file.c

  1// SPDX-License-Identifier: GPL-2.0
  2/*
  3 * Simple file system for zoned block devices exposing zones as files.
  4 *
  5 * Copyright (C) 2022 Western Digital Corporation or its affiliates.
  6 */
  7#include <linux/module.h>
  8#include <linux/pagemap.h>
  9#include <linux/iomap.h>
 10#include <linux/init.h>
 11#include <linux/slab.h>
 12#include <linux/blkdev.h>
 13#include <linux/statfs.h>
 14#include <linux/writeback.h>
 15#include <linux/quotaops.h>
 16#include <linux/seq_file.h>
 17#include <linux/parser.h>
 18#include <linux/uio.h>
 19#include <linux/mman.h>
 20#include <linux/sched/mm.h>
 21#include <linux/task_io_accounting_ops.h>
 22
 23#include "zonefs.h"
 24
 25#include "trace.h"
 26
 27static int zonefs_read_iomap_begin(struct inode *inode, loff_t offset,
 28				   loff_t length, unsigned int flags,
 29				   struct iomap *iomap, struct iomap *srcmap)
 30{
 31	struct zonefs_inode_info *zi = ZONEFS_I(inode);
 32	struct zonefs_zone *z = zonefs_inode_zone(inode);
 33	struct super_block *sb = inode->i_sb;
 34	loff_t isize;
 35
 36	/*
 37	 * All blocks are always mapped below EOF. If reading past EOF,
 38	 * act as if there is a hole up to the file maximum size.
 39	 */
 40	mutex_lock(&zi->i_truncate_mutex);
 41	iomap->bdev = inode->i_sb->s_bdev;
 42	iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize);
 43	isize = i_size_read(inode);
 44	if (iomap->offset >= isize) {
 45		iomap->type = IOMAP_HOLE;
 46		iomap->addr = IOMAP_NULL_ADDR;
 47		iomap->length = length;
 48	} else {
 49		iomap->type = IOMAP_MAPPED;
 50		iomap->addr = (z->z_sector << SECTOR_SHIFT) + iomap->offset;
 51		iomap->length = isize - iomap->offset;
 52	}
 53	mutex_unlock(&zi->i_truncate_mutex);
 54
 55	trace_zonefs_iomap_begin(inode, iomap);
 56
 57	return 0;
 58}
 59
 60static const struct iomap_ops zonefs_read_iomap_ops = {
 61	.iomap_begin	= zonefs_read_iomap_begin,
 62};
 63
 64static int zonefs_write_iomap_begin(struct inode *inode, loff_t offset,
 65				    loff_t length, unsigned int flags,
 66				    struct iomap *iomap, struct iomap *srcmap)
 67{
 68	struct zonefs_inode_info *zi = ZONEFS_I(inode);
 69	struct zonefs_zone *z = zonefs_inode_zone(inode);
 70	struct super_block *sb = inode->i_sb;
 71	loff_t isize;
 72
 73	/* All write I/Os should always be within the file maximum size */
 74	if (WARN_ON_ONCE(offset + length > z->z_capacity))
 75		return -EIO;
 76
 77	/*
 78	 * Sequential zones can only accept direct writes. This is already
 79	 * checked when writes are issued, so warn if we see a page writeback
 80	 * operation.
 81	 */
 82	if (WARN_ON_ONCE(zonefs_zone_is_seq(z) && !(flags & IOMAP_DIRECT)))
 83		return -EIO;
 84
 85	/*
 86	 * For conventional zones, all blocks are always mapped. For sequential
 87	 * zones, all blocks after always mapped below the inode size (zone
 88	 * write pointer) and unwritten beyond.
