?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: fix locking for res->tracking and dlm->tracking_list
[linux/fpc-iii.git] / fs / ceph / file.c
blobc8222bfe1e5669b00874274d2d4ee697d9ee2102
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/module.h>
4 #include <linux/sched.h>
5 #include <linux/slab.h>
6 #include <linux/file.h>
7 #include <linux/mount.h>
8 #include <linux/namei.h>
9 #include <linux/writeback.h>
10 #include <linux/falloc.h>
11
12 #include "super.h"
13 #include "mds_client.h"
14 #include "cache.h"
15
16 /*
17 * Ceph file operations
18 *
19 * Implement basic open/close functionality, and implement
20 * read/write.
21 *
22 * We implement three modes of file I/O:
23 * - buffered uses the generic_file_aio_{read,write} helpers
24 *
25 * - synchronous is used when there is multi-client read/write
26 * sharing, avoids the page cache, and synchronously waits for an
27 * ack from the OSD.
28 *
29 * - direct io takes the variant of the sync path that references
30 * user pages directly.
31 *
32 * fsync() flushes and waits on dirty pages, but just queues metadata
33 * for writeback: since the MDS can recover size and mtime there is no
34 * need to wait for MDS acknowledgement.
35 */
36
37 /*
38 * Calculate the length sum of direct io vectors that can
39 * be combined into one page vector.
40 */
41 static size_t dio_get_pagev_size(const struct iov_iter *it)
42 {
43 const struct iovec *iov = it->iov;
44 const struct iovec *iovend = iov + it->nr_segs;
45 size_t size;
46
47 size = iov->iov_len - it->iov_offset;
48 /*
49 * An iov can be page vectored when both the current tail
50 * and the next base are page aligned.
51 */
52 while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) &&
53 (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) {
54 size += iov->iov_len;
55 }
56 dout("dio_get_pagevlen len = %zu\n", size);
57 return size;
58 }
59
60 /*
61 * Allocate a page vector based on (@it, @nbytes).
62 * The return value is the tuple describing a page vector,
63 * that is (@pages, @page_align, @num_pages).
64 */
65 static struct page **
66 dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes,
67 size_t *page_align, int *num_pages)
68 {
69 struct iov_iter tmp_it = *it;
70 size_t align;
71 struct page **pages;
72 int ret = 0, idx, npages;
73
74 align = (unsigned long)(it->iov->iov_base + it->iov_offset) &
75 (PAGE_SIZE - 1);
76 npages = calc_pages_for(align, nbytes);
77 pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL);
78 if (!pages) {
79 pages = vmalloc(sizeof(*pages) * npages);
80 if (!pages)
81 return ERR_PTR(-ENOMEM);
82 }
83
84 for (idx = 0; idx < npages; ) {
85 size_t start;
86 ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes,
87 npages - idx, &start);
88 if (ret < 0)
89 goto fail;
90
91 iov_iter_advance(&tmp_it, ret);
92 nbytes -= ret;
93 idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE;
94 }
95
96 BUG_ON(nbytes != 0);
97 *num_pages = npages;
98 *page_align = align;
99 dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align);
100 return pages;
101 fail:
102 ceph_put_page_vector(pages, idx, false);
103 return ERR_PTR(ret);
104 }
105
106 /*
107 * Prepare an open request. Preallocate ceph_cap to avoid an
108 * inopportune ENOMEM later.
109 */
110 static struct ceph_mds_request *
111 prepare_open_request(struct super_block *sb, int flags, int create_mode)
112 {
113 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
114 struct ceph_mds_client *mdsc = fsc->mdsc;
115 struct ceph_mds_request *req;
116 int want_auth = USE_ANY_MDS;
117 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
118
119 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
120 want_auth = USE_AUTH_MDS;
121
122 req = ceph_mdsc_create_request(mdsc, op, want_auth);
123 if (IS_ERR(req))
124 goto out;
125 req->r_fmode = ceph_flags_to_mode(flags);
126 req->r_args.open.flags = cpu_to_le32(flags);
127 req->r_args.open.mode = cpu_to_le32(create_mode);
128 out:
129 return req;
130 }
131
132 /*
133 * initialize private struct file data.
134 * if we fail, clean up by dropping fmode reference on the ceph_inode
135 */
136 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
137 {
138 struct ceph_file_info *cf;
139 int ret = 0;
140 struct ceph_inode_info *ci = ceph_inode(inode);
141 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
142 struct ceph_mds_client *mdsc = fsc->mdsc;
143
144 switch (inode->i_mode & S_IFMT) {
145 case S_IFREG:
146 /* First file open request creates the cookie, we want to keep
147 * this cookie around for the filetime of the inode as not to
148 * have to worry about fscache register / revoke / operation
149 * races.
150 *
151 * Also, if we know the operation is going to invalidate data
152 * (non readonly) just nuke the cache right away.
153 */
154 ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
155 if ((fmode & CEPH_FILE_MODE_WR))
156 ceph_fscache_invalidate(inode);
157 case S_IFDIR:
158 dout("init_file %p %p 0%o (regular)\n", inode, file,
159 inode->i_mode);
160 cf = kmem_cache_alloc(ceph_file_cachep, GFP_KERNEL | __GFP_ZERO);
161 if (cf == NULL) {
162 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
163 return -ENOMEM;
164 }
165 cf->fmode = fmode;
166 cf->next_offset = 2;
167 cf->readdir_cache_idx = -1;
168 file->private_data = cf;
169 BUG_ON(inode->i_fop->release != ceph_release);
170 break;
171
172 case S_IFLNK:
173 dout("init_file %p %p 0%o (symlink)\n", inode, file,
174 inode->i_mode);
175 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
176 break;
177
178 default:
179 dout("init_file %p %p 0%o (special)\n", inode, file,
180 inode->i_mode);
181 /*
182 * we need to drop the open ref now, since we don't
183 * have .release set to ceph_release.
