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>
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 static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
28 unsigned int flags, struct iomap *iomap,
31 struct zonefs_inode_info *zi = ZONEFS_I(inode);
32 struct super_block *sb = inode->i_sb;
35 /* All I/Os should always be within the file maximum size */
36 if (WARN_ON_ONCE(offset + length > zi->i_max_size))
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
44 if (WARN_ON_ONCE(zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
45 (flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT)))
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.
53 mutex_lock(&zi->i_truncate_mutex);
54 isize = i_size_read(inode);
56 iomap->type = IOMAP_UNWRITTEN;
58 iomap->type = IOMAP_MAPPED;
59 if (flags & IOMAP_WRITE)
60 length = zi->i_max_size - offset;
62 length = min(length, isize - offset);
63 mutex_unlock(&zi->i_truncate_mutex);
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;
73 static const struct iomap_ops zonefs_iomap_ops = {
74 .iomap_begin = zonefs_iomap_begin,
77 static int zonefs_readpage(struct file *unused, struct page *page)
79 return iomap_readpage(page, &zonefs_iomap_ops);
82 static void zonefs_readahead(struct readahead_control *rac)
84 iomap_readahead(rac, &zonefs_iomap_ops);
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.
91 static int zonefs_map_blocks(struct iomap_writepage_ctx *wpc,
92 struct inode *inode, loff_t offset)
94 struct zonefs_inode_info *zi = ZONEFS_I(inode);
96 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
98 if (WARN_ON_ONCE(offset >= i_size_read(inode)))
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)
106 return zonefs_iomap_begin(inode, offset, zi->i_max_size - offset,
107 IOMAP_WRITE, &wpc->iomap, NULL);
110 static const struct iomap_writeback_ops zonefs_writeback_ops = {
111 .map_blocks = zonefs_map_blocks,
114 static int zonefs_writepage(struct page *page, struct writeback_control *wbc)
116 struct iomap_writepage_ctx wpc = { };
118 return iomap_writepage(page, wbc, &wpc, &zonefs_writeback_ops);
121 static int zonefs_writepages(struct address_space *mapping,
122 struct writeback_control *wbc)
124 struct iomap_writepage_ctx wpc = { };
126 return iomap_writepages(mapping, wbc, &wpc, &zonefs_writeback_ops);
129 static 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,
143 static void zonefs_update_stats(struct inode *inode, loff_t new_isize)
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);
150 if (new_isize == old_isize)
153 spin_lock(&sbi->s_lock);
156 * This may be called for an update after an IO error.
157 * So beware of the values seen.
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;
164 sbi->s_used_blocks = 0;
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;
172 spin_unlock(&sbi->s_lock);
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.
180 static loff_t zonefs_check_zone_condition(struct inode *inode,
181 struct blk_zone *zone, bool warn,
184 struct zonefs_inode_info *zi = ZONEFS_I(inode);
186 switch (zone->cond) {
187 case BLK_ZONE_COND_OFFLINE:
189 * Dead zone: make the inode immutable, disable all accesses
190 * and set the file size to 0 (zone wp set to zone start).
193 zonefs_warn(inode->i_sb, "inode %lu: offline zone\n",
195 inode->i_flags |= S_IMMUTABLE;
196 inode->i_mode &= ~0777;
197 zone->wp = zone->start;
199 case BLK_ZONE_COND_READONLY:
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.
209 zonefs_warn(inode->i_sb, "inode %lu: read-only zone\n",
211 inode->i_flags |= S_IMMUTABLE;
213 zone->cond = BLK_ZONE_COND_OFFLINE;
214 inode->i_mode &= ~0777;
215 zone->wp = zone->start;
218 inode->i_mode &= ~0222;
219 return i_size_read(inode);
221 if (zi->i_ztype == ZONEFS_ZTYPE_CNV)
222 return zi->i_max_size;
223 return (zone->wp - zone->start) << SECTOR_SHIFT;
227 struct zonefs_ioerr_data {
232 static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
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;
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).
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)
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).
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.
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);
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.
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",
288 if (zone->cond != BLK_ZONE_COND_OFFLINE) {
289 zone->cond = BLK_ZONE_COND_OFFLINE;
290 data_size = zonefs_check_zone_condition(inode, zone,
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",
297 if (zone->cond != BLK_ZONE_COND_READONLY) {
298 zone->cond = BLK_ZONE_COND_READONLY;
299 data_size = zonefs_check_zone_condition(inode, zone,
305 * If error=remount-ro was specified, any error result in remounting
306 * the volume as read-only.
