1 // SPDX-License-Identifier: GPL-2.0
3 * Simple file system for zoned block devices exposing zones as files.
5 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
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>
27 #define CREATE_TRACE_POINTS
31 * Get the name of a zone group directory.
33 static const char *zonefs_zgroup_name(enum zonefs_ztype ztype)
36 case ZONEFS_ZTYPE_CNV:
38 case ZONEFS_ZTYPE_SEQ:
47 * Manage the active zone count.
49 static void zonefs_account_active(struct super_block *sb,
50 struct zonefs_zone *z)
52 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
54 if (zonefs_zone_is_cnv(z))
58 * For zones that transitioned to the offline or readonly condition,
59 * we only need to clear the active state.
61 if (z->z_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY))
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.
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);
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);
86 * Manage the active zone count. Called with zi->i_truncate_mutex held.
88 void zonefs_inode_account_active(struct inode *inode)
90 lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
92 return zonefs_account_active(inode->i_sb, zonefs_inode_zone(inode));
96 * Execute a zone management operation.
98 static int zonefs_zone_mgmt(struct super_block *sb,
99 struct zonefs_zone *z, enum req_op op)
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
111 if (op == REQ_OP_ZONE_CLOSE && !z->z_wpoffset)
112 op = REQ_OP_ZONE_RESET;
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);
119 "Zone management operation %s at %llu failed %d\n",
120 blk_op_str(op), z->z_sector, ret);
127 int zonefs_inode_zone_mgmt(struct inode *inode, enum req_op op)
129 lockdep_assert_held(&ZONEFS_I(inode)->i_truncate_mutex);
131 return zonefs_zone_mgmt(inode->i_sb, zonefs_inode_zone(inode), op);
134 void zonefs_i_size_write(struct inode *inode, loff_t isize)
136 struct zonefs_zone *z = zonefs_inode_zone(inode);
138 i_size_write(inode, isize);
141 * A full zone is no longer open/active and does not need
144 if (isize >= z->z_capacity) {
145 struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
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);
153 void zonefs_update_stats(struct inode *inode, loff_t new_isize)
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);
160 if (new_isize == old_isize)
163 spin_lock(&sbi->s_lock);
166 * This may be called for an update after an IO error.
167 * So beware of the values seen.
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;
174 sbi->s_used_blocks = 0;
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;
182 spin_unlock(&sbi->s_lock);
186 * Check a zone condition. Return the amount of written (and still readable)
189 static loff_t zonefs_check_zone_condition(struct super_block *sb,
190 struct zonefs_zone *z,
191 struct blk_zone *zone)
193 switch (zone->cond) {
194 case BLK_ZONE_COND_OFFLINE:
195 zonefs_warn(sb, "Zone %llu: offline zone\n",
197 z->z_flags |= ZONEFS_ZONE_OFFLINE;
199 case BLK_ZONE_COND_READONLY:
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
208 zonefs_warn(sb, "Zone %llu: read-only zone\n",
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;
218 if (zonefs_zone_is_cnv(z))
219 return z->z_capacity;
220 return (zone->wp - zone->start) << SECTOR_SHIFT;
225 * Check a zone condition and adjust its inode access permissions for
226 * offline and readonly zones.
228 static void zonefs_inode_update_mode(struct inode *inode)
230 struct zonefs_zone *z = zonefs_inode_zone(inode);
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;
242 inode->i_mode &= ~0222;
245 z->z_flags &= ~ZONEFS_ZONE_INIT_MODE;
248 struct zonefs_ioerr_data {
253 static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
256 struct zonefs_ioerr_data *err = data;
257 struct inode *inode = err->inode;
258 struct zonefs_zone *z = zonefs_inode_zone(inode);
259 struct super_block *sb = inode->i_sb;
260 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
261 loff_t isize, data_size;
264 * Check the zone condition: if the zone is not "bad" (offline or
265 * read-only), read errors are simply signaled to the IO issuer as long
266 * as there is no inconsistency between the inode size and the amount of
267 * data writen in the zone (data_size).