 89	 */
 90	mutex_lock(&zi->i_truncate_mutex);
 91	iomap->bdev = inode->i_sb->s_bdev;
 92	iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize);
 93	iomap->addr = (z->z_sector << SECTOR_SHIFT) + iomap->offset;
 94	isize = i_size_read(inode);
 95	if (iomap->offset >= isize) {
 96		iomap->type = IOMAP_UNWRITTEN;
 97		iomap->length = z->z_capacity - iomap->offset;
 98	} else {
 99		iomap->type = IOMAP_MAPPED;
100		iomap->length = isize - iomap->offset;
101	}
102	mutex_unlock(&zi->i_truncate_mutex);
103
104	trace_zonefs_iomap_begin(inode, iomap);
105
106	return 0;
107}
108
109static const struct iomap_ops zonefs_write_iomap_ops = {
110	.iomap_begin	= zonefs_write_iomap_begin,
111};
112
113static int zonefs_read_folio(struct file *unused, struct folio *folio)
114{
115	iomap_bio_read_folio(folio, &zonefs_read_iomap_ops);
116	return 0;
117}
118
119static void zonefs_readahead(struct readahead_control *rac)
120{
121	iomap_bio_readahead(rac, &zonefs_read_iomap_ops);
122}
123
124/*
125 * Map blocks for page writeback. This is used only on conventional zone files,
126 * which implies that the page range can only be within the fixed inode size.
127 */
128static ssize_t zonefs_writeback_range(struct iomap_writepage_ctx *wpc,
129		struct folio *folio, u64 offset, unsigned len, u64 end_pos)
130{
131	struct zonefs_zone *z = zonefs_inode_zone(wpc->inode);
132
133	if (WARN_ON_ONCE(zonefs_zone_is_seq(z)))
134		return -EIO;
135	if (WARN_ON_ONCE(offset >= i_size_read(wpc->inode)))
136		return -EIO;
137
138	/* If the mapping is already OK, nothing needs to be done */
139	if (offset < wpc->iomap.offset ||
140	    offset >= wpc->iomap.offset + wpc->iomap.length) {
141		int error;
142
143		error = zonefs_write_iomap_begin(wpc->inode, offset,
144				z->z_capacity - offset, IOMAP_WRITE,
145				&wpc->iomap, NULL);
146		if (error)
147			return error;
148	}
149
150	return iomap_add_to_ioend(wpc, folio, offset, end_pos, len);
151}
152
153static const struct iomap_writeback_ops zonefs_writeback_ops = {
154	.writeback_range	= zonefs_writeback_range,
155	.writeback_submit	= iomap_ioend_writeback_submit,
156};
157
158static int zonefs_writepages(struct address_space *mapping,
159			     struct writeback_control *wbc)
160{
161	struct iomap_writepage_ctx wpc = {
162		.inode		= mapping->host,
163		.wbc		= wbc,
164		.ops		= &zonefs_writeback_ops,
165	};
166
167	return iomap_writepages(&wpc);
168}
169
170static int zonefs_swap_activate(struct swap_info_struct *sis,
171				struct file *swap_file, sector_t *span)
172{
173	struct inode *inode = file_inode(swap_file);
174
175	if (zonefs_inode_is_seq(inode)) {
176		zonefs_err(inode->i_sb,
177			   "swap file: not a conventional zone file\n");
178		return -EINVAL;
179	}
180
181	return iomap_swapfile_activate(sis, swap_file, span,
182				       &zonefs_read_iomap_ops);
183}
184
185const struct address_space_operations zonefs_file_aops = {
186	.read_folio		= zonefs_read_folio,
187	.readahead		= zonefs_readahead,
188	.writepages		= zonefs_writepages,
189	.dirty_folio		= iomap_dirty_folio,
190	.release_folio		= iomap_release_folio,
191	.invalidate_folio	= iomap_invalidate_folio,
192	.migrate_folio		= filemap_migrate_folio,
193	.is_partially_uptodate	= iomap_is_partially_uptodate,
194	.error_remove_folio	= generic_error_remove_folio,
195	.swap_activate		= zonefs_swap_activate,
196};
197
198int zonefs_file_truncate(struct inode *inode, loff_t isize)
199{
200	struct zonefs_inode_info *zi = ZONEFS_I(inode);
201	struct zonefs_zone *z = zonefs_inode_zone(inode);
202	loff_t old_isize;
203	enum req_op op;
204	int ret = 0;
205
206	/*
207	 * Only sequential zone files can be truncated and truncation is allowed
208	 * only down to a 0 size, which is equivalent to a zone reset, and to
209	 * the maximum file size, which is equivalent to a zone finish.