184 */
185 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
186 BUG_ON(inode->i_fop->release == ceph_release);
187
188 /* call the proper open fop */
189 ret = inode->i_fop->open(inode, file);
190 }
191 return ret;
192 }
193
194 /*
195 * If we already have the requisite capabilities, we can satisfy
196 * the open request locally (no need to request new caps from the
197 * MDS). We do, however, need to inform the MDS (asynchronously)
198 * if our wanted caps set expands.
199 */
200 int ceph_open(struct inode *inode, struct file *file)
201 {
202 struct ceph_inode_info *ci = ceph_inode(inode);
203 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
204 struct ceph_mds_client *mdsc = fsc->mdsc;
205 struct ceph_mds_request *req;
206 struct ceph_file_info *cf = file->private_data;
207 int err;
208 int flags, fmode, wanted;
209
210 if (cf) {
211 dout("open file %p is already opened\n", file);
212 return 0;
213 }
214
215 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
216 flags = file->f_flags & ~(O_CREAT|O_EXCL);
217 if (S_ISDIR(inode->i_mode))
218 flags = O_DIRECTORY; /* mds likes to know */
219
220 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
221 ceph_vinop(inode), file, flags, file->f_flags);
222 fmode = ceph_flags_to_mode(flags);
223 wanted = ceph_caps_for_mode(fmode);
224
225 /* snapped files are read-only */
226 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
227 return -EROFS;
228
229 /* trivially open snapdir */
230 if (ceph_snap(inode) == CEPH_SNAPDIR) {
231 spin_lock(&ci->i_ceph_lock);
232 __ceph_get_fmode(ci, fmode);
233 spin_unlock(&ci->i_ceph_lock);
234 return ceph_init_file(inode, file, fmode);
235 }
236
237 /*
238 * No need to block if we have caps on the auth MDS (for
239 * write) or any MDS (for read). Update wanted set
240 * asynchronously.
241 */
242 spin_lock(&ci->i_ceph_lock);
243 if (__ceph_is_any_real_caps(ci) &&
244 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
245 int mds_wanted = __ceph_caps_mds_wanted(ci);
246 int issued = __ceph_caps_issued(ci, NULL);
247
248 dout("open %p fmode %d want %s issued %s using existing\n",
249 inode, fmode, ceph_cap_string(wanted),
250 ceph_cap_string(issued));
251 __ceph_get_fmode(ci, fmode);
252 spin_unlock(&ci->i_ceph_lock);
253
254 /* adjust wanted? */
255 if ((issued & wanted) != wanted &&
256 (mds_wanted & wanted) != wanted &&
257 ceph_snap(inode) != CEPH_SNAPDIR)
258 ceph_check_caps(ci, 0, NULL);
259
260 return ceph_init_file(inode, file, fmode);
261 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
262 (ci->i_snap_caps & wanted) == wanted) {
263 __ceph_get_fmode(ci, fmode);
264 spin_unlock(&ci->i_ceph_lock);
265 return ceph_init_file(inode, file, fmode);
266 }
267
268 spin_unlock(&ci->i_ceph_lock);
269
270 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
271 req = prepare_open_request(inode->i_sb, flags, 0);
272 if (IS_ERR(req)) {
273 err = PTR_ERR(req);
274 goto out;
275 }
276 req->r_inode = inode;
277 ihold(inode);
278
279 req->r_num_caps = 1;
280 err = ceph_mdsc_do_request(mdsc, NULL, req);
281 if (!err)
282 err = ceph_init_file(inode, file, req->r_fmode);
283 ceph_mdsc_put_request(req);
284 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
285 out:
286 return err;
287 }
288
289
290 /*
291 * Do a lookup + open with a single request. If we get a non-existent
292 * file or symlink, return 1 so the VFS can retry.
