Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc
[~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/exportfs.h>
39 #include <linux/slab.h>
40
41 #include "exofs.h"
42
43 #define EXOFS_DBGMSG2(M...) do {} while (0)
44
45 /******************************************************************************
46  * MOUNT OPTIONS
47  *****************************************************************************/
48
49 /*
50  * struct to hold what we get from mount options
51  */
52 struct exofs_mountopt {
53         bool is_osdname;
54         const char *dev_name;
55         uint64_t pid;
56         int timeout;
57 };
58
59 /*
60  * exofs-specific mount-time options.
61  */
62 enum { Opt_name, Opt_pid, Opt_to, Opt_err };
63
64 /*
65  * Our mount-time options.  These should ideally be 64-bit unsigned, but the
66  * kernel's parsing functions do not currently support that.  32-bit should be
67  * sufficient for most applications now.
68  */
69 static match_table_t tokens = {
70         {Opt_name, "osdname=%s"},
71         {Opt_pid, "pid=%u"},
72         {Opt_to, "to=%u"},
73         {Opt_err, NULL}
74 };
75
76 /*
77  * The main option parsing method.  Also makes sure that all of the mandatory
78  * mount options were set.
79  */
80 static int parse_options(char *options, struct exofs_mountopt *opts)
81 {
82         char *p;
83         substring_t args[MAX_OPT_ARGS];
84         int option;
85         bool s_pid = false;
86
87         EXOFS_DBGMSG("parse_options %s\n", options);
88         /* defaults */
89         memset(opts, 0, sizeof(*opts));
90         opts->timeout = BLK_DEFAULT_SG_TIMEOUT;
91
92         while ((p = strsep(&options, ",")) != NULL) {
93                 int token;
94                 char str[32];
95
96                 if (!*p)
97                         continue;
98
99                 token = match_token(p, tokens, args);
100                 switch (token) {
101                 case Opt_name:
102                         opts->dev_name = match_strdup(&args[0]);
103                         if (unlikely(!opts->dev_name)) {
104                                 EXOFS_ERR("Error allocating dev_name");
105                                 return -ENOMEM;
106                         }
107                         opts->is_osdname = true;
108                         break;
109                 case Opt_pid:
110                         if (0 == match_strlcpy(str, &args[0], sizeof(str)))
111                                 return -EINVAL;
112                         opts->pid = simple_strtoull(str, NULL, 0);
113                         if (opts->pid < EXOFS_MIN_PID) {
114                                 EXOFS_ERR("Partition ID must be >= %u",
115                                           EXOFS_MIN_PID);
116                                 return -EINVAL;
117                         }
118                         s_pid = 1;
119                         break;
120                 case Opt_to:
121                         if (match_int(&args[0], &option))
122                                 return -EINVAL;
123                         if (option <= 0) {
124                                 EXOFS_ERR("Timout must be > 0");
125                                 return -EINVAL;
126                         }
127                         opts->timeout = option * HZ;
128                         break;
129                 }
130         }
131
132         if (!s_pid) {
133                 EXOFS_ERR("Need to specify the following options:\n");
134                 EXOFS_ERR("    -o pid=pid_no_to_use\n");
135                 return -EINVAL;
136         }
137
138         return 0;
139 }
140
141 /******************************************************************************
142  * INODE CACHE
143  *****************************************************************************/
144
145 /*
146  * Our inode cache.  Isn't it pretty?
147  */
148 static struct kmem_cache *exofs_inode_cachep;
149
150 /*
151  * Allocate an inode in the cache
152  */
153 static struct inode *exofs_alloc_inode(struct super_block *sb)
154 {
155         struct exofs_i_info *oi;
156
157         oi = kmem_cache_alloc(exofs_inode_cachep, GFP_KERNEL);
158         if (!oi)
159                 return NULL;
160
161         oi->vfs_inode.i_version = 1;
162         return &oi->vfs_inode;
163 }
164
165 static void exofs_i_callback(struct rcu_head *head)
166 {
167         struct inode *inode = container_of(head, struct inode, i_rcu);
168         INIT_LIST_HEAD(&inode->i_dentry);
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->comps, &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->comps, &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 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 comps;
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(&comps, &one_comp, sbi, EXOFS_SUPER_ID);
382
383         ret = ore_get_io_state(&sbi->layout, &comps, &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_le32(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 void exofs_free_sbi(struct exofs_sb_info *sbi)
433 {
434         while (sbi->comps.numdevs) {
435                 int i = --sbi->comps.numdevs;
436                 struct osd_dev *od = sbi->comps.ods[i];
437
438                 if (od) {
439                         sbi->comps.ods[i] = NULL;
440                         osduld_put_device(od);
441                 }
442         }
443         if (sbi->comps.ods != sbi->_min_one_dev)
444                 kfree(sbi->comps.ods);
445         kfree(sbi);
446 }
447
448 /*
449  * This function is called when the vfs is freeing the superblock.  We just
450  * need to free our own part.