210	 */
211	if (!zonefs_zone_is_seq(z))
212		return -EPERM;
213
214	if (!isize)
215		op = REQ_OP_ZONE_RESET;
216	else if (isize == z->z_capacity)
217		op = REQ_OP_ZONE_FINISH;
218	else
219		return -EPERM;
220
221	inode_dio_wait(inode);
222
223	/* Serialize against page faults */
224	filemap_invalidate_lock(inode->i_mapping);
225
226	/* Serialize against zonefs_iomap_begin() */
227	mutex_lock(&zi->i_truncate_mutex);
228
229	old_isize = i_size_read(inode);
230	if (isize == old_isize)
231		goto unlock;
232
233	ret = zonefs_inode_zone_mgmt(inode, op);
234	if (ret)
235		goto unlock;
236
237	/*
238	 * If the mount option ZONEFS_MNTOPT_EXPLICIT_OPEN is set,
239	 * take care of open zones.
240	 */
241	if (z->z_flags & ZONEFS_ZONE_OPEN) {
242		/*
243		 * Truncating a zone to EMPTY or FULL is the equivalent of
244		 * closing the zone. For a truncation to 0, we need to
245		 * re-open the zone to ensure new writes can be processed.
246		 * For a truncation to the maximum file size, the zone is
247		 * closed and writes cannot be accepted anymore, so clear
248		 * the open flag.
249		 */
250		if (!isize)
251			ret = zonefs_inode_zone_mgmt(inode, REQ_OP_ZONE_OPEN);
252		else
253			z->z_flags &= ~ZONEFS_ZONE_OPEN;
254	}
255
256	zonefs_update_stats(inode, isize);
257	truncate_setsize(inode, isize);
258	z->z_wpoffset = isize;
259	zonefs_inode_account_active(inode);
260
261unlock:
262	mutex_unlock(&zi->i_truncate_mutex);
263	filemap_invalidate_unlock(inode->i_mapping);
264
265	return ret;
266}
267
268static int zonefs_file_fsync(struct file *file, loff_t start, loff_t end,
269			     int datasync)
270{
271	struct inode *inode = file_inode(file);
272	int ret = 0;
273
274	if (unlikely(IS_IMMUTABLE(inode)))
275		return -EPERM;
276
277	/*
278	 * Since only direct writes are allowed in sequential files, page cache
279	 * flush is needed only for conventional zone files.
280	 */
281	if (zonefs_inode_is_cnv(inode))
282		ret = file_write_and_wait_range(file, start, end);
283	if (!ret)
284		ret = blkdev_issue_flush(inode->i_sb->s_bdev);
285
286	if (ret)
287		zonefs_io_error(inode, true);
288
289	return ret;
290}
291
292static vm_fault_t zonefs_filemap_page_mkwrite(struct vm_fault *vmf)
293{
294	struct inode *inode = file_inode(vmf->vma->vm_file);
295	vm_fault_t ret;
296
297	if (unlikely(IS_IMMUTABLE(inode)))
298		return VM_FAULT_SIGBUS;
299
300	/*
301	 * Sanity check: only conventional zone files can have shared
302	 * writeable mappings.
