diff options
Diffstat (limited to 'fs/gfs2/ops_file.c')
-rw-r--r-- | fs/gfs2/ops_file.c | 1597 |
1 files changed, 1597 insertions, 0 deletions
diff --git a/fs/gfs2/ops_file.c b/fs/gfs2/ops_file.c new file mode 100644 index 000000000000..1e6f51b74f44 --- /dev/null +++ b/fs/gfs2/ops_file.c | |||
@@ -0,0 +1,1597 @@ | |||
1 | /* | ||
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | ||
3 | * Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved. | ||
4 | * | ||
5 | * This copyrighted material is made available to anyone wishing to use, | ||
6 | * modify, copy, or redistribute it subject to the terms and conditions | ||
7 | * of the GNU General Public License v.2. | ||
8 | */ | ||
9 | |||
10 | #include <linux/sched.h> | ||
11 | #include <linux/slab.h> | ||
12 | #include <linux/spinlock.h> | ||
13 | #include <linux/completion.h> | ||
14 | #include <linux/buffer_head.h> | ||
15 | #include <linux/pagemap.h> | ||
16 | #include <linux/uio.h> | ||
17 | #include <linux/blkdev.h> | ||
18 | #include <linux/mm.h> | ||
19 | #include <linux/smp_lock.h> | ||
20 | #include <linux/gfs2_ioctl.h> | ||
21 | #include <asm/semaphore.h> | ||
22 | #include <asm/uaccess.h> | ||
23 | |||
24 | #include "gfs2.h" | ||
25 | #include "bmap.h" | ||
26 | #include "dir.h" | ||
27 | #include "glock.h" | ||
28 | #include "glops.h" | ||
29 | #include "inode.h" | ||
30 | #include "jdata.h" | ||
31 | #include "lm.h" | ||
32 | #include "log.h" | ||
33 | #include "meta_io.h" | ||
34 | #include "ops_file.h" | ||
35 | #include "ops_vm.h" | ||
36 | #include "quota.h" | ||
37 | #include "rgrp.h" | ||
38 | #include "trans.h" | ||
39 | |||
40 | /* "bad" is for NFS support */ | ||
41 | struct filldir_bad_entry { | ||
42 | char *fbe_name; | ||
43 | unsigned int fbe_length; | ||
44 | uint64_t fbe_offset; | ||
45 | struct gfs2_inum fbe_inum; | ||
46 | unsigned int fbe_type; | ||
47 | }; | ||
48 | |||
49 | struct filldir_bad { | ||
50 | struct gfs2_sbd *fdb_sbd; | ||
51 | |||
52 | struct filldir_bad_entry *fdb_entry; | ||
53 | unsigned int fdb_entry_num; | ||
54 | unsigned int fdb_entry_off; | ||
55 | |||
56 | char *fdb_name; | ||
57 | unsigned int fdb_name_size; | ||
58 | unsigned int fdb_name_off; | ||
59 | }; | ||
60 | |||
61 | /* For regular, non-NFS */ | ||
62 | struct filldir_reg { | ||
63 | struct gfs2_sbd *fdr_sbd; | ||
64 | int fdr_prefetch; | ||
65 | |||
66 | filldir_t fdr_filldir; | ||
67 | void *fdr_opaque; | ||
68 | }; | ||
69 | |||
70 | typedef ssize_t(*do_rw_t) (struct file *file, | ||
71 | char __user *buf, | ||
72 | size_t size, loff_t *offset, | ||
73 | unsigned int num_gh, struct gfs2_holder *ghs); | ||
74 | |||
75 | /** | ||
76 | * gfs2_llseek - seek to a location in a file | ||
77 | * @file: the file | ||
78 | * @offset: the offset | ||
79 | * @origin: Where to seek from (SEEK_SET, SEEK_CUR, or SEEK_END) | ||
80 | * | ||
81 | * SEEK_END requires the glock for the file because it references the | ||
82 | * file's size. | ||
83 | * | ||
84 | * Returns: The new offset, or errno | ||
85 | */ | ||
86 | |||
87 | static loff_t gfs2_llseek(struct file *file, loff_t offset, int origin) | ||
88 | { | ||
89 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
90 | struct gfs2_holder i_gh; | ||
91 | loff_t error; | ||
92 | |||
93 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
94 | |||
95 | if (origin == 2) { | ||
96 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, | ||
97 | &i_gh); | ||
98 | if (!error) { | ||
99 | error = remote_llseek(file, offset, origin); | ||
100 | gfs2_glock_dq_uninit(&i_gh); | ||
101 | } | ||
102 | } else | ||
103 | error = remote_llseek(file, offset, origin); | ||
104 | |||
105 | return error; | ||
106 | } | ||
107 | |||
108 | static inline unsigned int vma2state(struct vm_area_struct *vma) | ||
109 | { | ||
110 | if ((vma->vm_flags & (VM_MAYWRITE | VM_MAYSHARE)) == | ||
111 | (VM_MAYWRITE | VM_MAYSHARE)) | ||
112 | return LM_ST_EXCLUSIVE; | ||
113 | return LM_ST_SHARED; | ||
114 | } | ||
115 | |||
116 | static ssize_t walk_vm_hard(struct file *file, const char __user *buf, size_t size, | ||
117 | loff_t *offset, do_rw_t operation) | ||
118 | { | ||
119 | struct gfs2_holder *ghs; | ||
120 | unsigned int num_gh = 0; | ||
121 | ssize_t count; | ||
122 | struct super_block *sb = file->f_dentry->d_inode->i_sb; | ||
123 | struct mm_struct *mm = current->mm; | ||
124 | struct vm_area_struct *vma; | ||
125 | unsigned long start = (unsigned long)buf; | ||
126 | unsigned long end = start + size; | ||
127 | int dumping = (current->flags & PF_DUMPCORE); | ||
128 | unsigned int x = 0; | ||
129 | |||
130 | for (vma = find_vma(mm, start); vma; vma = vma->vm_next) { | ||
131 | if (end <= vma->vm_start) | ||
132 | break; | ||
133 | if (vma->vm_file && | ||
134 | vma->vm_file->f_dentry->d_inode->i_sb == sb) { | ||
135 | num_gh++; | ||
136 | } | ||
137 | } | ||
138 | |||
139 | ghs = kcalloc((num_gh + 1), sizeof(struct gfs2_holder), GFP_KERNEL); | ||
140 | if (!ghs) { | ||
141 | if (!dumping) | ||
142 | up_read(&mm->mmap_sem); | ||
143 | return -ENOMEM; | ||
144 | } | ||
145 | |||
146 | for (vma = find_vma(mm, start); vma; vma = vma->vm_next) { | ||
147 | if (end <= vma->vm_start) | ||
148 | break; | ||
149 | if (vma->vm_file) { | ||
150 | struct inode *inode = vma->vm_file->f_dentry->d_inode; | ||
151 | if (inode->i_sb == sb) | ||
152 | gfs2_holder_init(get_v2ip(inode)->i_gl, | ||
153 | vma2state(vma), 0, &ghs[x++]); | ||
154 | } | ||
155 | } | ||
156 | |||
157 | if (!dumping) | ||
158 | up_read(&mm->mmap_sem); | ||
159 | |||
160 | gfs2_assert(get_v2sdp(sb), x == num_gh); | ||
161 | |||
162 | count = operation(file, buf, size, offset, num_gh, ghs); | ||
163 | |||
164 | while (num_gh--) | ||
165 | gfs2_holder_uninit(&ghs[num_gh]); | ||
166 | kfree(ghs); | ||
167 | |||
168 | return count; | ||
169 | } | ||
170 | |||
171 | /** | ||
172 | * walk_vm - Walk the vmas associated with a buffer for read or write. | ||
173 | * If any of them are gfs2, pass the gfs2 inode down to the read/write | ||
174 | * worker function so that locks can be acquired in the correct order. | ||
175 | * @file: The file to read/write from/to | ||
176 | * @buf: The buffer to copy to/from | ||
177 | * @size: The amount of data requested | ||
178 | * @offset: The current file offset | ||
179 | * @operation: The read or write worker function | ||
180 | * | ||
181 | * Outputs: Offset - updated according to number of bytes written | ||
182 | * | ||
183 | * Returns: The number of bytes written, errno on failure | ||
184 | */ | ||
185 | |||
186 | static ssize_t walk_vm(struct file *file, const char __user *buf, size_t size, | ||
187 | loff_t *offset, do_rw_t operation) | ||
