Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Simple file system for zoned block devices exposing zones as files.
4 *
5 * Copyright (C) 2019 Western Digital Corporation or its affiliates.
6 */
7#include <linux/module.h>
8#include <linux/pagemap.h>
9#include <linux/magic.h>
10#include <linux/iomap.h>
11#include <linux/init.h>
12#include <linux/slab.h>
13#include <linux/blkdev.h>
14#include <linux/statfs.h>
15#include <linux/writeback.h>
16#include <linux/quotaops.h>
17#include <linux/seq_file.h>
18#include <linux/parser.h>
19#include <linux/uio.h>
20#include <linux/mman.h>
21#include <linux/sched/mm.h>
22#include <linux/crc32.h>
23#include <linux/task_io_accounting_ops.h>
24
25#include "zonefs.h"
26
27#define CREATE_TRACE_POINTS
28#include "trace.h"
29
30/*
31 * Manage the active zone count. Called with zi->i_truncate_mutex held.
32 */
33static void zonefs_account_active(struct inode *inode)
34{
35 struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
36 struct zonefs_inode_info *zi = ZONEFS_I(inode);
37
38 lockdep_assert_held(&zi->i_truncate_mutex);
39
40 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
41 return;
42
43 /*
44 * For zones that transitioned to the offline or readonly condition,
45 * we only need to clear the active state.
46 */
47 if (zi->i_flags & (ZONEFS_ZONE_OFFLINE | ZONEFS_ZONE_READONLY))
48 goto out;
49
50 /*
51 * If the zone is active, that is, if it is explicitly open or
52 * partially written, check if it was already accounted as active.
53 */
54 if ((zi->i_flags & ZONEFS_ZONE_OPEN) ||
55 (zi->i_wpoffset > 0 && zi->i_wpoffset < zi->i_max_size)) {
56 if (!(zi->i_flags & ZONEFS_ZONE_ACTIVE)) {
57 zi->i_flags |= ZONEFS_ZONE_ACTIVE;
58 atomic_inc(&sbi->s_active_seq_files);
59 }
60 return;
61 }
62
63out:
64 /* The zone is not active. If it was, update the active count */
65 if (zi->i_flags & ZONEFS_ZONE_ACTIVE) {
66 zi->i_flags &= ~ZONEFS_ZONE_ACTIVE;
67 atomic_dec(&sbi->s_active_seq_files);
68 }
69}
70
71static inline int zonefs_zone_mgmt(struct inode *inode, enum req_op op)
72{
73 struct zonefs_inode_info *zi = ZONEFS_I(inode);
74 int ret;
75
76 lockdep_assert_held(&zi->i_truncate_mutex);
77
78 /*
79 * With ZNS drives, closing an explicitly open zone that has not been
80 * written will change the zone state to "closed", that is, the zone
81 * will remain active. Since this can then cause failure of explicit
82 * open operation on other zones if the drive active zone resources
83 * are exceeded, make sure that the zone does not remain active by
84 * resetting it.
85 */
86 if (op == REQ_OP_ZONE_CLOSE && !zi->i_wpoffset)
87 op = REQ_OP_ZONE_RESET;
88
89 trace_zonefs_zone_mgmt(inode, op);
90 ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector,
91 zi->i_zone_size >> SECTOR_SHIFT, GFP_NOFS);
92 if (ret) {
93 zonefs_err(inode->i_sb,
94 "Zone management operation %s at %llu failed %d\n",
95 blk_op_str(op), zi->i_zsector, ret);
96 return ret;
97 }
98
99 return 0;
100}
101
102static inline void zonefs_i_size_write(struct inode *inode, loff_t isize)
103{
104 struct zonefs_inode_info *zi = ZONEFS_I(inode);
105
106 i_size_write(inode, isize);
107 /*
108 * A full zone is no longer open/active and does not need
109 * explicit closing.
110 */
111 if (isize >= zi->i_max_size) {
112 struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
113
114 if (zi->i_flags & ZONEFS_ZONE_ACTIVE)
115 atomic_dec(&sbi->s_active_seq_files);
116 zi->i_flags &= ~(ZONEFS_ZONE_OPEN | ZONEFS_ZONE_ACTIVE);
117 }
118}
119
120static int zonefs_read_iomap_begin(struct inode *inode, loff_t offset,
121 loff_t length, unsigned int flags,
122 struct iomap *iomap, struct iomap *srcmap)
123{
124 struct zonefs_inode_info *zi = ZONEFS_I(inode);
125 struct super_block *sb = inode->i_sb;
126 loff_t isize;
127
128 /*
129 * All blocks are always mapped below EOF. If reading past EOF,
130 * act as if there is a hole up to the file maximum size.
131 */
132 mutex_lock(&zi->i_truncate_mutex);
133 iomap->bdev = inode->i_sb->s_bdev;
134 iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize);
135 isize = i_size_read(inode);
136 if (iomap->offset >= isize) {
137 iomap->type = IOMAP_HOLE;
138 iomap->addr = IOMAP_NULL_ADDR;
139 iomap->length = length;
140 } else {
141 iomap->type = IOMAP_MAPPED;
142 iomap->addr = (zi->i_zsector << SECTOR_SHIFT) + iomap->offset;
143 iomap->length = isize - iomap->offset;
144 }
145 mutex_unlock(&zi->i_truncate_mutex);
146
147 trace_zonefs_iomap_begin(inode, iomap);
148
149 return 0;
150}
151
152static const struct iomap_ops zonefs_read_iomap_ops = {
153 .iomap_begin = zonefs_read_iomap_begin,
154};
155
156static int zonefs_write_iomap_begin(struct inode *inode, loff_t offset,
157 loff_t length, unsigned int flags,
158 struct iomap *iomap, struct iomap *srcmap)
159{
160 struct zonefs_inode_info *zi = ZONEFS_I(inode);
161 struct super_block *sb = inode->i_sb;
162 loff_t isize;
163
164 /* All write I/Os should always be within the file maximum size */
165 if (WARN_ON_ONCE(offset + length > zi->i_max_size))
166 return -EIO;
167
168 /*
169 * Sequential zones can only accept direct writes. This is already
170 * checked when writes are issued, so warn if we see a page writeback
171 * operation.
172 */
173 if (WARN_ON_ONCE(zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
174 !(flags & IOMAP_DIRECT)))
175 return -EIO;
176
177 /*
178 * For conventional zones, all blocks are always mapped. For sequential
179 * zones, all blocks after always mapped below the inode size (zone
180 * write pointer) and unwriten beyond.
181 */
182 mutex_lock(&zi->i_truncate_mutex);
183 iomap->bdev = inode->i_sb->s_bdev;
184 iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize);
185 iomap->addr = (zi->i_zsector << SECTOR_SHIFT) + iomap->offset;
186 isize = i_size_read(inode);
187 if (iomap->offset >= isize) {
188 iomap->type = IOMAP_UNWRITTEN;
189 iomap->length = zi->i_max_size - iomap->offset;
190 } else {
191 iomap->type = IOMAP_MAPPED;
192 iomap->length = isize - iomap->offset;
193 }
194 mutex_unlock(&zi->i_truncate_mutex);
195
196 trace_zonefs_iomap_begin(inode, iomap);
197
198 return 0;
199}
200
201static const struct iomap_ops zonefs_write_iomap_ops = {
202 .iomap_begin = zonefs_write_iomap_begin,
203};
204
205static int zonefs_read_folio(struct file *unused, struct folio *folio)
206{
207 return iomap_read_folio(folio, &zonefs_read_iomap_ops);
208}
209
210static void zonefs_readahead(struct readahead_control *rac)
211{
212 iomap_readahead(rac, &zonefs_read_iomap_ops);
213}
214
215/*
216 * Map blocks for page writeback. This is used only on conventional zone files,
217 * which implies that the page range can only be within the fixed inode size.
