blob: 7d0e4a82d898a83f2695eead6ffc767689ffe140 [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weil124e68e2009-10-06 11:31:08 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weil124e68e2009-10-06 11:31:08 -07004#include <linux/sched.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weil124e68e2009-10-06 11:31:08 -07006#include <linux/file.h>
7#include <linux/namei.h>
8#include <linux/writeback.h>
9
10#include "super.h"
11#include "mds_client.h"
12
13/*
14 * Ceph file operations
15 *
16 * Implement basic open/close functionality, and implement
17 * read/write.
18 *
19 * We implement three modes of file I/O:
20 * - buffered uses the generic_file_aio_{read,write} helpers
21 *
22 * - synchronous is used when there is multi-client read/write
23 * sharing, avoids the page cache, and synchronously waits for an
24 * ack from the OSD.
25 *
26 * - direct io takes the variant of the sync path that references
27 * user pages directly.
28 *
29 * fsync() flushes and waits on dirty pages, but just queues metadata
30 * for writeback: since the MDS can recover size and mtime there is no
31 * need to wait for MDS acknowledgement.
32 */
33
34
35/*
36 * Prepare an open request. Preallocate ceph_cap to avoid an
37 * inopportune ENOMEM later.
38 */
39static struct ceph_mds_request *
40prepare_open_request(struct super_block *sb, int flags, int create_mode)
41{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070042 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
43 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070044 struct ceph_mds_request *req;
45 int want_auth = USE_ANY_MDS;
46 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
47
48 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
49 want_auth = USE_AUTH_MDS;
50
51 req = ceph_mdsc_create_request(mdsc, op, want_auth);
52 if (IS_ERR(req))
53 goto out;
54 req->r_fmode = ceph_flags_to_mode(flags);
55 req->r_args.open.flags = cpu_to_le32(flags);
56 req->r_args.open.mode = cpu_to_le32(create_mode);
Sage Weil6a18be12009-11-04 11:40:05 -080057 req->r_args.open.preferred = cpu_to_le32(-1);
Sage Weil124e68e2009-10-06 11:31:08 -070058out:
59 return req;
60}
61
62/*
63 * initialize private struct file data.
64 * if we fail, clean up by dropping fmode reference on the ceph_inode
65 */
66static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
67{
68 struct ceph_file_info *cf;
69 int ret = 0;
70
71 switch (inode->i_mode & S_IFMT) {
72 case S_IFREG:
73 case S_IFDIR:
74 dout("init_file %p %p 0%o (regular)\n", inode, file,
75 inode->i_mode);
76 cf = kmem_cache_alloc(ceph_file_cachep, GFP_NOFS | __GFP_ZERO);
77 if (cf == NULL) {
78 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
79 return -ENOMEM;
80 }
81 cf->fmode = fmode;
82 cf->next_offset = 2;
83 file->private_data = cf;
84 BUG_ON(inode->i_fop->release != ceph_release);
85 break;
86
87 case S_IFLNK:
88 dout("init_file %p %p 0%o (symlink)\n", inode, file,
89 inode->i_mode);
90 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
91 break;
92
93 default:
94 dout("init_file %p %p 0%o (special)\n", inode, file,
95 inode->i_mode);
96 /*
97 * we need to drop the open ref now, since we don't
98 * have .release set to ceph_release.
99 */
100 ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */
101 BUG_ON(inode->i_fop->release == ceph_release);
102
103 /* call the proper open fop */
104 ret = inode->i_fop->open(inode, file);
105 }
106 return ret;
107}
108
109/*
110 * If the filp already has private_data, that means the file was
111 * already opened by intent during lookup, and we do nothing.
112 *
113 * If we already have the requisite capabilities, we can satisfy
114 * the open request locally (no need to request new caps from the
115 * MDS). We do, however, need to inform the MDS (asynchronously)
116 * if our wanted caps set expands.
