Merge branch 'for-linus' of git://git.open-osd.org/linux-open-osd
[~shefty/rdma-dev.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         kmem_cache_free(exofs_inode_cachep, exofs_i(inode));
170 }
171
172 /*
173  * Remove an inode from the cache
174  */
175 static void exofs_destroy_inode(struct inode *inode)
176 {
177         call_rcu(&inode->i_rcu, exofs_i_callback);
178 }
179
180 /*
181  * Initialize the inode
182  */
183 static void exofs_init_once(void *foo)
184 {
185         struct exofs_i_info *oi = foo;
186
187         inode_init_once(&oi->vfs_inode);
188 }
189
190 /*
191  * Create and initialize the inode cache
192  */
193 static int init_inodecache(void)
194 {
195         exofs_inode_cachep = kmem_cache_create("exofs_inode_cache",
196                                 sizeof(struct exofs_i_info), 0,
197                                 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
198                                 exofs_init_once);
199         if (exofs_inode_cachep == NULL)
200                 return -ENOMEM;
201         return 0;
202 }
203
204 /*
205  * Destroy the inode cache
206  */
207 static void destroy_inodecache(void)
208 {
209         kmem_cache_destroy(exofs_inode_cachep);
210 }
211
212 /******************************************************************************
213  * Some osd helpers
214  *****************************************************************************/
215 void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
216 {
217         osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
218 }
219
220 static int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
221                     u64 offset, void *p, unsigned length)
222 {
223         struct osd_request *or = osd_start_request(od, GFP_KERNEL);
224 /*      struct osd_sense_info osi = {.key = 0};*/
225         int ret;
226
227         if (unlikely(!or)) {
228                 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
229                 return -ENOMEM;
230         }
231         ret = osd_req_read_kern(or, obj, offset, p, length);
232         if (unlikely(ret)) {
233                 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
234                 goto out;
235         }
236
237         ret = osd_finalize_request(or, 0, cred, NULL);
238         if (unlikely(ret)) {
239                 EXOFS_DBGMSG("Failed to osd_finalize_request() => %d\n", ret);
240                 goto out;
241         }
242
243         ret = osd_execute_request(or);
244         if (unlikely(ret))
245                 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
246         /* osd_req_decode_sense(or, ret); */
247
248 out:
249         osd_end_request(or);
250         EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
251                       "length=0x%llx dev=%p ret=>%d\n",
252                       _LLU(obj->id), _LLU(offset), _LLU(length), od, ret);
253         return ret;
254 }
255
256 static const struct osd_attr g_attr_sb_stats = ATTR_DEF(
257         EXOFS_APAGE_SB_DATA,
258         EXOFS_ATTR_SB_STATS,
259         sizeof(struct exofs_sb_stats));
260
261 static int __sbi_read_stats(struct exofs_sb_info *sbi)
262 {
263         struct osd_attr attrs[] = {
264                 [0] = g_attr_sb_stats,
265         };
266         struct ore_io_state *ios;
267         int ret;
268
269         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
270         if (unlikely(ret)) {
271                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
272                 return ret;
273         }
274
275         ios->in_attr = attrs;
276         ios->in_attr_len = ARRAY_SIZE(attrs);
277
278         ret = ore_read(ios);
279         if (unlikely(ret)) {
280                 EXOFS_ERR("Error reading super_block stats => %d\n", ret);
281                 goto out;
282         }
283
284         ret = extract_attr_from_ios(ios, &attrs[0]);
285         if (ret) {
286                 EXOFS_ERR("%s: extract_attr of sb_stats failed\n", __func__);
287                 goto out;
288         }
289         if (attrs[0].len) {
290                 struct exofs_sb_stats *ess;
291
292                 if (unlikely(attrs[0].len != sizeof(*ess))) {
293                         EXOFS_ERR("%s: Wrong version of exofs_sb_stats "
294                                   "size(%d) != expected(%zd)\n",
295                                   __func__, attrs[0].len, sizeof(*ess));
296                         goto out;
297                 }
298
299                 ess = attrs[0].val_ptr;
300                 sbi->s_nextid = le64_to_cpu(ess->s_nextid);
301                 sbi->s_numfiles = le32_to_cpu(ess->s_numfiles);
302         }
303
304 out:
305         ore_put_io_state(ios);
306         return ret;
307 }
308
309 static void stats_done(struct ore_io_state *ios, void *p)
