]>
Commit | Line | Data |
---|---|---|
8dcc1a9d DLM |
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 | ||
24 | #include "zonefs.h" | |
25 | ||
26 | static int zonefs_iomap_begin(struct inode *inode, loff_t offset, loff_t length, | |
27 | unsigned int flags, struct iomap *iomap, | |
28 | struct iomap *srcmap) | |
29 | { | |
30 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
31 | struct super_block *sb = inode->i_sb; | |
32 | loff_t isize; | |
33 | ||
34 | /* All I/Os should always be within the file maximum size */ | |
35 | if (WARN_ON_ONCE(offset + length > zi->i_max_size)) | |
36 | return -EIO; | |
37 | ||
38 | /* | |
39 | * Sequential zones can only accept direct writes. This is already | |
40 | * checked when writes are issued, so warn if we see a page writeback | |
41 | * operation. | |
42 | */ | |
43 | if (WARN_ON_ONCE(zi->i_ztype == ZONEFS_ZTYPE_SEQ && | |
44 | (flags & IOMAP_WRITE) && !(flags & IOMAP_DIRECT))) | |
45 | return -EIO; | |
46 | ||
47 | /* | |
48 | * For conventional zones, all blocks are always mapped. For sequential | |
49 | * zones, all blocks after always mapped below the inode size (zone | |
50 | * write pointer) and unwriten beyond. | |
51 | */ | |
52 | mutex_lock(&zi->i_truncate_mutex); | |
53 | isize = i_size_read(inode); | |
54 | if (offset >= isize) | |
55 | iomap->type = IOMAP_UNWRITTEN; | |
56 | else | |
57 | iomap->type = IOMAP_MAPPED; | |
58 | if (flags & IOMAP_WRITE) | |
59 | length = zi->i_max_size - offset; | |
60 | else | |
61 | length = min(length, isize - offset); | |
62 | mutex_unlock(&zi->i_truncate_mutex); | |
63 | ||
64 | iomap->offset = ALIGN_DOWN(offset, sb->s_blocksize); | |
65 | iomap->length = ALIGN(offset + length, sb->s_blocksize) - iomap->offset; | |
66 | iomap->bdev = inode->i_sb->s_bdev; | |
67 | iomap->addr = (zi->i_zsector << SECTOR_SHIFT) + iomap->offset; | |
68 | ||
69 | return 0; | |
70 | } | |
71 | ||
72 | static const struct iomap_ops zonefs_iomap_ops = { | |
73 | .iomap_begin = zonefs_iomap_begin, | |
74 | }; | |
75 | ||
76 | static int zonefs_readpage(struct file *unused, struct page *page) | |
77 | { | |
78 | return iomap_readpage(page, &zonefs_iomap_ops); | |
79 | } | |
80 | ||
9d24a13a | 81 | static void zonefs_readahead(struct readahead_control *rac) |
8dcc1a9d | 82 | { |
9d24a13a | 83 | iomap_readahead(rac, &zonefs_iomap_ops); |
8dcc1a9d DLM |
84 | } |
85 | ||
86 | /* | |
87 | * Map blocks for page writeback. This is used only on conventional zone files, | |
88 | * which implies that the page range can only be within the fixed inode size. | |
89 | */ | |
90 | static int zonefs_map_blocks(struct iomap_writepage_ctx *wpc, | |
91 | struct inode *inode, loff_t offset) | |
92 | { | |
93 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
94 | ||
95 | if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV)) | |
96 | return -EIO; | |
97 | if (WARN_ON_ONCE(offset >= i_size_read(inode))) | |
98 | return -EIO; | |
99 | ||
100 | /* If the mapping is already OK, nothing needs to be done */ | |
101 | if (offset >= wpc->iomap.offset && | |
102 | offset < wpc->iomap.offset + wpc->iomap.length) | |
103 | return 0; | |
104 | ||
105 | return zonefs_iomap_begin(inode, offset, zi->i_max_size - offset, | |
106 | IOMAP_WRITE, &wpc->iomap, NULL); | |
107 | } | |
108 | ||
109 | static const struct iomap_writeback_ops zonefs_writeback_ops = { | |
110 | .map_blocks = zonefs_map_blocks, | |
111 | }; | |
112 | ||
113 | static int zonefs_writepage(struct page *page, struct writeback_control *wbc) | |
114 | { | |
115 | struct iomap_writepage_ctx wpc = { }; | |
116 | ||
117 | return iomap_writepage(page, wbc, &wpc, &zonefs_writeback_ops); | |
118 | } | |
119 | ||
120 | static int zonefs_writepages(struct address_space *mapping, | |
121 | struct writeback_control *wbc) | |
122 | { | |
123 | struct iomap_writepage_ctx wpc = { }; | |
124 | ||
125 | return iomap_writepages(mapping, wbc, &wpc, &zonefs_writeback_ops); | |
126 | } | |
127 | ||
128 | static const struct address_space_operations zonefs_file_aops = { | |
129 | .readpage = zonefs_readpage, | |
9d24a13a | 130 | .readahead = zonefs_readahead, |
8dcc1a9d DLM |
131 | .writepage = zonefs_writepage, |
132 | .writepages = zonefs_writepages, | |
133 | .set_page_dirty = iomap_set_page_dirty, | |
134 | .releasepage = iomap_releasepage, | |
135 | .invalidatepage = iomap_invalidatepage, | |
136 | .migratepage = iomap_migrate_page, | |
137 | .is_partially_uptodate = iomap_is_partially_uptodate, | |
138 | .error_remove_page = generic_error_remove_page, | |
139 | .direct_IO = noop_direct_IO, | |
140 | }; | |
141 | ||
142 | static void zonefs_update_stats(struct inode *inode, loff_t new_isize) | |
143 | { | |
144 | struct super_block *sb = inode->i_sb; | |
145 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
146 | loff_t old_isize = i_size_read(inode); | |
147 | loff_t nr_blocks; | |
148 | ||
149 | if (new_isize == old_isize) | |
150 | return; | |
151 | ||
152 | spin_lock(&sbi->s_lock); | |
153 | ||
154 | /* | |
155 | * This may be called for an update after an IO error. | |
156 | * So beware of the values seen. | |
157 | */ | |
158 | if (new_isize < old_isize) { | |
159 | nr_blocks = (old_isize - new_isize) >> sb->s_blocksize_bits; | |
160 | if (sbi->s_used_blocks > nr_blocks) | |
161 | sbi->s_used_blocks -= nr_blocks; | |
162 | else | |
163 | sbi->s_used_blocks = 0; | |
164 | } else { | |
165 | sbi->s_used_blocks += | |
166 | (new_isize - old_isize) >> sb->s_blocksize_bits; | |
167 | if (sbi->s_used_blocks > sbi->s_blocks) | |
168 | sbi->s_used_blocks = sbi->s_blocks; | |
169 | } | |
170 | ||
171 | spin_unlock(&sbi->s_lock); | |
172 | } | |
173 | ||
174 | /* | |
175 | * Check a zone condition and adjust its file inode access permissions for | |
176 | * offline and readonly zones. Return the inode size corresponding to the | |
177 | * amount of readable data in the zone. | |
178 | */ | |
179 | static loff_t zonefs_check_zone_condition(struct inode *inode, | |
ccf4ad7d DLM |
180 | struct blk_zone *zone, bool warn, |
181 | bool mount) | |
8dcc1a9d DLM |
182 | { |
183 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
184 | ||
185 | switch (zone->cond) { | |
186 | case BLK_ZONE_COND_OFFLINE: | |
187 | /* | |
188 | * Dead zone: make the inode immutable, disable all accesses | |
189 | * and set the file size to 0 (zone wp set to zone start). | |
190 | */ | |
