overlayfs: Implement splice-read
[linux-block.git] / fs / jfs / inode.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *   Copyright (C) International Business Machines Corp., 2000-2004
4  *   Portions Copyright (C) Christoph Hellwig, 2001-2002
5  */
6
7 #include <linux/fs.h>
8 #include <linux/mpage.h>
9 #include <linux/buffer_head.h>
10 #include <linux/pagemap.h>
11 #include <linux/quotaops.h>
12 #include <linux/uio.h>
13 #include <linux/writeback.h>
14 #include "jfs_incore.h"
15 #include "jfs_inode.h"
16 #include "jfs_filsys.h"
17 #include "jfs_imap.h"
18 #include "jfs_extent.h"
19 #include "jfs_unicode.h"
20 #include "jfs_debug.h"
21 #include "jfs_dmap.h"
22
23
24 struct inode *jfs_iget(struct super_block *sb, unsigned long ino)
25 {
26         struct inode *inode;
27         int ret;
28
29         inode = iget_locked(sb, ino);
30         if (!inode)
31                 return ERR_PTR(-ENOMEM);
32         if (!(inode->i_state & I_NEW))
33                 return inode;
34
35         ret = diRead(inode);
36         if (ret < 0) {
37                 iget_failed(inode);
38                 return ERR_PTR(ret);
39         }
40
41         if (S_ISREG(inode->i_mode)) {
42                 inode->i_op = &jfs_file_inode_operations;
43                 inode->i_fop = &jfs_file_operations;
44                 inode->i_mapping->a_ops = &jfs_aops;
45         } else if (S_ISDIR(inode->i_mode)) {
46                 inode->i_op = &jfs_dir_inode_operations;
47                 inode->i_fop = &jfs_dir_operations;
48         } else if (S_ISLNK(inode->i_mode)) {
49                 if (inode->i_size >= IDATASIZE) {
50                         inode->i_op = &page_symlink_inode_operations;
51                         inode_nohighmem(inode);
52                         inode->i_mapping->a_ops = &jfs_aops;
53                 } else {
54                         inode->i_op = &jfs_fast_symlink_inode_operations;
55                         inode->i_link = JFS_IP(inode)->i_inline;
56                         /*
57                          * The inline data should be null-terminated, but
58                          * don't let on-disk corruption crash the kernel
59                          */
60                         inode->i_link[inode->i_size] = '\0';
61                 }
62         } else {
63                 inode->i_op = &jfs_file_inode_operations;
64                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
65         }
66         unlock_new_inode(inode);
67         return inode;
68 }
69
70 /*
71  * Workhorse of both fsync & write_inode
72  */
73 int jfs_commit_inode(struct inode *inode, int wait)
74 {
75         int rc = 0;
76         tid_t tid;
77         static int noisy = 5;
78
79         jfs_info("In jfs_commit_inode, inode = 0x%p", inode);
80
81         /*
82          * Don't commit if inode has been committed since last being
83          * marked dirty, or if it has been deleted.
84          */
85         if (inode->i_nlink == 0 || !test_cflag(COMMIT_Dirty, inode))
86                 return 0;
87
88         if (isReadOnly(inode)) {
89                 /* kernel allows writes to devices on read-only
90                  * partitions and may think inode is dirty
91                  */
92                 if (!special_file(inode->i_mode) && noisy) {
93                         jfs_err("jfs_commit_inode(0x%p) called on read-only volume",
94                                 inode);
95                         jfs_err("Is remount racy?");
96                         noisy--;
97                 }
98                 return 0;
99         }
100
101         tid = txBegin(inode->i_sb, COMMIT_INODE);
102         mutex_lock(&JFS_IP(inode)->commit_mutex);
103
104         /*
105          * Retest inode state after taking commit_mutex
106          */
107         if (inode->i_nlink && test_cflag(COMMIT_Dirty, inode))
108                 rc = txCommit(tid, 1, &inode, wait ? COMMIT_SYNC : 0);
109
110         txEnd(tid);
111         mutex_unlock(&JFS_IP(inode)->commit_mutex);
112         return rc;
113 }
114
115 int jfs_write_inode(struct inode *inode, struct writeback_control *wbc)
116 {
117         int wait = wbc->sync_mode == WB_SYNC_ALL;
118
119         if (inode->i_nlink == 0)
120                 return 0;
121         /*
122          * If COMMIT_DIRTY is not set, the inode isn't really dirty.
