xref: /openbmc/linux/fs/ceph/file.c (revision 94c7b6fc)
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/aio.h>
11 #include <linux/falloc.h>
12 
13 #include "super.h"
14 #include "mds_client.h"
15 #include "cache.h"
16 
17 /*
18  * Ceph file operations
19  *
20  * Implement basic open/close functionality, and implement
21  * read/write.
22  *
23  * We implement three modes of file I/O:
24  *  - buffered uses the generic_file_aio_{read,write} helpers
25  *
26  *  - synchronous is used when there is multi-client read/write
27  *    sharing, avoids the page cache, and synchronously waits for an
28  *    ack from the OSD.
29  *
30  *  - direct io takes the variant of the sync path that references
31  *    user pages directly.
32  *
33  * fsync() flushes and waits on dirty pages, but just queues metadata
34  * for writeback: since the MDS can recover size and mtime there is no
35  * need to wait for MDS acknowledgement.
36  */
37 
38 
39 /*
40  * Prepare an open request.  Preallocate ceph_cap to avoid an
41  * inopportune ENOMEM later.
42  */
43 static struct ceph_mds_request *
44 prepare_open_request(struct super_block *sb, int flags, int create_mode)
45 {
46 	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
47 	struct ceph_mds_client *mdsc = fsc->mdsc;
48 	struct ceph_mds_request *req;
49 	int want_auth = USE_ANY_MDS;
50 	int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
51 
52 	if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
53 		want_auth = USE_AUTH_MDS;
54 
55 	req = ceph_mdsc_create_request(mdsc, op, want_auth);
56 	if (IS_ERR(req))
57 		goto out;
58 	req->r_fmode = ceph_flags_to_mode(flags);
59 	req->r_args.open.flags = cpu_to_le32(flags);
60 	req->r_args.open.mode = cpu_to_le32(create_mode);
61 out:
62 	return req;
63 }
64 
65 /*
66  * initialize private struct file data.
67  * if we fail, clean up by dropping fmode reference on the ceph_inode
68  */
69 static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
70 {
71 	struct ceph_file_info *cf;
72 	int ret = 0;
73 	struct ceph_inode_info *ci = ceph_inode(inode);
74 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
75 	struct ceph_mds_client *mdsc = fsc->mdsc;
76 
77 	switch (inode->i_mode & S_IFMT) {
78 	case S_IFREG:
79 		/* First file open request creates the cookie, we want to keep
80 		 * this cookie around for the filetime of the inode as not to
81 		 * have to worry about fscache register / revoke / operation
82 		 * races.
83 		 *
84 		 * Also, if we know the operation is going to invalidate data
85 		 * (non readonly) just nuke the cache right away.
86 		 */
87 		ceph_fscache_register_inode_cookie(mdsc->fsc, ci);
88 		if ((fmode & CEPH_FILE_MODE_WR))
89 			ceph_fscache_invalidate(inode);
90 	case S_IFDIR:
91 		dout("init_file %p %p 0%o (regular)\n", inode, file,
92 		     inode->i_mode);
93 		cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
94 		if (cf == NULL) {
95 			ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
96 			return -ENOMEM;
97 		}
98 		cf->fmode = fmode;
99 		cf->next_offset = 2;
100 		file->private_data = cf;
101 		BUG_ON(inode->i_fop->release != ceph_release);
102 		break;
103 
104 	case S_IFLNK:
105 		dout("init_file %p %p 0%o (symlink)\n", inode, file,
106 		     inode->i_mode);
107 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
108 		break;
109 
110 	default:
111 		dout("init_file %p %p 0%o (special)\n", inode, file,
112 		     inode->i_mode);
113 		/*
114 		 * we need to drop the open ref now, since we don't
115 		 * have .release set to ceph_release.
116 		 */
117 		ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
118 		BUG_ON(inode->i_fop->release == ceph_release);
119 
120 		/* call the proper open fop */
121 		ret = inode->i_fop->open(inode, file);
122 	}
123 	return ret;
124 }
125 
126 /*
127  * If we already have the requisite capabilities, we can satisfy
128  * the open request locally (no need to request new caps from the
129  * MDS).  We do, however, need to inform the MDS (asynchronously)
130  * if our wanted caps set expands.
131  */
132 int ceph_open(struct inode *inode, struct file *file)
133 {
134 	struct ceph_inode_info *ci = ceph_inode(inode);
135 	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
136 	struct ceph_mds_client *mdsc = fsc->mdsc;
137 	struct ceph_mds_request *req;
138 	struct ceph_file_info *cf = file->private_data;
139 	struct inode *parent_inode = NULL;
140 	int err;
141 	int flags, fmode, wanted;
142 
143 	if (cf) {
144 		dout("open file %p is already opened\n", file);
145 		return 0;
146 	}
147 
148 	/* filter out O_CREAT|O_EXCL; vfs did that already.  yuck. */
149 	flags = file->f_flags & ~(O_CREAT|O_EXCL);
150 	if (S_ISDIR(inode->i_mode))
151 		flags = O_DIRECTORY;  /* mds likes to know */
152 
153 	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
154 	     ceph_vinop(inode), file, flags, file->f_flags);
155 	fmode = ceph_flags_to_mode(flags);
156 	wanted = ceph_caps_for_mode(fmode);
157 
158 	/* snapped files are read-only */
159 	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
160 		return -EROFS;
161 
162 	/* trivially open snapdir */
163 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
164 		spin_lock(&ci->i_ceph_lock);
165 		__ceph_get_fmode(ci, fmode);
166 		spin_unlock(&ci->i_ceph_lock);
167 		return ceph_init_file(inode, file, fmode);
168 	}
169 
170 	/*
171 	 * No need to block if we have caps on the auth MDS (for
172 	 * write) or any MDS (for read).  Update wanted set
173 	 * asynchronously.