293 */
294 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
295 struct file *file, unsigned flags, umode_t mode,
296 int *opened)
297 {
298 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
299 struct ceph_mds_client *mdsc = fsc->mdsc;
300 struct ceph_mds_request *req;
301 struct dentry *dn;
302 struct ceph_acls_info acls = {};
303 int err;
304
305 dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n",
306 dir, dentry, dentry,
307 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
308
309 if (dentry->d_name.len > NAME_MAX)
310 return -ENAMETOOLONG;
311
312 err = ceph_init_dentry(dentry);
313 if (err < 0)
314 return err;
315
316 if (flags & O_CREAT) {
317 err = ceph_pre_init_acls(dir, &mode, &acls);
318 if (err < 0)
319 return err;
320 }
321
322 /* do the open */
323 req = prepare_open_request(dir->i_sb, flags, mode);
324 if (IS_ERR(req)) {
325 err = PTR_ERR(req);
326 goto out_acl;
327 }
328 req->r_dentry = dget(dentry);
329 req->r_num_caps = 2;
330 if (flags & O_CREAT) {
331 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
332 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
333 if (acls.pagelist) {
334 req->r_pagelist = acls.pagelist;
335 acls.pagelist = NULL;
336 }
337 }
338 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
339 err = ceph_mdsc_do_request(mdsc,
340 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
341 req);
342 err = ceph_handle_snapdir(req, dentry, err);
343 if (err)
344 goto out_req;
345
346 if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
347 err = ceph_handle_notrace_create(dir, dentry);
348
349 if (d_unhashed(dentry)) {
350 dn = ceph_finish_lookup(req, dentry, err);
351 if (IS_ERR(dn))
352 err = PTR_ERR(dn);
353 } else {
354 /* we were given a hashed negative dentry */
355 dn = NULL;
356 }
357 if (err)
358 goto out_req;
359 if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) {
360 /* make vfs retry on splice, ENOENT, or symlink */
361 dout("atomic_open finish_no_open on dn %p\n", dn);
362 err = finish_no_open(file, dn);
363 } else {
364 dout("atomic_open finish_open on dn %p\n", dn);
365 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
366 ceph_init_inode_acls(d_inode(dentry), &acls);
367 *opened |= FILE_CREATED;
368 }
369 err = finish_open(file, dentry, ceph_open, opened);
370 }
371 out_req:
372 if (!req->r_err && req->r_target_inode)
373 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
374 ceph_mdsc_put_request(req);
375 out_acl:
376 ceph_release_acls_info(&acls);
377 dout("atomic_open result=%d\n", err);
378 return err;
379 }
380
381 int ceph_release(struct inode *inode, struct file *file)
382 {
383 struct ceph_inode_info *ci = ceph_inode(inode);
384 struct ceph_file_info *cf = file->private_data;
385
386 dout("release inode %p file %p\n", inode, file);
387 ceph_put_fmode(ci, cf->fmode);
388 if (cf->last_readdir)
389 ceph_mdsc_put_request(cf->last_readdir);
390 kfree(cf->last_name);
391 kfree(cf->dir_info);
392 kmem_cache_free(ceph_file_cachep, cf);
393
394 /* wake up anyone waiting for caps on this inode */
395 wake_up_all(&ci->i_cap_wq);
396 return 0;
397 }
398
399 enum {
400 CHECK_EOF = 1,
401 READ_INLINE = 2,
402 };
403
404 /*
405 * Read a range of bytes striped over one or more objects. Iterate over
406 * objects we stripe over. (That's not atomic, but good enough for now.)
407 *
408 * If we get a short result from the OSD, check against i_size; we need to
409 * only return a short read to the caller if we hit EOF.
410 */
411 static int striped_read(struct inode *inode,
412 u64 off, u64 len,
413 struct page **pages, int num_pages,
414 int *checkeof, bool o_direct,
415 unsigned long buf_align)
416 {
417 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
418 struct ceph_inode_info *ci = ceph_inode(inode);
419 u64 pos, this_len, left;
420 int io_align, page_align;
421 int pages_left;
422 int read;
423 struct page **page_pos;
424 int ret;
425 bool hit_stripe, was_short;
426
427 /*
428 * we may need to do multiple reads. not atomic, unfortunately.
429 */
430 pos = off;
431 left = len;
432 page_pos = pages;
433 pages_left = num_pages;
434 read = 0;
435 io_align = off & ~PAGE_MASK;
436
437 more:
438 if (o_direct)
439 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
440 else
441 page_align = pos & ~PAGE_MASK;
442 this_len = left;
443 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
444 &ci->i_layout, pos, &this_len,
445 ci->i_truncate_seq,
446 ci->i_truncate_size,
447 page_pos, pages_left, page_align);
448 if (ret == -ENOENT)
449 ret = 0;
450 hit_stripe = this_len < left;
451 was_short = ret >= 0 && ret < this_len;
452 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
453 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
454
455 if (ret >= 0) {
456 int didpages;
457 if (was_short && (pos + ret < inode->i_size)) {
458 int zlen = min(this_len - ret,
459 inode->i_size - pos - ret);
460 int zoff = (o_direct ? buf_align : io_align) +
461 read + ret;
462 dout(" zero gap %llu to %llu\n",
463 pos + ret, pos + ret + zlen);
464 ceph_zero_page_vector_range(zoff, zlen, pages);
465 ret += zlen;
466 }
467
468 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
469 pos += ret;
470 read = pos - off;
471 left -= ret;
472 page_pos += didpages;
473 pages_left -= didpages;
474
475 /* hit stripe and need continue*/
476 if (left && hit_stripe && pos < inode->i_size)
477 goto more;
478 }
479
480 if (read > 0) {
481 ret = read;
482 /* did we bounce off eof? */
483 if (pos + left > inode->i_size)
484 *checkeof = CHECK_EOF;
485 }
486
487 dout("striped_read returns %d\n", ret);
488 return ret;
489 }
490
491 /*
492 * Completely synchronous read and write methods. Direct from __user
493 * buffer to osd, or directly to user pages (if O_DIRECT).
494 *
495 * If the read spans object boundary, just do multiple reads.
496 */
497 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
498 int *checkeof)
499 {
500 struct file *file = iocb->ki_filp;
501 struct inode *inode = file_inode(file);
502 struct page **pages;
503 u64 off = iocb->ki_pos;
504 int num_pages, ret;
505 size_t len = iov_iter_count(i);
506
507 dout("sync_read on file %p %llu~%u %s\n", file, off,
508 (unsigned)len,
509 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
510
511 if (!len)
512 return 0;
513 /*
514 * flush any page cache pages in this range. this
515 * will make concurrent normal and sync io slow,
516 * but it will at least behave sensibly when they are
517 * in sequence.