123          * It has been committed since the last change, but was still
124          * on the dirty inode list.
125          */
126         if (!test_cflag(COMMIT_Dirty, inode)) {
127                 /* Make sure committed changes hit the disk */
128                 jfs_flush_journal(JFS_SBI(inode->i_sb)->log, wait);
129                 return 0;
130         }
131
132         if (jfs_commit_inode(inode, wait)) {
133                 jfs_err("jfs_write_inode: jfs_commit_inode failed!");
134                 return -EIO;
135         } else
136                 return 0;
137 }
138
139 void jfs_evict_inode(struct inode *inode)
140 {
141         struct jfs_inode_info *ji = JFS_IP(inode);
142
143         jfs_info("In jfs_evict_inode, inode = 0x%p", inode);
144
145         if (!inode->i_nlink && !is_bad_inode(inode)) {
146                 dquot_initialize(inode);
147
148                 if (JFS_IP(inode)->fileset == FILESYSTEM_I) {
149                         struct inode *ipimap = JFS_SBI(inode->i_sb)->ipimap;
150                         truncate_inode_pages_final(&inode->i_data);
151
152                         if (test_cflag(COMMIT_Freewmap, inode))
153                                 jfs_free_zero_link(inode);
154
155                         if (ipimap && JFS_IP(ipimap)->i_imap)
156                                 diFree(inode);
157
158                         /*
159                          * Free the inode from the quota allocation.
160                          */
161                         dquot_free_inode(inode);
162                 }
163         } else {
164                 truncate_inode_pages_final(&inode->i_data);
165         }
166         clear_inode(inode);
167         dquot_drop(inode);
168
169         BUG_ON(!list_empty(&ji->anon_inode_list));
170
171         spin_lock_irq(&ji->ag_lock);
172         if (ji->active_ag != -1) {
173                 struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap;
174                 atomic_dec(&bmap->db_active[ji->active_ag]);
175                 ji->active_ag = -1;
176         }
177         spin_unlock_irq(&ji->ag_lock);
178 }
179
180 void jfs_dirty_inode(struct inode *inode, int flags)
181 {
182         static int noisy = 5;
183
184         if (isReadOnly(inode)) {
185                 if (!special_file(inode->i_mode) && noisy) {
186                         /* kernel allows writes to devices on read-only
187                          * partitions and may try to mark inode dirty
188                          */
189                         jfs_err("jfs_dirty_inode called on read-only volume");
190                         jfs_err("Is remount racy?");
191                         noisy--;
192                 }
193                 return;
194         }
195
196         set_cflag(COMMIT_Dirty, inode);
197 }
198
199 int jfs_get_block(struct inode *ip, sector_t lblock,
200                   struct buffer_head *bh_result, int create)
201 {
202         s64 lblock64 = lblock;
203         int rc = 0;
204         xad_t xad;
205         s64 xaddr;
206         int xflag;
207         s32 xlen = bh_result->b_size >> ip->i_blkbits;
208
209         /*
210          * Take appropriate lock on inode
211          */
212         if (create)
213                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
214         else
215                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
216
217         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
218             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
219             xaddr) {
220                 if (xflag & XAD_NOTRECORDED) {
221                         if (!create)
222                                 /*
223                                  * Allocated but not recorded, read treats
224                                  * this as a hole
225                                  */
226                                 goto unlock;
227                         XADoffset(&xad, lblock64);
228                         XADlength(&xad, xlen);
229                         XADaddress(&xad, xaddr);
230                         rc = extRecord(ip, &xad);
231                         if (rc)
232                                 goto unlock;
233                         set_buffer_new(bh_result);
234                 }
235
236                 map_bh(bh_result, ip->i_sb, xaddr);
237                 bh_result->b_size = xlen << ip->i_blkbits;
238                 goto unlock;
239         }
240         if (!create)
241                 goto unlock;
242
243         /*
244          * Allocate a new block
245          */
246         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
247                 goto unlock;
248         rc = extAlloc(ip, xlen, lblock64, &xad, false);
249         if (rc)
250                 goto unlock;
251
252         set_buffer_new(bh_result);
253         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
254         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
255
256       unlock:
257         /*
258          * Release lock on inode
259          */
260         if (create)
261                 IWRITE_UNLOCK(ip);
262         else
263                 IREAD_UNLOCK(ip);