451  */
452 static void exofs_put_super(struct super_block *sb)
453 {
454         int num_pend;
455         struct exofs_sb_info *sbi = sb->s_fs_info;
456
457         /* make sure there are no pending commands */
458         for (num_pend = atomic_read(&sbi->s_curr_pending); num_pend > 0;
459              num_pend = atomic_read(&sbi->s_curr_pending)) {
460                 wait_queue_head_t wq;
461
462                 printk(KERN_NOTICE "%s: !!Pending operations in flight. "
463                        "This is a BUG. please report to osd-dev@open-osd.org\n",
464                        __func__);
465                 init_waitqueue_head(&wq);
466                 wait_event_timeout(wq,
467                                   (atomic_read(&sbi->s_curr_pending) == 0),
468                                   msecs_to_jiffies(100));
469         }
470
471         _exofs_print_device("Unmounting", NULL, sbi->comps.ods[0],
472                             sbi->one_comp.obj.partition);
473
474         bdi_destroy(&sbi->bdi);
475         exofs_free_sbi(sbi);
476         sb->s_fs_info = NULL;
477 }
478
479 static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
480                                     struct exofs_device_table *dt)
481 {
482         u64 stripe_length;
483
484         sbi->data_map.odm_num_comps   =
485                                 le32_to_cpu(dt->dt_data_map.cb_num_comps);
486         sbi->data_map.odm_stripe_unit =
487                                 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
488         sbi->data_map.odm_group_width =
489                                 le32_to_cpu(dt->dt_data_map.cb_group_width);
490         sbi->data_map.odm_group_depth =
491                                 le32_to_cpu(dt->dt_data_map.cb_group_depth);
492         sbi->data_map.odm_mirror_cnt  =
493                                 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
494         sbi->data_map.odm_raid_algorithm  =
495                                 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
496
497 /* FIXME: Only raid0 for now. if not so, do not mount */
498         if (sbi->data_map.odm_num_comps != numdevs) {
499                 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
500                           sbi->data_map.odm_num_comps, numdevs);
501                 return -EINVAL;
502         }
503         if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
504                 EXOFS_ERR("Only RAID_0 for now\n");
505                 return -EINVAL;
506         }
507         if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
508                 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
509                           numdevs, sbi->data_map.odm_mirror_cnt);
510                 return -EINVAL;
511         }
512
513         if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
514                 EXOFS_ERR("Stripe Unit(0x%llx)"
515                           " must be Multples of PAGE_SIZE(0x%lx)\n",
516                           _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
517                 return -EINVAL;
518         }
519
520         sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
521         sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
522
523         if (sbi->data_map.odm_group_width) {
524                 sbi->layout.group_width = sbi->data_map.odm_group_width;
525                 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
526                 if (!sbi->layout.group_depth) {
527                         EXOFS_ERR("group_depth == 0 && group_width != 0\n");
528                         return -EINVAL;
529                 }
530                 sbi->layout.group_count = sbi->data_map.odm_num_comps /
531                                                 sbi->layout.mirrors_p1 /
532                                                 sbi->data_map.odm_group_width;
533         } else {
534                 if (sbi->data_map.odm_group_depth) {
535                         printk(KERN_NOTICE "Warning: group_depth ignored "
536                                 "group_width == 0 && group_depth == %d\n",
537                                 sbi->data_map.odm_group_depth);
538                         sbi->data_map.odm_group_depth = 0;
539                 }
540                 sbi->layout.group_width = sbi->data_map.odm_num_comps /
541                                                         sbi->layout.mirrors_p1;
542                 sbi->layout.group_depth = -1;
543                 sbi->layout.group_count = 1;
544         }
545
546         stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
547         if (stripe_length >= (1ULL << 32)) {
548                 EXOFS_ERR("Total Stripe length(0x%llx)"
549                           " >= 32bit is not supported\n", _LLU(stripe_length));
550                 return -EINVAL;
551         }
552
553         EXOFS_DBGMSG("exofs: layout: "
554                 "num_comps=%u stripe_unit=0x%x group_width=%u "