174 	 */
175 	spin_lock(&ci->i_ceph_lock);
176 	if (__ceph_is_any_real_caps(ci) &&
177 	    (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
178 		int mds_wanted = __ceph_caps_mds_wanted(ci);
179 		int issued = __ceph_caps_issued(ci, NULL);
180 
181 		dout("open %p fmode %d want %s issued %s using existing\n",
182 		     inode, fmode, ceph_cap_string(wanted),
183 		     ceph_cap_string(issued));
184 		__ceph_get_fmode(ci, fmode);
185 		spin_unlock(&ci->i_ceph_lock);
186 
187 		/* adjust wanted? */
188 		if ((issued & wanted) != wanted &&
189 		    (mds_wanted & wanted) != wanted &&
190 		    ceph_snap(inode) != CEPH_SNAPDIR)
191 			ceph_check_caps(ci, 0, NULL);
192 
193 		return ceph_init_file(inode, file, fmode);
194 	} else if (ceph_snap(inode) != CEPH_NOSNAP &&
195 		   (ci->i_snap_caps & wanted) == wanted) {
196 		__ceph_get_fmode(ci, fmode);
197 		spin_unlock(&ci->i_ceph_lock);
198 		return ceph_init_file(inode, file, fmode);
199 	}
200 
201 	spin_unlock(&ci->i_ceph_lock);
202 
203 	dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
204 	req = prepare_open_request(inode->i_sb, flags, 0);
205 	if (IS_ERR(req)) {
206 		err = PTR_ERR(req);
207 		goto out;
208 	}
209 	req->r_inode = inode;
210 	ihold(inode);
211 
212 	req->r_num_caps = 1;
213 	if (flags & O_CREAT)
214 		parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
215 	err = ceph_mdsc_do_request(mdsc, parent_inode, req);
216 	iput(parent_inode);
217 	if (!err)
218 		err = ceph_init_file(inode, file, req->r_fmode);
219 	ceph_mdsc_put_request(req);
220 	dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
221 out:
222 	return err;
223 }
224 
225 
226 /*
227  * Do a lookup + open with a single request.  If we get a non-existent
228  * file or symlink, return 1 so the VFS can retry.
229  */
230 int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
231 		     struct file *file, unsigned flags, umode_t mode,
232 		     int *opened)
233 {
234 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
235 	struct ceph_mds_client *mdsc = fsc->mdsc;
236 	struct ceph_mds_request *req;
237 	struct dentry *dn;
238 	int err;
239 
240 	dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
241 	     dir, dentry, dentry->d_name.len, dentry->d_name.name,
242 	     d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
243 
244 	if (dentry->d_name.len > NAME_MAX)
245 		return -ENAMETOOLONG;
246 
247 	err = ceph_init_dentry(dentry);
248 	if (err < 0)
249 		return err;
250 
251 	/* do the open */
252 	req = prepare_open_request(dir->i_sb, flags, mode);
253 	if (IS_ERR(req))
254 		return PTR_ERR(req);
255 	req->r_dentry = dget(dentry);
256 	req->r_num_caps = 2;
257 	if (flags & O_CREAT) {
258 		req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
259 		req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
260 	}
261 	req->r_locked_dir = dir;           /* caller holds dir->i_mutex */
262 	err = ceph_mdsc_do_request(mdsc,
263 				   (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
264 				   req);
265 	if (err)
266 		goto out_err;
267 
268 	err = ceph_handle_snapdir(req, dentry, err);
269 	if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
270 		err = ceph_handle_notrace_create(dir, dentry);
271 
272 	if (d_unhashed(dentry)) {
273 		dn = ceph_finish_lookup(req, dentry, err);
274 		if (IS_ERR(dn))
275 			err = PTR_ERR(dn);
276 	} else {
277 		/* we were given a hashed negative dentry */
278 		dn = NULL;
279 	}
280 	if (err)
281 		goto out_err;
282 	if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
283 		/* make vfs retry on splice, ENOENT, or symlink */
284 		dout("atomic_open finish_no_open on dn %p\n", dn);
285 		err = finish_no_open(file, dn);
286 	} else {
287 		dout("atomic_open finish_open on dn %p\n", dn);
288 		if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
289 			ceph_init_acl(dentry, dentry->d_inode, dir);
290 			*opened |= FILE_CREATED;
291 		}
292 		err = finish_open(file, dentry, ceph_open, opened);
293 	}
294 out_err:
295 	if (!req->r_err && req->r_target_inode)
296 		ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
297 	ceph_mdsc_put_request(req);
298 	dout("atomic_open result=%d\n", err);
299 	return err;
300 }
301 
302 int ceph_release(struct inode *inode, struct file *file)
303 {
304 	struct ceph_inode_info *ci = ceph_inode(inode);
305 	struct ceph_file_info *cf = file->private_data;
306 
307 	dout("release inode %p file %p\n", inode, file);
308 	ceph_put_fmode(ci, cf->fmode);
309 	if (cf->last_readdir)
310 		ceph_mdsc_put_request(cf->last_readdir);
311 	kfree(cf->last_name);
312 	kfree(cf->dir_info);
313 	dput(cf->dentry);
314 	kmem_cache_free(ceph_file_cachep, cf);
315 
316 	/* wake up anyone waiting for caps on this inode */
317 	wake_up_all(&ci->i_cap_wq);
318 	return 0;
319 }
320 
321 /*
322  * Read a range of bytes striped over one or more objects.  Iterate over
323  * objects we stripe over.  (That's not atomic, but good enough for now.)