264         return rc;
265 }
266
267 static int jfs_writepages(struct address_space *mapping,
268                         struct writeback_control *wbc)
269 {
270         return mpage_writepages(mapping, wbc, jfs_get_block);
271 }
272
273 static int jfs_read_folio(struct file *file, struct folio *folio)
274 {
275         return mpage_read_folio(folio, jfs_get_block);
276 }
277
278 static void jfs_readahead(struct readahead_control *rac)
279 {
280         mpage_readahead(rac, jfs_get_block);
281 }
282
283 static void jfs_write_failed(struct address_space *mapping, loff_t to)
284 {
285         struct inode *inode = mapping->host;
286
287         if (to > inode->i_size) {
288                 truncate_pagecache(inode, inode->i_size);
289                 jfs_truncate(inode);
290         }
291 }
292
293 static int jfs_write_begin(struct file *file, struct address_space *mapping,
294                                 loff_t pos, unsigned len,
295                                 struct page **pagep, void **fsdata)
296 {
297         int ret;
298
299         ret = block_write_begin(mapping, pos, len, pagep, jfs_get_block);
300         if (unlikely(ret))
301                 jfs_write_failed(mapping, pos + len);
302
303         return ret;
304 }
305
306 static int jfs_write_end(struct file *file, struct address_space *mapping,
307                 loff_t pos, unsigned len, unsigned copied, struct page *page,
308                 void *fsdata)
309 {
310         int ret;
311
312         ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
313         if (ret < len)
314                 jfs_write_failed(mapping, pos + len);
315         return ret;
316 }
317
318 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
319 {
320         return generic_block_bmap(mapping, block, jfs_get_block);
321 }
322
323 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
324 {
325         struct file *file = iocb->ki_filp;
326         struct address_space *mapping = file->f_mapping;
327         struct inode *inode = file->f_mapping->host;
328         size_t count = iov_iter_count(iter);
329         ssize_t ret;
330
331         ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block);
332
333         /*
334          * In case of error extending write may have instantiated a few
335          * blocks outside i_size. Trim these off again.
336          */
337         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
338                 loff_t isize = i_size_read(inode);
339                 loff_t end = iocb->ki_pos + count;
340
341                 if (end > isize)
342                         jfs_write_failed(mapping, end);
343         }
344
345         return ret;
346 }
347
348 const struct address_space_operations jfs_aops = {
349         .dirty_folio    = block_dirty_folio,
350         .invalidate_folio = block_invalidate_folio,
351         .read_folio     = jfs_read_folio,
352         .readahead      = jfs_readahead,
353         .writepages     = jfs_writepages,
354         .write_begin    = jfs_write_begin,
355         .write_end      = jfs_write_end,
356         .bmap           = jfs_bmap,
357         .direct_IO      = jfs_direct_IO,
358         .migrate_folio  = buffer_migrate_folio,
359 };
360
361 /*
362  * Guts of jfs_truncate.  Called with locks already held.  Can be called
363  * with directory for truncating directory index table.
364  */
365 void jfs_truncate_nolock(struct inode *ip, loff_t length)
366 {
367         loff_t newsize;
368         tid_t tid;
369
370         ASSERT(length >= 0);
371
372         if (test_cflag(COMMIT_Nolink, ip)) {
373                 xtTruncate(0, ip, length, COMMIT_WMAP);
374                 return;
375         }
376
377         do {
378                 tid = txBegin(ip->i_sb, 0);
379
380                 /*
381                  * The commit_mutex cannot be taken before txBegin.
382                  * txBegin may block and there is a chance the inode
383                  * could be marked dirty and need to be committed
384                  * before txBegin unblocks
385                  */
386                 mutex_lock(&JFS_IP(ip)->commit_mutex);
387
388                 newsize = xtTruncate(tid, ip, length,
389                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
390                 if (newsize < 0) {
391                         txEnd(tid);
392                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
393                         break;
394                 }
395
396                 ip->i_mtime = ip->i_ctime = current_time(ip);
397                 mark_inode_dirty(ip);
398
399                 txCommit(tid, 1, &ip, 0);
400                 txEnd(tid);
401                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
402         } while (newsize > length);     /* Truncate isn't always atomic */
403 }
404
405 void jfs_truncate(struct inode *ip)
406 {
407         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
408
409         block_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
410
411         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
412         jfs_truncate_nolock(ip, ip->i_size);
413         IWRITE_UNLOCK(ip);
414 }