117 */
118int ceph_open(struct inode *inode, struct file *file)
119{
120 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700121 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
122 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700123 struct ceph_mds_request *req;
124 struct ceph_file_info *cf = file->private_data;
125 struct inode *parent_inode = file->f_dentry->d_parent->d_inode;
126 int err;
127 int flags, fmode, wanted;
128
129 if (cf) {
130 dout("open file %p is already opened\n", file);
131 return 0;
132 }
133
134 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
135 flags = file->f_flags & ~(O_CREAT|O_EXCL);
136 if (S_ISDIR(inode->i_mode))
137 flags = O_DIRECTORY; /* mds likes to know */
138
139 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
140 ceph_vinop(inode), file, flags, file->f_flags);
141 fmode = ceph_flags_to_mode(flags);
142 wanted = ceph_caps_for_mode(fmode);
143
144 /* snapped files are read-only */
145 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
146 return -EROFS;
147
148 /* trivially open snapdir */
149 if (ceph_snap(inode) == CEPH_SNAPDIR) {
150 spin_lock(&inode->i_lock);
151 __ceph_get_fmode(ci, fmode);
152 spin_unlock(&inode->i_lock);
153 return ceph_init_file(inode, file, fmode);
154 }
155
156 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800157 * No need to block if we have caps on the auth MDS (for
158 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700159 * asynchronously.
160 */
161 spin_lock(&inode->i_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800162 if (__ceph_is_any_real_caps(ci) &&
163 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700164 int mds_wanted = __ceph_caps_mds_wanted(ci);
165 int issued = __ceph_caps_issued(ci, NULL);
166
167 dout("open %p fmode %d want %s issued %s using existing\n",
168 inode, fmode, ceph_cap_string(wanted),
169 ceph_cap_string(issued));
170 __ceph_get_fmode(ci, fmode);
171 spin_unlock(&inode->i_lock);
172
173 /* adjust wanted? */
174 if ((issued & wanted) != wanted &&
175 (mds_wanted & wanted) != wanted &&
176 ceph_snap(inode) != CEPH_SNAPDIR)
177 ceph_check_caps(ci, 0, NULL);
178
179 return ceph_init_file(inode, file, fmode);
180 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
181 (ci->i_snap_caps & wanted) == wanted) {
182 __ceph_get_fmode(ci, fmode);
183 spin_unlock(&inode->i_lock);
184 return ceph_init_file(inode, file, fmode);
185 }
186 spin_unlock(&inode->i_lock);
187
188 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
189 req = prepare_open_request(inode->i_sb, flags, 0);
190 if (IS_ERR(req)) {
191 err = PTR_ERR(req);
192 goto out;
193 }
194 req->r_inode = igrab(inode);
195 req->r_num_caps = 1;
196 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
197 if (!err)
198 err = ceph_init_file(inode, file, req->r_fmode);
199 ceph_mdsc_put_request(req);
200 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
201out:
202 return err;
203}
204
205
206/*
207 * Do a lookup + open with a single request.
208 *
209 * If this succeeds, but some subsequent check in the vfs
210 * may_open() fails, the struct *file gets cleaned up (i.e.
211 * ceph_release gets called). So fear not!
212 */
213/*
214 * flags
215 * path_lookup_open -> LOOKUP_OPEN
216 * path_lookup_create -> LOOKUP_OPEN|LOOKUP_CREATE
217 */
218struct dentry *ceph_lookup_open(struct inode *dir, struct dentry *dentry,
219 struct nameidata *nd, int mode,
220 int locked_dir)
221{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700222 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
223 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700224 struct file *file = nd->intent.open.file;
225 struct inode *parent_inode = get_dentry_parent_inode(file->f_dentry);
226 struct ceph_mds_request *req;
227 int err;
228 int flags = nd->intent.open.flags - 1; /* silly vfs! */
229
230 dout("ceph_lookup_open dentry %p '%.*s' flags %d mode 0%o\n",
231 dentry, dentry->d_name.len, dentry->d_name.name, flags, mode);
232
233 /* do the open */
234 req = prepare_open_request(dir->i_sb, flags, mode);
235 if (IS_ERR(req))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200236 return ERR_CAST(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700237 req->r_dentry = dget(dentry);
238 req->r_num_caps = 2;
239 if (flags & O_CREAT) {
240 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
241 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
242 }
243 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
244 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
245 dentry = ceph_finish_lookup(req, dentry, err);
246 if (!err && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
247 err = ceph_handle_notrace_create(dir, dentry);
248 if (!err)
249 err = ceph_init_file(req->r_dentry->d_inode, file,
250 req->r_fmode);
251 ceph_mdsc_put_request(req);
252 dout("ceph_lookup_open result=%p\n", dentry);
253 return dentry;
254}
255
256int ceph_release(struct inode *inode, struct file *file)
257{
258 struct ceph_inode_info *ci = ceph_inode(inode);
259 struct ceph_file_info *cf = file->private_data;
260
261 dout("release inode %p file %p\n", inode, file);
262 ceph_put_fmode(ci, cf->fmode);
263 if (cf->last_readdir)
264 ceph_mdsc_put_request(cf->last_readdir);
265 kfree(cf->last_name);
266 kfree(cf->dir_info);
267 dput(cf->dentry);
268 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800269
270 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700271 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700272 return 0;
273}
274
275/*
Sage Weil124e68e2009-10-06 11:31:08 -0700276 * Read a range of bytes striped over one or more objects. Iterate over
277 * objects we stripe over. (That's not atomic, but good enough for now.)