518 */
519 ret = filemap_write_and_wait_range(inode->i_mapping, off,
520 off + len);
521 if (ret < 0)
522 return ret;
523
524 if (iocb->ki_flags & IOCB_DIRECT) {
525 while (iov_iter_count(i)) {
526 size_t start;
527 ssize_t n;
528
529 n = dio_get_pagev_size(i);
530 pages = dio_get_pages_alloc(i, n, &start, &num_pages);
531 if (IS_ERR(pages))
532 return PTR_ERR(pages);
533
534 ret = striped_read(inode, off, n,
535 pages, num_pages, checkeof,
536 1, start);
537
538 ceph_put_page_vector(pages, num_pages, true);
539
540 if (ret <= 0)
541 break;
542 off += ret;
543 iov_iter_advance(i, ret);
544 if (ret < n)
545 break;
546 }
547 } else {
548 num_pages = calc_pages_for(off, len);
549 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
550 if (IS_ERR(pages))
551 return PTR_ERR(pages);
552 ret = striped_read(inode, off, len, pages,
553 num_pages, checkeof, 0, 0);
554 if (ret > 0) {
555 int l, k = 0;
556 size_t left = ret;
557
558 while (left) {
559 size_t page_off = off & ~PAGE_MASK;
560 size_t copy = min_t(size_t,
561 PAGE_SIZE - page_off, left);
562 l = copy_page_to_iter(pages[k++], page_off,
563 copy, i);
564 off += l;
565 left -= l;
566 if (l < copy)
567 break;
568 }
569 }
570 ceph_release_page_vector(pages, num_pages);
571 }
572
573 if (off > iocb->ki_pos) {
574 ret = off - iocb->ki_pos;
575 iocb->ki_pos = off;
576 }
577
578 dout("sync_read result %d\n", ret);
579 return ret;
580 }
581
582 /*
583 * Write commit request unsafe callback, called to tell us when a
584 * request is unsafe (that is, in flight--has been handed to the
585 * messenger to send to its target osd). It is called again when
586 * we've received a response message indicating the request is
587 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
588 * is completed early (and unsuccessfully) due to a timeout or
589 * interrupt.
590 *
591 * This is used if we requested both an ACK and ONDISK commit reply
592 * from the OSD.
593 */
594 static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
595 {
596 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
597
598 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
599 unsafe ? "un" : "");
600 if (unsafe) {
601 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
602 spin_lock(&ci->i_unsafe_lock);
603 list_add_tail(&req->r_unsafe_item,
604 &ci->i_unsafe_writes);
605 spin_unlock(&ci->i_unsafe_lock);
606 } else {
607 spin_lock(&ci->i_unsafe_lock);
608 list_del_init(&req->r_unsafe_item);
609 spin_unlock(&ci->i_unsafe_lock);
610 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
611 }
612 }
613
614
615 /*
616 * Synchronous write, straight from __user pointer or user pages.
617 *
618 * If write spans object boundary, just do multiple writes. (For a
619 * correct atomic write, we should e.g. take write locks on all
620 * objects, rollback on failure, etc.)
621 */
622 static ssize_t
623 ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
624 struct ceph_snap_context *snapc)
625 {
626 struct file *file = iocb->ki_filp;
627 struct inode *inode = file_inode(file);
628 struct ceph_inode_info *ci = ceph_inode(inode);
629 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
630 struct ceph_vino vino;
631 struct ceph_osd_request *req;
632 struct page **pages;
633 int num_pages;
634 int written = 0;
635 int flags;
636 int check_caps = 0;
637 int ret;
638 struct timespec mtime = CURRENT_TIME;
639 size_t count = iov_iter_count(from);
640
641 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
642 return -EROFS;
643
644 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
645 (unsigned)count);
646
647 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
648 if (ret < 0)
649 return ret;
650
651 ret = invalidate_inode_pages2_range(inode->i_mapping,
652 pos >> PAGE_CACHE_SHIFT,
653 (pos + count) >> PAGE_CACHE_SHIFT);
654 if (ret < 0)
655 dout("invalidate_inode_pages2_range returned %d\n", ret);
656
657 flags = CEPH_OSD_FLAG_ORDERSNAP |
658 CEPH_OSD_FLAG_ONDISK |
659 CEPH_OSD_FLAG_WRITE;
660
661 while (iov_iter_count(from) > 0) {
662 u64 len = dio_get_pagev_size(from);
663 size_t start;
664 ssize_t n;
665
666 vino = ceph_vino(inode);
667 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
668 vino, pos, &len, 0,
669 2,/*include a 'startsync' command*/
670 CEPH_OSD_OP_WRITE, flags, snapc,
671 ci->i_truncate_seq,
672 ci->i_truncate_size,
673 false);
674 if (IS_ERR(req)) {
675 ret = PTR_ERR(req);
676 break;
677 }
678
679 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0);
680
681 n = len;
682 pages = dio_get_pages_alloc(from, len, &start, &num_pages);
683 if (IS_ERR(pages)) {
684 ceph_osdc_put_request(req);
685 ret = PTR_ERR(pages);
686 break;
687 }
688
689 /*
690 * throw out any page cache pages in this range. this
691 * may block.