218 */
219static int zonefs_write_map_blocks(struct iomap_writepage_ctx *wpc,
220 struct inode *inode, loff_t offset)
221{
222 struct zonefs_inode_info *zi = ZONEFS_I(inode);
223
224 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
225 return -EIO;
226 if (WARN_ON_ONCE(offset >= i_size_read(inode)))
227 return -EIO;
228
229 /* If the mapping is already OK, nothing needs to be done */
230 if (offset >= wpc->iomap.offset &&
231 offset < wpc->iomap.offset + wpc->iomap.length)
232 return 0;
233
234 return zonefs_write_iomap_begin(inode, offset, zi->i_max_size - offset,
235 IOMAP_WRITE, &wpc->iomap, NULL);
236}
237
238static const struct iomap_writeback_ops zonefs_writeback_ops = {
239 .map_blocks = zonefs_write_map_blocks,
240};
241
242static int zonefs_writepages(struct address_space *mapping,
243 struct writeback_control *wbc)
244{
245 struct iomap_writepage_ctx wpc = { };
246
247 return iomap_writepages(mapping, wbc, &wpc, &zonefs_writeback_ops);
248}
249
250static int zonefs_swap_activate(struct swap_info_struct *sis,
251 struct file *swap_file, sector_t *span)
252{
253 struct inode *inode = file_inode(swap_file);
254 struct zonefs_inode_info *zi = ZONEFS_I(inode);
255
256 if (zi->i_ztype != ZONEFS_ZTYPE_CNV) {
257 zonefs_err(inode->i_sb,
258 "swap file: not a conventional zone file\n");
259 return -EINVAL;
260 }
261
262 return iomap_swapfile_activate(sis, swap_file, span,
263 &zonefs_read_iomap_ops);
264}
265
266static const struct address_space_operations zonefs_file_aops = {
267 .read_folio = zonefs_read_folio,
268 .readahead = zonefs_readahead,
269 .writepages = zonefs_writepages,
270 .dirty_folio = filemap_dirty_folio,
271 .release_folio = iomap_release_folio,
272 .invalidate_folio = iomap_invalidate_folio,
273 .migrate_folio = filemap_migrate_folio,
274 .is_partially_uptodate = iomap_is_partially_uptodate,
275 .error_remove_page = generic_error_remove_page,
276 .direct_IO = noop_direct_IO,
277 .swap_activate = zonefs_swap_activate,
278};
279
280static void zonefs_update_stats(struct inode *inode, loff_t new_isize)
281{
282 struct super_block *sb = inode->i_sb;
283 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
284 loff_t old_isize = i_size_read(inode);
285 loff_t nr_blocks;
286
287 if (new_isize == old_isize)
288 return;
289
290 spin_lock(&sbi->s_lock);
291
292 /*
293 * This may be called for an update after an IO error.
294 * So beware of the values seen.
295 */
296 if (new_isize < old_isize) {
297 nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits;
298 if (sbi->s_used_blocks > nr_blocks)
299 sbi->s_used_blocks -= nr_blocks;
300 else
301 sbi->s_used_blocks = 0;
302 } else {
303 sbi->s_used_blocks +=
304 (new_isize - old_isize) >> sb->s_blocksize_bits;
305 if (sbi->s_used_blocks > sbi->s_blocks)
306 sbi->s_used_blocks = sbi->s_blocks;
307 }
308
309 spin_unlock(&sbi->s_lock);
310}
311
312/*
313 * Check a zone condition and adjust its file inode access permissions for
314 * offline and readonly zones. Return the inode size corresponding to the
315 * amount of readable data in the zone.
316 */
317static loff_t zonefs_check_zone_condition(struct inode *inode,
318 struct blk_zone *zone, bool warn,
319 bool mount)
320{
321 struct zonefs_inode_info *zi = ZONEFS_I(inode);
322
323 switch (zone->cond) {
324 case BLK_ZONE_COND_OFFLINE:
325 /*
326 * Dead zone: make the inode immutable, disable all accesses
327 * and set the file size to 0 (zone wp set to zone start).
328 */
329 if (warn)
330 zonefs_warn(inode->i_sb, "inode %lu: offline zone\n",
331 inode->i_ino);
332 inode->i_flags |= S_IMMUTABLE;
333 inode->i_mode &= ~0777;
334 zone->wp = zone->start;
335 zi->i_flags |= ZONEFS_ZONE_OFFLINE;
336 return 0;
337 case BLK_ZONE_COND_READONLY:
338 /*
339 * The write pointer of read-only zones is invalid. If such a
340 * zone is found during mount, the file size cannot be retrieved
341 * so we treat the zone as offline (mount == true case).
342 * Otherwise, keep the file size as it was when last updated
343 * so that the user can recover data. In both cases, writes are
344 * always disabled for the zone.
345 */
346 if (warn)
347 zonefs_warn(inode->i_sb, "inode %lu: read-only zone\n",
348 inode->i_ino);
349 inode->i_flags |= S_IMMUTABLE;
350 if (mount) {
351 zone->cond = BLK_ZONE_COND_OFFLINE;
352 inode->i_mode &= ~0777;
353 zone->wp = zone->start;
354 zi->i_flags |= ZONEFS_ZONE_OFFLINE;
355 return 0;
356 }
357 zi->i_flags |= ZONEFS_ZONE_READONLY;
358 inode->i_mode &= ~0222;
359 return i_size_read(inode);
360 case BLK_ZONE_COND_FULL:
361 /* The write pointer of full zones is invalid. */
362 return zi->i_max_size;
363 default:
364 if (zi->i_ztype == ZONEFS_ZTYPE_CNV)
365 return zi->i_max_size;
366 return (zone->wp - zone->start) << SECTOR_SHIFT;
367 }
368}
369
370struct zonefs_ioerr_data {
371 struct inode *inode;
372 bool write;
373};
374
375static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx,
376 void *data)
377{
378 struct zonefs_ioerr_data *err = data;
379 struct inode *inode = err->inode;
380 struct zonefs_inode_info *zi = ZONEFS_I(inode);
381 struct super_block *sb = inode->i_sb;
382 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
383 loff_t isize, data_size;
384
385 /*
386 * Check the zone condition: if the zone is not "bad" (offline or
387 * read-only), read errors are simply signaled to the IO issuer as long
388 * as there is no inconsistency between the inode size and the amount of
389 * data writen in the zone (data_size).
390 */
391 data_size = zonefs_check_zone_condition(inode, zone, true, false);
392 isize = i_size_read(inode);
393 if (zone->cond != BLK_ZONE_COND_OFFLINE &&
394 zone->cond != BLK_ZONE_COND_READONLY &&
395 !err->write && isize == data_size)
396 return 0;
397
398 /*
399 * At this point, we detected either a bad zone or an inconsistency
400 * between the inode size and the amount of data written in the zone.
401 * For the latter case, the cause may be a write IO error or an external
402 * action on the device. Two error patterns exist:
403 * 1) The inode size is lower than the amount of data in the zone:
404 * a write operation partially failed and data was writen at the end
405 * of the file. This can happen in the case of a large direct IO
406 * needing several BIOs and/or write requests to be processed.
407 * 2) The inode size is larger than the amount of data in the zone:
408 * this can happen with a deferred write error with the use of the
409 * device side write cache after getting successful write IO
410 * completions. Other possibilities are (a) an external corruption,
411 * e.g. an application reset the zone directly, or (b) the device
412 * has a serious problem (e.g. firmware bug).
413 *
414 * In all cases, warn about inode size inconsistency and handle the
415 * IO error according to the zone condition and to the mount options.
416 */
417 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && isize != data_size)
418 zonefs_warn(sb, "inode %lu: invalid size %lld (should be %lld)\n",
419 inode->i_ino, isize, data_size);
420
421 /*
422 * First handle bad zones signaled by hardware. The mount options
423 * errors=zone-ro and errors=zone-offline result in changing the
424 * zone condition to read-only and offline respectively, as if the
425 * condition was signaled by the hardware.
426 */
427 if (zone->cond == BLK_ZONE_COND_OFFLINE ||
428 sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL) {
429 zonefs_warn(sb, "inode %lu: read/write access disabled\n",
430 inode->i_ino);
431 if (zone->cond != BLK_ZONE_COND_OFFLINE) {
432 zone->cond = BLK_ZONE_COND_OFFLINE;
433 data_size = zonefs_check_zone_condition(inode, zone,
434 false, false);
435 }
436 } else if (zone->cond == BLK_ZONE_COND_READONLY ||
437 sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO) {
438 zonefs_warn(sb, "inode %lu: write access disabled\n",
439 inode->i_ino);
440 if (zone->cond != BLK_ZONE_COND_READONLY) {
441 zone->cond = BLK_ZONE_COND_READONLY;
442 data_size = zonefs_check_zone_condition(inode, zone,
443 false, false);
444 }
445 } else if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO &&
446 data_size > isize) {
447 /* Do not expose garbage data */
448 data_size = isize;
449 }
450
451 /*
452 * If the filesystem is mounted with the explicit-open mount option, we
453 * need to clear the ZONEFS_ZONE_OPEN flag if the zone transitioned to
454 * the read-only or offline condition, to avoid attempting an explicit
455 * close of the zone when the inode file is closed.
456 */
457 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) &&
458 (zone->cond == BLK_ZONE_COND_OFFLINE ||
459 zone->cond == BLK_ZONE_COND_READONLY))
460 zi->i_flags &= ~ZONEFS_ZONE_OPEN;
461
462 /*
463 * If error=remount-ro was specified, any error result in remounting
464 * the volume as read-only.
