xref: /linux/fs/zonefs/super.c (revision e6a901a00822659181c93c86d8bbc2a17779fddc)
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/uio.h>
19 #include <linux/mman.h>
20 #include <linux/sched/mm.h>
21 #include <linux/crc32.h>
22 #include <linux/task_io_accounting_ops.h>
23 #include <linux/fs_parser.h>
24 #include <linux/fs_context.h>
25 
26 #include "zonefs.h"
27 
28 #define CREATE_TRACE_POINTS
29 #include "trace.h"
30 
31 /*
32  * Get the name of a zone group directory.
33  */
34 static const char *zonefs_zgroup_name(enum zonefs_ztype ztype)
35 {
36 	switch (ztype) {
37 	case ZONEFS_ZTYPE_CNV:
38 		return "cnv";
39 	case ZONEFS_ZTYPE_SEQ:
40 		return "seq";
41 	default:
42 		WARN_ON_ONCE(1);
43 		return "???";
44 	}
45 }
46 
47 /*
48  * Manage the active zone count.
49  */
50 static void zonefs_account_active(struct super_block *sb,
51 				  struct zonefs_zone *z)
52 {
53 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
54 
55 	if (zonefs_zone_is_cnv(z))
56 		return;
57 
58 	/*
59 	 * For zones that transitioned to the offline or readonly condition,
60 	 * we only need to clear the active state.
61 	 */
62 	if (z->z_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY))
63 		goto out;
64 
65 	/*
66 	 * If the zone is active, that is, if it is explicitly open or
67 	 * partially written, check if it was already accounted as active.
68 	 */
69 	if ((z->z_flags & ZONEFS_ZONE_OPEN) ||
70 	    (z->z_wpoffset > 0 && z->z_wpoffset < z->z_capacity)) {
71 		if (!(z->z_flags & ZONEFS_ZONE_ACTIVE)) {
72 			z->z_flags |= ZONEFS_ZONE_ACTIVE;
73 			atomic_inc(&sbi->s_active_seq_files);
74 		}
75 		return;
76 	}
77 
78 out:
79 	/* The zone is not active. If it was, update the active count */
80 	if (z->z_flags & ZONEFS_ZONE_ACTIVE) {
81 		z->z_flags &= ~ZONEFS_ZONE_ACTIVE;
82 		atomic_dec(&sbi->s_active_seq_files);
83 	}
84 }
85 
86 /*
87  * Manage the active zone count. Called with zi->i_truncate_mutex held.
88  */
89 void zonefs_inode_account_active(struct inode *inode)
90 {
91 	lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
92 
93 	return zonefs_account_active(inode->i_sb, zonefs_inode_zone(inode));
94 }
95 
96 /*
97  * Execute a zone management operation.
98  */
99 static int zonefs_zone_mgmt(struct super_block *sb,
100 			    struct zonefs_zone *z, enum req_op op)
101 {
102 	int ret;
103 
104 	/*
105 	 * With ZNS drives, closing an explicitly open zone that has not been
106 	 * written will change the zone state to "closed", that is, the zone
107 	 * will remain active. Since this can then cause failure of explicit
108 	 * open operation on other zones if the drive active zone resources
109 	 * are exceeded, make sure that the zone does not remain active by
110 	 * resetting it.
111 	 */
112 	if (op == REQ_OP_ZONE_CLOSE && !z->z_wpoffset)
113 		op = REQ_OP_ZONE_RESET;
114 
115 	trace_zonefs_zone_mgmt(sb, z, op);
116 	ret = blkdev_zone_mgmt(sb->s_bdev, op, z->z_sector,
117 			       z->z_size >> SECTOR_SHIFT);
118 	if (ret) {
119 		zonefs_err(sb,
120 			   "Zone management operation %s at %llu failed %d\n",
121 			   blk_op_str(op), z->z_sector, ret);
122 		return ret;
123 	}
124 
125 	return 0;
126 }
127 
128 int zonefs_inode_zone_mgmt(struct inode *inode, enum req_op op)
129 {
130 	lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
131 
132 	return zonefs_zone_mgmt(inode->i_sb, zonefs_inode_zone(inode), op);
133 }
134 
135 void zonefs_i_size_write(struct inode *inode, loff_t isize)
136 {
137 	struct zonefs_zone *z = zonefs_inode_zone(inode);
138 
139 	i_size_write(inode, isize);
140 
141 	/*
142 	 * A full zone is no longer open/active and does not need
143 	 * explicit closing.
144 	 */
145 	if (isize >= z->z_capacity) {
146 		struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
147 
148 		if (z->z_flags & ZONEFS_ZONE_ACTIVE)
149 			atomic_dec(&sbi->s_active_seq_files);
150 		z->z_flags &= ~(ZONEFS_ZONE_OPEN | ZONEFS_ZONE_ACTIVE);
151 	}
152 }
153 
154 void zonefs_update_stats(struct inode *inode, loff_t new_isize)
155 {
156 	struct super_block *sb = inode->i_sb;
157 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
158 	loff_t old_isize = i_size_read(inode);
159 	loff_t nr_blocks;
160 
161 	if (new_isize == old_isize)
162 		return;
163 
164 	spin_lock(&sbi->s_lock);
165 
166 	/*
167 	 * This may be called for an update after an IO error.
168 	 * So beware of the values seen.
169 	 */
170 	if (new_isize < old_isize) {
171 		nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits;
172 		if (sbi->s_used_blocks > nr_blocks)
173 			sbi->s_used_blocks -= nr_blocks;
174 		else
175 			sbi->s_used_blocks = 0;
176 	} else {
177 		sbi->s_used_blocks +=
178 			(new_isize - old_isize) >> sb->s_blocksize_bits;
179 		if (sbi->s_used_blocks > sbi->s_blocks)
180 			sbi->s_used_blocks = sbi->s_blocks;
181 	}
182 
183 	spin_unlock(&sbi->s_lock);
184 }
185 
186 /*
187  * Check a zone condition. Return the amount of written (and still readable)
188  * data in the zone.
189  */
190 static loff_t zonefs_check_zone_condition(struct super_block *sb,
191 					  struct zonefs_zone *z,
192 					  struct blk_zone *zone)
193 {
194 	switch (zone->cond) {
195 	case BLK_ZONE_COND_OFFLINE:
196 		zonefs_warn(sb, "Zone %llu: offline zone\n",
197 			    z->z_sector);
198 		z->z_flags |= ZONEFS_ZONE_OFFLINE;
199 		return 0;
200 	case BLK_ZONE_COND_READONLY:
201 		/*
202 		 * The write pointer of read-only zones is invalid, so we cannot
203 		 * determine the zone wpoffset (inode size). We thus keep the
204 		 * zone wpoffset as is, which leads to an empty file
205 		 * (wpoffset == 0) on mount. For a runtime error, this keeps
206 		 * the inode size as it was when last updated so that the user
207 		 * can recover data.