692 */
693 truncate_inode_pages_range(inode->i_mapping, pos,
694 (pos+n) | (PAGE_CACHE_SIZE-1));
695 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
696 false, false);
697
698 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
699 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
700
701 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
702 if (!ret)
703 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
704
705 ceph_put_page_vector(pages, num_pages, false);
706
707 ceph_osdc_put_request(req);
708 if (ret)
709 break;
710 pos += n;
711 written += n;
712 iov_iter_advance(from, n);
713
714 if (pos > i_size_read(inode)) {
715 check_caps = ceph_inode_set_size(inode, pos);
716 if (check_caps)
717 ceph_check_caps(ceph_inode(inode),
718 CHECK_CAPS_AUTHONLY,
719 NULL);
720 }
721 }
722
723 if (ret != -EOLDSNAPC && written > 0) {
724 iocb->ki_pos = pos;
725 ret = written;
726 }
727 return ret;
728 }
729
730
731 /*
732 * Synchronous write, straight from __user pointer or user pages.
733 *
734 * If write spans object boundary, just do multiple writes. (For a
735 * correct atomic write, we should e.g. take write locks on all
736 * objects, rollback on failure, etc.)
737 */
738 static ssize_t
739 ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
740 struct ceph_snap_context *snapc)
741 {
742 struct file *file = iocb->ki_filp;
743 struct inode *inode = file_inode(file);
744 struct ceph_inode_info *ci = ceph_inode(inode);
745 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
746 struct ceph_vino vino;
747 struct ceph_osd_request *req;
748 struct page **pages;
749 u64 len;
750 int num_pages;
751 int written = 0;
752 int flags;
753 int check_caps = 0;
754 int ret;
755 struct timespec mtime = CURRENT_TIME;
756 size_t count = iov_iter_count(from);
757
758 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
759 return -EROFS;
760
761 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
762
763 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
764 if (ret < 0)
765 return ret;
766
767 ret = invalidate_inode_pages2_range(inode->i_mapping,
768 pos >> PAGE_CACHE_SHIFT,
769 (pos + count) >> PAGE_CACHE_SHIFT);
770 if (ret < 0)
771 dout("invalidate_inode_pages2_range returned %d\n", ret);
772
773 flags = CEPH_OSD_FLAG_ORDERSNAP |
774 CEPH_OSD_FLAG_ONDISK |
775 CEPH_OSD_FLAG_WRITE |
776 CEPH_OSD_FLAG_ACK;
777
778 while ((len = iov_iter_count(from)) > 0) {
779 size_t left;
780 int n;
781
782 vino = ceph_vino(inode);
783 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
784 vino, pos, &len, 0, 1,
785 CEPH_OSD_OP_WRITE, flags, snapc,
786 ci->i_truncate_seq,
787 ci->i_truncate_size,
788 false);
789 if (IS_ERR(req)) {
790 ret = PTR_ERR(req);
791 break;
792 }
793
794 /*
795 * write from beginning of first page,
796 * regardless of io alignment
797 */
798 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
799
800 pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL);
801 if (IS_ERR(pages)) {
802 ret = PTR_ERR(pages);
803 goto out;
804 }
805
806 left = len;
807 for (n = 0; n < num_pages; n++) {
808 size_t plen = min_t(size_t, left, PAGE_SIZE);
809 ret = copy_page_from_iter(pages[n], 0, plen, from);
810 if (ret != plen) {
811 ret = -EFAULT;
812 break;
813 }
814 left -= ret;
815 }
816
817 if (ret < 0) {
818 ceph_release_page_vector(pages, num_pages);
819 goto out;
820 }
821
822 /* get a second commit callback */
823 req->r_unsafe_callback = ceph_sync_write_unsafe;
824 req->r_inode = inode;
825
826 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
827 false, true);
828
829 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
830 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
831
832 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
833 if (!ret)
834 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
835
836 out:
837 ceph_osdc_put_request(req);
838 if (ret == 0) {
839 pos += len;
840 written += len;
841
842 if (pos > i_size_read(inode)) {
843 check_caps = ceph_inode_set_size(inode, pos);
844 if (check_caps)
845 ceph_check_caps(ceph_inode(inode),
846 CHECK_CAPS_AUTHONLY,
847 NULL);
848 }
849 } else
850 break;
851 }
852
853 if (ret != -EOLDSNAPC && written > 0) {
854 ret = written;
855 iocb->ki_pos = pos;
856 }
857 return ret;
858 }
859
860 /*
861 * Wrap generic_file_aio_read with checks for cap bits on the inode.
862 * Atomically grab references, so that those bits are not released
863 * back to the MDS mid-read.
864 *
865 * Hmm, the sync read case isn't actually async... should it be?
