blob: 2eb02f80a0ab05de9071bc5d076f6c8fe952cbf5 [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>
Kent Overstreeta27bb332013-05-07 16:19:08 -070010#include <linux/aio.h>
Li Wangad7a60d2013-08-15 11:51:44 +080011#include <linux/falloc.h>
Sage Weil124e68e2009-10-06 11:31:08 -070012
13#include "super.h"
14#include "mds_client.h"
Milosz Tanski99ccbd22013-08-21 17:29:54 -040015#include "cache.h"
Sage Weil124e68e2009-10-06 11:31:08 -070016
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 */
43static struct ceph_mds_request *
44prepare_open_request(struct super_block *sb, int flags, int create_mode)
45{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070046 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
47 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -070048 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);
Sage Weil124e68e2009-10-06 11:31:08 -070061out:
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 */
69static int ceph_init_file(struct inode *inode, struct file *file, int fmode)
70{
71 struct ceph_file_info *cf;
72 int ret = 0;
Milosz Tanski99ccbd22013-08-21 17:29:54 -040073 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;
Sage Weil124e68e2009-10-06 11:31:08 -070076
77 switch (inode->i_mode & S_IFMT) {
78 case S_IFREG:
Milosz Tanski99ccbd22013-08-21 17:29:54 -040079 /* 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);
Sage Weil124e68e2009-10-06 11:31:08 -070090 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/*
Sage Weil124e68e2009-10-06 11:31:08 -0700127 * 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 */
132int ceph_open(struct inode *inode, struct file *file)
133{
134 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700135 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
136 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700137 struct ceph_mds_request *req;
138 struct ceph_file_info *cf = file->private_data;
Sage Weil5f21c962011-07-26 11:30:29 -0700139 struct inode *parent_inode = NULL;
Sage Weil124e68e2009-10-06 11:31:08 -0700140 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) {
Sage Weilbe655592011-11-30 09:47:09 -0800164 spin_lock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700165 __ceph_get_fmode(ci, fmode);
Sage Weilbe655592011-11-30 09:47:09 -0800166 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700167 return ceph_init_file(inode, file, fmode);
168 }
169
170 /*
Sage Weil7421ab82010-11-07 09:07:15 -0800171 * No need to block if we have caps on the auth MDS (for
172 * write) or any MDS (for read). Update wanted set
Sage Weil124e68e2009-10-06 11:31:08 -0700173 * asynchronously.
174 */
Sage Weilbe655592011-11-30 09:47:09 -0800175 spin_lock(&ci->i_ceph_lock);
Sage Weil7421ab82010-11-07 09:07:15 -0800176 if (__ceph_is_any_real_caps(ci) &&
177 (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700178 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);
Sage Weilbe655592011-11-30 09:47:09 -0800185 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700186
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);
Sage Weilbe655592011-11-30 09:47:09 -0800197 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700198 return ceph_init_file(inode, file, fmode);
199 }
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400200
Sage Weilbe655592011-11-30 09:47:09 -0800201 spin_unlock(&ci->i_ceph_lock);
Sage Weil124e68e2009-10-06 11:31:08 -0700202
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 }
Sage Weil70b666c2011-05-27 09:24:26 -0700209 req->r_inode = inode;
210 ihold(inode);
Milosz Tanski99ccbd22013-08-21 17:29:54 -0400211
Sage Weil124e68e2009-10-06 11:31:08 -0700212 req->r_num_caps = 1;
Sage Weil752c8bd2013-02-05 13:52:29 -0800213 if (flags & O_CREAT)
Sage Weil5f21c962011-07-26 11:30:29 -0700214 parent_inode = ceph_get_dentry_parent_inode(file->f_dentry);
Sage Weil124e68e2009-10-06 11:31:08 -0700215 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
Sage Weil5f21c962011-07-26 11:30:29 -0700216 iput(parent_inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700217 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));
221out:
222 return err;
223}
224
225
226/*
Sage Weil5ef50c32012-07-31 11:27:36 -0700227 * 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.
