTR)
169                 flags = ORANGEFS_OP_INTERRUPTIBLE;
170
171         ret = service_operation(new_op, "orangefs_statfs", flags);
172
173         if (new_op->downcall.status < 0)
174                 goto out_op_release;
175
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",
180                      __func__,
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);
186
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;
191
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;
198
199 out_op_release:
200         op_release(new_op);
201         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_statfs: returning %d\n", ret);
202         return ret;
203 }
204
205 /*
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.
208  */
209 static int orangefs_remount_fs(struct super_block *sb, int *flags, char *data)
210 {
211         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount_fs: called\n");
212         return parse_mount_options(sb, data, 1);
213 }
214
215 /*
216  * Remount as initiated by pvfs2-client-core on restart.  This is used to
217  * repopulate mount information left from previous pvfs2-client-core.
218  *
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!
229  */
230 int orangefs_remount(struct orangefs_sb_info_s *orangefs_sb)
231 {
232         struct orangefs_kernel_op_s *new_op;
233         int ret = -EINVAL;
234
235         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_remount: called\n");
236
237         new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
238         if (!new_op)
239                 return -ENOMEM;
240         strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
241                 orangefs_sb->devname,
242                 ORANGEFS_MAX_SERVER_ADDR_LEN);
243
244         gossip_debug(GOSSIP_SUPER_DEBUG,
245                      "Attempting ORANGEFS Remount via host %s\n",
246                      new_op->upcall.req.fs_mount.orangefs_config_server);
247
248         /*
249          * we assume that the calling function has already acquired the
250          * request_mutex to prevent other operations from bypassing
251          * this one
252          */
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",
257                      ret);
258         if (ret == 0) {
259                 /*
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
263                  */
264                 orangefs_sb->id = new_op->downcall.resp.fs_mount.id;
265                 orangefs_sb->mount_pending = 0;
266         }
267
268         op_release(new_op);
269
270         if (orangefs_userspace_version >= 20906) {
271                 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
272                 if (!new_op)
273                         return -ENOMEM;
274                 new_op->upcall.req.features.features = 0;
275                 ret = service_operation(new_op, "orangefs_features",
276                     ORANGEFS_OP_PRIORITY | ORANGEFS_OP_NO_MUTEX);
277                 if (!ret)
278                         orangefs_features =
279                             new_op->downcall.resp.features.features;
280                 else
281                         orangefs_features = 0;
282                 op_release(new_op);
283         } else {
284                 orangefs_features = 0;
285         }
286
287         return ret;
288 }
289
290 int fsid_key_table_initialize(void)
291 {
292         return 0;
293 }
294
295 void fsid_key_table_finalize(void)
296 {
297 }
298
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,
306 };
307
308 static struct dentry *orangefs_fh_to_dentry(struct super_block *sb,
309                                   struct fid *fid,
310                                   int fh_len,
311                                   int fh_type)
312 {
313         struct orangefs_object_kref refn;
314
315         if (fh_len < 5 || fh_type > 2)
316                 return NULL;
317
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",
322                      &refn.khandle,
323                      refn.fs_id);
324
325         return d_obtain_alias(orangefs_iget(sb, &refn));
326 }
327
328 static int orangefs_encode_fh(struct inode *inode,
329                     __u32 *fh,
330                     int *max_len,
331                     struct inode *parent)
332 {
333         int len = parent ? 10 : 5;
334         int type = 1;
335         struct orangefs_object_kref refn;
336
337         if (*max_len < len) {
338                 gossip_err("fh buffer is too small for encoding\n");
339                 *max_len = len;
