blob: ecebbc09bfc77f8040e3ef1d572bfe10396b6498 [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>
Sage Weil5ef50c32012-07-31 11:27:36 -07007#include <linux/mount.h>
Sage Weil124e68e2009-10-06 11:31:08 -07008#include <linux/namei.h>
9#include <linux/writeback.h>
10
11#include "super.h"
12#include "mds_client.h"
13
14/*
15 * Ceph file operations
16 *
17 * Implement basic open/close functionality, and implement
18 * read/write.
19 *
20 * We implement three modes of file I/O:
21 * - buffered uses the generic_file_aio_{read,write} helpers
22 *
23 * - synchronous is used when there is multi-client read/write
24 * sharing, avoids the page cache, and synchronously waits for an
25 * ack from the OSD.
26 *
27 * - direct io takes the variant of the sync path that references
28 * user pages directly.
29 *
30 * fsync() flushes and waits on dirty pages, but just queues metadata
31 * for writeback: since the MDS can recover size and mtime there is no
32 * need to wait for MDS acknowledgement.
33 */
34
35
36/*
37 * Prepare an open request. Preallocate ceph_cap to avoid an
38 * inopportune ENOMEM later.
39 */
40static struct ceph_mds_request *
41prepare_open_request(struct super_block *sb, int flags, int create_mode)
42{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070043 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
44 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070045 struct ceph_mds_request *req;
46 int want_auth = USE_ANY_MDS;
47 int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN;
48
49 if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC))
50 want_auth = USE_AUTH_MDS;
51
52 req = ceph_mdsc_create_request(mdsc, op, want_auth);
53 if (IS_ERR(req))
54 goto out;
55 req->r_fmode = ceph_flags_to_mode(flags);
56 req->r_args.open.flags = cpu_to_le32(flags);
57 req->r_args.open.mode = cpu_to_le32(create_mode);
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/*
Sage Weil124e68e2009-10-06 11:31:08 -0700110 * If we already have the requisite capabilities, we can satisfy
111 * the open request locally (no need to request new caps from the
112 * MDS). We do, however, need to inform the MDS (asynchronously)
113 * if our wanted caps set expands.
114 */
115int ceph_open(struct inode *inode, struct file *file)
116{
117 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700118 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
119 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700120 struct ceph_mds_request *req;
121 struct ceph_file_info *cf = file->private_data;
Sage Weil5f21c962011-07-26 11:30:29 -0700122 struct inode *parent_inode = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700123 int err;
124 int flags, fmode, wanted;
125
126 if (cf) {
127 dout("open file %p is already opened\n", file);
128 return 0;
129 }
130
131 /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */
132 flags = file->f_flags & ~(O_CREAT|O_EXCL);
133 if (S_ISDIR(inode->i_mode))
134 flags = O_DIRECTORY; /* mds likes to know */
135
136 dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
137 ceph_vinop(inode), file, flags, file->f_flags);
138 fmode = ceph_flags_to_mode(flags);
139 wanted = ceph_caps_for_mode(fmode);
140
141 /* snapped files are read-only */
142 if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
143 return -EROFS;
144
145 /* trivially open snapdir */
146 if (ceph_snap(inode) == CEPH_SNAPDIR) {
Sage Weilbe655592011-11-30 09:47:09 -0800147 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700148 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800149 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700150 return ceph_init_file(inode, file, fmode);
151 }
152
153 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800154 * No need to block if we have caps on the auth MDS (for
155 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700156 * asynchronously.