278 *
279 * If we get a short result from the OSD, check against i_size; we need to
280 * only return a short read to the caller if we hit EOF.
281 */
282static int striped_read(struct inode *inode,
283 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800284 struct page **pages, int num_pages,
Henry C Changab226e22010-12-15 20:41:54 -0800285 int *checkeof, bool align_to_pages,
286 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700287{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700288 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700289 struct ceph_inode_info *ci = ceph_inode(inode);
290 u64 pos, this_len;
Sage Weilb7495fc2010-11-09 12:43:12 -0800291 int io_align, page_align;
Yehuda Sadeh972f0d32010-02-04 13:41:41 -0800292 int page_off = off & ~PAGE_CACHE_MASK; /* first byte's offset in page */
Sage Weil124e68e2009-10-06 11:31:08 -0700293 int left, pages_left;
294 int read;
295 struct page **page_pos;
296 int ret;
297 bool hit_stripe, was_short;
298
299 /*
300 * we may need to do multiple reads. not atomic, unfortunately.
301 */
302 pos = off;
303 left = len;
304 page_pos = pages;
305 pages_left = num_pages;
306 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800307 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700308
309more:
Sage Weilb7495fc2010-11-09 12:43:12 -0800310 if (align_to_pages)
Henry C Changab226e22010-12-15 20:41:54 -0800311 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800312 else
313 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700314 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700315 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700316 &ci->i_layout, pos, &this_len,
317 ci->i_truncate_seq,
318 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800319 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700320 hit_stripe = this_len < left;
321 was_short = ret >= 0 && ret < this_len;
322 if (ret == -ENOENT)
323 ret = 0;
324 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
325 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
326
327 if (ret > 0) {
328 int didpages =
329 ((pos & ~PAGE_CACHE_MASK) + ret) >> PAGE_CACHE_SHIFT;
330
331 if (read < pos - off) {
332 dout(" zero gap %llu to %llu\n", off + read, pos);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700333 ceph_zero_page_vector_range(page_off + read,
334 pos - off - read, pages);
Sage Weil124e68e2009-10-06 11:31:08 -0700335 }
336 pos += ret;
337 read = pos - off;
338 left -= ret;
339 page_pos += didpages;
340 pages_left -= didpages;
341
342 /* hit stripe? */
343 if (left && hit_stripe)
344 goto more;
345 }
346
347 if (was_short) {
348 /* was original extent fully inside i_size? */
349 if (pos + left <= inode->i_size) {
350 dout("zero tail\n");
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700351 ceph_zero_page_vector_range(page_off + read, len - read,
352 pages);
Yehuda Sadeh972f0d32010-02-04 13:41:41 -0800353 read = len;
Sage Weil124e68e2009-10-06 11:31:08 -0700354 goto out;
355 }
356
357 /* check i_size */
Sage Weil6a026582010-02-09 14:04:02 -0800358 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700359 }
360
361out:
362 if (ret >= 0)
363 ret = read;
364 dout("striped_read returns %d\n", ret);
365 return ret;
366}
367
368/*
369 * Completely synchronous read and write methods. Direct from __user
370 * buffer to osd, or directly to user pages (if O_DIRECT).