866 */
867 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
868 {
869 struct file *filp = iocb->ki_filp;
870 struct ceph_file_info *fi = filp->private_data;
871 size_t len = iov_iter_count(to);
872 struct inode *inode = file_inode(filp);
873 struct ceph_inode_info *ci = ceph_inode(inode);
874 struct page *pinned_page = NULL;
875 ssize_t ret;
876 int want, got = 0;
877 int retry_op = 0, read = 0;
878
879 again:
880 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
881 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
882
883 if (fi->fmode & CEPH_FILE_MODE_LAZY)
884 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
885 else
886 want = CEPH_CAP_FILE_CACHE;
887 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
888 if (ret < 0)
889 return ret;
890
891 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
892 (iocb->ki_flags & IOCB_DIRECT) ||
893 (fi->flags & CEPH_F_SYNC)) {
894
895 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
896 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
897 ceph_cap_string(got));
898
899 if (ci->i_inline_version == CEPH_INLINE_NONE) {
900 /* hmm, this isn't really async... */
901 ret = ceph_sync_read(iocb, to, &retry_op);
902 } else {
903 retry_op = READ_INLINE;
904 }
905 } else {
906 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
907 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
908 ceph_cap_string(got));
909
910 ret = generic_file_read_iter(iocb, to);
911 }
912 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
913 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
914 if (pinned_page) {
915 page_cache_release(pinned_page);
916 pinned_page = NULL;
917 }
918 ceph_put_cap_refs(ci, got);
919 if (retry_op && ret >= 0) {
920 int statret;
921 struct page *page = NULL;
922 loff_t i_size;
923 if (retry_op == READ_INLINE) {
924 page = __page_cache_alloc(GFP_KERNEL);
925 if (!page)
926 return -ENOMEM;
927 }
928
929 statret = __ceph_do_getattr(inode, page,
930 CEPH_STAT_CAP_INLINE_DATA, !!page);
931 if (statret < 0) {
932 if (page)
933 __free_page(page);
934 if (statret == -ENODATA) {
935 BUG_ON(retry_op != READ_INLINE);
936 goto again;
937 }
938 return statret;
939 }
940
941 i_size = i_size_read(inode);
942 if (retry_op == READ_INLINE) {
943 BUG_ON(ret > 0 || read > 0);
944 if (iocb->ki_pos < i_size &&
945 iocb->ki_pos < PAGE_CACHE_SIZE) {
946 loff_t end = min_t(loff_t, i_size,
947 iocb->ki_pos + len);
948 end = min_t(loff_t, end, PAGE_CACHE_SIZE);
949 if (statret < end)
950 zero_user_segment(page, statret, end);
951 ret = copy_page_to_iter(page,
952 iocb->ki_pos & ~PAGE_MASK,
953 end - iocb->ki_pos, to);
954 iocb->ki_pos += ret;
955 read += ret;
956 }
957 if (iocb->ki_pos < i_size && read < len) {
958 size_t zlen = min_t(size_t, len - read,
959 i_size - iocb->ki_pos);
960 ret = iov_iter_zero(zlen, to);
961 iocb->ki_pos += ret;
962 read += ret;
963 }
964 __free_pages(page, 0);
965 return read;
966 }
967
968 /* hit EOF or hole? */
969 if (retry_op == CHECK_EOF && iocb->ki_pos < i_size &&
970 ret < len) {
971 dout("sync_read hit hole, ppos %lld < size %lld"
972 ", reading more\n", iocb->ki_pos,
973 inode->i_size);
974
975 read += ret;
976 len -= ret;
977 retry_op = 0;
978 goto again;
979 }
980 }
981
982 if (ret >= 0)
983 ret += read;
984
985 return ret;
986 }
987
988 /*
989 * Take cap references to avoid releasing caps to MDS mid-write.
990 *
991 * If we are synchronous, and write with an old snap context, the OSD
992 * may return EOLDSNAPC. In that case, retry the write.. _after_
993 * dropping our cap refs and allowing the pending snap to logically
994 * complete _before_ this write occurs.
995 *
996 * If we are near ENOSPC, write synchronously.
997 */
998 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
999 {
1000 struct file *file = iocb->ki_filp;
1001 struct ceph_file_info *fi = file->private_data;
1002 struct inode *inode = file_inode(file);
1003 struct ceph_inode_info *ci = ceph_inode(inode);
1004 struct ceph_osd_client *osdc =
1005 &ceph_sb_to_client(inode->i_sb)->client->osdc;
1006 struct ceph_cap_flush *prealloc_cf;
1007 ssize_t count, written = 0;
1008 int err, want, got;
1009 loff_t pos;
1010
1011 if (ceph_snap(inode) != CEPH_NOSNAP)
1012 return -EROFS;
1013
1014 prealloc_cf = ceph_alloc_cap_flush();
1015 if (!prealloc_cf)
1016 return -ENOMEM;
1017
1018 mutex_lock(&inode->i_mutex);
1019
1020 /* We can write back this queue in page reclaim */
1021 current->backing_dev_info = inode_to_bdi(inode);
1022
1023 if (iocb->ki_flags & IOCB_APPEND) {
1024 err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1025 if (err < 0)
1026 goto out;
1027 }
1028
1029 err = generic_write_checks(iocb, from);
1030 if (err <= 0)
1031 goto out;
1032
1033 pos = iocb->ki_pos;
1034 count = iov_iter_count(from);
1035 err = file_remove_privs(file);
1036 if (err)
1037 goto out;
1038
1039 err = file_update_time(file);
1040 if (err)
1041 goto out;
1042
1043 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1044 err = ceph_uninline_data(file, NULL);
1045 if (err < 0)
1046 goto out;
1047 }
1048
1049 retry_snap:
1050 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
1051 err = -ENOSPC;
1052 goto out;
1053 }
1054
1055 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
1056 inode, ceph_vinop(inode), pos, count, inode->i_size);
1057 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1058 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1059 else