157 */
Sage Weilbe655592011-11-30 09:47:09 -0800158 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800159 if (__ceph_is_any_real_caps(ci) &&
160 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700161 int mds_wanted = __ceph_caps_mds_wanted(ci);
162 int issued = __ceph_caps_issued(ci, NULL);
163
164 dout("open %p fmode %d want %s issued %s using existing\n",
165 inode, fmode, ceph_cap_string(wanted),
166 ceph_cap_string(issued));
167 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800168 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700169
170 /* adjust wanted? */
171 if ((issued & wanted) != wanted &&
172 (mds_wanted & wanted) != wanted &&
173 ceph_snap(inode) != CEPH_SNAPDIR)
174 ceph_check_caps(ci, 0, NULL);
175
176 return ceph_init_file(inode, file, fmode);
177 } else if (ceph_snap(inode) != CEPH_NOSNAP &&
178 (ci->i_snap_caps & wanted) == wanted) {
179 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800180 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700181 return ceph_init_file(inode, file, fmode);
182 }
Sage Weilbe655592011-11-30 09:47:09 -0800183 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700184
185 dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted));
186 req = prepare_open_request(inode->i_sb, flags, 0);
187 if (IS_ERR(req)) {
188 err = PTR_ERR(req);
189 goto out;
190 }
Sage Weil70b666c2011-05-27 09:24:26 -0700191 req->r_inode = inode;
192 ihold(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700193 req->r_num_caps = 1;
Sage Weil5f21c962011-07-26 11:30:29 -0700194 if (flags & (O_CREAT|O_TRUNC))
195 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
Sage Weil124e68e2009-10-06 11:31:08 -0700196 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -0700197 iput(parent_inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700198 if (!err)
199 err = ceph_init_file(inode, file, req->r_fmode);
200 ceph_mdsc_put_request(req);
201 dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode));
202out:
203 return err;
204}
205
206
207/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700208 * Do a lookup + open with a single request. If we get a non-existent
209 * file or symlink, return 1 so the VFS can retry.
Sage Weil124e68e2009-10-06 11:31:08 -0700210 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700211int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400212 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400213 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700214{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700215 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
216 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700217 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700218 struct dentry *dn;
Sage Weil124e68e2009-10-06 11:31:08 -0700219 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700220
Sage Weil5ef50c32012-07-31 11:27:36 -0700221 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
222 dir, dentry, dentry->d_name.len, dentry->d_name.name,
223 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
224
225 if (dentry->d_name.len > NAME_MAX)
226 return -ENAMETOOLONG;
227
228 err = ceph_init_dentry(dentry);
229 if (err < 0)
230 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700231
232 /* do the open */
233 req = prepare_open_request(dir->i_sb, flags, mode);
234 if (IS_ERR(req))
Al Virod9585272012-06-22 12:39:14 +0400235 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700236 req->r_dentry = dget(dentry);
237 req->r_num_caps = 2;
238 if (flags & O_CREAT) {
239 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
240 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
241 }
242 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700243 err = ceph_mdsc_do_request(mdsc,
244 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
245 req);
Sage Weil468640e2011-07-26 11:28:11 -0700246 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil5ef50c32012-07-31 11:27:36 -0700247 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700248 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700249
250 if (d_unhashed(dentry)) {
251 dn = ceph_finish_lookup(req, dentry, err);
252 if (IS_ERR(dn))
253 err = PTR_ERR(dn);
254 } else {
255 /* we were given a hashed negative dentry */
256 dn = NULL;
257 }
Sage Weil468640e2011-07-26 11:28:11 -0700258 if (err)
Sage Weil5ef50c32012-07-31 11:27:36 -0700259 goto out_err;
260 if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
261 /* make vfs retry on splice, ENOENT, or symlink */
262 dout("atomic_open finish_no_open on dn %p\n", dn);
263 err = finish_no_open(file, dn);
264 } else {
265 dout("atomic_open finish_open on dn %p\n", dn);
266 err = finish_open(file, dentry, ceph_open, opened);
267 }
268
269out_err:
Sage Weil124e68e2009-10-06 11:31:08 -0700270 ceph_mdsc_put_request(req);
Sage Weil5ef50c32012-07-31 11:27:36 -0700271 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400272 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700273}
274
275int ceph_release(struct inode *inode, struct file *file)
276{
277 struct ceph_inode_info *ci = ceph_inode(inode);
278 struct ceph_file_info *cf = file->private_data;
279
280 dout("release inode %p file %p\n", inode, file);
281 ceph_put_fmode(ci, cf->fmode);
282 if (cf->last_readdir)
283 ceph_mdsc_put_request(cf->last_readdir);
284 kfree(cf->last_name);
285 kfree(cf->dir_info);
286 dput(cf->dentry);
287 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800288
289 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700290 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700291 return 0;
292}
293
294/*
Sage Weil124e68e2009-10-06 11:31:08 -0700295 * Read a range of bytes striped over one or more objects. Iterate over
296 * objects we stripe over. (That's not atomic, but good enough for now.)