Sage Weil124e68e2009-10-06 11:31:08 -0700229 */
Sage Weil5ef50c32012-07-31 11:27:36 -0700230int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
Al Viro30d90492012-06-22 12:40:19 +0400231 struct file *file, unsigned flags, umode_t mode,
Al Virod9585272012-06-22 12:39:14 +0400232 int *opened)
Sage Weil124e68e2009-10-06 11:31:08 -0700233{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700234 struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
235 struct ceph_mds_client *mdsc = fsc->mdsc;
Sage Weil124e68e2009-10-06 11:31:08 -0700236 struct ceph_mds_request *req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700237 struct dentry *dn;
Sage Weil124e68e2009-10-06 11:31:08 -0700238 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700239
Sage Weil5ef50c32012-07-31 11:27:36 -0700240 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;
Sage Weil124e68e2009-10-06 11:31:08 -0700250
251 /* do the open */
252 req = prepare_open_request(dir->i_sb, flags, mode);
253 if (IS_ERR(req))
Al Virod9585272012-06-22 12:39:14 +0400254 return PTR_ERR(req);
Sage Weil124e68e2009-10-06 11:31:08 -0700255 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 */
Sage Weilacda7652011-07-26 11:27:48 -0700262 err = ceph_mdsc_do_request(mdsc,
263 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
264 req);
Sam Lang79aec9842012-12-19 09:44:23 -1000265 if (err)
266 goto out_err;
267
Sage Weil468640e2011-07-26 11:28:11 -0700268 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil5ef50c32012-07-31 11:27:36 -0700269 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700270 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700271
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 }
Sage Weil468640e2011-07-26 11:28:11 -0700280 if (err)
Sage Weil5ef50c32012-07-31 11:27:36 -0700281 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);
Sam Lang6e8575f2012-12-28 09:56:46 -0800288 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
Yan, Zhengb20a95a2014-02-11 12:55:05 +0800289 ceph_init_acl(dentry, dentry->d_inode, dir);
Sam Lang6e8575f2012-12-28 09:56:46 -0800290 *opened |= FILE_CREATED;
291 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700292 err = finish_open(file, dentry, ceph_open, opened);
293 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700294out_err:
Yan, Zhengab866542014-04-01 20:20:17 +0800295 if (!req->r_err && req->r_target_inode)
296 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700297 ceph_mdsc_put_request(req);
Sage Weil5ef50c32012-07-31 11:27:36 -0700298 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400299 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700300}
301
302int 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);
Sage Weil195d3ce2010-03-01 09:57:54 -0800315
316 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700317 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700318 return 0;
319}
320
321/*
Sage Weil124e68e2009-10-06 11:31:08 -0700322 * 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 */
328static int striped_read(struct inode *inode,
329 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800330 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700331 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800332 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700333{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700334 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700335 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300336 u64 pos, this_len, left;
Sage Weilb7495fc2010-11-09 12:43:12 -0800337 int io_align, page_align;
Dan Carpenter688bac42013-07-23 16:48:01 +0300338 int pages_left;
Sage Weil124e68e2009-10-06 11:31:08 -0700339 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;
Sage Weilb7495fc2010-11-09 12:43:12 -0800352 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700353
354more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700355 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800356 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800357 else
358 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700359 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700360 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700361 &ci->i_layout, pos, &this_len,
362 ci->i_truncate_seq,
363 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800364 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700365 if (ret == -ENOENT)
366 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700367 hit_stripe = this_len < left;
368 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300369 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700370 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
371
majianpeng02ae66d2013-08-06 16:20:38 +0800372 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;
Sage Weil124e68e2009-10-06 11:31:08 -0700382 }
majianpeng02ae66d2013-08-06 16:20:38 +0800383
384 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700385 pos += ret;
386 read = pos - off;
387 left -= ret;
388 page_pos += didpages;
389 pages_left -= didpages;
390
majianpeng02ae66d2013-08-06 16:20:38 +0800391 /* hit stripe and need continue*/
392 if (left && hit_stripe && pos < inode->i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700393 goto more;
394 }
395
majianpengee7289b2013-08-21 15:02:51 +0800396 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800397 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700398 /* did we bounce off eof? */
399 if (pos + left > inode->i_size)
400 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700401 }
402
Sage Weil124e68e2009-10-06 11:31:08 -0700403 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 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800413static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
414 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700415{
majianpeng8eb4efb2013-09-26 14:42:17 +0800416 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500417 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700418 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800419 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800420 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400421 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700422
majianpeng8eb4efb2013-09-26 14:42:17 +0800423 dout("sync_read on file %p %llu~%u %s\n", file, off,
424 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700425 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800426
427 if (!len)
428 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800429 /*
430 * flush any page cache pages in this range. this
431 * will make concurrent normal and sync io slow,
432 * but it will at least behave sensibly when they are
433 * in sequence.