188 | { | ||
189 | struct gfs2_holder gh; | ||
190 | |||
191 | if (current->mm) { | ||
192 | struct super_block *sb = file->f_dentry->d_inode->i_sb; | ||
193 | struct mm_struct *mm = current->mm; | ||
194 | struct vm_area_struct *vma; | ||
195 | unsigned long start = (unsigned long)buf; | ||
196 | unsigned long end = start + size; | ||
197 | int dumping = (current->flags & PF_DUMPCORE); | ||
198 | |||
199 | if (!dumping) | ||
200 | down_read(&mm->mmap_sem); | ||
201 | |||
202 | for (vma = find_vma(mm, start); vma; vma = vma->vm_next) { | ||
203 | if (end <= vma->vm_start) | ||
204 | break; | ||
205 | if (vma->vm_file && | ||
206 | vma->vm_file->f_dentry->d_inode->i_sb == sb) | ||
207 | goto do_locks; | ||
208 | } | ||
209 | |||
210 | if (!dumping) | ||
211 | up_read(&mm->mmap_sem); | ||
212 | } | ||
213 | |||
214 | return operation(file, buf, size, offset, 0, &gh); | ||
215 | |||
216 | do_locks: | ||
217 | return walk_vm_hard(file, buf, size, offset, operation); | ||
218 | } | ||
219 | |||
220 | static ssize_t do_jdata_read(struct file *file, char __user *buf, size_t size, | ||
221 | loff_t *offset) | ||
222 | { | ||
223 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
224 | ssize_t count = 0; | ||
225 | |||
226 | if (*offset < 0) | ||
227 | return -EINVAL; | ||
228 | if (!access_ok(VERIFY_WRITE, buf, size)) | ||
229 | return -EFAULT; | ||
230 | |||
231 | if (!(file->f_flags & O_LARGEFILE)) { | ||
232 | if (*offset >= MAX_NON_LFS) | ||
233 | return -EFBIG; | ||
234 | if (*offset + size > MAX_NON_LFS) | ||
235 | size = MAX_NON_LFS - *offset; | ||
236 | } | ||
237 | |||
238 | count = gfs2_jdata_read(ip, buf, *offset, size, gfs2_copy2user); | ||
239 | |||
240 | if (count > 0) | ||
241 | *offset += count; | ||
242 | |||
243 | return count; | ||
244 | } | ||
245 | |||
246 | /** | ||
247 | * do_read_direct - Read bytes from a file | ||
248 | * @file: The file to read from | ||
249 | * @buf: The buffer to copy into | ||
250 | * @size: The amount of data requested | ||
251 | * @offset: The current file offset | ||
252 | * @num_gh: The number of other locks we need to do the read | ||
253 | * @ghs: the locks we need plus one for our lock | ||
254 | * | ||
255 | * Outputs: Offset - updated according to number of bytes read | ||
256 | * | ||
257 | * Returns: The number of bytes read, errno on failure | ||
258 | */ | ||
259 | |||
260 | static ssize_t do_read_direct(struct file *file, char __user *buf, size_t size, | ||
261 | loff_t *offset, unsigned int num_gh, | ||
262 | struct gfs2_holder *ghs) | ||
263 | { | ||
264 | struct inode *inode = file->f_mapping->host; | ||
265 | struct gfs2_inode *ip = get_v2ip(inode); | ||
266 | unsigned int state = LM_ST_DEFERRED; | ||
267 | int flags = 0; | ||
268 | unsigned int x; | ||
269 | ssize_t count = 0; | ||
270 | int error; | ||
271 | |||
272 | for (x = 0; x < num_gh; x++) | ||
273 | if (ghs[x].gh_gl == ip->i_gl) { | ||
274 | state = LM_ST_SHARED; | ||
275 | flags |= GL_LOCAL_EXCL; | ||
276 | break; | ||
277 | } | ||
278 | |||
279 | gfs2_holder_init(ip->i_gl, state, flags, &ghs[num_gh]); | ||
280 | |||
281 | error = gfs2_glock_nq_m(num_gh + 1, ghs); | ||
282 | if (error) | ||
283 | goto out; | ||
284 | |||
285 | error = -EINVAL; | ||
286 | if (gfs2_is_jdata(ip)) | ||
287 | goto out_gunlock; | ||
288 | |||
289 | if (gfs2_is_stuffed(ip)) { | ||
290 | size_t mask = bdev_hardsect_size(inode->i_sb->s_bdev) - 1; | ||
291 | |||
292 | if (((*offset) & mask) || (((unsigned long)buf) & mask)) | ||
293 | goto out_gunlock; | ||
294 | |||
295 | count = do_jdata_read(file, buf, size & ~mask, offset); | ||
296 | } else | ||
297 | count = generic_file_read(file, buf, size, offset); | ||
298 | |||
299 | error = 0; | ||
300 | |||
301 | out_gunlock: | ||
302 | gfs2_glock_dq_m(num_gh + 1, ghs); | ||
303 | |||
304 | out: | ||
305 | gfs2_holder_uninit(&ghs[num_gh]); | ||
306 | |||
307 | return (count) ? count : error; | ||
308 | } | ||
309 | |||
310 | /** | ||
311 | * do_read_buf - Read bytes from a file | ||
312 | * @file: The file to read from | ||
313 | * @buf: The buffer to copy into | ||
314 | * @size: The amount of data requested | ||
315 | * @offset: The current file offset | ||
316 | * @num_gh: The number of other locks we need to do the read | ||
317 | * @ghs: the locks we need plus one for our lock | ||
318 | * | ||
319 | * Outputs: Offset - updated according to number of bytes read | ||
320 | * | ||
321 | * Returns: The number of bytes read, errno on failure | ||
322 | */ | ||
323 | |||
324 | static ssize_t do_read_buf(struct file *file, char __user *buf, size_t size, | ||
325 | loff_t *offset, unsigned int num_gh, | ||
326 | struct gfs2_holder *ghs) | ||
327 | { | ||
328 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
329 | ssize_t count = 0; | ||
330 | int error; | ||
331 | |||
332 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &ghs[num_gh]); | ||
333 | |||
334 | error = gfs2_glock_nq_m_atime(num_gh + 1, ghs); | ||
335 | if (error) | ||
336 | goto out; | ||
337 | |||
338 | if (gfs2_is_jdata(ip)) | ||
339 | count = do_jdata_read(file, buf, size, offset); | ||
340 | else | ||
341 | count = generic_file_read(file, buf, size, offset); | ||
342 | |||
343 | gfs2_glock_dq_m(num_gh + 1, ghs); | ||
344 | |||
345 | out: | ||
346 | gfs2_holder_uninit(&ghs[num_gh]); | ||
347 | |||
348 | return (count) ? count : error; | ||
349 | } | ||
350 | |||
351 | /** | ||
352 | * gfs2_read - Read bytes from a file | ||
353 | * @file: The file to read from | ||
354 | * @buf: The buffer to copy into | ||
355 | * @size: The amount of data requested | ||
356 | * @offset: The current file offset | ||
357 | * | ||
358 | * Outputs: Offset - updated according to number of bytes read | ||
359 | * | ||
360 | * Returns: The number of bytes read, errno on failure | ||
361 | */ | ||
362 | |||
363 | static ssize_t gfs2_read(struct file *file, char __user *buf, size_t size, | ||
364 | loff_t *offset) | ||
365 | { | ||
366 | atomic_inc(&get_v2sdp(file->f_mapping->host->i_sb)->sd_ops_file); | ||
367 | |||
368 | if (file->f_flags & O_DIRECT) | ||
369 | return walk_vm(file, buf, size, offset, do_read_direct); | ||
370 | else | ||
371 | return walk_vm(file, buf, size, offset, do_read_buf); | ||
372 | } | ||
373 | |||
374 | /** | ||
375 | * grope_mapping - feel up a mapping that needs to be written | ||
376 | * @buf: the start of the memory to be written | ||
377 | * @size: the size of the memory to be written | ||
378 | * | ||
379 | * We do this after acquiring the locks on the mapping, | ||
380 | * but before starting the write transaction. We need to make | ||
381 | * sure that we don't cause recursive transactions if blocks | ||