1060 want = CEPH_CAP_FILE_BUFFER;
1061 got = 0;
1062 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count,
1063 &got, NULL);
1064 if (err < 0)
1065 goto out;
1066
1067 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
1068 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
1069
1070 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
1071 (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
1072 struct ceph_snap_context *snapc;
1073 struct iov_iter data;
1074 mutex_unlock(&inode->i_mutex);
1075
1076 spin_lock(&ci->i_ceph_lock);
1077 if (__ceph_have_pending_cap_snap(ci)) {
1078 struct ceph_cap_snap *capsnap =
1079 list_last_entry(&ci->i_cap_snaps,
1080 struct ceph_cap_snap,
1081 ci_item);
1082 snapc = ceph_get_snap_context(capsnap->context);
1083 } else {
1084 BUG_ON(!ci->i_head_snapc);
1085 snapc = ceph_get_snap_context(ci->i_head_snapc);
1086 }
1087 spin_unlock(&ci->i_ceph_lock);
1088
1089 /* we might need to revert back to that point */
1090 data = *from;
1091 if (iocb->ki_flags & IOCB_DIRECT)
1092 written = ceph_sync_direct_write(iocb, &data, pos,
1093 snapc);
1094 else
1095 written = ceph_sync_write(iocb, &data, pos, snapc);
1096 if (written == -EOLDSNAPC) {
1097 dout("aio_write %p %llx.%llx %llu~%u"
1098 "got EOLDSNAPC, retrying\n",
1099 inode, ceph_vinop(inode),
1100 pos, (unsigned)count);
1101 mutex_lock(&inode->i_mutex);
1102 goto retry_snap;
1103 }
1104 if (written > 0)
1105 iov_iter_advance(from, written);
1106 ceph_put_snap_context(snapc);
1107 } else {
1108 loff_t old_size = inode->i_size;
1109 /*
1110 * No need to acquire the i_truncate_mutex. Because
1111 * the MDS revokes Fwb caps before sending truncate
1112 * message to us. We can't get Fwb cap while there
1113 * are pending vmtruncate. So write and vmtruncate
1114 * can not run at the same time
1115 */
1116 written = generic_perform_write(file, from, pos);
1117 if (likely(written >= 0))
1118 iocb->ki_pos = pos + written;
1119 if (inode->i_size > old_size)
1120 ceph_fscache_update_objectsize(inode);
1121 mutex_unlock(&inode->i_mutex);
1122 }
1123
1124 if (written >= 0) {
1125 int dirty;
1126 spin_lock(&ci->i_ceph_lock);
1127 ci->i_inline_version = CEPH_INLINE_NONE;
1128 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1129 &prealloc_cf);
1130 spin_unlock(&ci->i_ceph_lock);
1131 if (dirty)
1132 __mark_inode_dirty(inode, dirty);
1133 }
1134
1135 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1136 inode, ceph_vinop(inode), pos, (unsigned)count,
1137 ceph_cap_string(got));
1138 ceph_put_cap_refs(ci, got);
1139
1140 if (written >= 0 &&
1141 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
1142 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
1143 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
1144 if (err < 0)
1145 written = err;
1146 }
1147
1148 goto out_unlocked;
1149
1150 out:
1151 mutex_unlock(&inode->i_mutex);
1152 out_unlocked:
1153 ceph_free_cap_flush(prealloc_cf);
1154 current->backing_dev_info = NULL;
1155 return written ? written : err;
1156 }
1157
1158 /*
1159 * llseek. be sure to verify file size on SEEK_END.
1160 */
1161 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
1162 {
1163 struct inode *inode = file->f_mapping->host;
1164 int ret;
1165
1166 mutex_lock(&inode->i_mutex);
1167
1168 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
1169 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
1170 if (ret < 0) {
1171 offset = ret;
1172 goto out;
1173 }
1174 }
1175
1176 switch (whence) {
1177 case SEEK_END:
1178 offset += inode->i_size;
1179 break;
1180 case SEEK_CUR:
1181 /*
1182 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1183 * position-querying operation. Avoid rewriting the "same"
1184 * f_pos value back to the file because a concurrent read(),
1185 * write() or lseek() might have altered it
1186 */
1187 if (offset == 0) {
1188 offset = file->f_pos;
1189 goto out;
1190 }
1191 offset += file->f_pos;
1192 break;
1193 case SEEK_DATA:
1194 if (offset >= inode->i_size) {
1195 ret = -ENXIO;
1196 goto out;
1197 }
1198 break;
1199 case SEEK_HOLE:
1200 if (offset >= inode->i_size) {
1201 ret = -ENXIO;
1202 goto out;
1203 }
1204 offset = inode->i_size;
1205 break;
1206 }
1207
1208 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1209
1210 out:
1211 mutex_unlock(&inode->i_mutex);
1212 return offset;
1213 }
1214
1215 static inline void ceph_zero_partial_page(
1216 struct inode *inode, loff_t offset, unsigned size)
1217 {
1218 struct page *page;
1219 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1220
1221 page = find_lock_page(inode->i_mapping, index);
1222 if (page) {
1223 wait_on_page_writeback(page);
1224 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1225 unlock_page(page);
1226 page_cache_release(page);
1227 }
1228 }
1229
1230 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1231 loff_t length)
1232 {
1233 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1234 if (offset < nearly) {
1235 loff_t size = nearly - offset;
1236 if (length < size)
1237 size = length;
1238 ceph_zero_partial_page(inode, offset, size);
1239 offset += size;
1240 length -= size;
1241 }
1242 if (length >= PAGE_CACHE_SIZE) {
1243 loff_t size = round_down(length, PAGE_CACHE_SIZE);
1244 truncate_pagecache_range(inode, offset, offset + size - 1);
1245 offset += size;
1246 length -= size;
1247 }