324  *
325  * If we get a short result from the OSD, check against i_size; we need to
326  * only return a short read to the caller if we hit EOF.
327  */
328 static int striped_read(struct inode *inode,
329 			u64 off, u64 len,
330 			struct page **pages, int num_pages,
331 			int *checkeof, bool o_direct,
332 			unsigned long buf_align)
333 {
334 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
335 	struct ceph_inode_info *ci = ceph_inode(inode);
336 	u64 pos, this_len, left;
337 	int io_align, page_align;
338 	int pages_left;
339 	int read;
340 	struct page **page_pos;
341 	int ret;
342 	bool hit_stripe, was_short;
343 
344 	/*
345 	 * we may need to do multiple reads.  not atomic, unfortunately.
346 	 */
347 	pos = off;
348 	left = len;
349 	page_pos = pages;
350 	pages_left = num_pages;
351 	read = 0;
352 	io_align = off & ~PAGE_MASK;
353 
354 more:
355 	if (o_direct)
356 		page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
357 	else
358 		page_align = pos & ~PAGE_MASK;
359 	this_len = left;
360 	ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
361 				  &ci->i_layout, pos, &this_len,
362 				  ci->i_truncate_seq,
363 				  ci->i_truncate_size,
364 				  page_pos, pages_left, page_align);
365 	if (ret == -ENOENT)
366 		ret = 0;
367 	hit_stripe = this_len < left;
368 	was_short = ret >= 0 && ret < this_len;
369 	dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
370 	     ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
371 
372 	if (ret >= 0) {
373 		int didpages;
374 		if (was_short && (pos + ret < inode->i_size)) {
375 			u64 tmp = min(this_len - ret,
376 					inode->i_size - pos - ret);
377 			dout(" zero gap %llu to %llu\n",
378 				pos + ret, pos + ret + tmp);
379 			ceph_zero_page_vector_range(page_align + read + ret,
380 							tmp, pages);
381 			ret += tmp;
382 		}
383 
384 		didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
385 		pos += ret;
386 		read = pos - off;
387 		left -= ret;
388 		page_pos += didpages;
389 		pages_left -= didpages;
390 
391 		/* hit stripe and need continue*/
392 		if (left && hit_stripe && pos < inode->i_size)
393 			goto more;
394 	}
395 
396 	if (read > 0) {
397 		ret = read;
398 		/* did we bounce off eof? */
399 		if (pos + left > inode->i_size)
400 			*checkeof = 1;
401 	}
402 
403 	dout("striped_read returns %d\n", ret);
404 	return ret;
405 }
406 
407 /*
408  * Completely synchronous read and write methods.  Direct from __user
409  * buffer to osd, or directly to user pages (if O_DIRECT).
410  *
411  * If the read spans object boundary, just do multiple reads.
412  */
413 static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
414 				int *checkeof)
415 {
416 	struct file *file = iocb->ki_filp;
417 	struct inode *inode = file_inode(file);
418 	struct page **pages;
419 	u64 off = iocb->ki_pos;
420 	int num_pages, ret;
421 	size_t len = iov_iter_count(i);
422 
423 	dout("sync_read on file %p %llu~%u %s\n", file, off,
424 	     (unsigned)len,
425 	     (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
426 	/*
427 	 * flush any page cache pages in this range.  this
428 	 * will make concurrent normal and sync io slow,
429 	 * but it will at least behave sensibly when they are
430 	 * in sequence.
