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 */
parse_options(char * options,struct exofs_mountopt * opts)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 */
exofs_alloc_inode(struct super_block * sb)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
exofs_i_callback(struct rcu_head * head)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 */
exofs_destroy_inode(struct inode * inode)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 */
exofs_init_once(void * foo)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 */
init_inodecache(void)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 */
destroy_inodecache(void)207 static void destroy_inodecache(void)
208 {
209 kmem_cache_destroy(exofs_inode_cachep);
210 }
211
212 /******************************************************************************
213 * Some osd helpers
214 *****************************************************************************/
exofs_make_credential(u8 cred_a[OSD_CAP_LEN],const struct osd_obj_id * obj)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
exofs_read_kern(struct osd_dev * od,u8 * cred,struct osd_obj_id * obj,u64 offset,void * p,unsigned length)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
__sbi_read_stats(struct exofs_sb_info * sbi)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
stats_done(struct ore_io_state * ios,void * p)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 */
exofs_sbi_write_stats(struct exofs_sb_info * sbi)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 */
exofs_sync_fs(struct super_block * sb,int wait)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_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
exofs_write_super(struct super_block * sb)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
_exofs_print_device(const char * msg,const char * dev_path,struct osd_dev * od,u64 pid)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
exofs_free_sbi(struct exofs_sb_info * sbi)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 */
exofs_put_super(struct super_block * sb)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
_read_and_match_data_map(struct exofs_sb_info * sbi,unsigned numdevs,struct exofs_device_table * dt)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
__ra_pages(struct ore_layout * layout)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 */
exofs_devs_2_odi(struct exofs_dt_device_info * dt_dev,struct osd_dev_info * odi)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 memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
533
534 odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
535 odi->osdname = dt_dev->osdname;
536
537 /* FIXME support long names. Will need a _put function */
538 if (dt_dev->long_name_offset)
539 return -EINVAL;
540
541 /* Make sure osdname is printable!
542 * mkexofs should give us space for a null-terminator else the
543 * device-table is invalid.
544 */
545 if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
546 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
547 dt_dev->osdname[odi->osdname_len] = 0;
548
549 /* If it's all zeros something is bad we read past end-of-obj */
550 return !(odi->systemid_len || odi->osdname_len);
551 }
552
__alloc_dev_table(struct exofs_sb_info * sbi,unsigned numdevs,struct exofs_dev ** peds)553 int __alloc_dev_table(struct exofs_sb_info *sbi, unsigned numdevs,
554 struct exofs_dev **peds)
555 {
556 struct __alloc_ore_devs_and_exofs_devs {
557 /* Twice bigger table: See exofs_init_comps() and comment at
558 * exofs_read_lookup_dev_table()
559 */
560 struct ore_dev *oreds[numdevs * 2 - 1];
561 struct exofs_dev eds[numdevs];
562 } *aoded;
