tools headers UAPI: Sync openat2.h with the kernel sources
[linux-2.6-microblaze.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                         truncate_inode_pages_final(&inode->i_data);
150
151                         if (test_cflag(COMMIT_Freewmap, inode))
152                                 jfs_free_zero_link(inode);
153
154                         diFree(inode);
155
156                         /*
157                          * Free the inode from the quota allocation.
158                          */
159                         dquot_free_inode(inode);
160                 }
161         } else {
162                 truncate_inode_pages_final(&inode->i_data);
163         }
164         clear_inode(inode);
165         dquot_drop(inode);
166
167         BUG_ON(!list_empty(&ji->anon_inode_list));
168
169         spin_lock_irq(&ji->ag_lock);
170         if (ji->active_ag != -1) {
171                 struct bmap *bmap = JFS_SBI(inode->i_sb)->bmap;
172                 atomic_dec(&bmap->db_active[ji->active_ag]);
173                 ji->active_ag = -1;
174         }
175         spin_unlock_irq(&ji->ag_lock);
176 }
177
178 void jfs_dirty_inode(struct inode *inode, int flags)
179 {
180         static int noisy = 5;
181
182         if (isReadOnly(inode)) {
183                 if (!special_file(inode->i_mode) && noisy) {
184                         /* kernel allows writes to devices on read-only
185                          * partitions and may try to mark inode dirty
186                          */
187                         jfs_err("jfs_dirty_inode called on read-only volume");
188                         jfs_err("Is remount racy?");
189                         noisy--;
190                 }
191                 return;
192         }
193
194         set_cflag(COMMIT_Dirty, inode);
195 }
196
197 int jfs_get_block(struct inode *ip, sector_t lblock,
198                   struct buffer_head *bh_result, int create)
199 {
200         s64 lblock64 = lblock;
201         int rc = 0;
202         xad_t xad;
203         s64 xaddr;
204         int xflag;
205         s32 xlen = bh_result->b_size >> ip->i_blkbits;
206
207         /*
208          * Take appropriate lock on inode
209          */
210         if (create)
211                 IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
212         else
213                 IREAD_LOCK(ip, RDWRLOCK_NORMAL);
214
215         if (((lblock64 << ip->i_sb->s_blocksize_bits) < ip->i_size) &&
216             (!xtLookup(ip, lblock64, xlen, &xflag, &xaddr, &xlen, 0)) &&
217             xaddr) {
218                 if (xflag & XAD_NOTRECORDED) {
219                         if (!create)
220                                 /*
221                                  * Allocated but not recorded, read treats
222                                  * this as a hole
223                                  */
224                                 goto unlock;
225 #ifdef _JFS_4K
226                         XADoffset(&xad, lblock64);
227                         XADlength(&xad, xlen);
228                         XADaddress(&xad, xaddr);
229 #else                           /* _JFS_4K */
230                         /*
231                          * As long as block size = 4K, this isn't a problem.
232                          * We should mark the whole page not ABNR, but how
233                          * will we know to mark the other blocks BH_New?
234                          */
235                         BUG();
236 #endif                          /* _JFS_4K */
237                         rc = extRecord(ip, &xad);
238                         if (rc)
239                                 goto unlock;
240                         set_buffer_new(bh_result);
241                 }
242
243                 map_bh(bh_result, ip->i_sb, xaddr);
244                 bh_result->b_size = xlen << ip->i_blkbits;
245                 goto unlock;
246         }
247         if (!create)
248                 goto unlock;
249
250         /*
251          * Allocate a new block
252          */
253 #ifdef _JFS_4K
254         if ((rc = extHint(ip, lblock64 << ip->i_sb->s_blocksize_bits, &xad)))
255                 goto unlock;
256         rc = extAlloc(ip, xlen, lblock64, &xad, false);
257         if (rc)
258                 goto unlock;
259
260         set_buffer_new(bh_result);
261         map_bh(bh_result, ip->i_sb, addressXAD(&xad));
262         bh_result->b_size = lengthXAD(&xad) << ip->i_blkbits;
263
264 #else                           /* _JFS_4K */
265         /*
266          * We need to do whatever it takes to keep all but the last buffers
267          * in 4K pages - see jfs_write.c
268          */
269         BUG();
270 #endif                          /* _JFS_4K */
271
272       unlock:
273         /*
274          * Release lock on inode
275          */
276         if (create)
277                 IWRITE_UNLOCK(ip);
278         else