191 | if (warn) | |
192 | zonefs_warn(inode->i_sb, "inode %lu: offline zone\n", | |
193 | inode->i_ino); | |
194 | inode->i_flags |= S_IMMUTABLE; | |
195 | inode->i_mode &= ~0777; | |
196 | zone->wp = zone->start; | |
197 | return 0; | |
198 | case BLK_ZONE_COND_READONLY: | |
ccf4ad7d DLM |
199 | /* |
200 | * The write pointer of read-only zones is invalid. If such a | |
201 | * zone is found during mount, the file size cannot be retrieved | |
202 | * so we treat the zone as offline (mount == true case). | |
203 | * Otherwise, keep the file size as it was when last updated | |
204 | * so that the user can recover data. In both cases, writes are | |
205 | * always disabled for the zone. | |
206 | */ | |
8dcc1a9d DLM |
207 | if (warn) |
208 | zonefs_warn(inode->i_sb, "inode %lu: read-only zone\n", | |
209 | inode->i_ino); | |
210 | inode->i_flags |= S_IMMUTABLE; | |
ccf4ad7d DLM |
211 | if (mount) { |
212 | zone->cond = BLK_ZONE_COND_OFFLINE; | |
213 | inode->i_mode &= ~0777; | |
214 | zone->wp = zone->start; | |
215 | return 0; | |
216 | } | |
8dcc1a9d | 217 | inode->i_mode &= ~0222; |
ccf4ad7d | 218 | return i_size_read(inode); |
8dcc1a9d DLM |
219 | default: |
220 | if (zi->i_ztype == ZONEFS_ZTYPE_CNV) | |
221 | return zi->i_max_size; | |
222 | return (zone->wp - zone->start) << SECTOR_SHIFT; | |
223 | } | |
224 | } | |
225 | ||
226 | struct zonefs_ioerr_data { | |
227 | struct inode *inode; | |
228 | bool write; | |
229 | }; | |
230 | ||
231 | static int zonefs_io_error_cb(struct blk_zone *zone, unsigned int idx, | |
232 | void *data) | |
233 | { | |
234 | struct zonefs_ioerr_data *err = data; | |
235 | struct inode *inode = err->inode; | |
236 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
237 | struct super_block *sb = inode->i_sb; | |
238 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
239 | loff_t isize, data_size; | |
240 | ||
241 | /* | |
242 | * Check the zone condition: if the zone is not "bad" (offline or | |
243 | * read-only), read errors are simply signaled to the IO issuer as long | |
244 | * as there is no inconsistency between the inode size and the amount of | |
245 | * data writen in the zone (data_size). | |
246 | */ | |
ccf4ad7d | 247 | data_size = zonefs_check_zone_condition(inode, zone, true, false); |
8dcc1a9d DLM |
248 | isize = i_size_read(inode); |
249 | if (zone->cond != BLK_ZONE_COND_OFFLINE && | |
250 | zone->cond != BLK_ZONE_COND_READONLY && | |
251 | !err->write && isize == data_size) | |
252 | return 0; | |
253 | ||
254 | /* | |
255 | * At this point, we detected either a bad zone or an inconsistency | |
256 | * between the inode size and the amount of data written in the zone. | |
257 | * For the latter case, the cause may be a write IO error or an external | |
258 | * action on the device. Two error patterns exist: | |
259 | * 1) The inode size is lower than the amount of data in the zone: | |
260 | * a write operation partially failed and data was writen at the end | |
261 | * of the file. This can happen in the case of a large direct IO | |
262 | * needing several BIOs and/or write requests to be processed. | |
263 | * 2) The inode size is larger than the amount of data in the zone: | |
264 | * this can happen with a deferred write error with the use of the | |
265 | * device side write cache after getting successful write IO | |
266 | * completions. Other possibilities are (a) an external corruption, | |
267 | * e.g. an application reset the zone directly, or (b) the device | |
268 | * has a serious problem (e.g. firmware bug). | |
269 | * | |
270 | * In all cases, warn about inode size inconsistency and handle the | |
271 | * IO error according to the zone condition and to the mount options. | |
272 | */ | |
273 | if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && isize != data_size) | |
274 | zonefs_warn(sb, "inode %lu: invalid size %lld (should be %lld)\n", | |
275 | inode->i_ino, isize, data_size); | |
276 | ||
277 | /* | |
278 | * First handle bad zones signaled by hardware. The mount options | |
279 | * errors=zone-ro and errors=zone-offline result in changing the | |
280 | * zone condition to read-only and offline respectively, as if the | |
281 | * condition was signaled by the hardware. | |
282 | */ | |
283 | if (zone->cond == BLK_ZONE_COND_OFFLINE || | |
284 | sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL) { | |
285 | zonefs_warn(sb, "inode %lu: read/write access disabled\n", | |
286 | inode->i_ino); | |
287 | if (zone->cond != BLK_ZONE_COND_OFFLINE) { | |
288 | zone->cond = BLK_ZONE_COND_OFFLINE; | |
289 | data_size = zonefs_check_zone_condition(inode, zone, | |
ccf4ad7d | 290 | false, false); |
8dcc1a9d DLM |
291 | } |
292 | } else if (zone->cond == BLK_ZONE_COND_READONLY || | |
293 | sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO) { | |
294 | zonefs_warn(sb, "inode %lu: write access disabled\n", | |
295 | inode->i_ino); | |
296 | if (zone->cond != BLK_ZONE_COND_READONLY) { | |
297 | zone->cond = BLK_ZONE_COND_READONLY; | |
298 | data_size = zonefs_check_zone_condition(inode, zone, | |
ccf4ad7d | 299 | false, false); |
8dcc1a9d DLM |
300 | } |
301 | } | |
302 | ||
303 | /* | |
304 | * If error=remount-ro was specified, any error result in remounting | |
305 | * the volume as read-only. | |
306 | */ | |
307 | if ((sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) && !sb_rdonly(sb)) { | |
308 | zonefs_warn(sb, "remounting filesystem read-only\n"); | |
309 | sb->s_flags |= SB_RDONLY; | |
310 | } | |
311 | ||
312 | /* | |
313 | * Update block usage stats and the inode size to prevent access to | |
314 | * invalid data. | |
315 | */ | |
316 | zonefs_update_stats(inode, data_size); | |
317 | i_size_write(inode, data_size); | |
318 | zi->i_wpoffset = data_size; | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
323 | /* | |
324 | * When an file IO error occurs, check the file zone to see if there is a change | |
325 | * in the zone condition (e.g. offline or read-only). For a failed write to a | |
326 | * sequential zone, the zone write pointer position must also be checked to | |
327 | * eventually correct the file size and zonefs inode write pointer offset | |
328 | * (which can be out of sync with the drive due to partial write failures). | |
329 | */ | |
330 | static void zonefs_io_error(struct inode *inode, bool write) | |
331 | { | |
