Linux-libre 3.2.73-gnu1
[librecmc/linux-libre.git] / fs / exofs / super.c
1 /*
2  * Copyright (C) 2005, 2006
3  * Avishay Traeger (avishay@gmail.com)
4  * Copyright (C) 2008, 2009
5  * Boaz Harrosh <bharrosh@panasas.com>
6  *
7  * Copyrights for code taken from ext2:
8  *     Copyright (C) 1992, 1993, 1994, 1995
9  *     Remy Card (card@masi.ibp.fr)
10  *     Laboratoire MASI - Institut Blaise Pascal
11  *     Universite Pierre et Marie Curie (Paris VI)
12  *     from
13  *     linux/fs/minix/inode.c
14  *     Copyright (C) 1991, 1992  Linus Torvalds
15  *
16  * This file is part of exofs.
17  *
18  * exofs is free software; you can redistribute it and/or modify
19  * it under the terms of the GNU General Public License as published by
20  * the Free Software Foundation.  Since it is based on ext2, and the only
21  * valid version of GPL for the Linux kernel is version 2, the only valid
22  * version of GPL for exofs is version 2.
23  *
24  * exofs is distributed in the hope that it will be useful,
25  * but WITHOUT ANY WARRANTY; without even the implied warranty of
26  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27  * GNU General Public License for more details.
28  *
29  * You should have received a copy of the GNU General Public License
30  * along with exofs; if not, write to the Free Software
31  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
32  */
33
34 #include <linux/string.h>
35 #include <linux/parser.h>
36 #include <linux/vfs.h>
37 #include <linux/random.h>
38 #include <linux/module.h>
39 #include <linux/exportfs.h>
40 #include <linux/slab.h>
41
42 #include "exofs.h"
43
44 #define EXOFS_DBGMSG2(M...) do {} while (0)
45
46 /******************************************************************************
47  * MOUNT OPTIONS
48  *****************************************************************************/
49
50 /*
51  * struct to hold what we get from mount options
52  */
53 struct exofs_mountopt {
54         bool is_osdname;
55         const char *dev_name;
56         uint64_t pid;
57         int timeout;
58 };
59
60 /*
61  * exofs-specific mount-time options.
62  */
63 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
64
65 /*
66  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
67  * kernel's parsing functions do not currently support that.  32-bit should be
68  * sufficient for most applications now.
69  */
70 static match_table_t tokens = {
71         {Opt_name, "osdname=%s"},
72         {Opt_pid, "pid=%u"},
73         {Opt_to, "to=%u"},
74         {Opt_err, NULL}
75 };
76
77 /*
78  * The main option parsing method.  Also makes sure that all of the mandatory
79  * mount options were set.
80  */
81 static int parse_options(char *options, struct exofs_mountopt *opts)
82 {
83         char *p;
84         substring_t args[MAX_OPT_ARGS];
85         int option;
86         bool s_pid = false;
87
88         EXOFS_DBGMSG("parse_options %s\n", options);
89         /* defaults */
90         memset(opts, 0, sizeof(*opts));
91         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
92
93         while ((p = strsep(&options, ",")) != NULL) {
94                 int token;
95                 char str[32];
96
97                 if (!*p)
98                         continue;
99
100                 token = match_token(p, tokens, args);
101                 switch (token) {
102                 case Opt_name:
103                         opts->dev_name = match_strdup(&args[0]);
104                         if (unlikely(!opts->dev_name)) {
105                                 EXOFS_ERR("Error allocating dev_name");
106                                 return -ENOMEM;
107                         }
108                         opts->is_osdname = true;
109                         break;
110                 case Opt_pid:
111                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
112                                 return -EINVAL;
113                         opts->pid = simple_strtoull(str, NULL, 0);
114                         if (opts->pid < EXOFS_MIN_PID) {
115                                 EXOFS_ERR("Partition ID must be >= %u",
116                                           EXOFS_MIN_PID);
117                                 return -EINVAL;
118                         }
119                         s_pid = 1;
120                         break;
121                 case Opt_to:
122                         if (match_int(&args[0], &option))
123                                 return -EINVAL;
124                         if (option <= 0) {
125                                 EXOFS_ERR("Timout must be > 0");
126                                 return -EINVAL;
127                         }
128                         opts->timeout = option * HZ;
129                         break;
130                 }
131         }
132
133         if (!s_pid) {
134                 EXOFS_ERR("Need to specify the following options:\n");
135                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
136                 return -EINVAL;
137         }
138
139         return 0;
140 }
141
142 /******************************************************************************
143  * INODE CACHE
144  *****************************************************************************/
145
146 /*
147  * Our inode cache.  Isn't it pretty?