310 {
311         ore_put_io_state(ios);
312         /* Good thanks nothing to do anymore */
313 }
314
315 /* Asynchronously write the stats attribute */
316 int exofs_sbi_write_stats(struct exofs_sb_info *sbi)
317 {
318         struct osd_attr attrs[] = {
319                 [0] = g_attr_sb_stats,
320         };
321         struct ore_io_state *ios;
322         int ret;
323
324         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
325         if (unlikely(ret)) {
326                 EXOFS_ERR("%s: ore_get_io_state failed.\n", __func__);
327                 return ret;
328         }
329
330         sbi->s_ess.s_nextid   = cpu_to_le64(sbi->s_nextid);
331         sbi->s_ess.s_numfiles = cpu_to_le64(sbi->s_numfiles);
332         attrs[0].val_ptr = &sbi->s_ess;
333
334
335         ios->done = stats_done;
336         ios->private = sbi;
337         ios->out_attr = attrs;
338         ios->out_attr_len = ARRAY_SIZE(attrs);
339
340         ret = ore_write(ios);
341         if (unlikely(ret)) {
342                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
343                 ore_put_io_state(ios);
344         }
345
346         return ret;
347 }
348
349 /******************************************************************************
350  * SUPERBLOCK FUNCTIONS
351  *****************************************************************************/
352 static const struct super_operations exofs_sops;
353 static const struct export_operations exofs_export_ops;
354
355 /*
356  * Write the superblock to the OSD
357  */
358 static int exofs_sync_fs(struct super_block *sb, int wait)
359 {
360         struct exofs_sb_info *sbi;
361         struct exofs_fscb *fscb;
362         struct ore_comp one_comp;
363         struct ore_components oc;
364         struct ore_io_state *ios;
365         int ret = -ENOMEM;
366
367         fscb = kmalloc(sizeof(*fscb), GFP_KERNEL);
368         if (unlikely(!fscb))
369                 return -ENOMEM;
370
371         sbi = sb->s_fs_info;
372
373         /* NOTE: We no longer dirty the super_block anywhere in exofs. The
374          * reason we write the fscb here on unmount is so we can stay backwards
375          * compatible with fscb->s_version == 1. (What we are not compatible
376          * with is if a new version FS crashed and then we try to mount an old
377          * version). Otherwise the exofs_fscb is read-only from mkfs time. All
378          * the writeable info is set in exofs_sbi_write_stats() above.
379          */
380
381         exofs_init_comps(&oc, &one_comp, sbi, EXOFS_SUPER_ID);
382
383         ret = ore_get_io_state(&sbi->layout, &oc, &ios);
384         if (unlikely(ret))
385                 goto out;
386
387         lock_super(sb);
388
389         ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
390         memset(fscb, 0, ios->length);
391         fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
392         fscb->s_numfiles = cpu_to_le64(sbi->s_numfiles);
393         fscb->s_magic = cpu_to_le16(sb->s_magic);
394         fscb->s_newfs = 0;
395         fscb->s_version = EXOFS_FSCB_VER;
396
397         ios->offset = 0;
398         ios->kern_buff = fscb;
399
400         ret = ore_write(ios);
401         if (unlikely(ret))
402                 EXOFS_ERR("%s: ore_write failed.\n", __func__);
403         else
404                 sb->s_dirt = 0;
405
406
407         unlock_super(sb);
408 out:
409         EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
410         ore_put_io_state(ios);
411         kfree(fscb);
412         return ret;
413 }
414
415 static void exofs_write_super(struct super_block *sb)
416 {
417         if (!(sb->s_flags & MS_RDONLY))
418                 exofs_sync_fs(sb, 1);
419         else
420                 sb->s_dirt = 0;
421 }
422
423 static void _exofs_print_device(const char *msg, const char *dev_path,
424                                 struct osd_dev *od, u64 pid)
425 {
426         const struct osd_dev_info *odi = osduld_device_info(od);
427
428         printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
429                 msg, dev_path ?: "", odi->osdname, _LLU(pid));
430 }
431
432 static void exofs_free_sbi(struct exofs_sb_info *sbi)
433 {
434         unsigned numdevs = sbi->oc.numdevs;
435
436         while (numdevs) {
437                 unsigned i = --numdevs;
438                 struct osd_dev *od = ore_comp_dev(&sbi->oc, i);
439
440                 if (od) {
441                         ore_comp_set_dev(&sbi->oc, i, NULL);
442                         osduld_put_device(od);
443                 }
444         }
445         kfree(sbi->oc.ods);
446         kfree(sbi);
447 }
448
449 /*
450  * This function is called when the vfs is freeing the superblock.  We just
451  * need to free our own part.
