Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Simple file system for zoned block devices exposing zones as files.
4 *
5 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
6 */
7#include <linux/module.h>
8#include <linux/pagemap.h>
9#include <linux/magic.h>
10#include <linux/iomap.h>
11#include <linux/init.h>
12#include <linux/slab.h>
13#include <linux/blkdev.h>
14#include <linux/statfs.h>
15#include <linux/writeback.h>
16#include <linux/quotaops.h>
17#include <linux/seq_file.h>
18#include <linux/parser.h>
19#include <linux/uio.h>
20#include <linux/mman.h>
21#include <linux/sched/mm.h>
22#include <linux/crc32.h>
23#include <linux/task_io_accounting_ops.h>
24
25#include "zonefs.h"
26
27#define CREATE_TRACE_POINTS
28#include "trace.h"
29
30/*
31 * Get the name of a zone group directory.
32 */
33static const char *zonefs_zgroup_name(enum zonefs_ztype ztype)
34{
35 switch (ztype) {
36 case ZONEFS_ZTYPE_CNV:
37 return "cnv";
38 case ZONEFS_ZTYPE_SEQ:
39 return "seq";
40 default:
41 WARN_ON_ONCE(1);
42 return "???";
43 }
44}
45
46/*
47 * Manage the active zone count.
48 */
49static void zonefs_account_active(struct super_block *sb,
50 struct zonefs_zone *z)
51{
52 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
53
54 if (zonefs_zone_is_cnv(z))
55 return;
56
57 /*
58 * For zones that transitioned to the offline or readonly condition,
59 * we only need to clear the active state.
60 */
61 if (z->z_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY))
62 goto out;
63
64 /*
65 * If the zone is active, that is, if it is explicitly open or
66 * partially written, check if it was already accounted as active.
67 */
68 if ((z->z_flags & ZONEFS_ZONE_OPEN) ||
69 (z->z_wpoffset > 0 && z->z_wpoffset < z->z_capacity)) {
70 if (!(z->z_flags & ZONEFS_ZONE_ACTIVE)) {
71 z->z_flags |= ZONEFS_ZONE_ACTIVE;
72 atomic_inc(&sbi->s_active_seq_files);
73 }
74 return;
75 }
76
77out:
78 /* The zone is not active. If it was, update the active count */
79 if (z->z_flags & ZONEFS_ZONE_ACTIVE) {
80 z->z_flags &= ~ZONEFS_ZONE_ACTIVE;
81 atomic_dec(&sbi->s_active_seq_files);
82 }
83}
84
85/*
86 * Manage the active zone count. Called with zi->i_truncate_mutex held.
87 */
88void zonefs_inode_account_active(struct inode *inode)
89{
90 lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
91
92 return zonefs_account_active(inode->i_sb, zonefs_inode_zone(inode));
93}
94
95/*
96 * Execute a zone management operation.
97 */
98static int zonefs_zone_mgmt(struct super_block *sb,
99 struct zonefs_zone *z, enum req_op op)
100{
101 int ret;
102
103 /*
104 * With ZNS drives, closing an explicitly open zone that has not been
105 * written will change the zone state to "closed", that is, the zone
106 * will remain active. Since this can then cause failure of explicit
107 * open operation on other zones if the drive active zone resources
108 * are exceeded, make sure that the zone does not remain active by
109 * resetting it.
110 */
111 if (op == REQ_OP_ZONE_CLOSE && !z->z_wpoffset)
112 op = REQ_OP_ZONE_RESET;
113
114 trace_zonefs_zone_mgmt(sb, z, op);
115 ret = blkdev_zone_mgmt(sb->s_bdev, op, z->z_sector,
116 z->z_size >> SECTOR_SHIFT, GFP_NOFS);
117 if (ret) {
118 zonefs_err(sb,
119 "Zone management operation %s at %llu failed %d\n",
120 blk_op_str(op), z->z_sector, ret);
121 return ret;
122 }
123
124 return 0;
125}
126
127int zonefs_inode_zone_mgmt(struct inode *inode, enum req_op op)
128{
129 lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
130
131 return zonefs_zone_mgmt(inode->i_sb, zonefs_inode_zone(inode), op);
132}
133
134void zonefs_i_size_write(struct inode *inode, loff_t isize)
135{
136 struct zonefs_zone *z = zonefs_inode_zone(inode);
137
138 i_size_write(inode, isize);
139
140 /*
141 * A full zone is no longer open/active and does not need
142 * explicit closing.
143 */
144 if (isize >= z->z_capacity) {
145 struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
146
147 if (z->z_flags & ZONEFS_ZONE_ACTIVE)
148 atomic_dec(&sbi->s_active_seq_files);
149 z->z_flags &= ~(ZONEFS_ZONE_OPEN | ZONEFS_ZONE_ACTIVE);
150 }
151}
152
153void zonefs_update_stats(struct inode *inode, loff_t new_isize)
154{
155 struct super_block *sb = inode->i_sb;
156 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
157 loff_t old_isize = i_size_read(inode);
158 loff_t nr_blocks;
159
160 if (new_isize == old_isize)
161 return;
162
163 spin_lock(&sbi->s_lock);
164
165 /*
166 * This may be called for an update after an IO error.
167 * So beware of the values seen.
168 */
169 if (new_isize < old_isize) {
170 nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits;
171 if (sbi->s_used_blocks > nr_blocks)
172 sbi->s_used_blocks -= nr_blocks;
173 else
174 sbi->s_used_blocks = 0;
175 } else {
176 sbi->s_used_blocks +=
177 (new_isize - old_isize) >> sb->s_blocksize_bits;
178 if (sbi->s_used_blocks > sbi->s_blocks)
179 sbi->s_used_blocks = sbi->s_blocks;
180 }
181
182 spin_unlock(&sbi->s_lock);
183}
184
185/*
186 * Check a zone condition. Return the amount of written (and still readable)
187 * data in the zone.
188 */
189static loff_t zonefs_check_zone_condition(struct super_block *sb,
190 struct zonefs_zone *z,
191 struct blk_zone *zone)
192{
193 switch (zone->cond) {
194 case BLK_ZONE_COND_OFFLINE:
195 zonefs_warn(sb, "Zone %llu: offline zone\n",
196 z->z_sector);
197 z->z_flags |= ZONEFS_ZONE_OFFLINE;
198 return 0;
199 case BLK_ZONE_COND_READONLY:
200 /*
201 * The write pointer of read-only zones is invalid, so we cannot
202 * determine the zone wpoffset (inode size). We thus keep the
203 * zone wpoffset as is, which leads to an empty file
204 * (wpoffset == 0) on mount. For a runtime error, this keeps
205 * the inode size as it was when last updated so that the user
206 * can recover data.
207 */
208 zonefs_warn(sb, "Zone %llu: read-only zone\n",
209 z->z_sector);
210 z->z_flags |= ZONEFS_ZONE_READONLY;
211 if (zonefs_zone_is_cnv(z))
212 return z->z_capacity;
213 return z->z_wpoffset;
214 case BLK_ZONE_COND_FULL:
215 /* The write pointer of full zones is invalid. */
216 return z->z_capacity;
217 default:
218 if (zonefs_zone_is_cnv(z))
219 return z->z_capacity;
220 return (zone->wp - zone->start) << SECTOR_SHIFT;
221 }
222}
223
224/*
225 * Check a zone condition and adjust its inode access permissions for
226 * offline and readonly zones.
227 */
228static void zonefs_inode_update_mode(struct inode *inode)
229{
230 struct zonefs_zone *z = zonefs_inode_zone(inode);
231
232 if (z->z_flags & ZONEFS_ZONE_OFFLINE) {
233 /* Offline zones cannot be read nor written */
234 inode->i_flags |= S_IMMUTABLE;
235 inode->i_mode &= ~0777;
236 } else if (z->z_flags & ZONEFS_ZONE_READONLY) {
237 /* Readonly zones cannot be written */
238 inode->i_flags |= S_IMMUTABLE;
239 if (z->z_flags & ZONEFS_ZONE_INIT_MODE)
240 inode->i_mode &= ~0777;
241 else
242 inode->i_mode &= ~0222;
243 }
244
245 z->z_flags &= ~ZONEFS_ZONE_INIT_MODE;
246 z->z_mode = inode->i_mode;
247}
248
249static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
250 void *data)
251{
252 struct blk_zone *z = data;
253
254 *z = *zone;
255 return 0;
256}
257
258static void zonefs_handle_io_error(struct inode *inode, struct blk_zone *zone,
259 bool write)
260{
261 struct zonefs_zone *z = zonefs_inode_zone(inode);
262 struct super_block *sb = inode->i_sb;
263 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
264 loff_t isize, data_size;
265
266 /*
267 * Check the zone condition: if the zone is not "bad" (offline or
268 * read-only), read errors are simply signaled to the IO issuer as long
269 * as there is no inconsistency between the inode size and the amount of
270 * data writen in the zone (data_size).
271 */
272 data_size = zonefs_check_zone_condition(sb, z, zone);
273 isize = i_size_read(inode);
274 if (!(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)) &&
275 !write && isize == data_size)
276 return;
277
278 /*
279 * At this point, we detected either a bad zone or an inconsistency
280 * between the inode size and the amount of data written in the zone.
281 * For the latter case, the cause may be a write IO error or an external
282 * action on the device. Two error patterns exist:
283 * 1) The inode size is lower than the amount of data in the zone:
284 * a write operation partially failed and data was writen at the end
285 * of the file. This can happen in the case of a large direct IO
286 * needing several BIOs and/or write requests to be processed.
287 * 2) The inode size is larger than the amount of data in the zone:
288 * this can happen with a deferred write error with the use of the
289 * device side write cache after getting successful write IO
290 * completions. Other possibilities are (a) an external corruption,
291 * e.g. an application reset the zone directly, or (b) the device
292 * has a serious problem (e.g. firmware bug).
293 *
294 * In all cases, warn about inode size inconsistency and handle the
295 * IO error according to the zone condition and to the mount options.
296 */
297 if (isize != data_size)
298 zonefs_warn(sb,
299 "inode %lu: invalid size %lld (should be %lld)\n",
300 inode->i_ino, isize, data_size);
301
302 /*
303 * First handle bad zones signaled by hardware. The mount options
304 * errors=zone-ro and errors=zone-offline result in changing the
305 * zone condition to read-only and offline respectively, as if the
306 * condition was signaled by the hardware.
307 */
308 if ((z->z_flags & ZONEFS_ZONE_OFFLINE) ||
309 (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)) {
310 zonefs_warn(sb, "inode %lu: read/write access disabled\n",
311 inode->i_ino);
312 if (!(z->z_flags & ZONEFS_ZONE_OFFLINE))
313 z->z_flags |= ZONEFS_ZONE_OFFLINE;
314 zonefs_inode_update_mode(inode);
315 data_size = 0;
316 } else if ((z->z_flags & ZONEFS_ZONE_READONLY) ||
317 (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)) {
318 zonefs_warn(sb, "inode %lu: write access disabled\n",
319 inode->i_ino);
320 if (!(z->z_flags & ZONEFS_ZONE_READONLY))
321 z->z_flags |= ZONEFS_ZONE_READONLY;
322 zonefs_inode_update_mode(inode);
323 data_size = isize;
324 } else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO &&
325 data_size > isize) {
326 /* Do not expose garbage data */
327 data_size = isize;
328 }
329
330 /*
331 * If the filesystem is mounted with the explicit-open mount option, we
332 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
333 * the read-only or offline condition, to avoid attempting an explicit
334 * close of the zone when the inode file is closed.
335 */
336 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
337 (z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)))
338 z->z_flags &= ~ZONEFS_ZONE_OPEN;
339
340 /*
341 * If error=remount-ro was specified, any error result in remounting
342 * the volume as read-only.
343 */
344 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
345 zonefs_warn(sb, "remounting filesystem read-only\n");
346 sb->s_flags |= SB_RDONLY;
347 }
348
349 /*
350 * Update block usage stats and the inode size to prevent access to
351 * invalid data.