465 */
466 if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) {
467 zonefs_warn(sb, "remounting filesystem read-only\n");
468 sb->s_flags |= SB_RDONLY;
469 }
470
471 /*
472 * Update block usage stats and the inode size to prevent access to
473 * invalid data.
474 */
475 zonefs_update_stats(inode, data_size);
476 zonefs_i_size_write(inode, data_size);
477 zi->i_wpoffset = data_size;
478 zonefs_account_active(inode);
479
480 return 0;
481}
482
483/*
484 * When an file IO error occurs, check the file zone to see if there is a change
485 * in the zone condition (e.g. offline or read-only). For a failed write to a
486 * sequential zone, the zone write pointer position must also be checked to
487 * eventually correct the file size and zonefs inode write pointer offset
488 * (which can be out of sync with the drive due to partial write failures).
489 */
490static void __zonefs_io_error(struct inode *inode, bool write)
491{
492 struct zonefs_inode_info *zi = ZONEFS_I(inode);
493 struct super_block *sb = inode->i_sb;
494 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
495 unsigned int noio_flag;
496 unsigned int nr_zones = 1;
497 struct zonefs_ioerr_data err = {
498 .inode = inode,
499 .write = write,
500 };
501 int ret;
502
503 /*
504 * The only files that have more than one zone are conventional zone
505 * files with aggregated conventional zones, for which the inode zone
506 * size is always larger than the device zone size.
507 */
508 if (zi->i_zone_size > bdev_zone_sectors(sb->s_bdev))
509 nr_zones = zi->i_zone_size >>
510 (sbi->s_zone_sectors_shift + SECTOR_SHIFT);
511
512 /*
513 * Memory allocations in blkdev_report_zones() can trigger a memory
514 * reclaim which may in turn cause a recursion into zonefs as well as
515 * struct request allocations for the same device. The former case may
516 * end up in a deadlock on the inode truncate mutex, while the latter
517 * may prevent IO forward progress. Executing the report zones under
518 * the GFP_NOIO context avoids both problems.
519 */
520 noio_flag = memalloc_noio_save();
521 ret = blkdev_report_zones(sb->s_bdev, zi->i_zsector, nr_zones,
522 zonefs_io_error_cb, &err);
523 if (ret != nr_zones)
524 zonefs_err(sb, "Get inode %lu zone information failed %d\n",
525 inode->i_ino, ret);
526 memalloc_noio_restore(noio_flag);
527}
528
529static void zonefs_io_error(struct inode *inode, bool write)
530{
531 struct zonefs_inode_info *zi = ZONEFS_I(inode);
532
533 mutex_lock(&zi->i_truncate_mutex);
534 __zonefs_io_error(inode, write);
535 mutex_unlock(&zi->i_truncate_mutex);
536}
537
538static int zonefs_file_truncate(struct inode *inode, loff_t isize)
539{
540 struct zonefs_inode_info *zi = ZONEFS_I(inode);
541 loff_t old_isize;
542 enum req_op op;
543 int ret = 0;
544
545 /*
546 * Only sequential zone files can be truncated and truncation is allowed
547 * only down to a 0 size, which is equivalent to a zone reset, and to
548 * the maximum file size, which is equivalent to a zone finish.
549 */
550 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
551 return -EPERM;
552
553 if (!isize)
554 op = REQ_OP_ZONE_RESET;
555 else if (isize == zi->i_max_size)
556 op = REQ_OP_ZONE_FINISH;
557 else
558 return -EPERM;
559
560 inode_dio_wait(inode);
561
562 /* Serialize against page faults */
563 filemap_invalidate_lock(inode->i_mapping);
564
565 /* Serialize against zonefs_iomap_begin() */
566 mutex_lock(&zi->i_truncate_mutex);
567
568 old_isize = i_size_read(inode);
569 if (isize == old_isize)
570 goto unlock;
571
572 ret = zonefs_zone_mgmt(inode, op);
573 if (ret)
574 goto unlock;
575
576 /*
577 * If the mount option ZONEFS_MNTOPT_EXPLICIT_OPEN is set,
578 * take care of open zones.
579 */
580 if (zi->i_flags & ZONEFS_ZONE_OPEN) {
581 /*
582 * Truncating a zone to EMPTY or FULL is the equivalent of
583 * closing the zone. For a truncation to 0, we need to
584 * re-open the zone to ensure new writes can be processed.
585 * For a truncation to the maximum file size, the zone is
586 * closed and writes cannot be accepted anymore, so clear
587 * the open flag.
588 */
589 if (!isize)
590 ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
591 else
592 zi->i_flags &= ~ZONEFS_ZONE_OPEN;
593 }
594
595 zonefs_update_stats(inode, isize);
596 truncate_setsize(inode, isize);
597 zi->i_wpoffset = isize;
598 zonefs_account_active(inode);
599
600unlock:
601 mutex_unlock(&zi->i_truncate_mutex);
602 filemap_invalidate_unlock(inode->i_mapping);
603
604 return ret;
605}
606
607static int zonefs_inode_setattr(struct user_namespace *mnt_userns,
608 struct dentry *dentry, struct iattr *iattr)
609{
610 struct inode *inode = d_inode(dentry);
611 int ret;
612
613 if (unlikely(IS_IMMUTABLE(inode)))
614 return -EPERM;
615
616 ret = setattr_prepare(&init_user_ns, dentry, iattr);
617 if (ret)
618 return ret;
619
620 /*
621 * Since files and directories cannot be created nor deleted, do not
622 * allow setting any write attributes on the sub-directories grouping
623 * files by zone type.
624 */
625 if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) &&
626 (iattr->ia_mode & 0222))
627 return -EPERM;
628
629 if (((iattr->ia_valid & ATTR_UID) &&
630 !uid_eq(iattr->ia_uid, inode->i_uid)) ||
631 ((iattr->ia_valid & ATTR_GID) &&
632 !gid_eq(iattr->ia_gid, inode->i_gid))) {
633 ret = dquot_transfer(mnt_userns, inode, iattr);
634 if (ret)
635 return ret;
636 }
637
638 if (iattr->ia_valid & ATTR_SIZE) {
639 ret = zonefs_file_truncate(inode, iattr->ia_size);
640 if (ret)
641 return ret;
642 }
643
644 setattr_copy(&init_user_ns, inode, iattr);
645
646 return 0;
647}
648
649static const struct inode_operations zonefs_file_inode_operations = {
650 .setattr = zonefs_inode_setattr,
651};
652
653static int zonefs_file_fsync(struct file *file, loff_t start, loff_t end,
654 int datasync)
655{
656 struct inode *inode = file_inode(file);
657 int ret = 0;
658
659 if (unlikely(IS_IMMUTABLE(inode)))
660 return -EPERM;
661
662 /*
663 * Since only direct writes are allowed in sequential files, page cache
664 * flush is needed only for conventional zone files.
665 */
666 if (ZONEFS_I(inode)->i_ztype == ZONEFS_ZTYPE_CNV)
667 ret = file_write_and_wait_range(file, start, end);
668 if (!ret)
669 ret = blkdev_issue_flush(inode->i_sb->s_bdev);
670
671 if (ret)
672 zonefs_io_error(inode, true);
673
674 return ret;
675}
676
677static vm_fault_t zonefs_filemap_page_mkwrite(struct vm_fault *vmf)
678{
679 struct inode *inode = file_inode(vmf->vma->vm_file);
680 struct zonefs_inode_info *zi = ZONEFS_I(inode);
681 vm_fault_t ret;
682
683 if (unlikely(IS_IMMUTABLE(inode)))
684 return VM_FAULT_SIGBUS;
685
686 /*
687 * Sanity check: only conventional zone files can have shared
688 * writeable mappings.
689 */
690 if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV))
691 return VM_FAULT_NOPAGE;
692
693 sb_start_pagefault(inode->i_sb);
694 file_update_time(vmf->vma->vm_file);
695
696 /* Serialize against truncates */
697 filemap_invalidate_lock_shared(inode->i_mapping);
698 ret = iomap_page_mkwrite(vmf, &zonefs_write_iomap_ops);
699 filemap_invalidate_unlock_shared(inode->i_mapping);
700
701 sb_end_pagefault(inode->i_sb);
702 return ret;
703}
704
705static const struct vm_operations_struct zonefs_file_vm_ops = {
706 .fault = filemap_fault,
707 .map_pages = filemap_map_pages,
708 .page_mkwrite = zonefs_filemap_page_mkwrite,
709};
710
711static int zonefs_file_mmap(struct file *file, struct vm_area_struct *vma)
712{
713 /*
714 * Conventional zones accept random writes, so their files can support
715 * shared writable mappings. For sequential zone files, only read
716 * mappings are possible since there are no guarantees for write
717 * ordering between msync() and page cache writeback.