555                 "group_depth=0x%llx mirrors_p1=%u raid_algorithm=%u\n",
556                 numdevs,
557                 sbi->layout.stripe_unit,
558                 sbi->layout.group_width,
559                 _LLU(sbi->layout.group_depth),
560                 sbi->layout.mirrors_p1,
561                 sbi->data_map.odm_raid_algorithm);
562         return 0;
563 }
564
565 static unsigned __ra_pages(struct ore_layout *layout)
566 {
567         const unsigned _MIN_RA = 32; /* min 128K read-ahead */
568         unsigned ra_pages = layout->group_width * layout->stripe_unit /
569                                 PAGE_SIZE;
570         unsigned max_io_pages = exofs_max_io_pages(layout, ~0);
571
572         ra_pages *= 2; /* two stripes */
573         if (ra_pages < _MIN_RA)
574                 ra_pages = roundup(_MIN_RA, ra_pages / 2);
575
576         if (ra_pages > max_io_pages)
577                 ra_pages = max_io_pages;
578
579         return ra_pages;
580 }
581
582 /* @odi is valid only as long as @fscb_dev is valid */
583 static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
584                              struct osd_dev_info *odi)
585 {
586         odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
587         memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
588
589         odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
590         odi->osdname = dt_dev->osdname;
591
592         /* FIXME support long names. Will need a _put function */
593         if (dt_dev->long_name_offset)
594                 return -EINVAL;
595
596         /* Make sure osdname is printable!
597          * mkexofs should give us space for a null-terminator else the
598          * device-table is invalid.
599          */
600         if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
601                 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
602         dt_dev->osdname[odi->osdname_len] = 0;
603
604         /* If it's all zeros something is bad we read past end-of-obj */
605         return !(odi->systemid_len || odi->osdname_len);
606 }
607
608 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
609                                        struct osd_dev *fscb_od,
610                                        unsigned table_count)
611 {
612         struct ore_comp comp;
613         struct exofs_device_table *dt;
614         unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
615                                              sizeof(*dt);
616         unsigned numdevs, i;
617         int ret;
618
619         dt = kmalloc(table_bytes, GFP_KERNEL);
620         if (unlikely(!dt)) {
621                 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
622                           table_bytes);
623                 return -ENOMEM;
624         }
625
626         sbi->comps.numdevs = 0;
627
628         comp.obj.partition = sbi->one_comp.obj.partition;
629         comp.obj.id = EXOFS_DEVTABLE_ID;
630         exofs_make_credential(comp.cred, &comp.obj);
631
632         ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
633                               table_bytes);
634         if (unlikely(ret)) {
635                 EXOFS_ERR("ERROR: reading device table\n");
636                 goto out;
637         }
638
639         numdevs = le64_to_cpu(dt->dt_num_devices);
640         if (unlikely(!numdevs)) {
641                 ret = -EINVAL;
642                 goto out;
643         }
644         WARN_ON(table_count != numdevs);
645
646         ret = _read_and_match_data_map(sbi, numdevs, dt);
647         if (unlikely(ret))
648                 goto out;
649
650         if (likely(numdevs > 1)) {
651                 unsigned size = numdevs * sizeof(sbi->comps.ods[0]);
652
653                 /* Twice bigger table: See exofs_init_comps() and below
654                  * comment
655                  */
656                 sbi->comps.ods = kzalloc(size + size - 1, GFP_KERNEL);
657                 if (unlikely(!sbi->comps.ods)) {
658                         EXOFS_ERR("ERROR: faild allocating Device array[%d]\n",
659                                   numdevs);
660                         ret = -ENOMEM;
661                         goto out;
662                 }
663         }
664
665         for (i = 0; i < numdevs; i++) {
666                 struct exofs_fscb fscb;
667                 struct osd_dev_info odi;
668                 struct osd_dev *od;
669
670                 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
671                         EXOFS_ERR("ERROR: Read all-zeros device entry\n");
672                         ret = -EINVAL;
673                         goto out;
674                 }
675
676                 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
677                        i, odi.osdname);
678
679                 /* On all devices the device table is identical. The user can
680                  * specify any one of the participating devices on the command
681                  * line. We always keep them in device-table order.