269 data_size = zonefs_check_zone_condition(sb, z, zone);
270 isize = i_size_read(inode);
271 if (!(z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)) &&
272 !err->write && isize == data_size)
276 * At this point, we detected either a bad zone or an inconsistency
277 * between the inode size and the amount of data written in the zone.
278 * For the latter case, the cause may be a write IO error or an external
279 * action on the device. Two error patterns exist:
280 * 1) The inode size is lower than the amount of data in the zone:
281 * a write operation partially failed and data was writen at the end
282 * of the file. This can happen in the case of a large direct IO
283 * needing several BIOs and/or write requests to be processed.
284 * 2) The inode size is larger than the amount of data in the zone:
285 * this can happen with a deferred write error with the use of the
286 * device side write cache after getting successful write IO
287 * completions. Other possibilities are (a) an external corruption,
288 * e.g. an application reset the zone directly, or (b) the device
289 * has a serious problem (e.g. firmware bug).
291 * In all cases, warn about inode size inconsistency and handle the
292 * IO error according to the zone condition and to the mount options.
294 if (zonefs_zone_is_seq(z) && isize != data_size)
296 "inode %lu: invalid size %lld (should be %lld)\n",
297 inode->i_ino, isize, data_size);
300 * First handle bad zones signaled by hardware. The mount options
301 * errors=zone-ro and errors=zone-offline result in changing the
302 * zone condition to read-only and offline respectively, as if the
303 * condition was signaled by the hardware.
305 if ((z->z_flags & ZONEFS_ZONE_OFFLINE) ||
306 (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)) {
307 zonefs_warn(sb, "inode %lu: read/write access disabled\n",
309 if (!(z->z_flags & ZONEFS_ZONE_OFFLINE))
310 z->z_flags |= ZONEFS_ZONE_OFFLINE;
311 zonefs_inode_update_mode(inode);
313 } else if ((z->z_flags & ZONEFS_ZONE_READONLY) ||
314 (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)) {
315 zonefs_warn(sb, "inode %lu: write access disabled\n",
317 if (!(z->z_flags & ZONEFS_ZONE_READONLY))
318 z->z_flags |= ZONEFS_ZONE_READONLY;
319 zonefs_inode_update_mode(inode);
321 } else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO &&
323 /* Do not expose garbage data */
328 * If the filesystem is mounted with the explicit-open mount option, we
329 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
330 * the read-only or offline condition, to avoid attempting an explicit
331 * close of the zone when the inode file is closed.
333 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
334 (z->z_flags & (ZONEFS_ZONE_READONLY | ZONEFS_ZONE_OFFLINE)))
335 z->z_flags &= ~ZONEFS_ZONE_OPEN;
338 * If error=remount-ro was specified, any error result in remounting
339 * the volume as read-only.
341 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
342 zonefs_warn(sb, "remounting filesystem read-only\n");
343 sb->s_flags |= SB_RDONLY;
347 * Update block usage stats and the inode size to prevent access to
350 zonefs_update_stats(inode, data_size);
351 zonefs_i_size_write(inode, data_size);
352 z->z_wpoffset = data_size;
353 zonefs_inode_account_active(inode);
359 * When an file IO error occurs, check the file zone to see if there is a change
360 * in the zone condition (e.g. offline or read-only). For a failed write to a
361 * sequential zone, the zone write pointer position must also be checked to
362 * eventually correct the file size and zonefs inode write pointer offset
363 * (which can be out of sync with the drive due to partial write failures).
365 void __zonefs_io_error(struct inode *inode, bool write)
367 struct zonefs_zone *z = zonefs_inode_zone(inode);
368 struct super_block *sb = inode->i_sb;
369 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
370 unsigned int noio_flag;
371 unsigned int nr_zones = 1;
372 struct zonefs_ioerr_data err = {
379 * The only files that have more than one zone are conventional zone
380 * files with aggregated conventional zones, for which the inode zone
381 * size is always larger than the device zone size.