563 struct exofs_dev *eds;
564 unsigned i;
565
566 aoded = kzalloc(sizeof(*aoded), GFP_KERNEL);
567 if (unlikely(!aoded)) {
568 EXOFS_ERR("ERROR: faild allocating Device array[%d]\n",
569 numdevs);
570 return -ENOMEM;
571 }
572
573 sbi->oc.ods = aoded->oreds;
574 *peds = eds = aoded->eds;
575 for (i = 0; i < numdevs; ++i)
576 aoded->oreds[i] = &eds[i].ored;
577 return 0;
578 }
579
exofs_read_lookup_dev_table(struct exofs_sb_info * sbi,struct osd_dev * fscb_od,unsigned table_count)580 static int exofs_read_lookup_dev_table(struct exofs_sb_info *sbi,
581 struct osd_dev *fscb_od,
582 unsigned table_count)
583 {
584 struct ore_comp comp;
585 struct exofs_device_table *dt;
586 struct exofs_dev *eds;
587 unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
588 sizeof(*dt);
589 unsigned numdevs, i;
590 int ret;
591
592 dt = kmalloc(table_bytes, GFP_KERNEL);
593 if (unlikely(!dt)) {
594 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
595 table_bytes);
596 return -ENOMEM;
597 }
598
599 sbi->oc.numdevs = 0;
600
601 comp.obj.partition = sbi->one_comp.obj.partition;
602 comp.obj.id = EXOFS_DEVTABLE_ID;
603 exofs_make_credential(comp.cred, &comp.obj);
604
605 ret = exofs_read_kern(fscb_od, comp.cred, &comp.obj, 0, dt,
606 table_bytes);
607 if (unlikely(ret)) {
608 EXOFS_ERR("ERROR: reading device table\n");
609 goto out;
610 }
611
612 numdevs = le64_to_cpu(dt->dt_num_devices);
613 if (unlikely(!numdevs)) {
614 ret = -EINVAL;
615 goto out;
616 }
617 WARN_ON(table_count != numdevs);
618
619 ret = _read_and_match_data_map(sbi, numdevs, dt);
620 if (unlikely(ret))
621 goto out;
622
623 ret = __alloc_dev_table(sbi, numdevs, &eds);
624 if (unlikely(ret))
625 goto out;
626 /* exofs round-robins the device table view according to inode
627 * number. We hold a: twice bigger table hence inodes can point
628 * to any device and have a sequential view of the table
629 * starting at this device. See exofs_init_comps()
630 */
631 memcpy(&sbi->oc.ods[numdevs], &sbi->oc.ods[0],
632 (numdevs - 1) * sizeof(sbi->oc.ods[0]));
633
634 for (i = 0; i < numdevs; i++) {
635 struct exofs_fscb fscb;
636 struct osd_dev_info odi;
637 struct osd_dev *od;
638
639 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
640 EXOFS_ERR("ERROR: Read all-zeros device entry\n");
641 ret = -EINVAL;
642 goto out;
643 }
644
645 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
646 i, odi.osdname);
647
648 /* the exofs id is currently the table index */
649 eds[i].did = i;
650
651 /* On all devices the device table is identical. The user can
652 * specify any one of the participating devices on the command
653 * line. We always keep them in device-table order.
654 */
655 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
656 eds[i].ored.od = fscb_od;
657 ++sbi->oc.numdevs;
658 fscb_od = NULL;
659 continue;
660 }
661
662 od = osduld_info_lookup(&odi);
663 if (IS_ERR(od)) {
664 ret = PTR_ERR(od);
665 EXOFS_ERR("ERROR: device requested is not found "
666 "osd_name-%s =>%d\n", odi.osdname, ret);
667 goto out;
668 }
669
670 eds[i].ored.od = od;
671 ++sbi->oc.numdevs;
672
673 /* Read the fscb of the other devices to make sure the FS
674 * partition is there.
675 */
676 ret = exofs_read_kern(od, comp.cred, &comp.obj, 0, &fscb,
677 sizeof(fscb));
678 if (unlikely(ret)) {
679 EXOFS_ERR("ERROR: Malformed participating device "
680 "error reading fscb osd_name-%s\n",
681 odi.osdname);
682 goto out;
683 }
684
685 /* TODO: verify other information is correct and FS-uuid
686 * matches. Benny what did you say about device table
687 * generation and old devices?
688 */
689 }
690
691 out:
692 kfree(dt);
693 if (unlikely(fscb_od && !ret)) {