148  */
149 static struct kmem_cache *exofs_inode_cachep;
150
151 /*
152  * Allocate an inode in the cache
153  */
154 static struct inode *exofs_alloc_inode(struct super_block *sb)
155 {
156         struct exofs_i_info *oi;
157
158         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
159         if (!oi)
160                 return NULL;
161
162         oi->vfs_inode.i_version = 1;
163         return &oi->vfs_inode;
164 }
165
166 static void exofs_i_callback(struct rcu_head *head)
167 {
168         struct inode *inode = container_of(head, struct inode, i_rcu);
169         INIT_LIST_HEAD(&inode->i_dentry);
170         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
171 }
172
173 /*
174  * Remove an inode from the cache
175  */
176 static void exofs_destroy_inode(struct inode *inode)
177 {
178         call_rcu(&inode->i_rcu, exofs_i_callback);
179 }
180
181 /*
182  * Initialize the inode
183  */
184 static void exofs_init_once(void *foo)
185 {
186         struct exofs_i_info *oi = foo;
187
188         inode_init_once(&oi->vfs_inode);
189 }
190
191 /*
192  * Create and initialize the inode cache
193  */
194 static int init_inodecache(void)
195 {
196         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
197                                 sizeof(struct exofs_i_info), 0,
198                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
199                                 exofs_init_once);
200         if (exofs_inode_cachep == NULL)
201                 return -ENOMEM;
202         return 0;
203 }
204
205 /*
206  * Destroy the inode cache
207  */
208 static void destroy_inodecache(void)
209 {
210         kmem_cache_destroy(exofs_inode_cachep);
211 }
212
213 /******************************************************************************
214  * Some osd helpers
215  *****************************************************************************/
216 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
217 {
218         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
219 }
220
221 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
222                     u64 offset, void *p, unsigned length)
223 {
224         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
225 /*      struct osd_sense_info osi = {.key = 0};*/
226         int ret;
227
228         if (unlikely(!or)) {
229                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
230                 return -ENOMEM;
231         }
232         ret = osd_req_read_kern(or, obj, offset, p, length);
233         if (unlikely(ret)) {
234                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
235                 goto out;
236         }
237
238         ret = osd_finalize_request(or, 0, cred, NULL);
239         if (unlikely(ret)) {
240                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
241                 goto out;
242         }
243
244         ret = osd_execute_request(or);
245         if (unlikely(ret))
246                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
247         /* osd_req_decode_sense(or, ret); */
248
249 out:
250         osd_end_request(or);
251         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
252                       "length=0x%llx dev=%p ret=>%d\n",
253                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
254         return ret;
255 }
256
257 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
258         EXOFS_APAGE_SB_DATA,
259         EXOFS_ATTR_SB_STATS,
260         sizeof(struct exofs_sb_stats));
261
262 static int __sbi_read_stats(struct exofs_sb_info *sbi)
263 {
264         struct osd_attr attrs[] = {
265                 [0] = g_attr_sb_stats,
266         };
267         struct ore_io_state *ios;
268         int ret;
269
270         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
271         if (unlikely(ret)) {
272                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
273                 return ret;
274         }
275
276         ios->in_attr = attrs;
277         ios->in_attr_len = ARRAY_SIZE(attrs);
278
279         ret = ore_read(ios);
280         if (unlikely(ret)) {
281                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
282                 goto out;
283         }
284
285         ret = extract_attr_from_ios(ios, &attrs[0]);
286         if (ret) {
287                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
288                 goto out;
289         }
290         if (attrs[0].len) {
291                 struct exofs_sb_stats *ess;
292
293                 if (unlikely(attrs[0].len != sizeof(*ess))) {
294                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
295                                   "size(%d) != expected(%zd)\n",
296                                   __func__, attrs[0].len, sizeof(*ess));
297                         goto out;
298                 }
299
300                 ess = attrs[0].val_ptr;
301                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