452  */
453 static void exofs_put_super(struct super_block *sb)
454 {
455         int num_pend;
456         struct exofs_sb_info *sbi = sb->s_fs_info;
457
458         /* make sure there are no pending commands */
459         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
460              num_pend = atomic_read(&sbi->s_curr_pending)) {
461                 wait_queue_head_t wq;
462
463                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
464                        "This is a BUG. please report to osd-dev@open-osd.org\n",
465                        __func__);
466                 init_waitqueue_head(&wq);
467                 wait_event_timeout(wq,
468                                   (atomic_read(&sbi->s_curr_pending) == 0),
469                                   msecs_to_jiffies(100));
470         }
471
472         _exofs_print_device("Unmounting", NULL, ore_comp_dev(&sbi->oc, 0),
473                             sbi->one_comp.obj.partition);
474
475         bdi_destroy(&sbi->bdi);
476         exofs_free_sbi(sbi);
477         sb->s_fs_info = NULL;
478 }
479
480 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
481                                     struct exofs_device_table *dt)
482 {
483         int ret;
484
485         sbi->layout.stripe_unit =
486                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
487         sbi->layout.group_width =
488                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
489         sbi->layout.group_depth =
490                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
491         sbi->layout.mirrors_p1  =
492                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt) + 1;
493         sbi->layout.raid_algorithm  =
494                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
495
496         ret = ore_verify_layout(numdevs, &sbi->layout);
497
498         EXOFS_DBGMSG("exofs: layout: "
499                 "num_comps=%u stripe_unit=0x%x group_width=%u "
500                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
501                 numdevs,
502                 sbi->layout.stripe_unit,
503                 sbi->layout.group_width,
504                 _LLU(sbi->layout.group_depth),
505                 sbi->layout.mirrors_p1,
506                 sbi->layout.raid_algorithm);
507         return ret;
508 }
509
510 static unsigned __ra_pages(struct ore_layout *layout)
511 {
512         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
513         unsigned ra_pages = layout->group_width * layout->stripe_unit /
514                                 PAGE_SIZE;
515         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
516
517         ra_pages *= 2; /* two stripes */
518         if (ra_pages < _MIN_RA)
519                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
520
521         if (ra_pages > max_io_pages)
522                 ra_pages = max_io_pages;
523
524         return ra_pages;
525 }
526
527 /* @odi is valid only as long as @fscb_dev is valid */
528 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
529                              struct osd_dev_info *odi)
530 {
531         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
532         if (likely(odi->systemid_len))
533                 memcpy(odi->systemid, dt_dev->systemid, OSD_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: failed 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.numdevs = 1;
749         sbi->oc.single_comp = EC_SINGLE_COMP;
750         sbi->oc.comps = &sbi->one_comp;
751
752         /* fill in some other data by hand */
753         memset(sb->s_id, 0, sizeof(sb->s_id));
754         strcpy(sb->s_id, "exofs");
755         sb->s_blocksize = EXOFS_BLKSIZE;
756         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
757         sb->s_maxbytes = MAX_LFS_FILESIZE;
758         sb->s_max_links = EXOFS_LINK_MAX;
759         atomic_set(&sbi->s_curr_pending, 0);