694 EXOFS_ERR("ERROR: Bad device-table container device not present\n");
695 osduld_put_device(fscb_od);
696 return -EINVAL;
697 }
698 return ret;
699 }
700
701 /*
702 * Read the superblock from the OSD and fill in the fields
703 */
exofs_fill_super(struct super_block * sb,void * data,int silent)704 static int exofs_fill_super(struct super_block *sb, void *data, int silent)
705 {
706 struct inode *root;
707 struct exofs_mountopt *opts = data;
708 struct exofs_sb_info *sbi; /*extended info */
709 struct osd_dev *od; /* Master device */
710 struct exofs_fscb fscb; /*on-disk superblock info */
711 struct ore_comp comp;
712 unsigned table_count;
713 int ret;
714
715 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
716 if (!sbi)
717 return -ENOMEM;
718
719 /* use mount options to fill superblock */
720 if (opts->is_osdname) {
721 struct osd_dev_info odi = {.systemid_len = 0};
722
723 odi.osdname_len = strlen(opts->dev_name);
724 odi.osdname = (u8 *)opts->dev_name;
725 od = osduld_info_lookup(&odi);
726 kfree(opts->dev_name);
727 opts->dev_name = NULL;
728 } else {
729 od = osduld_path_lookup(opts->dev_name);
730 }
731 if (IS_ERR(od)) {
732 ret = -EINVAL;
733 goto free_sbi;
734 }
735
736 /* Default layout in case we do not have a device-table */
737 sbi->layout.stripe_unit = PAGE_SIZE;
738 sbi->layout.mirrors_p1 = 1;
739 sbi->layout.group_width = 1;
740 sbi->layout.group_depth = -1;
741 sbi->layout.group_count = 1;
742 sbi->s_timeout = opts->timeout;
743
744 sbi->one_comp.obj.partition = opts->pid;
745 sbi->one_comp.obj.id = 0;
746 exofs_make_credential(sbi->one_comp.cred, &sbi->one_comp.obj);
747 sbi->oc.numdevs = 1;
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 }
806
807 __sbi_read_stats(sbi);
808
809 /* set up operation vectors */
810 sbi->bdi.ra_pages = __ra_pages(&sbi->layout);
811 sb->s_bdi = &sbi->bdi;
812 sb->s_fs_info = sbi;
813 sb->s_op = &exofs_sops;
814 sb->s_export_op = &exofs_export_ops;
815 root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
816 if (IS_ERR(root)) {
817 EXOFS_ERR("ERROR: exofs_iget failed\n");
818 ret = PTR_ERR(root);
819 goto free_sbi;
820 }
821 sb->s_root = d_alloc_root(root);
822 if (!sb->s_root) {
823 iput(root);
824 EXOFS_ERR("ERROR: get root inode failed\n");
825 ret = -ENOMEM;
826 goto free_sbi;
827 }
828
829 if (!S_ISDIR(root->i_mode)) {
830 dput(sb->s_root);
831 sb->s_root = NULL;
832 EXOFS_ERR("ERROR: corrupt root inode (mode = %hd)\n",
833 root->i_mode);
834 ret = -EINVAL;
835 goto free_sbi;
836 }
837
838 ret = bdi_setup_and_register(&sbi->bdi, "exofs", BDI_CAP_MAP_COPY);
839 if (ret) {
840 EXOFS_DBGMSG("Failed to bdi_setup_and_register\n");
841 dput(sb->s_root);
842 sb->s_root = NULL;
843 goto free_sbi;
844 }
845
846 _exofs_print_device("Mounting", opts->dev_name,
847 ore_comp_dev(&sbi->oc, 0),
848 sbi->one_comp.obj.partition);
849 return 0;
850
851 free_sbi:
852 EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
853 opts->dev_name, sbi->one_comp.obj.partition, ret);
854 exofs_free_sbi(sbi);
855 return ret;
856 }
857
858 /*
859 * Set up the superblock (calls exofs_fill_super eventually)
860 */
exofs_mount(struct file_system_type * type,int flags,const char * dev_name,void * data)861 static struct dentry *exofs_mount(struct file_system_type *type,
862 int flags, const char *dev_name,
863 void *data)
864 {
865 struct exofs_mountopt opts;
866 int ret;
867
868 ret = parse_options(data, &opts);
869 if (ret)
870 return ERR_PTR(ret);
871
872 if (!opts.dev_name)
873 opts.dev_name = dev_name;
874 return mount_nodev(type, flags, &opts, exofs_fill_super);
875 }
876
877 /*
878 * Return information about the file system state in the buffer. This is used
879 * by the 'df' command, for example.