371 *
372 * If the read spans object boundary, just do multiple reads.
373 */
374static ssize_t ceph_sync_read(struct file *file, char __user *data,
Sage Weil6a026582010-02-09 14:04:02 -0800375 unsigned len, loff_t *poff, int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700376{
377 struct inode *inode = file->f_dentry->d_inode;
378 struct page **pages;
379 u64 off = *poff;
Henry C Changab226e22010-12-15 20:41:54 -0800380 int num_pages, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700381
382 dout("sync_read on file %p %llu~%u %s\n", file, off, len,
383 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
384
Henry C Changab226e22010-12-15 20:41:54 -0800385 if (file->f_flags & O_DIRECT) {
386 num_pages = calc_pages_for((unsigned long)data, len);
Henry C Changb6aa5902010-12-15 20:45:41 -0800387 pages = ceph_get_direct_page_vector(data, num_pages, true);
Henry C Changab226e22010-12-15 20:41:54 -0800388 } else {
389 num_pages = calc_pages_for(off, len);
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700390 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Henry C Changab226e22010-12-15 20:41:54 -0800391 }
Sage Weil124e68e2009-10-06 11:31:08 -0700392 if (IS_ERR(pages))
393 return PTR_ERR(pages);
394
Sage Weile98b6fe2010-11-09 12:24:53 -0800395 /*
396 * flush any page cache pages in this range. this
397 * will make concurrent normal and sync io slow,
398 * but it will at least behave sensibly when they are
399 * in sequence.
400 */
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800401 ret = filemap_write_and_wait(inode->i_mapping);
402 if (ret < 0)
403 goto done;
404
Sage Weilb7495fc2010-11-09 12:43:12 -0800405 ret = striped_read(inode, off, len, pages, num_pages, checkeof,
Henry C Changab226e22010-12-15 20:41:54 -0800406 file->f_flags & O_DIRECT,
407 (unsigned long)data & ~PAGE_MASK);
Sage Weil124e68e2009-10-06 11:31:08 -0700408
409 if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700410 ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
Sage Weil124e68e2009-10-06 11:31:08 -0700411 if (ret >= 0)
412 *poff = off + ret;
413
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800414done:
Sage Weil124e68e2009-10-06 11:31:08 -0700415 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800416 ceph_put_page_vector(pages, num_pages, true);
Sage Weil124e68e2009-10-06 11:31:08 -0700417 else
418 ceph_release_page_vector(pages, num_pages);
419 dout("sync_read result %d\n", ret);
420 return ret;
421}
422
423/*
424 * Write commit callback, called if we requested both an ACK and
425 * ONDISK commit reply from the OSD.
426 */
427static void sync_write_commit(struct ceph_osd_request *req,
428 struct ceph_msg *msg)
429{
430 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
431
432 dout("sync_write_commit %p tid %llu\n", req, req->r_tid);
433 spin_lock(&ci->i_unsafe_lock);
434 list_del_init(&req->r_unsafe_item);
435 spin_unlock(&ci->i_unsafe_lock);
436 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
437}
438
439/*
440 * Synchronous write, straight from __user pointer or user pages (if
441 * O_DIRECT).
442 *
443 * If write spans object boundary, just do multiple writes. (For a
444 * correct atomic write, we should e.g. take write locks on all
445 * objects, rollback on failure, etc.)