718 */
719 if (ZONEFS_I(file_inode(file))->i_ztype == ZONEFS_ZTYPE_SEQ &&
720 (vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
721 return -EINVAL;
722
723 file_accessed(file);
724 vma->vm_ops = &zonefs_file_vm_ops;
725
726 return 0;
727}
728
729static loff_t zonefs_file_llseek(struct file *file, loff_t offset, int whence)
730{
731 loff_t isize = i_size_read(file_inode(file));
732
733 /*
734 * Seeks are limited to below the zone size for conventional zones
735 * and below the zone write pointer for sequential zones. In both
736 * cases, this limit is the inode size.
737 */
738 return generic_file_llseek_size(file, offset, whence, isize, isize);
739}
740
741static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
742 int error, unsigned int flags)
743{
744 struct inode *inode = file_inode(iocb->ki_filp);
745 struct zonefs_inode_info *zi = ZONEFS_I(inode);
746
747 if (error) {
748 zonefs_io_error(inode, true);
749 return error;
750 }
751
752 if (size && zi->i_ztype != ZONEFS_ZTYPE_CNV) {
753 /*
754 * Note that we may be seeing completions out of order,
755 * but that is not a problem since a write completed
756 * successfully necessarily means that all preceding writes
757 * were also successful. So we can safely increase the inode
758 * size to the write end location.
759 */
760 mutex_lock(&zi->i_truncate_mutex);
761 if (i_size_read(inode) < iocb->ki_pos + size) {
762 zonefs_update_stats(inode, iocb->ki_pos + size);
763 zonefs_i_size_write(inode, iocb->ki_pos + size);
764 }
765 mutex_unlock(&zi->i_truncate_mutex);
766 }
767
768 return 0;
769}
770
771static const struct iomap_dio_ops zonefs_write_dio_ops = {
772 .end_io = zonefs_file_write_dio_end_io,
773};
774
775static ssize_t zonefs_file_dio_append(struct kiocb *iocb, struct iov_iter *from)
776{
777 struct inode *inode = file_inode(iocb->ki_filp);
778 struct zonefs_inode_info *zi = ZONEFS_I(inode);
779 struct block_device *bdev = inode->i_sb->s_bdev;
780 unsigned int max = bdev_max_zone_append_sectors(bdev);
781 struct bio *bio;
782 ssize_t size;
783 int nr_pages;
784 ssize_t ret;
785
786 max = ALIGN_DOWN(max << SECTOR_SHIFT, inode->i_sb->s_blocksize);
787 iov_iter_truncate(from, max);
788
789 nr_pages = iov_iter_npages(from, BIO_MAX_VECS);
790 if (!nr_pages)
791 return 0;
792
793 bio = bio_alloc(bdev, nr_pages,
794 REQ_OP_ZONE_APPEND | REQ_SYNC | REQ_IDLE, GFP_NOFS);
795 bio->bi_iter.bi_sector = zi->i_zsector;
796 bio->bi_ioprio = iocb->ki_ioprio;
797 if (iocb_is_dsync(iocb))
798 bio->bi_opf |= REQ_FUA;
799
800 ret = bio_iov_iter_get_pages(bio, from);
801 if (unlikely(ret))
802 goto out_release;
803
804 size = bio->bi_iter.bi_size;
805 task_io_account_write(size);
806
807 if (iocb->ki_flags & IOCB_HIPRI)
808 bio_set_polled(bio, iocb);
809
810 ret = submit_bio_wait(bio);
811
812 /*
813 * If the file zone was written underneath the file system, the zone
814 * write pointer may not be where we expect it to be, but the zone
815 * append write can still succeed. So check manually that we wrote where
816 * we intended to, that is, at zi->i_wpoffset.
817 */
818 if (!ret) {
819 sector_t wpsector =
820 zi->i_zsector + (zi->i_wpoffset >> SECTOR_SHIFT);
821
822 if (bio->bi_iter.bi_sector != wpsector) {
823 zonefs_warn(inode->i_sb,
824 "Corrupted write pointer %llu for zone at %llu\n",
825 wpsector, zi->i_zsector);
826 ret = -EIO;
827 }
828 }
829
830 zonefs_file_write_dio_end_io(iocb, size, ret, 0);
831 trace_zonefs_file_dio_append(inode, size, ret);
832
833out_release:
834 bio_release_pages(bio, false);
835 bio_put(bio);
836
837 if (ret >= 0) {
838 iocb->ki_pos += size;
839 return size;
840 }
841
842 return ret;
843}
844
845/*
846 * Do not exceed the LFS limits nor the file zone size. If pos is under the
847 * limit it becomes a short access. If it exceeds the limit, return -EFBIG.
848 */
849static loff_t zonefs_write_check_limits(struct file *file, loff_t pos,
850 loff_t count)
851{
852 struct inode *inode = file_inode(file);
853 struct zonefs_inode_info *zi = ZONEFS_I(inode);
854 loff_t limit = rlimit(RLIMIT_FSIZE);
855 loff_t max_size = zi->i_max_size;
856
857 if (limit != RLIM_INFINITY) {
858 if (pos >= limit) {
859 send_sig(SIGXFSZ, current, 0);
860 return -EFBIG;
861 }
862 count = min(count, limit - pos);
863 }
864
865 if (!(file->f_flags & O_LARGEFILE))
866 max_size = min_t(loff_t, MAX_NON_LFS, max_size);
867
868 if (unlikely(pos >= max_size))
869 return -EFBIG;
870
871 return min(count, max_size - pos);
872}
873
874static ssize_t zonefs_write_checks(struct kiocb *iocb, struct iov_iter *from)
875{
876 struct file *file = iocb->ki_filp;
877 struct inode *inode = file_inode(file);
878 struct zonefs_inode_info *zi = ZONEFS_I(inode);
879 loff_t count;
880
881 if (IS_SWAPFILE(inode))
882 return -ETXTBSY;
883
884 if (!iov_iter_count(from))
885 return 0;
886
887 if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
888 return -EINVAL;
889
890 if (iocb->ki_flags & IOCB_APPEND) {
891 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
892 return -EINVAL;
893 mutex_lock(&zi->i_truncate_mutex);
894 iocb->ki_pos = zi->i_wpoffset;
895 mutex_unlock(&zi->i_truncate_mutex);
896 }
897
898 count = zonefs_write_check_limits(file, iocb->ki_pos,
899 iov_iter_count(from));
900 if (count < 0)
901 return count;
902
903 iov_iter_truncate(from, count);
904 return iov_iter_count(from);
905}
906
907/*
908 * Handle direct writes. For sequential zone files, this is the only possible
909 * write path. For these files, check that the user is issuing writes
910 * sequentially from the end of the file. This code assumes that the block layer
911 * delivers write requests to the device in sequential order. This is always the
912 * case if a block IO scheduler implementing the ELEVATOR_F_ZBD_SEQ_WRITE
913 * elevator feature is being used (e.g. mq-deadline). The block layer always
914 * automatically select such an elevator for zoned block devices during the
915 * device initialization.
916 */
917static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
918{
919 struct inode *inode = file_inode(iocb->ki_filp);
920 struct zonefs_inode_info *zi = ZONEFS_I(inode);
921 struct super_block *sb = inode->i_sb;
922 bool sync = is_sync_kiocb(iocb);
923 bool append = false;
924 ssize_t ret, count;
925
926 /*
927 * For async direct IOs to sequential zone files, refuse IOCB_NOWAIT
928 * as this can cause write reordering (e.g. the first aio gets EAGAIN
929 * on the inode lock but the second goes through but is now unaligned).