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;
314 * Update block usage stats and the inode size to prevent access to
317 zonefs_update_stats(inode, data_size);
318 i_size_write(inode, data_size);
319 zi->i_wpoffset = data_size;
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).
331 static void zonefs_io_error(struct inode *inode, bool write)
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_max_size >> (sbi->s_zone_sectors_shift + SECTOR_SHIFT);
339 struct zonefs_ioerr_data err = {
345 mutex_lock(&zi->i_truncate_mutex);
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.
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);
359 zonefs_err(sb, "Get inode %lu zone information failed %d\n",
361 memalloc_noio_restore(noio_flag);
363 mutex_unlock(&zi->i_truncate_mutex);
366 static int zonefs_file_truncate(struct inode *inode, loff_t isize)
368 struct zonefs_inode_info *zi = ZONEFS_I(inode);
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.
378 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
382 op = REQ_OP_ZONE_RESET;
383 else if (isize == zi->i_max_size)
384 op = REQ_OP_ZONE_FINISH;
388 inode_dio_wait(inode);
390 /* Serialize against page faults */
391 down_write(&zi->i_mmap_sem);
393 /* Serialize against zonefs_iomap_begin() */
394 mutex_lock(&zi->i_truncate_mutex);
396 old_isize = i_size_read(inode);
397 if (isize == old_isize)
400 ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector,
401 zi->i_max_size >> SECTOR_SHIFT, GFP_NOFS);
403 zonefs_err(inode->i_sb,
404 "Zone management operation at %llu failed %d",
409 zonefs_update_stats(inode, isize);
410 truncate_setsize(inode, isize);
411 zi->i_wpoffset = isize;
414 mutex_unlock(&zi->i_truncate_mutex);
415 up_write(&zi->i_mmap_sem);
420 static int zonefs_inode_setattr(struct dentry *dentry, struct iattr *iattr)
422 struct inode *inode = d_inode(dentry);
425 if (unlikely(IS_IMMUTABLE(inode)))
428 ret = setattr_prepare(dentry, iattr);
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.
437 if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
438 (iattr->ia_mode & 0222))
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);
450 if (iattr->ia_valid & ATTR_SIZE) {
451 ret = zonefs_file_truncate(inode, iattr->ia_size);
456 setattr_copy(inode, iattr);
461 static const struct inode_operations zonefs_file_inode_operations = {
462 .setattr = zonefs_inode_setattr,
465 static int zonefs_file_fsync(struct file *file, loff_t start, loff_t end,
468 struct inode *inode = file_inode(file);
471 if (unlikely(IS_IMMUTABLE(inode)))
475 * Since only direct writes are allowed in sequential files, page cache
476 * flush is needed only for conventional zone files.
478 if (ZONEFS_I(inode)->i_ztype == ZONEFS_ZTYPE_CNV)
479 ret = file_write_and_wait_range(file, start, end);
481 ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL);
484 zonefs_io_error(inode, true);
489 static vm_fault_t zonefs_filemap_fault(struct vm_fault *vmf)
491 struct zonefs_inode_info *zi = ZONEFS_I(file_inode(vmf->vma->vm_file));
494 down_read(&zi->i_mmap_sem);
495 ret = filemap_fault(vmf);
496 up_read(&zi->i_mmap_sem);
501 static vm_fault_t zonefs_filemap_page_mkwrite(struct vm_fault *vmf)
503 struct inode *inode = file_inode(vmf->vma->vm_file);
504 struct zonefs_inode_info *zi = ZONEFS_I(inode);
507 if (unlikely(IS_IMMUTABLE(inode)))
508 return VM_FAULT_SIGBUS;
511 * Sanity check: only conventional zone files can have shared
512 * writeable mappings.
514 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
515 return VM_FAULT_NOPAGE;
517 sb_start_pagefault(inode->i_sb);
518 file_update_time(vmf->vma->vm_file);
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);
525 sb_end_pagefault(inode->i_sb);
529 static 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,
535 static int zonefs_file_mmap(struct file *file, struct vm_area_struct *vma)
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.
543 if (ZONEFS_I(file_inode(file))->i_ztype == ZONEFS_ZTYPE_SEQ &&
544 (vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
548 vma->vm_ops = &zonefs_file_vm_ops;
553 static loff_t zonefs_file_llseek(struct file *file, loff_t offset, int whence)
555 loff_t isize = i_size_read(file_inode(file));
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.