431 	 */
432 	ret = filemap_write_and_wait_range(inode->i_mapping, off,
433 						off + len);
434 	if (ret < 0)
435 		return ret;
436 
437 	if (file->f_flags & O_DIRECT) {
438 		while (iov_iter_count(i)) {
439 			size_t start;
440 			ssize_t n;
441 
442 			n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
443 			if (n < 0)
444 				return n;
445 
446 			num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
447 
448 			ret = striped_read(inode, off, n,
449 					   pages, num_pages, checkeof,
450 					   1, start);
451 
452 			ceph_put_page_vector(pages, num_pages, true);
453 
454 			if (ret <= 0)
455 				break;
456 			off += ret;
457 			iov_iter_advance(i, ret);
458 			if (ret < n)
459 				break;
460 		}
461 	} else {
462 		num_pages = calc_pages_for(off, len);
463 		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
464 		if (IS_ERR(pages))
465 			return PTR_ERR(pages);
466 		ret = striped_read(inode, off, len, pages,
467 					num_pages, checkeof, 0, 0);
468 		if (ret > 0) {
469 			int l, k = 0;
470 			size_t left = ret;
471 
472 			while (left) {
473 				int copy = min_t(size_t, PAGE_SIZE, left);
474 				l = copy_page_to_iter(pages[k++], 0, copy, i);
475 				off += l;
476 				left -= l;
477 				if (l < copy)
478 					break;
479 			}
480 		}
481 		ceph_release_page_vector(pages, num_pages);
482 	}
483 
484 	if (off > iocb->ki_pos) {
485 		ret = off - iocb->ki_pos;
486 		iocb->ki_pos = off;
487 	}
488 
489 	dout("sync_read result %d\n", ret);
490 	return ret;
491 }
492 
493 /*
494  * Write commit request unsafe callback, called to tell us when a
495  * request is unsafe (that is, in flight--has been handed to the
496  * messenger to send to its target osd).  It is called again when
497  * we've received a response message indicating the request is
498  * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
499  * is completed early (and unsuccessfully) due to a timeout or
500  * interrupt.
501  *
502  * This is used if we requested both an ACK and ONDISK commit reply
503  * from the OSD.
504  */
505 static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
506 {
507 	struct ceph_inode_info *ci = ceph_inode(req->r_inode);
508 
509 	dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
510 		unsafe ? "un" : "");
511 	if (unsafe) {
512 		ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
513 		spin_lock(&ci->i_unsafe_lock);
514 		list_add_tail(&req->r_unsafe_item,
515 			      &ci->i_unsafe_writes);
516 		spin_unlock(&ci->i_unsafe_lock);
517 	} else {
518 		spin_lock(&ci->i_unsafe_lock);
519 		list_del_init(&req->r_unsafe_item);
520 		spin_unlock(&ci->i_unsafe_lock);
521 		ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
522 	}
523 }
524 
525 
526 /*
527  * Synchronous write, straight from __user pointer or user pages.
528  *
529  * If write spans object boundary, just do multiple writes.  (For a
530  * correct atomic write, we should e.g. take write locks on all
531  * objects, rollback on failure, etc.)
532  */
533 static ssize_t
534 ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from)
535 {
536 	struct file *file = iocb->ki_filp;
537 	struct inode *inode = file_inode(file);
538 	struct ceph_inode_info *ci = ceph_inode(inode);
539 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
540 	struct ceph_snap_context *snapc;
541 	struct ceph_vino vino;
542 	struct ceph_osd_request *req;
543 	struct page **pages;
544 	int num_pages;
545 	int written = 0;
546 	int flags;
547 	int check_caps = 0;
548 	int ret;
549 	struct timespec mtime = CURRENT_TIME;
550 	loff_t pos = iocb->ki_pos;
551 	size_t count = iov_iter_count(from);
552 
553 	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
554 		return -EROFS;
555 
556 	dout("sync_direct_write on file %p %lld~%u\n", file, pos,
557 	     (unsigned)count);
558 
559 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
560 	if (ret < 0)
561 		return ret;
562 
563 	ret = invalidate_inode_pages2_range(inode->i_mapping,
564 					    pos >> PAGE_CACHE_SHIFT,
565 					    (pos + count) >> PAGE_CACHE_SHIFT);
566 	if (ret < 0)
567 		dout("invalidate_inode_pages2_range returned %d\n", ret);
568 
569 	flags = CEPH_OSD_FLAG_ORDERSNAP |
570 		CEPH_OSD_FLAG_ONDISK |
571 		CEPH_OSD_FLAG_WRITE;
572 
573 	while (iov_iter_count(from) > 0) {
574 		u64 len = iov_iter_single_seg_count(from);
575 		size_t start;
576 		ssize_t n;
577 
578 		snapc = ci->i_snap_realm->cached_context;
579 		vino = ceph_vino(inode);
580 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
581 					    vino, pos, &len,
582 					    2,/*include a 'startsync' command*/
583 					    CEPH_OSD_OP_WRITE, flags, snapc,
584 					    ci->i_truncate_seq,
585 					    ci->i_truncate_size,
586 					    false);
587 		if (IS_ERR(req)) {
588 			ret = PTR_ERR(req);
589 			break;
590 		}
591 
592 		n = iov_iter_get_pages_alloc(from, &pages, len, &start);
593 		if (unlikely(n < 0)) {
594 			ret = n;
595 			ceph_osdc_put_request(req);
596 			break;
597 		}
598 
599 		num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
600 		/*
601 		 * throw out any page cache pages in this range. this
602 		 * may block.