446 */
447static ssize_t ceph_sync_write(struct file *file, const char __user *data,
448 size_t left, loff_t *offset)
449{
450 struct inode *inode = file->f_dentry->d_inode;
451 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700452 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700453 struct ceph_osd_request *req;
454 struct page **pages;
455 int num_pages;
456 long long unsigned pos;
457 u64 len;
458 int written = 0;
459 int flags;
460 int do_sync = 0;
461 int check_caps = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800462 int page_align, io_align;
Henry C Changab226e22010-12-15 20:41:54 -0800463 unsigned long buf_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700464 int ret;
465 struct timespec mtime = CURRENT_TIME;
466
467 if (ceph_snap(file->f_dentry->d_inode) != CEPH_NOSNAP)
468 return -EROFS;
469
470 dout("sync_write on file %p %lld~%u %s\n", file, *offset,
471 (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
472
473 if (file->f_flags & O_APPEND)
474 pos = i_size_read(inode);
475 else
476 pos = *offset;
477
Sage Weilb7495fc2010-11-09 12:43:12 -0800478 io_align = pos & ~PAGE_MASK;
Henry C Changab226e22010-12-15 20:41:54 -0800479 buf_align = (unsigned long)data & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800480
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800481 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left);
482 if (ret < 0)
483 return ret;
484
485 ret = invalidate_inode_pages2_range(inode->i_mapping,
486 pos >> PAGE_CACHE_SHIFT,
487 (pos + left) >> PAGE_CACHE_SHIFT);
488 if (ret < 0)
489 dout("invalidate_inode_pages2_range returned %d\n", ret);
490
Sage Weil124e68e2009-10-06 11:31:08 -0700491 flags = CEPH_OSD_FLAG_ORDERSNAP |
492 CEPH_OSD_FLAG_ONDISK |
493 CEPH_OSD_FLAG_WRITE;
494 if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
495 flags |= CEPH_OSD_FLAG_ACK;
496 else
497 do_sync = 1;
498
499 /*
500 * we may need to do multiple writes here if we span an object
501 * boundary. this isn't atomic, unfortunately. :(
502 */
503more:
504 len = left;
Henry C Changab226e22010-12-15 20:41:54 -0800505 if (file->f_flags & O_DIRECT) {
Sage Weilb7495fc2010-11-09 12:43:12 -0800506 /* write from beginning of first page, regardless of
507 io alignment */
Henry C Changab226e22010-12-15 20:41:54 -0800508 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
509 num_pages = calc_pages_for((unsigned long)data, len);
510 } else {
Sage Weilb7495fc2010-11-09 12:43:12 -0800511 page_align = pos & ~PAGE_MASK;
Henry C Changab226e22010-12-15 20:41:54 -0800512 num_pages = calc_pages_for(pos, len);
513 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700514 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Sage Weil124e68e2009-10-06 11:31:08 -0700515 ceph_vino(inode), pos, &len,
516 CEPH_OSD_OP_WRITE, flags,
517 ci->i_snap_realm->cached_context,
518 do_sync,
519 ci->i_truncate_seq, ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800520 &mtime, false, 2, page_align);
Sage Weila79832f2010-04-01 16:06:19 -0700521 if (!req)
522 return -ENOMEM;
Sage Weil124e68e2009-10-06 11:31:08 -0700523
Sage Weil124e68e2009-10-06 11:31:08 -0700524 if (file->f_flags & O_DIRECT) {
Henry C Changb6aa5902010-12-15 20:45:41 -0800525 pages = ceph_get_direct_page_vector(data, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700526 if (IS_ERR(pages)) {
527 ret = PTR_ERR(pages);
528 goto out;
529 }
530
531 /*
532 * throw out any page cache pages in this range. this
533 * may block.
534 */
Sage Weil213c99e2010-08-03 10:25:11 -0700535 truncate_inode_pages_range(inode->i_mapping, pos,
Sage Weil5c6a2cd2010-04-22 13:48:59 -0700536 (pos+len) | (PAGE_CACHE_SIZE-1));
Sage Weil124e68e2009-10-06 11:31:08 -0700537 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700538 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700539 if (IS_ERR(pages)) {
540 ret = PTR_ERR(pages);
541 goto out;
542 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700543 ret = ceph_copy_user_to_page_vector(pages, data, pos, len);
Sage Weil124e68e2009-10-06 11:31:08 -0700544 if (ret < 0) {
545 ceph_release_page_vector(pages, num_pages);
546 goto out;
547 }
548
549 if ((file->f_flags & O_SYNC) == 0) {
550 /* get a second commit callback */
551 req->r_safe_callback = sync_write_commit;
552 req->r_own_pages = 1;
553 }
554 }
555 req->r_pages = pages;
556 req->r_num_pages = num_pages;
557 req->r_inode = inode;
558
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700559 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700560 if (!ret) {
561 if (req->r_safe_callback) {
562 /*
563 * Add to inode unsafe list only after we
564 * start_request so that a tid has been assigned.