562 return generic_file_llseek_size(file, offset, whence, isize, isize);
565 static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
566 int error, unsigned int flags)
568 struct inode *inode = file_inode(iocb->ki_filp);
569 struct zonefs_inode_info *zi = ZONEFS_I(inode);
572 zonefs_io_error(inode, true);
576 if (size && zi->i_ztype != ZONEFS_ZTYPE_CNV) {
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.
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);
589 mutex_unlock(&zi->i_truncate_mutex);
595 static const struct iomap_dio_ops zonefs_write_dio_ops = {
596 .end_io = zonefs_file_write_dio_end_io,
599 static ssize_t zonefs_file_dio_append(struct kiocb *iocb, struct iov_iter *from)
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;
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);
614 nr_pages = iov_iter_npages(from, BIO_MAX_PAGES);
618 bio = bio_alloc_bioset(GFP_NOFS, nr_pages, &fs_bio_set);
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;
630 ret = bio_iov_iter_get_pages(bio, from);
635 size = bio->bi_iter.bi_size;
636 task_io_account_write(ret);
638 if (iocb->ki_flags & IOCB_HIPRI)
639 bio_set_polled(bio, iocb);
641 ret = submit_bio_wait(bio);
645 zonefs_file_write_dio_end_io(iocb, size, ret, 0);
647 iocb->ki_pos += size;
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.
664 static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
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);
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).
679 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !sync &&
680 (iocb->ki_flags & IOCB_NOWAIT))
683 if (iocb->ki_flags & IOCB_NOWAIT) {
684 if (!inode_trylock(inode))
690 ret = generic_write_checks(iocb, from);
694 iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos);
695 count = iov_iter_count(from);
697 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
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);
710 mutex_unlock(&zi->i_truncate_mutex);
715 ret = zonefs_file_dio_append(iocb, from);
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)) {
723 mutex_lock(&zi->i_truncate_mutex);
724 zi->i_wpoffset += count;
725 mutex_unlock(&zi->i_truncate_mutex);
734 static ssize_t zonefs_file_buffered_write(struct kiocb *iocb,
735 struct iov_iter *from)
737 struct inode *inode = file_inode(iocb->ki_filp);
738 struct zonefs_inode_info *zi = ZONEFS_I(inode);
742 * Direct IO writes are mandatory for sequential zone files so that the
743 * write IO issuing order is preserved.
745 if (zi->i_ztype != ZONEFS_ZTYPE_CNV)
748 if (iocb->ki_flags & IOCB_NOWAIT) {
749 if (!inode_trylock(inode))
755 ret = generic_write_checks(iocb, from);
759 iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos);
761 ret = iomap_file_buffered_write(iocb, from, &zonefs_iomap_ops);
764 else if (ret == -EIO)
765 zonefs_io_error(inode, true);
770 ret = generic_write_sync(iocb, ret);
775 static ssize_t zonefs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
777 struct inode *inode = file_inode(iocb->ki_filp);
779 if (unlikely(IS_IMMUTABLE(inode)))
782 if (sb_rdonly(inode->i_sb))
785 /* Write operations beyond the zone size are not allowed */
786 if (iocb->ki_pos >= ZONEFS_I(inode)->i_max_size)
789 if (iocb->ki_flags & IOCB_DIRECT) {
790 ssize_t ret = zonefs_file_dio_write(iocb, from);
795 return zonefs_file_buffered_write(iocb, from);
798 static int zonefs_file_read_dio_end_io(struct kiocb *iocb, ssize_t size,
799 int error, unsigned int flags)
802 zonefs_io_error(file_inode(iocb->ki_filp), false);
809 static const struct iomap_dio_ops zonefs_read_dio_ops = {
810 .end_io = zonefs_file_read_dio_end_io,
813 static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
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;
821 /* Offline zones cannot be read */
822 if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777)))
825 if (iocb->ki_pos >= zi->i_max_size)
828 if (iocb->ki_flags & IOCB_NOWAIT) {
829 if (!inode_trylock_shared(inode))
832 inode_lock_shared(inode);
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);
843 iov_iter_truncate(to, isize - iocb->ki_pos);
844 mutex_unlock(&zi->i_truncate_mutex);
846 if (iocb->ki_flags & IOCB_DIRECT) {
847 size_t count = iov_iter_count(to);
849 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
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));
857 ret = generic_file_read_iter(iocb, to);
859 zonefs_io_error(inode, false);