603 		 */
604 		truncate_inode_pages_range(inode->i_mapping, pos,
605 				   (pos+n) | (PAGE_CACHE_SIZE-1));
606 		osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
607 						false, false);
608 
609 		/* BUG_ON(vino.snap != CEPH_NOSNAP); */
610 		ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
611 
612 		ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
613 		if (!ret)
614 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
615 
616 		ceph_put_page_vector(pages, num_pages, false);
617 
618 		ceph_osdc_put_request(req);
619 		if (ret)
620 			break;
621 		pos += n;
622 		written += n;
623 		iov_iter_advance(from, n);
624 
625 		if (pos > i_size_read(inode)) {
626 			check_caps = ceph_inode_set_size(inode, pos);
627 			if (check_caps)
628 				ceph_check_caps(ceph_inode(inode),
629 						CHECK_CAPS_AUTHONLY,
630 						NULL);
631 		}
632 	}
633 
634 	if (ret != -EOLDSNAPC && written > 0) {
635 		iocb->ki_pos = pos;
636 		ret = written;
637 	}
638 	return ret;
639 }
640 
641 
642 /*
643  * Synchronous write, straight from __user pointer or user pages.
644  *
645  * If write spans object boundary, just do multiple writes.  (For a
646  * correct atomic write, we should e.g. take write locks on all
647  * objects, rollback on failure, etc.)
648  */
649 static ssize_t ceph_sync_write(struct kiocb *iocb, struct iov_iter *from)
650 {
651 	struct file *file = iocb->ki_filp;
652 	struct inode *inode = file_inode(file);
653 	struct ceph_inode_info *ci = ceph_inode(inode);
654 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
655 	struct ceph_snap_context *snapc;
656 	struct ceph_vino vino;
657 	struct ceph_osd_request *req;
658 	struct page **pages;
659 	u64 len;
660 	int num_pages;
661 	int written = 0;
662 	int flags;
663 	int check_caps = 0;
664 	int ret;
665 	struct timespec mtime = CURRENT_TIME;
666 	loff_t pos = iocb->ki_pos;
667 	size_t count = iov_iter_count(from);
668 
669 	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
670 		return -EROFS;
671 
672 	dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
673 
674 	ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
675 	if (ret < 0)
676 		return ret;
677 
678 	ret = invalidate_inode_pages2_range(inode->i_mapping,
679 					    pos >> PAGE_CACHE_SHIFT,
680 					    (pos + count) >> PAGE_CACHE_SHIFT);
681 	if (ret < 0)
682 		dout("invalidate_inode_pages2_range returned %d\n", ret);
683 
684 	flags = CEPH_OSD_FLAG_ORDERSNAP |
685 		CEPH_OSD_FLAG_ONDISK |
686 		CEPH_OSD_FLAG_WRITE |
687 		CEPH_OSD_FLAG_ACK;
688 
689 	while ((len = iov_iter_count(from)) > 0) {
690 		size_t left;
691 		int n;
692 
693 		snapc = ci->i_snap_realm->cached_context;
694 		vino = ceph_vino(inode);
695 		req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
696 					    vino, pos, &len, 1,
697 					    CEPH_OSD_OP_WRITE, flags, snapc,
698 					    ci->i_truncate_seq,
699 					    ci->i_truncate_size,
700 					    false);
701 		if (IS_ERR(req)) {
702 			ret = PTR_ERR(req);
703 			break;
704 		}
705 
706 		/*
707 		 * write from beginning of first page,
708 		 * regardless of io alignment
709 		 */
710 		num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
711 
712 		pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
713 		if (IS_ERR(pages)) {
714 			ret = PTR_ERR(pages);
715 			goto out;
716 		}
717 
718 		left = len;
719 		for (n = 0; n < num_pages; n++) {
720 			size_t plen = min_t(size_t, left, PAGE_SIZE);
721 			ret = copy_page_from_iter(pages[n], 0, plen, from);
722 			if (ret != plen) {
723 				ret = -EFAULT;
724 				break;
725 			}
726 			left -= ret;
727 		}
728 
729 		if (ret < 0) {
730 			ceph_release_page_vector(pages, num_pages);
731 			goto out;
732 		}
733 
734 		/* get a second commit callback */
735 		req->r_unsafe_callback = ceph_sync_write_unsafe;
736 		req->r_inode = inode;
737 
738 		osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
739 						false, true);
740 
741 		/* BUG_ON(vino.snap != CEPH_NOSNAP); */
742 		ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
743 
744 		ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
745 		if (!ret)
746 			ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
747 
748 out:
749 		ceph_osdc_put_request(req);
750 		if (ret == 0) {
751 			pos += len;
752 			written += len;
753 
754 			if (pos > i_size_read(inode)) {
755 				check_caps = ceph_inode_set_size(inode, pos);
756 				if (check_caps)
757 					ceph_check_caps(ceph_inode(inode),
758 							CHECK_CAPS_AUTHONLY,
759 							NULL);
760 			}
761 		} else
762 			break;
763 	}
764 
765 	if (ret != -EOLDSNAPC && written > 0) {
766 		ret = written;
767 		iocb->ki_pos = pos;
768 	}
769 	return ret;
770 }
771 
772 /*
773  * Wrap generic_file_aio_read with checks for cap bits on the inode.