930 */
931 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !sync &&
932 (iocb->ki_flags & IOCB_NOWAIT))
933 return -EOPNOTSUPP;
934
935 if (iocb->ki_flags & IOCB_NOWAIT) {
936 if (!inode_trylock(inode))
937 return -EAGAIN;
938 } else {
939 inode_lock(inode);
940 }
941
942 count = zonefs_write_checks(iocb, from);
943 if (count <= 0) {
944 ret = count;
945 goto inode_unlock;
946 }
947
948 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
949 ret = -EINVAL;
950 goto inode_unlock;
951 }
952
953 /* Enforce sequential writes (append only) in sequential zones */
954 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ) {
955 mutex_lock(&zi->i_truncate_mutex);
956 if (iocb->ki_pos != zi->i_wpoffset) {
957 mutex_unlock(&zi->i_truncate_mutex);
958 ret = -EINVAL;
959 goto inode_unlock;
960 }
961 mutex_unlock(&zi->i_truncate_mutex);
962 append = sync;
963 }
964
965 if (append)
966 ret = zonefs_file_dio_append(iocb, from);
967 else
968 ret = iomap_dio_rw(iocb, from, &zonefs_write_iomap_ops,
969 &zonefs_write_dio_ops, 0, NULL, 0);
970 if (zi->i_ztype == ZONEFS_ZTYPE_SEQ &&
971 (ret > 0 || ret == -EIOCBQUEUED)) {
972 if (ret > 0)
973 count = ret;
974
975 /*
976 * Update the zone write pointer offset assuming the write
977 * operation succeeded. If it did not, the error recovery path
978 * will correct it. Also do active seq file accounting.
979 */
980 mutex_lock(&zi->i_truncate_mutex);
981 zi->i_wpoffset += count;
982 zonefs_account_active(inode);
983 mutex_unlock(&zi->i_truncate_mutex);
984 }
985
986inode_unlock:
987 inode_unlock(inode);
988
989 return ret;
990}
991
992static ssize_t zonefs_file_buffered_write(struct kiocb *iocb,
993 struct iov_iter *from)
994{
995 struct inode *inode = file_inode(iocb->ki_filp);
996 struct zonefs_inode_info *zi = ZONEFS_I(inode);
997 ssize_t ret;
998
999 /*
1000 * Direct IO writes are mandatory for sequential zone files so that the
1001 * write IO issuing order is preserved.
1002 */
1003 if (zi->i_ztype != ZONEFS_ZTYPE_CNV)
1004 return -EIO;
1005
1006 if (iocb->ki_flags & IOCB_NOWAIT) {
1007 if (!inode_trylock(inode))
1008 return -EAGAIN;
1009 } else {
1010 inode_lock(inode);
1011 }
1012
1013 ret = zonefs_write_checks(iocb, from);
1014 if (ret <= 0)
1015 goto inode_unlock;
1016
1017 ret = iomap_file_buffered_write(iocb, from, &zonefs_write_iomap_ops);
1018 if (ret > 0)
1019 iocb->ki_pos += ret;
1020 else if (ret == -EIO)
1021 zonefs_io_error(inode, true);
1022
1023inode_unlock:
1024 inode_unlock(inode);
1025 if (ret > 0)
1026 ret = generic_write_sync(iocb, ret);
1027
1028 return ret;
1029}
1030
1031static ssize_t zonefs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
1032{
1033 struct inode *inode = file_inode(iocb->ki_filp);
1034
1035 if (unlikely(IS_IMMUTABLE(inode)))
1036 return -EPERM;
1037
1038 if (sb_rdonly(inode->i_sb))
1039 return -EROFS;
1040
1041 /* Write operations beyond the zone size are not allowed */
1042 if (iocb->ki_pos >= ZONEFS_I(inode)->i_max_size)
1043 return -EFBIG;
1044
1045 if (iocb->ki_flags & IOCB_DIRECT) {
1046 ssize_t ret = zonefs_file_dio_write(iocb, from);
1047 if (ret != -ENOTBLK)
1048 return ret;
1049 }
1050
1051 return zonefs_file_buffered_write(iocb, from);
1052}
1053
1054static int zonefs_file_read_dio_end_io(struct kiocb *iocb, ssize_t size,
1055 int error, unsigned int flags)
1056{
1057 if (error) {
1058 zonefs_io_error(file_inode(iocb->ki_filp), false);
1059 return error;
1060 }
1061
1062 return 0;
1063}
1064
1065static const struct iomap_dio_ops zonefs_read_dio_ops = {
1066 .end_io = zonefs_file_read_dio_end_io,
1067};
1068
1069static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1070{
1071 struct inode *inode = file_inode(iocb->ki_filp);
1072 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1073 struct super_block *sb = inode->i_sb;
1074 loff_t isize;
1075 ssize_t ret;
1076
1077 /* Offline zones cannot be read */
1078 if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777)))
1079 return -EPERM;
1080
1081 if (iocb->ki_pos >= zi->i_max_size)
1082 return 0;
1083
1084 if (iocb->ki_flags & IOCB_NOWAIT) {
1085 if (!inode_trylock_shared(inode))
1086 return -EAGAIN;
1087 } else {
1088 inode_lock_shared(inode);
1089 }
1090
1091 /* Limit read operations to written data */
1092 mutex_lock(&zi->i_truncate_mutex);
1093 isize = i_size_read(inode);
1094 if (iocb->ki_pos >= isize) {
1095 mutex_unlock(&zi->i_truncate_mutex);
1096 ret = 0;
1097 goto inode_unlock;
1098 }
1099 iov_iter_truncate(to, isize - iocb->ki_pos);
1100 mutex_unlock(&zi->i_truncate_mutex);
1101
1102 if (iocb->ki_flags & IOCB_DIRECT) {
1103 size_t count = iov_iter_count(to);
1104
1105 if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
1106 ret = -EINVAL;
1107 goto inode_unlock;
1108 }
1109 file_accessed(iocb->ki_filp);
1110 ret = iomap_dio_rw(iocb, to, &zonefs_read_iomap_ops,
1111 &zonefs_read_dio_ops, 0, NULL, 0);
1112 } else {
1113 ret = generic_file_read_iter(iocb, to);
1114 if (ret == -EIO)
1115 zonefs_io_error(inode, false);
1116 }
1117
1118inode_unlock:
1119 inode_unlock_shared(inode);
1120
1121 return ret;
1122}
1123
1124/*
1125 * Write open accounting is done only for sequential files.
1126 */
1127static inline bool zonefs_seq_file_need_wro(struct inode *inode,
1128 struct file *file)
1129{
1130 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1131
1132 if (zi->i_ztype != ZONEFS_ZTYPE_SEQ)
1133 return false;
1134
1135 if (!(file->f_mode & FMODE_WRITE))
1136 return false;
1137
1138 return true;
1139}
1140
1141static int zonefs_seq_file_write_open(struct inode *inode)
1142{
1143 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1144 int ret = 0;
1145
1146 mutex_lock(&zi->i_truncate_mutex);
1147
1148 if (!zi->i_wr_refcnt) {
1149 struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
1150 unsigned int wro = atomic_inc_return(&sbi->s_wro_seq_files);
1151
1152 if (sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1153
1154 if (sbi->s_max_wro_seq_files
1155 && wro > sbi->s_max_wro_seq_files) {
1156 atomic_dec(&sbi->s_wro_seq_files);
1157 ret = -EBUSY;
1158 goto unlock;
1159 }
1160
1161 if (i_size_read(inode) < zi->i_max_size) {
1162 ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
1163 if (ret) {
1164 atomic_dec(&sbi->s_wro_seq_files);
1165 goto unlock;
1166 }
1167 zi->i_flags |= ZONEFS_ZONE_OPEN;
1168 zonefs_account_active(inode);
1169 }
1170 }
1171 }
1172
1173 zi->i_wr_refcnt++;
1174
1175unlock:
1176 mutex_unlock(&zi->i_truncate_mutex);
1177
1178 return ret;
1179}
1180
1181static int zonefs_file_open(struct inode *inode, struct file *file)
1182{
1183 int ret;
1184
1185 ret = generic_file_open(inode, file);
1186 if (ret)
1187 return ret;
1188
1189 if (zonefs_seq_file_need_wro(inode, file))
1190 return zonefs_seq_file_write_open(inode);
1191
1192 return 0;
1193}
1194
1195static void zonefs_seq_file_write_close(struct inode *inode)
1196{
1197 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1198 struct super_block *sb = inode->i_sb;
1199 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1200 int ret = 0;
1201
1202 mutex_lock(&zi->i_truncate_mutex);
1203
1204 zi->i_wr_refcnt--;
1205 if (zi->i_wr_refcnt)
1206 goto unlock;
1207
1208 /*
1209 * The file zone may not be open anymore (e.g. the file was truncated to
1210 * its maximum size or it was fully written). For this case, we only
1211 * need to decrement the write open count.
1212 */
1213 if (zi->i_flags & ZONEFS_ZONE_OPEN) {
1214 ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_CLOSE);
1215 if (ret) {
1216 __zonefs_io_error(inode, false);
1217 /*
1218 * Leaving zones explicitly open may lead to a state
1219 * where most zones cannot be written (zone resources
1220 * exhausted). So take preventive action by remounting
1221 * read-only.