302                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
303         }
304
305 out:
306         ore_put_io_state(ios);
307         return ret;
308 }
309
310 static void stats_done(struct ore_io_state *ios, void *p)
311 {
312         ore_put_io_state(ios);
313         /* Good thanks nothing to do anymore */
314 }
315
316 /* Asynchronously write the stats attribute */
317 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
318 {
319         struct osd_attr attrs[] = {
320                 [0] = g_attr_sb_stats,
321         };
322         struct ore_io_state *ios;
323         int ret;
324
325         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
326         if (unlikely(ret)) {
327                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
328                 return ret;
329         }
330
331         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
332         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
333         attrs[0].val_ptr = &sbi->s_ess;
334
335
336         ios->done = stats_done;
337         ios->private = sbi;
338         ios->out_attr = attrs;
339         ios->out_attr_len = ARRAY_SIZE(attrs);
340
341         ret = ore_write(ios);
342         if (unlikely(ret)) {
343                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
344                 ore_put_io_state(ios);
345         }
346
347         return ret;
348 }
349
350 /******************************************************************************
351  * SUPERBLOCK FUNCTIONS
352  *****************************************************************************/
353 static const struct super_operations exofs_sops;
354 static const struct export_operations exofs_export_ops;
355
356 /*
357  * Write the superblock to the OSD
358  */
359 static int exofs_sync_fs(struct super_block *sb, int wait)
360 {
361         struct exofs_sb_info *sbi;
362         struct exofs_fscb *fscb;
363         struct ore_comp one_comp;
364         struct ore_components oc;
365         struct ore_io_state *ios;
366         int ret = -ENOMEM;
367
368         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
369         if (unlikely(!fscb))
370                 return -ENOMEM;
371
372         sbi = sb->s_fs_info;
373
374         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
375          * reason we write the fscb here on unmount is so we can stay backwards
376          * compatible with fscb->s_version == 1. (What we are not compatible
377          * with is if a new version FS crashed and then we try to mount an old
378          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
379          * the writeable info is set in exofs_sbi_write_stats() above.
380          */
381
382         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
383
384         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
385         if (unlikely(ret))
386                 goto out;
387
388         lock_super(sb);
389
390         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
391         memset(fscb, 0, ios->length);
392         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
393         fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
394         fscb->s_magic = cpu_to_le16(sb->s_magic);
395         fscb->s_newfs = 0;
396         fscb->s_version = EXOFS_FSCB_VER;
397
398         ios->offset = 0;
399         ios->kern_buff = fscb;
400
401         ret = ore_write(ios);
402         if (unlikely(ret))
403                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
404         else
405                 sb->s_dirt = 0;
406
407
408         unlock_super(sb);
409 out:
410         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
411         ore_put_io_state(ios);
412         kfree(fscb);
413         return ret;
414 }
415
416 static void exofs_write_super(struct super_block *sb)
417 {
418         if (!(sb->s_flags & MS_RDONLY))
419                 exofs_sync_fs(sb, 1);
420         else
421                 sb->s_dirt = 0;
422 }
423
424 static void _exofs_print_device(const char *msg, const char *dev_path,
425                                 struct osd_dev *od, u64 pid)
426 {
427         const struct osd_dev_info *odi = osduld_device_info(od);
428
429         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
430                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
431 }
432
433 static void exofs_free_sbi(struct exofs_sb_info *sbi)
434 {
435         unsigned numdevs = sbi->oc.numdevs;
436
437         while (numdevs) {
438                 unsigned i = --numdevs;
439                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
440
441                 if (od) {
442                         ore_comp_set_dev(&sbi->oc, i, NULL);
443                         osduld_put_device(od);
444                 }
445         }
446         kfree(sbi->oc.ods);
447         kfree(sbi);
448 }
449
450 /*
451  * This function is called when the vfs is freeing the superblock.  We just
452  * need to free our own part.