303	 */
304	if (zonefs_inode_is_seq(inode))
305		return VM_FAULT_NOPAGE;
306
307	sb_start_pagefault(inode->i_sb);
308	file_update_time(vmf->vma->vm_file);
309
310	/* Serialize against truncates */
311	filemap_invalidate_lock_shared(inode->i_mapping);
312	ret = iomap_page_mkwrite(vmf, &zonefs_write_iomap_ops, NULL);
313	filemap_invalidate_unlock_shared(inode->i_mapping);
314
315	sb_end_pagefault(inode->i_sb);
316	return ret;
317}
318
319static const struct vm_operations_struct zonefs_file_vm_ops = {
320	.fault		= filemap_fault,
321	.map_pages	= filemap_map_pages,
322	.page_mkwrite	= zonefs_filemap_page_mkwrite,
323};
324
325static int zonefs_file_mmap_prepare(struct vm_area_desc *desc)
326{
327	struct file *file = desc->file;
328
329	/*
330	 * Conventional zones accept random writes, so their files can support
331	 * shared writable mappings. For sequential zone files, only read
332	 * mappings are possible since there are no guarantees for write
333	 * ordering between msync() and page cache writeback.
334	 */
335	if (zonefs_inode_is_seq(file_inode(file)) &&
336	    (desc->vm_flags & VM_SHARED) && (desc->vm_flags & VM_MAYWRITE))
337		return -EINVAL;
338
339	file_accessed(file);
340	desc->vm_ops = &zonefs_file_vm_ops;
341
342	return 0;
343}
344
345static loff_t zonefs_file_llseek(struct file *file, loff_t offset, int whence)
346{
347	loff_t isize = i_size_read(file_inode(file));
348
349	/*
350	 * Seeks are limited to below the zone size for conventional zones
351	 * and below the zone write pointer for sequential zones. In both
352	 * cases, this limit is the inode size.
353	 */
354	return generic_file_llseek_size(file, offset, whence, isize, isize);
355}
356
357static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size,
358					int error, unsigned int flags)
359{
360	struct inode *inode = file_inode(iocb->ki_filp);
361	struct zonefs_inode_info *zi = ZONEFS_I(inode);
362
363	if (error) {
364		/*
365		 * For Sync IOs, error recovery is called from
366		 * zonefs_file_dio_write().
367		 */
368		if (!is_sync_kiocb(iocb))
369			zonefs_io_error(inode, true);
370		return error;
371	}
372
373	if (size && zonefs_inode_is_seq(inode)) {
374		/*
375		 * Note that we may be seeing completions out of order,
376		 * but that is not a problem since a write completed
377		 * successfully necessarily means that all preceding writes
378		 * were also successful. So we can safely increase the inode
379		 * size to the write end location.
380		 */
381		mutex_lock(&zi->i_truncate_mutex);
382		if (i_size_read(inode) < iocb->ki_pos + size) {
383			zonefs_update_stats(inode, iocb->ki_pos + size);
384			zonefs_i_size_write(inode, iocb->ki_pos + size);
385		}
386		mutex_unlock(&zi->i_truncate_mutex);
387	}
388
389	return 0;
390}
391
392static const struct iomap_dio_ops zonefs_write_dio_ops = {
393	.end_io		= zonefs_file_write_dio_end_io,
394};
395
396/*
397 * Do not exceed the LFS limits nor the file zone size. If pos is under the
398 * limit it becomes a short access. If it exceeds the limit, return -EFBIG.