880 */
exofs_statfs(struct dentry * dentry,struct kstatfs * buf)881 static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
882 {
883 struct super_block *sb = dentry->d_sb;
884 struct exofs_sb_info *sbi = sb->s_fs_info;
885 struct ore_io_state *ios;
886 struct osd_attr attrs[] = {
887 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
888 OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
889 ATTR_DEF(OSD_APAGE_PARTITION_INFORMATION,
890 OSD_ATTR_PI_USED_CAPACITY, sizeof(__be64)),
891 };
892 uint64_t capacity = ULLONG_MAX;
893 uint64_t used = ULLONG_MAX;
894 int ret;
895
896 ret = ore_get_io_state(&sbi->layout, &sbi->oc, &ios);
897 if (ret) {
898 EXOFS_DBGMSG("ore_get_io_state failed.\n");
899 return ret;
900 }
901
902 ios->in_attr = attrs;
903 ios->in_attr_len = ARRAY_SIZE(attrs);
904
905 ret = ore_read(ios);
906 if (unlikely(ret))
907 goto out;
908
909 ret = extract_attr_from_ios(ios, &attrs[0]);
910 if (likely(!ret)) {
911 capacity = get_unaligned_be64(attrs[0].val_ptr);
912 if (unlikely(!capacity))
913 capacity = ULLONG_MAX;
914 } else
915 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
916
917 ret = extract_attr_from_ios(ios, &attrs[1]);
918 if (likely(!ret))
919 used = get_unaligned_be64(attrs[1].val_ptr);
920 else
921 EXOFS_DBGMSG("exofs_statfs: get used-space failed.\n");
922
923 /* fill in the stats buffer */
924 buf->f_type = EXOFS_SUPER_MAGIC;
925 buf->f_bsize = EXOFS_BLKSIZE;
926 buf->f_blocks = capacity >> 9;
927 buf->f_bfree = (capacity - used) >> 9;
928 buf->f_bavail = buf->f_bfree;
929 buf->f_files = sbi->s_numfiles;
930 buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
931 buf->f_namelen = EXOFS_NAME_LEN;
932
933 out:
934 ore_put_io_state(ios);
935 return ret;
936 }
937
938 static const struct super_operations exofs_sops = {
939 .alloc_inode = exofs_alloc_inode,
940 .destroy_inode = exofs_destroy_inode,
941 .write_inode = exofs_write_inode,
942 .evict_inode = exofs_evict_inode,
943 .put_super = exofs_put_super,
944 .write_super = exofs_write_super,
945 .sync_fs = exofs_sync_fs,
946 .statfs = exofs_statfs,
947 };
948
949 /******************************************************************************
950 * EXPORT OPERATIONS
951 *****************************************************************************/
952
exofs_get_parent(struct dentry * child)953 static struct dentry *exofs_get_parent(struct dentry *child)
954 {
955 unsigned long ino = exofs_parent_ino(child);
956
957 if (!ino)
958 return ERR_PTR(-ESTALE);
959
960 return d_obtain_alias(exofs_iget(child->d_inode->i_sb, ino));
961 }
962
exofs_nfs_get_inode(struct super_block * sb,u64 ino,u32 generation)963 static struct inode *exofs_nfs_get_inode(struct super_block *sb,
964 u64 ino, u32 generation)
965 {
966 struct inode *inode;
967
968 inode = exofs_iget(sb, ino);
969 if (IS_ERR(inode))
970 return ERR_CAST(inode);
971 if (generation && inode->i_generation != generation) {
972 /* we didn't find the right inode.. */
973 iput(inode);
974 return ERR_PTR(-ESTALE);
975 }
976 return inode;
977 }
978
exofs_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)979 static struct dentry *exofs_fh_to_dentry(struct super_block *sb,
980 struct fid *fid, int fh_len, int fh_type)
981 {
982 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
983 exofs_nfs_get_inode);
984 }
985
exofs_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)986 static struct dentry *exofs_fh_to_parent(struct super_block *sb,
987 struct fid *fid, int fh_len, int fh_type)
988 {
989 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
990 exofs_nfs_get_inode);
991 }
992
993 static const struct export_operations exofs_export_ops = {
994 .fh_to_dentry = exofs_fh_to_dentry,
995 .fh_to_parent = exofs_fh_to_parent,
996 .get_parent = exofs_get_parent,
997 };
998
999 /******************************************************************************
1000 * INSMOD/RMMOD
1001 *****************************************************************************/
1002
1003 /*
1004 * struct that describes this file system
1005 */
1006 static struct file_system_type exofs_type = {
1007 .owner = THIS_MODULE,
1008 .name = "exofs",
1009 .mount = exofs_mount,
1010 .kill_sb = generic_shutdown_super,
1011 };
1012
init_exofs(void)1013 static int __init init_exofs(void)
1014 {
1015 int err;
1016
1017 err = init_inodecache();
1018 if (err)
1019 goto out;
1020
1021 err = register_filesystem(&exofs_type);
1022 if (err)
1023 goto out_d;
1024
1025 return 0;
1026 out_d:
1027 destroy_inodecache();
1028 out:
1029 return err;
1030 }
1031
exit_exofs(void)1032 static void __exit exit_exofs(void)
1033 {
1034 unregister_filesystem(&exofs_type);
1035 destroy_inodecache();
1036 }
1037
1038 MODULE_AUTHOR("Avishay Traeger <avishay@gmail.com>");
1039 MODULE_DESCRIPTION("exofs");
1040 MODULE_LICENSE("GPL");
1041
1042 module_init(init_exofs)
1043 module_exit(exit_exofs)
1044