297 *
298 * If we get a short result from the OSD, check against i_size; we need to
299 * only return a short read to the caller if we hit EOF.
300 */
301static int striped_read(struct inode *inode,
302 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800303 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700304 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800305 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700306{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700307 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700308 struct ceph_inode_info *ci = ceph_inode(inode);
309 u64 pos, this_len;
Sage Weilb7495fc2010-11-09 12:43:12 -0800310 int io_align, page_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700311 int left, pages_left;
312 int read;
313 struct page **page_pos;
314 int ret;
315 bool hit_stripe, was_short;
316
317 /*
318 * we may need to do multiple reads. not atomic, unfortunately.
319 */
320 pos = off;
321 left = len;
322 page_pos = pages;
323 pages_left = num_pages;
324 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800325 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700326
327more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700328 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800329 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800330 else
331 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700332 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700333 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700334 &ci->i_layout, pos, &this_len,
335 ci->i_truncate_seq,
336 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800337 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700338 if (ret == -ENOENT)
339 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700340 hit_stripe = this_len < left;
341 was_short = ret >= 0 && ret < this_len;
Sage Weil124e68e2009-10-06 11:31:08 -0700342 dout("striped_read %llu~%u (read %u) got %d%s%s\n", pos, left, read,
343 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
344
345 if (ret > 0) {
Sage Weil773e9b42011-06-07 20:57:14 -0700346 int didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700347
348 if (read < pos - off) {
349 dout(" zero gap %llu to %llu\n", off + read, pos);
Sage Weil773e9b42011-06-07 20:57:14 -0700350 ceph_zero_page_vector_range(page_align + read,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700351 pos - off - read, pages);
Sage Weil124e68e2009-10-06 11:31:08 -0700352 }
353 pos += ret;
354 read = pos - off;
355 left -= ret;
356 page_pos += didpages;
357 pages_left -= didpages;
358
359 /* hit stripe? */
360 if (left && hit_stripe)
361 goto more;
362 }
363
364 if (was_short) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700365 /* did we bounce off eof? */
366 if (pos + left > inode->i_size)
367 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700368
Sage Weilc3cd6282011-06-01 16:08:44 -0700369 /* zero trailing bytes (inside i_size) */
370 if (left > 0 && pos < inode->i_size) {
371 if (pos + left > inode->i_size)
372 left = inode->i_size - pos;
373
374 dout("zero tail %d\n", left);
Sage Weil773e9b42011-06-07 20:57:14 -0700375 ceph_zero_page_vector_range(page_align + read, left,
Sage Weilc3cd6282011-06-01 16:08:44 -0700376 pages);
377 read += left;
378 }
Sage Weil124e68e2009-10-06 11:31:08 -0700379 }
380
Sage Weil124e68e2009-10-06 11:31:08 -0700381 if (ret >= 0)
382 ret = read;
383 dout("striped_read returns %d\n", ret);
384 return ret;
385}
386
387/*
388 * Completely synchronous read and write methods. Direct from __user
389 * buffer to osd, or directly to user pages (if O_DIRECT).
390 *
391 * If the read spans object boundary, just do multiple reads.