453  */
454 static void exofs_put_super(struct super_block *sb)
455 {
456         int num_pend;
457         struct exofs_sb_info *sbi = sb->s_fs_info;
458
459         /* make sure there are no pending commands */
460         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
461              num_pend = atomic_read(&sbi->s_curr_pending)) {
462                 wait_queue_head_t wq;
463
464                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
465                        "This is a BUG. please report to osd-dev@open-osd.org\n",
466                        __func__);
467                 init_waitqueue_head(&wq);
468                 wait_event_timeout(wq,
469                                   (atomic_read(&sbi->s_curr_pending) == 0),
470                                   msecs_to_jiffies(100));
471         }
472
473         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
474                             sbi->one_comp.obj.partition);
475
476         bdi_destroy(&sbi->bdi);
477         exofs_free_sbi(sbi);
478         sb->s_fs_info = NULL;
479 }
480
481 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
482                                     struct exofs_device_table *dt)
483 {
484         int ret;
485
486         sbi->layout.stripe_unit =
487                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
488         sbi->layout.group_width =
489                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
490         sbi->layout.group_depth =
491                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
492         sbi->layout.mirrors_p1  =
493                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
494         sbi->layout.raid_algorithm  =
495                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
496
497         ret = ore_verify_layout(numdevs, &sbi->layout);
498
499         EXOFS_DBGMSG("exofs: layout: "
500                 "num_comps=%u stripe_unit=0x%x group_width=%u "
501                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
502                 numdevs,
503                 sbi->layout.stripe_unit,
504                 sbi->layout.group_width,
505                 _LLU(sbi->layout.group_depth),
506                 sbi->layout.mirrors_p1,
507                 sbi->layout.raid_algorithm);
508         return ret;
509 }
510
511 static unsigned __ra_pages(struct ore_layout *layout)
512 {
513         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
514         unsigned ra_pages = layout->group_width * layout->stripe_unit /
515                                 PAGE_SIZE;
516         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
517
518         ra_pages *= 2; /* two stripes */
519         if (ra_pages < _MIN_RA)
520                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
521
522         if (ra_pages > max_io_pages)
523                 ra_pages = max_io_pages;
524
525         return ra_pages;
526 }
527
528 /* @odi is valid only as long as @fscb_dev is valid */
529 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
530                              struct osd_dev_info *odi)
531 {
532         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
533         memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
534
535         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
536         odi->osdname = dt_dev->osdname;
537
538         /* FIXME support long names. Will need a _put function */
539         if (dt_dev->long_name_offset)
540                 return -EINVAL;
541
542         /* Make sure osdname is printable!
543          * mkexofs should give us space for a null-terminator else the
544          * device-table is invalid.
545          */
546         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
547                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
548         dt_dev->osdname[odi->osdname_len] = 0;
549
550         /* If it's all zeros something is bad we read past end-of-obj */
551         return !(odi->systemid_len || odi->osdname_len);
552 }
553
554 int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
555                       struct exofs_dev **peds)
556 {
557         struct __alloc_ore_devs_and_exofs_devs {
558                 /* Twice bigger table: See exofs_init_comps() and comment at
559                  * exofs_read_lookup_dev_table()
560                  */
561                 struct ore_dev *oreds[numdevs * 2 - 1];
562                 struct exofs_dev eds[numdevs];
563         } *aoded;
564         struct exofs_dev *eds;
565         unsigned i;
566
567         aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
568         if (unlikely(!aoded)) {
569                 EXOFS_ERR("ERROR: faild allocating Device array[%d]\n",
570                           numdevs);
571                 return -ENOMEM;
572         }
573
574         sbi->oc.ods = aoded->oreds;
575         *peds = eds = aoded->eds;
576         for (i = 0; i < numdevs; ++i)
577                 aoded->oreds[i] = &eds[i].ored;
578         return 0;
579 }
580
581 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
582                                        struct osd_dev *fscb_od,
583                                        unsigned table_count)
584 {
585         struct ore_comp comp;
586         struct exofs_device_table *dt;
587         struct exofs_dev *eds;
588         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
589                                              sizeof(*dt);
590         unsigned numdevs, i;
591         int ret;
592
593         dt = kmalloc(table_bytes, GFP_KERNEL);
594         if (unlikely(!dt)) {
595                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
596                           table_bytes);
597                 return -ENOMEM;
598         }
599
600         sbi->oc.numdevs = 0;
601
602         comp.obj.partition = sbi->one_comp.obj.partition;
603         comp.obj.id = EXOFS_DEVTABLE_ID;
604         exofs_make_credential(comp.cred, &comp.obj);
605
606         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
607                               table_bytes);
608         if (unlikely(ret)) {
609                 EXOFS_ERR("ERROR: reading device table\n");
610                 goto out;
611         }
612
613         numdevs = le64_to_cpu(dt->dt_num_devices);
614         if (unlikely(!numdevs)) {
615                 ret = -EINVAL;
616                 goto out;
617         }
618         WARN_ON(table_count != numdevs);
619
620         ret = _read_and_match_data_map(sbi, numdevs, dt);
621         if (unlikely(ret))
622                 goto out;
623
624         ret = __alloc_dev_table(sbi, numdevs, &eds);
625         if (unlikely(ret))
626                 goto out;
627         /* exofs round-robins the device table view according to inode
628          * number. We hold a: twice bigger table hence inodes can point
629          * to any device and have a sequential view of the table
630          * starting at this device. See exofs_init_comps()
631          */
632         memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
633                 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
634
635         for (i = 0; i < numdevs; i++) {
636                 struct exofs_fscb fscb;
637                 struct osd_dev_info odi;
638                 struct osd_dev *od;
639
640                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
641                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
642                         ret = -EINVAL;
643                         goto out;
644                 }
645
646                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
647                        i, odi.osdname);
648
649                 /* the exofs id is currently the table index */
650                 eds[i].did = i;
651
652                 /* On all devices the device table is identical. The user can
653                  * specify any one of the participating devices on the command
654                  * line. We always keep them in device-table order.