383 if (z->z_size > bdev_zone_sectors(sb->s_bdev))
384 nr_zones = z->z_size >>
385 (sbi->s_zone_sectors_shift + SECTOR_SHIFT);
388 * Memory allocations in blkdev_report_zones() can trigger a memory
389 * reclaim which may in turn cause a recursion into zonefs as well as
390 * struct request allocations for the same device. The former case may
391 * end up in a deadlock on the inode truncate mutex, while the latter
392 * may prevent IO forward progress. Executing the report zones under
393 * the GFP_NOIO context avoids both problems.
395 noio_flag = memalloc_noio_save();
396 ret = blkdev_report_zones(sb->s_bdev, z->z_sector, nr_zones,
397 zonefs_io_error_cb, &err);
399 zonefs_err(sb, "Get inode %lu zone information failed %d\n",
401 memalloc_noio_restore(noio_flag);
404 static struct kmem_cache *zonefs_inode_cachep;
406 static struct inode *zonefs_alloc_inode(struct super_block *sb)
408 struct zonefs_inode_info *zi;
410 zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL);
414 inode_init_once(&zi->i_vnode);
415 mutex_init(&zi->i_truncate_mutex);
421 static void zonefs_free_inode(struct inode *inode)
423 kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
429 static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
431 struct super_block *sb = dentry->d_sb;
432 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
435 buf->f_type = ZONEFS_MAGIC;
436 buf->f_bsize = sb->s_blocksize;
437 buf->f_namelen = ZONEFS_NAME_MAX;
439 spin_lock(&sbi->s_lock);
441 buf->f_blocks = sbi->s_blocks;
442 if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
445 buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
446 buf->f_bavail = buf->f_bfree;
448 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
449 if (sbi->s_zgroup[t].g_nr_zones)
450 buf->f_files += sbi->s_zgroup[t].g_nr_zones + 1;
454 spin_unlock(&sbi->s_lock);
456 buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b);
462 Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
463 Opt_explicit_open, Opt_err,
466 static const match_table_t tokens = {
467 { Opt_errors_ro, "errors=remount-ro"},
468 { Opt_errors_zro, "errors=zone-ro"},
469 { Opt_errors_zol, "errors=zone-offline"},
470 { Opt_errors_repair, "errors=repair"},
471 { Opt_explicit_open, "explicit-open" },
475 static int zonefs_parse_options(struct super_block *sb, char *options)
477 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
478 substring_t args[MAX_OPT_ARGS];
484 while ((p = strsep(&options, ",")) != NULL) {
490 token = match_token(p, tokens, args);
493 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
494 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
497 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
498 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
501 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
502 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
504 case Opt_errors_repair:
505 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
506 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
508 case Opt_explicit_open:
509 sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
519 static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
521 struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
523 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
524 seq_puts(seq, ",errors=remount-ro");
525 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
526 seq_puts(seq, ",errors=zone-ro");
527 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
528 seq_puts(seq, ",errors=zone-offline");
529 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
530 seq_puts(seq, ",errors=repair");
535 static int zonefs_remount(struct super_block *sb, int *flags, char *data)
539 return zonefs_parse_options(sb, data);
542 static int zonefs_inode_setattr(struct user_namespace *mnt_userns,
543 struct dentry *dentry, struct iattr *iattr)
545 struct inode *inode = d_inode(dentry);
548 if (unlikely(IS_IMMUTABLE(inode)))
551 ret = setattr_prepare(&init_user_ns, dentry, iattr);
556 * Since files and directories cannot be created nor deleted, do not
557 * allow setting any write attributes on the sub-directories grouping
558 * files by zone type.