382 | * need to be allocated to the file backing the mapping. | ||
383 | * | ||
384 | * Returns: errno | ||
385 | */ | ||
386 | |||
387 | static int grope_mapping(const char __user *buf, size_t size) | ||
388 | { | ||
389 | const char __user *stop = buf + size; | ||
390 | char c; | ||
391 | |||
392 | while (buf < stop) { | ||
393 | if (copy_from_user(&c, buf, 1)) | ||
394 | return -EFAULT; | ||
395 | buf += PAGE_CACHE_SIZE; | ||
396 | buf = (const char __user *)PAGE_ALIGN((unsigned long)buf); | ||
397 | } | ||
398 | |||
399 | return 0; | ||
400 | } | ||
401 | |||
402 | /** | ||
403 | * do_write_direct_alloc - Write bytes to a file | ||
404 | * @file: The file to write to | ||
405 | * @buf: The buffer to copy from | ||
406 | * @size: The amount of data requested | ||
407 | * @offset: The current file offset | ||
408 | * | ||
409 | * Outputs: Offset - updated according to number of bytes written | ||
410 | * | ||
411 | * Returns: The number of bytes written, errno on failure | ||
412 | */ | ||
413 | |||
414 | static ssize_t do_write_direct_alloc(struct file *file, const char __user *buf, size_t size, | ||
415 | loff_t *offset) | ||
416 | { | ||
417 | struct inode *inode = file->f_mapping->host; | ||
418 | struct gfs2_inode *ip = get_v2ip(inode); | ||
419 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
420 | struct gfs2_alloc *al = NULL; | ||
421 | struct iovec local_iov = { .iov_base = buf, .iov_len = size }; | ||
422 | struct buffer_head *dibh; | ||
423 | unsigned int data_blocks, ind_blocks; | ||
424 | ssize_t count; | ||
425 | int error; | ||
426 | |||
427 | gfs2_write_calc_reserv(ip, size, &data_blocks, &ind_blocks); | ||
428 | |||
429 | al = gfs2_alloc_get(ip); | ||
430 | |||
431 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | ||
432 | if (error) | ||
433 | goto fail; | ||
434 | |||
435 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | ||
436 | if (error) | ||
437 | goto fail_gunlock_q; | ||
438 | |||
439 | al->al_requested = data_blocks + ind_blocks; | ||
440 | |||
441 | error = gfs2_inplace_reserve(ip); | ||
442 | if (error) | ||
443 | goto fail_gunlock_q; | ||
444 | |||
445 | error = gfs2_trans_begin(sdp, | ||
446 | al->al_rgd->rd_ri.ri_length + ind_blocks + | ||
447 | RES_DINODE + RES_STATFS + RES_QUOTA, 0); | ||
448 | if (error) | ||
449 | goto fail_ipres; | ||
450 | |||
451 | if ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID)) { | ||
452 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
453 | if (error) | ||
454 | goto fail_end_trans; | ||
455 | |||
456 | ip->i_di.di_mode &= (ip->i_di.di_mode & S_IXGRP) ? | ||
457 | (~(S_ISUID | S_ISGID)) : (~S_ISUID); | ||
458 | |||
459 | gfs2_trans_add_bh(ip->i_gl, dibh); | ||
460 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
461 | brelse(dibh); | ||
462 | } | ||
463 | |||
464 | if (gfs2_is_stuffed(ip)) { | ||
465 | error = gfs2_unstuff_dinode(ip, gfs2_unstuffer_sync, NULL); | ||
466 | if (error) | ||
467 | goto fail_end_trans; | ||
468 | } | ||
469 | |||
470 | count = generic_file_write_nolock(file, &local_iov, 1, offset); | ||
471 | if (count < 0) { | ||
472 | error = count; | ||
473 | goto fail_end_trans; | ||
474 | } | ||
475 | |||
476 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
477 | if (error) | ||
478 | goto fail_end_trans; | ||
479 | |||
480 | if (ip->i_di.di_size < inode->i_size) | ||
481 | ip->i_di.di_size = inode->i_size; | ||
482 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | ||
483 | |||
484 | gfs2_trans_add_bh(ip->i_gl, dibh); | ||
485 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
486 | brelse(dibh); | ||
487 | |||
488 | gfs2_trans_end(sdp); | ||
489 | |||
490 | if (file->f_flags & O_SYNC) | ||
491 | gfs2_log_flush_glock(ip->i_gl); | ||
492 | |||
493 | gfs2_inplace_release(ip); | ||
494 | gfs2_quota_unlock(ip); | ||
495 | gfs2_alloc_put(ip); | ||
496 | |||
497 | if (file->f_mapping->nrpages) { | ||
498 | error = filemap_fdatawrite(file->f_mapping); | ||
499 | if (!error) | ||
500 | error = filemap_fdatawait(file->f_mapping); | ||
501 | } | ||
502 | if (error) | ||
503 | return error; | ||
504 | |||
505 | return count; | ||
506 | |||
507 | fail_end_trans: | ||
508 | gfs2_trans_end(sdp); | ||
509 | |||
510 | fail_ipres: | ||
511 | gfs2_inplace_release(ip); | ||
512 | |||
513 | fail_gunlock_q: | ||
514 | gfs2_quota_unlock(ip); | ||
515 | |||
516 | fail: | ||
517 | gfs2_alloc_put(ip); | ||
518 | |||
519 | return error; | ||
520 | } | ||
521 | |||
522 | /** | ||
523 | * do_write_direct - Write bytes to a file | ||
524 | * @file: The file to write to | ||
525 | * @buf: The buffer to copy from | ||
526 | * @size: The amount of data requested | ||
527 | * @offset: The current file offset | ||
528 | * @num_gh: The number of other locks we need to do the read | ||
529 | * @gh: the locks we need plus one for our lock | ||
530 | * | ||
531 | * Outputs: Offset - updated according to number of bytes written | ||
532 | * | ||
533 | * Returns: The number of bytes written, errno on failure | ||
534 | */ | ||
535 | |||
536 | static ssize_t do_write_direct(struct file *file, const char __user *buf, size_t size, | ||
537 | loff_t *offset, unsigned int num_gh, | ||
538 | struct gfs2_holder *ghs) | ||
539 | { | ||
540 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
541 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
542 | struct gfs2_file *fp = get_v2fp(file); | ||
543 | unsigned int state = LM_ST_DEFERRED; | ||
544 | int alloc_required; | ||
545 | unsigned int x; | ||
546 | size_t s; | ||
547 | ssize_t count = 0; | ||
548 | int error; | ||
549 | |||
550 | if (test_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags)) | ||
551 | state = LM_ST_EXCLUSIVE; | ||
552 | else | ||
553 | for (x = 0; x < num_gh; x++) | ||
554 | if (ghs[x].gh_gl == ip->i_gl) { | ||
555 | state = LM_ST_EXCLUSIVE; | ||
556 | break; | ||
557 | } | ||
558 | |||
559 | restart: | ||
560 | gfs2_holder_init(ip->i_gl, state, 0, &ghs[num_gh]); | ||
561 | |||
562 | error = gfs2_glock_nq_m(num_gh + 1, ghs); | ||
563 | if (error) | ||
564 | goto out; | ||
565 | |||
566 | error = -EINVAL; | ||
567 | if (gfs2_is_jdata(ip)) | ||
568 | goto out_gunlock; | ||
569 | |||
570 | if (num_gh) { | ||
571 | error = grope_mapping(buf, size); | ||
572 | if (error) | ||
573 | goto out_gunlock; | ||
574 | } | ||
575 | |||
576 | if (file->f_flags & O_APPEND) | ||
577 | *offset = ip->i_di.di_size; | ||
578 | |||
579 | if (!(file->f_flags & O_LARGEFILE)) { | ||
580 | error = -EFBIG; | ||
581 | if (*offset >= MAX_NON_LFS) | ||
582 | goto out_gunlock; | ||
583 | if (*offset + size > MAX_NON_LFS) | ||
584 | size = MAX_NON_LFS - *offset; | ||