655                  */
656                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
657                         eds[i].ored.od = fscb_od;
658                         ++sbi->oc.numdevs;
659                         fscb_od = NULL;
660                         continue;
661                 }
662
663                 od = osduld_info_lookup(&odi);
664                 if (IS_ERR(od)) {
665                         ret = PTR_ERR(od);
666                         EXOFS_ERR("ERROR: device requested is not found "
667                                   "osd_name-%s =>%d\n", odi.osdname, ret);
668                         goto out;
669                 }
670
671                 eds[i].ored.od = od;
672                 ++sbi->oc.numdevs;
673
674                 /* Read the fscb of the other devices to make sure the FS
675                  * partition is there.
676                  */
677                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
678                                       sizeof(fscb));
679                 if (unlikely(ret)) {
680                         EXOFS_ERR("ERROR: Malformed participating device "
681                                   "error reading fscb osd_name-%s\n",
682                                   odi.osdname);
683                         goto out;
684                 }
685
686                 /* TODO: verify other information is correct and FS-uuid
687                  *       matches. Benny what did you say about device table
688                  *       generation and old devices?
689                  */
690         }
691
692 out:
693         kfree(dt);
694         if (unlikely(fscb_od && !ret)) {
695                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
696                         osduld_put_device(fscb_od);
697                         return -EINVAL;
698         }
699         return ret;
700 }
701
702 /*
703  * Read the superblock from the OSD and fill in the fields
704  */
705 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
706 {
707         struct inode *root;
708         struct exofs_mountopt *opts = data;
709         struct exofs_sb_info *sbi;      /*extended info                  */
710         struct osd_dev *od;             /* Master device                 */
711         struct exofs_fscb fscb;         /*on-disk superblock info        */
712         struct ore_comp comp;
713         unsigned table_count;
714         int ret;
715
716         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
717         if (!sbi)
718                 return -ENOMEM;
719
720         /* use mount options to fill superblock */
721         if (opts->is_osdname) {
722                 struct osd_dev_info odi = {.systemid_len = 0};
723
724                 odi.osdname_len = strlen(opts->dev_name);
725                 odi.osdname = (u8 *)opts->dev_name;
726                 od = osduld_info_lookup(&odi);
727                 kfree(opts->dev_name);
728                 opts->dev_name = NULL;
729         } else {
730                 od = osduld_path_lookup(opts->dev_name);
731         }
732         if (IS_ERR(od)) {
733                 ret = -EINVAL;
734                 goto free_sbi;
735         }
736
737         /* Default layout in case we do not have a device-table */
738         sbi->layout.stripe_unit = PAGE_SIZE;
739         sbi->layout.mirrors_p1 = 1;
740         sbi->layout.group_width = 1;
741         sbi->layout.group_depth = -1;
742         sbi->layout.group_count = 1;
743         sbi->s_timeout = opts->timeout;
744
745         sbi->one_comp.obj.partition = opts->pid;
746         sbi->one_comp.obj.id = 0;
747         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
748         sbi->oc.single_comp = EC_SINGLE_COMP;
749         sbi->oc.comps = &sbi->one_comp;
750
751         /* fill in some other data by hand */
752         memset(sb->s_id, 0, sizeof(sb->s_id));
753         strcpy(sb->s_id, "exofs");
754         sb->s_blocksize = EXOFS_BLKSIZE;
755         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
756         sb->s_maxbytes = MAX_LFS_FILESIZE;
757         atomic_set(&sbi->s_curr_pending, 0);
758         sb->s_bdev = NULL;
759         sb->s_dev = 0;
760
761         comp.obj.partition = sbi->one_comp.obj.partition;
762         comp.obj.id = EXOFS_SUPER_ID;