434 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800435 ret = filemap_write_and_wait_range(inode->i_mapping, off,
436 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800437 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800438 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800439
majianpeng8eb4efb2013-09-26 14:42:17 +0800440 if (file->f_flags & O_DIRECT) {
441 while (iov_iter_count(i)) {
Al Viro2b777c92014-04-03 22:31:22 -0400442 size_t start;
443 ssize_t n;
Sage Weil124e68e2009-10-06 11:31:08 -0700444
Al Viro2b777c92014-04-03 22:31:22 -0400445 n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
446 if (n < 0)
447 return n;
Sage Weil124e68e2009-10-06 11:31:08 -0700448
Al Viro2b777c92014-04-03 22:31:22 -0400449 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
450
451 ret = striped_read(inode, off, n,
majianpeng8eb4efb2013-09-26 14:42:17 +0800452 pages, num_pages, checkeof,
Al Viro2b777c92014-04-03 22:31:22 -0400453 1, start);
454
majianpeng8eb4efb2013-09-26 14:42:17 +0800455 ceph_put_page_vector(pages, num_pages, true);
456
457 if (ret <= 0)
458 break;
459 off += ret;
460 iov_iter_advance(i, ret);
Al Viro2b777c92014-04-03 22:31:22 -0400461 if (ret < n)
majianpeng8eb4efb2013-09-26 14:42:17 +0800462 break;
463 }
464 } else {
465 num_pages = calc_pages_for(off, len);
466 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
467 if (IS_ERR(pages))
468 return PTR_ERR(pages);
469 ret = striped_read(inode, off, len, pages,
470 num_pages, checkeof, 0, 0);
471 if (ret > 0) {
472 int l, k = 0;
Al Viro2b777c92014-04-03 22:31:22 -0400473 size_t left = ret;
majianpeng8eb4efb2013-09-26 14:42:17 +0800474
475 while (left) {
Yan, Zheng5aaa4322014-07-02 10:13:00 +0800476 size_t page_off = off & ~PAGE_MASK;
477 size_t copy = min_t(size_t,
478 PAGE_SIZE - page_off, left);
479 l = copy_page_to_iter(pages[k++], page_off,
480 copy, i);
Al Viro2b777c92014-04-03 22:31:22 -0400481 off += l;
482 left -= l;
483 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800484 break;
485 }
486 }
Sage Weil124e68e2009-10-06 11:31:08 -0700487 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800488 }
489
490 if (off > iocb->ki_pos) {
491 ret = off - iocb->ki_pos;
492 iocb->ki_pos = off;
493 }
494
Sage Weil124e68e2009-10-06 11:31:08 -0700495 dout("sync_read result %d\n", ret);
496 return ret;
497}
498
499/*
Alex Elder26be8802013-04-15 11:20:42 -0500500 * Write commit request unsafe callback, called to tell us when a
501 * request is unsafe (that is, in flight--has been handed to the
502 * messenger to send to its target osd). It is called again when
503 * we've received a response message indicating the request is
504 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
505 * is completed early (and unsuccessfully) due to a timeout or
506 * interrupt.