565 */
566 spin_lock(&ci->i_unsafe_lock);
Henry C Chang936aeb52010-09-22 20:21:17 -0700567 list_add(&req->r_unsafe_item, &ci->i_unsafe_writes);
Sage Weil124e68e2009-10-06 11:31:08 -0700568 spin_unlock(&ci->i_unsafe_lock);
569 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
570 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700571 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700572 }
573
574 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800575 ceph_put_page_vector(pages, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700576 else if (file->f_flags & O_SYNC)
577 ceph_release_page_vector(pages, num_pages);
578
579out:
580 ceph_osdc_put_request(req);
581 if (ret == 0) {
582 pos += len;
583 written += len;
584 left -= len;
585 if (left)
586 goto more;
587
588 ret = written;
589 *offset = pos;
590 if (pos > i_size_read(inode))
591 check_caps = ceph_inode_set_size(inode, pos);
592 if (check_caps)
593 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
594 NULL);
595 }
596 return ret;
597}
598
599/*
600 * Wrap generic_file_aio_read with checks for cap bits on the inode.
601 * Atomically grab references, so that those bits are not released
602 * back to the MDS mid-read.
603 *
604 * Hmm, the sync read case isn't actually async... should it be?
605 */
606static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
607 unsigned long nr_segs, loff_t pos)
608{
609 struct file *filp = iocb->ki_filp;
Sage Weil2962507ca2010-05-27 10:40:43 -0700610 struct ceph_file_info *fi = filp->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700611 loff_t *ppos = &iocb->ki_pos;
612 size_t len = iov->iov_len;
613 struct inode *inode = filp->f_dentry->d_inode;
614 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700615 void __user *base = iov->iov_base;
Sage Weil124e68e2009-10-06 11:31:08 -0700616 ssize_t ret;
Sage Weil2962507ca2010-05-27 10:40:43 -0700617 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800618 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700619
620 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
621 inode, ceph_vinop(inode), pos, (unsigned)len, inode);
Sage Weil6a026582010-02-09 14:04:02 -0800622again:
Sage Weil124e68e2009-10-06 11:31:08 -0700623 __ceph_do_pending_vmtruncate(inode);
Sage Weil2962507ca2010-05-27 10:40:43 -0700624 if (fi->fmode & CEPH_FILE_MODE_LAZY)
625 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
626 else
627 want = CEPH_CAP_FILE_CACHE;
628 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700629 if (ret < 0)
630 goto out;
631 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
632 inode, ceph_vinop(inode), pos, (unsigned)len,
633 ceph_cap_string(got));
634
Sage Weil2962507ca2010-05-27 10:40:43 -0700635 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700636 (iocb->ki_filp->f_flags & O_DIRECT) ||
637 (inode->i_sb->s_flags & MS_SYNCHRONOUS))
638 /* hmm, this isn't really async... */
Sage Weil6a026582010-02-09 14:04:02 -0800639 ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
Sage Weil124e68e2009-10-06 11:31:08 -0700640 else
641 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
642
643out:
644 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
645 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
646 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800647
648 if (checkeof && ret >= 0) {
649 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
650
651 /* hit EOF or hole? */
652 if (statret == 0 && *ppos < inode->i_size) {
653 dout("aio_read sync_read hit hole, reading more\n");
654 read += ret;
655 base += ret;
656 len -= ret;
657 checkeof = 0;
658 goto again;
659 }
660 }
661 if (ret >= 0)
662 ret += read;
663
Sage Weil124e68e2009-10-06 11:31:08 -0700664 return ret;
665}
666
667/*
668 * Take cap references to avoid releasing caps to MDS mid-write.
669 *
670 * If we are synchronous, and write with an old snap context, the OSD
671 * may return EOLDSNAPC. In that case, retry the write.. _after_
672 * dropping our cap refs and allowing the pending snap to logically
673 * complete _before_ this write occurs.
674 *
675 * If we are near ENOSPC, write synchronously.