340                 type = 255;
341                 goto out;
342         }
343
344         refn = ORANGEFS_I(inode)->refn;
345         ORANGEFS_khandle_to(&refn.khandle, fh, 16);
346         fh[4] = refn.fs_id;
347
348         gossip_debug(GOSSIP_SUPER_DEBUG,
349                      "Encoding fh: handle %pU, fsid %u\n",
350                      &refn.khandle,
351                      refn.fs_id);
352
353
354         if (parent) {
355                 refn = ORANGEFS_I(parent)->refn;
356                 ORANGEFS_khandle_to(&refn.khandle, (char *) fh + 20, 16);
357                 fh[9] = refn.fs_id;
358
359                 type = 2;
360                 gossip_debug(GOSSIP_SUPER_DEBUG,
361                              "Encoding parent: handle %pU, fsid %u\n",
362                              &refn.khandle,
363                              refn.fs_id);
364         }
365         *max_len = len;
366
367 out:
368         return type;
369 }
370
371 static const struct export_operations orangefs_export_ops = {
372         .encode_fh = orangefs_encode_fh,
373         .fh_to_dentry = orangefs_fh_to_dentry,
374 };
375
376 static int orangefs_unmount(int id, __s32 fs_id, const char *devname)
377 {
378         struct orangefs_kernel_op_s *op;
379         int r;
380         op = op_alloc(ORANGEFS_VFS_OP_FS_UMOUNT);
381         if (!op)
382                 return -ENOMEM;
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. */
389         if (r)
390                 gossip_err("orangefs_unmount: service_operation %d\n", r);
391         op_release(op);
392         return r;
393 }
394
395 static int orangefs_fill_sb(struct super_block *sb,
396                 struct orangefs_fs_mount_response *fs_mount,
397                 void *data, int silent)
398 {
399         int ret = -EINVAL;
400         struct inode *root = NULL;
401         struct dentry *root_dentry = NULL;
402         struct orangefs_object_kref root_object;
403
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))
407                 return -ENOMEM;
408         ORANGEFS_SB(sb)->sb = sb;
409
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;
413
414         if (data) {
415                 ret = parse_mount_options(sb, data, silent);
416                 if (ret)
417                         return ret;
418         }
419
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;
425
426         sb->s_blocksize = orangefs_bufmap_size_query();
427         sb->s_blocksize_bits = orangefs_bufmap_shift_query();
428         sb->s_maxbytes = MAX_LFS_FILESIZE;
429
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,
435                      root_object.fs_id);
436
437         root = orangefs_iget(sb, &root_object);
438         if (IS_ERR(root))
439                 return PTR_ERR(root);
440
441         gossip_debug(GOSSIP_SUPER_DEBUG,
442                      "Allocated root inode [%p] with mode %x\n",
443                      root,
444                      root->i_mode);
445
446         /* allocates and places root dentry in dcache */
447         root_dentry = d_make_root(root);
448         if (!root_dentry)
449                 return -ENOMEM;
450
451         sb->s_export_op = &orangefs_export_ops;
452         sb->s_root = root_dentry;
453         return 0;
454 }
455
456 struct dentry *orangefs_mount(struct file_system_type *fst,
457                            int flags,
458                            const char *devname,
459                            void *data)
460 {
461         int ret = -EINVAL;
462         struct super_block *sb = ERR_PTR(-EINVAL);
463         struct orangefs_kernel_op_s *new_op;
464         struct dentry *d = ERR_PTR(-EINVAL);
465
466         gossip_debug(GOSSIP_SUPER_DEBUG,
467                      "orangefs_mount: called with devname %s\n",
468                      devname);
469
470         if (!devname) {
471                 gossip_err("ERROR: device name not specified.\n");
472                 return ERR_PTR(-EINVAL);
473         }
474
475         new_op = op_alloc(ORANGEFS_VFS_OP_FS_MOUNT);
476         if (!new_op)
477                 return ERR_PTR(-ENOMEM);
478
479         strncpy(new_op->upcall.req.fs_mount.orangefs_config_server,
480                 devname,
481                 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
482
483         gossip_debug(GOSSIP_SUPER_DEBUG,
484                      "Attempting ORANGEFS Mount via host %s\n",
485                      new_op->upcall.req.fs_mount.orangefs_config_server);
486
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);