507 *
508 * This is used if we requested both an ACK and ONDISK commit reply
509 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700510 */
Alex Elder26be8802013-04-15 11:20:42 -0500511static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700512{
513 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
514
Alex Elder26be8802013-04-15 11:20:42 -0500515 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
516 unsafe ? "un" : "");
517 if (unsafe) {
518 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
519 spin_lock(&ci->i_unsafe_lock);
520 list_add_tail(&req->r_unsafe_item,
521 &ci->i_unsafe_writes);
522 spin_unlock(&ci->i_unsafe_lock);
523 } else {
524 spin_lock(&ci->i_unsafe_lock);
525 list_del_init(&req->r_unsafe_item);
526 spin_unlock(&ci->i_unsafe_lock);
527 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
528 }
Sage Weil124e68e2009-10-06 11:31:08 -0700529}
530
majianpenge8344e62013-09-12 13:54:26 +0800531
Sage Weil124e68e2009-10-06 11:31:08 -0700532/*
majianpenge8344e62013-09-12 13:54:26 +0800533 * Synchronous write, straight from __user pointer or user pages.
Sage Weil124e68e2009-10-06 11:31:08 -0700534 *
535 * If write spans object boundary, just do multiple writes. (For a
536 * correct atomic write, we should e.g. take write locks on all
537 * objects, rollback on failure, etc.)
538 */
majianpenge8344e62013-09-12 13:54:26 +0800539static ssize_t
Yan, Zheng06fee302014-07-28 14:33:46 +0800540ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Sage Weil124e68e2009-10-06 11:31:08 -0700541{
majianpenge8344e62013-09-12 13:54:26 +0800542 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500543 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700544 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700545 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500546 struct ceph_snap_context *snapc;
547 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700548 struct ceph_osd_request *req;
549 struct page **pages;
550 int num_pages;
Sage Weil124e68e2009-10-06 11:31:08 -0700551 int written = 0;
552 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700553 int check_caps = 0;
554 int ret;
555 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400556 size_t count = iov_iter_count(from);
Sage Weil124e68e2009-10-06 11:31:08 -0700557
Al Viro496ad9a2013-01-23 17:07:38 -0500558 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700559 return -EROFS;
560
majianpenge8344e62013-09-12 13:54:26 +0800561 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
562 (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700563
majianpenge8344e62013-09-12 13:54:26 +0800564 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800565 if (ret < 0)
566 return ret;
567
568 ret = invalidate_inode_pages2_range(inode->i_mapping,
569 pos >> PAGE_CACHE_SHIFT,
majianpenge8344e62013-09-12 13:54:26 +0800570 (pos + count) >> PAGE_CACHE_SHIFT);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800571 if (ret < 0)
572 dout("invalidate_inode_pages2_range returned %d\n", ret);
573
Sage Weil124e68e2009-10-06 11:31:08 -0700574 flags = CEPH_OSD_FLAG_ORDERSNAP |
575 CEPH_OSD_FLAG_ONDISK |
576 CEPH_OSD_FLAG_WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700577
Al Viro4908b822014-04-03 23:09:01 -0400578 while (iov_iter_count(from) > 0) {
579 u64 len = iov_iter_single_seg_count(from);
Al Viro64c31312014-04-03 22:58:25 -0400580 size_t start;
581 ssize_t n;
majianpenge8344e62013-09-12 13:54:26 +0800582
583 snapc = ci->i_snap_realm->cached_context;
584 vino = ceph_vino(inode);
585 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
586 vino, pos, &len,
587 2,/*include a 'startsync' command*/
588 CEPH_OSD_OP_WRITE, flags, snapc,
589 ci->i_truncate_seq,
590 ci->i_truncate_size,
591 false);
592 if (IS_ERR(req)) {
593 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400594 break;
majianpenge8344e62013-09-12 13:54:26 +0800595 }
596
Al Viro4908b822014-04-03 23:09:01 -0400597 n = iov_iter_get_pages_alloc(from, &pages, len, &start);
Al Viro64c31312014-04-03 22:58:25 -0400598 if (unlikely(n < 0)) {
599 ret = n;
600 ceph_osdc_put_request(req);
601 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700602 }
603
Al Viro64c31312014-04-03 22:58:25 -0400604 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
Sage Weil124e68e2009-10-06 11:31:08 -0700605 /*
606 * throw out any page cache pages in this range. this
607 * may block.