352 */
353 zonefs_update_stats(inode, data_size);
354 zonefs_i_size_write(inode, data_size);
355 z->z_wpoffset = data_size;
356 zonefs_inode_account_active(inode);
357}
358
359/*
360 * When an file IO error occurs, check the file zone to see if there is a change
361 * in the zone condition (e.g. offline or read-only). For a failed write to a
362 * sequential zone, the zone write pointer position must also be checked to
363 * eventually correct the file size and zonefs inode write pointer offset
364 * (which can be out of sync with the drive due to partial write failures).
365 */
366void __zonefs_io_error(struct inode *inode, bool write)
367{
368 struct zonefs_zone *z = zonefs_inode_zone(inode);
369 struct super_block *sb = inode->i_sb;
370 unsigned int noio_flag;
371 struct blk_zone zone;
372 int ret;
373
374 /*
375 * Conventional zone have no write pointer and cannot become read-only
376 * or offline. So simply fake a report for a single or aggregated zone
377 * and let zonefs_handle_io_error() correct the zone inode information
378 * according to the mount options.
379 */
380 if (!zonefs_zone_is_seq(z)) {
381 zone.start = z->z_sector;
382 zone.len = z->z_size >> SECTOR_SHIFT;
383 zone.wp = zone.start + zone.len;
384 zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
385 zone.cond = BLK_ZONE_COND_NOT_WP;
386 zone.capacity = zone.len;
387 goto handle_io_error;
388 }
389
390 /*
391 * Memory allocations in blkdev_report_zones() can trigger a memory
392 * reclaim which may in turn cause a recursion into zonefs as well as
393 * struct request allocations for the same device. The former case may
394 * end up in a deadlock on the inode truncate mutex, while the latter
395 * may prevent IO forward progress. Executing the report zones under
396 * the GFP_NOIO context avoids both problems.
397 */
398 noio_flag = memalloc_noio_save();
399 ret = blkdev_report_zones(sb->s_bdev, z->z_sector, 1,
400 zonefs_io_error_cb, &zone);
401 memalloc_noio_restore(noio_flag);
402
403 if (ret != 1) {
404 zonefs_err(sb, "Get inode %lu zone information failed %d\n",
405 inode->i_ino, ret);
406 zonefs_warn(sb, "remounting filesystem read-only\n");
407 sb->s_flags |= SB_RDONLY;
408 return;
409 }
410
411handle_io_error:
412 zonefs_handle_io_error(inode, &zone, write);
413}
414
415static struct kmem_cache *zonefs_inode_cachep;
416
417static struct inode *zonefs_alloc_inode(struct super_block *sb)
418{
419 struct zonefs_inode_info *zi;
420
421 zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL);
422 if (!zi)
423 return NULL;
424
425 inode_init_once(&zi->i_vnode);
426 mutex_init(&zi->i_truncate_mutex);
427 zi->i_wr_refcnt = 0;
428
429 return &zi->i_vnode;
430}
431
432static void zonefs_free_inode(struct inode *inode)
433{
434 kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
435}
436
437/*
438 * File system stat.
439 */
440static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
441{
442 struct super_block *sb = dentry->d_sb;
443 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
444 enum zonefs_ztype t;
445
446 buf->f_type = ZONEFS_MAGIC;
447 buf->f_bsize = sb->s_blocksize;
448 buf->f_namelen = ZONEFS_NAME_MAX;
449
450 spin_lock(&sbi->s_lock);
451
452 buf->f_blocks = sbi->s_blocks;
453 if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
454 buf->f_bfree = 0;
455 else
456 buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
457 buf->f_bavail = buf->f_bfree;
458
459 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
460 if (sbi->s_zgroup[t].g_nr_zones)
461 buf->f_files += sbi->s_zgroup[t].g_nr_zones + 1;
462 }
463 buf->f_ffree = 0;
464
465 spin_unlock(&sbi->s_lock);
466
467 buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b);
468
469 return 0;
470}
471
472enum {
473 Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
474 Opt_explicit_open, Opt_err,
475};
476
477static const match_table_t tokens = {
478 { Opt_errors_ro, "errors=remount-ro"},
479 { Opt_errors_zro, "errors=zone-ro"},
480 { Opt_errors_zol, "errors=zone-offline"},
481 { Opt_errors_repair, "errors=repair"},
482 { Opt_explicit_open, "explicit-open" },
483 { Opt_err, NULL}
484};
485
486static int zonefs_parse_options(struct super_block *sb, char *options)
487{
488 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
489 substring_t args[MAX_OPT_ARGS];
490 char *p;
491
492 if (!options)
493 return 0;
494
495 while ((p = strsep(&options, ",")) != NULL) {
496 int token;
497
498 if (!*p)
499 continue;
500
501 token = match_token(p, tokens, args);
502 switch (token) {
503 case Opt_errors_ro:
504 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
505 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
506 break;
507 case Opt_errors_zro:
508 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
509 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
510 break;
511 case Opt_errors_zol:
512 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
513 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
514 break;
515 case Opt_errors_repair:
516 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
517 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
518 break;
519 case Opt_explicit_open:
520 sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
521 break;
522 default:
523 return -EINVAL;
524 }
525 }
526
527 return 0;
528}
529
530static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
531{
532 struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
533
534 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
535 seq_puts(seq, ",errors=remount-ro");
536 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
537 seq_puts(seq, ",errors=zone-ro");
538 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
539 seq_puts(seq, ",errors=zone-offline");
540 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
541 seq_puts(seq, ",errors=repair");
542
543 return 0;
544}
545
546static int zonefs_remount(struct super_block *sb, int *flags, char *data)
547{
548 sync_filesystem(sb);
549
550 return zonefs_parse_options(sb, data);
551}
552
553static int zonefs_inode_setattr(struct mnt_idmap *idmap,
554 struct dentry *dentry, struct iattr *iattr)
555{
556 struct inode *inode = d_inode(dentry);
557 int ret;
558
559 if (unlikely(IS_IMMUTABLE(inode)))
560 return -EPERM;
561
562 ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
563 if (ret)
564 return ret;
565
566 /*
567 * Since files and directories cannot be created nor deleted, do not
568 * allow setting any write attributes on the sub-directories grouping
569 * files by zone type.
570 */
571 if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
572 (iattr->ia_mode & 0222))
573 return -EPERM;
574
575 if (((iattr->ia_valid & ATTR_UID) &&
576 !uid_eq(iattr->ia_uid, inode->i_uid)) ||
577 ((iattr->ia_valid & ATTR_GID) &&
578 !gid_eq(iattr->ia_gid, inode->i_gid))) {
579 ret = dquot_transfer(&nop_mnt_idmap, inode, iattr);
580 if (ret)
581 return ret;
582 }
583
584 if (iattr->ia_valid & ATTR_SIZE) {
585 ret = zonefs_file_truncate(inode, iattr->ia_size);
586 if (ret)
587 return ret;
588 }
589
590 setattr_copy(&nop_mnt_idmap, inode, iattr);
591
592 if (S_ISREG(inode->i_mode)) {
593 struct zonefs_zone *z = zonefs_inode_zone(inode);
594
595 z->z_mode = inode->i_mode;
596 z->z_uid = inode->i_uid;
597 z->z_gid = inode->i_gid;
598 }
599
600 return 0;
601}
602
603static const struct inode_operations zonefs_file_inode_operations = {
604 .setattr = zonefs_inode_setattr,
605};
606
607static long zonefs_fname_to_fno(const struct qstr *fname)
608{
609 const char *name = fname->name;
610 unsigned int len = fname->len;
611 long fno = 0, shift = 1;
612 const char *rname;
613 char c = *name;
614 unsigned int i;
615
616 /*
617 * File names are always a base-10 number string without any
618 * leading 0s.
619 */
620 if (!isdigit(c))
621 return -ENOENT;
622
623 if (len > 1 && c == '0')
624 return -ENOENT;
625
626 if (len == 1)
627 return c - '0';
628
629 for (i = 0, rname = name + len - 1; i < len; i++, rname--) {
630 c = *rname;
631 if (!isdigit(c))
632 return -ENOENT;
633 fno += (c - '0') * shift;
634 shift *= 10;
635 }
636
637 return fno;
638}
639
640static struct inode *zonefs_get_file_inode(struct inode *dir,
641 struct dentry *dentry)
642{
643 struct zonefs_zone_group *zgroup = dir->i_private;
644 struct super_block *sb = dir->i_sb;
645 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
646 struct zonefs_zone *z;
647 struct inode *inode;
648 ino_t ino;
649 long fno;
650
651 /* Get the file number from the file name */
652 fno = zonefs_fname_to_fno(&dentry->d_name);
653 if (fno < 0)
654 return ERR_PTR(fno);
655
656 if (!zgroup->g_nr_zones || fno >= zgroup->g_nr_zones)
657 return ERR_PTR(-ENOENT);
658
659 z = &zgroup->g_zones[fno];
660 ino = z->z_sector >> sbi->s_zone_sectors_shift;
661 inode = iget_locked(sb, ino);
662 if (!inode)
663 return ERR_PTR(-ENOMEM);
664 if (!(inode->i_state & I_NEW)) {
665 WARN_ON_ONCE(inode->i_private != z);
666 return inode;
667 }
668
669 inode->i_ino = ino;
670 inode->i_mode = z->z_mode;
671 inode_set_mtime_to_ts(inode,
672 inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(dir))));
673 inode->i_uid = z->z_uid;
674 inode->i_gid = z->z_gid;
675 inode->i_size = z->z_wpoffset;
676 inode->i_blocks = z->z_capacity >> SECTOR_SHIFT;
677 inode->i_private = z;
678
679 inode->i_op = &zonefs_file_inode_operations;
680 inode->i_fop = &zonefs_file_operations;
681 inode->i_mapping->a_ops = &zonefs_file_aops;
682
683 /* Update the inode access rights depending on the zone condition */
684 zonefs_inode_update_mode(inode);
685
686 unlock_new_inode(inode);
687
688 return inode;
689}
690
691static struct inode *zonefs_get_zgroup_inode(struct super_block *sb,
692 enum zonefs_ztype ztype)
693{
694 struct inode *root = d_inode(sb->s_root);
695 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
696 struct inode *inode;
697 ino_t ino = bdev_nr_zones(sb->s_bdev) + ztype + 1;
698
699 inode = iget_locked(sb, ino);
700 if (!inode)
701 return ERR_PTR(-ENOMEM);
702 if (!(inode->i_state & I_NEW))
703 return inode;
704
705 inode->i_ino = ino;
706 inode_init_owner(&nop_mnt_idmap, inode, root, S_IFDIR | 0555);
707 inode->i_size = sbi->s_zgroup[ztype].g_nr_zones;
708 inode_set_mtime_to_ts(inode,
709 inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(root))));
710 inode->i_private = &sbi->s_zgroup[ztype];
711 set_nlink(inode, 2);
712
713 inode->i_op = &zonefs_dir_inode_operations;
714 inode->i_fop = &zonefs_dir_operations;
715
716 unlock_new_inode(inode);
717
718 return inode;
719}
720
721
722static struct inode *zonefs_get_dir_inode(struct inode *dir,
723 struct dentry *dentry)
724{
725 struct super_block *sb = dir->i_sb;
726 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
727 const char *name = dentry->d_name.name;
728 enum zonefs_ztype ztype;
729
730 /*
731 * We only need to check for the "seq" directory and
732 * the "cnv" directory if we have conventional zones.