1222 */
1223 if (zi->i_flags & ZONEFS_ZONE_OPEN &&
1224 !(sb->s_flags & SB_RDONLY)) {
1225 zonefs_warn(sb,
1226 "closing zone at %llu failed %d\n",
1227 zi->i_zsector, ret);
1228 zonefs_warn(sb,
1229 "remounting filesystem read-only\n");
1230 sb->s_flags |= SB_RDONLY;
1231 }
1232 goto unlock;
1233 }
1234
1235 zi->i_flags &= ~ZONEFS_ZONE_OPEN;
1236 zonefs_account_active(inode);
1237 }
1238
1239 atomic_dec(&sbi->s_wro_seq_files);
1240
1241unlock:
1242 mutex_unlock(&zi->i_truncate_mutex);
1243}
1244
1245static int zonefs_file_release(struct inode *inode, struct file *file)
1246{
1247 /*
1248 * If we explicitly open a zone we must close it again as well, but the
1249 * zone management operation can fail (either due to an IO error or as
1250 * the zone has gone offline or read-only). Make sure we don't fail the
1251 * close(2) for user-space.
1252 */
1253 if (zonefs_seq_file_need_wro(inode, file))
1254 zonefs_seq_file_write_close(inode);
1255
1256 return 0;
1257}
1258
1259static const struct file_operations zonefs_file_operations = {
1260 .open = zonefs_file_open,
1261 .release = zonefs_file_release,
1262 .fsync = zonefs_file_fsync,
1263 .mmap = zonefs_file_mmap,
1264 .llseek = zonefs_file_llseek,
1265 .read_iter = zonefs_file_read_iter,
1266 .write_iter = zonefs_file_write_iter,
1267 .splice_read = generic_file_splice_read,
1268 .splice_write = iter_file_splice_write,
1269 .iopoll = iocb_bio_iopoll,
1270};
1271
1272static struct kmem_cache *zonefs_inode_cachep;
1273
1274static struct inode *zonefs_alloc_inode(struct super_block *sb)
1275{
1276 struct zonefs_inode_info *zi;
1277
1278 zi = alloc_inode_sb(sb, zonefs_inode_cachep, GFP_KERNEL);
1279 if (!zi)
1280 return NULL;
1281
1282 inode_init_once(&zi->i_vnode);
1283 mutex_init(&zi->i_truncate_mutex);
1284 zi->i_wr_refcnt = 0;
1285 zi->i_flags = 0;
1286
1287 return &zi->i_vnode;
1288}
1289
1290static void zonefs_free_inode(struct inode *inode)
1291{
1292 kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode));
1293}
1294
1295/*
1296 * File system stat.
1297 */
1298static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf)
1299{
1300 struct super_block *sb = dentry->d_sb;
1301 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1302 enum zonefs_ztype t;
1303
1304 buf->f_type = ZONEFS_MAGIC;
1305 buf->f_bsize = sb->s_blocksize;
1306 buf->f_namelen = ZONEFS_NAME_MAX;
1307
1308 spin_lock(&sbi->s_lock);
1309
1310 buf->f_blocks = sbi->s_blocks;
1311 if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks))
1312 buf->f_bfree = 0;
1313 else
1314 buf->f_bfree = buf->f_blocks - sbi->s_used_blocks;
1315 buf->f_bavail = buf->f_bfree;
1316
1317 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
1318 if (sbi->s_nr_files[t])
1319 buf->f_files += sbi->s_nr_files[t] + 1;
1320 }
1321 buf->f_ffree = 0;
1322
1323 spin_unlock(&sbi->s_lock);
1324
1325 buf->f_fsid = uuid_to_fsid(sbi->s_uuid.b);
1326
1327 return 0;
1328}
1329
1330enum {
1331 Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair,
1332 Opt_explicit_open, Opt_err,
1333};
1334
1335static const match_table_t tokens = {
1336 { Opt_errors_ro, "errors=remount-ro"},
1337 { Opt_errors_zro, "errors=zone-ro"},
1338 { Opt_errors_zol, "errors=zone-offline"},
1339 { Opt_errors_repair, "errors=repair"},
1340 { Opt_explicit_open, "explicit-open" },
1341 { Opt_err, NULL}
1342};
1343
1344static int zonefs_parse_options(struct super_block *sb, char *options)
1345{
1346 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1347 substring_t args[MAX_OPT_ARGS];
1348 char *p;
1349
1350 if (!options)
1351 return 0;
1352
1353 while ((p = strsep(&options, ",")) != NULL) {
1354 int token;
1355
1356 if (!*p)
1357 continue;
1358
1359 token = match_token(p, tokens, args);
1360 switch (token) {
1361 case Opt_errors_ro:
1362 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
1363 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO;
1364 break;
1365 case Opt_errors_zro:
1366 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
1367 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO;
1368 break;
1369 case Opt_errors_zol:
1370 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
1371 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL;
1372 break;
1373 case Opt_errors_repair:
1374 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK;
1375 sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR;
1376 break;
1377 case Opt_explicit_open:
1378 sbi->s_mount_opts |= ZONEFS_MNTOPT_EXPLICIT_OPEN;
1379 break;
1380 default:
1381 return -EINVAL;
1382 }
1383 }
1384
1385 return 0;
1386}
1387
1388static int zonefs_show_options(struct seq_file *seq, struct dentry *root)
1389{
1390 struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb);
1391
1392 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO)
1393 seq_puts(seq, ",errors=remount-ro");
1394 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO)
1395 seq_puts(seq, ",errors=zone-ro");
1396 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL)
1397 seq_puts(seq, ",errors=zone-offline");
1398 if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR)
1399 seq_puts(seq, ",errors=repair");
1400
1401 return 0;
1402}
1403
1404static int zonefs_remount(struct super_block *sb, int *flags, char *data)
1405{
1406 sync_filesystem(sb);
1407
1408 return zonefs_parse_options(sb, data);
1409}
1410
1411static const struct super_operations zonefs_sops = {
1412 .alloc_inode = zonefs_alloc_inode,
1413 .free_inode = zonefs_free_inode,
1414 .statfs = zonefs_statfs,
1415 .remount_fs = zonefs_remount,
1416 .show_options = zonefs_show_options,
1417};
1418
1419static const struct inode_operations zonefs_dir_inode_operations = {
1420 .lookup = simple_lookup,
1421 .setattr = zonefs_inode_setattr,
1422};
1423
1424static void zonefs_init_dir_inode(struct inode *parent, struct inode *inode,
1425 enum zonefs_ztype type)
1426{
1427 struct super_block *sb = parent->i_sb;
1428
1429 inode->i_ino = bdev_nr_zones(sb->s_bdev) + type + 1;
1430 inode_init_owner(&init_user_ns, inode, parent, S_IFDIR | 0555);
1431 inode->i_op = &zonefs_dir_inode_operations;
1432 inode->i_fop = &simple_dir_operations;
1433 set_nlink(inode, 2);
1434 inc_nlink(parent);
1435}
1436
1437static int zonefs_init_file_inode(struct inode *inode, struct blk_zone *zone,
1438 enum zonefs_ztype type)
1439{
1440 struct super_block *sb = inode->i_sb;
1441 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1442 struct zonefs_inode_info *zi = ZONEFS_I(inode);
1443 int ret = 0;
1444
1445 inode->i_ino = zone->start >> sbi->s_zone_sectors_shift;
1446 inode->i_mode = S_IFREG | sbi->s_perm;
1447
1448 zi->i_ztype = type;
1449 zi->i_zsector = zone->start;
1450 zi->i_zone_size = zone->len << SECTOR_SHIFT;
1451 if (zi->i_zone_size > bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT &&
1452 !(sbi->s_features & ZONEFS_F_AGGRCNV)) {
1453 zonefs_err(sb,
1454 "zone size %llu doesn't match device's zone sectors %llu\n",
1455 zi->i_zone_size,
1456 bdev_zone_sectors(sb->s_bdev) << SECTOR_SHIFT);
1457 return -EINVAL;
1458 }
1459
1460 zi->i_max_size = min_t(loff_t, MAX_LFS_FILESIZE,
1461 zone->capacity << SECTOR_SHIFT);
1462 zi->i_wpoffset = zonefs_check_zone_condition(inode, zone, true, true);
1463
1464 inode->i_uid = sbi->s_uid;
1465 inode->i_gid = sbi->s_gid;
1466 inode->i_size = zi->i_wpoffset;
1467 inode->i_blocks = zi->i_max_size >> SECTOR_SHIFT;
1468
1469 inode->i_op = &zonefs_file_inode_operations;
1470 inode->i_fop = &zonefs_file_operations;
1471 inode->i_mapping->a_ops = &zonefs_file_aops;
1472
1473 sb->s_maxbytes = max(zi->i_max_size, sb->s_maxbytes);
1474 sbi->s_blocks += zi->i_max_size >> sb->s_blocksize_bits;
1475 sbi->s_used_blocks += zi->i_wpoffset >> sb->s_blocksize_bits;
1476
1477 mutex_lock(&zi->i_truncate_mutex);
1478
1479 /*
1480 * For sequential zones, make sure that any open zone is closed first
1481 * to ensure that the initial number of open zones is 0, in sync with
1482 * the open zone accounting done when the mount option
1483 * ZONEFS_MNTOPT_EXPLICIT_OPEN is used.