608 */
Sage Weil213c99e2010-08-03 10:25:11 -0700609 truncate_inode_pages_range(inode->i_mapping, pos,
Al Viro64c31312014-04-03 22:58:25 -0400610 (pos+n) | (PAGE_CACHE_SIZE-1));
611 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
majianpenge8344e62013-09-12 13:54:26 +0800612 false, false);
613
614 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
615 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
616
617 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
618 if (!ret)
619 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
620
621 ceph_put_page_vector(pages, num_pages, false);
622
majianpenge8344e62013-09-12 13:54:26 +0800623 ceph_osdc_put_request(req);
Al Viro64c31312014-04-03 22:58:25 -0400624 if (ret)
majianpenge8344e62013-09-12 13:54:26 +0800625 break;
Al Viro64c31312014-04-03 22:58:25 -0400626 pos += n;
627 written += n;
Al Viro4908b822014-04-03 23:09:01 -0400628 iov_iter_advance(from, n);
Al Viro64c31312014-04-03 22:58:25 -0400629
630 if (pos > i_size_read(inode)) {
631 check_caps = ceph_inode_set_size(inode, pos);
632 if (check_caps)
633 ceph_check_caps(ceph_inode(inode),
634 CHECK_CAPS_AUTHONLY,
635 NULL);
636 }
majianpenge8344e62013-09-12 13:54:26 +0800637 }
638
639 if (ret != -EOLDSNAPC && written > 0) {
640 iocb->ki_pos = pos;
641 ret = written;
642 }
643 return ret;
644}
645
646
647/*
648 * Synchronous write, straight from __user pointer or user pages.
649 *
650 * If write spans object boundary, just do multiple writes. (For a
651 * correct atomic write, we should e.g. take write locks on all
652 * objects, rollback on failure, etc.)
653 */
Yan, Zheng06fee302014-07-28 14:33:46 +0800654static ssize_t
655ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
majianpenge8344e62013-09-12 13:54:26 +0800656{
657 struct file *file = iocb->ki_filp;
658 struct inode *inode = file_inode(file);
659 struct ceph_inode_info *ci = ceph_inode(inode);
660 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
661 struct ceph_snap_context *snapc;
662 struct ceph_vino vino;
663 struct ceph_osd_request *req;
664 struct page **pages;
665 u64 len;
666 int num_pages;
667 int written = 0;
668 int flags;
669 int check_caps = 0;
670 int ret;
671 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400672 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +0800673
674 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
675 return -EROFS;
676
677 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
678
679 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
680 if (ret < 0)
681 return ret;
682
683 ret = invalidate_inode_pages2_range(inode->i_mapping,
684 pos >> PAGE_CACHE_SHIFT,
685 (pos + count) >> PAGE_CACHE_SHIFT);
686 if (ret < 0)
687 dout("invalidate_inode_pages2_range returned %d\n", ret);
688
689 flags = CEPH_OSD_FLAG_ORDERSNAP |
690 CEPH_OSD_FLAG_ONDISK |
691 CEPH_OSD_FLAG_WRITE |
692 CEPH_OSD_FLAG_ACK;
693
Al Viro4908b822014-04-03 23:09:01 -0400694 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +0800695 size_t left;
696 int n;
697
698 snapc = ci->i_snap_realm->cached_context;
699 vino = ceph_vino(inode);
700 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
701 vino, pos, &len, 1,
702 CEPH_OSD_OP_WRITE, flags, snapc,
703 ci->i_truncate_seq,
704 ci->i_truncate_size,
705 false);
706 if (IS_ERR(req)) {
707 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400708 break;
majianpenge8344e62013-09-12 13:54:26 +0800709 }
710
711 /*
712 * write from beginning of first page,
713 * regardless of io alignment
714 */
715 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
716
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700717 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700718 if (IS_ERR(pages)) {
719 ret = PTR_ERR(pages);
720 goto out;
721 }
majianpenge8344e62013-09-12 13:54:26 +0800722
723 left = len;
724 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +0200725 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -0400726 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +0800727 if (ret != plen) {
728 ret = -EFAULT;
729 break;
730 }
731 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +0800732 }
733
Sage Weil124e68e2009-10-06 11:31:08 -0700734 if (ret < 0) {
735 ceph_release_page_vector(pages, num_pages);
736 goto out;
737 }
738
majianpenge8344e62013-09-12 13:54:26 +0800739 /* get a second commit callback */