585 | } | ||
586 | |||
587 | if (gfs2_is_stuffed(ip) || | ||
588 | *offset + size > ip->i_di.di_size || | ||
589 | ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID))) | ||
590 | alloc_required = 1; | ||
591 | else { | ||
592 | error = gfs2_write_alloc_required(ip, *offset, size, | ||
593 | &alloc_required); | ||
594 | if (error) | ||
595 | goto out_gunlock; | ||
596 | } | ||
597 | |||
598 | if (alloc_required && state != LM_ST_EXCLUSIVE) { | ||
599 | gfs2_glock_dq_m(num_gh + 1, ghs); | ||
600 | gfs2_holder_uninit(&ghs[num_gh]); | ||
601 | state = LM_ST_EXCLUSIVE; | ||
602 | goto restart; | ||
603 | } | ||
604 | |||
605 | if (alloc_required) { | ||
606 | set_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags); | ||
607 | |||
608 | /* split large writes into smaller atomic transactions */ | ||
609 | while (size) { | ||
610 | s = gfs2_tune_get(sdp, gt_max_atomic_write); | ||
611 | if (s > size) | ||
612 | s = size; | ||
613 | |||
614 | error = do_write_direct_alloc(file, buf, s, offset); | ||
615 | if (error < 0) | ||
616 | goto out_gunlock; | ||
617 | |||
618 | buf += error; | ||
619 | size -= error; | ||
620 | count += error; | ||
621 | } | ||
622 | } else { | ||
623 | struct iovec local_iov = { .iov_base = buf, .iov_len = size }; | ||
624 | struct gfs2_holder t_gh; | ||
625 | |||
626 | clear_bit(GFF_DID_DIRECT_ALLOC, &fp->f_flags); | ||
627 | |||
628 | error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, | ||
629 | GL_NEVER_RECURSE, &t_gh); | ||
630 | if (error) | ||
631 | goto out_gunlock; | ||
632 | |||
633 | count = generic_file_write_nolock(file, &local_iov, 1, offset); | ||
634 | |||
635 | gfs2_glock_dq_uninit(&t_gh); | ||
636 | } | ||
637 | |||
638 | error = 0; | ||
639 | |||
640 | out_gunlock: | ||
641 | gfs2_glock_dq_m(num_gh + 1, ghs); | ||
642 | |||
643 | out: | ||
644 | gfs2_holder_uninit(&ghs[num_gh]); | ||
645 | |||
646 | return (count) ? count : error; | ||
647 | } | ||
648 | |||
649 | /** | ||
650 | * do_do_write_buf - Write bytes to a file | ||
651 | * @file: The file to write to | ||
652 | * @buf: The buffer to copy from | ||
653 | * @size: The amount of data requested | ||
654 | * @offset: The current file offset | ||
655 | * | ||
656 | * Outputs: Offset - updated according to number of bytes written | ||
657 | * | ||
658 | * Returns: The number of bytes written, errno on failure | ||
659 | */ | ||
660 | |||
661 | static ssize_t do_do_write_buf(struct file *file, const char __user *buf, size_t size, | ||
662 | loff_t *offset) | ||
663 | { | ||
664 | struct inode *inode = file->f_mapping->host; | ||
665 | struct gfs2_inode *ip = get_v2ip(inode); | ||
666 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
667 | struct gfs2_alloc *al = NULL; | ||
668 | struct buffer_head *dibh; | ||
669 | unsigned int data_blocks, ind_blocks; | ||
670 | int alloc_required, journaled; | ||
671 | ssize_t count; | ||
672 | int error; | ||
673 | |||
674 | journaled = gfs2_is_jdata(ip); | ||
675 | |||
676 | gfs2_write_calc_reserv(ip, size, &data_blocks, &ind_blocks); | ||
677 | |||
678 | error = gfs2_write_alloc_required(ip, *offset, size, &alloc_required); | ||
679 | if (error) | ||
680 | return error; | ||
681 | |||
682 | if (alloc_required) { | ||
683 | al = gfs2_alloc_get(ip); | ||
684 | |||
685 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | ||
686 | if (error) | ||
687 | goto fail; | ||
688 | |||
689 | error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid); | ||
690 | if (error) | ||
691 | goto fail_gunlock_q; | ||
692 | |||
693 | al->al_requested = data_blocks + ind_blocks; | ||
694 | |||
695 | error = gfs2_inplace_reserve(ip); | ||
696 | if (error) | ||
697 | goto fail_gunlock_q; | ||
698 | |||
699 | error = gfs2_trans_begin(sdp, | ||
700 | al->al_rgd->rd_ri.ri_length + | ||
701 | ind_blocks + | ||
702 | ((journaled) ? data_blocks : 0) + | ||
703 | RES_DINODE + RES_STATFS + RES_QUOTA, | ||
704 | 0); | ||
705 | if (error) | ||
706 | goto fail_ipres; | ||
707 | } else { | ||
708 | error = gfs2_trans_begin(sdp, | ||
709 | ((journaled) ? data_blocks : 0) + | ||
710 | RES_DINODE, | ||
711 | 0); | ||
712 | if (error) | ||
713 | goto fail_ipres; | ||
714 | } | ||
715 | |||
716 | if ((ip->i_di.di_mode & (S_ISUID | S_ISGID)) && !capable(CAP_FSETID)) { | ||
717 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
718 | if (error) | ||
719 | goto fail_end_trans; | ||
720 | |||
721 | ip->i_di.di_mode &= (ip->i_di.di_mode & S_IXGRP) ? | ||
722 | (~(S_ISUID | S_ISGID)) : (~S_ISUID); | ||
723 | |||
724 | gfs2_trans_add_bh(ip->i_gl, dibh); | ||
725 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
726 | brelse(dibh); | ||
727 | } | ||
728 | |||
729 | if (journaled) { | ||
730 | count = gfs2_jdata_write(ip, buf, *offset, size, | ||
731 | gfs2_copy_from_user); | ||
732 | if (count < 0) { | ||
733 | error = count; | ||
734 | goto fail_end_trans; | ||
735 | } | ||
736 | |||
737 | *offset += count; | ||
738 | } else { | ||
739 | struct iovec local_iov = { .iov_base = buf, .iov_len = size }; | ||
740 | |||
741 | count = generic_file_write_nolock(file, &local_iov, 1, offset); | ||
742 | if (count < 0) { | ||
743 | error = count; | ||
744 | goto fail_end_trans; | ||
745 | } | ||
746 | |||
747 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
748 | if (error) | ||
749 | goto fail_end_trans; | ||
750 | |||
751 | if (ip->i_di.di_size < inode->i_size) | ||
752 | ip->i_di.di_size = inode->i_size; | ||
753 | ip->i_di.di_mtime = ip->i_di.di_ctime = get_seconds(); | ||
754 | |||
755 | gfs2_trans_add_bh(ip->i_gl, dibh); | ||
756 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
757 | brelse(dibh); | ||
758 | } | ||
759 | |||
760 | gfs2_trans_end(sdp); | ||
761 | |||
762 | if (file->f_flags & O_SYNC || IS_SYNC(inode)) { | ||
763 | gfs2_log_flush_glock(ip->i_gl); | ||
764 | error = filemap_fdatawrite(file->f_mapping); | ||
765 | if (error == 0) | ||
766 | error = filemap_fdatawait(file->f_mapping); | ||
767 | if (error) | ||
768 | goto fail_ipres; | ||
769 | } | ||
770 | |||
771 | if (alloc_required) { | ||
772 | gfs2_assert_warn(sdp, count != size || | ||
773 | al->al_alloced); | ||
774 | gfs2_inplace_release(ip); | ||
775 | gfs2_quota_unlock(ip); | ||
776 | gfs2_alloc_put(ip); | ||
777 | } | ||
778 | |||
779 | return count; | ||
780 | |||
781 | fail_end_trans: | ||
782 | gfs2_trans_end(sdp); | ||
783 | |||
784 | fail_ipres: | ||
785 | if (alloc_required) | ||
786 | gfs2_inplace_release(ip); | ||
787 | |||
788 | fail_gunlock_q: | ||
789 | if (alloc_required) | ||
790 | gfs2_quota_unlock(ip); | ||
791 | |||
792 | fail: | ||
793 | if (alloc_required) | ||
794 | gfs2_alloc_put(ip); | ||