279                 IREAD_UNLOCK(ip);
280         return rc;
281 }
282
283 static int jfs_writepage(struct page *page, struct writeback_control *wbc)
284 {
285         return block_write_full_page(page, jfs_get_block, wbc);
286 }
287
288 static int jfs_writepages(struct address_space *mapping,
289                         struct writeback_control *wbc)
290 {
291         return mpage_writepages(mapping, wbc, jfs_get_block);
292 }
293
294 static int jfs_readpage(struct file *file, struct page *page)
295 {
296         return mpage_readpage(page, jfs_get_block);
297 }
298
299 static void jfs_readahead(struct readahead_control *rac)
300 {
301         mpage_readahead(rac, jfs_get_block);
302 }
303
304 static void jfs_write_failed(struct address_space *mapping, loff_t to)
305 {
306         struct inode *inode = mapping->host;
307
308         if (to > inode->i_size) {
309                 truncate_pagecache(inode, inode->i_size);
310                 jfs_truncate(inode);
311         }
312 }
313
314 static int jfs_write_begin(struct file *file, struct address_space *mapping,
315                                 loff_t pos, unsigned len, unsigned flags,
316                                 struct page **pagep, void **fsdata)
317 {
318         int ret;
319
320         ret = nobh_write_begin(mapping, pos, len, flags, pagep, fsdata,
321                                 jfs_get_block);
322         if (unlikely(ret))
323                 jfs_write_failed(mapping, pos + len);
324
325         return ret;
326 }
327
328 static sector_t jfs_bmap(struct address_space *mapping, sector_t block)
329 {
330         return generic_block_bmap(mapping, block, jfs_get_block);
331 }
332
333 static ssize_t jfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
334 {
335         struct file *file = iocb->ki_filp;
336         struct address_space *mapping = file->f_mapping;
337         struct inode *inode = file->f_mapping->host;
338         size_t count = iov_iter_count(iter);
339         ssize_t ret;
340
341         ret = blockdev_direct_IO(iocb, inode, iter, jfs_get_block);
342
343         /*
344          * In case of error extending write may have instantiated a few
345          * blocks outside i_size. Trim these off again.
346          */
347         if (unlikely(iov_iter_rw(iter) == WRITE && ret < 0)) {
348                 loff_t isize = i_size_read(inode);
349                 loff_t end = iocb->ki_pos + count;
350
351                 if (end > isize)
352                         jfs_write_failed(mapping, end);
353         }
354
355         return ret;
356 }
357
358 const struct address_space_operations jfs_aops = {
359         .readpage       = jfs_readpage,
360         .readahead      = jfs_readahead,
361         .writepage      = jfs_writepage,
362         .writepages     = jfs_writepages,
363         .write_begin    = jfs_write_begin,
364         .write_end      = nobh_write_end,
365         .bmap           = jfs_bmap,
366         .direct_IO      = jfs_direct_IO,
367 };
368
369 /*
370  * Guts of jfs_truncate.  Called with locks already held.  Can be called
371  * with directory for truncating directory index table.
372  */
373 void jfs_truncate_nolock(struct inode *ip, loff_t length)
374 {
375         loff_t newsize;
376         tid_t tid;
377
378         ASSERT(length >= 0);
379
380         if (test_cflag(COMMIT_Nolink, ip)) {
381                 xtTruncate(0, ip, length, COMMIT_WMAP);
382                 return;
383         }
384
385         do {
386                 tid = txBegin(ip->i_sb, 0);
387
388                 /*
389                  * The commit_mutex cannot be taken before txBegin.
390                  * txBegin may block and there is a chance the inode
391                  * could be marked dirty and need to be committed
392                  * before txBegin unblocks
393                  */
394                 mutex_lock(&JFS_IP(ip)->commit_mutex);
395
396                 newsize = xtTruncate(tid, ip, length,
397                                      COMMIT_TRUNCATE | COMMIT_PWMAP);
398                 if (newsize < 0) {
399                         txEnd(tid);
400                         mutex_unlock(&JFS_IP(ip)->commit_mutex);
401                         break;
402                 }
403
404                 ip->i_mtime = ip->i_ctime = current_time(ip);
405                 mark_inode_dirty(ip);
406
407                 txCommit(tid, 1, &ip, 0);
408                 txEnd(tid);
409                 mutex_unlock(&JFS_IP(ip)->commit_mutex);
410         } while (newsize > length);     /* Truncate isn't always atomic */
411 }
412
413 void jfs_truncate(struct inode *ip)
414 {
415         jfs_info("jfs_truncate: size = 0x%lx", (ulong) ip->i_size);
416
417         nobh_truncate_page(ip->i_mapping, ip->i_size, jfs_get_block);
418
419         IWRITE_LOCK(ip, RDWRLOCK_NORMAL);
420         jfs_truncate_nolock(ip, ip->i_size);
421         IWRITE_UNLOCK(ip);
422 }