740 req->r_unsafe_callback = ceph_sync_write_unsafe;
741 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -0700742
majianpenge8344e62013-09-12 13:54:26 +0800743 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
744 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -0500745
majianpenge8344e62013-09-12 13:54:26 +0800746 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
747 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Sage Weil124e68e2009-10-06 11:31:08 -0700748
majianpenge8344e62013-09-12 13:54:26 +0800749 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
750 if (!ret)
751 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700752
753out:
majianpenge8344e62013-09-12 13:54:26 +0800754 ceph_osdc_put_request(req);
755 if (ret == 0) {
756 pos += len;
757 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -0700758
majianpenge8344e62013-09-12 13:54:26 +0800759 if (pos > i_size_read(inode)) {
760 check_caps = ceph_inode_set_size(inode, pos);
761 if (check_caps)
762 ceph_check_caps(ceph_inode(inode),
763 CHECK_CAPS_AUTHONLY,
764 NULL);
765 }
766 } else
767 break;
768 }
769
770 if (ret != -EOLDSNAPC && written > 0) {
Sage Weil124e68e2009-10-06 11:31:08 -0700771 ret = written;
majianpenge8344e62013-09-12 13:54:26 +0800772 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700773 }
774 return ret;
775}
776
777/*
778 * Wrap generic_file_aio_read with checks for cap bits on the inode.
779 * Atomically grab references, so that those bits are not released
780 * back to the MDS mid-read.
781 *
782 * Hmm, the sync read case isn't actually async... should it be?
783 */
Al Viro36444242014-04-02 20:28:01 -0400784static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -0700785{
786 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700787 struct ceph_file_info *fi = filp->private_data;
majianpeng8eb4efb2013-09-26 14:42:17 +0800788 size_t len = iocb->ki_nbytes;
Al Viro496ad9a2013-01-23 17:07:38 -0500789 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700790 struct ceph_inode_info *ci = ceph_inode(inode);
791 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700792 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800793 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700794
Sage Weil6a026582010-02-09 14:04:02 -0800795again:
majianpeng8eb4efb2013-09-26 14:42:17 +0800796 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
797 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
798
Sage Weil29625072010-05-27 10:40:43 -0700799 if (fi->fmode & CEPH_FILE_MODE_LAZY)
800 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
801 else
802 want = CEPH_CAP_FILE_CACHE;
803 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700804 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800805 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700806
Sage Weil29625072010-05-27 10:40:43 -0700807 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700808 (iocb->ki_filp->f_flags & O_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +0800809 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700810
majianpeng8eb4efb2013-09-26 14:42:17 +0800811 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
812 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
813 ceph_cap_string(got));
814
majianpeng8eb4efb2013-09-26 14:42:17 +0800815 /* hmm, this isn't really async... */
Al Viro36444242014-04-02 20:28:01 -0400816 ret = ceph_sync_read(iocb, to, &checkeof);
majianpeng8eb4efb2013-09-26 14:42:17 +0800817 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +0800818 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -0400819 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +0800820 ceph_cap_string(got));
821
Al Viro36444242014-04-02 20:28:01 -0400822 ret = generic_file_read_iter(iocb, to);
majianpeng8eb4efb2013-09-26 14:42:17 +0800823 }
Sage Weil124e68e2009-10-06 11:31:08 -0700824 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
825 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
826 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800827
828 if (checkeof && ret >= 0) {
majianpeng8eb4efb2013-09-26 14:42:17 +0800829 int statret = ceph_do_getattr(inode,
830 CEPH_STAT_CAP_SIZE);
Sage Weil6a026582010-02-09 14:04:02 -0800831
832 /* hit EOF or hole? */