399 */
400static loff_t zonefs_write_check_limits(struct file *file, loff_t pos,
401					loff_t count)
402{
403	struct inode *inode = file_inode(file);
404	struct zonefs_zone *z = zonefs_inode_zone(inode);
405	loff_t limit = rlimit(RLIMIT_FSIZE);
406	loff_t max_size = z->z_capacity;
407
408	if (limit != RLIM_INFINITY) {
409		if (pos >= limit) {
410			send_sig(SIGXFSZ, current, 0);
411			return -EFBIG;
412		}
413		count = min(count, limit - pos);
414	}
415
416	if (!(file->f_flags & O_LARGEFILE))
417		max_size = min_t(loff_t, MAX_NON_LFS, max_size);
418
419	if (unlikely(pos >= max_size))
420		return -EFBIG;
421
422	return min(count, max_size - pos);
423}
424
425static ssize_t zonefs_write_checks(struct kiocb *iocb, struct iov_iter *from)
426{
427	struct file *file = iocb->ki_filp;
428	struct inode *inode = file_inode(file);
429	struct zonefs_inode_info *zi = ZONEFS_I(inode);
430	struct zonefs_zone *z = zonefs_inode_zone(inode);
431	loff_t count;
432
433	if (IS_SWAPFILE(inode))
434		return -ETXTBSY;
435
436	if (!iov_iter_count(from))
437		return 0;
438
439	if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
440		return -EINVAL;
441
442	if (iocb->ki_flags & IOCB_APPEND) {
443		if (zonefs_zone_is_cnv(z))
444			return -EINVAL;
445		mutex_lock(&zi->i_truncate_mutex);
446		iocb->ki_pos = z->z_wpoffset;
447		mutex_unlock(&zi->i_truncate_mutex);
448	}
449
450	count = zonefs_write_check_limits(file, iocb->ki_pos,
451					  iov_iter_count(from));
452	if (count < 0)
453		return count;
454
455	iov_iter_truncate(from, count);
456	return iov_iter_count(from);
457}
458
459/*
460 * Handle direct writes. For sequential zone files, this is the only possible
461 * write path. For these files, check that the user is issuing writes
462 * sequentially from the end of the file. This code assumes that the block layer
463 * delivers write requests to the device in sequential order. This is always the
464 * case if a block IO scheduler implementing the ELEVATOR_F_ZBD_SEQ_WRITE
465 * elevator feature is being used (e.g. mq-deadline). The block layer always
466 * automatically select such an elevator for zoned block devices during the
467 * device initialization.
468 */
469static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from)
470{
471	struct inode *inode = file_inode(iocb->ki_filp);
472	struct zonefs_inode_info *zi = ZONEFS_I(inode);
473	struct zonefs_zone *z = zonefs_inode_zone(inode);
474	struct super_block *sb = inode->i_sb;
475	ssize_t ret, count;
476
477	/*
478	 * For async direct IOs to sequential zone files, refuse IOCB_NOWAIT
479	 * as this can cause write reordering (e.g. the first aio gets EAGAIN
480	 * on the inode lock but the second goes through but is now unaligned).
481	 */
482	if (zonefs_zone_is_seq(z) && !is_sync_kiocb(iocb) &&
483	    (iocb->ki_flags & IOCB_NOWAIT))
484		return -EOPNOTSUPP;
485
486	if (iocb->ki_flags & IOCB_NOWAIT) {
487		if (!inode_trylock(inode))
488			return -EAGAIN;
489	} else {
490		inode_lock(inode);
491	}
492
493	count = zonefs_write_checks(iocb, from);
494	if (count <= 0) {
495		ret = count;
496		goto inode_unlock;
497	}
498
499	if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
500		ret = -EINVAL;
501		goto inode_unlock;
502	}
503
504	/* Enforce sequential writes (append only) in sequential zones */
505	if (zonefs_zone_is_seq(z)) {
506		mutex_lock(&zi->i_truncate_mutex);
507		if (iocb->ki_pos != z->z_wpoffset) {
508			mutex_unlock(&zi->i_truncate_mutex);
509			ret = -EINVAL;
510			goto inode_unlock;
511		}
512		/*
513		 * Advance the zone write pointer offset. This assumes that the
514		 * IO will succeed, which is OK to do because we do not allow
515		 * partial writes (IOMAP_DIO_PARTIAL is not set) and if the IO
516		 * fails, the error path will correct the write pointer offset.
517		 */
518		z->z_wpoffset += count;
519		zonefs_inode_account_active(inode);
520		mutex_unlock(&zi->i_truncate_mutex);
521	}
522
523	/*
524	 * iomap_dio_rw() may return ENOTBLK if there was an issue with
525	 * page invalidation. Overwrite that error code with EBUSY so that
526	 * the user can make sense of the error.