332 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
333 | struct super_block *sb = inode->i_sb; | |
334 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
335 | unsigned int noio_flag; | |
336 | unsigned int nr_zones = | |
337 | zi->i_max_size >> (sbi->s_zone_sectors_shift + SECTOR_SHIFT); | |
338 | struct zonefs_ioerr_data err = { | |
339 | .inode = inode, | |
340 | .write = write, | |
341 | }; | |
342 | int ret; | |
343 | ||
344 | mutex_lock(&zi->i_truncate_mutex); | |
345 | ||
346 | /* | |
347 | * Memory allocations in blkdev_report_zones() can trigger a memory | |
348 | * reclaim which may in turn cause a recursion into zonefs as well as | |
349 | * struct request allocations for the same device. The former case may | |
350 | * end up in a deadlock on the inode truncate mutex, while the latter | |
351 | * may prevent IO forward progress. Executing the report zones under | |
352 | * the GFP_NOIO context avoids both problems. | |
353 | */ | |
354 | noio_flag = memalloc_noio_save(); | |
355 | ret = blkdev_report_zones(sb->s_bdev, zi->i_zsector, nr_zones, | |
356 | zonefs_io_error_cb, &err); | |
357 | if (ret != nr_zones) | |
358 | zonefs_err(sb, "Get inode %lu zone information failed %d\n", | |
359 | inode->i_ino, ret); | |
360 | memalloc_noio_restore(noio_flag); | |
361 | ||
362 | mutex_unlock(&zi->i_truncate_mutex); | |
363 | } | |
364 | ||
365 | static int zonefs_file_truncate(struct inode *inode, loff_t isize) | |
366 | { | |
367 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
368 | loff_t old_isize; | |
369 | enum req_opf op; | |
370 | int ret = 0; | |
371 | ||
372 | /* | |
373 | * Only sequential zone files can be truncated and truncation is allowed | |
374 | * only down to a 0 size, which is equivalent to a zone reset, and to | |
375 | * the maximum file size, which is equivalent to a zone finish. | |
376 | */ | |
377 | if (zi->i_ztype != ZONEFS_ZTYPE_SEQ) | |
378 | return -EPERM; | |
379 | ||
380 | if (!isize) | |
381 | op = REQ_OP_ZONE_RESET; | |
382 | else if (isize == zi->i_max_size) | |
383 | op = REQ_OP_ZONE_FINISH; | |
384 | else | |
385 | return -EPERM; | |
386 | ||
387 | inode_dio_wait(inode); | |
388 | ||
389 | /* Serialize against page faults */ | |
390 | down_write(&zi->i_mmap_sem); | |
391 | ||
392 | /* Serialize against zonefs_iomap_begin() */ | |
393 | mutex_lock(&zi->i_truncate_mutex); | |
394 | ||
395 | old_isize = i_size_read(inode); | |
396 | if (isize == old_isize) | |
397 | goto unlock; | |
398 | ||
399 | ret = blkdev_zone_mgmt(inode->i_sb->s_bdev, op, zi->i_zsector, | |
400 | zi->i_max_size >> SECTOR_SHIFT, GFP_NOFS); | |
401 | if (ret) { | |
402 | zonefs_err(inode->i_sb, | |
403 | "Zone management operation at %llu failed %d", | |
404 | zi->i_zsector, ret); | |
405 | goto unlock; | |
406 | } | |
407 | ||
408 | zonefs_update_stats(inode, isize); | |
409 | truncate_setsize(inode, isize); | |
410 | zi->i_wpoffset = isize; | |
411 | ||
412 | unlock: | |
413 | mutex_unlock(&zi->i_truncate_mutex); | |
414 | up_write(&zi->i_mmap_sem); | |
415 | ||
416 | return ret; | |
417 | } | |
418 | ||
419 | static int zonefs_inode_setattr(struct dentry *dentry, struct iattr *iattr) | |
420 | { | |
421 | struct inode *inode = d_inode(dentry); | |
422 | int ret; | |
423 | ||
424 | if (unlikely(IS_IMMUTABLE(inode))) | |
425 | return -EPERM; | |
426 | ||
427 | ret = setattr_prepare(dentry, iattr); | |
428 | if (ret) | |
429 | return ret; | |
430 | ||
431 | /* | |
432 | * Since files and directories cannot be created nor deleted, do not | |
433 | * allow setting any write attributes on the sub-directories grouping | |
434 | * files by zone type. | |
435 | */ | |
436 | if ((iattr->ia_valid & ATTR_MODE) && S_ISDIR(inode->i_mode) && | |
437 | (iattr->ia_mode & 0222)) | |
438 | return -EPERM; | |
439 | ||
440 | if (((iattr->ia_valid & ATTR_UID) && | |
441 | !uid_eq(iattr->ia_uid, inode->i_uid)) || | |
442 | ((iattr->ia_valid & ATTR_GID) && | |
443 | !gid_eq(iattr->ia_gid, inode->i_gid))) { | |
444 | ret = dquot_transfer(inode, iattr); | |
445 | if (ret) | |
446 | return ret; | |
447 | } | |
448 | ||
449 | if (iattr->ia_valid & ATTR_SIZE) { | |
450 | ret = zonefs_file_truncate(inode, iattr->ia_size); | |
451 | if (ret) | |
452 | return ret; | |
453 | } | |
454 | ||
455 | setattr_copy(inode, iattr); | |
456 | ||
457 | return 0; | |
458 | } | |
459 | ||
460 | static const struct inode_operations zonefs_file_inode_operations = { | |
461 | .setattr = zonefs_inode_setattr, | |
462 | }; | |
463 | ||
464 | static int zonefs_file_fsync(struct file *file, loff_t start, loff_t end, | |
465 | int datasync) | |
466 | { | |
467 | struct inode *inode = file_inode(file); | |
468 | int ret = 0; | |
469 | ||
470 | if (unlikely(IS_IMMUTABLE(inode))) | |
471 | return -EPERM; | |
472 | ||
473 | /* | |
474 | * Since only direct writes are allowed in sequential files, page cache | |
475 | * flush is needed only for conventional zone files. | |
476 | */ | |
477 | if (ZONEFS_I(inode)->i_ztype == ZONEFS_ZTYPE_CNV) | |
478 | ret = file_write_and_wait_range(file, start, end); | |
479 | if (!ret) | |
480 | ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); | |
481 | ||
482 | if (ret) | |
483 | zonefs_io_error(inode, true); | |
484 | ||
485 | return ret; | |
486 | } | |
487 | ||
488 | static vm_fault_t zonefs_filemap_fault(struct vm_fault *vmf) | |
489 | { | |
490 | struct zonefs_inode_info *zi = ZONEFS_I(file_inode(vmf->vma->vm_file)); | |
491 | vm_fault_t ret; | |
492 | ||
493 | down_read(&zi->i_mmap_sem); | |
494 | ret = filemap_fault(vmf); | |
495 | up_read(&zi->i_mmap_sem); | |
496 | ||
497 | return ret; | |
498 | } | |
499 | ||
500 | static vm_fault_t zonefs_filemap_page_mkwrite(struct vm_fault *vmf) | |
501 | { | |
502 | struct inode *inode = file_inode(vmf->vma->vm_file); | |
503 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
504 | vm_fault_t ret; | |
505 | ||
506 | if (unlikely(IS_IMMUTABLE(inode))) | |
507 | return VM_FAULT_SIGBUS; | |
508 | ||
509 | /* | |
510 | * Sanity check: only conventional zone files can have shared | |
511 | * writeable mappings. | |
512 | */ | |
513 | if (WARN_ON_ONCE(zi->i_ztype != ZONEFS_ZTYPE_CNV)) | |
514 | return VM_FAULT_NOPAGE; | |
515 | ||
516 | sb_start_pagefault(inode->i_sb); | |
517 | file_update_time(vmf->vma->vm_file); | |
518 | ||
519 | /* Serialize against truncates */ | |
520 | down_read(&zi->i_mmap_sem); | |