733 */
734 if (dentry->d_name.len != 3)
735 return ERR_PTR(-ENOENT);
736
737 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
738 if (sbi->s_zgroup[ztype].g_nr_zones &&
739 memcmp(name, zonefs_zgroup_name(ztype), 3) == 0)
740 break;
741 }
742 if (ztype == ZONEFS_ZTYPE_MAX)
743 return ERR_PTR(-ENOENT);
744
745 return zonefs_get_zgroup_inode(sb, ztype);
746}
747
748static struct dentry *zonefs_lookup(struct inode *dir, struct dentry *dentry,
749 unsigned int flags)
750{
751 struct inode *inode;
752
753 if (dentry->d_name.len > ZONEFS_NAME_MAX)
754 return ERR_PTR(-ENAMETOOLONG);
755
756 if (dir == d_inode(dir->i_sb->s_root))
757 inode = zonefs_get_dir_inode(dir, dentry);
758 else
759 inode = zonefs_get_file_inode(dir, dentry);
760
761 return d_splice_alias(inode, dentry);
762}
763
764static int zonefs_readdir_root(struct file *file, struct dir_context *ctx)
765{
766 struct inode *inode = file_inode(file);
767 struct super_block *sb = inode->i_sb;
768 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
769 enum zonefs_ztype ztype = ZONEFS_ZTYPE_CNV;
770 ino_t base_ino = bdev_nr_zones(sb->s_bdev) + 1;
771
772 if (ctx->pos >= inode->i_size)
773 return 0;
774
775 if (!dir_emit_dots(file, ctx))
776 return 0;
777
778 if (ctx->pos == 2) {
779 if (!sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones)
780 ztype = ZONEFS_ZTYPE_SEQ;
781
782 if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
783 base_ino + ztype, DT_DIR))
784 return 0;
785 ctx->pos++;
786 }
787
788 if (ctx->pos == 3 && ztype != ZONEFS_ZTYPE_SEQ) {
789 ztype = ZONEFS_ZTYPE_SEQ;
790 if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
791 base_ino + ztype, DT_DIR))
792 return 0;
793 ctx->pos++;
794 }
795
796 return 0;
797}
798
799static int zonefs_readdir_zgroup(struct file *file,
800 struct dir_context *ctx)
801{
802 struct inode *inode = file_inode(file);
803 struct zonefs_zone_group *zgroup = inode->i_private;
804 struct super_block *sb = inode->i_sb;
805 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
806 struct zonefs_zone *z;
807 int fname_len;
808 char *fname;
809 ino_t ino;
810 int f;
811
812 /*
813 * The size of zone group directories is equal to the number
814 * of zone files in the group and does note include the "." and
815 * ".." entries. Hence the "+ 2" here.
816 */
817 if (ctx->pos >= inode->i_size + 2)
818 return 0;
819
820 if (!dir_emit_dots(file, ctx))
821 return 0;
822
823 fname = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
824 if (!fname)
825 return -ENOMEM;
826
827 for (f = ctx->pos - 2; f < zgroup->g_nr_zones; f++) {
828 z = &zgroup->g_zones[f];
829 ino = z->z_sector >> sbi->s_zone_sectors_shift;
830 fname_len = snprintf(fname, ZONEFS_NAME_MAX - 1, "%u", f);
831 if (!dir_emit(ctx, fname, fname_len, ino, DT_REG))
832 break;
833 ctx->pos++;
834 }
835
836 kfree(fname);
837
838 return 0;
839}
840
841static int zonefs_readdir(struct file *file, struct dir_context *ctx)
842{
843 struct inode *inode = file_inode(file);
844
845 if (inode == d_inode(inode->i_sb->s_root))
846 return zonefs_readdir_root(file, ctx);
847
848 return zonefs_readdir_zgroup(file, ctx);
849}
850
851const struct inode_operations zonefs_dir_inode_operations = {
852 .lookup = zonefs_lookup,
853 .setattr = zonefs_inode_setattr,
854};
855
856const struct file_operations zonefs_dir_operations = {
857 .llseek = generic_file_llseek,
858 .read = generic_read_dir,
859 .iterate_shared = zonefs_readdir,
860};
861
862struct zonefs_zone_data {
863 struct super_block *sb;
864 unsigned int nr_zones[ZONEFS_ZTYPE_MAX];
865 sector_t cnv_zone_start;
866 struct blk_zone *zones;
867};
868
869static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
870 void *data)
871{
872 struct zonefs_zone_data *zd = data;
873 struct super_block *sb = zd->sb;
874 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
875
876 /*
877 * We do not care about the first zone: it contains the super block
878 * and not exposed as a file.
879 */
880 if (!idx)
881 return 0;
882
883 /*
884 * Count the number of zones that will be exposed as files.
885 * For sequential zones, we always have as many files as zones.
886 * FOr conventional zones, the number of files depends on if we have
887 * conventional zones aggregation enabled.
888 */
889 switch (zone->type) {
890 case BLK_ZONE_TYPE_CONVENTIONAL:
891 if (sbi->s_features & ZONEFS_F_AGGRCNV) {
892 /* One file per set of contiguous conventional zones */
893 if (!(sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) ||
894 zone->start != zd->cnv_zone_start)
895 sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
896 zd->cnv_zone_start = zone->start + zone->len;
897 } else {
898 /* One file per zone */
899 sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
900 }
901 break;
902 case BLK_ZONE_TYPE_SEQWRITE_REQ:
903 case BLK_ZONE_TYPE_SEQWRITE_PREF:
904 sbi->s_zgroup[ZONEFS_ZTYPE_SEQ].g_nr_zones++;
905 break;
906 default:
907 zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
908 zone->type);
909 return -EIO;
910 }
911
912 memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
913
914 return 0;
915}
916
917static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
918{
919 struct block_device *bdev = zd->sb->s_bdev;
920 int ret;
921
922 zd->zones = kvcalloc(bdev_nr_zones(bdev), sizeof(struct blk_zone),
923 GFP_KERNEL);
924 if (!zd->zones)
925 return -ENOMEM;
926
927 /* Get zones information from the device */
928 ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
929 zonefs_get_zone_info_cb, zd);
930 if (ret < 0) {
931 zonefs_err(zd->sb, "Zone report failed %d\n", ret);
932 return ret;
933 }
934
935 if (ret != bdev_nr_zones(bdev)) {
936 zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
937 ret, bdev_nr_zones(bdev));
938 return -EIO;
939 }
940
941 return 0;
942}
943
944static inline void zonefs_free_zone_info(struct zonefs_zone_data *zd)
945{
946 kvfree(zd->zones);
947}
948
949/*
950 * Create a zone group and populate it with zone files.
951 */
952static int zonefs_init_zgroup(struct super_block *sb,
953 struct zonefs_zone_data *zd,
954 enum zonefs_ztype ztype)
955{
956 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
957 struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype];
958 struct blk_zone *zone, *next, *end;
959 struct zonefs_zone *z;
960 unsigned int n = 0;
961 int ret;
962
963 /* Allocate the zone group. If it is empty, we have nothing to do. */
964 if (!zgroup->g_nr_zones)
965 return 0;
966
967 zgroup->g_zones = kvcalloc(zgroup->g_nr_zones,
968 sizeof(struct zonefs_zone), GFP_KERNEL);
969 if (!zgroup->g_zones)
970 return -ENOMEM;
971
972 /*
973 * Initialize the zone groups using the device zone information.
974 * We always skip the first zone as it contains the super block
975 * and is not use to back a file.
976 */
977 end = zd->zones + bdev_nr_zones(sb->s_bdev);
978 for (zone = &zd->zones[1]; zone < end; zone = next) {
979
980 next = zone + 1;
981 if (zonefs_zone_type(zone) != ztype)
982 continue;
983
984 if (WARN_ON_ONCE(n >= zgroup->g_nr_zones))
985 return -EINVAL;
986
987 /*
988 * For conventional zones, contiguous zones can be aggregated
989 * together to form larger files. Note that this overwrites the
990 * length of the first zone of the set of contiguous zones
991 * aggregated together. If one offline or read-only zone is
992 * found, assume that all zones aggregated have the same
993 * condition.
994 */
995 if (ztype == ZONEFS_ZTYPE_CNV &&
996 (sbi->s_features & ZONEFS_F_AGGRCNV)) {
997 for (; next < end; next++) {
998 if (zonefs_zone_type(next) != ztype)
999 break;
1000 zone->len += next->len;
1001 zone->capacity += next->capacity;
1002 if (next->cond == BLK_ZONE_COND_READONLY &&
1003 zone->cond != BLK_ZONE_COND_OFFLINE)
1004 zone->cond = BLK_ZONE_COND_READONLY;
1005 else if (next->cond == BLK_ZONE_COND_OFFLINE)
1006 zone->cond = BLK_ZONE_COND_OFFLINE;
1007 }
1008 }
1009
1010 z = &zgroup->g_zones[n];
1011 if (ztype == ZONEFS_ZTYPE_CNV)
1012 z->z_flags |= ZONEFS_ZONE_CNV;
1013 z->z_sector = zone->start;
1014 z->z_size = zone->len << SECTOR_SHIFT;
1015 if (z->z_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT &&
1016 !(sbi->s_features & ZONEFS_F_AGGRCNV)) {
1017 zonefs_err(sb,
1018 "Invalid zone size %llu (device zone sectors %llu)\n",
1019 z->z_size,
1020 bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT);
1021 return -EINVAL;
1022 }
1023
1024 z->z_capacity = min_t(loff_t, MAX_LFS_FILESIZE,
1025 zone->capacity << SECTOR_SHIFT);
1026 z->z_wpoffset = zonefs_check_zone_condition(sb, z, zone);
1027
1028 z->z_mode = S_IFREG | sbi->s_perm;
1029 z->z_uid = sbi->s_uid;
1030 z->z_gid = sbi->s_gid;
1031
1032 /*
1033 * Let zonefs_inode_update_mode() know that we will need
1034 * special initialization of the inode mode the first time
1035 * it is accessed.
1036 */
1037 z->z_flags |= ZONEFS_ZONE_INIT_MODE;
1038
1039 sb->s_maxbytes = max(z->z_capacity, sb->s_maxbytes);
1040 sbi->s_blocks += z->z_capacity >> sb->s_blocksize_bits;
1041 sbi->s_used_blocks += z->z_wpoffset >> sb->s_blocksize_bits;
1042
1043 /*
1044 * For sequential zones, make sure that any open zone is closed
1045 * first to ensure that the initial number of open zones is 0,
1046 * in sync with the open zone accounting done when the mount
1047 * option ZONEFS_MNTOPT_EXPLICIT_OPEN is used.
1048 */
1049 if (ztype == ZONEFS_ZTYPE_SEQ &&
1050 (zone->cond == BLK_ZONE_COND_IMP_OPEN ||
1051 zone->cond == BLK_ZONE_COND_EXP_OPEN)) {
1052 ret = zonefs_zone_mgmt(sb, z, REQ_OP_ZONE_CLOSE);
1053 if (ret)
1054 return ret;
1055 }
1056
1057 zonefs_account_active(sb, z);
1058
1059 n++;
1060 }
1061
1062 if (WARN_ON_ONCE(n != zgroup->g_nr_zones))
1063 return -EINVAL;
1064
1065 zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1066 zonefs_zgroup_name(ztype),
1067 zgroup->g_nr_zones,
1068 zgroup->g_nr_zones > 1 ? "s" : "");
1069
1070 return 0;
1071}
1072
1073static void zonefs_free_zgroups(struct super_block *sb)
1074{
1075 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1076 enum zonefs_ztype ztype;
1077
1078 if (!sbi)
1079 return;
1080
1081 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1082 kvfree(sbi->s_zgroup[ztype].g_zones);
1083 sbi->s_zgroup[ztype].g_zones = NULL;
1084 }
1085}
1086
1087/*
1088 * Create a zone group and populate it with zone files.
1089 */
1090static int zonefs_init_zgroups(struct super_block *sb)
1091{
1092 struct zonefs_zone_data zd;
1093 enum zonefs_ztype ztype;
1094 int ret;
1095
1096 /* First get the device zone information */
1097 memset(&zd, 0, sizeof(struct zonefs_zone_data));
1098 zd.sb = sb;
1099 ret = zonefs_get_zone_info(&zd);
1100 if (ret)
1101 goto cleanup;
1102
1103 /* Allocate and initialize the zone groups */
1104 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1105 ret = zonefs_init_zgroup(sb, &zd, ztype);
1106 if (ret) {
1107 zonefs_info(sb,
1108 "Zone group \"%s\" initialization failed\n",
1109 zonefs_zgroup_name(ztype));
1110 break;
1111 }
1112 }
1113
1114cleanup:
1115 zonefs_free_zone_info(&zd);
1116 if (ret)
1117 zonefs_free_zgroups(sb);
1118
1119 return ret;
1120}
1121
1122/*
1123 * Read super block information from the device.