795 | |||
796 | return error; | ||
797 | } | ||
798 | |||
799 | /** | ||
800 | * do_write_buf - Write bytes to a file | ||
801 | * @file: The file to write to | ||
802 | * @buf: The buffer to copy from | ||
803 | * @size: The amount of data requested | ||
804 | * @offset: The current file offset | ||
805 | * @num_gh: The number of other locks we need to do the read | ||
806 | * @gh: the locks we need plus one for our lock | ||
807 | * | ||
808 | * Outputs: Offset - updated according to number of bytes written | ||
809 | * | ||
810 | * Returns: The number of bytes written, errno on failure | ||
811 | */ | ||
812 | |||
813 | static ssize_t do_write_buf(struct file *file, const char __user *buf, size_t size, | ||
814 | loff_t *offset, unsigned int num_gh, | ||
815 | struct gfs2_holder *ghs) | ||
816 | { | ||
817 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
818 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
819 | size_t s; | ||
820 | ssize_t count = 0; | ||
821 | int error; | ||
822 | |||
823 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ghs[num_gh]); | ||
824 | |||
825 | error = gfs2_glock_nq_m(num_gh + 1, ghs); | ||
826 | if (error) | ||
827 | goto out; | ||
828 | |||
829 | if (num_gh) { | ||
830 | error = grope_mapping(buf, size); | ||
831 | if (error) | ||
832 | goto out_gunlock; | ||
833 | } | ||
834 | |||
835 | if (file->f_flags & O_APPEND) | ||
836 | *offset = ip->i_di.di_size; | ||
837 | |||
838 | if (!(file->f_flags & O_LARGEFILE)) { | ||
839 | error = -EFBIG; | ||
840 | if (*offset >= MAX_NON_LFS) | ||
841 | goto out_gunlock; | ||
842 | if (*offset + size > MAX_NON_LFS) | ||
843 | size = MAX_NON_LFS - *offset; | ||
844 | } | ||
845 | |||
846 | /* split large writes into smaller atomic transactions */ | ||
847 | while (size) { | ||
848 | s = gfs2_tune_get(sdp, gt_max_atomic_write); | ||
849 | if (s > size) | ||
850 | s = size; | ||
851 | |||
852 | error = do_do_write_buf(file, buf, s, offset); | ||
853 | if (error < 0) | ||
854 | goto out_gunlock; | ||
855 | |||
856 | buf += error; | ||
857 | size -= error; | ||
858 | count += error; | ||
859 | } | ||
860 | |||
861 | error = 0; | ||
862 | |||
863 | out_gunlock: | ||
864 | gfs2_glock_dq_m(num_gh + 1, ghs); | ||
865 | |||
866 | out: | ||
867 | gfs2_holder_uninit(&ghs[num_gh]); | ||
868 | |||
869 | return (count) ? count : error; | ||
870 | } | ||
871 | |||
872 | /** | ||
873 | * gfs2_write - Write bytes to a file | ||
874 | * @file: The file to write to | ||
875 | * @buf: The buffer to copy from | ||
876 | * @size: The amount of data requested | ||
877 | * @offset: The current file offset | ||
878 | * | ||
879 | * Outputs: Offset - updated according to number of bytes written | ||
880 | * | ||
881 | * Returns: The number of bytes written, errno on failure | ||
882 | */ | ||
883 | |||
884 | static ssize_t gfs2_write(struct file *file, const char __user *buf, | ||
885 | size_t size, loff_t *offset) | ||
886 | { | ||
887 | struct inode *inode = file->f_mapping->host; | ||
888 | ssize_t count; | ||
889 | |||
890 | atomic_inc(&get_v2sdp(inode->i_sb)->sd_ops_file); | ||
891 | |||
892 | if (*offset < 0) | ||
893 | return -EINVAL; | ||
894 | if (!access_ok(VERIFY_READ, buf, size)) | ||
895 | return -EFAULT; | ||
896 | |||
897 | mutex_lock(&inode->i_mutex); | ||
898 | if (file->f_flags & O_DIRECT) | ||
899 | count = walk_vm(file, buf, size, offset, | ||
900 | do_write_direct); | ||
901 | else | ||
902 | count = walk_vm(file, buf, size, offset, do_write_buf); | ||
903 | mutex_unlock(&inode->i_mutex); | ||
904 | |||
905 | return count; | ||
906 | } | ||
907 | |||
908 | /** | ||
909 | * filldir_reg_func - Report a directory entry to the caller of gfs2_dir_read() | ||
910 | * @opaque: opaque data used by the function | ||
911 | * @name: the name of the directory entry | ||
912 | * @length: the length of the name | ||
913 | * @offset: the entry's offset in the directory | ||
914 | * @inum: the inode number the entry points to | ||
915 | * @type: the type of inode the entry points to | ||
916 | * | ||
917 | * Returns: 0 on success, 1 if buffer full | ||
918 | */ | ||
919 | |||
920 | static int filldir_reg_func(void *opaque, const char *name, unsigned int length, | ||
921 | uint64_t offset, struct gfs2_inum *inum, | ||
922 | unsigned int type) | ||
923 | { | ||
924 | struct filldir_reg *fdr = (struct filldir_reg *)opaque; | ||
925 | struct gfs2_sbd *sdp = fdr->fdr_sbd; | ||
926 | int error; | ||
927 | |||
928 | error = fdr->fdr_filldir(fdr->fdr_opaque, name, length, offset, | ||
929 | inum->no_formal_ino, type); | ||
930 | if (error) | ||
931 | return 1; | ||
932 | |||
933 | if (fdr->fdr_prefetch && !(length == 1 && *name == '.')) { | ||
934 | gfs2_glock_prefetch_num(sdp, | ||
935 | inum->no_addr, &gfs2_inode_glops, | ||
936 | LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY); | ||
937 | gfs2_glock_prefetch_num(sdp, | ||
938 | inum->no_addr, &gfs2_iopen_glops, | ||
939 | LM_ST_SHARED, LM_FLAG_TRY); | ||
940 | } | ||
941 | |||
942 | return 0; | ||
943 | } | ||
944 | |||
945 | /** | ||
946 | * readdir_reg - Read directory entries from a directory | ||
947 | * @file: The directory to read from | ||
948 | * @dirent: Buffer for dirents | ||
949 | * @filldir: Function used to do the copying | ||
950 | * | ||
951 | * Returns: errno | ||
952 | */ | ||
953 | |||
954 | static int readdir_reg(struct file *file, void *dirent, filldir_t filldir) | ||
955 | { | ||
956 | struct gfs2_inode *dip = get_v2ip(file->f_mapping->host); | ||
957 | struct filldir_reg fdr; | ||
958 | struct gfs2_holder d_gh; | ||
959 | uint64_t offset = file->f_pos; | ||
960 | int error; | ||
961 | |||
962 | fdr.fdr_sbd = dip->i_sbd; | ||
963 | fdr.fdr_prefetch = 1; | ||
964 | fdr.fdr_filldir = filldir; | ||
965 | fdr.fdr_opaque = dirent; | ||
966 | |||
967 | gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh); | ||
968 | error = gfs2_glock_nq_atime(&d_gh); | ||
969 | if (error) { | ||
970 | gfs2_holder_uninit(&d_gh); | ||
971 | return error; | ||
972 | } | ||
973 | |||
974 | error = gfs2_dir_read(dip, &offset, &fdr, filldir_reg_func); | ||
975 | |||
976 | gfs2_glock_dq_uninit(&d_gh); | ||
977 | |||
978 | file->f_pos = offset; | ||
979 | |||
980 | return error; | ||
981 | } | ||
982 | |||
983 | /** | ||
984 | * filldir_bad_func - Report a directory entry to the caller of gfs2_dir_read() | ||
985 | * @opaque: opaque data used by the function | ||
986 | * @name: the name of the directory entry | ||
987 | * @length: the length of the name | ||
988 | * @offset: the entry's offset in the directory | ||
989 | * @inum: the inode number the entry points to | ||