774  * Atomically grab references, so that those bits are not released
775  * back to the MDS mid-read.
776  *
777  * Hmm, the sync read case isn't actually async... should it be?
778  */
779 static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
780 {
781 	struct file *filp = iocb->ki_filp;
782 	struct ceph_file_info *fi = filp->private_data;
783 	size_t len = iocb->ki_nbytes;
784 	struct inode *inode = file_inode(filp);
785 	struct ceph_inode_info *ci = ceph_inode(inode);
786 	ssize_t ret;
787 	int want, got = 0;
788 	int checkeof = 0, read = 0;
789 
790 again:
791 	dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
792 	     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
793 
794 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
795 		want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
796 	else
797 		want = CEPH_CAP_FILE_CACHE;
798 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
799 	if (ret < 0)
800 		return ret;
801 
802 	if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
803 	    (iocb->ki_filp->f_flags & O_DIRECT) ||
804 	    (fi->flags & CEPH_F_SYNC)) {
805 
806 		dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
807 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
808 		     ceph_cap_string(got));
809 
810 		/* hmm, this isn't really async... */
811 		ret = ceph_sync_read(iocb, to, &checkeof);
812 	} else {
813 		dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
814 		     inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
815 		     ceph_cap_string(got));
816 
817 		ret = generic_file_read_iter(iocb, to);
818 	}
819 	dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
820 	     inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
821 	ceph_put_cap_refs(ci, got);
822 
823 	if (checkeof && ret >= 0) {
824 		int statret = ceph_do_getattr(inode,
825 					      CEPH_STAT_CAP_SIZE);
826 
827 		/* hit EOF or hole? */
828 		if (statret == 0 && iocb->ki_pos < inode->i_size &&
829 			ret < len) {
830 			dout("sync_read hit hole, ppos %lld < size %lld"
831 			     ", reading more\n", iocb->ki_pos,
832 			     inode->i_size);
833 
834 			iov_iter_advance(to, ret);
835 			read += ret;
836 			len -= ret;
837 			checkeof = 0;
838 			goto again;
839 		}
840 	}
841 
842 	if (ret >= 0)
843 		ret += read;
844 
845 	return ret;
846 }
847 
848 /*
849  * Take cap references to avoid releasing caps to MDS mid-write.
850  *
851  * If we are synchronous, and write with an old snap context, the OSD
852  * may return EOLDSNAPC.  In that case, retry the write.. _after_
853  * dropping our cap refs and allowing the pending snap to logically
854  * complete _before_ this write occurs.
855  *
856  * If we are near ENOSPC, write synchronously.
857  */
858 static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
859 {
860 	struct file *file = iocb->ki_filp;
861 	struct ceph_file_info *fi = file->private_data;
862 	struct inode *inode = file_inode(file);
863 	struct ceph_inode_info *ci = ceph_inode(inode);
864 	struct ceph_osd_client *osdc =
865 		&ceph_sb_to_client(inode->i_sb)->client->osdc;
866 	ssize_t count = iov_iter_count(from), written = 0;
867 	int err, want, got;
868 	loff_t pos = iocb->ki_pos;
869 
870 	if (ceph_snap(inode) != CEPH_NOSNAP)
871 		return -EROFS;
872 
873 	mutex_lock(&inode->i_mutex);
874 
875 	/* We can write back this queue in page reclaim */
876 	current->backing_dev_info = file->f_mapping->backing_dev_info;
877 
878 	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
879 	if (err)
880 		goto out;
881 
882 	if (count == 0)
883 		goto out;
884 	iov_iter_truncate(from, count);
885 
886 	err = file_remove_suid(file);
887 	if (err)
888 		goto out;
889 
890 	err = file_update_time(file);
891 	if (err)
892 		goto out;
893 
894 retry_snap:
895 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
896 		err = -ENOSPC;
897 		goto out;
898 	}
899 
900 	dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
901 	     inode, ceph_vinop(inode), pos, count, inode->i_size);
902 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
903 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
904 	else
905 		want = CEPH_CAP_FILE_BUFFER;
906 	got = 0;
907 	err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count);
908 	if (err < 0)
909 		goto out;
910 
911 	dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
912 	     inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
913 
914 	if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
915 	    (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
916 		struct iov_iter data;
917 		mutex_unlock(&inode->i_mutex);
918 		/* we might need to revert back to that point */
919 		data = *from;
920 		if (file->f_flags & O_DIRECT)
921 			written = ceph_sync_direct_write(iocb, &data);