521 | ret = iomap_page_mkwrite(vmf, &zonefs_iomap_ops); | |
522 | up_read(&zi->i_mmap_sem); | |
523 | ||
524 | sb_end_pagefault(inode->i_sb); | |
525 | return ret; | |
526 | } | |
527 | ||
528 | static const struct vm_operations_struct zonefs_file_vm_ops = { | |
529 | .fault = zonefs_filemap_fault, | |
530 | .map_pages = filemap_map_pages, | |
531 | .page_mkwrite = zonefs_filemap_page_mkwrite, | |
532 | }; | |
533 | ||
534 | static int zonefs_file_mmap(struct file *file, struct vm_area_struct *vma) | |
535 | { | |
536 | /* | |
537 | * Conventional zones accept random writes, so their files can support | |
538 | * shared writable mappings. For sequential zone files, only read | |
539 | * mappings are possible since there are no guarantees for write | |
540 | * ordering between msync() and page cache writeback. | |
541 | */ | |
542 | if (ZONEFS_I(file_inode(file))->i_ztype == ZONEFS_ZTYPE_SEQ && | |
543 | (vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) | |
544 | return -EINVAL; | |
545 | ||
546 | file_accessed(file); | |
547 | vma->vm_ops = &zonefs_file_vm_ops; | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | static loff_t zonefs_file_llseek(struct file *file, loff_t offset, int whence) | |
553 | { | |
554 | loff_t isize = i_size_read(file_inode(file)); | |
555 | ||
556 | /* | |
557 | * Seeks are limited to below the zone size for conventional zones | |
558 | * and below the zone write pointer for sequential zones. In both | |
559 | * cases, this limit is the inode size. | |
560 | */ | |
561 | return generic_file_llseek_size(file, offset, whence, isize, isize); | |
562 | } | |
563 | ||
564 | static int zonefs_file_write_dio_end_io(struct kiocb *iocb, ssize_t size, | |
565 | int error, unsigned int flags) | |
566 | { | |
567 | struct inode *inode = file_inode(iocb->ki_filp); | |
568 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
569 | ||
570 | if (error) { | |
571 | zonefs_io_error(inode, true); | |
572 | return error; | |
573 | } | |
574 | ||
575 | if (size && zi->i_ztype != ZONEFS_ZTYPE_CNV) { | |
576 | /* | |
577 | * Note that we may be seeing completions out of order, | |
578 | * but that is not a problem since a write completed | |
579 | * successfully necessarily means that all preceding writes | |
580 | * were also successful. So we can safely increase the inode | |
581 | * size to the write end location. | |
582 | */ | |
583 | mutex_lock(&zi->i_truncate_mutex); | |
584 | if (i_size_read(inode) < iocb->ki_pos + size) { | |
585 | zonefs_update_stats(inode, iocb->ki_pos + size); | |
586 | i_size_write(inode, iocb->ki_pos + size); | |
587 | } | |
588 | mutex_unlock(&zi->i_truncate_mutex); | |
589 | } | |
590 | ||
591 | return 0; | |
592 | } | |
593 | ||
594 | static const struct iomap_dio_ops zonefs_write_dio_ops = { | |
595 | .end_io = zonefs_file_write_dio_end_io, | |
596 | }; | |
597 | ||
598 | /* | |
599 | * Handle direct writes. For sequential zone files, this is the only possible | |
600 | * write path. For these files, check that the user is issuing writes | |
601 | * sequentially from the end of the file. This code assumes that the block layer | |
602 | * delivers write requests to the device in sequential order. This is always the | |
603 | * case if a block IO scheduler implementing the ELEVATOR_F_ZBD_SEQ_WRITE | |
604 | * elevator feature is being used (e.g. mq-deadline). The block layer always | |
605 | * automatically select such an elevator for zoned block devices during the | |
606 | * device initialization. | |
607 | */ | |
608 | static ssize_t zonefs_file_dio_write(struct kiocb *iocb, struct iov_iter *from) | |
609 | { | |
610 | struct inode *inode = file_inode(iocb->ki_filp); | |
611 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
612 | struct super_block *sb = inode->i_sb; | |
613 | size_t count; | |
614 | ssize_t ret; | |
615 | ||
616 | /* | |
7c69eb84 | 617 | * For async direct IOs to sequential zone files, refuse IOCB_NOWAIT |
8dcc1a9d DLM |
618 | * as this can cause write reordering (e.g. the first aio gets EAGAIN |
619 | * on the inode lock but the second goes through but is now unaligned). | |
620 | */ | |
7c69eb84 CH |
621 | if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && !is_sync_kiocb(iocb) && |
622 | (iocb->ki_flags & IOCB_NOWAIT)) | |
623 | return -EOPNOTSUPP; | |
8dcc1a9d DLM |
624 | |
625 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
626 | if (!inode_trylock(inode)) | |
627 | return -EAGAIN; | |
628 | } else { | |
629 | inode_lock(inode); | |
630 | } | |
631 | ||
632 | ret = generic_write_checks(iocb, from); | |
633 | if (ret <= 0) | |
634 | goto inode_unlock; | |
635 | ||
636 | iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos); | |
637 | count = iov_iter_count(from); | |
638 | ||
639 | if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) { | |
640 | ret = -EINVAL; | |
641 | goto inode_unlock; | |
642 | } | |
643 | ||
644 | /* Enforce sequential writes (append only) in sequential zones */ | |
645 | mutex_lock(&zi->i_truncate_mutex); | |
646 | if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && iocb->ki_pos != zi->i_wpoffset) { | |
647 | mutex_unlock(&zi->i_truncate_mutex); | |
648 | ret = -EINVAL; | |
649 | goto inode_unlock; | |
650 | } | |
651 | mutex_unlock(&zi->i_truncate_mutex); | |
652 | ||
653 | ret = iomap_dio_rw(iocb, from, &zonefs_iomap_ops, | |
654 | &zonefs_write_dio_ops, is_sync_kiocb(iocb)); | |
655 | if (zi->i_ztype == ZONEFS_ZTYPE_SEQ && | |
656 | (ret > 0 || ret == -EIOCBQUEUED)) { | |
657 | if (ret > 0) | |
658 | count = ret; | |
659 | mutex_lock(&zi->i_truncate_mutex); | |
660 | zi->i_wpoffset += count; | |
661 | mutex_unlock(&zi->i_truncate_mutex); | |
662 | } | |
663 | ||
664 | inode_unlock: | |
665 | inode_unlock(inode); | |
666 | ||
667 | return ret; | |
668 | } | |
669 | ||
670 | static ssize_t zonefs_file_buffered_write(struct kiocb *iocb, | |
671 | struct iov_iter *from) | |
672 | { | |
673 | struct inode *inode = file_inode(iocb->ki_filp); | |
674 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
675 | ssize_t ret; | |
676 | ||
677 | /* | |
678 | * Direct IO writes are mandatory for sequential zone files so that the | |
679 | * write IO issuing order is preserved. | |
680 | */ | |
681 | if (zi->i_ztype != ZONEFS_ZTYPE_CNV) | |
682 | return -EIO; | |
683 | ||
684 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
685 | if (!inode_trylock(inode)) | |
