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