392 */
393static ssize_t ceph_sync_read(struct file *file, char __user *data,
Sage Weil6a026582010-02-09 14:04:02 -0800394 unsigned len, loff_t *poff, int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700395{
396 struct inode *inode = file->f_dentry->d_inode;
397 struct page **pages;
398 u64 off = *poff;
Henry C Changab226e22010-12-15 20:41:54 -0800399 int num_pages, ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700400
401 dout("sync_read on file %p %llu~%u %s\n", file, off, len,
402 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
403
Henry C Changab226e22010-12-15 20:41:54 -0800404 if (file->f_flags & O_DIRECT) {
405 num_pages = calc_pages_for((unsigned long)data, len);
Henry C Changb6aa5902010-12-15 20:45:41 -0800406 pages = ceph_get_direct_page_vector(data, num_pages, true);
Henry C Changab226e22010-12-15 20:41:54 -0800407 } else {
408 num_pages = calc_pages_for(off, len);
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700409 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Henry C Changab226e22010-12-15 20:41:54 -0800410 }
Sage Weil124e68e2009-10-06 11:31:08 -0700411 if (IS_ERR(pages))
412 return PTR_ERR(pages);
413
Sage Weile98b6fe2010-11-09 12:24:53 -0800414 /*
415 * flush any page cache pages in this range. this
416 * will make concurrent normal and sync io slow,
417 * but it will at least behave sensibly when they are
418 * in sequence.
419 */
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800420 ret = filemap_write_and_wait(inode->i_mapping);
421 if (ret < 0)
422 goto done;
423
Sage Weilb7495fc2010-11-09 12:43:12 -0800424 ret = striped_read(inode, off, len, pages, num_pages, checkeof,
Henry C Changab226e22010-12-15 20:41:54 -0800425 file->f_flags & O_DIRECT,
426 (unsigned long)data & ~PAGE_MASK);
Sage Weil124e68e2009-10-06 11:31:08 -0700427
428 if (ret >= 0 && (file->f_flags & O_DIRECT) == 0)
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700429 ret = ceph_copy_page_vector_to_user(pages, data, off, ret);
Sage Weil124e68e2009-10-06 11:31:08 -0700430 if (ret >= 0)
431 *poff = off + ret;
432
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800433done:
Sage Weil124e68e2009-10-06 11:31:08 -0700434 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800435 ceph_put_page_vector(pages, num_pages, true);
Sage Weil124e68e2009-10-06 11:31:08 -0700436 else
437 ceph_release_page_vector(pages, num_pages);
438 dout("sync_read result %d\n", ret);
439 return ret;
440}
441
442/*
443 * Write commit callback, called if we requested both an ACK and
444 * ONDISK commit reply from the OSD.
445 */
446static void sync_write_commit(struct ceph_osd_request *req,
447 struct ceph_msg *msg)
448{
449 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
450
451 dout("sync_write_commit %p tid %llu\n", req, req->r_tid);
452 spin_lock(&ci->i_unsafe_lock);
453 list_del_init(&req->r_unsafe_item);
454 spin_unlock(&ci->i_unsafe_lock);
455 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
456}
457
458/*
459 * Synchronous write, straight from __user pointer or user pages (if
460 * O_DIRECT).
461 *
462 * If write spans object boundary, just do multiple writes. (For a
463 * correct atomic write, we should e.g. take write locks on all
464 * objects, rollback on failure, etc.)