763         exofs_make_credential(comp.cred, &comp.obj);
764
765         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
766         if (unlikely(ret))
767                 goto free_sbi;
768
769         sb->s_magic = le16_to_cpu(fscb.s_magic);
770         /* NOTE: we read below to be backward compatible with old versions */
771         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
772         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
773
774         /* make sure what we read from the object store is correct */
775         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
776                 if (!silent)
777                         EXOFS_ERR("ERROR: Bad magic value\n");
778                 ret = -EINVAL;
779                 goto free_sbi;
780         }
781         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
782                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
783                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
784                 ret = -EINVAL;
785                 goto free_sbi;
786         }
787
788         /* start generation numbers from a random point */
789         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
790         spin_lock_init(&sbi->s_next_gen_lock);
791
792         table_count = le64_to_cpu(fscb.s_dev_table_count);
793         if (table_count) {
794                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
795                 if (unlikely(ret))
796                         goto free_sbi;
797         } else {
798                 struct exofs_dev *eds;
799
800                 ret = __alloc_dev_table(sbi, 1, &eds);
801                 if (unlikely(ret))
802                         goto free_sbi;
803
804                 ore_comp_set_dev(&sbi->oc, 0, od);
805                 sbi->oc.numdevs = 1;
806         }
807
808         __sbi_read_stats(sbi);
809
810         /* set up operation vectors */
811         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
812         sb->s_bdi = &sbi->bdi;
813         sb->s_fs_info = sbi;
814         sb->s_op = &exofs_sops;
815         sb->s_export_op = &exofs_export_ops;
816         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
817         if (IS_ERR(root)) {
818                 EXOFS_ERR("ERROR: exofs_iget failed\n");
819                 ret = PTR_ERR(root);
820                 goto free_sbi;
821         }
822         sb->s_root = d_alloc_root(root);
823         if (!sb->s_root) {
824                 iput(root);
825                 EXOFS_ERR("ERROR: get root inode failed\n");
826                 ret = -ENOMEM;
827                 goto free_sbi;
828         }
829
830         if (!S_ISDIR(root->i_mode)) {
831                 dput(sb->s_root);
832                 sb->s_root = NULL;
833                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
834                        root->i_mode);
835                 ret = -EINVAL;
836                 goto free_sbi;
837         }
838
839         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
840         if (ret) {
841                 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
842                 goto free_sbi;
843         }
844
845         _exofs_print_device("Mounting", opts->dev_name,
846                             ore_comp_dev(&sbi->oc, 0),
847                             sbi->one_comp.obj.partition);
848         return 0;
849
850 free_sbi:
851         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
852                   opts->dev_name, sbi->one_comp.obj.partition, ret);
853         exofs_free_sbi(sbi);
854         return ret;
855 }
856
857 /*
858  * Set up the superblock (calls exofs_fill_super eventually)
859  */
860 static struct dentry *exofs_mount(struct file_system_type *type,
861                           int flags, const char *dev_name,
862                           void *data)
863 {
864         struct exofs_mountopt opts;
865         int ret;
866
867         ret = parse_options(data, &opts);
868         if (ret)
869                 return ERR_PTR(ret);
870
871         if (!opts.dev_name)
872                 opts.dev_name = dev_name;
873         return mount_nodev(type, flags, &opts, exofs_fill_super);
874 }
875
876 /*
877  * Return information about the file system state in the buffer.  This is used
878  * by the 'df' command, for example.