208 		 */
209 		zonefs_warn(sb, "Zone %llu: read-only zone\n",
210 			    z->z_sector);
211 		z->z_flags |= ZONEFS_ZONE_READONLY;
212 		if (zonefs_zone_is_cnv(z))
213 			return z->z_capacity;
214 		return z->z_wpoffset;
215 	case BLK_ZONE_COND_FULL:
216 		/* The write pointer of full zones is invalid. */
217 		return z->z_capacity;
218 	default:
219 		if (zonefs_zone_is_cnv(z))
220 			return z->z_capacity;
221 		return (zone->wp - zone->start) << SECTOR_SHIFT;
222 	}
223 }
224 
225 /*
226  * Check a zone condition and adjust its inode access permissions for
227  * offline and readonly zones.
228  */
229 static void zonefs_inode_update_mode(struct inode *inode)
230 {
231 	struct zonefs_zone *z = zonefs_inode_zone(inode);
232 
233 	if (z->z_flags & ZONEFS_ZONE_OFFLINE) {
234 		/* Offline zones cannot be read nor written */
235 		inode->i_flags |= S_IMMUTABLE;
236 		inode->i_mode &= ~0777;
237 	} else if (z->z_flags & ZONEFS_ZONE_READONLY) {
238 		/* Readonly zones cannot be written */
239 		inode->i_flags |= S_IMMUTABLE;
240 		if (z->z_flags & ZONEFS_ZONE_INIT_MODE)
241 			inode->i_mode &= ~0777;
242 		else
243 			inode->i_mode &= ~0222;
244 	}
245 
246 	z->z_flags &= ~ZONEFS_ZONE_INIT_MODE;
247 	z->z_mode = inode->i_mode;
248 }
249 
250 static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
251 			      void *data)
252 {
253 	struct blk_zone *z = data;
254 
255 	*z = *zone;
256 	return 0;
257 }
258 
259 static void zonefs_handle_io_error(struct inode *inode, struct blk_zone *zone,
260 				   bool write)
261 {
262 	struct zonefs_zone *z = zonefs_inode_zone(inode);
263 	struct super_block *sb = inode->i_sb;
264 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
265 	loff_t isize, data_size;
266 
267 	/*
268 	 * Check the zone condition: if the zone is not "bad" (offline or
269 	 * read-only), read errors are simply signaled to the IO issuer as long
270 	 * as there is no inconsistency between the inode size and the amount of
271 	 * data writen in the zone (data_size).
272 	 */
273 	data_size = zonefs_check_zone_condition(sb, z, zone);
274 	isize = i_size_read(inode);
275 	if (!(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)) &&
276 	    !write && isize == data_size)
277 		return;
278 
279 	/*
280 	 * At this point, we detected either a bad zone or an inconsistency
281 	 * between the inode size and the amount of data written in the zone.
282 	 * For the latter case, the cause may be a write IO error or an external
283 	 * action on the device. Two error patterns exist:
284 	 * 1) The inode size is lower than the amount of data in the zone:
285 	 *    a write operation partially failed and data was writen at the end
286 	 *    of the file. This can happen in the case of a large direct IO
287 	 *    needing several BIOs and/or write requests to be processed.
288 	 * 2) The inode size is larger than the amount of data in the zone:
289 	 *    this can happen with a deferred write error with the use of the
290 	 *    device side write cache after getting successful write IO
291 	 *    completions. Other possibilities are (a) an external corruption,
292 	 *    e.g. an application reset the zone directly, or (b) the device
293 	 *    has a serious problem (e.g. firmware bug).
294 	 *
295 	 * In all cases, warn about inode size inconsistency and handle the
296 	 * IO error according to the zone condition and to the mount options.
297 	 */
298 	if (isize != data_size)
299 		zonefs_warn(sb,
300 			    "inode %lu: invalid size %lld (should be %lld)\n",
301 			    inode->i_ino, isize, data_size);
302 
303 	/*
304 	 * First handle bad zones signaled by hardware. The mount options
305 	 * errors=zone-ro and errors=zone-offline result in changing the
306 	 * zone condition to read-only and offline respectively, as if the
307 	 * condition was signaled by the hardware.
308 	 */
309 	if ((z->z_flags & ZONEFS_ZONE_OFFLINE) ||
310 	    (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)) {
311 		zonefs_warn(sb, "inode %lu: read/write access disabled\n",
312 			    inode->i_ino);
313 		if (!(z->z_flags & ZONEFS_ZONE_OFFLINE))
314 			z->z_flags |= ZONEFS_ZONE_OFFLINE;
315 		zonefs_inode_update_mode(inode);
316 		data_size = 0;
317 	} else if ((z->z_flags & ZONEFS_ZONE_READONLY) ||
318 		   (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)) {
319 		zonefs_warn(sb, "inode %lu: write access disabled\n",
320 			    inode->i_ino);
321 		if (!(z->z_flags & ZONEFS_ZONE_READONLY))
322 			z->z_flags |= ZONEFS_ZONE_READONLY;
323 		zonefs_inode_update_mode(inode);
324 		data_size = isize;
325 	} else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO &&
326 		   data_size > isize) {
327 		/* Do not expose garbage data */
328 		data_size = isize;
329 	}
330 
331 	/*
332 	 * If the filesystem is mounted with the explicit-open mount option, we
333 	 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
334 	 * the read-only or offline condition, to avoid attempting an explicit
335 	 * close of the zone when the inode file is closed.
336 	 */
337 	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
338 	    (z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)))
339 		z->z_flags &= ~ZONEFS_ZONE_OPEN;
340 
341 	/*
342 	 * If error=remount-ro was specified, any error result in remounting
343 	 * the volume as read-only.
344 	 */
345 	if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
346 		zonefs_warn(sb, "remounting filesystem read-only\n");
347 		sb->s_flags |= SB_RDONLY;
348 	}
349 
350 	/*
351 	 * Update block usage stats and the inode size  to prevent access to
352 	 * invalid data.
353 	 */
354 	zonefs_update_stats(inode, data_size);
355 	zonefs_i_size_write(inode, data_size);
356 	z->z_wpoffset = data_size;
357 	zonefs_inode_account_active(inode);
358 }
359 
360 /*
361  * When an file IO error occurs, check the file zone to see if there is a change
362  * in the zone condition (e.g. offline or read-only). For a failed write to a
363  * sequential zone, the zone write pointer position must also be checked to
364  * eventually correct the file size and zonefs inode write pointer offset
365  * (which can be out of sync with the drive due to partial write failures).