922 		else
923 			written = ceph_sync_write(iocb, &data);
924 		if (written == -EOLDSNAPC) {
925 			dout("aio_write %p %llx.%llx %llu~%u"
926 				"got EOLDSNAPC, retrying\n",
927 				inode, ceph_vinop(inode),
928 				pos, (unsigned)count);
929 			mutex_lock(&inode->i_mutex);
930 			goto retry_snap;
931 		}
932 		if (written > 0)
933 			iov_iter_advance(from, written);
934 	} else {
935 		loff_t old_size = inode->i_size;
936 		/*
937 		 * No need to acquire the i_truncate_mutex. Because
938 		 * the MDS revokes Fwb caps before sending truncate
939 		 * message to us. We can't get Fwb cap while there
940 		 * are pending vmtruncate. So write and vmtruncate
941 		 * can not run at the same time
942 		 */
943 		written = generic_perform_write(file, from, pos);
944 		if (likely(written >= 0))
945 			iocb->ki_pos = pos + written;
946 		if (inode->i_size > old_size)
947 			ceph_fscache_update_objectsize(inode);
948 		mutex_unlock(&inode->i_mutex);
949 	}
950 
951 	if (written >= 0) {
952 		int dirty;
953 		spin_lock(&ci->i_ceph_lock);
954 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
955 		spin_unlock(&ci->i_ceph_lock);
956 		if (dirty)
957 			__mark_inode_dirty(inode, dirty);
958 	}
959 
960 	dout("aio_write %p %llx.%llx %llu~%u  dropping cap refs on %s\n",
961 	     inode, ceph_vinop(inode), pos, (unsigned)count,
962 	     ceph_cap_string(got));
963 	ceph_put_cap_refs(ci, got);
964 
965 	if (written >= 0 &&
966 	    ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
967 	     ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
968 		err = vfs_fsync_range(file, pos, pos + written - 1, 1);
969 		if (err < 0)
970 			written = err;
971 	}
972 
973 	goto out_unlocked;
974 
975 out:
976 	mutex_unlock(&inode->i_mutex);
977 out_unlocked:
978 	current->backing_dev_info = NULL;
979 	return written ? written : err;
980 }
981 
982 /*
983  * llseek.  be sure to verify file size on SEEK_END.
984  */
985 static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
986 {
987 	struct inode *inode = file->f_mapping->host;
988 	int ret;
989 
990 	mutex_lock(&inode->i_mutex);
991 
992 	if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
993 		ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
994 		if (ret < 0) {
995 			offset = ret;
996 			goto out;
997 		}
998 	}
999 
1000 	switch (whence) {
1001 	case SEEK_END:
1002 		offset += inode->i_size;
1003 		break;
1004 	case SEEK_CUR:
1005 		/*
1006 		 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1007 		 * position-querying operation.  Avoid rewriting the "same"
1008 		 * f_pos value back to the file because a concurrent read(),
1009 		 * write() or lseek() might have altered it
1010 		 */
1011 		if (offset == 0) {
1012 			offset = file->f_pos;
1013 			goto out;
1014 		}
1015 		offset += file->f_pos;
1016 		break;
1017 	case SEEK_DATA:
1018 		if (offset >= inode->i_size) {
1019 			ret = -ENXIO;
1020 			goto out;
1021 		}
1022 		break;
1023 	case SEEK_HOLE:
1024 		if (offset >= inode->i_size) {
1025 			ret = -ENXIO;
1026 			goto out;
1027 		}
1028 		offset = inode->i_size;
1029 		break;
1030 	}
1031 
1032 	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1033 
1034 out:
1035 	mutex_unlock(&inode->i_mutex);
1036 	return offset;
1037 }
1038 
1039 static inline void ceph_zero_partial_page(
1040 	struct inode *inode, loff_t offset, unsigned size)
1041 {
1042 	struct page *page;
1043 	pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1044 
1045 	page = find_lock_page(inode->i_mapping, index);
1046 	if (page) {
1047 		wait_on_page_writeback(page);
1048 		zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1049 		unlock_page(page);
1050 		page_cache_release(page);
1051 	}
1052 }
1053 
1054 static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1055 				      loff_t length)
1056 {
1057 	loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1058 	if (offset < nearly) {
1059 		loff_t size = nearly - offset;
1060 		if (length < size)
1061 			size = length;
1062 		ceph_zero_partial_page(inode, offset, size);
1063 		offset += size;
1064 		length -= size;
1065 	}
1066 	if (length >= PAGE_CACHE_SIZE) {
1067 		loff_t size = round_down(length, PAGE_CACHE_SIZE);
1068 		truncate_pagecache_range(inode, offset, offset + size - 1);
1069 		offset += size;
1070 		length -= size;
1071 	}
1072 	if (length)
1073 		ceph_zero_partial_page(inode, offset, length);
1074 }
1075 
1076 static int ceph_zero_partial_object(struct inode *inode,
1077 				    loff_t offset, loff_t *length)
1078 {
1079 	struct ceph_inode_info *ci = ceph_inode(inode);
1080 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1081 	struct ceph_osd_request *req;
1082 	int ret = 0;
1083 	loff_t zero = 0;
1084 	int op;
1085 
1086 	if (!length) {