465 */
466static ssize_t ceph_sync_write(struct file *file, const char __user *data,
467 size_t left, loff_t *offset)
468{
469 struct inode *inode = file->f_dentry->d_inode;
470 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700471 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700472 struct ceph_osd_request *req;
473 struct page **pages;
474 int num_pages;
475 long long unsigned pos;
476 u64 len;
477 int written = 0;
478 int flags;
479 int do_sync = 0;
480 int check_caps = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800481 int page_align, io_align;
Henry C Changab226e22010-12-15 20:41:54 -0800482 unsigned long buf_align;
Sage Weil124e68e2009-10-06 11:31:08 -0700483 int ret;
484 struct timespec mtime = CURRENT_TIME;
485
486 if (ceph_snap(file->f_dentry->d_inode) != CEPH_NOSNAP)
487 return -EROFS;
488
489 dout("sync_write on file %p %lld~%u %s\n", file, *offset,
490 (unsigned)left, (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
491
492 if (file->f_flags & O_APPEND)
493 pos = i_size_read(inode);
494 else
495 pos = *offset;
496
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800497 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + left);
498 if (ret < 0)
499 return ret;
500
501 ret = invalidate_inode_pages2_range(inode->i_mapping,
502 pos >> PAGE_CACHE_SHIFT,
503 (pos + left) >> PAGE_CACHE_SHIFT);
504 if (ret < 0)
505 dout("invalidate_inode_pages2_range returned %d\n", ret);
506
Sage Weil124e68e2009-10-06 11:31:08 -0700507 flags = CEPH_OSD_FLAG_ORDERSNAP |
508 CEPH_OSD_FLAG_ONDISK |
509 CEPH_OSD_FLAG_WRITE;
510 if ((file->f_flags & (O_SYNC|O_DIRECT)) == 0)
511 flags |= CEPH_OSD_FLAG_ACK;
512 else
513 do_sync = 1;
514
515 /*
516 * we may need to do multiple writes here if we span an object
517 * boundary. this isn't atomic, unfortunately. :(
518 */
519more:
Sage Weild7f124f2011-06-13 16:22:18 -0700520 io_align = pos & ~PAGE_MASK;
521 buf_align = (unsigned long)data & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700522 len = left;
Henry C Changab226e22010-12-15 20:41:54 -0800523 if (file->f_flags & O_DIRECT) {
Sage Weilb7495fc2010-11-09 12:43:12 -0800524 /* write from beginning of first page, regardless of
525 io alignment */
Henry C Changab226e22010-12-15 20:41:54 -0800526 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
527 num_pages = calc_pages_for((unsigned long)data, len);
528 } else {
Sage Weilb7495fc2010-11-09 12:43:12 -0800529 page_align = pos & ~PAGE_MASK;
Henry C Changab226e22010-12-15 20:41:54 -0800530 num_pages = calc_pages_for(pos, len);
531 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700532 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
Sage Weil124e68e2009-10-06 11:31:08 -0700533 ceph_vino(inode), pos, &len,
534 CEPH_OSD_OP_WRITE, flags,
535 ci->i_snap_realm->cached_context,
536 do_sync,
537 ci->i_truncate_seq, ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800538 &mtime, false, 2, page_align);
Sage Weila79832f2010-04-01 16:06:19 -0700539 if (!req)
540 return -ENOMEM;
Sage Weil124e68e2009-10-06 11:31:08 -0700541
Sage Weil124e68e2009-10-06 11:31:08 -0700542 if (file->f_flags & O_DIRECT) {
Henry C Changb6aa5902010-12-15 20:45:41 -0800543 pages = ceph_get_direct_page_vector(data, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700544 if (IS_ERR(pages)) {
545 ret = PTR_ERR(pages);
546 goto out;
547 }
548
549 /*
550 * throw out any page cache pages in this range. this
551 * may block.
552 */
Sage Weil213c99e2010-08-03 10:25:11 -0700553 truncate_inode_pages_range(inode->i_mapping, pos,
Sage Weil5c6a2cd2010-04-22 13:48:59 -0700554 (pos+len) | (PAGE_CACHE_SIZE-1));
Sage Weil124e68e2009-10-06 11:31:08 -0700555 } else {
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700556 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700557 if (IS_ERR(pages)) {
558 ret = PTR_ERR(pages);
559 goto out;
560 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700561 ret = ceph_copy_user_to_page_vector(pages, data, pos, len);
Sage Weil124e68e2009-10-06 11:31:08 -0700562 if (ret < 0) {
563 ceph_release_page_vector(pages, num_pages);
564 goto out;
565 }
566
567 if ((file->f_flags & O_SYNC) == 0) {
568 /* get a second commit callback */
569 req->r_safe_callback = sync_write_commit;
570 req->r_own_pages = 1;
571 }
572 }
573 req->r_pages = pages;
574 req->r_num_pages = num_pages;
575 req->r_inode = inode;
576
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700577 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700578 if (!ret) {
579 if (req->r_safe_callback) {
580 /*
581 * Add to inode unsafe list only after we
582 * start_request so that a tid has been assigned.