366  */
367 void __zonefs_io_error(struct inode *inode, bool write)
368 {
369 	struct zonefs_zone *z = zonefs_inode_zone(inode);
370 	struct super_block *sb = inode->i_sb;
371 	unsigned int noio_flag;
372 	struct blk_zone zone;
373 	int ret;
374 
375 	/*
376 	 * Conventional zone have no write pointer and cannot become read-only
377 	 * or offline. So simply fake a report for a single or aggregated zone
378 	 * and let zonefs_handle_io_error() correct the zone inode information
379 	 * according to the mount options.
380 	 */
381 	if (!zonefs_zone_is_seq(z)) {
382 		zone.start = z->z_sector;
383 		zone.len = z->z_size >> SECTOR_SHIFT;
384 		zone.wp = zone.start + zone.len;
385 		zone.type = BLK_ZONE_TYPE_CONVENTIONAL;
386 		zone.cond = BLK_ZONE_COND_NOT_WP;
387 		zone.capacity = zone.len;
388 		goto handle_io_error;
389 	}
390 
391 	/*
392 	 * Memory allocations in blkdev_report_zones() can trigger a memory
393 	 * reclaim which may in turn cause a recursion into zonefs as well as
394 	 * struct request allocations for the same device. The former case may
395 	 * end up in a deadlock on the inode truncate mutex, while the latter
396 	 * may prevent IO forward progress. Executing the report zones under
397 	 * the GFP_NOIO context avoids both problems.
398 	 */
399 	noio_flag = memalloc_noio_save();
400 	ret = blkdev_report_zones(sb->s_bdev, z->z_sector, 1,
401 				  zonefs_io_error_cb, &zone);
402 	memalloc_noio_restore(noio_flag);
403 
404 	if (ret != 1) {
405 		zonefs_err(sb, "Get inode %lu zone information failed %d\n",
406 			   inode->i_ino, ret);
407 		zonefs_warn(sb, "remounting filesystem read-only\n");
408 		sb->s_flags |= SB_RDONLY;
409 		return;
410 	}
411 
412 handle_io_error:
413 	zonefs_handle_io_error(inode, &zone, write);
414 }
415 
416 static struct kmem_cache *zonefs_inode_cachep;
417 
418 static struct inode *zonefs_alloc_inode(struct super_block *sb)
419 {
420 	struct zonefs_inode_info *zi;
421 
422 	zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL);
423 	if (!zi)
424 		return NULL;
425 
426 	inode_init_once(&zi->i_vnode);
427 	mutex_init(&zi->i_truncate_mutex);
428 	zi->i_wr_refcnt = 0;
429 
430 	return &zi->i_vnode;
431 }
432 
433 static void zonefs_free_inode(struct inode *inode)
434 {
435 	kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
436 }
437 
438 /*
439  * File system stat.
440  */
441 static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
442 {
443 	struct super_block *sb = dentry->d_sb;
444 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
445 	enum zonefs_ztype t;
446 
447 	buf->f_type = ZONEFS_MAGIC;
448 	buf->f_bsize = sb->s_blocksize;
449 	buf->f_namelen = ZONEFS_NAME_MAX;
450 
451 	spin_lock(&sbi->s_lock);
452 
453 	buf->f_blocks = sbi->s_blocks;
454 	if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
455 		buf->f_bfree = 0;
456 	else
457 		buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
458 	buf->f_bavail = buf->f_bfree;
459 
460 	for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
461 		if (sbi->s_zgroup[t].g_nr_zones)
462 			buf->f_files += sbi->s_zgroup[t].g_nr_zones + 1;
463 	}
464 	buf->f_ffree = 0;
465 
466 	spin_unlock(&sbi->s_lock);
467 
468 	buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b);
469 
470 	return 0;
471 }
472 
473 enum {
474 	Opt_errors, Opt_explicit_open,
475 };
476 
477 struct zonefs_context {
478 	unsigned long s_mount_opts;
479 };
480 
481 static const struct constant_table zonefs_param_errors[] = {
482 	{"remount-ro",		ZONEFS_MNTOPT_ERRORS_RO},
483 	{"zone-ro",		ZONEFS_MNTOPT_ERRORS_ZRO},
484 	{"zone-offline",	ZONEFS_MNTOPT_ERRORS_ZOL},
485 	{"repair", 		ZONEFS_MNTOPT_ERRORS_REPAIR},
486 	{}
487 };
488 
489 static const struct fs_parameter_spec zonefs_param_spec[] = {
490 	fsparam_enum	("errors",		Opt_errors, zonefs_param_errors),
491 	fsparam_flag	("explicit-open",	Opt_explicit_open),
492 	{}
493 };
494 
495 static int zonefs_parse_param(struct fs_context *fc, struct fs_parameter *param)
496 {
497 	struct zonefs_context *ctx = fc->fs_private;
498 	struct fs_parse_result result;
499 	int opt;
500 
501 	opt = fs_parse(fc, zonefs_param_spec, param, &result);
502 	if (opt < 0)
503 		return opt;
504 
505 	switch (opt) {
506 	case Opt_errors:
507 		ctx->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
508 		ctx->s_mount_opts |= result.uint_32;
509 		break;
510 	case Opt_explicit_open:
511 		ctx->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
512 		break;
513 	default:
514 		return -EINVAL;
515 	}
516 
517 	return 0;
518 }
519 
520 static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
521 {
522 	struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
523 
524 	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
525 		seq_puts(seq, ",errors=remount-ro");
526 	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
527 		seq_puts(seq, ",errors=zone-ro");
528 	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
529 		seq_puts(seq, ",errors=zone-offline");
530 	if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
531 		seq_puts(seq, ",errors=repair");
532 
533 	return 0;
534 }
535 
536 static int zonefs_inode_setattr(struct mnt_idmap *idmap,
537 				struct dentry *dentry, struct iattr *iattr)
538 {
539 	struct inode *inode = d_inode(dentry);
540 	int ret;
541 
542 	if (unlikely(IS_IMMUTABLE(inode)))
543 		return -EPERM;
544 
545 	ret = setattr_prepare(&nop_mnt_idmap, dentry, iattr);
546 	if (ret)
547 		return ret;
548 
549 	/*
550 	 * Since files and directories cannot be created nor deleted, do not
551 	 * allow setting any write attributes on the sub-directories grouping
552 	 * files by zone type.