1124 */
1125static int zonefs_read_super(struct super_block *sb)
1126{
1127 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1128 struct zonefs_super *super;
1129 u32 crc, stored_crc;
1130 struct page *page;
1131 struct bio_vec bio_vec;
1132 struct bio bio;
1133 int ret;
1134
1135 page = alloc_page(GFP_KERNEL);
1136 if (!page)
1137 return -ENOMEM;
1138
1139 bio_init(&bio, sb->s_bdev, &bio_vec, 1, REQ_OP_READ);
1140 bio.bi_iter.bi_sector = 0;
1141 __bio_add_page(&bio, page, PAGE_SIZE, 0);
1142
1143 ret = submit_bio_wait(&bio);
1144 if (ret)
1145 goto free_page;
1146
1147 super = page_address(page);
1148
1149 ret = -EINVAL;
1150 if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1151 goto free_page;
1152
1153 stored_crc = le32_to_cpu(super->s_crc);
1154 super->s_crc = 0;
1155 crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1156 if (crc != stored_crc) {
1157 zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1158 crc, stored_crc);
1159 goto free_page;
1160 }
1161
1162 sbi->s_features = le64_to_cpu(super->s_features);
1163 if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1164 zonefs_err(sb, "Unknown features set 0x%llx\n",
1165 sbi->s_features);
1166 goto free_page;
1167 }
1168
1169 if (sbi->s_features & ZONEFS_F_UID) {
1170 sbi->s_uid = make_kuid(current_user_ns(),
1171 le32_to_cpu(super->s_uid));
1172 if (!uid_valid(sbi->s_uid)) {
1173 zonefs_err(sb, "Invalid UID feature\n");
1174 goto free_page;
1175 }
1176 }
1177
1178 if (sbi->s_features & ZONEFS_F_GID) {
1179 sbi->s_gid = make_kgid(current_user_ns(),
1180 le32_to_cpu(super->s_gid));
1181 if (!gid_valid(sbi->s_gid)) {
1182 zonefs_err(sb, "Invalid GID feature\n");
1183 goto free_page;
1184 }
1185 }
1186
1187 if (sbi->s_features & ZONEFS_F_PERM)
1188 sbi->s_perm = le32_to_cpu(super->s_perm);
1189
1190 if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1191 zonefs_err(sb, "Reserved area is being used\n");
1192 goto free_page;
1193 }
1194
1195 import_uuid(&sbi->s_uuid, super->s_uuid);
1196 ret = 0;
1197
1198free_page:
1199 __free_page(page);
1200
1201 return ret;
1202}
1203
1204static const struct super_operations zonefs_sops = {
1205 .alloc_inode = zonefs_alloc_inode,
1206 .free_inode = zonefs_free_inode,
1207 .statfs = zonefs_statfs,
1208 .remount_fs = zonefs_remount,
1209 .show_options = zonefs_show_options,
1210};
1211
1212static int zonefs_get_zgroup_inodes(struct super_block *sb)
1213{
1214 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1215 struct inode *dir_inode;
1216 enum zonefs_ztype ztype;
1217
1218 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1219 if (!sbi->s_zgroup[ztype].g_nr_zones)
1220 continue;
1221
1222 dir_inode = zonefs_get_zgroup_inode(sb, ztype);
1223 if (IS_ERR(dir_inode))
1224 return PTR_ERR(dir_inode);
1225
1226 sbi->s_zgroup[ztype].g_inode = dir_inode;
1227 }
1228
1229 return 0;
1230}
1231
1232static void zonefs_release_zgroup_inodes(struct super_block *sb)
1233{
1234 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1235 enum zonefs_ztype ztype;
1236
1237 if (!sbi)
1238 return;
1239
1240 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1241 if (sbi->s_zgroup[ztype].g_inode) {
1242 iput(sbi->s_zgroup[ztype].g_inode);
1243 sbi->s_zgroup[ztype].g_inode = NULL;
1244 }
1245 }
1246}
1247
1248/*
1249 * Check that the device is zoned. If it is, get the list of zones and create
1250 * sub-directories and files according to the device zone configuration and
1251 * format options.
1252 */
1253static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1254{
1255 struct zonefs_sb_info *sbi;
1256 struct inode *inode;
1257 enum zonefs_ztype ztype;
1258 int ret;
1259
1260 if (!bdev_is_zoned(sb->s_bdev)) {
1261 zonefs_err(sb, "Not a zoned block device\n");
1262 return -EINVAL;
1263 }
1264
1265 /*
1266 * Initialize super block information: the maximum file size is updated
1267 * when the zone files are created so that the format option
1268 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1269 * beyond the zone size is taken into account.
1270 */
1271 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1272 if (!sbi)
1273 return -ENOMEM;
1274
1275 spin_lock_init(&sbi->s_lock);
1276 sb->s_fs_info = sbi;
1277 sb->s_magic = ZONEFS_MAGIC;
1278 sb->s_maxbytes = 0;
1279 sb->s_op = &zonefs_sops;
1280 sb->s_time_gran = 1;
1281
1282 /*
1283 * The block size is set to the device zone write granularity to ensure
1284 * that write operations are always aligned according to the device
1285 * interface constraints.
1286 */
1287 sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev));
1288 sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1289 sbi->s_uid = GLOBAL_ROOT_UID;
1290 sbi->s_gid = GLOBAL_ROOT_GID;
1291 sbi->s_perm = 0640;
1292 sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1293
1294 atomic_set(&sbi->s_wro_seq_files, 0);
1295 sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev);
1296 atomic_set(&sbi->s_active_seq_files, 0);
1297 sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev);
1298
1299 ret = zonefs_read_super(sb);
1300 if (ret)
1301 return ret;
1302
1303 ret = zonefs_parse_options(sb, data);
1304 if (ret)
1305 return ret;
1306
1307 zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev));
1308
1309 if (!sbi->s_max_wro_seq_files &&
1310 !sbi->s_max_active_seq_files &&
1311 sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1312 zonefs_info(sb,
1313 "No open and active zone limits. Ignoring explicit_open mount option\n");
1314 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
1315 }
1316
1317 /* Initialize the zone groups */
1318 ret = zonefs_init_zgroups(sb);
1319 if (ret)
1320 goto cleanup;
1321
1322 /* Create the root directory inode */
1323 ret = -ENOMEM;
1324 inode = new_inode(sb);
1325 if (!inode)
1326 goto cleanup;
1327
1328 inode->i_ino = bdev_nr_zones(sb->s_bdev);
1329 inode->i_mode = S_IFDIR | 0555;
1330 simple_inode_init_ts(inode);
1331 inode->i_op = &zonefs_dir_inode_operations;
1332 inode->i_fop = &zonefs_dir_operations;
1333 inode->i_size = 2;
1334 set_nlink(inode, 2);
1335 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1336 if (sbi->s_zgroup[ztype].g_nr_zones) {
1337 inc_nlink(inode);
1338 inode->i_size++;
1339 }
1340 }
1341
1342 sb->s_root = d_make_root(inode);
1343 if (!sb->s_root)
1344 goto cleanup;
1345
1346 /*
1347 * Take a reference on the zone groups directory inodes
1348 * to keep them in the inode cache.
1349 */
1350 ret = zonefs_get_zgroup_inodes(sb);
1351 if (ret)
1352 goto cleanup;
1353
1354 ret = zonefs_sysfs_register(sb);
1355 if (ret)
1356 goto cleanup;
1357
1358 return 0;
1359
1360cleanup:
1361 zonefs_release_zgroup_inodes(sb);
1362 zonefs_free_zgroups(sb);
1363
1364 return ret;
1365}
1366
1367static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1368 int flags, const char *dev_name, void *data)
1369{
1370 return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1371}
1372
1373static void zonefs_kill_super(struct super_block *sb)
1374{
1375 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1376
1377 /* Release the reference on the zone group directory inodes */
1378 zonefs_release_zgroup_inodes(sb);
1379
1380 kill_block_super(sb);
1381
1382 zonefs_sysfs_unregister(sb);
1383 zonefs_free_zgroups(sb);
1384 kfree(sbi);
1385}
1386
1387/*
1388 * File system definition and registration.
1389 */
1390static struct file_system_type zonefs_type = {
1391 .owner = THIS_MODULE,
1392 .name = "zonefs",
1393 .mount = zonefs_mount,
1394 .kill_sb = zonefs_kill_super,
1395 .fs_flags = FS_REQUIRES_DEV,
1396};
1397
1398static int __init zonefs_init_inodecache(void)
1399{
1400 zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1401 sizeof(struct zonefs_inode_info), 0,
1402 (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1403 NULL);
1404 if (zonefs_inode_cachep == NULL)
1405 return -ENOMEM;
1406 return 0;
1407}
1408
1409static void zonefs_destroy_inodecache(void)
1410{
1411 /*
1412 * Make sure all delayed rcu free inodes are flushed before we
1413 * destroy the inode cache.
1414 */
1415 rcu_barrier();
1416 kmem_cache_destroy(zonefs_inode_cachep);
1417}
1418
1419static int __init zonefs_init(void)
1420{
1421 int ret;
1422
1423 BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1424
1425 ret = zonefs_init_inodecache();
1426 if (ret)
1427 return ret;
1428
1429 ret = zonefs_sysfs_init();
1430 if (ret)
1431 goto destroy_inodecache;
1432
1433 ret = register_filesystem(&zonefs_type);
1434 if (ret)
1435 goto sysfs_exit;
1436
1437 return 0;
1438
1439sysfs_exit:
1440 zonefs_sysfs_exit();
1441destroy_inodecache:
1442 zonefs_destroy_inodecache();
1443
1444 return ret;
1445}
1446
1447static void __exit zonefs_exit(void)
1448{
1449 unregister_filesystem(&zonefs_type);
1450 zonefs_sysfs_exit();
1451 zonefs_destroy_inodecache();
1452}
1453
1454MODULE_AUTHOR("Damien Le Moal");
1455MODULE_DESCRIPTION("Zone file system for zoned block devices");
1456MODULE_LICENSE("GPL");
1457MODULE_ALIAS_FS("zonefs");
1458module_init(zonefs_init);
1459module_exit(zonefs_exit);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Simple file system for zoned block devices exposing zones as files.
4 *
5 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
6 */
7#include <linux/module.h>
8#include <linux/fs.h>
9#include <linux/magic.h>
10#include <linux/iomap.h>
11#include <linux/init.h>
12#include <linux/slab.h>
13#include <linux/blkdev.h>
14#include <linux/statfs.h>
15#include <linux/writeback.h>
16#include <linux/quotaops.h>
17#include <linux/seq_file.h>
18#include <linux/parser.h>
19#include <linux/uio.h>
20#include <linux/mman.h>
21#include <linux/sched/mm.h>
22#include <linux/crc32.h>
23#include <linux/task_io_accounting_ops.h>
24
25#include "zonefs.h"
26
27static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
28 unsigned int flags, struct iomap *iomap,
29 struct iomap *srcmap)
30{
31 struct zonefs_inode_info *zi = ZONEFS_I(inode);
32 struct super_block *sb = inode->i_sb;
33 loff_t isize;
34
35 /* All I/Os should always be within the file maximum size */
36 if (WARN_ON_ONCE(offset + length > zi->i_max_size))
37 return -EIO;
38
39 /*
40 * Sequential zones can only accept direct writes. This is already
41 * checked when writes are issued, so warn if we see a page writeback
42 * operation.
43 */
44 if (WARN_ON_ONCE(zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
45 (flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT)))
46 return -EIO;
47
48 /*
49 * For conventional zones, all blocks are always mapped. For sequential
50 * zones, all blocks after always mapped below the inode size (zone
51 * write pointer) and unwriten beyond.