583 */
584 spin_lock(&ci->i_unsafe_lock);
Henry C Chang49bcb932011-03-15 09:18:02 +0000585 list_add_tail(&req->r_unsafe_item,
586 &ci->i_unsafe_writes);
Sage Weil124e68e2009-10-06 11:31:08 -0700587 spin_unlock(&ci->i_unsafe_lock);
588 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
589 }
Henry C Chang78a25562011-03-15 09:18:01 +0000590
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700591 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Henry C Chang78a25562011-03-15 09:18:01 +0000592 if (ret < 0 && req->r_safe_callback) {
593 spin_lock(&ci->i_unsafe_lock);
594 list_del_init(&req->r_unsafe_item);
595 spin_unlock(&ci->i_unsafe_lock);
596 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
597 }
Sage Weil124e68e2009-10-06 11:31:08 -0700598 }
599
600 if (file->f_flags & O_DIRECT)
Henry C Changb6aa5902010-12-15 20:45:41 -0800601 ceph_put_page_vector(pages, num_pages, false);
Sage Weil124e68e2009-10-06 11:31:08 -0700602 else if (file->f_flags & O_SYNC)
603 ceph_release_page_vector(pages, num_pages);
604
605out:
606 ceph_osdc_put_request(req);
607 if (ret == 0) {
608 pos += len;
609 written += len;
610 left -= len;
Sage Weild7f124f2011-06-13 16:22:18 -0700611 data += written;
Sage Weil124e68e2009-10-06 11:31:08 -0700612 if (left)
613 goto more;
614
615 ret = written;
616 *offset = pos;
617 if (pos > i_size_read(inode))
618 check_caps = ceph_inode_set_size(inode, pos);
619 if (check_caps)
620 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY,
621 NULL);
622 }
623 return ret;
624}
625
626/*
627 * Wrap generic_file_aio_read with checks for cap bits on the inode.
628 * Atomically grab references, so that those bits are not released
629 * back to the MDS mid-read.
630 *
631 * Hmm, the sync read case isn't actually async... should it be?
632 */
633static ssize_t ceph_aio_read(struct kiocb *iocb, const struct iovec *iov,
634 unsigned long nr_segs, loff_t pos)
635{
636 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700637 struct ceph_file_info *fi = filp->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700638 loff_t *ppos = &iocb->ki_pos;
639 size_t len = iov->iov_len;
640 struct inode *inode = filp->f_dentry->d_inode;
641 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadehcd84db62010-06-11 16:58:48 -0700642 void __user *base = iov->iov_base;
Sage Weil124e68e2009-10-06 11:31:08 -0700643 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700644 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800645 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700646
647 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
648 inode, ceph_vinop(inode), pos, (unsigned)len, inode);
Sage Weil6a026582010-02-09 14:04:02 -0800649again:
Sage Weil124e68e2009-10-06 11:31:08 -0700650 __ceph_do_pending_vmtruncate(inode);
Sage Weil29625072010-05-27 10:40:43 -0700651 if (fi->fmode & CEPH_FILE_MODE_LAZY)
652 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
653 else
654 want = CEPH_CAP_FILE_CACHE;
655 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700656 if (ret < 0)
657 goto out;
658 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
659 inode, ceph_vinop(inode), pos, (unsigned)len,
660 ceph_cap_string(got));
661
Sage Weil29625072010-05-27 10:40:43 -0700662 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700663 (iocb->ki_filp->f_flags & O_DIRECT) ||
Sage Weil4918b6d2011-07-26 11:26:07 -0700664 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
665 (fi->flags & CEPH_F_SYNC))
Sage Weil124e68e2009-10-06 11:31:08 -0700666 /* hmm, this isn't really async... */
Sage Weil6a026582010-02-09 14:04:02 -0800667 ret = ceph_sync_read(filp, base, len, ppos, &checkeof);
Sage Weil124e68e2009-10-06 11:31:08 -0700668 else
669 ret = generic_file_aio_read(iocb, iov, nr_segs, pos);
670
671out:
672 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
673 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
674 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800675
676 if (checkeof && ret >= 0) {
677 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
678
679 /* hit EOF or hole? */
680 if (statret == 0 && *ppos < inode->i_size) {
Sage Weilc3cd6282011-06-01 16:08:44 -0700681 dout("aio_read sync_read hit hole, ppos %lld < size %lld, reading more\n", *ppos, inode->i_size);
Sage Weil6a026582010-02-09 14:04:02 -0800682 read += ret;
683 base += ret;
684 len -= ret;
685 checkeof = 0;
686 goto again;
687 }
688 }
689 if (ret >= 0)
690 ret += read;
691
Sage Weil124e68e2009-10-06 11:31:08 -0700692 return ret;
693}
694
695/*
696 * Take cap references to avoid releasing caps to MDS mid-write.