560 if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
561 (iattr->ia_mode & 0222))
564 if (((iattr->ia_valid & ATTR_UID) &&
565 !uid_eq(iattr->ia_uid, inode->i_uid)) ||
566 ((iattr->ia_valid & ATTR_GID) &&
567 !gid_eq(iattr->ia_gid, inode->i_gid))) {
568 ret = dquot_transfer(mnt_userns, inode, iattr);
573 if (iattr->ia_valid & ATTR_SIZE) {
574 ret = zonefs_file_truncate(inode, iattr->ia_size);
579 setattr_copy(&init_user_ns, inode, iattr);
584 static const struct inode_operations zonefs_dir_inode_operations = {
585 .lookup = simple_lookup,
586 .setattr = zonefs_inode_setattr,
589 static void zonefs_init_dir_inode(struct inode *parent, struct inode *inode,
590 enum zonefs_ztype ztype)
592 struct super_block *sb = parent->i_sb;
594 inode->i_ino = bdev_nr_zones(sb->s_bdev) + ztype + 1;
595 inode_init_owner(&init_user_ns, inode, parent, S_IFDIR | 0555);
596 inode->i_op = &zonefs_dir_inode_operations;
597 inode->i_fop = &simple_dir_operations;
602 static const struct inode_operations zonefs_file_inode_operations = {
603 .setattr = zonefs_inode_setattr,
606 static void zonefs_init_file_inode(struct inode *inode,
607 struct zonefs_zone *z)
609 struct super_block *sb = inode->i_sb;
610 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
612 inode->i_private = z;
614 inode->i_ino = z->z_sector >> sbi->s_zone_sectors_shift;
615 inode->i_mode = S_IFREG | sbi->s_perm;
616 inode->i_uid = sbi->s_uid;
617 inode->i_gid = sbi->s_gid;
618 inode->i_size = z->z_wpoffset;
619 inode->i_blocks = z->z_capacity >> SECTOR_SHIFT;
621 inode->i_op = &zonefs_file_inode_operations;
622 inode->i_fop = &zonefs_file_operations;
623 inode->i_mapping->a_ops = &zonefs_file_aops;
625 /* Update the inode access rights depending on the zone condition */
626 z->z_flags |= ZONEFS_ZONE_INIT_MODE;
627 zonefs_inode_update_mode(inode);
630 static struct dentry *zonefs_create_inode(struct dentry *parent,
632 struct zonefs_zone *z,
633 enum zonefs_ztype ztype)
635 struct inode *dir = d_inode(parent);
636 struct dentry *dentry;
640 dentry = d_alloc_name(parent, name);
644 inode = new_inode(parent->d_sb);
648 inode->i_ctime = inode->i_mtime = inode->i_atime = dir->i_ctime;
650 zonefs_init_file_inode(inode, z);
652 zonefs_init_dir_inode(dir, inode, ztype);
654 d_add(dentry, inode);
665 struct zonefs_zone_data {
666 struct super_block *sb;
667 unsigned int nr_zones[ZONEFS_ZTYPE_MAX];
668 sector_t cnv_zone_start;
669 struct blk_zone *zones;
673 * Create the inodes for a zone group.
675 static int zonefs_create_zgroup_inodes(struct super_block *sb,
676 enum zonefs_ztype ztype)
678 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
679 struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype];
680 struct dentry *dir, *dent;
687 /* If the group is empty, there is nothing to do */
688 if (!zgroup->g_nr_zones)
691 file_name = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
695 dir = zonefs_create_inode(sb->s_root, zonefs_zgroup_name(ztype),
702 for (i = 0; i < zgroup->g_nr_zones; i++) {
703 /* Use the zone number within its group as the file name */
704 snprintf(file_name, ZONEFS_NAME_MAX - 1, "%u", i);
705 dent = zonefs_create_inode(dir, file_name,
706 &zgroup->g_zones[i], ztype);
719 static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
722 struct zonefs_zone_data *zd = data;
723 struct super_block *sb = zd->sb;
724 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
727 * We do not care about the first zone: it contains the super block
728 * and not exposed as a file.
734 * Count the number of zones that will be exposed as files.
735 * For sequential zones, we always have as many files as zones.
736 * FOr conventional zones, the number of files depends on if we have
737 * conventional zones aggregation enabled.