553 	 */
554 	if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
555 	    (iattr->ia_mode & 0222))
556 		return -EPERM;
557 
558 	if (((iattr->ia_valid & ATTR_UID) &&
559 	     !uid_eq(iattr->ia_uid, inode->i_uid)) ||
560 	    ((iattr->ia_valid & ATTR_GID) &&
561 	     !gid_eq(iattr->ia_gid, inode->i_gid))) {
562 		ret = dquot_transfer(&nop_mnt_idmap, inode, iattr);
563 		if (ret)
564 			return ret;
565 	}
566 
567 	if (iattr->ia_valid & ATTR_SIZE) {
568 		ret = zonefs_file_truncate(inode, iattr->ia_size);
569 		if (ret)
570 			return ret;
571 	}
572 
573 	setattr_copy(&nop_mnt_idmap, inode, iattr);
574 
575 	if (S_ISREG(inode->i_mode)) {
576 		struct zonefs_zone *z = zonefs_inode_zone(inode);
577 
578 		z->z_mode = inode->i_mode;
579 		z->z_uid = inode->i_uid;
580 		z->z_gid = inode->i_gid;
581 	}
582 
583 	return 0;
584 }
585 
586 static const struct inode_operations zonefs_file_inode_operations = {
587 	.setattr	= zonefs_inode_setattr,
588 };
589 
590 static long zonefs_fname_to_fno(const struct qstr *fname)
591 {
592 	const char *name = fname->name;
593 	unsigned int len = fname->len;
594 	long fno = 0, shift = 1;
595 	const char *rname;
596 	char c = *name;
597 	unsigned int i;
598 
599 	/*
600 	 * File names are always a base-10 number string without any
601 	 * leading 0s.
602 	 */
603 	if (!isdigit(c))
604 		return -ENOENT;
605 
606 	if (len > 1 && c == '0')
607 		return -ENOENT;
608 
609 	if (len == 1)
610 		return c - '0';
611 
612 	for (i = 0, rname = name + len - 1; i < len; i++, rname--) {
613 		c = *rname;
614 		if (!isdigit(c))
615 			return -ENOENT;
616 		fno += (c - '0') * shift;
617 		shift *= 10;
618 	}
619 
620 	return fno;
621 }
622 
623 static struct inode *zonefs_get_file_inode(struct inode *dir,
624 					   struct dentry *dentry)
625 {
626 	struct zonefs_zone_group *zgroup = dir->i_private;
627 	struct super_block *sb = dir->i_sb;
628 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
629 	struct zonefs_zone *z;
630 	struct inode *inode;
631 	ino_t ino;
632 	long fno;
633 
634 	/* Get the file number from the file name */
635 	fno = zonefs_fname_to_fno(&dentry->d_name);
636 	if (fno < 0)
637 		return ERR_PTR(fno);
638 
639 	if (!zgroup->g_nr_zones || fno >= zgroup->g_nr_zones)
640 		return ERR_PTR(-ENOENT);
641 
642 	z = &zgroup->g_zones[fno];
643 	ino = z->z_sector >> sbi->s_zone_sectors_shift;
644 	inode = iget_locked(sb, ino);
645 	if (!inode)
646 		return ERR_PTR(-ENOMEM);
647 	if (!(inode->i_state & I_NEW)) {
648 		WARN_ON_ONCE(inode->i_private != z);
649 		return inode;
650 	}
651 
652 	inode->i_ino = ino;
653 	inode->i_mode = z->z_mode;
654 	inode_set_mtime_to_ts(inode,
655 			      inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(dir))));
656 	inode->i_uid = z->z_uid;
657 	inode->i_gid = z->z_gid;
658 	inode->i_size = z->z_wpoffset;
659 	inode->i_blocks = z->z_capacity >> SECTOR_SHIFT;
660 	inode->i_private = z;
661 
662 	inode->i_op = &zonefs_file_inode_operations;
663 	inode->i_fop = &zonefs_file_operations;
664 	inode->i_mapping->a_ops = &zonefs_file_aops;
665 
666 	/* Update the inode access rights depending on the zone condition */
667 	zonefs_inode_update_mode(inode);
668 
669 	unlock_new_inode(inode);
670 
671 	return inode;
672 }
673 
674 static struct inode *zonefs_get_zgroup_inode(struct super_block *sb,
675 					     enum zonefs_ztype ztype)
676 {
677 	struct inode *root = d_inode(sb->s_root);
678 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
679 	struct inode *inode;
680 	ino_t ino = bdev_nr_zones(sb->s_bdev) + ztype + 1;
681 
682 	inode = iget_locked(sb, ino);
683 	if (!inode)
684 		return ERR_PTR(-ENOMEM);
685 	if (!(inode->i_state & I_NEW))
686 		return inode;
687 
688 	inode->i_ino = ino;
689 	inode_init_owner(&nop_mnt_idmap, inode, root, S_IFDIR | 0555);
690 	inode->i_size = sbi->s_zgroup[ztype].g_nr_zones;
691 	inode_set_mtime_to_ts(inode,
692 			      inode_set_atime_to_ts(inode, inode_set_ctime_to_ts(inode, inode_get_ctime(root))));
693 	inode->i_private = &sbi->s_zgroup[ztype];
694 	set_nlink(inode, 2);
695 
696 	inode->i_op = &zonefs_dir_inode_operations;
697 	inode->i_fop = &zonefs_dir_operations;
698 
699 	unlock_new_inode(inode);
700 
701 	return inode;
702 }
703 
704 
705 static struct inode *zonefs_get_dir_inode(struct inode *dir,
706 					  struct dentry *dentry)
707 {
708 	struct super_block *sb = dir->i_sb;
709 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
710 	const char *name = dentry->d_name.name;
711 	enum zonefs_ztype ztype;
712 
713 	/*
714 	 * We only need to check for the "seq" directory and
715 	 * the "cnv" directory if we have conventional zones.