879  */
880 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
881 {
882         struct super_block *sb = dentry->d_sb;
883         struct exofs_sb_info *sbi = sb->s_fs_info;
884         struct ore_io_state *ios;
885         struct osd_attr attrs[] = {
886                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
887                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
888                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
889                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
890         };
891         uint64_t capacity = ULLONG_MAX;
892         uint64_t used = ULLONG_MAX;
893         int ret;
894
895         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
896         if (ret) {
897                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
898                 return ret;
899         }
900
901         ios->in_attr = attrs;
902         ios->in_attr_len = ARRAY_SIZE(attrs);
903
904         ret = ore_read(ios);
905         if (unlikely(ret))
906                 goto out;
907
908         ret = extract_attr_from_ios(ios, &attrs[0]);
909         if (likely(!ret)) {
910                 capacity = get_unaligned_be64(attrs[0].val_ptr);
911                 if (unlikely(!capacity))
912                         capacity = ULLONG_MAX;
913         } else
914                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
915
916         ret = extract_attr_from_ios(ios, &attrs[1]);
917         if (likely(!ret))
918                 used = get_unaligned_be64(attrs[1].val_ptr);
919         else
920                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
921
922         /* fill in the stats buffer */
923         buf->f_type = EXOFS_SUPER_MAGIC;
924         buf->f_bsize = EXOFS_BLKSIZE;
925         buf->f_blocks = capacity >> 9;
926         buf->f_bfree = (capacity - used) >> 9;
927         buf->f_bavail = buf->f_bfree;
928         buf->f_files = sbi->s_numfiles;
929         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
930         buf->f_namelen = EXOFS_NAME_LEN;
931
932 out:
933         ore_put_io_state(ios);
934         return ret;
935 }
936
937 static const struct super_operations exofs_sops = {
938         .alloc_inode    = exofs_alloc_inode,
939         .destroy_inode  = exofs_destroy_inode,
940         .write_inode    = exofs_write_inode,
941         .evict_inode    = exofs_evict_inode,
942         .put_super      = exofs_put_super,
943         .write_super    = exofs_write_super,
944         .sync_fs        = exofs_sync_fs,
945         .statfs         = exofs_statfs,
946 };
947
948 /******************************************************************************
949  * EXPORT OPERATIONS
950  *****************************************************************************/
951
952 static struct dentry *exofs_get_parent(struct dentry *child)
953 {
954         unsigned long ino = exofs_parent_ino(child);
955
956         if (!ino)
957                 return ERR_PTR(-ESTALE);
958
959         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
960 }
961
962 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
963                 u64 ino, u32 generation)
964 {
965         struct inode *inode;
966
967         inode = exofs_iget(sb, ino);
968         if (IS_ERR(inode))
969                 return ERR_CAST(inode);
970         if (generation && inode->i_generation != generation) {
971                 /* we didn't find the right inode.. */
972                 iput(inode);
973                 return ERR_PTR(-ESTALE);
974         }
975         return inode;
976 }
977
978 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
979                                 struct fid *fid, int fh_len, int fh_type)
980 {
981         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
982                                     exofs_nfs_get_inode);
983 }
984
985 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
986                                 struct fid *fid, int fh_len, int fh_type)
987 {
988         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
989                                     exofs_nfs_get_inode);
990 }
991
992 static const struct export_operations exofs_export_ops = {
993         .fh_to_dentry = exofs_fh_to_dentry,
994         .fh_to_parent = exofs_fh_to_parent,
995         .get_parent = exofs_get_parent,
996 };
997
998 /******************************************************************************
999  * INSMOD/RMMOD
1000  *****************************************************************************/
1001
1002 /*
1003  * struct that describes this file system
1004  */
1005 static struct file_system_type exofs_type = {
1006         .owner          = THIS_MODULE,
1007         .name           = "exofs",
1008         .mount          = exofs_mount,
1009         .kill_sb        = generic_shutdown_super,
1010 };
1011
1012 static int __init init_exofs(void)
1013 {
1014         int err;
1015
1016         err = init_inodecache();
1017         if (err)
1018                 goto out;
1019
1020         err = register_filesystem(&exofs_type);
1021         if (err)
1022                 goto out_d;
1023
1024         return 0;
1025 out_d:
1026         destroy_inodecache();
1027 out:
1028         return err;
1029 }
1030
1031 static void __exit exit_exofs(void)
1032 {
1033         unregister_filesystem(&exofs_type);
1034         destroy_inodecache();
1035 }
1036
1037 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1038 MODULE_DESCRIPTION("exofs");
1039 MODULE_LICENSE("GPL");
1040
1041 module_init(init_exofs)
1042 module_exit(exit_exofs)