863 inode_unlock_shared(inode);
868 static 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,
880 static struct kmem_cache *zonefs_inode_cachep;
882 static struct inode *zonefs_alloc_inode(struct super_block *sb)
884 struct zonefs_inode_info *zi;
886 zi = kmem_cache_alloc(zonefs_inode_cachep, GFP_KERNEL);
890 inode_init_once(&zi->i_vnode);
891 mutex_init(&zi->i_truncate_mutex);
892 init_rwsem(&zi->i_mmap_sem);
897 static void zonefs_free_inode(struct inode *inode)
899 kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
905 static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
907 struct super_block *sb = dentry->d_sb;
908 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
912 buf->f_type = ZONEFS_MAGIC;
913 buf->f_bsize = sb->s_blocksize;
914 buf->f_namelen = ZONEFS_NAME_MAX;
916 spin_lock(&sbi->s_lock);
918 buf->f_blocks = sbi->s_blocks;
919 if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
922 buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
923 buf->f_bavail = buf->f_bfree;
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;
931 spin_unlock(&sbi->s_lock);
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);
942 Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
946 static 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"},
954 static int zonefs_parse_options(struct super_block *sb, char *options)
956 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
957 substring_t args[MAX_OPT_ARGS];
963 while ((p = strsep(&options, ",")) != NULL) {
969 token = match_token(p, tokens, args);
972 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
973 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
976 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
977 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
980 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
981 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
983 case Opt_errors_repair:
984 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
985 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
995 static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
997 struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
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");
1011 static int zonefs_remount(struct super_block *sb, int *flags, char *data)
1013 sync_filesystem(sb);
1015 return zonefs_parse_options(sb, data);
1018 static 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,
1026 static const struct inode_operations zonefs_dir_inode_operations = {
1027 .lookup = simple_lookup,
1028 .setattr = zonefs_inode_setattr,
1031 static void zonefs_init_dir_inode(struct inode *parent, struct inode *inode,
1032 enum zonefs_ztype type)
1034 struct super_block *sb = parent->i_sb;
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);
1044 static void zonefs_init_file_inode(struct inode *inode, struct blk_zone *zone,
1045 enum zonefs_ztype type)
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);
1051 inode->i_ino = zone->start >> sbi->s_zone_sectors_shift;
1052 inode->i_mode = S_IFREG | sbi->s_perm;
1055 zi->i_zsector = zone->start;
1056 zi->i_max_size = min_t(loff_t, MAX_LFS_FILESIZE,
1057 zone->len << SECTOR_SHIFT);
1058 zi->i_wpoffset = zonefs_check_zone_condition(inode, zone, true, true);
1060 inode->i_uid = sbi->s_uid;
1061 inode->i_gid = sbi->s_gid;
1062 inode->i_size = zi->i_wpoffset;
1063 inode->i_blocks = zone->len;
1065 inode->i_op = &zonefs_file_inode_operations;
1066 inode->i_fop = &zonefs_file_operations;
1067 inode->i_mapping->a_ops = &zonefs_file_aops;
1069 sb->s_maxbytes = max(zi->i_max_size, sb->s_maxbytes);
1070 sbi->s_blocks += zi->i_max_size >> sb->s_blocksize_bits;
1071 sbi->s_used_blocks += zi->i_wpoffset >> sb->s_blocksize_bits;
1074 static struct dentry *zonefs_create_inode(struct dentry *parent,
1075 const char *name, struct blk_zone *zone,
1076 enum zonefs_ztype type)
1078 struct inode *dir = d_inode(parent);
1079 struct dentry *dentry;
1080 struct inode *inode;
1082 dentry = d_alloc_name(parent, name);
1086 inode = new_inode(parent->d_sb);
1090 inode->i_ctime = inode->i_mtime = inode->i_atime = dir->i_ctime;
1092 zonefs_init_file_inode(inode, zone, type);
1094 zonefs_init_dir_inode(dir, inode, type);
1095 d_add(dentry, inode);
1106 struct zonefs_zone_data {
1107 struct super_block *sb;
1108 unsigned int nr_zones[ZONEFS_ZTYPE_MAX];
1109 struct blk_zone *zones;
1113 * Create a zone group and populate it with zone files.
