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