697 *
698 * If we are synchronous, and write with an old snap context, the OSD
699 * may return EOLDSNAPC. In that case, retry the write.. _after_
700 * dropping our cap refs and allowing the pending snap to logically
701 * complete _before_ this write occurs.
702 *
703 * If we are near ENOSPC, write synchronously.
704 */
705static ssize_t ceph_aio_write(struct kiocb *iocb, const struct iovec *iov,
706 unsigned long nr_segs, loff_t pos)
707{
708 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700709 struct ceph_file_info *fi = file->private_data;
Sage Weil124e68e2009-10-06 11:31:08 -0700710 struct inode *inode = file->f_dentry->d_inode;
711 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700712 struct ceph_osd_client *osdc =
713 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Sage Weil124e68e2009-10-06 11:31:08 -0700714 loff_t endoff = pos + iov->iov_len;
Sage Weil33caad32010-05-26 14:31:27 -0700715 int want, got = 0;
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000716 int ret, err;
Sage Weil124e68e2009-10-06 11:31:08 -0700717
718 if (ceph_snap(inode) != CEPH_NOSNAP)
719 return -EROFS;
720
721retry_snap:
722 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL))
723 return -ENOSPC;
724 __ceph_do_pending_vmtruncate(inode);
725 dout("aio_write %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
726 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
727 inode->i_size);
Sage Weil33caad32010-05-26 14:31:27 -0700728 if (fi->fmode & CEPH_FILE_MODE_LAZY)
729 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
730 else
731 want = CEPH_CAP_FILE_BUFFER;
732 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
Sage Weil124e68e2009-10-06 11:31:08 -0700733 if (ret < 0)
Sage Weild8de9ab2011-07-26 11:27:34 -0700734 goto out_put;
Sage Weil124e68e2009-10-06 11:31:08 -0700735
736 dout("aio_write %p %llx.%llx %llu~%u got cap refs on %s\n",
737 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
738 ceph_cap_string(got));
739
Sage Weil33caad32010-05-26 14:31:27 -0700740 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700741 (iocb->ki_filp->f_flags & O_DIRECT) ||
Sage Weil4918b6d2011-07-26 11:26:07 -0700742 (inode->i_sb->s_flags & MS_SYNCHRONOUS) ||
743 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700744 ret = ceph_sync_write(file, iov->iov_base, iov->iov_len,
745 &iocb->ki_pos);
746 } else {
Sage Weild8de9ab2011-07-26 11:27:34 -0700747 /*
748 * buffered write; drop Fw early to avoid slow
749 * revocation if we get stuck on balance_dirty_pages
750 */
751 int dirty;
Sage Weil124e68e2009-10-06 11:31:08 -0700752
Sage Weilbe655592011-11-30 09:47:09 -0800753 spin_lock(&ci->i_ceph_lock);
Sage Weild8de9ab2011-07-26 11:27:34 -0700754 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800755 spin_unlock(&ci->i_ceph_lock);
Sage Weild8de9ab2011-07-26 11:27:34 -0700756 ceph_put_cap_refs(ci, got);
757
758 ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
Sage Weil124e68e2009-10-06 11:31:08 -0700759 if ((ret >= 0 || ret == -EIOCBQUEUED) &&
760 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host)
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000761 || ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Christoph Hellwig8018ab02010-03-22 17:32:25 +0100762 err = vfs_fsync_range(file, pos, pos + ret - 1, 1);
Yehuda Sadeh88d892a2010-02-23 18:16:23 +0000763 if (err < 0)
764 ret = err;
765 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700766
767 if (dirty)
768 __mark_inode_dirty(inode, dirty);