716 	 */
717 	if (dentry->d_name.len != 3)
718 		return ERR_PTR(-ENOENT);
719 
720 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
721 		if (sbi->s_zgroup[ztype].g_nr_zones &&
722 		    memcmp(name, zonefs_zgroup_name(ztype), 3) == 0)
723 			break;
724 	}
725 	if (ztype == ZONEFS_ZTYPE_MAX)
726 		return ERR_PTR(-ENOENT);
727 
728 	return zonefs_get_zgroup_inode(sb, ztype);
729 }
730 
731 static struct dentry *zonefs_lookup(struct inode *dir, struct dentry *dentry,
732 				    unsigned int flags)
733 {
734 	struct inode *inode;
735 
736 	if (dentry->d_name.len > ZONEFS_NAME_MAX)
737 		return ERR_PTR(-ENAMETOOLONG);
738 
739 	if (dir == d_inode(dir->i_sb->s_root))
740 		inode = zonefs_get_dir_inode(dir, dentry);
741 	else
742 		inode = zonefs_get_file_inode(dir, dentry);
743 
744 	return d_splice_alias(inode, dentry);
745 }
746 
747 static int zonefs_readdir_root(struct file *file, struct dir_context *ctx)
748 {
749 	struct inode *inode = file_inode(file);
750 	struct super_block *sb = inode->i_sb;
751 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
752 	enum zonefs_ztype ztype = ZONEFS_ZTYPE_CNV;
753 	ino_t base_ino = bdev_nr_zones(sb->s_bdev) + 1;
754 
755 	if (ctx->pos >= inode->i_size)
756 		return 0;
757 
758 	if (!dir_emit_dots(file, ctx))
759 		return 0;
760 
761 	if (ctx->pos == 2) {
762 		if (!sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones)
763 			ztype = ZONEFS_ZTYPE_SEQ;
764 
765 		if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
766 			      base_ino + ztype, DT_DIR))
767 			return 0;
768 		ctx->pos++;
769 	}
770 
771 	if (ctx->pos == 3 && ztype != ZONEFS_ZTYPE_SEQ) {
772 		ztype = ZONEFS_ZTYPE_SEQ;
773 		if (!dir_emit(ctx, zonefs_zgroup_name(ztype), 3,
774 			      base_ino + ztype, DT_DIR))
775 			return 0;
776 		ctx->pos++;
777 	}
778 
779 	return 0;
780 }
781 
782 static int zonefs_readdir_zgroup(struct file *file,
783 				 struct dir_context *ctx)
784 {
785 	struct inode *inode = file_inode(file);
786 	struct zonefs_zone_group *zgroup = inode->i_private;
787 	struct super_block *sb = inode->i_sb;
788 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
789 	struct zonefs_zone *z;
790 	int fname_len;
791 	char *fname;
792 	ino_t ino;
793 	int f;
794 
795 	/*
796 	 * The size of zone group directories is equal to the number
797 	 * of zone files in the group and does note include the "." and
798 	 * ".." entries. Hence the "+ 2" here.
799 	 */
800 	if (ctx->pos >= inode->i_size + 2)
801 		return 0;
802 
803 	if (!dir_emit_dots(file, ctx))
804 		return 0;
805 
806 	fname = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
807 	if (!fname)
808 		return -ENOMEM;
809 
810 	for (f = ctx->pos - 2; f < zgroup->g_nr_zones; f++) {
811 		z = &zgroup->g_zones[f];
812 		ino = z->z_sector >> sbi->s_zone_sectors_shift;
813 		fname_len = snprintf(fname, ZONEFS_NAME_MAX - 1, "%u", f);
814 		if (!dir_emit(ctx, fname, fname_len, ino, DT_REG))
815 			break;
816 		ctx->pos++;
817 	}
818 
819 	kfree(fname);
820 
821 	return 0;
822 }
823 
824 static int zonefs_readdir(struct file *file, struct dir_context *ctx)
825 {
826 	struct inode *inode = file_inode(file);
827 
828 	if (inode == d_inode(inode->i_sb->s_root))
829 		return zonefs_readdir_root(file, ctx);
830 
831 	return zonefs_readdir_zgroup(file, ctx);
832 }
833 
834 const struct inode_operations zonefs_dir_inode_operations = {
835 	.lookup		= zonefs_lookup,
836 	.setattr	= zonefs_inode_setattr,
837 };
838 
839 const struct file_operations zonefs_dir_operations = {
840 	.llseek		= generic_file_llseek,
841 	.read		= generic_read_dir,
842 	.iterate_shared	= zonefs_readdir,
843 };
844 
845 struct zonefs_zone_data {
846 	struct super_block	*sb;
847 	unsigned int		nr_zones[ZONEFS_ZTYPE_MAX];
848 	sector_t		cnv_zone_start;
849 	struct blk_zone		*zones;
850 };
851 
852 static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
853 				   void *data)
854 {
855 	struct zonefs_zone_data *zd = data;
856 	struct super_block *sb = zd->sb;
857 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
858 
859 	/*
860 	 * We do not care about the first zone: it contains the super block
861 	 * and not exposed as a file.
862 	 */
863 	if (!idx)
864 		return 0;
865 
866 	/*
867 	 * Count the number of zones that will be exposed as files.
868 	 * For sequential zones, we always have as many files as zones.
869 	 * FOr conventional zones, the number of files depends on if we have
870 	 * conventional zones aggregation enabled.
871 	 */
872 	switch (zone->type) {
873 	case BLK_ZONE_TYPE_CONVENTIONAL:
874 		if (sbi->s_features & ZONEFS_F_AGGRCNV) {
875 			/* One file per set of contiguous conventional zones */
876 			if (!(sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) ||
877 			    zone->start != zd->cnv_zone_start)
878 				sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
879 			zd->cnv_zone_start = zone->start + zone->len;
880 		} else {
881 			/* One file per zone */
882 			sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
883 		}
884 		break;
885 	case BLK_ZONE_TYPE_SEQWRITE_REQ:
886 	case BLK_ZONE_TYPE_SEQWRITE_PREF:
887 		sbi->s_zgroup[ZONEFS_ZTYPE_SEQ].g_nr_zones++;
888 		break;
889 	default:
890 		zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
891 			   zone->type);
892 		return -EIO;
893 	}
894 
895 	memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
896 
897 	return 0;
898 }
899 
900 static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
901 {
902 	struct block_device *bdev = zd->sb->s_bdev;
903 	int ret;
904 
905 	zd->zones = kvcalloc(bdev_nr_zones(bdev), sizeof(struct blk_zone),
906 			     GFP_KERNEL);
907 	if (!zd->zones)
908 		return -ENOMEM;
909 
910 	/* Get zones information from the device */
911 	ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
912 				  zonefs_get_zone_info_cb, zd);
913 	if (ret < 0) {
914 		zonefs_err(zd->sb, "Zone report failed %d\n", ret);
915 		return ret;
916 	}
917 
918 	if (ret != bdev_nr_zones(bdev)) {
919 		zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
920 			   ret, bdev_nr_zones(bdev));
921 		return -EIO;
922 	}
923 
924 	return 0;
925 }
926 
927 static inline void zonefs_free_zone_info(struct zonefs_zone_data *zd)
928 {
929 	kvfree(zd->zones);
930 }
931 
932 /*
933  * Create a zone group and populate it with zone files.