majianpeng8eb4efb2013-09-26 14:42:17 +0800833 if (statret == 0 && iocb->ki_pos < inode->i_size &&
834 ret < len) {
835 dout("sync_read hit hole, ppos %lld < size %lld"
836 ", reading more\n", iocb->ki_pos,
837 inode->i_size);
838
Al Viro36444242014-04-02 20:28:01 -0400839 iov_iter_advance(to, ret);
Sage Weil6a026582010-02-09 14:04:02 -0800840 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -0800841 len -= ret;
842 checkeof = 0;
843 goto again;
844 }
845 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800846
Sage Weil6a026582010-02-09 14:04:02 -0800847 if (ret >= 0)
848 ret += read;
849
Sage Weil124e68e2009-10-06 11:31:08 -0700850 return ret;
851}
852
853/*
854 * Take cap references to avoid releasing caps to MDS mid-write.
855 *
856 * If we are synchronous, and write with an old snap context, the OSD
857 * may return EOLDSNAPC. In that case, retry the write.. _after_
858 * dropping our cap refs and allowing the pending snap to logically
859 * complete _before_ this write occurs.
860 *
861 * If we are near ENOSPC, write synchronously.
862 */
Al Viro4908b822014-04-03 23:09:01 -0400863static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -0700864{
865 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700866 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -0500867 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700868 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700869 struct ceph_osd_client *osdc =
870 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Al Viro4908b822014-04-03 23:09:01 -0400871 ssize_t count = iov_iter_count(from), written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800872 int err, want, got;
Al Viro4908b822014-04-03 23:09:01 -0400873 loff_t pos = iocb->ki_pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700874
875 if (ceph_snap(inode) != CEPH_NOSNAP)
876 return -EROFS;
877
Yan, Zheng03d254e2013-04-12 16:11:13 +0800878 mutex_lock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800879
Yan, Zheng03d254e2013-04-12 16:11:13 +0800880 /* We can write back this queue in page reclaim */
881 current->backing_dev_info = file->f_mapping->backing_dev_info;
882
883 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
884 if (err)
885 goto out;
886
887 if (count == 0)
888 goto out;
Al Viro4908b822014-04-03 23:09:01 -0400889 iov_iter_truncate(from, count);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800890
891 err = file_remove_suid(file);
892 if (err)
893 goto out;
894
895 err = file_update_time(file);
896 if (err)
897 goto out;
898
Sage Weil124e68e2009-10-06 11:31:08 -0700899retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800900 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800901 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800902 goto out;
903 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800904
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700905 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800906 inode, ceph_vinop(inode), pos, count, inode->i_size);
Sage Weil7971bd92013-05-01 21:15:58 -0700907 if (fi->fmode & CEPH_FILE_MODE_LAZY)
908 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
909 else
910 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800911 got = 0;
912 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count);
913 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +0800914 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -0700915
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700916 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800917 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -0700918
919 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
majianpenge8344e62013-09-12 13:54:26 +0800920 (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Al Viro4908b822014-04-03 23:09:01 -0400921 struct iov_iter data;
Yan, Zheng37505d52013-04-12 16:11:10 +0800922 mutex_unlock(&inode->i_mutex);
Al Viro4908b822014-04-03 23:09:01 -0400923 /* we might need to revert back to that point */
924 data = *from;
majianpenge8344e62013-09-12 13:54:26 +0800925 if (file->f_flags & O_DIRECT)
Yan, Zheng06fee302014-07-28 14:33:46 +0800926 written = ceph_sync_direct_write(iocb, &data, pos);
majianpenge8344e62013-09-12 13:54:26 +0800927 else
Yan, Zheng06fee302014-07-28 14:33:46 +0800928 written = ceph_sync_write(iocb, &data, pos);