1248 if (length)
1249 ceph_zero_partial_page(inode, offset, length);
1250 }
1251
1252 static int ceph_zero_partial_object(struct inode *inode,
1253 loff_t offset, loff_t *length)
1254 {
1255 struct ceph_inode_info *ci = ceph_inode(inode);
1256 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1257 struct ceph_osd_request *req;
1258 int ret = 0;
1259 loff_t zero = 0;
1260 int op;
1261
1262 if (!length) {
1263 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1264 length = &zero;
1265 } else {
1266 op = CEPH_OSD_OP_ZERO;
1267 }
1268
1269 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1270 ceph_vino(inode),
1271 offset, length,
1272 0, 1, op,
1273 CEPH_OSD_FLAG_WRITE |
1274 CEPH_OSD_FLAG_ONDISK,
1275 NULL, 0, 0, false);
1276 if (IS_ERR(req)) {
1277 ret = PTR_ERR(req);
1278 goto out;
1279 }
1280
1281 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1282 &inode->i_mtime);
1283
1284 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1285 if (!ret) {
1286 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1287 if (ret == -ENOENT)
1288 ret = 0;
1289 }
1290 ceph_osdc_put_request(req);
1291
1292 out:
1293 return ret;
1294 }
1295
1296 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1297 {
1298 int ret = 0;
1299 struct ceph_inode_info *ci = ceph_inode(inode);
1300 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1301 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1302 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1303 u64 object_set_size = object_size * stripe_count;
1304 u64 nearly, t;
1305
1306 /* round offset up to next period boundary */
1307 nearly = offset + object_set_size - 1;
1308 t = nearly;
1309 nearly -= do_div(t, object_set_size);
1310
1311 while (length && offset < nearly) {
1312 loff_t size = length;
1313 ret = ceph_zero_partial_object(inode, offset, &size);
1314 if (ret < 0)
1315 return ret;
1316 offset += size;
1317 length -= size;
1318 }
1319 while (length >= object_set_size) {
1320 int i;
1321 loff_t pos = offset;
1322 for (i = 0; i < stripe_count; ++i) {
1323 ret = ceph_zero_partial_object(inode, pos, NULL);
1324 if (ret < 0)
1325 return ret;
1326 pos += stripe_unit;
1327 }
1328 offset += object_set_size;
1329 length -= object_set_size;
1330 }
1331 while (length) {
1332 loff_t size = length;
1333 ret = ceph_zero_partial_object(inode, offset, &size);
1334 if (ret < 0)
1335 return ret;
1336 offset += size;
1337 length -= size;
1338 }
1339 return ret;
1340 }
1341
1342 static long ceph_fallocate(struct file *file, int mode,
1343 loff_t offset, loff_t length)
1344 {
1345 struct ceph_file_info *fi = file->private_data;
1346 struct inode *inode = file_inode(file);
1347 struct ceph_inode_info *ci = ceph_inode(inode);
1348 struct ceph_osd_client *osdc =
1349 &ceph_inode_to_client(inode)->client->osdc;
1350 struct ceph_cap_flush *prealloc_cf;
1351 int want, got = 0;
1352 int dirty;
1353 int ret = 0;
1354 loff_t endoff = 0;
1355 loff_t size;
1356
1357 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1358 return -EOPNOTSUPP;
1359
1360 if (!S_ISREG(inode->i_mode))
1361 return -EOPNOTSUPP;
1362
1363 prealloc_cf = ceph_alloc_cap_flush();
1364 if (!prealloc_cf)
1365 return -ENOMEM;
1366
1367 mutex_lock(&inode->i_mutex);
1368
1369 if (ceph_snap(inode) != CEPH_NOSNAP) {
1370 ret = -EROFS;
1371 goto unlock;
1372 }
1373
1374 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1375 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1376 ret = -ENOSPC;
1377 goto unlock;
1378 }
1379
1380 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1381 ret = ceph_uninline_data(file, NULL);
1382 if (ret < 0)
1383 goto unlock;
1384 }
1385
1386 size = i_size_read(inode);
1387 if (!(mode & FALLOC_FL_KEEP_SIZE))
1388 endoff = offset + length;
1389
1390 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1391 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1392 else
1393 want = CEPH_CAP_FILE_BUFFER;
1394
1395 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL);
1396 if (ret < 0)
1397 goto unlock;
1398
1399 if (mode & FALLOC_FL_PUNCH_HOLE) {
1400 if (offset < size)
1401 ceph_zero_pagecache_range(inode, offset, length);
1402 ret = ceph_zero_objects(inode, offset, length);
1403 } else if (endoff > size) {
1404 truncate_pagecache_range(inode, size, -1);
1405 if (ceph_inode_set_size(inode, endoff))
1406 ceph_check_caps(ceph_inode(inode),
1407 CHECK_CAPS_AUTHONLY, NULL);
1408 }
1409
1410 if (!ret) {
1411 spin_lock(&ci->i_ceph_lock);
1412 ci->i_inline_version = CEPH_INLINE_NONE;
1413 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1414 &prealloc_cf);
1415 spin_unlock(&ci->i_ceph_lock);
1416 if (dirty)
1417 __mark_inode_dirty(inode, dirty);
1418 }
1419
1420 ceph_put_cap_refs(ci, got);
1421 unlock:
1422 mutex_unlock(&inode->i_mutex);
1423 ceph_free_cap_flush(prealloc_cf);
1424 return ret;
1425 }
1426
1427 const struct file_operations ceph_file_fops = {
1428 .open = ceph_open,
1429 .release = ceph_release,
1430 .llseek = ceph_llseek,
1431 .read_iter = ceph_read_iter,
1432 .write_iter = ceph_write_iter,
1433 .mmap = ceph_mmap,
1434 .fsync = ceph_fsync,
1435 .lock = ceph_lock,
1436 .flock = ceph_flock,
1437 .splice_read = generic_file_splice_read,
1438 .splice_write = iter_file_splice_write,
1439 .unlocked_ioctl = ceph_ioctl,
1440 .compat_ioctl = ceph_ioctl,
1441 .fallocate = ceph_fallocate,
1442 };
1443
Linux with added FPC-III support