52 */
53 mutex_lock(&zi->i_truncate_mutex);
54 isize = i_size_read(inode);
55 if (offset >= isize)
56 iomap->type = IOMAP_UNWRITTEN;
57 else
58 iomap->type = IOMAP_MAPPED;
59 if (flags & IOMAP_WRITE)
60 length = zi->i_max_size - offset;
61 else
62 length = min(length, isize - offset);
63 mutex_unlock(&zi->i_truncate_mutex);
64
65 iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize);
66 iomap->length = ALIGN(offset + length, sb->s_blocksize) - iomap->offset;
67 iomap->bdev = inode->i_sb->s_bdev;
68 iomap->addr = (zi->i_zsector << SECTOR_SHIFT) + iomap->offset;
69
70 return 0;
71}
72
73static const struct iomap_ops zonefs_iomap_ops = {
74 .iomap_begin = zonefs_iomap_begin,
75};
76
77static int zonefs_readpage(struct file *unused, struct page *page)
78{
79 return iomap_readpage(page, &zonefs_iomap_ops);
80}
81
82static void zonefs_readahead(struct readahead_control *rac)
83{
84 iomap_readahead(rac, &zonefs_iomap_ops);
85}
86
87/*
88 * Map blocks for page writeback. This is used only on conventional zone files,
89 * which implies that the page range can only be within the fixed inode size.
90 */
91static int zonefs_map_blocks(struct iomap_writepage_ctx *wpc,
92 struct inode *inode, loff_t offset)
93{
94 struct zonefs_inode_info *zi = ZONEFS_I(inode);
95
96 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
97 return -EIO;
98 if (WARN_ON_ONCE(offset >= i_size_read(inode)))
99 return -EIO;
100
101 /* If the mapping is already OK, nothing needs to be done */
102 if (offset >= wpc->iomap.offset &&
103 offset < wpc->iomap.offset + wpc->iomap.length)
104 return 0;
105
106 return zonefs_iomap_begin(inode, offset, zi->i_max_size - offset,
107 IOMAP_WRITE, &wpc->iomap, NULL);
108}
109
110static const struct iomap_writeback_ops zonefs_writeback_ops = {
111 .map_blocks = zonefs_map_blocks,
112};
113
114static int zonefs_writepage(struct page *page, struct writeback_control *wbc)
115{
116 struct iomap_writepage_ctx wpc = { };
117
118 return iomap_writepage(page, wbc, &wpc, &zonefs_writeback_ops);
119}
120
121static int zonefs_writepages(struct address_space *mapping,
122 struct writeback_control *wbc)
123{
124 struct iomap_writepage_ctx wpc = { };
125
126 return iomap_writepages(mapping, wbc, &wpc, &zonefs_writeback_ops);
127}
128
129static const struct address_space_operations zonefs_file_aops = {
130 .readpage = zonefs_readpage,
131 .readahead = zonefs_readahead,
132 .writepage = zonefs_writepage,
133 .writepages = zonefs_writepages,
134 .set_page_dirty = iomap_set_page_dirty,
135 .releasepage = iomap_releasepage,
136 .invalidatepage = iomap_invalidatepage,
137 .migratepage = iomap_migrate_page,
138 .is_partially_uptodate = iomap_is_partially_uptodate,
139 .error_remove_page = generic_error_remove_page,
140 .direct_IO = noop_direct_IO,
141};
142
143static void zonefs_update_stats(struct inode *inode, loff_t new_isize)
144{
145 struct super_block *sb = inode->i_sb;
146 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
147 loff_t old_isize = i_size_read(inode);
148 loff_t nr_blocks;
149
150 if (new_isize == old_isize)
151 return;
152
153 spin_lock(&sbi->s_lock);
154
155 /*
156 * This may be called for an update after an IO error.
157 * So beware of the values seen.
158 */
159 if (new_isize < old_isize) {
160 nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits;
161 if (sbi->s_used_blocks > nr_blocks)
162 sbi->s_used_blocks -= nr_blocks;
163 else
164 sbi->s_used_blocks = 0;
165 } else {
166 sbi->s_used_blocks +=
167 (new_isize - old_isize) >> sb->s_blocksize_bits;
168 if (sbi->s_used_blocks > sbi->s_blocks)
169 sbi->s_used_blocks = sbi->s_blocks;
170 }
171
172 spin_unlock(&sbi->s_lock);
173}
174
175/*
176 * Check a zone condition and adjust its file inode access permissions for
177 * offline and readonly zones. Return the inode size corresponding to the
178 * amount of readable data in the zone.
179 */
180static loff_t zonefs_check_zone_condition(struct inode *inode,
181 struct blk_zone *zone, bool warn,
182 bool mount)
183{
184 struct zonefs_inode_info *zi = ZONEFS_I(inode);
185
186 switch (zone->cond) {
187 case BLK_ZONE_COND_OFFLINE:
188 /*
189 * Dead zone: make the inode immutable, disable all accesses
190 * and set the file size to 0 (zone wp set to zone start).
191 */
192 if (warn)
193 zonefs_warn(inode->i_sb, "inode %lu: offline zone\n",
194 inode->i_ino);
195 inode->i_flags |= S_IMMUTABLE;
196 inode->i_mode &= ~0777;
197 zone->wp = zone->start;
198 return 0;
199 case BLK_ZONE_COND_READONLY:
200 /*
201 * The write pointer of read-only zones is invalid. If such a
202 * zone is found during mount, the file size cannot be retrieved
203 * so we treat the zone as offline (mount == true case).
204 * Otherwise, keep the file size as it was when last updated
205 * so that the user can recover data. In both cases, writes are
206 * always disabled for the zone.
207 */
208 if (warn)
209 zonefs_warn(inode->i_sb, "inode %lu: read-only zone\n",
210 inode->i_ino);
211 inode->i_flags |= S_IMMUTABLE;
212 if (mount) {
213 zone->cond = BLK_ZONE_COND_OFFLINE;
214 inode->i_mode &= ~0777;
215 zone->wp = zone->start;
216 return 0;
217 }
218 inode->i_mode &= ~0222;
219 return i_size_read(inode);
220 default:
221 if (zi->i_ztype == ZONEFS_ZTYPE_CNV)
222 return zi->i_max_size;
223 return (zone->wp - zone->start) << SECTOR_SHIFT;
224 }
225}
226
227struct zonefs_ioerr_data {
228 struct inode *inode;
229 bool write;
230};
231
232static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
233 void *data)
234{
235 struct zonefs_ioerr_data *err = data;
236 struct inode *inode = err->inode;
237 struct zonefs_inode_info *zi = ZONEFS_I(inode);
238 struct super_block *sb = inode->i_sb;
239 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
240 loff_t isize, data_size;
241
242 /*
243 * Check the zone condition: if the zone is not "bad" (offline or
244 * read-only), read errors are simply signaled to the IO issuer as long
245 * as there is no inconsistency between the inode size and the amount of
246 * data writen in the zone (data_size).
247 */
248 data_size = zonefs_check_zone_condition(inode, zone, true, false);
249 isize = i_size_read(inode);
250 if (zone->cond != BLK_ZONE_COND_OFFLINE &&
251 zone->cond != BLK_ZONE_COND_READONLY &&
252 !err->write && isize == data_size)
253 return 0;
254
255 /*
256 * At this point, we detected either a bad zone or an inconsistency
257 * between the inode size and the amount of data written in the zone.
258 * For the latter case, the cause may be a write IO error or an external
259 * action on the device. Two error patterns exist:
260 * 1) The inode size is lower than the amount of data in the zone:
261 * a write operation partially failed and data was writen at the end
262 * of the file. This can happen in the case of a large direct IO
263 * needing several BIOs and/or write requests to be processed.
264 * 2) The inode size is larger than the amount of data in the zone:
265 * this can happen with a deferred write error with the use of the
266 * device side write cache after getting successful write IO
267 * completions. Other possibilities are (a) an external corruption,
268 * e.g. an application reset the zone directly, or (b) the device
269 * has a serious problem (e.g. firmware bug).
270 *
271 * In all cases, warn about inode size inconsistency and handle the
272 * IO error according to the zone condition and to the mount options.
273 */
274 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && isize != data_size)
275 zonefs_warn(sb, "inode %lu: invalid size %lld (should be %lld)\n",
276 inode->i_ino, isize, data_size);
277
278 /*
279 * First handle bad zones signaled by hardware. The mount options
280 * errors=zone-ro and errors=zone-offline result in changing the
281 * zone condition to read-only and offline respectively, as if the
282 * condition was signaled by the hardware.
283 */
284 if (zone->cond == BLK_ZONE_COND_OFFLINE ||
285 sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL) {
286 zonefs_warn(sb, "inode %lu: read/write access disabled\n",
287 inode->i_ino);
288 if (zone->cond != BLK_ZONE_COND_OFFLINE) {
289 zone->cond = BLK_ZONE_COND_OFFLINE;
290 data_size = zonefs_check_zone_condition(inode, zone,
291 false, false);
292 }
293 } else if (zone->cond == BLK_ZONE_COND_READONLY ||
294 sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO) {
295 zonefs_warn(sb, "inode %lu: write access disabled\n",
296 inode->i_ino);
297 if (zone->cond != BLK_ZONE_COND_READONLY) {
298 zone->cond = BLK_ZONE_COND_READONLY;
299 data_size = zonefs_check_zone_condition(inode, zone,
300 false, false);
301 }
302 }
303
304 /*
305 * If error=remount-ro was specified, any error result in remounting
306 * the volume as read-only.
307 */
308 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
309 zonefs_warn(sb, "remounting filesystem read-only\n");
310 sb->s_flags |= SB_RDONLY;
311 }
312
313 /*
314 * Update block usage stats and the inode size to prevent access to
315 * invalid data.
316 */
317 zonefs_update_stats(inode, data_size);
318 i_size_write(inode, data_size);
319 zi->i_wpoffset = data_size;
320
321 return 0;
322}
323
324/*
325 * When an file IO error occurs, check the file zone to see if there is a change
326 * in the zone condition (e.g. offline or read-only). For a failed write to a
327 * sequential zone, the zone write pointer position must also be checked to
328 * eventually correct the file size and zonefs inode write pointer offset
329 * (which can be out of sync with the drive due to partial write failures).
330 */
331static void zonefs_io_error(struct inode *inode, bool write)
332{
333 struct zonefs_inode_info *zi = ZONEFS_I(inode);
334 struct super_block *sb = inode->i_sb;
335 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
336 unsigned int noio_flag;
337 unsigned int nr_zones =
338 zi->i_zone_size >> (sbi->s_zone_sectors_shift + SECTOR_SHIFT);
339 struct zonefs_ioerr_data err = {
340 .inode = inode,
341 .write = write,
342 };
343 int ret;
344
345 mutex_lock(&zi->i_truncate_mutex);
346
347 /*
348 * Memory allocations in blkdev_report_zones() can trigger a memory
349 * reclaim which may in turn cause a recursion into zonefs as well as
350 * struct request allocations for the same device. The former case may
351 * end up in a deadlock on the inode truncate mutex, while the latter
352 * may prevent IO forward progress. Executing the report zones under
353 * the GFP_NOIO context avoids both problems.
354 */
355 noio_flag = memalloc_noio_save();
356 ret = blkdev_report_zones(sb->s_bdev, zi->i_zsector, nr_zones,
357 zonefs_io_error_cb, &err);
358 if (ret != nr_zones)
359 zonefs_err(sb, "Get inode %lu zone information failed %d\n",
360 inode->i_ino, ret);
361 memalloc_noio_restore(noio_flag);
362
363 mutex_unlock(&zi->i_truncate_mutex);
364}
365
366static int zonefs_file_truncate(struct inode *inode, loff_t isize)
367{
368 struct zonefs_inode_info *zi = ZONEFS_I(inode);
369 loff_t old_isize;
370 enum req_opf op;
371 int ret = 0;
372
373 /*
374 * Only sequential zone files can be truncated and truncation is allowed
375 * only down to a 0 size, which is equivalent to a zone reset, and to
376 * the maximum file size, which is equivalent to a zone finish.