686 | return -EAGAIN; | |
687 | } else { | |
688 | inode_lock(inode); | |
689 | } | |
690 | ||
691 | ret = generic_write_checks(iocb, from); | |
692 | if (ret <= 0) | |
693 | goto inode_unlock; | |
694 | ||
695 | iov_iter_truncate(from, zi->i_max_size - iocb->ki_pos); | |
696 | ||
697 | ret = iomap_file_buffered_write(iocb, from, &zonefs_iomap_ops); | |
698 | if (ret > 0) | |
699 | iocb->ki_pos += ret; | |
700 | else if (ret == -EIO) | |
701 | zonefs_io_error(inode, true); | |
702 | ||
703 | inode_unlock: | |
704 | inode_unlock(inode); | |
705 | if (ret > 0) | |
706 | ret = generic_write_sync(iocb, ret); | |
707 | ||
708 | return ret; | |
709 | } | |
710 | ||
711 | static ssize_t zonefs_file_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
712 | { | |
713 | struct inode *inode = file_inode(iocb->ki_filp); | |
714 | ||
715 | if (unlikely(IS_IMMUTABLE(inode))) | |
716 | return -EPERM; | |
717 | ||
718 | if (sb_rdonly(inode->i_sb)) | |
719 | return -EROFS; | |
720 | ||
721 | /* Write operations beyond the zone size are not allowed */ | |
722 | if (iocb->ki_pos >= ZONEFS_I(inode)->i_max_size) | |
723 | return -EFBIG; | |
724 | ||
725 | if (iocb->ki_flags & IOCB_DIRECT) | |
726 | return zonefs_file_dio_write(iocb, from); | |
727 | ||
728 | return zonefs_file_buffered_write(iocb, from); | |
729 | } | |
730 | ||
731 | static int zonefs_file_read_dio_end_io(struct kiocb *iocb, ssize_t size, | |
732 | int error, unsigned int flags) | |
733 | { | |
734 | if (error) { | |
735 | zonefs_io_error(file_inode(iocb->ki_filp), false); | |
736 | return error; | |
737 | } | |
738 | ||
739 | return 0; | |
740 | } | |
741 | ||
742 | static const struct iomap_dio_ops zonefs_read_dio_ops = { | |
743 | .end_io = zonefs_file_read_dio_end_io, | |
744 | }; | |
745 | ||
746 | static ssize_t zonefs_file_read_iter(struct kiocb *iocb, struct iov_iter *to) | |
747 | { | |
748 | struct inode *inode = file_inode(iocb->ki_filp); | |
749 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
750 | struct super_block *sb = inode->i_sb; | |
751 | loff_t isize; | |
752 | ssize_t ret; | |
753 | ||
754 | /* Offline zones cannot be read */ | |
755 | if (unlikely(IS_IMMUTABLE(inode) && !(inode->i_mode & 0777))) | |
756 | return -EPERM; | |
757 | ||
758 | if (iocb->ki_pos >= zi->i_max_size) | |
759 | return 0; | |
760 | ||
761 | if (iocb->ki_flags & IOCB_NOWAIT) { | |
762 | if (!inode_trylock_shared(inode)) | |
763 | return -EAGAIN; | |
764 | } else { | |
765 | inode_lock_shared(inode); | |
766 | } | |
767 | ||
768 | /* Limit read operations to written data */ | |
769 | mutex_lock(&zi->i_truncate_mutex); | |
770 | isize = i_size_read(inode); | |
771 | if (iocb->ki_pos >= isize) { | |
772 | mutex_unlock(&zi->i_truncate_mutex); | |
773 | ret = 0; | |
774 | goto inode_unlock; | |
775 | } | |
776 | iov_iter_truncate(to, isize - iocb->ki_pos); | |
777 | mutex_unlock(&zi->i_truncate_mutex); | |
778 | ||
779 | if (iocb->ki_flags & IOCB_DIRECT) { | |
780 | size_t count = iov_iter_count(to); | |
781 | ||
782 | if ((iocb->ki_pos | count) & (sb->s_blocksize - 1)) { | |
783 | ret = -EINVAL; | |
784 | goto inode_unlock; | |
785 | } | |
786 | file_accessed(iocb->ki_filp); | |
787 | ret = iomap_dio_rw(iocb, to, &zonefs_iomap_ops, | |
788 | &zonefs_read_dio_ops, is_sync_kiocb(iocb)); | |
789 | } else { | |
790 | ret = generic_file_read_iter(iocb, to); | |
791 | if (ret == -EIO) | |
792 | zonefs_io_error(inode, false); | |
793 | } | |
794 | ||
795 | inode_unlock: | |
796 | inode_unlock_shared(inode); | |
797 | ||
798 | return ret; | |
799 | } | |
800 | ||
801 | static const struct file_operations zonefs_file_operations = { | |
802 | .open = generic_file_open, | |
803 | .fsync = zonefs_file_fsync, | |
804 | .mmap = zonefs_file_mmap, | |
805 | .llseek = zonefs_file_llseek, | |
806 | .read_iter = zonefs_file_read_iter, | |
807 | .write_iter = zonefs_file_write_iter, | |
808 | .splice_read = generic_file_splice_read, | |
809 | .splice_write = iter_file_splice_write, | |
810 | .iopoll = iomap_dio_iopoll, | |
811 | }; | |
812 | ||
813 | static struct kmem_cache *zonefs_inode_cachep; | |
814 | ||
815 | static struct inode *zonefs_alloc_inode(struct super_block *sb) | |
816 | { | |
817 | struct zonefs_inode_info *zi; | |
818 | ||
819 | zi = kmem_cache_alloc(zonefs_inode_cachep, GFP_KERNEL); | |
820 | if (!zi) | |
821 | return NULL; | |
822 | ||
823 | inode_init_once(&zi->i_vnode); | |
824 | mutex_init(&zi->i_truncate_mutex); | |
825 | init_rwsem(&zi->i_mmap_sem); | |
826 | ||
827 | return &zi->i_vnode; | |
828 | } | |
829 | ||
830 | static void zonefs_free_inode(struct inode *inode) | |
831 | { | |
832 | kmem_cache_free(zonefs_inode_cachep, ZONEFS_I(inode)); | |
833 | } | |
834 | ||
835 | /* | |
836 | * File system stat. | |
837 | */ | |
838 | static int zonefs_statfs(struct dentry *dentry, struct kstatfs *buf) | |
839 | { | |
840 | struct super_block *sb = dentry->d_sb; | |
841 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
842 | enum zonefs_ztype t; | |
843 | u64 fsid; | |
844 | ||
845 | buf->f_type = ZONEFS_MAGIC; | |
846 | buf->f_bsize = sb->s_blocksize; | |
847 | buf->f_namelen = ZONEFS_NAME_MAX; | |
848 | ||
849 | spin_lock(&sbi->s_lock); | |
850 | ||
851 | buf->f_blocks = sbi->s_blocks; | |
852 | if (WARN_ON(sbi->s_used_blocks > sbi->s_blocks)) | |
853 | buf->f_bfree = 0; | |
854 | else | |
855 | buf->f_bfree = buf->f_blocks - sbi->s_used_blocks; | |
856 | buf->f_bavail = buf->f_bfree; | |
857 | ||
858 | for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) { | |
859 | if (sbi->s_nr_files[t]) | |
860 | buf->f_files += sbi->s_nr_files[t] + 1; | |
861 | } | |
862 | buf->f_ffree = 0; | |
863 | ||
864 | spin_unlock(&sbi->s_lock); | |
865 | ||
866 | fsid = le64_to_cpup((void *)sbi->s_uuid.b) ^ | |
867 | le64_to_cpup((void *)sbi->s_uuid.b + sizeof(u64)); | |
868 | buf->f_fsid.val[0] = (u32)fsid; | |
869 | buf->f_fsid.val[1] = (u32)(fsid >> 32); | |
870 | ||
871 | return 0; | |
872 | } | |
873 | ||
874 | enum { | |
875 | Opt_errors_ro, Opt_errors_zro, Opt_errors_zol, Opt_errors_repair, | |
876 | Opt_err, | |
877 | }; | |
878 | ||
879 | static const match_table_t tokens = { | |
880 | { Opt_errors_ro, "errors=remount-ro"}, | |
881 | { Opt_errors_zro, "errors=zone-ro"}, | |
882 | { Opt_errors_zol, "errors=zone-offline"}, | |
883 | { Opt_errors_repair, "errors=repair"}, | |
884 | { Opt_err, NULL} | |