1484 */
1485 if (type == ZONEFS_ZTYPE_SEQ &&
1486 (zone->cond == BLK_ZONE_COND_IMP_OPEN ||
1487 zone->cond == BLK_ZONE_COND_EXP_OPEN)) {
1488 ret = zonefs_zone_mgmt(inode, REQ_OP_ZONE_CLOSE);
1489 if (ret)
1490 goto unlock;
1491 }
1492
1493 zonefs_account_active(inode);
1494
1495unlock:
1496 mutex_unlock(&zi->i_truncate_mutex);
1497
1498 return ret;
1499}
1500
1501static struct dentry *zonefs_create_inode(struct dentry *parent,
1502 const char *name, struct blk_zone *zone,
1503 enum zonefs_ztype type)
1504{
1505 struct inode *dir = d_inode(parent);
1506 struct dentry *dentry;
1507 struct inode *inode;
1508 int ret = -ENOMEM;
1509
1510 dentry = d_alloc_name(parent, name);
1511 if (!dentry)
1512 return ERR_PTR(ret);
1513
1514 inode = new_inode(parent->d_sb);
1515 if (!inode)
1516 goto dput;
1517
1518 inode->i_ctime = inode->i_mtime = inode->i_atime = dir->i_ctime;
1519 if (zone) {
1520 ret = zonefs_init_file_inode(inode, zone, type);
1521 if (ret) {
1522 iput(inode);
1523 goto dput;
1524 }
1525 } else {
1526 zonefs_init_dir_inode(dir, inode, type);
1527 }
1528
1529 d_add(dentry, inode);
1530 dir->i_size++;
1531
1532 return dentry;
1533
1534dput:
1535 dput(dentry);
1536
1537 return ERR_PTR(ret);
1538}
1539
1540struct zonefs_zone_data {
1541 struct super_block *sb;
1542 unsigned int nr_zones[ZONEFS_ZTYPE_MAX];
1543 struct blk_zone *zones;
1544};
1545
1546/*
1547 * Create a zone group and populate it with zone files.
1548 */
1549static int zonefs_create_zgroup(struct zonefs_zone_data *zd,
1550 enum zonefs_ztype type)
1551{
1552 struct super_block *sb = zd->sb;
1553 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1554 struct blk_zone *zone, *next, *end;
1555 const char *zgroup_name;
1556 char *file_name;
1557 struct dentry *dir, *dent;
1558 unsigned int n = 0;
1559 int ret;
1560
1561 /* If the group is empty, there is nothing to do */
1562 if (!zd->nr_zones[type])
1563 return 0;
1564
1565 file_name = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL);
1566 if (!file_name)
1567 return -ENOMEM;
1568
1569 if (type == ZONEFS_ZTYPE_CNV)
1570 zgroup_name = "cnv";
1571 else
1572 zgroup_name = "seq";
1573
1574 dir = zonefs_create_inode(sb->s_root, zgroup_name, NULL, type);
1575 if (IS_ERR(dir)) {
1576 ret = PTR_ERR(dir);
1577 goto free;
1578 }
1579
1580 /*
1581 * The first zone contains the super block: skip it.
1582 */
1583 end = zd->zones + bdev_nr_zones(sb->s_bdev);
1584 for (zone = &zd->zones[1]; zone < end; zone = next) {
1585
1586 next = zone + 1;
1587 if (zonefs_zone_type(zone) != type)
1588 continue;
1589
1590 /*
1591 * For conventional zones, contiguous zones can be aggregated
1592 * together to form larger files. Note that this overwrites the
1593 * length of the first zone of the set of contiguous zones
1594 * aggregated together. If one offline or read-only zone is
1595 * found, assume that all zones aggregated have the same
1596 * condition.
1597 */
1598 if (type == ZONEFS_ZTYPE_CNV &&
1599 (sbi->s_features & ZONEFS_F_AGGRCNV)) {
1600 for (; next < end; next++) {
1601 if (zonefs_zone_type(next) != type)
1602 break;
1603 zone->len += next->len;
1604 zone->capacity += next->capacity;
1605 if (next->cond == BLK_ZONE_COND_READONLY &&
1606 zone->cond != BLK_ZONE_COND_OFFLINE)
1607 zone->cond = BLK_ZONE_COND_READONLY;
1608 else if (next->cond == BLK_ZONE_COND_OFFLINE)
1609 zone->cond = BLK_ZONE_COND_OFFLINE;
1610 }
1611 if (zone->capacity != zone->len) {
1612 zonefs_err(sb, "Invalid conventional zone capacity\n");
1613 ret = -EINVAL;
1614 goto free;
1615 }
1616 }
1617
1618 /*
1619 * Use the file number within its group as file name.
1620 */
1621 snprintf(file_name, ZONEFS_NAME_MAX - 1, "%u", n);
1622 dent = zonefs_create_inode(dir, file_name, zone, type);
1623 if (IS_ERR(dent)) {
1624 ret = PTR_ERR(dent);
1625 goto free;
1626 }
1627
1628 n++;
1629 }
1630
1631 zonefs_info(sb, "Zone group \"%s\" has %u file%s\n",
1632 zgroup_name, n, n > 1 ? "s" : "");
1633
1634 sbi->s_nr_files[type] = n;
1635 ret = 0;
1636
1637free:
1638 kfree(file_name);
1639
1640 return ret;
1641}
1642
1643static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx,
1644 void *data)
1645{
1646 struct zonefs_zone_data *zd = data;
1647
1648 /*
1649 * Count the number of usable zones: the first zone at index 0 contains
1650 * the super block and is ignored.
1651 */
1652 switch (zone->type) {
1653 case BLK_ZONE_TYPE_CONVENTIONAL:
1654 zone->wp = zone->start + zone->len;
1655 if (idx)
1656 zd->nr_zones[ZONEFS_ZTYPE_CNV]++;
1657 break;
1658 case BLK_ZONE_TYPE_SEQWRITE_REQ:
1659 case BLK_ZONE_TYPE_SEQWRITE_PREF:
1660 if (idx)
1661 zd->nr_zones[ZONEFS_ZTYPE_SEQ]++;
1662 break;
1663 default:
1664 zonefs_err(zd->sb, "Unsupported zone type 0x%x\n",
1665 zone->type);
1666 return -EIO;
1667 }
1668
1669 memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone));
1670
1671 return 0;
1672}
1673
1674static int zonefs_get_zone_info(struct zonefs_zone_data *zd)
1675{
1676 struct block_device *bdev = zd->sb->s_bdev;
1677 int ret;
1678
1679 zd->zones = kvcalloc(bdev_nr_zones(bdev), sizeof(struct blk_zone),
1680 GFP_KERNEL);
1681 if (!zd->zones)
1682 return -ENOMEM;
1683
1684 /* Get zones information from the device */
1685 ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES,
1686 zonefs_get_zone_info_cb, zd);
1687 if (ret < 0) {
1688 zonefs_err(zd->sb, "Zone report failed %d\n", ret);
1689 return ret;
1690 }
1691
1692 if (ret != bdev_nr_zones(bdev)) {
1693 zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n",
1694 ret, bdev_nr_zones(bdev));
1695 return -EIO;
1696 }
1697
1698 return 0;
1699}
1700
1701static inline void zonefs_cleanup_zone_info(struct zonefs_zone_data *zd)
1702{
1703 kvfree(zd->zones);
1704}
1705
1706/*
1707 * Read super block information from the device.