1115 static int zonefs_create_zgroup(struct zonefs_zone_data *zd,
1116 enum zonefs_ztype type)
1118 struct super_block *sb = zd->sb;
1119 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1120 struct blk_zone *zone, *next, *end;
1121 const char *zgroup_name;
1127 /* If the group is empty, there is nothing to do */
1128 if (!zd->nr_zones[type])
1131 file_name = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
1135 if (type == ZONEFS_ZTYPE_CNV)
1136 zgroup_name = "cnv";
1138 zgroup_name = "seq";
1140 dir = zonefs_create_inode(sb->s_root, zgroup_name, NULL, type);
1147 * The first zone contains the super block: skip it.
1149 end = zd->zones + blkdev_nr_zones(sb->s_bdev->bd_disk);
1150 for (zone = &zd->zones[1]; zone < end; zone = next) {
1153 if (zonefs_zone_type(zone) != type)
1157 * For conventional zones, contiguous zones can be aggregated
1158 * together to form larger files. Note that this overwrites the
1159 * length of the first zone of the set of contiguous zones
1160 * aggregated together. If one offline or read-only zone is
1161 * found, assume that all zones aggregated have the same
1164 if (type == ZONEFS_ZTYPE_CNV &&
1165 (sbi->s_features & ZONEFS_F_AGGRCNV)) {
1166 for (; next < end; next++) {
1167 if (zonefs_zone_type(next) != type)
1169 zone->len += next->len;
1170 if (next->cond == BLK_ZONE_COND_READONLY &&
1171 zone->cond != BLK_ZONE_COND_OFFLINE)
1172 zone->cond = BLK_ZONE_COND_READONLY;
1173 else if (next->cond == BLK_ZONE_COND_OFFLINE)
1174 zone->cond = BLK_ZONE_COND_OFFLINE;
1179 * Use the file number within its group as file name.
1181 snprintf(file_name, ZONEFS_NAME_MAX - 1, "%u", n);
1182 if (!zonefs_create_inode(dir, file_name, zone, type)) {
1190 zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1191 zgroup_name, n, n > 1 ? "s" : "");
1193 sbi->s_nr_files[type] = n;
1202 static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
1205 struct zonefs_zone_data *zd = data;
1208 * Count the number of usable zones: the first zone at index 0 contains
1209 * the super block and is ignored.
1211 switch (zone->type) {
1212 case BLK_ZONE_TYPE_CONVENTIONAL:
1213 zone->wp = zone->start + zone->len;
1215 zd->nr_zones[ZONEFS_ZTYPE_CNV]++;
1217 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1218 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1220 zd->nr_zones[ZONEFS_ZTYPE_SEQ]++;
1223 zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
1228 memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
1233 static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
1235 struct block_device *bdev = zd->sb->s_bdev;
1238 zd->zones = kvcalloc(blkdev_nr_zones(bdev->bd_disk),
1239 sizeof(struct blk_zone), GFP_KERNEL);
1243 /* Get zones information from the device */
1244 ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
1245 zonefs_get_zone_info_cb, zd);
1247 zonefs_err(zd->sb, "Zone report failed %d\n", ret);
1251 if (ret != blkdev_nr_zones(bdev->bd_disk)) {
1252 zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
1253 ret, blkdev_nr_zones(bdev->bd_disk));
1260 static inline void zonefs_cleanup_zone_info(struct zonefs_zone_data *zd)
1266 * Read super block information from the device.
1268 static int zonefs_read_super(struct super_block *sb)
1270 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1271 struct zonefs_super *super;
1272 u32 crc, stored_crc;
1274 struct bio_vec bio_vec;
1278 page = alloc_page(GFP_KERNEL);
1282 bio_init(&bio, &bio_vec, 1);
1283 bio.bi_iter.bi_sector = 0;
1284 bio.bi_opf = REQ_OP_READ;
1285 bio_set_dev(&bio, sb->s_bdev);
1286 bio_add_page(&bio, page, PAGE_SIZE, 0);
1288 ret = submit_bio_wait(&bio);
1295 if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1298 stored_crc = le32_to_cpu(super->s_crc);
1300 crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1301 if (crc != stored_crc) {
1302 zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1307 sbi->s_features = le64_to_cpu(super->s_features);
1308 if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1309 zonefs_err(sb, "Unknown features set 0x%llx\n",
1314 if (sbi->s_features & ZONEFS_F_UID) {
1315 sbi->s_uid = make_kuid(current_user_ns(),
1316 le32_to_cpu(super->s_uid));
1317 if (!uid_valid(sbi->s_uid)) {
1318 zonefs_err(sb, "Invalid UID feature\n");
1323 if (sbi->s_features & ZONEFS_F_GID) {
1324 sbi->s_gid = make_kgid(current_user_ns(),
1325 le32_to_cpu(super->s_gid));
1326 if (!gid_valid(sbi->s_gid)) {
1327 zonefs_err(sb, "Invalid GID feature\n");
1332 if (sbi->s_features & ZONEFS_F_PERM)
1333 sbi->s_perm = le32_to_cpu(super->s_perm);
1335 if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1336 zonefs_err(sb, "Reserved area is being used\n");
1340 import_uuid(&sbi->s_uuid, super->s_uuid);
1352 * Check that the device is zoned. If it is, get the list of zones and create
1353 * sub-directories and files according to the device zone configuration and
1356 static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1358 struct zonefs_zone_data zd;
1359 struct zonefs_sb_info *sbi;
1360 struct inode *inode;
1361 enum zonefs_ztype t;
1364 if (!bdev_is_zoned(sb->s_bdev)) {
1365 zonefs_err(sb, "Not a zoned block device\n");
1370 * Initialize super block information: the maximum file size is updated
1371 * when the zone files are created so that the format option
1372 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1373 * beyond the zone size is taken into account.