377 */
378 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
379 return -EPERM;
380
381 if (!isize)
382 op = REQ_OP_ZONE_RESET;
383 else if (isize == zi->i_max_size)
384 op = REQ_OP_ZONE_FINISH;
385 else
386 return -EPERM;
387
388 inode_dio_wait(inode);
389
390 /* Serialize against page faults */
391 down_write(&zi->i_mmap_sem);
392
393 /* Serialize against zonefs_iomap_begin() */
394 mutex_lock(&zi->i_truncate_mutex);
395
396 old_isize = i_size_read(inode);
397 if (isize == old_isize)
398 goto unlock;
399
400 ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector,
401 zi->i_zone_size >> SECTOR_SHIFT, GFP_NOFS);
402 if (ret) {
403 zonefs_err(inode->i_sb,
404 "Zone management operation at %llu failed %d",
405 zi->i_zsector, ret);
406 goto unlock;
407 }
408
409 zonefs_update_stats(inode, isize);
410 truncate_setsize(inode, isize);
411 zi->i_wpoffset = isize;
412
413unlock:
414 mutex_unlock(&zi->i_truncate_mutex);
415 up_write(&zi->i_mmap_sem);
416
417 return ret;
418}
419
420static int zonefs_inode_setattr(struct dentry *dentry, struct iattr *iattr)
421{
422 struct inode *inode = d_inode(dentry);
423 int ret;
424
425 if (unlikely(IS_IMMUTABLE(inode)))
426 return -EPERM;
427
428 ret = setattr_prepare(dentry, iattr);
429 if (ret)
430 return ret;
431
432 /*
433 * Since files and directories cannot be created nor deleted, do not
434 * allow setting any write attributes on the sub-directories grouping
435 * files by zone type.
436 */
437 if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
438 (iattr->ia_mode & 0222))
439 return -EPERM;
440
441 if (((iattr->ia_valid & ATTR_UID) &&
442 !uid_eq(iattr->ia_uid, inode->i_uid)) ||
443 ((iattr->ia_valid & ATTR_GID) &&
444 !gid_eq(iattr->ia_gid, inode->i_gid))) {
445 ret = dquot_transfer(inode, iattr);
446 if (ret)
447 return ret;
448 }
449
450 if (iattr->ia_valid & ATTR_SIZE) {
451 ret = zonefs_file_truncate(inode, iattr->ia_size);
452 if (ret)
453 return ret;
454 }
455
456 setattr_copy(inode, iattr);
457
458 return 0;
459}
460
461static const struct inode_operations zonefs_file_inode_operations = {
462 .setattr = zonefs_inode_setattr,
463};
464
465static int zonefs_file_fsync(struct file *file, loff_t start, loff_t end,
466 int datasync)
467{
468 struct inode *inode = file_inode(file);
469 int ret = 0;
470
471 if (unlikely(IS_IMMUTABLE(inode)))
472 return -EPERM;
473
474 /*
475 * Since only direct writes are allowed in sequential files, page cache
476 * flush is needed only for conventional zone files.
477 */
478 if (ZONEFS_I(inode)->i_ztype == ZONEFS_ZTYPE_CNV)
479 ret = file_write_and_wait_range(file, start, end);
480 if (!ret)
481 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL);
482
483 if (ret)
484 zonefs_io_error(inode, true);
485
486 return ret;
487}
488
489static vm_fault_t zonefs_filemap_fault(struct vm_fault *vmf)
490{
491 struct zonefs_inode_info *zi = ZONEFS_I(file_inode(vmf->vma->vm_file));
492 vm_fault_t ret;
493
494 down_read(&zi->i_mmap_sem);
495 ret = filemap_fault(vmf);
496 up_read(&zi->i_mmap_sem);
497
498 return ret;
499}
500
501static vm_fault_t zonefs_filemap_page_mkwrite(struct vm_fault *vmf)
502{
503 struct inode *inode = file_inode(vmf->vma->vm_file);
504 struct zonefs_inode_info *zi = ZONEFS_I(inode);
505 vm_fault_t ret;
506
507 if (unlikely(IS_IMMUTABLE(inode)))
508 return VM_FAULT_SIGBUS;
509
510 /*
511 * Sanity check: only conventional zone files can have shared
512 * writeable mappings.
513 */
514 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
515 return VM_FAULT_NOPAGE;
516
517 sb_start_pagefault(inode->i_sb);
518 file_update_time(vmf->vma->vm_file);
519
520 /* Serialize against truncates */
521 down_read(&zi->i_mmap_sem);
522 ret = iomap_page_mkwrite(vmf, &zonefs_iomap_ops);
523 up_read(&zi->i_mmap_sem);
524
525 sb_end_pagefault(inode->i_sb);
526 return ret;
527}
528
529static const struct vm_operations_struct zonefs_file_vm_ops = {
530 .fault = zonefs_filemap_fault,
531 .map_pages = filemap_map_pages,
532 .page_mkwrite = zonefs_filemap_page_mkwrite,
533};
534
535static int zonefs_file_mmap(struct file *file, struct vm_area_struct *vma)
536{
537 /*
538 * Conventional zones accept random writes, so their files can support
539 * shared writable mappings. For sequential zone files, only read
540 * mappings are possible since there are no guarantees for write
541 * ordering between msync() and page cache writeback.
542 */
543 if (ZONEFS_I(file_inode(file))->i_ztype == ZONEFS_ZTYPE_SEQ &&
544 (vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
545 return -EINVAL;
546
547 file_accessed(file);
548 vma->vm_ops = &zonefs_file_vm_ops;
549
550 return 0;
551}
552
553static loff_t zonefs_file_llseek(struct file *file, loff_t offset, int whence)
554{
555 loff_t isize = i_size_read(file_inode(file));
556
557 /*
558 * Seeks are limited to below the zone size for conventional zones
559 * and below the zone write pointer for sequential zones. In both
560 * cases, this limit is the inode size.
561 */
562 return generic_file_llseek_size(file, offset, whence, isize, isize);
563}
564
565static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
566 int error, unsigned int flags)
567{
568 struct inode *inode = file_inode(iocb->ki_filp);
569 struct zonefs_inode_info *zi = ZONEFS_I(inode);
570
571 if (error) {
572 zonefs_io_error(inode, true);
573 return error;
574 }
575
576 if (size && zi->i_ztype != ZONEFS_ZTYPE_CNV) {
577 /*
578 * Note that we may be seeing completions out of order,
579 * but that is not a problem since a write completed
580 * successfully necessarily means that all preceding writes
581 * were also successful. So we can safely increase the inode
582 * size to the write end location.
583 */
584 mutex_lock(&zi->i_truncate_mutex);
585 if (i_size_read(inode) < iocb->ki_pos + size) {
586 zonefs_update_stats(inode, iocb->ki_pos + size);
587 i_size_write(inode, iocb->ki_pos + size);
588 }
589 mutex_unlock(&zi->i_truncate_mutex);
590 }
591
592 return 0;
593}
594
595static const struct iomap_dio_ops zonefs_write_dio_ops = {
596 .end_io = zonefs_file_write_dio_end_io,
597};
598
599static ssize_t zonefs_file_dio_append(struct kiocb *iocb, struct iov_iter *from)
600{
601 struct inode *inode = file_inode(iocb->ki_filp);
602 struct zonefs_inode_info *zi = ZONEFS_I(inode);
603 struct block_device *bdev = inode->i_sb->s_bdev;
604 unsigned int max;
605 struct bio *bio;
606 ssize_t size;
607 int nr_pages;
608 ssize_t ret;
609
610 max = queue_max_zone_append_sectors(bdev_get_queue(bdev));
611 max = ALIGN_DOWN(max << SECTOR_SHIFT, inode->i_sb->s_blocksize);
612 iov_iter_truncate(from, max);
613
614 nr_pages = iov_iter_npages(from, BIO_MAX_PAGES);
615 if (!nr_pages)
616 return 0;
617
618 bio = bio_alloc_bioset(GFP_NOFS, nr_pages, &fs_bio_set);
619 if (!bio)
620 return -ENOMEM;
621
622 bio_set_dev(bio, bdev);
623 bio->bi_iter.bi_sector = zi->i_zsector;
624 bio->bi_write_hint = iocb->ki_hint;
625 bio->bi_ioprio = iocb->ki_ioprio;
626 bio->bi_opf = REQ_OP_ZONE_APPEND | REQ_SYNC | REQ_IDLE;
627 if (iocb->ki_flags & IOCB_DSYNC)
628 bio->bi_opf |= REQ_FUA;
629
630 ret = bio_iov_iter_get_pages(bio, from);
631 if (unlikely(ret)) {
632 bio_io_error(bio);
633 return ret;
634 }
635 size = bio->bi_iter.bi_size;
636 task_io_account_write(ret);
637
638 if (iocb->ki_flags & IOCB_HIPRI)
639 bio_set_polled(bio, iocb);
640
641 ret = submit_bio_wait(bio);
642
643 bio_put(bio);
644
645 zonefs_file_write_dio_end_io(iocb, size, ret, 0);
646 if (ret >= 0) {
647 iocb->ki_pos += size;
648 return size;
649 }
650
651 return ret;
652}
653
654/*
655 * Handle direct writes. For sequential zone files, this is the only possible
656 * write path. For these files, check that the user is issuing writes
657 * sequentially from the end of the file. This code assumes that the block layer
658 * delivers write requests to the device in sequential order. This is always the
659 * case if a block IO scheduler implementing the ELEVATOR_F_ZBD_SEQ_WRITE
660 * elevator feature is being used (e.g. mq-deadline). The block layer always
661 * automatically select such an elevator for zoned block devices during the
662 * device initialization.
663 */
664static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
665{
666 struct inode *inode = file_inode(iocb->ki_filp);
667 struct zonefs_inode_info *zi = ZONEFS_I(inode);
668 struct super_block *sb = inode->i_sb;
669 bool sync = is_sync_kiocb(iocb);
670 bool append = false;
671 size_t count;
672 ssize_t ret;
673
674 /*
675 * For async direct IOs to sequential zone files, refuse IOCB_NOWAIT
676 * as this can cause write reordering (e.g. the first aio gets EAGAIN
677 * on the inode lock but the second goes through but is now unaligned).
678 */
679 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !sync &&
680 (iocb->ki_flags & IOCB_NOWAIT))
681 return -EOPNOTSUPP;
682
683 if (iocb->ki_flags & IOCB_NOWAIT) {
684 if (!inode_trylock(inode))
685 return -EAGAIN;
686 } else {
687 inode_lock(inode);
688 }
689
690 ret = generic_write_checks(iocb, from);
691 if (ret <= 0)
692 goto inode_unlock;
693
694 iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos);
695 count = iov_iter_count(from);
696
697 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
698 ret = -EINVAL;
699 goto inode_unlock;
700 }
701
702 /* Enforce sequential writes (append only) in sequential zones */
703 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ) {
704 mutex_lock(&zi->i_truncate_mutex);
705 if (iocb->ki_pos != zi->i_wpoffset) {
706 mutex_unlock(&zi->i_truncate_mutex);
707 ret = -EINVAL;
708 goto inode_unlock;
709 }
710 mutex_unlock(&zi->i_truncate_mutex);
711 append = sync;
712 }
713
714 if (append)
715 ret = zonefs_file_dio_append(iocb, from);
716 else
717 ret = iomap_dio_rw(iocb, from, &zonefs_iomap_ops,
718 &zonefs_write_dio_ops, sync);
719 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
720 (ret > 0 || ret == -EIOCBQUEUED)) {
721 if (ret > 0)
722 count = ret;
723 mutex_lock(&zi->i_truncate_mutex);
724 zi->i_wpoffset += count;
725 mutex_unlock(&zi->i_truncate_mutex);
726 }
727
728inode_unlock:
729 inode_unlock(inode);
730
731 return ret;
732}
733
734static ssize_t zonefs_file_buffered_write(struct kiocb *iocb,
735 struct iov_iter *from)
736{
737 struct inode *inode = file_inode(iocb->ki_filp);
738 struct zonefs_inode_info *zi = ZONEFS_I(inode);
739 ssize_t ret;
740
741 /*
742 * Direct IO writes are mandatory for sequential zone files so that the
743 * write IO issuing order is preserved.