760         sb->s_bdev = NULL;
761         sb->s_dev = 0;
762
763         comp.obj.partition = sbi->one_comp.obj.partition;
764         comp.obj.id = EXOFS_SUPER_ID;
765         exofs_make_credential(comp.cred, &comp.obj);
766
767         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
768         if (unlikely(ret))
769                 goto free_sbi;
770
771         sb->s_magic = le16_to_cpu(fscb.s_magic);
772         /* NOTE: we read below to be backward compatible with old versions */
773         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
774         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
775
776         /* make sure what we read from the object store is correct */
777         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
778                 if (!silent)
779                         EXOFS_ERR("ERROR: Bad magic value\n");
780                 ret = -EINVAL;
781                 goto free_sbi;
782         }
783         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
784                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
785                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
786                 ret = -EINVAL;
787                 goto free_sbi;
788         }
789
790         /* start generation numbers from a random point */
791         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
792         spin_lock_init(&sbi->s_next_gen_lock);
793
794         table_count = le64_to_cpu(fscb.s_dev_table_count);
795         if (table_count) {
796                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
797                 if (unlikely(ret))
798                         goto free_sbi;
799         } else {
800                 struct exofs_dev *eds;
801
802                 ret = __alloc_dev_table(sbi, 1, &eds);
803                 if (unlikely(ret))
804                         goto free_sbi;
805
806                 ore_comp_set_dev(&sbi->oc, 0, od);
807         }
808
809         __sbi_read_stats(sbi);
810
811         /* set up operation vectors */
812         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
813         sb->s_bdi = &sbi->bdi;
814         sb->s_fs_info = sbi;
815         sb->s_op = &exofs_sops;
816         sb->s_export_op = &exofs_export_ops;
817         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
818         if (IS_ERR(root)) {
819                 EXOFS_ERR("ERROR: exofs_iget failed\n");
820                 ret = PTR_ERR(root);
821                 goto free_sbi;
822         }
823         sb->s_root = d_make_root(root);
824         if (!sb->s_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                 dput(sb->s_root);
843                 sb->s_root = NULL;
844                 goto free_sbi;
845         }
846
847         _exofs_print_device("Mounting", opts->dev_name,
848                             ore_comp_dev(&sbi->oc, 0),
849                             sbi->one_comp.obj.partition);
850         return 0;
851
852 free_sbi:
853         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
854                   opts->dev_name, sbi->one_comp.obj.partition, ret);
855         exofs_free_sbi(sbi);
856         return ret;
857 }
858
859 /*
860  * Set up the superblock (calls exofs_fill_super eventually)
861  */
862 static struct dentry *exofs_mount(struct file_system_type *type,
863                           int flags, const char *dev_name,
864                           void *data)
865 {
866         struct exofs_mountopt opts;
867         int ret;
868
869         ret = parse_options(data, &opts);
870         if (ret)
871                 return ERR_PTR(ret);
872
873         if (!opts.dev_name)
874                 opts.dev_name = dev_name;
875         return mount_nodev(type, flags, &opts, exofs_fill_super);
876 }
877
878 /*
879  * Return information about the file system state in the buffer.  This is used
880  * by the 'df' command, for example.