885 | }; | |
886 | ||
887 | static int zonefs_parse_options(struct super_block *sb, char *options) | |
888 | { | |
889 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
890 | substring_t args[MAX_OPT_ARGS]; | |
891 | char *p; | |
892 | ||
893 | if (!options) | |
894 | return 0; | |
895 | ||
896 | while ((p = strsep(&options, ",")) != NULL) { | |
897 | int token; | |
898 | ||
899 | if (!*p) | |
900 | continue; | |
901 | ||
902 | token = match_token(p, tokens, args); | |
903 | switch (token) { | |
904 | case Opt_errors_ro: | |
905 | sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK; | |
906 | sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_RO; | |
907 | break; | |
908 | case Opt_errors_zro: | |
909 | sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK; | |
910 | sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZRO; | |
911 | break; | |
912 | case Opt_errors_zol: | |
913 | sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK; | |
914 | sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_ZOL; | |
915 | break; | |
916 | case Opt_errors_repair: | |
917 | sbi->s_mount_opts &= ~ZONEFS_MNTOPT_ERRORS_MASK; | |
918 | sbi->s_mount_opts |= ZONEFS_MNTOPT_ERRORS_REPAIR; | |
919 | break; | |
920 | default: | |
921 | return -EINVAL; | |
922 | } | |
923 | } | |
924 | ||
925 | return 0; | |
926 | } | |
927 | ||
928 | static int zonefs_show_options(struct seq_file *seq, struct dentry *root) | |
929 | { | |
930 | struct zonefs_sb_info *sbi = ZONEFS_SB(root->d_sb); | |
931 | ||
932 | if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_RO) | |
933 | seq_puts(seq, ",errors=remount-ro"); | |
934 | if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZRO) | |
935 | seq_puts(seq, ",errors=zone-ro"); | |
936 | if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_ZOL) | |
937 | seq_puts(seq, ",errors=zone-offline"); | |
938 | if (sbi->s_mount_opts & ZONEFS_MNTOPT_ERRORS_REPAIR) | |
939 | seq_puts(seq, ",errors=repair"); | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
944 | static int zonefs_remount(struct super_block *sb, int *flags, char *data) | |
945 | { | |
946 | sync_filesystem(sb); | |
947 | ||
948 | return zonefs_parse_options(sb, data); | |
949 | } | |
950 | ||
951 | static const struct super_operations zonefs_sops = { | |
952 | .alloc_inode = zonefs_alloc_inode, | |
953 | .free_inode = zonefs_free_inode, | |
954 | .statfs = zonefs_statfs, | |
955 | .remount_fs = zonefs_remount, | |
956 | .show_options = zonefs_show_options, | |
957 | }; | |
958 | ||
959 | static const struct inode_operations zonefs_dir_inode_operations = { | |
960 | .lookup = simple_lookup, | |
961 | .setattr = zonefs_inode_setattr, | |
962 | }; | |
963 | ||
964 | static void zonefs_init_dir_inode(struct inode *parent, struct inode *inode, | |
965 | enum zonefs_ztype type) | |
966 | { | |
967 | struct super_block *sb = parent->i_sb; | |
968 | ||
969 | inode->i_ino = blkdev_nr_zones(sb->s_bdev->bd_disk) + type + 1; | |
970 | inode_init_owner(inode, parent, S_IFDIR | 0555); | |
971 | inode->i_op = &zonefs_dir_inode_operations; | |
972 | inode->i_fop = &simple_dir_operations; | |
973 | set_nlink(inode, 2); | |
974 | inc_nlink(parent); | |
975 | } | |
976 | ||
977 | static void zonefs_init_file_inode(struct inode *inode, struct blk_zone *zone, | |
978 | enum zonefs_ztype type) | |
979 | { | |
980 | struct super_block *sb = inode->i_sb; | |
981 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
982 | struct zonefs_inode_info *zi = ZONEFS_I(inode); | |
983 | ||
984 | inode->i_ino = zone->start >> sbi->s_zone_sectors_shift; | |
985 | inode->i_mode = S_IFREG | sbi->s_perm; | |
986 | ||
987 | zi->i_ztype = type; | |
988 | zi->i_zsector = zone->start; | |
989 | zi->i_max_size = min_t(loff_t, MAX_LFS_FILESIZE, | |
990 | zone->len << SECTOR_SHIFT); | |
ccf4ad7d | 991 | zi->i_wpoffset = zonefs_check_zone_condition(inode, zone, true, true); |
8dcc1a9d DLM |
992 | |
993 | inode->i_uid = sbi->s_uid; | |
994 | inode->i_gid = sbi->s_gid; | |
995 | inode->i_size = zi->i_wpoffset; | |
996 | inode->i_blocks = zone->len; | |
997 | ||
998 | inode->i_op = &zonefs_file_inode_operations; | |
999 | inode->i_fop = &zonefs_file_operations; | |
1000 | inode->i_mapping->a_ops = &zonefs_file_aops; | |
1001 | ||
1002 | sb->s_maxbytes = max(zi->i_max_size, sb->s_maxbytes); | |
1003 | sbi->s_blocks += zi->i_max_size >> sb->s_blocksize_bits; | |
1004 | sbi->s_used_blocks += zi->i_wpoffset >> sb->s_blocksize_bits; | |
1005 | } | |
1006 | ||
1007 | static struct dentry *zonefs_create_inode(struct dentry *parent, | |
1008 | const char *name, struct blk_zone *zone, | |
1009 | enum zonefs_ztype type) | |
1010 | { | |
1011 | struct inode *dir = d_inode(parent); | |
1012 | struct dentry *dentry; | |
1013 | struct inode *inode; | |
1014 | ||
1015 | dentry = d_alloc_name(parent, name); | |
1016 | if (!dentry) | |
1017 | return NULL; | |
1018 | ||
1019 | inode = new_inode(parent->d_sb); | |
1020 | if (!inode) | |
1021 | goto dput; | |
1022 | ||
1023 | inode->i_ctime = inode->i_mtime = inode->i_atime = dir->i_ctime; | |
1024 | if (zone) | |
1025 | zonefs_init_file_inode(inode, zone, type); | |
1026 | else | |
1027 | zonefs_init_dir_inode(dir, inode, type); | |
1028 | d_add(dentry, inode); | |
1029 | dir->i_size++; | |
1030 | ||
1031 | return dentry; | |
1032 | ||
1033 | dput: | |
1034 | dput(dentry); | |
1035 | ||
1036 | return NULL; | |
1037 | } | |
1038 | ||
1039 | struct zonefs_zone_data { | |
1040 | struct super_block *sb; | |
1041 | unsigned int nr_zones[ZONEFS_ZTYPE_MAX]; | |
1042 | struct blk_zone *zones; | |
1043 | }; | |
1044 | ||
1045 | /* | |
1046 | * Create a zone group and populate it with zone files. | |
1047 | */ | |
1048 | static int zonefs_create_zgroup(struct zonefs_zone_data *zd, | |
1049 | enum zonefs_ztype type) | |
1050 | { | |
1051 | struct super_block *sb = zd->sb; | |
1052 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
1053 | struct blk_zone *zone, *next, *end; | |
1054 | const char *zgroup_name; | |
1055 | char *file_name; | |
1056 | struct dentry *dir; | |
1057 | unsigned int n = 0; | |
1058 | int ret = -ENOMEM; | |
1059 | ||
1060 | /* If the group is empty, there is nothing to do */ | |
1061 | if (!zd->nr_zones[type]) | |
1062 | return 0; | |
1063 | ||
1064 | file_name = kmalloc(ZONEFS_NAME_MAX, GFP_KERNEL); | |
1065 | if (!file_name) | |
1066 | return -ENOMEM; | |
1067 | ||