934  */
935 static int zonefs_init_zgroup(struct super_block *sb,
936 			      struct zonefs_zone_data *zd,
937 			      enum zonefs_ztype ztype)
938 {
939 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
940 	struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype];
941 	struct blk_zone *zone, *next, *end;
942 	struct zonefs_zone *z;
943 	unsigned int n = 0;
944 	int ret;
945 
946 	/* Allocate the zone group. If it is empty, we have nothing to do. */
947 	if (!zgroup->g_nr_zones)
948 		return 0;
949 
950 	zgroup->g_zones = kvcalloc(zgroup->g_nr_zones,
951 				   sizeof(struct zonefs_zone), GFP_KERNEL);
952 	if (!zgroup->g_zones)
953 		return -ENOMEM;
954 
955 	/*
956 	 * Initialize the zone groups using the device zone information.
957 	 * We always skip the first zone as it contains the super block
958 	 * and is not use to back a file.
959 	 */
960 	end = zd->zones + bdev_nr_zones(sb->s_bdev);
961 	for (zone = &zd->zones[1]; zone < end; zone = next) {
962 
963 		next = zone + 1;
964 		if (zonefs_zone_type(zone) != ztype)
965 			continue;
966 
967 		if (WARN_ON_ONCE(n >= zgroup->g_nr_zones))
968 			return -EINVAL;
969 
970 		/*
971 		 * For conventional zones, contiguous zones can be aggregated
972 		 * together to form larger files. Note that this overwrites the
973 		 * length of the first zone of the set of contiguous zones
974 		 * aggregated together. If one offline or read-only zone is
975 		 * found, assume that all zones aggregated have the same
976 		 * condition.
977 		 */
978 		if (ztype == ZONEFS_ZTYPE_CNV &&
979 		    (sbi->s_features & ZONEFS_F_AGGRCNV)) {
980 			for (; next < end; next++) {
981 				if (zonefs_zone_type(next) != ztype)
982 					break;
983 				zone->len += next->len;
984 				zone->capacity += next->capacity;
985 				if (next->cond == BLK_ZONE_COND_READONLY &&
986 				    zone->cond != BLK_ZONE_COND_OFFLINE)
987 					zone->cond = BLK_ZONE_COND_READONLY;
988 				else if (next->cond == BLK_ZONE_COND_OFFLINE)
989 					zone->cond = BLK_ZONE_COND_OFFLINE;
990 			}
991 		}
992 
993 		z = &zgroup->g_zones[n];
994 		if (ztype == ZONEFS_ZTYPE_CNV)
995 			z->z_flags |= ZONEFS_ZONE_CNV;
996 		z->z_sector = zone->start;
997 		z->z_size = zone->len << SECTOR_SHIFT;
998 		if (z->z_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT &&
999 		    !(sbi->s_features & ZONEFS_F_AGGRCNV)) {
1000 			zonefs_err(sb,
1001 				"Invalid zone size %llu (device zone sectors %llu)\n",
1002 				z->z_size,
1003 				bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT);
1004 			return -EINVAL;
1005 		}
1006 
1007 		z->z_capacity = min_t(loff_t, MAX_LFS_FILESIZE,
1008 				      zone->capacity << SECTOR_SHIFT);
1009 		z->z_wpoffset = zonefs_check_zone_condition(sb, z, zone);
1010 
1011 		z->z_mode = S_IFREG | sbi->s_perm;
1012 		z->z_uid = sbi->s_uid;
1013 		z->z_gid = sbi->s_gid;
1014 
1015 		/*
1016 		 * Let zonefs_inode_update_mode() know that we will need
1017 		 * special initialization of the inode mode the first time
1018 		 * it is accessed.
1019 		 */
1020 		z->z_flags |= ZONEFS_ZONE_INIT_MODE;
1021 
1022 		sb->s_maxbytes = max(z->z_capacity, sb->s_maxbytes);
1023 		sbi->s_blocks += z->z_capacity >> sb->s_blocksize_bits;
1024 		sbi->s_used_blocks += z->z_wpoffset >> sb->s_blocksize_bits;
1025 
1026 		/*
1027 		 * For sequential zones, make sure that any open zone is closed
1028 		 * first to ensure that the initial number of open zones is 0,
1029 		 * in sync with the open zone accounting done when the mount
1030 		 * option ZONEFS_MNTOPT_EXPLICIT_OPEN is used.
1031 		 */
1032 		if (ztype == ZONEFS_ZTYPE_SEQ &&
1033 		    (zone->cond == BLK_ZONE_COND_IMP_OPEN ||
1034 		     zone->cond == BLK_ZONE_COND_EXP_OPEN)) {
1035 			ret = zonefs_zone_mgmt(sb, z, REQ_OP_ZONE_CLOSE);
1036 			if (ret)
1037 				return ret;
1038 		}
1039 
1040 		zonefs_account_active(sb, z);
1041 
1042 		n++;
1043 	}
1044 
1045 	if (WARN_ON_ONCE(n != zgroup->g_nr_zones))
1046 		return -EINVAL;
1047 
1048 	zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1049 		    zonefs_zgroup_name(ztype),
1050 		    zgroup->g_nr_zones,
1051 		    str_plural(zgroup->g_nr_zones));
1052 
1053 	return 0;
1054 }
1055 
1056 static void zonefs_free_zgroups(struct super_block *sb)
1057 {
1058 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1059 	enum zonefs_ztype ztype;
1060 
1061 	if (!sbi)
1062 		return;
1063 
1064 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1065 		kvfree(sbi->s_zgroup[ztype].g_zones);
1066 		sbi->s_zgroup[ztype].g_zones = NULL;
1067 	}
1068 }
1069 
1070 /*
1071  * Create a zone group and populate it with zone files.
1072  */
1073 static int zonefs_init_zgroups(struct super_block *sb)
1074 {
1075 	struct zonefs_zone_data zd;
1076 	enum zonefs_ztype ztype;
1077 	int ret;
1078 
1079 	/* First get the device zone information */
1080 	memset(&zd, 0, sizeof(struct zonefs_zone_data));
1081 	zd.sb = sb;
1082 	ret = zonefs_get_zone_info(&zd);
1083 	if (ret)
1084 		goto cleanup;
1085 
1086 	/* Allocate and initialize the zone groups */
1087 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1088 		ret = zonefs_init_zgroup(sb, &zd, ztype);
1089 		if (ret) {
1090 			zonefs_info(sb,
1091 				    "Zone group \"%s\" initialization failed\n",
1092 				    zonefs_zgroup_name(ztype));
1093 			break;
1094 		}
1095 	}
1096 
1097 cleanup:
1098 	zonefs_free_zone_info(&zd);
1099 	if (ret)
1100 		zonefs_free_zgroups(sb);
1101 
1102 	return ret;
1103 }
1104 
1105 /*
1106  * Read super block information from the device.