1375 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1379 spin_lock_init(&sbi->s_lock);
1380 sb->s_fs_info = sbi;
1381 sb->s_magic = ZONEFS_MAGIC;
1383 sb->s_op = &zonefs_sops;
1384 sb->s_time_gran = 1;
1387 * The block size is set to the device physical sector size to ensure
1388 * that write operations on 512e devices (512B logical block and 4KB
1389 * physical block) are always aligned to the device physical blocks,
1390 * as mandated by the ZBC/ZAC specifications.
1392 sb_set_blocksize(sb, bdev_physical_block_size(sb->s_bdev));
1393 sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1394 sbi->s_uid = GLOBAL_ROOT_UID;
1395 sbi->s_gid = GLOBAL_ROOT_GID;
1397 sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1399 ret = zonefs_read_super(sb);
1403 ret = zonefs_parse_options(sb, data);
1407 memset(&zd, 0, sizeof(struct zonefs_zone_data));
1409 ret = zonefs_get_zone_info(&zd);
1413 zonefs_info(sb, "Mounting %u zones",
1414 blkdev_nr_zones(sb->s_bdev->bd_disk));
1416 /* Create root directory inode */
1418 inode = new_inode(sb);
1422 inode->i_ino = blkdev_nr_zones(sb->s_bdev->bd_disk);
1423 inode->i_mode = S_IFDIR | 0555;
1424 inode->i_ctime = inode->i_mtime = inode->i_atime = current_time(inode);
1425 inode->i_op = &zonefs_dir_inode_operations;
1426 inode->i_fop = &simple_dir_operations;
1427 set_nlink(inode, 2);
1429 sb->s_root = d_make_root(inode);
1433 /* Create and populate files in zone groups directories */
1434 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
1435 ret = zonefs_create_zgroup(&zd, t);
1441 zonefs_cleanup_zone_info(&zd);
1446 static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1447 int flags, const char *dev_name, void *data)
1449 return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1452 static void zonefs_kill_super(struct super_block *sb)
1454 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1457 d_genocide(sb->s_root);
1458 kill_block_super(sb);
1463 * File system definition and registration.
1465 static struct file_system_type zonefs_type = {
1466 .owner = THIS_MODULE,
1468 .mount = zonefs_mount,
1469 .kill_sb = zonefs_kill_super,
1470 .fs_flags = FS_REQUIRES_DEV,
1473 static int __init zonefs_init_inodecache(void)
1475 zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1476 sizeof(struct zonefs_inode_info), 0,
1477 (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1479 if (zonefs_inode_cachep == NULL)
1484 static void zonefs_destroy_inodecache(void)
1487 * Make sure all delayed rcu free inodes are flushed before we
1488 * destroy the inode cache.
1491 kmem_cache_destroy(zonefs_inode_cachep);
1494 static int __init zonefs_init(void)
1498 BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1500 ret = zonefs_init_inodecache();
1504 ret = register_filesystem(&zonefs_type);
1506 zonefs_destroy_inodecache();
1513 static void __exit zonefs_exit(void)
1515 zonefs_destroy_inodecache();
1516 unregister_filesystem(&zonefs_type);
1519 MODULE_AUTHOR("Damien Le Moal");
1520 MODULE_DESCRIPTION("Zone file system for zoned block devices");
1521 MODULE_LICENSE("GPL");
1522 module_init(zonefs_init);
1523 module_exit(zonefs_exit);