676 */
677static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
678 unsigned long nr_segs, loff_t pos)
679{
680 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700681 struct ceph_file_info *fi = file->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700682 struct inode *inode = file->f_dentry->d_inode;
683 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700684 struct ceph_osd_client *osdc =
685 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Sage Weil124e68e2009-10-06 11:31:08 -0700686 loff_t endoff = pos + iov->iov_len;
Sage Weil33caad32010-05-26 14:31:27 -0700687 int want, got = 0;
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000688 int ret, err;
Sage Weil124e68e2009-10-06 11:31:08 -0700689
690 if (ceph_snap(inode) != CEPH_NOSNAP)
691 return -EROFS;
692
693retry_snap:
694 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
695 return -ENOSPC;
696 __ceph_do_pending_vmtruncate(inode);
697 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
698 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
699 inode->i_size);
Sage Weil33caad32010-05-26 14:31:27 -0700700 if (fi->fmode & CEPH_FILE_MODE_LAZY)
701 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
702 else
703 want = CEPH_CAP_FILE_BUFFER;
704 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
Sage Weil124e68e2009-10-06 11:31:08 -0700705 if (ret < 0)
706 goto out;
707
708 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
709 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
710 ceph_cap_string(got));
711
Sage Weil33caad32010-05-26 14:31:27 -0700712 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700713 (iocb->ki_filp->f_flags & O_DIRECT) ||
714 (inode->i_sb->s_flags & MS_SYNCHRONOUS)) {
715 ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
716 &iocb->ki_pos);
717 } else {
718 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
719
720 if ((ret >= 0 || ret == -EIOCBQUEUED) &&
721 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host)
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000722 || ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100723 err = vfs_fsync_range(file, pos, pos + ret - 1, 1);
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000724 if (err < 0)
725 ret = err;
726 }
Sage Weil124e68e2009-10-06 11:31:08 -0700727 }
728 if (ret >= 0) {
729 spin_lock(&inode->i_lock);
730 __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
731 spin_unlock(&inode->i_lock);
732 }
733
734out:
735 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
736 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
737 ceph_cap_string(got));
738 ceph_put_cap_refs(ci, got);
739
740 if (ret == -EOLDSNAPC) {
741 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
742 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
743 goto retry_snap;
744 }
745
746 return ret;
747}
748
749/*
750 * llseek. be sure to verify file size on SEEK_END.
751 */
752static loff_t ceph_llseek(struct file *file, loff_t offset, int origin)
753{
754 struct inode *inode = file->f_mapping->host;
755 int ret;
756
757 mutex_lock(&inode->i_mutex);
758 __ceph_do_pending_vmtruncate(inode);
759 switch (origin) {
760 case SEEK_END:
761 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
762 if (ret < 0) {
763 offset = ret;
764 goto out;
765 }
766 offset += inode->i_size;
767 break;
768 case SEEK_CUR:
769 /*
770 * Here we special-case the lseek(fd, 0, SEEK_CUR)
771 * position-querying operation. Avoid rewriting the "same"
772 * f_pos value back to the file because a concurrent read(),
773 * write() or lseek() might have altered it
774 */
775 if (offset == 0) {
776 offset = file->f_pos;
777 goto out;
778 }
779 offset += file->f_pos;
780 break;
781 }
782
783 if (offset < 0 || offset > inode->i_sb->s_maxbytes) {
784 offset = -EINVAL;
785 goto out;
786 }
787
788 /* Special lock needed here? */
789 if (offset != file->f_pos) {
790 file->f_pos = offset;
791 file->f_version = 0;
792 }
793
794out:
795 mutex_unlock(&inode->i_mutex);
796 return offset;
797}
798
799const struct file_operations ceph_file_fops = {
800 .open = ceph_open,
801 .release = ceph_release,
802 .llseek = ceph_llseek,
803 .read = do_sync_read,
804 .write = do_sync_write,
805 .aio_read = ceph_aio_read,
806 .aio_write = ceph_aio_write,
807 .mmap = ceph_mmap,
808 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -0700809 .lock = ceph_lock,
810 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -0700811 .splice_read = generic_file_splice_read,
812 .splice_write = generic_file_splice_write,
813 .unlocked_ioctl = ceph_ioctl,
814 .compat_ioctl = ceph_ioctl,
815};
816