990 | * @type: the type of inode the entry points to | ||
991 | * | ||
992 | * For supporting NFS. | ||
993 | * | ||
994 | * Returns: 0 on success, 1 if buffer full | ||
995 | */ | ||
996 | |||
997 | static int filldir_bad_func(void *opaque, const char *name, unsigned int length, | ||
998 | uint64_t offset, struct gfs2_inum *inum, | ||
999 | unsigned int type) | ||
1000 | { | ||
1001 | struct filldir_bad *fdb = (struct filldir_bad *)opaque; | ||
1002 | struct gfs2_sbd *sdp = fdb->fdb_sbd; | ||
1003 | struct filldir_bad_entry *fbe; | ||
1004 | |||
1005 | if (fdb->fdb_entry_off == fdb->fdb_entry_num || | ||
1006 | fdb->fdb_name_off + length > fdb->fdb_name_size) | ||
1007 | return 1; | ||
1008 | |||
1009 | fbe = &fdb->fdb_entry[fdb->fdb_entry_off]; | ||
1010 | fbe->fbe_name = fdb->fdb_name + fdb->fdb_name_off; | ||
1011 | memcpy(fbe->fbe_name, name, length); | ||
1012 | fbe->fbe_length = length; | ||
1013 | fbe->fbe_offset = offset; | ||
1014 | fbe->fbe_inum = *inum; | ||
1015 | fbe->fbe_type = type; | ||
1016 | |||
1017 | fdb->fdb_entry_off++; | ||
1018 | fdb->fdb_name_off += length; | ||
1019 | |||
1020 | if (!(length == 1 && *name == '.')) { | ||
1021 | gfs2_glock_prefetch_num(sdp, | ||
1022 | inum->no_addr, &gfs2_inode_glops, | ||
1023 | LM_ST_SHARED, LM_FLAG_TRY | LM_FLAG_ANY); | ||
1024 | gfs2_glock_prefetch_num(sdp, | ||
1025 | inum->no_addr, &gfs2_iopen_glops, | ||
1026 | LM_ST_SHARED, LM_FLAG_TRY); | ||
1027 | } | ||
1028 | |||
1029 | return 0; | ||
1030 | } | ||
1031 | |||
1032 | /** | ||
1033 | * readdir_bad - Read directory entries from a directory | ||
1034 | * @file: The directory to read from | ||
1035 | * @dirent: Buffer for dirents | ||
1036 | * @filldir: Function used to do the copying | ||
1037 | * | ||
1038 | * For supporting NFS. | ||
1039 | * | ||
1040 | * Returns: errno | ||
1041 | */ | ||
1042 | |||
1043 | static int readdir_bad(struct file *file, void *dirent, filldir_t filldir) | ||
1044 | { | ||
1045 | struct gfs2_inode *dip = get_v2ip(file->f_mapping->host); | ||
1046 | struct gfs2_sbd *sdp = dip->i_sbd; | ||
1047 | struct filldir_reg fdr; | ||
1048 | unsigned int entries, size; | ||
1049 | struct filldir_bad *fdb; | ||
1050 | struct gfs2_holder d_gh; | ||
1051 | uint64_t offset = file->f_pos; | ||
1052 | unsigned int x; | ||
1053 | struct filldir_bad_entry *fbe; | ||
1054 | int error; | ||
1055 | |||
1056 | entries = gfs2_tune_get(sdp, gt_entries_per_readdir); | ||
1057 | size = sizeof(struct filldir_bad) + | ||
1058 | entries * (sizeof(struct filldir_bad_entry) + GFS2_FAST_NAME_SIZE); | ||
1059 | |||
1060 | fdb = kzalloc(size, GFP_KERNEL); | ||
1061 | if (!fdb) | ||
1062 | return -ENOMEM; | ||
1063 | |||
1064 | fdb->fdb_sbd = sdp; | ||
1065 | fdb->fdb_entry = (struct filldir_bad_entry *)(fdb + 1); | ||
1066 | fdb->fdb_entry_num = entries; | ||
1067 | fdb->fdb_name = ((char *)fdb) + sizeof(struct filldir_bad) + | ||
1068 | entries * sizeof(struct filldir_bad_entry); | ||
1069 | fdb->fdb_name_size = entries * GFS2_FAST_NAME_SIZE; | ||
1070 | |||
1071 | gfs2_holder_init(dip->i_gl, LM_ST_SHARED, GL_ATIME, &d_gh); | ||
1072 | error = gfs2_glock_nq_atime(&d_gh); | ||
1073 | if (error) { | ||
1074 | gfs2_holder_uninit(&d_gh); | ||
1075 | goto out; | ||
1076 | } | ||
1077 | |||
1078 | error = gfs2_dir_read(dip, &offset, fdb, filldir_bad_func); | ||
1079 | |||
1080 | gfs2_glock_dq_uninit(&d_gh); | ||
1081 | |||
1082 | fdr.fdr_sbd = sdp; | ||
1083 | fdr.fdr_prefetch = 0; | ||
1084 | fdr.fdr_filldir = filldir; | ||
1085 | fdr.fdr_opaque = dirent; | ||
1086 | |||
1087 | for (x = 0; x < fdb->fdb_entry_off; x++) { | ||
1088 | fbe = &fdb->fdb_entry[x]; | ||
1089 | |||
1090 | error = filldir_reg_func(&fdr, | ||
1091 | fbe->fbe_name, fbe->fbe_length, | ||
1092 | fbe->fbe_offset, | ||
1093 | &fbe->fbe_inum, fbe->fbe_type); | ||
1094 | if (error) { | ||
1095 | file->f_pos = fbe->fbe_offset; | ||
1096 | error = 0; | ||
1097 | goto out; | ||
1098 | } | ||
1099 | } | ||
1100 | |||
1101 | file->f_pos = offset; | ||
1102 | |||
1103 | out: | ||
1104 | kfree(fdb); | ||
1105 | |||
1106 | return error; | ||
1107 | } | ||
1108 | |||
1109 | /** | ||
1110 | * gfs2_readdir - Read directory entries from a directory | ||
1111 | * @file: The directory to read from | ||
1112 | * @dirent: Buffer for dirents | ||
1113 | * @filldir: Function used to do the copying | ||
1114 | * | ||
1115 | * Returns: errno | ||
1116 | */ | ||
1117 | |||
1118 | static int gfs2_readdir(struct file *file, void *dirent, filldir_t filldir) | ||
1119 | { | ||
1120 | int error; | ||
1121 | |||
1122 | atomic_inc(&get_v2sdp(file->f_mapping->host->i_sb)->sd_ops_file); | ||
1123 | |||
1124 | if (strcmp(current->comm, "nfsd") != 0) | ||
1125 | error = readdir_reg(file, dirent, filldir); | ||
1126 | else | ||
1127 | error = readdir_bad(file, dirent, filldir); | ||
1128 | |||
1129 | return error; | ||
1130 | } | ||
1131 | |||
1132 | static int gfs2_ioctl_flags(struct gfs2_inode *ip, unsigned int cmd, unsigned long arg) | ||
1133 | { | ||
1134 | unsigned int lmode = (cmd == GFS2_IOCTL_SETFLAGS) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | ||
1135 | struct buffer_head *dibh; | ||
1136 | struct gfs2_holder i_gh; | ||
1137 | int error; | ||
1138 | __u32 flags = 0, change; | ||
1139 | |||
1140 | if (cmd == GFS2_IOCTL_SETFLAGS) { | ||
1141 | error = get_user(flags, (__u32 __user *)arg); | ||
1142 | if (error) | ||
1143 | return -EFAULT; | ||
1144 | } | ||
1145 | |||
1146 | error = gfs2_glock_nq_init(ip->i_gl, lmode, 0, &i_gh); | ||
1147 | if (error) | ||
1148 | return error; | ||
1149 | |||
1150 | if (cmd == GFS2_IOCTL_SETFLAGS) { | ||
1151 | change = flags ^ ip->i_di.di_flags; | ||
1152 | error = -EPERM; | ||
1153 | if (change & (GFS2_DIF_IMMUTABLE|GFS2_DIF_APPENDONLY)) { | ||
1154 | if (!capable(CAP_LINUX_IMMUTABLE)) | ||
1155 | goto out; | ||
1156 | } | ||
1157 | error = -EINVAL; | ||
1158 | if (flags & (GFS2_DIF_JDATA|GFS2_DIF_DIRECTIO)) { | ||
1159 | if (!S_ISREG(ip->i_di.di_mode)) | ||
1160 | goto out; | ||
1161 | /* FIXME: Would be nice not to require the following test */ | ||
1162 | if ((flags & GFS2_DIF_JDATA) && ip->i_di.di_size) | ||
1163 | goto out; | ||
1164 | } | ||
1165 | if (flags & (GFS2_DIF_INHERIT_JDATA|GFS2_DIF_INHERIT_DIRECTIO)) { | ||
1166 | if (!S_ISDIR(ip->i_di.di_mode)) | ||
1167 | goto out; | ||
1168 | } | ||
1169 | |||
1170 | error = gfs2_trans_begin(ip->i_sbd, RES_DINODE, 0); | ||
1171 | if (error) | ||
1172 | goto out; | ||
1173 | |||
1174 | error = gfs2_meta_inode_buffer(ip, &dibh); | ||
1175 | if (error) | ||
1176 | goto out_trans_end; | ||
1177 | |||
1178 | ip->i_di.di_flags = flags; | ||
1179 | |||
1180 | gfs2_trans_add_bh(ip->i_gl, dibh); | ||