1708 */
1709static int zonefs_read_super(struct super_block *sb)
1710{
1711 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1712 struct zonefs_super *super;
1713 u32 crc, stored_crc;
1714 struct page *page;
1715 struct bio_vec bio_vec;
1716 struct bio bio;
1717 int ret;
1718
1719 page = alloc_page(GFP_KERNEL);
1720 if (!page)
1721 return -ENOMEM;
1722
1723 bio_init(&bio, sb->s_bdev, &bio_vec, 1, REQ_OP_READ);
1724 bio.bi_iter.bi_sector = 0;
1725 bio_add_page(&bio, page, PAGE_SIZE, 0);
1726
1727 ret = submit_bio_wait(&bio);
1728 if (ret)
1729 goto free_page;
1730
1731 super = page_address(page);
1732
1733 ret = -EINVAL;
1734 if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC)
1735 goto free_page;
1736
1737 stored_crc = le32_to_cpu(super->s_crc);
1738 super->s_crc = 0;
1739 crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super));
1740 if (crc != stored_crc) {
1741 zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)",
1742 crc, stored_crc);
1743 goto free_page;
1744 }
1745
1746 sbi->s_features = le64_to_cpu(super->s_features);
1747 if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) {
1748 zonefs_err(sb, "Unknown features set 0x%llx\n",
1749 sbi->s_features);
1750 goto free_page;
1751 }
1752
1753 if (sbi->s_features & ZONEFS_F_UID) {
1754 sbi->s_uid = make_kuid(current_user_ns(),
1755 le32_to_cpu(super->s_uid));
1756 if (!uid_valid(sbi->s_uid)) {
1757 zonefs_err(sb, "Invalid UID feature\n");
1758 goto free_page;
1759 }
1760 }
1761
1762 if (sbi->s_features & ZONEFS_F_GID) {
1763 sbi->s_gid = make_kgid(current_user_ns(),
1764 le32_to_cpu(super->s_gid));
1765 if (!gid_valid(sbi->s_gid)) {
1766 zonefs_err(sb, "Invalid GID feature\n");
1767 goto free_page;
1768 }
1769 }
1770
1771 if (sbi->s_features & ZONEFS_F_PERM)
1772 sbi->s_perm = le32_to_cpu(super->s_perm);
1773
1774 if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) {
1775 zonefs_err(sb, "Reserved area is being used\n");
1776 goto free_page;
1777 }
1778
1779 import_uuid(&sbi->s_uuid, super->s_uuid);
1780 ret = 0;
1781
1782free_page:
1783 __free_page(page);
1784
1785 return ret;
1786}
1787
1788/*
1789 * Check that the device is zoned. If it is, get the list of zones and create
1790 * sub-directories and files according to the device zone configuration and
1791 * format options.
1792 */
1793static int zonefs_fill_super(struct super_block *sb, void *data, int silent)
1794{
1795 struct zonefs_zone_data zd;
1796 struct zonefs_sb_info *sbi;
1797 struct inode *inode;
1798 enum zonefs_ztype t;
1799 int ret;
1800
1801 if (!bdev_is_zoned(sb->s_bdev)) {
1802 zonefs_err(sb, "Not a zoned block device\n");
1803 return -EINVAL;
1804 }
1805
1806 /*
1807 * Initialize super block information: the maximum file size is updated
1808 * when the zone files are created so that the format option
1809 * ZONEFS_F_AGGRCNV which increases the maximum file size of a file
1810 * beyond the zone size is taken into account.
1811 */
1812 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
1813 if (!sbi)
1814 return -ENOMEM;
1815
1816 spin_lock_init(&sbi->s_lock);
1817 sb->s_fs_info = sbi;
1818 sb->s_magic = ZONEFS_MAGIC;
1819 sb->s_maxbytes = 0;
1820 sb->s_op = &zonefs_sops;
1821 sb->s_time_gran = 1;
1822
1823 /*
1824 * The block size is set to the device zone write granularity to ensure
1825 * that write operations are always aligned according to the device
1826 * interface constraints.
1827 */
1828 sb_set_blocksize(sb, bdev_zone_write_granularity(sb->s_bdev));
1829 sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev));
1830 sbi->s_uid = GLOBAL_ROOT_UID;
1831 sbi->s_gid = GLOBAL_ROOT_GID;
1832 sbi->s_perm = 0640;
1833 sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO;
1834
1835 atomic_set(&sbi->s_wro_seq_files, 0);
1836 sbi->s_max_wro_seq_files = bdev_max_open_zones(sb->s_bdev);
1837 atomic_set(&sbi->s_active_seq_files, 0);
1838 sbi->s_max_active_seq_files = bdev_max_active_zones(sb->s_bdev);
1839
1840 ret = zonefs_read_super(sb);
1841 if (ret)
1842 return ret;
1843
1844 ret = zonefs_parse_options(sb, data);
1845 if (ret)
1846 return ret;
1847
1848 memset(&zd, 0, sizeof(struct zonefs_zone_data));
1849 zd.sb = sb;
1850 ret = zonefs_get_zone_info(&zd);
1851 if (ret)
1852 goto cleanup;
1853
1854 ret = zonefs_sysfs_register(sb);
1855 if (ret)
1856 goto cleanup;
1857
1858 zonefs_info(sb, "Mounting %u zones", bdev_nr_zones(sb->s_bdev));
1859
1860 if (!sbi->s_max_wro_seq_files &&
1861 !sbi->s_max_active_seq_files &&
1862 sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
1863 zonefs_info(sb,
1864 "No open and active zone limits. Ignoring explicit_open mount option\n");
1865 sbi->s_mount_opts &= ~ZONEFS_MNTOPT_EXPLICIT_OPEN;
1866 }
1867
1868 /* Create root directory inode */
1869 ret = -ENOMEM;
1870 inode = new_inode(sb);
1871 if (!inode)
1872 goto cleanup;
1873
1874 inode->i_ino = bdev_nr_zones(sb->s_bdev);
1875 inode->i_mode = S_IFDIR | 0555;
1876 inode->i_ctime = inode->i_mtime = inode->i_atime = current_time(inode);
1877 inode->i_op = &zonefs_dir_inode_operations;
1878 inode->i_fop = &simple_dir_operations;
1879 set_nlink(inode, 2);
1880
1881 sb->s_root = d_make_root(inode);
1882 if (!sb->s_root)
1883 goto cleanup;
1884
1885 /* Create and populate files in zone groups directories */
1886 for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) {
1887 ret = zonefs_create_zgroup(&zd, t);
1888 if (ret)
1889 break;
1890 }
1891
1892cleanup:
1893 zonefs_cleanup_zone_info(&zd);
1894
1895 return ret;
1896}
1897
1898static struct dentry *zonefs_mount(struct file_system_type *fs_type,
1899 int flags, const char *dev_name, void *data)
1900{
1901 return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super);
1902}
1903
1904static void zonefs_kill_super(struct super_block *sb)
1905{
1906 struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
1907
1908 if (sb->s_root)
1909 d_genocide(sb->s_root);
1910
1911 zonefs_sysfs_unregister(sb);
1912 kill_block_super(sb);
1913 kfree(sbi);
1914}
1915
1916/*
1917 * File system definition and registration.
1918 */
1919static struct file_system_type zonefs_type = {
1920 .owner = THIS_MODULE,
1921 .name = "zonefs",
1922 .mount = zonefs_mount,
1923 .kill_sb = zonefs_kill_super,
1924 .fs_flags = FS_REQUIRES_DEV,
1925};
1926
1927static int __init zonefs_init_inodecache(void)
1928{
1929 zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache",
1930 sizeof(struct zonefs_inode_info), 0,
1931 (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT),
1932 NULL);
1933 if (zonefs_inode_cachep == NULL)
1934 return -ENOMEM;
1935 return 0;
1936}
1937
1938static void zonefs_destroy_inodecache(void)
1939{
1940 /*
1941 * Make sure all delayed rcu free inodes are flushed before we
1942 * destroy the inode cache.
1943 */
1944 rcu_barrier();
1945 kmem_cache_destroy(zonefs_inode_cachep);
1946}
1947
1948static int __init zonefs_init(void)
1949{
1950 int ret;
1951
1952 BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE);
1953
1954 ret = zonefs_init_inodecache();
1955 if (ret)
1956 return ret;
1957
1958 ret = zonefs_sysfs_init();
1959 if (ret)
1960 goto destroy_inodecache;
1961
1962 ret = register_filesystem(&zonefs_type);
1963 if (ret)
1964 goto sysfs_exit;
1965
1966 return 0;
1967
1968sysfs_exit:
1969 zonefs_sysfs_exit();
1970destroy_inodecache:
1971 zonefs_destroy_inodecache();
1972
1973 return ret;
1974}
1975
1976static void __exit zonefs_exit(void)
1977{
1978 unregister_filesystem(&zonefs_type);
1979 zonefs_sysfs_exit();
1980 zonefs_destroy_inodecache();
1981}
1982
1983MODULE_AUTHOR("Damien Le Moal");
1984MODULE_DESCRIPTION("Zone file system for zoned block devices");
1985MODULE_LICENSE("GPL");
1986MODULE_ALIAS_FS("zonefs");
1987module_init(zonefs_init);
1988module_exit(zonefs_exit);