682                  */
683                 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
684                         sbi->comps.ods[i] = fscb_od;
685                         ++sbi->comps.numdevs;
686                         fscb_od = NULL;
687                         continue;
688                 }
689
690                 od = osduld_info_lookup(&odi);
691                 if (IS_ERR(od)) {
692                         ret = PTR_ERR(od);
693                         EXOFS_ERR("ERROR: device requested is not found "
694                                   "osd_name-%s =>%d\n", odi.osdname, ret);
695                         goto out;
696                 }
697
698                 sbi->comps.ods[i] = od;
699                 ++sbi->comps.numdevs;
700
701                 /* Read the fscb of the other devices to make sure the FS
702                  * partition is there.
703                  */
704                 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
705                                       sizeof(fscb));
706                 if (unlikely(ret)) {
707                         EXOFS_ERR("ERROR: Malformed participating device "
708                                   "error reading fscb osd_name-%s\n",
709                                   odi.osdname);
710                         goto out;
711                 }
712
713                 /* TODO: verify other information is correct and FS-uuid
714                  *       matches. Benny what did you say about device table
715                  *       generation and old devices?
716                  */
717         }
718
719 out:
720         kfree(dt);
721         if (likely(!ret)) {
722                 unsigned numdevs = sbi->comps.numdevs;
723
724                 if (unlikely(fscb_od)) {
725                         EXOFS_ERR("ERROR: Bad device-table container device not present\n");
726                         osduld_put_device(fscb_od);
727                         return -EINVAL;
728                 }
729                 /* exofs round-robins the device table view according to inode
730                  * number. We hold a: twice bigger table hence inodes can point
731                  * to any device and have a sequential view of the table
732                  * starting at this device. See exofs_init_comps()
733                  */
734                 for (i = 0; i < numdevs - 1; ++i)
735                         sbi->comps.ods[i + numdevs] = sbi->comps.ods[i];
736         }
737         return ret;
738 }
739
740 /*
741  * Read the superblock from the OSD and fill in the fields
742  */
743 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
744 {
745         struct inode *root;
746         struct exofs_mountopt *opts = data;
747         struct exofs_sb_info *sbi;      /*extended info                  */
748         struct osd_dev *od;             /* Master device                 */
749         struct exofs_fscb fscb;         /*on-disk superblock info        */
750         struct ore_comp comp;
751         unsigned table_count;
752         int ret;
753
754         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
755         if (!sbi)
756                 return -ENOMEM;
757
758         /* use mount options to fill superblock */
759         if (opts->is_osdname) {
760                 struct osd_dev_info odi = {.systemid_len = 0};
761
762                 odi.osdname_len = strlen(opts->dev_name);
763                 odi.osdname = (u8 *)opts->dev_name;
764                 od = osduld_info_lookup(&odi);
765                 kfree(opts->dev_name);
766                 opts->dev_name = NULL;
767         } else {
768                 od = osduld_path_lookup(opts->dev_name);
769         }
770         if (IS_ERR(od)) {
771                 ret = -EINVAL;
772                 goto free_sbi;
773         }
774
775         /* Default layout in case we do not have a device-table */
776         sbi->layout.stripe_unit = PAGE_SIZE;
777         sbi->layout.mirrors_p1 = 1;
778         sbi->layout.group_width = 1;
779         sbi->layout.group_depth = -1;
780         sbi->layout.group_count = 1;
781         sbi->s_timeout = opts->timeout;
782
783         sbi->one_comp.obj.partition = opts->pid;
784         sbi->one_comp.obj.id = 0;
785         exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
786         sbi->comps.numdevs = 1;
787         sbi->comps.single_comp = EC_SINGLE_COMP;
788         sbi->comps.comps = &sbi->one_comp;
789         sbi->comps.ods = sbi->_min_one_dev;
790
791         /* fill in some other data by hand */