1087 		op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1088 		length = &zero;
1089 	} else {
1090 		op = CEPH_OSD_OP_ZERO;
1091 	}
1092 
1093 	req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1094 					ceph_vino(inode),
1095 					offset, length,
1096 					1, op,
1097 					CEPH_OSD_FLAG_WRITE |
1098 					CEPH_OSD_FLAG_ONDISK,
1099 					NULL, 0, 0, false);
1100 	if (IS_ERR(req)) {
1101 		ret = PTR_ERR(req);
1102 		goto out;
1103 	}
1104 
1105 	ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1106 				&inode->i_mtime);
1107 
1108 	ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1109 	if (!ret) {
1110 		ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1111 		if (ret == -ENOENT)
1112 			ret = 0;
1113 	}
1114 	ceph_osdc_put_request(req);
1115 
1116 out:
1117 	return ret;
1118 }
1119 
1120 static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1121 {
1122 	int ret = 0;
1123 	struct ceph_inode_info *ci = ceph_inode(inode);
1124 	s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1125 	s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1126 	s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1127 	u64 object_set_size = object_size * stripe_count;
1128 	u64 nearly, t;
1129 
1130 	/* round offset up to next period boundary */
1131 	nearly = offset + object_set_size - 1;
1132 	t = nearly;
1133 	nearly -= do_div(t, object_set_size);
1134 
1135 	while (length && offset < nearly) {
1136 		loff_t size = length;
1137 		ret = ceph_zero_partial_object(inode, offset, &size);
1138 		if (ret < 0)
1139 			return ret;
1140 		offset += size;
1141 		length -= size;
1142 	}
1143 	while (length >= object_set_size) {
1144 		int i;
1145 		loff_t pos = offset;
1146 		for (i = 0; i < stripe_count; ++i) {
1147 			ret = ceph_zero_partial_object(inode, pos, NULL);
1148 			if (ret < 0)
1149 				return ret;
1150 			pos += stripe_unit;
1151 		}
1152 		offset += object_set_size;
1153 		length -= object_set_size;
1154 	}
1155 	while (length) {
1156 		loff_t size = length;
1157 		ret = ceph_zero_partial_object(inode, offset, &size);
1158 		if (ret < 0)
1159 			return ret;
1160 		offset += size;
1161 		length -= size;
1162 	}
1163 	return ret;
1164 }
1165 
1166 static long ceph_fallocate(struct file *file, int mode,
1167 				loff_t offset, loff_t length)
1168 {
1169 	struct ceph_file_info *fi = file->private_data;
1170 	struct inode *inode = file_inode(file);
1171 	struct ceph_inode_info *ci = ceph_inode(inode);
1172 	struct ceph_osd_client *osdc =
1173 		&ceph_inode_to_client(inode)->client->osdc;
1174 	int want, got = 0;
1175 	int dirty;
1176 	int ret = 0;
1177 	loff_t endoff = 0;
1178 	loff_t size;
1179 
1180 	if (!S_ISREG(inode->i_mode))
1181 		return -EOPNOTSUPP;
1182 
1183 	mutex_lock(&inode->i_mutex);
1184 
1185 	if (ceph_snap(inode) != CEPH_NOSNAP) {
1186 		ret = -EROFS;
1187 		goto unlock;
1188 	}
1189 
1190 	if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1191 		!(mode & FALLOC_FL_PUNCH_HOLE)) {
1192 		ret = -ENOSPC;
1193 		goto unlock;
1194 	}
1195 
1196 	size = i_size_read(inode);
1197 	if (!(mode & FALLOC_FL_KEEP_SIZE))
1198 		endoff = offset + length;
1199 
1200 	if (fi->fmode & CEPH_FILE_MODE_LAZY)
1201 		want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1202 	else
1203 		want = CEPH_CAP_FILE_BUFFER;
1204 
1205 	ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
1206 	if (ret < 0)
1207 		goto unlock;
1208 
1209 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1210 		if (offset < size)
1211 			ceph_zero_pagecache_range(inode, offset, length);
1212 		ret = ceph_zero_objects(inode, offset, length);
1213 	} else if (endoff > size) {
1214 		truncate_pagecache_range(inode, size, -1);
1215 		if (ceph_inode_set_size(inode, endoff))
1216 			ceph_check_caps(ceph_inode(inode),
1217 				CHECK_CAPS_AUTHONLY, NULL);
1218 	}
1219 
1220 	if (!ret) {
1221 		spin_lock(&ci->i_ceph_lock);
1222 		dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1223 		spin_unlock(&ci->i_ceph_lock);
1224 		if (dirty)
1225 			__mark_inode_dirty(inode, dirty);
1226 	}
1227 
1228 	ceph_put_cap_refs(ci, got);
1229 unlock:
1230 	mutex_unlock(&inode->i_mutex);
1231 	return ret;
1232 }
1233 
1234 const struct file_operations ceph_file_fops = {
1235 	.open = ceph_open,
1236 	.release = ceph_release,
1237 	.llseek = ceph_llseek,
1238 	.read = new_sync_read,
1239 	.write = new_sync_write,
1240 	.read_iter = ceph_read_iter,
1241 	.write_iter = ceph_write_iter,
1242 	.mmap = ceph_mmap,
1243 	.fsync = ceph_fsync,
1244 	.lock = ceph_lock,
1245 	.flock = ceph_flock,
1246 	.splice_read = generic_file_splice_read,
1247 	.splice_write = iter_file_splice_write,
1248 	.unlocked_ioctl = ceph_ioctl,
1249 	.compat_ioctl	= ceph_ioctl,
1250 	.fallocate	= ceph_fallocate,
1251 };
1252 
1253