490         if (ret)
491                 goto free_op;
492
493         if (new_op->downcall.resp.fs_mount.fs_id == ORANGEFS_FS_ID_NULL) {
494                 gossip_err("ERROR: Retrieved null fs_id\n");
495                 ret = -EINVAL;
496                 goto free_op;
497         }
498
499         sb = sget(fst, NULL, set_anon_super, flags, NULL);
500
501         if (IS_ERR(sb)) {
502                 d = ERR_CAST(sb);
503                 orangefs_unmount(new_op->downcall.resp.fs_mount.id,
504                     new_op->downcall.resp.fs_mount.fs_id, devname);
505                 goto free_op;
506         }
507
508         ret = orangefs_fill_sb(sb,
509               &new_op->downcall.resp.fs_mount, data,
510               flags & SB_SILENT ? 1 : 0);
511
512         if (ret) {
513                 d = ERR_PTR(ret);
514                 goto free_sb_and_op;
515         }
516
517         /*
518          * on successful mount, store the devname and data
519          * used
520          */
521         strncpy(ORANGEFS_SB(sb)->devname,
522                 devname,
523                 ORANGEFS_MAX_SERVER_ADDR_LEN - 1);
524
525         /* mount_pending must be cleared */
526         ORANGEFS_SB(sb)->mount_pending = 0;
527
528         /*
529          * finally, add this sb to our list of known orangefs
530          * sb's
531          */
532         gossip_debug(GOSSIP_SUPER_DEBUG,
533                      "Adding SB %p to orangefs superblocks\n",
534                      ORANGEFS_SB(sb));
535         spin_lock(&orangefs_superblocks_lock);
536         list_add_tail(&ORANGEFS_SB(sb)->list, &orangefs_superblocks);
537         spin_unlock(&orangefs_superblocks_lock);
538         op_release(new_op);
539
540         /* Must be removed from the list now. */
541         ORANGEFS_SB(sb)->no_list = 0;
542
543         if (orangefs_userspace_version >= 20906) {
544                 new_op = op_alloc(ORANGEFS_VFS_OP_FEATURES);
545                 if (!new_op)
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;
550                 op_release(new_op);
551         } else {
552                 orangefs_features = 0;
553         }
554
555         return dget(sb->s_root);
556
557 free_sb_and_op:
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);
562 free_op:
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");
567         }
568
569         op_release(new_op);
570
571         return d;
572 }
573
574 void orangefs_kill_sb(struct super_block *sb)
575 {
576         int r;
577         gossip_debug(GOSSIP_SUPER_DEBUG, "orangefs_kill_sb: called\n");
578
579         /* provided sb cleanup */
580         kill_anon_super(sb);
581
582         /*
583          * issue the unmount to userspace to tell it to remove the
584          * dynamic mount info it has for this superblock
585          */
586         r = orangefs_unmount(ORANGEFS_SB(sb)->id, ORANGEFS_SB(sb)->fs_id,
587             ORANGEFS_SB(sb)->devname);
588         if (!r)
589                 ORANGEFS_SB(sb)->mount_pending = 1;
590
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);
598         }
599
600         /*
601          * make sure that ORANGEFS_DEV_REMOUNT_ALL loop that might've seen us
602          * gets completed before we free the dang thing.
603          */
604         mutex_lock(&orangefs_request_mutex);
605         mutex_unlock(&orangefs_request_mutex);
606
607         /* free the orangefs superblock private data */
608         kfree(ORANGEFS_SB(sb));
609 }
610
611 int orangefs_inode_cache_initialize(void)
612 {
613         orangefs_inode_cache = kmem_cache_create_usercopy(
614                                         "orangefs_inode_cache",
615                                         sizeof(struct orangefs_inode_s),
616                                         0,
617                                         ORANGEFS_CACHE_CREATE_FLAGS,
618                                         offsetof(struct orangefs_inode_s,
619                                                 link_target),
620                                         sizeof_field(struct orangefs_inode_s,
621                                                 link_target),
622                                         orangefs_inode_cache_ctor);
623
624         if (!orangefs_inode_cache) {
625                 gossip_err("Cannot create orangefs_inode_cache\n");
626                 return -ENOMEM;
627         }
628         return 0;
629 }
630
631 int orangefs_inode_cache_finalize(void)
632 {
633         kmem_cache_destroy(orangefs_inode_cache);
634         return 0;
635 }