792         memset(sb->s_id, 0, sizeof(sb->s_id));
793         strcpy(sb->s_id, "exofs");
794         sb->s_blocksize = EXOFS_BLKSIZE;
795         sb->s_blocksize_bits = EXOFS_BLKSHIFT;
796         sb->s_maxbytes = MAX_LFS_FILESIZE;
797         atomic_set(&sbi->s_curr_pending, 0);
798         sb->s_bdev = NULL;
799         sb->s_dev = 0;
800
801         comp.obj.partition = sbi->one_comp.obj.partition;
802         comp.obj.id = EXOFS_SUPER_ID;
803         exofs_make_credential(comp.cred, &comp.obj);
804
805         ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb, sizeof(fscb));
806         if (unlikely(ret))
807                 goto free_sbi;
808
809         sb->s_magic = le16_to_cpu(fscb.s_magic);
810         /* NOTE: we read below to be backward compatible with old versions */
811         sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
812         sbi->s_numfiles = le32_to_cpu(fscb.s_numfiles);
813
814         /* make sure what we read from the object store is correct */
815         if (sb->s_magic != EXOFS_SUPER_MAGIC) {
816                 if (!silent)
817                         EXOFS_ERR("ERROR: Bad magic value\n");
818                 ret = -EINVAL;
819                 goto free_sbi;
820         }
821         if (le32_to_cpu(fscb.s_version) > EXOFS_FSCB_VER) {
822                 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
823                           EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
824                 ret = -EINVAL;
825                 goto free_sbi;
826         }
827
828         /* start generation numbers from a random point */
829         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
830         spin_lock_init(&sbi->s_next_gen_lock);
831
832         table_count = le64_to_cpu(fscb.s_dev_table_count);
833         if (table_count) {
834                 ret = exofs_read_lookup_dev_table(sbi, od, table_count);
835                 if (unlikely(ret))
836                         goto free_sbi;
837         } else {
838                 sbi->comps.ods[0] = od;
839         }
840
841         __sbi_read_stats(sbi);
842
843         /* set up operation vectors */
844         sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
845         sb->s_bdi = &sbi->bdi;
846         sb->s_fs_info = sbi;
847         sb->s_op = &exofs_sops;
848         sb->s_export_op = &exofs_export_ops;
849         root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
850         if (IS_ERR(root)) {
851                 EXOFS_ERR("ERROR: exofs_iget failed\n");
852                 ret = PTR_ERR(root);
853                 goto free_sbi;
854         }
855         sb->s_root = d_alloc_root(root);
856         if (!sb->s_root) {
857                 iput(root);
858                 EXOFS_ERR("ERROR: get root inode failed\n");
859                 ret = -ENOMEM;
860                 goto free_sbi;
861         }
862
863         if (!S_ISDIR(root->i_mode)) {
864                 dput(sb->s_root);
865                 sb->s_root = NULL;
866                 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
867                        root->i_mode);
868                 ret = -EINVAL;
869                 goto free_sbi;
870         }
871
872         ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
873         if (ret) {
874                 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
875                 goto free_sbi;
876         }
877
878         _exofs_print_device("Mounting", opts->dev_name, sbi->comps.ods[0],
879                             sbi->one_comp.obj.partition);
880         return 0;
881
882 free_sbi:
883         EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
884                   opts->dev_name, sbi->one_comp.obj.partition, ret);
885         exofs_free_sbi(sbi);
886         return ret;
887 }
888
889 /*
890  * Set up the superblock (calls exofs_fill_super eventually)
891  */
892 static struct dentry *exofs_mount(struct file_system_type *type,
893                           int flags, const char *dev_name,
894                           void *data)
895 {
896         struct exofs_mountopt opts;
897         int ret;
898
899         ret = parse_options(data, &opts);
900         if (ret)
901                 return ERR_PTR(ret);
902
903         if (!opts.dev_name)
904                 opts.dev_name = dev_name;
905         return mount_nodev(type, flags, &opts, exofs_fill_super);
906 }
907
908 /*
909  * Return information about the file system state in the buffer.  This is used
910  * by the 'df' command, for example.