881  */
882 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
883 {
884         struct super_block *sb = dentry->d_sb;
885         struct exofs_sb_info *sbi = sb->s_fs_info;
886         struct ore_io_state *ios;
887         struct osd_attr attrs[] = {
888                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
889                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
890                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
891                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
892         };
893         uint64_t capacity = ULLONG_MAX;
894         uint64_t used = ULLONG_MAX;
895         int ret;
896
897         ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
898         if (ret) {
899                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
900                 return ret;
901         }
902
903         ios->in_attr = attrs;
904         ios->in_attr_len = ARRAY_SIZE(attrs);
905
906         ret = ore_read(ios);
907         if (unlikely(ret))
908                 goto out;
909
910         ret = extract_attr_from_ios(ios, &attrs[0]);
911         if (likely(!ret)) {
912                 capacity = get_unaligned_be64(attrs[0].val_ptr);
913                 if (unlikely(!capacity))
914                         capacity = ULLONG_MAX;
915         } else
916                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
917
918         ret = extract_attr_from_ios(ios, &attrs[1]);
919         if (likely(!ret))
920                 used = get_unaligned_be64(attrs[1].val_ptr);
921         else
922                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
923
924         /* fill in the stats buffer */
925         buf->f_type = EXOFS_SUPER_MAGIC;
926         buf->f_bsize = EXOFS_BLKSIZE;
927         buf->f_blocks = capacity >> 9;
928         buf->f_bfree = (capacity - used) >> 9;
929         buf->f_bavail = buf->f_bfree;
930         buf->f_files = sbi->s_numfiles;
931         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
932         buf->f_namelen = EXOFS_NAME_LEN;
933
934 out:
935         ore_put_io_state(ios);
936         return ret;
937 }
938
939 static const struct super_operations exofs_sops = {
940         .alloc_inode    = exofs_alloc_inode,
941         .destroy_inode  = exofs_destroy_inode,
942         .write_inode    = exofs_write_inode,
943         .evict_inode    = exofs_evict_inode,
944         .put_super      = exofs_put_super,
945         .write_super    = exofs_write_super,
946         .sync_fs        = exofs_sync_fs,
947         .statfs         = exofs_statfs,
948 };
949
950 /******************************************************************************
951  * EXPORT OPERATIONS
952  *****************************************************************************/
953
954 static struct dentry *exofs_get_parent(struct dentry *child)
955 {
956         unsigned long ino = exofs_parent_ino(child);
957
958         if (!ino)
959                 return ERR_PTR(-ESTALE);
960
961         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
962 }
963
964 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
965                 u64 ino, u32 generation)
966 {
967         struct inode *inode;
968
969         inode = exofs_iget(sb, ino);
970         if (IS_ERR(inode))
971                 return ERR_CAST(inode);
972         if (generation && inode->i_generation != generation) {
973                 /* we didn't find the right inode.. */
974                 iput(inode);
975                 return ERR_PTR(-ESTALE);
976         }
977         return inode;
978 }
979
980 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
981                                 struct fid *fid, int fh_len, int fh_type)
982 {
983         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
984                                     exofs_nfs_get_inode);
985 }
986
987 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
988                                 struct fid *fid, int fh_len, int fh_type)
989 {
990         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
991                                     exofs_nfs_get_inode);
992 }
993
994 static const struct export_operations exofs_export_ops = {
995         .fh_to_dentry = exofs_fh_to_dentry,
996         .fh_to_parent = exofs_fh_to_parent,
997         .get_parent = exofs_get_parent,
998 };
999
1000 /******************************************************************************
1001  * INSMOD/RMMOD
1002  *****************************************************************************/
1003
1004 /*
1005  * struct that describes this file system
1006  */
1007 static struct file_system_type exofs_type = {
1008         .owner          = THIS_MODULE,
1009         .name           = "exofs",
1010         .mount          = exofs_mount,
1011         .kill_sb        = generic_shutdown_super,
1012 };
1013
1014 static int __init init_exofs(void)
1015 {
1016         int err;
1017
1018         err = init_inodecache();
1019         if (err)
1020                 goto out;
1021
1022         err = register_filesystem(&exofs_type);
1023         if (err)
1024                 goto out_d;
1025
1026         return 0;
1027 out_d:
1028         destroy_inodecache();
1029 out:
1030         return err;
1031 }
1032
1033 static void __exit exit_exofs(void)
1034 {
1035         unregister_filesystem(&exofs_type);
1036         destroy_inodecache();
1037 }
1038
1039 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1040 MODULE_DESCRIPTION("exofs");
1041 MODULE_LICENSE("GPL");
1042
1043 module_init(init_exofs)
1044 module_exit(exit_exofs)