769 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -0700770 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700771
Sage Weil124e68e2009-10-06 11:31:08 -0700772 if (ret >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -0700773 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -0800774 spin_lock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700775 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800776 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700777 if (dirty)
778 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -0700779 }
780
Sage Weild8de9ab2011-07-26 11:27:34 -0700781out_put:
Sage Weil124e68e2009-10-06 11:31:08 -0700782 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
783 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len,
784 ceph_cap_string(got));
785 ceph_put_cap_refs(ci, got);
786
Sage Weild8de9ab2011-07-26 11:27:34 -0700787out:
Sage Weil124e68e2009-10-06 11:31:08 -0700788 if (ret == -EOLDSNAPC) {
789 dout("aio_write %p %llx.%llx %llu~%u got EOLDSNAPC, retrying\n",
790 inode, ceph_vinop(inode), pos, (unsigned)iov->iov_len);
791 goto retry_snap;
792 }
793
794 return ret;
795}
796
797/*
798 * llseek. be sure to verify file size on SEEK_END.
799 */
800static loff_t ceph_llseek(struct file *file, loff_t offset, int origin)
801{
802 struct inode *inode = file->f_mapping->host;
803 int ret;
804
805 mutex_lock(&inode->i_mutex);
806 __ceph_do_pending_vmtruncate(inode);
Sage Weil6a82c472011-12-13 09:19:26 -0800807
808 if (origin == SEEK_END || origin == SEEK_DATA || origin == SEEK_HOLE) {
Sage Weil124e68e2009-10-06 11:31:08 -0700809 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
810 if (ret < 0) {
811 offset = ret;
812 goto out;
813 }
Josef Bacik06222e42011-07-18 13:21:38 -0400814 }
815
816 switch (origin) {
817 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -0700818 offset += inode->i_size;
819 break;
820 case SEEK_CUR:
821 /*
822 * Here we special-case the lseek(fd, 0, SEEK_CUR)
823 * position-querying operation. Avoid rewriting the "same"
824 * f_pos value back to the file because a concurrent read(),
825 * write() or lseek() might have altered it
826 */
827 if (offset == 0) {
828 offset = file->f_pos;
829 goto out;
830 }
831 offset += file->f_pos;
832 break;
Josef Bacik06222e42011-07-18 13:21:38 -0400833 case SEEK_DATA:
834 if (offset >= inode->i_size) {
835 ret = -ENXIO;
836 goto out;
837 }
838 break;
839 case SEEK_HOLE:
840 if (offset >= inode->i_size) {
841 ret = -ENXIO;
842 goto out;
843 }
844 offset = inode->i_size;
845 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700846 }
847
848 if (offset < 0 || offset > inode->i_sb->s_maxbytes) {
849 offset = -EINVAL;
850 goto out;
851 }
852
853 /* Special lock needed here? */
854 if (offset != file->f_pos) {
855 file->f_pos = offset;
856 file->f_version = 0;
857 }
858
859out:
860 mutex_unlock(&inode->i_mutex);
861 return offset;
862}
863
864const struct file_operations ceph_file_fops = {
865 .open = ceph_open,
866 .release = ceph_release,
867 .llseek = ceph_llseek,
868 .read = do_sync_read,
869 .write = do_sync_write,
870 .aio_read = ceph_aio_read,
871 .aio_write = ceph_aio_write,
872 .mmap = ceph_mmap,
873 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -0700874 .lock = ceph_lock,
875 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -0700876 .splice_read = generic_file_splice_read,
877 .splice_write = generic_file_splice_write,
878 .unlocked_ioctl = ceph_ioctl,
879 .compat_ioctl = ceph_ioctl,
880};
881