911  */
912 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
913 {
914         struct super_block *sb = dentry->d_sb;
915         struct exofs_sb_info *sbi = sb->s_fs_info;
916         struct ore_io_state *ios;
917         struct osd_attr attrs[] = {
918                 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
919                         OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
920                 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
921                         OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
922         };
923         uint64_t capacity = ULLONG_MAX;
924         uint64_t used = ULLONG_MAX;
925         int ret;
926
927         ret = ore_get_io_state(&sbi->layout, &sbi->comps, &ios);
928         if (ret) {
929                 EXOFS_DBGMSG("ore_get_io_state failed.\n");
930                 return ret;
931         }
932
933         ios->in_attr = attrs;
934         ios->in_attr_len = ARRAY_SIZE(attrs);
935
936         ret = ore_read(ios);
937         if (unlikely(ret))
938                 goto out;
939
940         ret = extract_attr_from_ios(ios, &attrs[0]);
941         if (likely(!ret)) {
942                 capacity = get_unaligned_be64(attrs[0].val_ptr);
943                 if (unlikely(!capacity))
944                         capacity = ULLONG_MAX;
945         } else
946                 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
947
948         ret = extract_attr_from_ios(ios, &attrs[1]);
949         if (likely(!ret))
950                 used = get_unaligned_be64(attrs[1].val_ptr);
951         else
952                 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
953
954         /* fill in the stats buffer */
955         buf->f_type = EXOFS_SUPER_MAGIC;
956         buf->f_bsize = EXOFS_BLKSIZE;
957         buf->f_blocks = capacity >> 9;
958         buf->f_bfree = (capacity - used) >> 9;
959         buf->f_bavail = buf->f_bfree;
960         buf->f_files = sbi->s_numfiles;
961         buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
962         buf->f_namelen = EXOFS_NAME_LEN;
963
964 out:
965         ore_put_io_state(ios);
966         return ret;
967 }
968
969 static const struct super_operations exofs_sops = {
970         .alloc_inode    = exofs_alloc_inode,
971         .destroy_inode  = exofs_destroy_inode,
972         .write_inode    = exofs_write_inode,
973         .evict_inode    = exofs_evict_inode,
974         .put_super      = exofs_put_super,
975         .write_super    = exofs_write_super,
976         .sync_fs        = exofs_sync_fs,
977         .statfs         = exofs_statfs,
978 };
979
980 /******************************************************************************
981  * EXPORT OPERATIONS
982  *****************************************************************************/
983
984 struct dentry *exofs_get_parent(struct dentry *child)
985 {
986         unsigned long ino = exofs_parent_ino(child);
987
988         if (!ino)
989                 return ERR_PTR(-ESTALE);
990
991         return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
992 }
993
994 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
995                 u64 ino, u32 generation)
996 {
997         struct inode *inode;
998
999         inode = exofs_iget(sb, ino);
1000         if (IS_ERR(inode))
1001                 return ERR_CAST(inode);
1002         if (generation && inode->i_generation != generation) {
1003                 /* we didn't find the right inode.. */
1004                 iput(inode);
1005                 return ERR_PTR(-ESTALE);
1006         }
1007         return inode;
1008 }
1009
1010 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
1011                                 struct fid *fid, int fh_len, int fh_type)
1012 {
1013         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1014                                     exofs_nfs_get_inode);
1015 }
1016
1017 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
1018                                 struct fid *fid, int fh_len, int fh_type)
1019 {
1020         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1021                                     exofs_nfs_get_inode);
1022 }
1023
1024 static const struct export_operations exofs_export_ops = {
1025         .fh_to_dentry = exofs_fh_to_dentry,
1026         .fh_to_parent = exofs_fh_to_parent,
1027         .get_parent = exofs_get_parent,
1028 };
1029
1030 /******************************************************************************
1031  * INSMOD/RMMOD
1032  *****************************************************************************/
1033
1034 /*
1035  * struct that describes this file system
1036  */
1037 static struct file_system_type exofs_type = {
1038         .owner          = THIS_MODULE,
1039         .name           = "exofs",
1040         .mount          = exofs_mount,
1041         .kill_sb        = generic_shutdown_super,
1042 };
1043
1044 static int __init init_exofs(void)
1045 {
1046         int err;
1047
1048         err = init_inodecache();
1049         if (err)
1050                 goto out;
1051
1052         err = register_filesystem(&exofs_type);
1053         if (err)
1054                 goto out_d;
1055
1056         return 0;
1057 out_d:
1058         destroy_inodecache();
1059 out:
1060         return err;
1061 }
1062
1063 static void __exit exit_exofs(void)
1064 {
1065         unregister_filesystem(&exofs_type);
1066         destroy_inodecache();
1067 }
1068
1069 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1070 MODULE_DESCRIPTION("exofs");
1071 MODULE_LICENSE("GPL");
1072
1073 module_init(init_exofs)
1074 module_exit(exit_exofs)