739 switch (zone->type) {
740 case BLK_ZONE_TYPE_CONVENTIONAL:
741 if (sbi->s_features & ZONEFS_F_AGGRCNV) {
742 /* One file per set of contiguous conventional zones */
743 if (!(sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones) ||
744 zone->start != zd->cnv_zone_start)
745 sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
746 zd->cnv_zone_start = zone->start + zone->len;
748 /* One file per zone */
749 sbi->s_zgroup[ZONEFS_ZTYPE_CNV].g_nr_zones++;
752 case BLK_ZONE_TYPE_SEQWRITE_REQ:
753 case BLK_ZONE_TYPE_SEQWRITE_PREF:
754 sbi->s_zgroup[ZONEFS_ZTYPE_SEQ].g_nr_zones++;
757 zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
762 memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
767 static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
769 struct block_device *bdev = zd->sb->s_bdev;
772 zd->zones = kvcalloc(bdev_nr_zones(bdev), sizeof(struct blk_zone),
777 /* Get zones information from the device */
778 ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
779 zonefs_get_zone_info_cb, zd);
781 zonefs_err(zd->sb, "Zone report failed %d\n", ret);
785 if (ret != bdev_nr_zones(bdev)) {
786 zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
787 ret, bdev_nr_zones(bdev));
794 static inline void zonefs_free_zone_info(struct zonefs_zone_data *zd)
800 * Create a zone group and populate it with zone files.
802 static int zonefs_init_zgroup(struct super_block *sb,
803 struct zonefs_zone_data *zd,
804 enum zonefs_ztype ztype)
806 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
807 struct zonefs_zone_group *zgroup = &sbi->s_zgroup[ztype];
808 struct blk_zone *zone, *next, *end;
809 struct zonefs_zone *z;
813 /* Allocate the zone group. If it is empty, we have nothing to do. */
814 if (!zgroup->g_nr_zones)
817 zgroup->g_zones = kvcalloc(zgroup->g_nr_zones,
818 sizeof(struct zonefs_zone), GFP_KERNEL);
819 if (!zgroup->g_zones)
823 * Initialize the zone groups using the device zone information.
824 * We always skip the first zone as it contains the super block
825 * and is not use to back a file.
827 end = zd->zones + bdev_nr_zones(sb->s_bdev);
828 for (zone = &zd->zones[1]; zone < end; zone = next) {
831 if (zonefs_zone_type(zone) != ztype)
834 if (WARN_ON_ONCE(n >= zgroup->g_nr_zones))
838 * For conventional zones, contiguous zones can be aggregated
839 * together to form larger files. Note that this overwrites the
840 * length of the first zone of the set of contiguous zones
841 * aggregated together. If one offline or read-only zone is
842 * found, assume that all zones aggregated have the same
845 if (ztype == ZONEFS_ZTYPE_CNV &&
846 (sbi->s_features & ZONEFS_F_AGGRCNV)) {
847 for (; next < end; next++) {
848 if (zonefs_zone_type(next) != ztype)
850 zone->len += next->len;
851 zone->capacity += next->capacity;
852 if (next->cond == BLK_ZONE_COND_READONLY &&
853 zone->cond != BLK_ZONE_COND_OFFLINE)
854 zone->cond = BLK_ZONE_COND_READONLY;
855 else if (next->cond == BLK_ZONE_COND_OFFLINE)
856 zone->cond = BLK_ZONE_COND_OFFLINE;
860 z = &zgroup->g_zones[n];
861 if (ztype == ZONEFS_ZTYPE_CNV)
862 z->z_flags |= ZONEFS_ZONE_CNV;
863 z->z_sector = zone->start;
864 z->z_size = zone->len << SECTOR_SHIFT;
865 if (z->z_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT &&
866 !(sbi->s_features & ZONEFS_F_AGGRCNV)) {
868 "Invalid zone size %llu (device zone sectors %llu)\n",
870 bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT);
874 z->z_capacity = min_t(loff_t, MAX_LFS_FILESIZE,
875 zone->capacity << SECTOR_SHIFT);
876 z->z_wpoffset = zonefs_check_zone_condition(sb, z, zone);
878 sb->s_maxbytes = max(z->z_capacity, sb->s_maxbytes);
879 sbi->s_blocks += z->z_capacity >> sb->s_blocksize_bits;
880 sbi->s_used_blocks += z->z_wpoffset >> sb->s_blocksize_bits;
883 * For sequential zones, make sure that any open zone is closed
884 * first to ensure that the initial number of open zones is 0,
885 * in sync with the open zone accounting done when the mount
886 * option ZONEFS_MNTOPT_EXPLICIT_OPEN is used.