1181 | gfs2_dinode_out(&ip->i_di, dibh->b_data); | ||
1182 | |||
1183 | brelse(dibh); | ||
1184 | |||
1185 | out_trans_end: | ||
1186 | gfs2_trans_end(ip->i_sbd); | ||
1187 | } else { | ||
1188 | flags = ip->i_di.di_flags; | ||
1189 | } | ||
1190 | out: | ||
1191 | gfs2_glock_dq_uninit(&i_gh); | ||
1192 | if (cmd == GFS2_IOCTL_GETFLAGS) { | ||
1193 | if (put_user(flags, (__u32 __user *)arg)) | ||
1194 | return -EFAULT; | ||
1195 | } | ||
1196 | return error; | ||
1197 | } | ||
1198 | |||
1199 | /** | ||
1200 | * gfs2_ioctl - do an ioctl on a file | ||
1201 | * @inode: the inode | ||
1202 | * @file: the file pointer | ||
1203 | * @cmd: the ioctl command | ||
1204 | * @arg: the argument | ||
1205 | * | ||
1206 | * Returns: errno | ||
1207 | */ | ||
1208 | |||
1209 | static int gfs2_ioctl(struct inode *inode, struct file *file, unsigned int cmd, | ||
1210 | unsigned long arg) | ||
1211 | { | ||
1212 | struct gfs2_inode *ip = get_v2ip(inode); | ||
1213 | |||
1214 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1215 | |||
1216 | switch (cmd) { | ||
1217 | case GFS2_IOCTL_IDENTIFY: { | ||
1218 | unsigned int x = GFS2_MAGIC; | ||
1219 | if (copy_to_user((unsigned int __user *)arg, &x, sizeof(unsigned int))) | ||
1220 | return -EFAULT; | ||
1221 | return 0; | ||
1222 | |||
1223 | case GFS2_IOCTL_SETFLAGS: | ||
1224 | case GFS2_IOCTL_GETFLAGS: | ||
1225 | return gfs2_ioctl_flags(ip, cmd, arg); | ||
1226 | } | ||
1227 | |||
1228 | default: | ||
1229 | return -ENOTTY; | ||
1230 | } | ||
1231 | } | ||
1232 | |||
1233 | /** | ||
1234 | * gfs2_mmap - | ||
1235 | * @file: The file to map | ||
1236 | * @vma: The VMA which described the mapping | ||
1237 | * | ||
1238 | * Returns: 0 or error code | ||
1239 | */ | ||
1240 | |||
1241 | static int gfs2_mmap(struct file *file, struct vm_area_struct *vma) | ||
1242 | { | ||
1243 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
1244 | struct gfs2_holder i_gh; | ||
1245 | int error; | ||
1246 | |||
1247 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1248 | |||
1249 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &i_gh); | ||
1250 | error = gfs2_glock_nq_atime(&i_gh); | ||
1251 | if (error) { | ||
1252 | gfs2_holder_uninit(&i_gh); | ||
1253 | return error; | ||
1254 | } | ||
1255 | |||
1256 | if (gfs2_is_jdata(ip)) { | ||
1257 | if (vma->vm_flags & VM_MAYSHARE) | ||
1258 | error = -EOPNOTSUPP; | ||
1259 | else | ||
1260 | vma->vm_ops = &gfs2_vm_ops_private; | ||
1261 | } else { | ||
1262 | /* This is VM_MAYWRITE instead of VM_WRITE because a call | ||
1263 | to mprotect() can turn on VM_WRITE later. */ | ||
1264 | |||
1265 | if ((vma->vm_flags & (VM_MAYSHARE | VM_MAYWRITE)) == | ||
1266 | (VM_MAYSHARE | VM_MAYWRITE)) | ||
1267 | vma->vm_ops = &gfs2_vm_ops_sharewrite; | ||
1268 | else | ||
1269 | vma->vm_ops = &gfs2_vm_ops_private; | ||
1270 | } | ||
1271 | |||
1272 | gfs2_glock_dq_uninit(&i_gh); | ||
1273 | |||
1274 | return error; | ||
1275 | } | ||
1276 | |||
1277 | /** | ||
1278 | * gfs2_open - open a file | ||
1279 | * @inode: the inode to open | ||
1280 | * @file: the struct file for this opening | ||
1281 | * | ||
1282 | * Returns: errno | ||
1283 | */ | ||
1284 | |||
1285 | static int gfs2_open(struct inode *inode, struct file *file) | ||
1286 | { | ||
1287 | struct gfs2_inode *ip = get_v2ip(inode); | ||
1288 | struct gfs2_holder i_gh; | ||
1289 | struct gfs2_file *fp; | ||
1290 | int error; | ||
1291 | |||
1292 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1293 | |||
1294 | fp = kzalloc(sizeof(struct gfs2_file), GFP_KERNEL); | ||
1295 | if (!fp) | ||
1296 | return -ENOMEM; | ||
1297 | |||
1298 | init_MUTEX(&fp->f_fl_mutex); | ||
1299 | |||
1300 | fp->f_inode = ip; | ||
1301 | fp->f_vfile = file; | ||
1302 | |||
1303 | gfs2_assert_warn(ip->i_sbd, !get_v2fp(file)); | ||
1304 | set_v2fp(file, fp); | ||
1305 | |||
1306 | if (S_ISREG(ip->i_di.di_mode)) { | ||
1307 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, | ||
1308 | &i_gh); | ||
1309 | if (error) | ||
1310 | goto fail; | ||
1311 | |||
1312 | if (!(file->f_flags & O_LARGEFILE) && | ||
1313 | ip->i_di.di_size > MAX_NON_LFS) { | ||
1314 | error = -EFBIG; | ||
1315 | goto fail_gunlock; | ||
1316 | } | ||
1317 | |||
1318 | /* Listen to the Direct I/O flag */ | ||
1319 | |||
1320 | if (ip->i_di.di_flags & GFS2_DIF_DIRECTIO) | ||
1321 | file->f_flags |= O_DIRECT; | ||
1322 | |||
1323 | /* Don't let the user open O_DIRECT on a jdata file */ | ||
1324 | |||
1325 | if ((file->f_flags & O_DIRECT) && gfs2_is_jdata(ip)) { | ||
1326 | error = -EINVAL; | ||
1327 | goto fail_gunlock; | ||
1328 | } | ||
1329 | |||
1330 | gfs2_glock_dq_uninit(&i_gh); | ||
1331 | } | ||
1332 | |||
1333 | return 0; | ||
1334 | |||
1335 | fail_gunlock: | ||
1336 | gfs2_glock_dq_uninit(&i_gh); | ||
1337 | |||
1338 | fail: | ||
1339 | set_v2fp(file, NULL); | ||
1340 | kfree(fp); | ||
1341 | |||
1342 | return error; | ||
1343 | } | ||
1344 | |||
1345 | /** | ||
1346 | * gfs2_close - called to close a struct file | ||
1347 | * @inode: the inode the struct file belongs to | ||
1348 | * @file: the struct file being closed | ||
1349 | * | ||
1350 | * Returns: errno | ||
1351 | */ | ||
1352 | |||
1353 | static int gfs2_close(struct inode *inode, struct file *file) | ||
1354 | { | ||
1355 | struct gfs2_sbd *sdp = get_v2sdp(inode->i_sb); | ||
1356 | struct gfs2_file *fp; | ||
1357 | |||
1358 | atomic_inc(&sdp->sd_ops_file); | ||
1359 | |||
1360 | fp = get_v2fp(file); | ||
1361 | set_v2fp(file, NULL); | ||
1362 | |||
1363 | if (gfs2_assert_warn(sdp, fp)) | ||
1364 | return -EIO; | ||
1365 | |||
1366 | kfree(fp); | ||
1367 | |||
1368 | return 0; | ||
1369 | } | ||
1370 | |||
1371 | /** | ||
1372 | * gfs2_fsync - sync the dirty data for a file (across the cluster) | ||
1373 | * @file: the file that points to the dentry (we ignore this) | ||
1374 | * @dentry: the dentry that points to the inode to sync | ||
1375 | * | ||
1376 | * Returns: errno | ||
1377 | */ | ||
1378 | |||
1379 | static int gfs2_fsync(struct file *file, struct dentry *dentry, int datasync) | ||
1380 | { | ||
1381 | struct gfs2_inode *ip = get_v2ip(dentry->d_inode); | ||
1382 | |||
1383 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1384 | gfs2_log_flush_glock(ip->i_gl); | ||
1385 | |||
1386 | return 0; | ||
1387 | } | ||
1388 | |||
1389 | /** | ||
1390 | * gfs2_lock - acquire/release a posix lock on a file | ||
1391 | * @file: the file pointer | ||
1392 | * @cmd: either modify or retrieve lock state, possibly wait | ||
1393 | * @fl: type and range of lock | ||
1394 | * | ||
1395 | * Returns: errno | ||
1396 | */ | ||
1397 | |||