1107  */
1108 static int zonefs_read_super(struct super_block *sb)
1109 {
1110 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1111 	struct zonefs_super *super;
1112 	u32 crc, stored_crc;
1113 	struct page *page;
1114 	struct bio_vec bio_vec;
1115 	struct bio bio;
1116 	int ret;
1117 
1118 	page = alloc_page(GFP_KERNEL);
1119 	if (!page)
1120 		return -ENOMEM;
1121 
1122 	bio_init(&bio, sb->s_bdev, &bio_vec, 1, REQ_OP_READ);
1123 	bio.bi_iter.bi_sector = 0;
1124 	__bio_add_page(&bio, page, PAGE_SIZE, 0);
1125 
1126 	ret = submit_bio_wait(&bio);
1127 	if (ret)
1128 		goto free_page;
1129 
1130 	super = page_address(page);
1131 
1132 	ret = -EINVAL;
1133 	if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1134 		goto free_page;
1135 
1136 	stored_crc = le32_to_cpu(super->s_crc);
1137 	super->s_crc = 0;
1138 	crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1139 	if (crc != stored_crc) {
1140 		zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1141 			   crc, stored_crc);
1142 		goto free_page;
1143 	}
1144 
1145 	sbi->s_features = le64_to_cpu(super->s_features);
1146 	if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1147 		zonefs_err(sb, "Unknown features set 0x%llx\n",
1148 			   sbi->s_features);
1149 		goto free_page;
1150 	}
1151 
1152 	if (sbi->s_features & ZONEFS_F_UID) {
1153 		sbi->s_uid = make_kuid(current_user_ns(),
1154 				       le32_to_cpu(super->s_uid));
1155 		if (!uid_valid(sbi->s_uid)) {
1156 			zonefs_err(sb, "Invalid UID feature\n");
1157 			goto free_page;
1158 		}
1159 	}
1160 
1161 	if (sbi->s_features & ZONEFS_F_GID) {
1162 		sbi->s_gid = make_kgid(current_user_ns(),
1163 				       le32_to_cpu(super->s_gid));
1164 		if (!gid_valid(sbi->s_gid)) {
1165 			zonefs_err(sb, "Invalid GID feature\n");
1166 			goto free_page;
1167 		}
1168 	}
1169 
1170 	if (sbi->s_features & ZONEFS_F_PERM)
1171 		sbi->s_perm = le32_to_cpu(super->s_perm);
1172 
1173 	if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1174 		zonefs_err(sb, "Reserved area is being used\n");
1175 		goto free_page;
1176 	}
1177 
1178 	import_uuid(&sbi->s_uuid, super->s_uuid);
1179 	ret = 0;
1180 
1181 free_page:
1182 	__free_page(page);
1183 
1184 	return ret;
1185 }
1186 
1187 static const struct super_operations zonefs_sops = {
1188 	.alloc_inode	= zonefs_alloc_inode,
1189 	.free_inode	= zonefs_free_inode,
1190 	.statfs		= zonefs_statfs,
1191 	.show_options	= zonefs_show_options,
1192 };
1193 
1194 static int zonefs_get_zgroup_inodes(struct super_block *sb)
1195 {
1196 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1197 	struct inode *dir_inode;
1198 	enum zonefs_ztype ztype;
1199 
1200 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1201 		if (!sbi->s_zgroup[ztype].g_nr_zones)
1202 			continue;
1203 
1204 		dir_inode = zonefs_get_zgroup_inode(sb, ztype);
1205 		if (IS_ERR(dir_inode))
1206 			return PTR_ERR(dir_inode);
1207 
1208 		sbi->s_zgroup[ztype].g_inode = dir_inode;
1209 	}
1210 
1211 	return 0;
1212 }
1213 
1214 static void zonefs_release_zgroup_inodes(struct super_block *sb)
1215 {
1216 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1217 	enum zonefs_ztype ztype;
1218 
1219 	if (!sbi)
1220 		return;
1221 
1222 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1223 		if (sbi->s_zgroup[ztype].g_inode) {
1224 			iput(sbi->s_zgroup[ztype].g_inode);
1225 			sbi->s_zgroup[ztype].g_inode = NULL;
1226 		}
1227 	}
1228 }
1229 
1230 /*
1231  * Check that the device is zoned. If it is, get the list of zones and create
1232  * sub-directories and files according to the device zone configuration and
1233  * format options.
1234  */
1235 static int zonefs_fill_super(struct super_block *sb, struct fs_context *fc)
1236 {
1237 	struct zonefs_sb_info *sbi;
1238 	struct zonefs_context *ctx = fc->fs_private;
1239 	struct inode *inode;
1240 	enum zonefs_ztype ztype;
1241 	int ret;
1242 
1243 	if (!bdev_is_zoned(sb->s_bdev)) {
1244 		zonefs_err(sb, "Not a zoned block device\n");
1245 		return -EINVAL;
1246 	}
1247 
1248 	/*
1249 	 * Initialize super block information: the maximum file size is updated
1250 	 * when the zone files are created so that the format option
1251 	 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1252 	 * beyond the zone size is taken into account.
1253 	 */
1254 	sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1255 	if (!sbi)
1256 		return -ENOMEM;
1257 
1258 	spin_lock_init(&sbi->s_lock);
1259 	sb->s_fs_info = sbi;
1260 	sb->s_magic = ZONEFS_MAGIC;
1261 	sb->s_maxbytes = 0;
1262 	sb->s_op = &zonefs_sops;
1263 	sb->s_time_gran	= 1;
1264 
1265 	/*
1266 	 * The block size is set to the device zone write granularity to ensure
1267 	 * that write operations are always aligned according to the device
1268 	 * interface constraints.