1068 | if (type == ZONEFS_ZTYPE_CNV) | |
1069 | zgroup_name = "cnv"; | |
1070 | else | |
1071 | zgroup_name = "seq"; | |
1072 | ||
1073 | dir = zonefs_create_inode(sb->s_root, zgroup_name, NULL, type); | |
1074 | if (!dir) | |
1075 | goto free; | |
1076 | ||
1077 | /* | |
1078 | * The first zone contains the super block: skip it. | |
1079 | */ | |
1080 | end = zd->zones + blkdev_nr_zones(sb->s_bdev->bd_disk); | |
1081 | for (zone = &zd->zones[1]; zone < end; zone = next) { | |
1082 | ||
1083 | next = zone + 1; | |
1084 | if (zonefs_zone_type(zone) != type) | |
1085 | continue; | |
1086 | ||
1087 | /* | |
1088 | * For conventional zones, contiguous zones can be aggregated | |
1089 | * together to form larger files. Note that this overwrites the | |
1090 | * length of the first zone of the set of contiguous zones | |
1091 | * aggregated together. If one offline or read-only zone is | |
1092 | * found, assume that all zones aggregated have the same | |
1093 | * condition. | |
1094 | */ | |
1095 | if (type == ZONEFS_ZTYPE_CNV && | |
1096 | (sbi->s_features & ZONEFS_F_AGGRCNV)) { | |
1097 | for (; next < end; next++) { | |
1098 | if (zonefs_zone_type(next) != type) | |
1099 | break; | |
1100 | zone->len += next->len; | |
1101 | if (next->cond == BLK_ZONE_COND_READONLY && | |
1102 | zone->cond != BLK_ZONE_COND_OFFLINE) | |
1103 | zone->cond = BLK_ZONE_COND_READONLY; | |
1104 | else if (next->cond == BLK_ZONE_COND_OFFLINE) | |
1105 | zone->cond = BLK_ZONE_COND_OFFLINE; | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | /* | |
1110 | * Use the file number within its group as file name. | |
1111 | */ | |
1112 | snprintf(file_name, ZONEFS_NAME_MAX - 1, "%u", n); | |
1113 | if (!zonefs_create_inode(dir, file_name, zone, type)) | |
1114 | goto free; | |
1115 | ||
1116 | n++; | |
1117 | } | |
1118 | ||
1119 | zonefs_info(sb, "Zone group \"%s\" has %u file%s\n", | |
1120 | zgroup_name, n, n > 1 ? "s" : ""); | |
1121 | ||
1122 | sbi->s_nr_files[type] = n; | |
1123 | ret = 0; | |
1124 | ||
1125 | free: | |
1126 | kfree(file_name); | |
1127 | ||
1128 | return ret; | |
1129 | } | |
1130 | ||
1131 | static int zonefs_get_zone_info_cb(struct blk_zone *zone, unsigned int idx, | |
1132 | void *data) | |
1133 | { | |
1134 | struct zonefs_zone_data *zd = data; | |
1135 | ||
1136 | /* | |
1137 | * Count the number of usable zones: the first zone at index 0 contains | |
1138 | * the super block and is ignored. | |
1139 | */ | |
1140 | switch (zone->type) { | |
1141 | case BLK_ZONE_TYPE_CONVENTIONAL: | |
1142 | zone->wp = zone->start + zone->len; | |
1143 | if (idx) | |
1144 | zd->nr_zones[ZONEFS_ZTYPE_CNV]++; | |
1145 | break; | |
1146 | case BLK_ZONE_TYPE_SEQWRITE_REQ: | |
1147 | case BLK_ZONE_TYPE_SEQWRITE_PREF: | |
1148 | if (idx) | |
1149 | zd->nr_zones[ZONEFS_ZTYPE_SEQ]++; | |
1150 | break; | |
1151 | default: | |
1152 | zonefs_err(zd->sb, "Unsupported zone type 0x%x\n", | |
1153 | zone->type); | |
1154 | return -EIO; | |
1155 | } | |
1156 | ||
1157 | memcpy(&zd->zones[idx], zone, sizeof(struct blk_zone)); | |
1158 | ||
1159 | return 0; | |
1160 | } | |
1161 | ||
1162 | static int zonefs_get_zone_info(struct zonefs_zone_data *zd) | |
1163 | { | |
1164 | struct block_device *bdev = zd->sb->s_bdev; | |
1165 | int ret; | |
1166 | ||
1167 | zd->zones = kvcalloc(blkdev_nr_zones(bdev->bd_disk), | |
1168 | sizeof(struct blk_zone), GFP_KERNEL); | |
1169 | if (!zd->zones) | |
1170 | return -ENOMEM; | |
1171 | ||
1172 | /* Get zones information from the device */ | |
1173 | ret = blkdev_report_zones(bdev, 0, BLK_ALL_ZONES, | |
1174 | zonefs_get_zone_info_cb, zd); | |
1175 | if (ret < 0) { | |
1176 | zonefs_err(zd->sb, "Zone report failed %d\n", ret); | |
1177 | return ret; | |
1178 | } | |
1179 | ||
1180 | if (ret != blkdev_nr_zones(bdev->bd_disk)) { | |
1181 | zonefs_err(zd->sb, "Invalid zone report (%d/%u zones)\n", | |
1182 | ret, blkdev_nr_zones(bdev->bd_disk)); | |
1183 | return -EIO; | |
1184 | } | |
1185 | ||
1186 | return 0; | |
1187 | } | |
1188 | ||
1189 | static inline void zonefs_cleanup_zone_info(struct zonefs_zone_data *zd) | |
1190 | { | |
1191 | kvfree(zd->zones); | |
1192 | } | |
1193 | ||
1194 | /* | |
1195 | * Read super block information from the device. | |
1196 | */ | |
1197 | static int zonefs_read_super(struct super_block *sb) | |
1198 | { | |
1199 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
1200 | struct zonefs_super *super; | |
1201 | u32 crc, stored_crc; | |
1202 | struct page *page; | |
1203 | struct bio_vec bio_vec; | |
1204 | struct bio bio; | |
1205 | int ret; | |
1206 | ||
1207 | page = alloc_page(GFP_KERNEL); | |
1208 | if (!page) | |
1209 | return -ENOMEM; | |
1210 | ||
1211 | bio_init(&bio, &bio_vec, 1); | |
1212 | bio.bi_iter.bi_sector = 0; | |
1213 | bio.bi_opf = REQ_OP_READ; | |
1214 | bio_set_dev(&bio, sb->s_bdev); | |
1215 | bio_add_page(&bio, page, PAGE_SIZE, 0); | |
1216 | ||
1217 | ret = submit_bio_wait(&bio); | |
1218 | if (ret) | |
1219 | goto free_page; | |
1220 | ||
1221 | super = kmap(page); | |
1222 | ||
1223 | ret = -EINVAL; | |
1224 | if (le32_to_cpu(super->s_magic) != ZONEFS_MAGIC) | |
1225 | goto unmap; | |
1226 | ||
1227 | stored_crc = le32_to_cpu(super->s_crc); | |
1228 | super->s_crc = 0; | |
1229 | crc = crc32(~0U, (unsigned char *)super, sizeof(struct zonefs_super)); | |
1230 | if (crc != stored_crc) { | |
1231 | zonefs_err(sb, "Invalid checksum (Expected 0x%08x, got 0x%08x)", | |
1232 | crc, stored_crc); | |
1233 | goto unmap; | |
1234 | } | |
1235 | ||
1236 | sbi->s_features = le64_to_cpu(super->s_features); | |
1237 | if (sbi->s_features & ~ZONEFS_F_DEFINED_FEATURES) { | |
1238 | zonefs_err(sb, "Unknown features set 0x%llx\n", | |
1239 | sbi->s_features); | |
1240 | goto unmap; | |
1241 | } | |
1242 | ||
1243 | if (sbi->s_features & ZONEFS_F_UID) { | |
1244 | sbi->s_uid = make_kuid(current_user_ns(), | |
1245 | le32_to_cpu(super->s_uid)); | |
1246 | if (!uid_valid(sbi->s_uid)) { | |
1247 | zonefs_err(sb, "Invalid UID feature\n"); | |
1248 | goto unmap; | |
1249 | } | |
1250 | } | |
1251 | ||
1252 | if (sbi->s_features & ZONEFS_F_GID) { | |
1253 | sbi->s_gid = make_kgid(current_user_ns(), | |
1254 | le32_to_cpu(super->s_gid)); | |
1255 | if (!gid_valid(sbi->s_gid)) { | |
1256 | zonefs_err(sb, "Invalid GID feature\n"); | |
1257 | goto unmap; | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | if (sbi->s_features & ZONEFS_F_PERM) | |