888 if (ztype == ZONEFS_ZTYPE_SEQ &&
889 (zone->cond == BLK_ZONE_COND_IMP_OPEN ||
890 zone->cond == BLK_ZONE_COND_EXP_OPEN)) {
891 ret = zonefs_zone_mgmt(sb, z, REQ_OP_ZONE_CLOSE);
896 zonefs_account_active(sb, z);
901 if (WARN_ON_ONCE(n != zgroup->g_nr_zones))
904 zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
905 zonefs_zgroup_name(ztype),
907 zgroup->g_nr_zones > 1 ? "s" : "");
912 static void zonefs_free_zgroups(struct super_block *sb)
914 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
915 enum zonefs_ztype ztype;
920 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
921 kvfree(sbi->s_zgroup[ztype].g_zones);
922 sbi->s_zgroup[ztype].g_zones = NULL;
927 * Create a zone group and populate it with zone files.
929 static int zonefs_init_zgroups(struct super_block *sb)
931 struct zonefs_zone_data zd;
932 enum zonefs_ztype ztype;
935 /* First get the device zone information */
936 memset(&zd, 0, sizeof(struct zonefs_zone_data));
938 ret = zonefs_get_zone_info(&zd);
942 /* Allocate and initialize the zone groups */
943 for (ztype = 0; ztype < ZONEFS_ZTYPE_MAX; ztype++) {
944 ret = zonefs_init_zgroup(sb, &zd, ztype);
947 "Zone group \"%s\" initialization failed\n",
948 zonefs_zgroup_name(ztype));
954 zonefs_free_zone_info(&zd);
956 zonefs_free_zgroups(sb);
962 * Read super block information from the device.
964 static int zonefs_read_super(struct super_block *sb)
966 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
967 struct zonefs_super *super;
970 struct bio_vec bio_vec;
974 page = alloc_page(GFP_KERNEL);
978 bio_init(&bio, sb->s_bdev, &bio_vec, 1, REQ_OP_READ);
979 bio.bi_iter.bi_sector = 0;
980 bio_add_page(&bio, page, PAGE_SIZE, 0);
982 ret = submit_bio_wait(&bio);
986 super = page_address(page);
989 if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
992 stored_crc = le32_to_cpu(super->s_crc);
994 crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
995 if (crc != stored_crc) {
996 zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1001 sbi->s_features = le64_to_cpu(super->s_features);
1002 if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1003 zonefs_err(sb, "Unknown features set 0x%llx\n",
1008 if (sbi->s_features & ZONEFS_F_UID) {
1009 sbi->s_uid = make_kuid(current_user_ns(),
1010 le32_to_cpu(super->s_uid));
1011 if (!uid_valid(sbi->s_uid)) {
1012 zonefs_err(sb, "Invalid UID feature\n");
1017 if (sbi->s_features & ZONEFS_F_GID) {
1018 sbi->s_gid = make_kgid(current_user_ns(),
1019 le32_to_cpu(super->s_gid));
1020 if (!gid_valid(sbi->s_gid)) {
1021 zonefs_err(sb, "Invalid GID feature\n");
1026 if (sbi->s_features & ZONEFS_F_PERM)
1027 sbi->s_perm = le32_to_cpu(super->s_perm);
1029 if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1030 zonefs_err(sb, "Reserved area is being used\n");
1034 import_uuid(&sbi->s_uuid, super->s_uuid);
1043 static const struct super_operations zonefs_sops = {
1044 .alloc_inode = zonefs_alloc_inode,
1045 .free_inode = zonefs_free_inode,
1046 .statfs = zonefs_statfs,
1047 .remount_fs = zonefs_remount,
1048 .show_options = zonefs_show_options,
1052 * Check that the device is zoned. If it is, get the list of zones and create
1053 * sub-directories and files according to the device zone configuration and
1056 static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1058 struct zonefs_sb_info *sbi;
1059 struct inode *inode;
1060 enum zonefs_ztype t;
1063 if (!bdev_is_zoned(sb->s_bdev)) {
1064 zonefs_err(sb, "Not a zoned block device\n");
1069 * Initialize super block information: the maximum file size is updated
1070 * when the zone files are created so that the format option
1071 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1072 * beyond the zone size is taken into account.