527	 */
528	ret = iomap_dio_rw(iocb, from, &zonefs_write_iomap_ops,
529			   &zonefs_write_dio_ops, 0, NULL, 0);
530	if (ret == -ENOTBLK)
531		ret = -EBUSY;
532
533	/*
534	 * For a failed IO or partial completion, trigger error recovery
535	 * to update the zone write pointer offset to a correct value.
536	 * For asynchronous IOs, zonefs_file_write_dio_end_io() may already
537	 * have executed error recovery if the IO already completed when we
538	 * reach here. However, we cannot know that and execute error recovery
539	 * again (that will not change anything).
540	 */
541	if (zonefs_zone_is_seq(z)) {
542		if (ret > 0 && ret != count)
543			ret = -EIO;
544		if (ret < 0 && ret != -EIOCBQUEUED)
545			zonefs_io_error(inode, true);
546	}
547
548inode_unlock:
549	inode_unlock(inode);
550
551	return ret;
552}
553
554static ssize_t zonefs_file_buffered_write(struct kiocb *iocb,
555					  struct iov_iter *from)
556{
557	struct inode *inode = file_inode(iocb->ki_filp);
558	ssize_t ret;
559
560	/*
561	 * Direct IO writes are mandatory for sequential zone files so that the
562	 * write IO issuing order is preserved.
563	 */
564	if (zonefs_inode_is_seq(inode))
565		return -EIO;
566
567	if (iocb->ki_flags & IOCB_NOWAIT) {
568		if (!inode_trylock(inode))
569			return -EAGAIN;
570	} else {
571		inode_lock(inode);
572	}
573
574	ret = zonefs_write_checks(iocb, from);
575	if (ret <= 0)
576		goto inode_unlock;
577
578	ret = iomap_file_buffered_write(iocb, from, &zonefs_write_iomap_ops,
579			NULL, NULL);
580	if (ret == -EIO)
581		zonefs_io_error(inode, true);
582
583inode_unlock:
584	inode_unlock(inode);
585	if (ret > 0)
586		ret = generic_write_sync(iocb, ret);
587
588	return ret;
589}
590
591static ssize_t zonefs_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
592{
593	struct inode *inode = file_inode(iocb->ki_filp);
594	struct zonefs_zone *z = zonefs_inode_zone(inode);
595
596	if (unlikely(IS_IMMUTABLE(inode)))
597		return -EPERM;
598
599	if (sb_rdonly(inode->i_sb))
600		return -EROFS;
601
602	/* Write operations beyond the zone capacity are not allowed */
603	if (iocb->ki_pos >= z->z_capacity)
604		return -EFBIG;
605
606	if (iocb->ki_flags & IOCB_DIRECT) {
607		ssize_t ret = zonefs_file_dio_write(iocb, from);
608
609		if (ret != -ENOTBLK)
610			return ret;
611	}
612
613	return zonefs_file_buffered_write(iocb, from);
614}
615
616static int zonefs_file_read_dio_end_io(struct kiocb *iocb, ssize_t size,
617				       int error, unsigned int flags)
618{
619	if (error) {
620		zonefs_io_error(file_inode(iocb->ki_filp), false);
621		return error;
622	}
623
624	return 0;
625}
626
627static const struct iomap_dio_ops zonefs_read_dio_ops = {
628	.end_io			= zonefs_file_read_dio_end_io,
629};
630
631static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
632{
633	struct inode *inode = file_inode(iocb->ki_filp);
634	struct zonefs_inode_info *zi = ZONEFS_I(inode);
635	struct zonefs_zone *z = zonefs_inode_zone(inode);
636	struct super_block *sb = inode->i_sb;
637	loff_t isize;
638	ssize_t ret;
639
640	/* Offline zones cannot be read */
641	if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777)))