1262 | sbi->s_perm = le32_to_cpu(super->s_perm); | |
1263 | ||
1264 | if (memchr_inv(super->s_reserved, 0, sizeof(super->s_reserved))) { | |
1265 | zonefs_err(sb, "Reserved area is being used\n"); | |
1266 | goto unmap; | |
1267 | } | |
1268 | ||
1269 | uuid_copy(&sbi->s_uuid, (uuid_t *)super->s_uuid); | |
1270 | ret = 0; | |
1271 | ||
1272 | unmap: | |
1273 | kunmap(page); | |
1274 | free_page: | |
1275 | __free_page(page); | |
1276 | ||
1277 | return ret; | |
1278 | } | |
1279 | ||
1280 | /* | |
1281 | * Check that the device is zoned. If it is, get the list of zones and create | |
1282 | * sub-directories and files according to the device zone configuration and | |
1283 | * format options. | |
1284 | */ | |
1285 | static int zonefs_fill_super(struct super_block *sb, void *data, int silent) | |
1286 | { | |
1287 | struct zonefs_zone_data zd; | |
1288 | struct zonefs_sb_info *sbi; | |
1289 | struct inode *inode; | |
1290 | enum zonefs_ztype t; | |
1291 | int ret; | |
1292 | ||
1293 | if (!bdev_is_zoned(sb->s_bdev)) { | |
1294 | zonefs_err(sb, "Not a zoned block device\n"); | |
1295 | return -EINVAL; | |
1296 | } | |
1297 | ||
1298 | /* | |
1299 | * Initialize super block information: the maximum file size is updated | |
1300 | * when the zone files are created so that the format option | |
1301 | * ZONEFS_F_AGGRCNV which increases the maximum file size of a file | |
1302 | * beyond the zone size is taken into account. | |
1303 | */ | |
1304 | sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); | |
1305 | if (!sbi) | |
1306 | return -ENOMEM; | |
1307 | ||
1308 | spin_lock_init(&sbi->s_lock); | |
1309 | sb->s_fs_info = sbi; | |
1310 | sb->s_magic = ZONEFS_MAGIC; | |
1311 | sb->s_maxbytes = 0; | |
1312 | sb->s_op = &zonefs_sops; | |
1313 | sb->s_time_gran = 1; | |
1314 | ||
1315 | /* | |
1316 | * The block size is set to the device physical sector size to ensure | |
1317 | * that write operations on 512e devices (512B logical block and 4KB | |
1318 | * physical block) are always aligned to the device physical blocks, | |
1319 | * as mandated by the ZBC/ZAC specifications. | |
1320 | */ | |
1321 | sb_set_blocksize(sb, bdev_physical_block_size(sb->s_bdev)); | |
1322 | sbi->s_zone_sectors_shift = ilog2(bdev_zone_sectors(sb->s_bdev)); | |
1323 | sbi->s_uid = GLOBAL_ROOT_UID; | |
1324 | sbi->s_gid = GLOBAL_ROOT_GID; | |
1325 | sbi->s_perm = 0640; | |
1326 | sbi->s_mount_opts = ZONEFS_MNTOPT_ERRORS_RO; | |
1327 | ||
1328 | ret = zonefs_read_super(sb); | |
1329 | if (ret) | |
1330 | return ret; | |
1331 | ||
1332 | ret = zonefs_parse_options(sb, data); | |
1333 | if (ret) | |
1334 | return ret; | |
1335 | ||
1336 | memset(&zd, 0, sizeof(struct zonefs_zone_data)); | |
1337 | zd.sb = sb; | |
1338 | ret = zonefs_get_zone_info(&zd); | |
1339 | if (ret) | |
1340 | goto cleanup; | |
1341 | ||
1342 | zonefs_info(sb, "Mounting %u zones", | |
1343 | blkdev_nr_zones(sb->s_bdev->bd_disk)); | |
1344 | ||
1345 | /* Create root directory inode */ | |
1346 | ret = -ENOMEM; | |
1347 | inode = new_inode(sb); | |
1348 | if (!inode) | |
1349 | goto cleanup; | |
1350 | ||
1351 | inode->i_ino = blkdev_nr_zones(sb->s_bdev->bd_disk); | |
1352 | inode->i_mode = S_IFDIR | 0555; | |
1353 | inode->i_ctime = inode->i_mtime = inode->i_atime = current_time(inode); | |
1354 | inode->i_op = &zonefs_dir_inode_operations; | |
1355 | inode->i_fop = &simple_dir_operations; | |
1356 | set_nlink(inode, 2); | |
1357 | ||
1358 | sb->s_root = d_make_root(inode); | |
1359 | if (!sb->s_root) | |
1360 | goto cleanup; | |
1361 | ||
1362 | /* Create and populate files in zone groups directories */ | |
1363 | for (t = 0; t < ZONEFS_ZTYPE_MAX; t++) { | |
1364 | ret = zonefs_create_zgroup(&zd, t); | |
1365 | if (ret) | |
1366 | break; | |
1367 | } | |
1368 | ||
1369 | cleanup: | |
1370 | zonefs_cleanup_zone_info(&zd); | |
1371 | ||
1372 | return ret; | |
1373 | } | |
1374 | ||
1375 | static struct dentry *zonefs_mount(struct file_system_type *fs_type, | |
1376 | int flags, const char *dev_name, void *data) | |
1377 | { | |
1378 | return mount_bdev(fs_type, flags, dev_name, data, zonefs_fill_super); | |
1379 | } | |
1380 | ||
1381 | static void zonefs_kill_super(struct super_block *sb) | |
1382 | { | |
1383 | struct zonefs_sb_info *sbi = ZONEFS_SB(sb); | |
1384 | ||
1385 | if (sb->s_root) | |
1386 | d_genocide(sb->s_root); | |
1387 | kill_block_super(sb); | |
1388 | kfree(sbi); | |
1389 | } | |
1390 | ||
1391 | /* | |
1392 | * File system definition and registration. | |
1393 | */ | |
1394 | static struct file_system_type zonefs_type = { | |
1395 | .owner = THIS_MODULE, | |
1396 | .name = "zonefs", | |
1397 | .mount = zonefs_mount, | |
1398 | .kill_sb = zonefs_kill_super, | |
1399 | .fs_flags = FS_REQUIRES_DEV, | |
1400 | }; | |
1401 | ||
1402 | static int __init zonefs_init_inodecache(void) | |
1403 | { | |
1404 | zonefs_inode_cachep = kmem_cache_create("zonefs_inode_cache", | |
1405 | sizeof(struct zonefs_inode_info), 0, | |
1406 | (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT), | |
1407 | NULL); | |
1408 | if (zonefs_inode_cachep == NULL) | |
1409 | return -ENOMEM; | |
1410 | return 0; | |
1411 | } | |
1412 | ||
1413 | static void zonefs_destroy_inodecache(void) | |
1414 | { | |
1415 | /* | |
1416 | * Make sure all delayed rcu free inodes are flushed before we | |
1417 | * destroy the inode cache. | |
1418 | */ | |
1419 | rcu_barrier(); | |
1420 | kmem_cache_destroy(zonefs_inode_cachep); | |
1421 | } | |
1422 | ||
1423 | static int __init zonefs_init(void) | |
1424 | { | |
1425 | int ret; | |
1426 | ||
1427 | BUILD_BUG_ON(sizeof(struct zonefs_super) != ZONEFS_SUPER_SIZE); | |
1428 | ||
1429 | ret = zonefs_init_inodecache(); | |
1430 | if (ret) | |
1431 | return ret; | |
1432 | ||
1433 | ret = register_filesystem(&zonefs_type); | |
1434 | if (ret) { | |
1435 | zonefs_destroy_inodecache(); | |
1436 | return ret; | |
1437 | } | |
1438 | ||
1439 | return 0; | |
1440 | } | |
1441 | ||
1442 | static void __exit zonefs_exit(void) | |
1443 | { | |
1444 | zonefs_destroy_inodecache(); | |
1445 | unregister_filesystem(&zonefs_type); | |
1446 | } | |
1447 | ||
1448 | MODULE_AUTHOR("Damien Le Moal"); | |
1449 | MODULE_DESCRIPTION("Zone file system for zoned block devices"); | |
1450 | MODULE_LICENSE("GPL"); | |
1451 | module_init(zonefs_init); | |
1452 | module_exit(zonefs_exit); |