1269 	 */
1270 	sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev));
1271 	sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1272 	sbi->s_uid = GLOBAL_ROOT_UID;
1273 	sbi->s_gid = GLOBAL_ROOT_GID;
1274 	sbi->s_perm = 0640;
1275 	sbi->s_mount_opts = ctx->s_mount_opts;
1276 
1277 	atomic_set(&sbi->s_wro_seq_files, 0);
1278 	sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev);
1279 	atomic_set(&sbi->s_active_seq_files, 0);
1280 	sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev);
1281 
1282 	ret = zonefs_read_super(sb);
1283 	if (ret)
1284 		return ret;
1285 
1286 	zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev));
1287 
1288 	if (!sbi->s_max_wro_seq_files &&
1289 	    !sbi->s_max_active_seq_files &&
1290 	    sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1291 		zonefs_info(sb,
1292 			"No open and active zone limits. Ignoring explicit_open mount option\n");
1293 		sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
1294 	}
1295 
1296 	/* Initialize the zone groups */
1297 	ret = zonefs_init_zgroups(sb);
1298 	if (ret)
1299 		goto cleanup;
1300 
1301 	/* Create the root directory inode */
1302 	ret = -ENOMEM;
1303 	inode = new_inode(sb);
1304 	if (!inode)
1305 		goto cleanup;
1306 
1307 	inode->i_ino = bdev_nr_zones(sb->s_bdev);
1308 	inode->i_mode = S_IFDIR | 0555;
1309 	simple_inode_init_ts(inode);
1310 	inode->i_op = &zonefs_dir_inode_operations;
1311 	inode->i_fop = &zonefs_dir_operations;
1312 	inode->i_size = 2;
1313 	set_nlink(inode, 2);
1314 	for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
1315 		if (sbi->s_zgroup[ztype].g_nr_zones) {
1316 			inc_nlink(inode);
1317 			inode->i_size++;
1318 		}
1319 	}
1320 
1321 	sb->s_root = d_make_root(inode);
1322 	if (!sb->s_root)
1323 		goto cleanup;
1324 
1325 	/*
1326 	 * Take a reference on the zone groups directory inodes
1327 	 * to keep them in the inode cache.
1328 	 */
1329 	ret = zonefs_get_zgroup_inodes(sb);
1330 	if (ret)
1331 		goto cleanup;
1332 
1333 	ret = zonefs_sysfs_register(sb);
1334 	if (ret)
1335 		goto cleanup;
1336 
1337 	return 0;
1338 
1339 cleanup:
1340 	zonefs_release_zgroup_inodes(sb);
1341 	zonefs_free_zgroups(sb);
1342 
1343 	return ret;
1344 }
1345 
1346 static void zonefs_kill_super(struct super_block *sb)
1347 {
1348 	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1349 
1350 	/* Release the reference on the zone group directory inodes */
1351 	zonefs_release_zgroup_inodes(sb);
1352 
1353 	kill_block_super(sb);
1354 
1355 	zonefs_sysfs_unregister(sb);
1356 	zonefs_free_zgroups(sb);
1357 	kfree(sbi);
1358 }
1359 
1360 static void zonefs_free_fc(struct fs_context *fc)
1361 {
1362 	struct zonefs_context *ctx = fc->fs_private;
1363 
1364 	kfree(ctx);
1365 }
1366 
1367 static int zonefs_get_tree(struct fs_context *fc)
1368 {
1369 	return get_tree_bdev(fc, zonefs_fill_super);
1370 }
1371 
1372 static int zonefs_reconfigure(struct fs_context *fc)
1373 {
1374 	struct zonefs_context *ctx = fc->fs_private;
1375 	struct super_block *sb = fc->root->d_sb;
1376 	struct zonefs_sb_info *sbi = sb->s_fs_info;
1377 
1378 	sync_filesystem(fc->root->d_sb);
1379 	/* Copy new options from ctx into sbi. */
1380 	sbi->s_mount_opts = ctx->s_mount_opts;
1381 
1382 	return 0;
1383 }
1384 
1385 static const struct fs_context_operations zonefs_context_ops = {
1386 	.parse_param    = zonefs_parse_param,
1387 	.get_tree       = zonefs_get_tree,
1388 	.reconfigure	= zonefs_reconfigure,
1389 	.free           = zonefs_free_fc,
1390 };
1391 
1392 /*
1393  * Set up the filesystem mount context.
1394  */
1395 static int zonefs_init_fs_context(struct fs_context *fc)
1396 {
1397 	struct zonefs_context *ctx;
1398 
1399 	ctx = kzalloc(sizeof(struct zonefs_context), GFP_KERNEL);
1400 	if (!ctx)
1401 		return -ENOMEM;
1402 	ctx->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1403 	fc->ops = &zonefs_context_ops;
1404 	fc->fs_private = ctx;
1405 
1406 	return 0;
1407 }
1408 
1409 /*
1410  * File system definition and registration.
1411  */
1412 static struct file_system_type zonefs_type = {
1413 	.owner			= THIS_MODULE,
1414 	.name			= "zonefs",
1415 	.kill_sb		= zonefs_kill_super,
1416 	.fs_flags		= FS_REQUIRES_DEV,
1417 	.init_fs_context	= zonefs_init_fs_context,
1418 	.parameters		= zonefs_param_spec,
1419 };
1420 
1421 static int __init zonefs_init_inodecache(void)
1422 {
1423 	zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1424 			sizeof(struct zonefs_inode_info), 0,
1425 			SLAB_RECLAIM_ACCOUNT | SLAB_ACCOUNT,
1426 			NULL);
1427 	if (zonefs_inode_cachep == NULL)
1428 		return -ENOMEM;
1429 	return 0;
1430 }
1431 
1432 static void zonefs_destroy_inodecache(void)
1433 {
1434 	/*
1435 	 * Make sure all delayed rcu free inodes are flushed before we
1436 	 * destroy the inode cache.
1437 	 */
1438 	rcu_barrier();
1439 	kmem_cache_destroy(zonefs_inode_cachep);
1440 }
1441 
1442 static int __init zonefs_init(void)
1443 {
1444 	int ret;
1445 
1446 	BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1447 
1448 	ret = zonefs_init_inodecache();
1449 	if (ret)
1450 		return ret;
1451 
1452 	ret = zonefs_sysfs_init();
1453 	if (ret)
1454 		goto destroy_inodecache;
1455 
1456 	ret = register_filesystem(&zonefs_type);
1457 	if (ret)
1458 		goto sysfs_exit;
1459 
1460 	return 0;
1461 
1462 sysfs_exit:
1463 	zonefs_sysfs_exit();
1464 destroy_inodecache:
1465 	zonefs_destroy_inodecache();
1466 
1467 	return ret;
1468 }
1469 
1470 static void __exit zonefs_exit(void)
1471 {
1472 	unregister_filesystem(&zonefs_type);
1473 	zonefs_sysfs_exit();
1474 	zonefs_destroy_inodecache();
1475 }
1476 
1477 MODULE_AUTHOR("Damien Le Moal");
1478 MODULE_DESCRIPTION("Zone file system for zoned block devices");
1479 MODULE_LICENSE("GPL");
1480 MODULE_ALIAS_FS("zonefs");
1481 module_init(zonefs_init);
1482 module_exit(zonefs_exit);
1483