744 */
745 if (zi->i_ztype != ZONEFS_ZTYPE_CNV)
746 return -EIO;
747
748 if (iocb->ki_flags & IOCB_NOWAIT) {
749 if (!inode_trylock(inode))
750 return -EAGAIN;
751 } else {
752 inode_lock(inode);
753 }
754
755 ret = generic_write_checks(iocb, from);
756 if (ret <= 0)
757 goto inode_unlock;
758
759 iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos);
760
761 ret = iomap_file_buffered_write(iocb, from, &zonefs_iomap_ops);
762 if (ret > 0)
763 iocb->ki_pos += ret;
764 else if (ret == -EIO)
765 zonefs_io_error(inode, true);
766
767inode_unlock:
768 inode_unlock(inode);
769 if (ret > 0)
770 ret = generic_write_sync(iocb, ret);
771
772 return ret;
773}
774
775static ssize_t zonefs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
776{
777 struct inode *inode = file_inode(iocb->ki_filp);
778
779 if (unlikely(IS_IMMUTABLE(inode)))
780 return -EPERM;
781
782 if (sb_rdonly(inode->i_sb))
783 return -EROFS;
784
785 /* Write operations beyond the zone size are not allowed */
786 if (iocb->ki_pos >= ZONEFS_I(inode)->i_max_size)
787 return -EFBIG;
788
789 if (iocb->ki_flags & IOCB_DIRECT) {
790 ssize_t ret = zonefs_file_dio_write(iocb, from);
791 if (ret != -ENOTBLK)
792 return ret;
793 }
794
795 return zonefs_file_buffered_write(iocb, from);
796}
797
798static int zonefs_file_read_dio_end_io(struct kiocb *iocb, ssize_t size,
799 int error, unsigned int flags)
800{
801 if (error) {
802 zonefs_io_error(file_inode(iocb->ki_filp), false);
803 return error;
804 }
805
806 return 0;
807}
808
809static const struct iomap_dio_ops zonefs_read_dio_ops = {
810 .end_io = zonefs_file_read_dio_end_io,
811};
812
813static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
814{
815 struct inode *inode = file_inode(iocb->ki_filp);
816 struct zonefs_inode_info *zi = ZONEFS_I(inode);
817 struct super_block *sb = inode->i_sb;
818 loff_t isize;
819 ssize_t ret;
820
821 /* Offline zones cannot be read */
822 if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777)))
823 return -EPERM;
824
825 if (iocb->ki_pos >= zi->i_max_size)
826 return 0;
827
828 if (iocb->ki_flags & IOCB_NOWAIT) {
829 if (!inode_trylock_shared(inode))
830 return -EAGAIN;
831 } else {
832 inode_lock_shared(inode);
833 }
834
835 /* Limit read operations to written data */
836 mutex_lock(&zi->i_truncate_mutex);
837 isize = i_size_read(inode);
838 if (iocb->ki_pos >= isize) {
839 mutex_unlock(&zi->i_truncate_mutex);
840 ret = 0;
841 goto inode_unlock;
842 }
843 iov_iter_truncate(to, isize - iocb->ki_pos);
844 mutex_unlock(&zi->i_truncate_mutex);
845
846 if (iocb->ki_flags & IOCB_DIRECT) {
847 size_t count = iov_iter_count(to);
848
849 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
850 ret = -EINVAL;
851 goto inode_unlock;
852 }
853 file_accessed(iocb->ki_filp);
854 ret = iomap_dio_rw(iocb, to, &zonefs_iomap_ops,
855 &zonefs_read_dio_ops, is_sync_kiocb(iocb));
856 } else {
857 ret = generic_file_read_iter(iocb, to);
858 if (ret == -EIO)
859 zonefs_io_error(inode, false);
860 }
861
862inode_unlock:
863 inode_unlock_shared(inode);
864
865 return ret;
866}
867
868static const struct file_operations zonefs_file_operations = {
869 .open = generic_file_open,
870 .fsync = zonefs_file_fsync,
871 .mmap = zonefs_file_mmap,
872 .llseek = zonefs_file_llseek,
873 .read_iter = zonefs_file_read_iter,
874 .write_iter = zonefs_file_write_iter,
875 .splice_read = generic_file_splice_read,
876 .splice_write = iter_file_splice_write,
877 .iopoll = iomap_dio_iopoll,
878};
879
880static struct kmem_cache *zonefs_inode_cachep;
881
882static struct inode *zonefs_alloc_inode(struct super_block *sb)
883{
884 struct zonefs_inode_info *zi;
885
886 zi = kmem_cache_alloc(zonefs_inode_cachep, GFP_KERNEL);
887 if (!zi)
888 return NULL;
889
890 inode_init_once(&zi->i_vnode);
891 mutex_init(&zi->i_truncate_mutex);
892 init_rwsem(&zi->i_mmap_sem);
893
894 return &zi->i_vnode;
895}
896
897static void zonefs_free_inode(struct inode *inode)
898{
899 kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
900}
901
902/*
903 * File system stat.
904 */
905static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
906{
907 struct super_block *sb = dentry->d_sb;
908 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
909 enum zonefs_ztype t;
910 u64 fsid;
911
912 buf->f_type = ZONEFS_MAGIC;
913 buf->f_bsize = sb->s_blocksize;
914 buf->f_namelen = ZONEFS_NAME_MAX;
915
916 spin_lock(&sbi->s_lock);
917
918 buf->f_blocks = sbi->s_blocks;
919 if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
920 buf->f_bfree = 0;
921 else
922 buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
923 buf->f_bavail = buf->f_bfree;
924
925 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
926 if (sbi->s_nr_files[t])
927 buf->f_files += sbi->s_nr_files[t] + 1;
928 }
929 buf->f_ffree = 0;
930
931 spin_unlock(&sbi->s_lock);
932
933 fsid = le64_to_cpup((void *)sbi->s_uuid.b) ^
934 le64_to_cpup((void *)sbi->s_uuid.b + sizeof(u64));
935 buf->f_fsid.val[0] = (u32)fsid;
936 buf->f_fsid.val[1] = (u32)(fsid >> 32);
937
938 return 0;
939}
940
941enum {
942 Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
943 Opt_err,
944};
945
946static const match_table_t tokens = {
947 { Opt_errors_ro, "errors=remount-ro"},
948 { Opt_errors_zro, "errors=zone-ro"},
949 { Opt_errors_zol, "errors=zone-offline"},
950 { Opt_errors_repair, "errors=repair"},
951 { Opt_err, NULL}
952};
953
954static int zonefs_parse_options(struct super_block *sb, char *options)
955{
956 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
957 substring_t args[MAX_OPT_ARGS];
958 char *p;
959
960 if (!options)
961 return 0;
962
963 while ((p = strsep(&options, ",")) != NULL) {
964 int token;
965
966 if (!*p)
967 continue;
968
969 token = match_token(p, tokens, args);
970 switch (token) {
971 case Opt_errors_ro:
972 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
973 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
974 break;
975 case Opt_errors_zro:
976 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
977 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
978 break;
979 case Opt_errors_zol:
980 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
981 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
982 break;
983 case Opt_errors_repair:
984 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
985 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
986 break;
987 default:
988 return -EINVAL;
989 }
990 }
991
992 return 0;
993}
994
995static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
996{
997 struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
998
999 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
1000 seq_puts(seq, ",errors=remount-ro");
1001 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
1002 seq_puts(seq, ",errors=zone-ro");
1003 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
1004 seq_puts(seq, ",errors=zone-offline");
1005 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
1006 seq_puts(seq, ",errors=repair");
1007
1008 return 0;
1009}
1010
1011static int zonefs_remount(struct super_block *sb, int *flags, char *data)
1012{
1013 sync_filesystem(sb);
1014
1015 return zonefs_parse_options(sb, data);
1016}
1017
1018static const struct super_operations zonefs_sops = {
1019 .alloc_inode = zonefs_alloc_inode,
1020 .free_inode = zonefs_free_inode,
1021 .statfs = zonefs_statfs,
1022 .remount_fs = zonefs_remount,
1023 .show_options = zonefs_show_options,
1024};
1025
1026static const struct inode_operations zonefs_dir_inode_operations = {
1027 .lookup = simple_lookup,
1028 .setattr = zonefs_inode_setattr,
1029};
1030
1031static void zonefs_init_dir_inode(struct inode *parent, struct inode *inode,
1032 enum zonefs_ztype type)
1033{
1034 struct super_block *sb = parent->i_sb;
1035
1036 inode->i_ino = blkdev_nr_zones(sb->s_bdev->bd_disk) + type + 1;
1037 inode_init_owner(inode, parent, S_IFDIR | 0555);
1038 inode->i_op = &zonefs_dir_inode_operations;
1039 inode->i_fop = &simple_dir_operations;
1040 set_nlink(inode, 2);
1041 inc_nlink(parent);
1042}
1043
1044static void zonefs_init_file_inode(struct inode *inode, struct blk_zone *zone,
1045 enum zonefs_ztype type)
1046{
1047 struct super_block *sb = inode->i_sb;
1048 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1049 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1050
1051 inode->i_ino = zone->start >> sbi->s_zone_sectors_shift;
1052 inode->i_mode = S_IFREG | sbi->s_perm;
1053
1054 zi->i_ztype = type;
1055 zi->i_zsector = zone->start;
1056 zi->i_zone_size = zone->len << SECTOR_SHIFT;
1057
1058 zi->i_max_size = min_t(loff_t, MAX_LFS_FILESIZE,
1059 zone->capacity << SECTOR_SHIFT);
1060 zi->i_wpoffset = zonefs_check_zone_condition(inode, zone, true, true);
1061
1062 inode->i_uid = sbi->s_uid;
1063 inode->i_gid = sbi->s_gid;
1064 inode->i_size = zi->i_wpoffset;
1065 inode->i_blocks = zi->i_max_size >> SECTOR_SHIFT;
1066
1067 inode->i_op = &zonefs_file_inode_operations;
1068 inode->i_fop = &zonefs_file_operations;
1069 inode->i_mapping->a_ops = &zonefs_file_aops;
1070
1071 sb->s_maxbytes = max(zi->i_max_size, sb->s_maxbytes);
1072 sbi->s_blocks += zi->i_max_size >> sb->s_blocksize_bits;
1073 sbi->s_used_blocks += zi->i_wpoffset >> sb->s_blocksize_bits;
1074}
1075
1076static struct dentry *zonefs_create_inode(struct dentry *parent,
1077 const char *name, struct blk_zone *zone,
1078 enum zonefs_ztype type)
1079{
1080 struct inode *dir = d_inode(parent);
1081 struct dentry *dentry;
1082 struct inode *inode;
1083
1084 dentry = d_alloc_name(parent, name);
1085 if (!dentry)
1086 return NULL;
1087
1088 inode = new_inode(parent->d_sb);
1089 if (!inode)
1090 goto dput;
1091
1092 inode->i_ctime = inode->i_mtime = inode->i_atime = dir->i_ctime;
1093 if (zone)
1094 zonefs_init_file_inode(inode, zone, type);
1095 else
1096 zonefs_init_dir_inode(dir, inode, type);
1097 d_add(dentry, inode);
1098 dir->i_size++;
1099
1100 return dentry;
1101
1102dput:
1103 dput(dentry);
1104
1105 return NULL;
1106}
1107
1108struct zonefs_zone_data {
1109 struct super_block *sb;
1110 unsigned int nr_zones[ZONEFS_ZTYPE_MAX];
1111 struct blk_zone *zones;
1112};
1113
1114/*
1115 * Create a zone group and populate it with zone files.
1116 */
1117static int zonefs_create_zgroup(struct zonefs_zone_data *zd,
1118 enum zonefs_ztype type)
1119{
1120 struct super_block *sb = zd->sb;
1121 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1122 struct blk_zone *zone, *next, *end;
1123 const char *zgroup_name;
1124 char *file_name;
1125 struct dentry *dir;
1126 unsigned int n = 0;
1127 int ret;
1128
1129 /* If the group is empty, there is nothing to do */
1130 if (!zd->nr_zones[type])
1131 return 0;
1132
1133 file_name = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
1134 if (!file_name)
1135 return -ENOMEM;
1136
1137 if (type == ZONEFS_ZTYPE_CNV)
1138 zgroup_name = "cnv";
1139 else
1140 zgroup_name = "seq";
1141
1142 dir = zonefs_create_inode(sb->s_root, zgroup_name, NULL, type);
1143 if (!dir) {
1144 ret = -ENOMEM;
1145 goto free;
1146 }
1147
1148 /*
1149 * The first zone contains the super block: skip it.