1074 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1078 spin_lock_init(&sbi->s_lock);
1079 sb->s_fs_info = sbi;
1080 sb->s_magic = ZONEFS_MAGIC;
1082 sb->s_op = &zonefs_sops;
1083 sb->s_time_gran = 1;
1086 * The block size is set to the device zone write granularity to ensure
1087 * that write operations are always aligned according to the device
1088 * interface constraints.
1090 sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev));
1091 sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1092 sbi->s_uid = GLOBAL_ROOT_UID;
1093 sbi->s_gid = GLOBAL_ROOT_GID;
1095 sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1097 atomic_set(&sbi->s_wro_seq_files, 0);
1098 sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev);
1099 atomic_set(&sbi->s_active_seq_files, 0);
1100 sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev);
1102 ret = zonefs_read_super(sb);
1106 ret = zonefs_parse_options(sb, data);
1110 zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev));
1112 if (!sbi->s_max_wro_seq_files &&
1113 !sbi->s_max_active_seq_files &&
1114 sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1116 "No open and active zone limits. Ignoring explicit_open mount option\n");
1117 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
1120 /* Initialize the zone groups */
1121 ret = zonefs_init_zgroups(sb);
1125 /* Create root directory inode */
1127 inode = new_inode(sb);
1131 inode->i_ino = bdev_nr_zones(sb->s_bdev);
1132 inode->i_mode = S_IFDIR | 0555;
1133 inode->i_ctime = inode->i_mtime = inode->i_atime = current_time(inode);
1134 inode->i_op = &zonefs_dir_inode_operations;
1135 inode->i_fop = &simple_dir_operations;
1136 set_nlink(inode, 2);
1138 sb->s_root = d_make_root(inode);
1142 /* Create and populate files in zone groups directories */
1143 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
1144 ret = zonefs_create_zgroup_inodes(sb, t);
1149 ret = zonefs_sysfs_register(sb);
1156 zonefs_free_zgroups(sb);
1161 static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1162 int flags, const char *dev_name, void *data)
1164 return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1167 static void zonefs_kill_super(struct super_block *sb)
1169 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1172 d_genocide(sb->s_root);
1174 zonefs_sysfs_unregister(sb);
1175 zonefs_free_zgroups(sb);
1176 kill_block_super(sb);
1181 * File system definition and registration.
1183 static struct file_system_type zonefs_type = {
1184 .owner = THIS_MODULE,
1186 .mount = zonefs_mount,
1187 .kill_sb = zonefs_kill_super,
1188 .fs_flags = FS_REQUIRES_DEV,
1191 static int __init zonefs_init_inodecache(void)
1193 zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1194 sizeof(struct zonefs_inode_info), 0,
1195 (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1197 if (zonefs_inode_cachep == NULL)
1202 static void zonefs_destroy_inodecache(void)
1205 * Make sure all delayed rcu free inodes are flushed before we
1206 * destroy the inode cache.
1209 kmem_cache_destroy(zonefs_inode_cachep);
1212 static int __init zonefs_init(void)
1216 BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1218 ret = zonefs_init_inodecache();
1222 ret = zonefs_sysfs_init();
1224 goto destroy_inodecache;
1226 ret = register_filesystem(&zonefs_type);
1233 zonefs_sysfs_exit();
1235 zonefs_destroy_inodecache();
1240 static void __exit zonefs_exit(void)
1242 unregister_filesystem(&zonefs_type);
1243 zonefs_sysfs_exit();
1244 zonefs_destroy_inodecache();
1247 MODULE_AUTHOR("Damien Le Moal");
1248 MODULE_DESCRIPTION("Zone file system for zoned block devices");
1249 MODULE_LICENSE("GPL");
1250 MODULE_ALIAS_FS("zonefs");
1251 module_init(zonefs_init);
1252 module_exit(zonefs_exit);