642		return -EPERM;
643
644	if (iocb->ki_pos >= z->z_capacity)
645		return 0;
646
647	if (iocb->ki_flags & IOCB_NOWAIT) {
648		if (!inode_trylock_shared(inode))
649			return -EAGAIN;
650	} else {
651		inode_lock_shared(inode);
652	}
653
654	/* Limit read operations to written data */
655	mutex_lock(&zi->i_truncate_mutex);
656	isize = i_size_read(inode);
657	if (iocb->ki_pos >= isize) {
658		mutex_unlock(&zi->i_truncate_mutex);
659		ret = 0;
660		goto inode_unlock;
661	}
662	iov_iter_truncate(to, isize - iocb->ki_pos);
663	mutex_unlock(&zi->i_truncate_mutex);
664
665	if (iocb->ki_flags & IOCB_DIRECT) {
666		size_t count = iov_iter_count(to);
667
668		if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) {
669			ret = -EINVAL;
670			goto inode_unlock;
671		}
672		file_accessed(iocb->ki_filp);
673		ret = iomap_dio_rw(iocb, to, &zonefs_read_iomap_ops,
674				   &zonefs_read_dio_ops, 0, NULL, 0);
675	} else {
676		ret = generic_file_read_iter(iocb, to);
677		if (ret == -EIO)
678			zonefs_io_error(inode, false);
679	}
680
681inode_unlock:
682	inode_unlock_shared(inode);
683
684	return ret;
685}
686
687static ssize_t zonefs_file_splice_read(struct file *in, loff_t *ppos,
688				       struct pipe_inode_info *pipe,
689				       size_t len, unsigned int flags)
690{
691	struct inode *inode = file_inode(in);
692	struct zonefs_inode_info *zi = ZONEFS_I(inode);
693	struct zonefs_zone *z = zonefs_inode_zone(inode);
694	loff_t isize;
695	ssize_t ret = 0;
696
697	/* Offline zones cannot be read */
698	if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777)))
699		return -EPERM;
700
701	if (*ppos >= z->z_capacity)
702		return 0;
703
704	inode_lock_shared(inode);
705
706	/* Limit read operations to written data */
707	mutex_lock(&zi->i_truncate_mutex);
708	isize = i_size_read(inode);
709	if (*ppos >= isize)
710		len = 0;
711	else
712		len = min_t(loff_t, len, isize - *ppos);
713	mutex_unlock(&zi->i_truncate_mutex);
714
715	if (len > 0) {
716		ret = filemap_splice_read(in, ppos, pipe, len, flags);
717		if (ret == -EIO)
718			zonefs_io_error(inode, false);
719	}
720
721	inode_unlock_shared(inode);
722	return ret;
723}
724
725/*
726 * Write open accounting is done only for sequential files.
727 */
728static inline bool zonefs_seq_file_need_wro(struct inode *inode,
729					    struct file *file)
730{
731	if (zonefs_inode_is_cnv(inode))
732		return false;
733
734	if (!(file->f_mode & FMODE_WRITE))
735		return false;
736
737	return true;
738}
739
740static int zonefs_seq_file_write_open(struct inode *inode)
741{
742	struct zonefs_inode_info *zi = ZONEFS_I(inode);
743	struct zonefs_zone *z = zonefs_inode_zone(inode);
744	int ret = 0;
745
746	mutex_lock(&zi->i_truncate_mutex);
747
748	if (!zi->i_wr_refcnt) {
749		struct zonefs_sb_info *sbi = ZONEFS_SB(inode->i_sb);
750		unsigned int wro = atomic_inc_return(&sbi->s_wro_seq_files);
751
752		if (sbi->s_mount_opts & ZONEFS_MNTOPT_EXPLICIT_OPEN) {
753
754			if (sbi->s_max_wro_seq_files
755			    && wro > sbi->s_max_wro_seq_files) {
756				atomic_dec(&sbi->s_wro_seq_files);
757				ret = -EBUSY;
758				goto unlock;
759			}
760
761			if (i_size_read(inode) < z->z_capacity) {