1150 */
1151 end = zd->zones + blkdev_nr_zones(sb->s_bdev->bd_disk);
1152 for (zone = &zd->zones[1]; zone < end; zone = next) {
1153
1154 next = zone + 1;
1155 if (zonefs_zone_type(zone) != type)
1156 continue;
1157
1158 /*
1159 * For conventional zones, contiguous zones can be aggregated
1160 * together to form larger files. Note that this overwrites the
1161 * length of the first zone of the set of contiguous zones
1162 * aggregated together. If one offline or read-only zone is
1163 * found, assume that all zones aggregated have the same
1164 * condition.
1165 */
1166 if (type == ZONEFS_ZTYPE_CNV &&
1167 (sbi->s_features & ZONEFS_F_AGGRCNV)) {
1168 for (; next < end; next++) {
1169 if (zonefs_zone_type(next) != type)
1170 break;
1171 zone->len += next->len;
1172 zone->capacity += next->capacity;
1173 if (next->cond == BLK_ZONE_COND_READONLY &&
1174 zone->cond != BLK_ZONE_COND_OFFLINE)
1175 zone->cond = BLK_ZONE_COND_READONLY;
1176 else if (next->cond == BLK_ZONE_COND_OFFLINE)
1177 zone->cond = BLK_ZONE_COND_OFFLINE;
1178 }
1179 if (zone->capacity != zone->len) {
1180 zonefs_err(sb, "Invalid conventional zone capacity\n");
1181 ret = -EINVAL;
1182 goto free;
1183 }
1184 }
1185
1186 /*
1187 * Use the file number within its group as file name.
1188 */
1189 snprintf(file_name, ZONEFS_NAME_MAX - 1, "%u", n);
1190 if (!zonefs_create_inode(dir, file_name, zone, type)) {
1191 ret = -ENOMEM;
1192 goto free;
1193 }
1194
1195 n++;
1196 }
1197
1198 zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1199 zgroup_name, n, n > 1 ? "s" : "");
1200
1201 sbi->s_nr_files[type] = n;
1202 ret = 0;
1203
1204free:
1205 kfree(file_name);
1206
1207 return ret;
1208}
1209
1210static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
1211 void *data)
1212{
1213 struct zonefs_zone_data *zd = data;
1214
1215 /*
1216 * Count the number of usable zones: the first zone at index 0 contains
1217 * the super block and is ignored.
1218 */
1219 switch (zone->type) {
1220 case BLK_ZONE_TYPE_CONVENTIONAL:
1221 zone->wp = zone->start + zone->len;
1222 if (idx)
1223 zd->nr_zones[ZONEFS_ZTYPE_CNV]++;
1224 break;
1225 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1226 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1227 if (idx)
1228 zd->nr_zones[ZONEFS_ZTYPE_SEQ]++;
1229 break;
1230 default:
1231 zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
1232 zone->type);
1233 return -EIO;
1234 }
1235
1236 memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
1237
1238 return 0;
1239}
1240
1241static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
1242{
1243 struct block_device *bdev = zd->sb->s_bdev;
1244 int ret;
1245
1246 zd->zones = kvcalloc(blkdev_nr_zones(bdev->bd_disk),
1247 sizeof(struct blk_zone), GFP_KERNEL);
1248 if (!zd->zones)
1249 return -ENOMEM;
1250
1251 /* Get zones information from the device */
1252 ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
1253 zonefs_get_zone_info_cb, zd);
1254 if (ret < 0) {
1255 zonefs_err(zd->sb, "Zone report failed %d\n", ret);
1256 return ret;
1257 }
1258
1259 if (ret != blkdev_nr_zones(bdev->bd_disk)) {
1260 zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
1261 ret, blkdev_nr_zones(bdev->bd_disk));
1262 return -EIO;
1263 }
1264
1265 return 0;
1266}
1267
1268static inline void zonefs_cleanup_zone_info(struct zonefs_zone_data *zd)
1269{
1270 kvfree(zd->zones);
1271}
1272
1273/*
1274 * Read super block information from the device.
1275 */
1276static int zonefs_read_super(struct super_block *sb)
1277{
1278 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1279 struct zonefs_super *super;
1280 u32 crc, stored_crc;
1281 struct page *page;
1282 struct bio_vec bio_vec;
1283 struct bio bio;
1284 int ret;
1285
1286 page = alloc_page(GFP_KERNEL);
1287 if (!page)
1288 return -ENOMEM;
1289
1290 bio_init(&bio, &bio_vec, 1);
1291 bio.bi_iter.bi_sector = 0;
1292 bio.bi_opf = REQ_OP_READ;
1293 bio_set_dev(&bio, sb->s_bdev);
1294 bio_add_page(&bio, page, PAGE_SIZE, 0);
1295
1296 ret = submit_bio_wait(&bio);
1297 if (ret)
1298 goto free_page;
1299
1300 super = kmap(page);
1301
1302 ret = -EINVAL;
1303 if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1304 goto unmap;
1305
1306 stored_crc = le32_to_cpu(super->s_crc);
1307 super->s_crc = 0;
1308 crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1309 if (crc != stored_crc) {
1310 zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1311 crc, stored_crc);
1312 goto unmap;
1313 }
1314
1315 sbi->s_features = le64_to_cpu(super->s_features);
1316 if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1317 zonefs_err(sb, "Unknown features set 0x%llx\n",
1318 sbi->s_features);
1319 goto unmap;
1320 }
1321
1322 if (sbi->s_features & ZONEFS_F_UID) {
1323 sbi->s_uid = make_kuid(current_user_ns(),
1324 le32_to_cpu(super->s_uid));
1325 if (!uid_valid(sbi->s_uid)) {
1326 zonefs_err(sb, "Invalid UID feature\n");
1327 goto unmap;
1328 }
1329 }
1330
1331 if (sbi->s_features & ZONEFS_F_GID) {
1332 sbi->s_gid = make_kgid(current_user_ns(),
1333 le32_to_cpu(super->s_gid));
1334 if (!gid_valid(sbi->s_gid)) {
1335 zonefs_err(sb, "Invalid GID feature\n");
1336 goto unmap;
1337 }
1338 }
1339
1340 if (sbi->s_features & ZONEFS_F_PERM)
1341 sbi->s_perm = le32_to_cpu(super->s_perm);
1342
1343 if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1344 zonefs_err(sb, "Reserved area is being used\n");
1345 goto unmap;
1346 }
1347
1348 import_uuid(&sbi->s_uuid, super->s_uuid);
1349 ret = 0;
1350
1351unmap:
1352 kunmap(page);
1353free_page:
1354 __free_page(page);
1355
1356 return ret;
1357}
1358
1359/*
1360 * Check that the device is zoned. If it is, get the list of zones and create
1361 * sub-directories and files according to the device zone configuration and
1362 * format options.
1363 */
1364static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1365{
1366 struct zonefs_zone_data zd;
1367 struct zonefs_sb_info *sbi;
1368 struct inode *inode;
1369 enum zonefs_ztype t;
1370 int ret;
1371
1372 if (!bdev_is_zoned(sb->s_bdev)) {
1373 zonefs_err(sb, "Not a zoned block device\n");
1374 return -EINVAL;
1375 }
1376
1377 /*
1378 * Initialize super block information: the maximum file size is updated
1379 * when the zone files are created so that the format option
1380 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1381 * beyond the zone size is taken into account.
1382 */
1383 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1384 if (!sbi)
1385 return -ENOMEM;
1386
1387 spin_lock_init(&sbi->s_lock);
1388 sb->s_fs_info = sbi;
1389 sb->s_magic = ZONEFS_MAGIC;
1390 sb->s_maxbytes = 0;
1391 sb->s_op = &zonefs_sops;
1392 sb->s_time_gran = 1;
1393
1394 /*
1395 * The block size is set to the device physical sector size to ensure
1396 * that write operations on 512e devices (512B logical block and 4KB
1397 * physical block) are always aligned to the device physical blocks,
1398 * as mandated by the ZBC/ZAC specifications.
1399 */
1400 sb_set_blocksize(sb, bdev_physical_block_size(sb->s_bdev));
1401 sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1402 sbi->s_uid = GLOBAL_ROOT_UID;
1403 sbi->s_gid = GLOBAL_ROOT_GID;
1404 sbi->s_perm = 0640;
1405 sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1406
1407 ret = zonefs_read_super(sb);
1408 if (ret)
1409 return ret;
1410
1411 ret = zonefs_parse_options(sb, data);
1412 if (ret)
1413 return ret;
1414
1415 memset(&zd, 0, sizeof(struct zonefs_zone_data));
1416 zd.sb = sb;
1417 ret = zonefs_get_zone_info(&zd);
1418 if (ret)
1419 goto cleanup;
1420
1421 zonefs_info(sb, "Mounting %u zones",
1422 blkdev_nr_zones(sb->s_bdev->bd_disk));
1423
1424 /* Create root directory inode */
1425 ret = -ENOMEM;
1426 inode = new_inode(sb);
1427 if (!inode)
1428 goto cleanup;
1429
1430 inode->i_ino = blkdev_nr_zones(sb->s_bdev->bd_disk);
1431 inode->i_mode = S_IFDIR | 0555;
1432 inode->i_ctime = inode->i_mtime = inode->i_atime = current_time(inode);
1433 inode->i_op = &zonefs_dir_inode_operations;
1434 inode->i_fop = &simple_dir_operations;
1435 set_nlink(inode, 2);
1436
1437 sb->s_root = d_make_root(inode);
1438 if (!sb->s_root)
1439 goto cleanup;
1440
1441 /* Create and populate files in zone groups directories */
1442 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
1443 ret = zonefs_create_zgroup(&zd, t);
1444 if (ret)
1445 break;
1446 }
1447
1448cleanup:
1449 zonefs_cleanup_zone_info(&zd);
1450
1451 return ret;
1452}
1453
1454static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1455 int flags, const char *dev_name, void *data)
1456{
1457 return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1458}
1459
1460static void zonefs_kill_super(struct super_block *sb)
1461{
1462 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1463
1464 if (sb->s_root)
1465 d_genocide(sb->s_root);
1466 kill_block_super(sb);
1467 kfree(sbi);
1468}
1469
1470/*
1471 * File system definition and registration.
1472 */
1473static struct file_system_type zonefs_type = {
1474 .owner = THIS_MODULE,
1475 .name = "zonefs",
1476 .mount = zonefs_mount,
1477 .kill_sb = zonefs_kill_super,
1478 .fs_flags = FS_REQUIRES_DEV,
1479};
1480
1481static int __init zonefs_init_inodecache(void)
1482{
1483 zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1484 sizeof(struct zonefs_inode_info), 0,
1485 (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1486 NULL);
1487 if (zonefs_inode_cachep == NULL)
1488 return -ENOMEM;
1489 return 0;
1490}
1491
1492static void zonefs_destroy_inodecache(void)
1493{
1494 /*
1495 * Make sure all delayed rcu free inodes are flushed before we
1496 * destroy the inode cache.
1497 */
1498 rcu_barrier();
1499 kmem_cache_destroy(zonefs_inode_cachep);
1500}
1501
1502static int __init zonefs_init(void)
1503{
1504 int ret;
1505
1506 BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1507
1508 ret = zonefs_init_inodecache();
1509 if (ret)
1510 return ret;
1511
1512 ret = register_filesystem(&zonefs_type);
1513 if (ret) {
1514 zonefs_destroy_inodecache();
1515 return ret;
1516 }
1517
1518 return 0;
1519}
1520
1521static void __exit zonefs_exit(void)
1522{
1523 zonefs_destroy_inodecache();
1524 unregister_filesystem(&zonefs_type);
1525}
1526
1527MODULE_AUTHOR("Damien Le Moal");
1528MODULE_DESCRIPTION("Zone file system for zoned block devices");
1529MODULE_LICENSE("GPL");
1530module_init(zonefs_init);
1531module_exit(zonefs_exit);