1398 | static int gfs2_lock(struct file *file, int cmd, struct file_lock *fl) | ||
1399 | { | ||
1400 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
1401 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
1402 | struct lm_lockname name = | ||
1403 | { .ln_number = ip->i_num.no_addr, | ||
1404 | .ln_type = LM_TYPE_PLOCK }; | ||
1405 | |||
1406 | atomic_inc(&sdp->sd_ops_file); | ||
1407 | |||
1408 | if (!(fl->fl_flags & FL_POSIX)) | ||
1409 | return -ENOLCK; | ||
1410 | if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID) | ||
1411 | return -ENOLCK; | ||
1412 | |||
1413 | if (sdp->sd_args.ar_localflocks) { | ||
1414 | if (IS_GETLK(cmd)) { | ||
1415 | struct file_lock *tmp; | ||
1416 | lock_kernel(); | ||
1417 | tmp = posix_test_lock(file, fl); | ||
1418 | fl->fl_type = F_UNLCK; | ||
1419 | if (tmp) | ||
1420 | memcpy(fl, tmp, sizeof(struct file_lock)); | ||
1421 | unlock_kernel(); | ||
1422 | return 0; | ||
1423 | } else { | ||
1424 | int error; | ||
1425 | lock_kernel(); | ||
1426 | error = posix_lock_file_wait(file, fl); | ||
1427 | unlock_kernel(); | ||
1428 | return error; | ||
1429 | } | ||
1430 | } | ||
1431 | |||
1432 | if (IS_GETLK(cmd)) | ||
1433 | return gfs2_lm_plock_get(sdp, &name, file, fl); | ||
1434 | else if (fl->fl_type == F_UNLCK) | ||
1435 | return gfs2_lm_punlock(sdp, &name, file, fl); | ||
1436 | else | ||
1437 | return gfs2_lm_plock(sdp, &name, file, cmd, fl); | ||
1438 | } | ||
1439 | |||
1440 | /** | ||
1441 | * gfs2_sendfile - Send bytes to a file or socket | ||
1442 | * @in_file: The file to read from | ||
1443 | * @out_file: The file to write to | ||
1444 | * @count: The amount of data | ||
1445 | * @offset: The beginning file offset | ||
1446 | * | ||
1447 | * Outputs: offset - updated according to number of bytes read | ||
1448 | * | ||
1449 | * Returns: The number of bytes sent, errno on failure | ||
1450 | */ | ||
1451 | |||
1452 | static ssize_t gfs2_sendfile(struct file *in_file, loff_t *offset, size_t count, | ||
1453 | read_actor_t actor, void *target) | ||
1454 | { | ||
1455 | struct gfs2_inode *ip = get_v2ip(in_file->f_mapping->host); | ||
1456 | struct gfs2_holder gh; | ||
1457 | ssize_t retval; | ||
1458 | |||
1459 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1460 | |||
1461 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); | ||
1462 | |||
1463 | retval = gfs2_glock_nq_atime(&gh); | ||
1464 | if (retval) | ||
1465 | goto out; | ||
1466 | |||
1467 | if (gfs2_is_jdata(ip)) | ||
1468 | retval = -EOPNOTSUPP; | ||
1469 | else | ||
1470 | retval = generic_file_sendfile(in_file, offset, count, actor, | ||
1471 | target); | ||
1472 | |||
1473 | gfs2_glock_dq(&gh); | ||
1474 | |||
1475 | out: | ||
1476 | gfs2_holder_uninit(&gh); | ||
1477 | |||
1478 | return retval; | ||
1479 | } | ||
1480 | |||
1481 | static int do_flock(struct file *file, int cmd, struct file_lock *fl) | ||
1482 | { | ||
1483 | struct gfs2_file *fp = get_v2fp(file); | ||
1484 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; | ||
1485 | struct gfs2_inode *ip = fp->f_inode; | ||
1486 | struct gfs2_glock *gl; | ||
1487 | unsigned int state; | ||
1488 | int flags; | ||
1489 | int error = 0; | ||
1490 | |||
1491 | state = (fl->fl_type == F_WRLCK) ? LM_ST_EXCLUSIVE : LM_ST_SHARED; | ||
1492 | flags = ((IS_SETLKW(cmd)) ? 0 : LM_FLAG_TRY) | GL_EXACT | GL_NOCACHE; | ||
1493 | |||
1494 | down(&fp->f_fl_mutex); | ||
1495 | |||
1496 | gl = fl_gh->gh_gl; | ||
1497 | if (gl) { | ||
1498 | if (fl_gh->gh_state == state) | ||
1499 | goto out; | ||
1500 | gfs2_glock_hold(gl); | ||
1501 | flock_lock_file_wait(file, | ||
1502 | &(struct file_lock){.fl_type = F_UNLCK}); | ||
1503 | gfs2_glock_dq_uninit(fl_gh); | ||
1504 | } else { | ||
1505 | error = gfs2_glock_get(ip->i_sbd, | ||
1506 | ip->i_num.no_addr, &gfs2_flock_glops, | ||
1507 | CREATE, &gl); | ||
1508 | if (error) | ||
1509 | goto out; | ||
1510 | } | ||
1511 | |||
1512 | gfs2_holder_init(gl, state, flags, fl_gh); | ||
1513 | gfs2_glock_put(gl); | ||
1514 | |||
1515 | error = gfs2_glock_nq(fl_gh); | ||
1516 | if (error) { | ||
1517 | gfs2_holder_uninit(fl_gh); | ||
1518 | if (error == GLR_TRYFAILED) | ||
1519 | error = -EAGAIN; | ||
1520 | } else { | ||
1521 | error = flock_lock_file_wait(file, fl); | ||
1522 | gfs2_assert_warn(ip->i_sbd, !error); | ||
1523 | } | ||
1524 | |||
1525 | out: | ||
1526 | up(&fp->f_fl_mutex); | ||
1527 | |||
1528 | return error; | ||
1529 | } | ||
1530 | |||
1531 | static void do_unflock(struct file *file, struct file_lock *fl) | ||
1532 | { | ||
1533 | struct gfs2_file *fp = get_v2fp(file); | ||
1534 | struct gfs2_holder *fl_gh = &fp->f_fl_gh; | ||
1535 | |||
1536 | down(&fp->f_fl_mutex); | ||
1537 | flock_lock_file_wait(file, fl); | ||
1538 | if (fl_gh->gh_gl) | ||
1539 | gfs2_glock_dq_uninit(fl_gh); | ||
1540 | up(&fp->f_fl_mutex); | ||
1541 | } | ||
1542 | |||
1543 | /** | ||
1544 | * gfs2_flock - acquire/release a flock lock on a file | ||
1545 | * @file: the file pointer | ||
1546 | * @cmd: either modify or retrieve lock state, possibly wait | ||
1547 | * @fl: type and range of lock | ||
1548 | * | ||
1549 | * Returns: errno | ||
1550 | */ | ||
1551 | |||
1552 | static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl) | ||
1553 | { | ||
1554 | struct gfs2_inode *ip = get_v2ip(file->f_mapping->host); | ||
1555 | struct gfs2_sbd *sdp = ip->i_sbd; | ||
1556 | |||
1557 | atomic_inc(&ip->i_sbd->sd_ops_file); | ||
1558 | |||
1559 | if (!(fl->fl_flags & FL_FLOCK)) | ||
1560 | return -ENOLCK; | ||
1561 | if ((ip->i_di.di_mode & (S_ISGID | S_IXGRP)) == S_ISGID) | ||
1562 | return -ENOLCK; | ||
1563 | |||
1564 | if (sdp->sd_args.ar_localflocks) | ||
1565 | return flock_lock_file_wait(file, fl); | ||
1566 | |||
1567 | if (fl->fl_type == F_UNLCK) { | ||
1568 | do_unflock(file, fl); | ||
1569 | return 0; | ||
1570 | } else | ||
1571 | return do_flock(file, cmd, fl); | ||
1572 | } | ||
1573 | |||
1574 | struct file_operations gfs2_file_fops = { | ||
1575 | .llseek = gfs2_llseek, | ||
1576 | .read = gfs2_read, | ||
1577 | .write = gfs2_write, | ||
1578 | .ioctl = gfs2_ioctl, | ||
1579 | .mmap = gfs2_mmap, | ||
1580 | .open = gfs2_open, | ||
1581 | .release = gfs2_close, | ||
1582 | .fsync = gfs2_fsync, | ||
1583 | .lock = gfs2_lock, | ||
1584 | .sendfile = gfs2_sendfile, | ||
1585 | .flock = gfs2_flock, | ||
1586 | }; | ||
1587 | |||
1588 | struct file_operations gfs2_dir_fops = { | ||
1589 | .readdir = gfs2_readdir, | ||
1590 | .ioctl = gfs2_ioctl, | ||
1591 | .open = gfs2_open, | ||
1592 | .release = gfs2_close, | ||
1593 | .fsync = gfs2_fsync, | ||
1594 | .lock = gfs2_lock, | ||
1595 | .flock = gfs2_flock, | ||
1596 | }; | ||
1597 | |||