762				ret = zonefs_inode_zone_mgmt(inode,
763							     REQ_OP_ZONE_OPEN);
764				if (ret) {
765					atomic_dec(&sbi->s_wro_seq_files);
766					goto unlock;
767				}
768				z->z_flags |= ZONEFS_ZONE_OPEN;
769				zonefs_inode_account_active(inode);
770			}
771		}
772	}
773
774	zi->i_wr_refcnt++;
775
776unlock:
777	mutex_unlock(&zi->i_truncate_mutex);
778
779	return ret;
780}
781
782static int zonefs_file_open(struct inode *inode, struct file *file)
783{
784	int ret;
785
786	file->f_mode |= FMODE_CAN_ODIRECT;
787	ret = generic_file_open(inode, file);
788	if (ret)
789		return ret;
790
791	if (zonefs_seq_file_need_wro(inode, file))
792		return zonefs_seq_file_write_open(inode);
793
794	return 0;
795}
796
797static void zonefs_seq_file_write_close(struct inode *inode)
798{
799	struct zonefs_inode_info *zi = ZONEFS_I(inode);
800	struct zonefs_zone *z = zonefs_inode_zone(inode);
801	struct super_block *sb = inode->i_sb;
802	struct zonefs_sb_info *sbi = ZONEFS_SB(sb);
803	int ret = 0;
804
805	mutex_lock(&zi->i_truncate_mutex);
806
807	zi->i_wr_refcnt--;
808	if (zi->i_wr_refcnt)
809		goto unlock;
810
811	/*
812	 * The file zone may not be open anymore (e.g. the file was truncated to
813	 * its maximum size or it was fully written). For this case, we only
814	 * need to decrement the write open count.
815	 */
816	if (z->z_flags & ZONEFS_ZONE_OPEN) {
817		ret = zonefs_inode_zone_mgmt(inode, REQ_OP_ZONE_CLOSE);
818		if (ret) {
819			__zonefs_io_error(inode, false);
820			/*
821			 * Leaving zones explicitly open may lead to a state
822			 * where most zones cannot be written (zone resources
823			 * exhausted). So take preventive action by remounting
824			 * read-only.
825			 */
826			if (z->z_flags & ZONEFS_ZONE_OPEN &&
827			    !(sb->s_flags & SB_RDONLY)) {
828				zonefs_warn(sb,
829					"closing zone at %llu failed %d\n",
830					z->z_sector, ret);
831				zonefs_warn(sb,
832					"remounting filesystem read-only\n");
833				sb->s_flags |= SB_RDONLY;
834			}
835			goto unlock;
836		}
837
838		z->z_flags &= ~ZONEFS_ZONE_OPEN;
839		zonefs_inode_account_active(inode);
840	}
841
842	atomic_dec(&sbi->s_wro_seq_files);
843
844unlock:
845	mutex_unlock(&zi->i_truncate_mutex);
846}
847
848static int zonefs_file_release(struct inode *inode, struct file *file)
849{
850	/*
851	 * If we explicitly open a zone we must close it again as well, but the
852	 * zone management operation can fail (either due to an IO error or as
853	 * the zone has gone offline or read-only). Make sure we don't fail the
854	 * close(2) for user-space.
855	 */
856	if (zonefs_seq_file_need_wro(inode, file))
857		zonefs_seq_file_write_close(inode);
858
859	return 0;
860}
861
862const struct file_operations zonefs_file_operations = {
863	.open		= zonefs_file_open,
864	.release	= zonefs_file_release,
865	.fsync		= zonefs_file_fsync,
866	.mmap_prepare	= zonefs_file_mmap_prepare,
867	.llseek		= zonefs_file_llseek,
868	.read_iter	= zonefs_file_read_iter,
869	.write_iter	= zonefs_file_write_iter,
870	.splice_read	= zonefs_file_splice_read,
871	.splice_write	= iter_file_splice_write,
872	.iopoll		= iocb_bio_iopoll,
873};