majianpeng0e5dd452013-08-08 15:32:19 +0800929 if (written == -EOLDSNAPC) {
930 dout("aio_write %p %llx.%llx %llu~%u"
931 "got EOLDSNAPC, retrying\n",
932 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -0400933 pos, (unsigned)count);
majianpeng0e5dd452013-08-08 15:32:19 +0800934 mutex_lock(&inode->i_mutex);
majianpeng0e5dd452013-08-08 15:32:19 +0800935 goto retry_snap;
936 }
Al Viro4908b822014-04-03 23:09:01 -0400937 if (written > 0)
938 iov_iter_advance(from, written);
Sage Weil7971bd92013-05-01 21:15:58 -0700939 } else {
Yunchuan Wen32d3e142013-12-26 06:29:27 -0800940 loff_t old_size = inode->i_size;
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700941 /*
942 * No need to acquire the i_truncate_mutex. Because
943 * the MDS revokes Fwb caps before sending truncate
944 * message to us. We can't get Fwb cap while there
945 * are pending vmtruncate. So write and vmtruncate
946 * can not run at the same time
947 */
Al Viro4908b822014-04-03 23:09:01 -0400948 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -0500949 if (likely(written >= 0))
950 iocb->ki_pos = pos + written;
Yunchuan Wen32d3e142013-12-26 06:29:27 -0800951 if (inode->i_size > old_size)
952 ceph_fscache_update_objectsize(inode);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800953 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -0700954 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700955
Yan, Zheng03d254e2013-04-12 16:11:13 +0800956 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -0700957 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -0800958 spin_lock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700959 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800960 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700961 if (dirty)
962 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -0700963 }
Sage Weil7971bd92013-05-01 21:15:58 -0700964
Sage Weil124e68e2009-10-06 11:31:08 -0700965 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -0400966 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -0700967 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -0700968 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -0700969
Yan, Zheng03d254e2013-04-12 16:11:13 +0800970 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800971 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
972 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800973 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800974 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +0800975 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800976 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800977
Sage Weil2f75e9e2013-08-09 09:57:58 -0700978 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -0700979
Sage Weil2f75e9e2013-08-09 09:57:58 -0700980out:
981 mutex_unlock(&inode->i_mutex);
982out_unlocked:
983 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800984 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -0700985}
986
987/*
988 * llseek. be sure to verify file size on SEEK_END.
989 */
Andrew Morton965c8e52012-12-17 15:59:39 -0800990static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -0700991{
992 struct inode *inode = file->f_mapping->host;
993 int ret;
994
995 mutex_lock(&inode->i_mutex);
Sage Weil6a82c472011-12-13 09:19:26 -0800996
Andrew Morton965c8e52012-12-17 15:59:39 -0800997 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Sage Weil124e68e2009-10-06 11:31:08 -0700998 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE);
999 if (ret < 0) {
1000 offset = ret;
1001 goto out;
1002 }
Josef Bacik06222e42011-07-18 13:21:38 -04001003 }
1004
Andrew Morton965c8e52012-12-17 15:59:39 -08001005 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001006 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -07001007 offset += inode->i_size;
1008 break;
1009 case SEEK_CUR:
1010 /*
1011 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1012 * position-querying operation. Avoid rewriting the "same"
1013 * f_pos value back to the file because a concurrent read(),
1014 * write() or lseek() might have altered it
1015 */
1016 if (offset == 0) {
1017 offset = file->f_pos;
1018 goto out;
1019 }
1020 offset += file->f_pos;
1021 break;
Josef Bacik06222e42011-07-18 13:21:38 -04001022 case SEEK_DATA:
1023 if (offset >= inode->i_size) {
1024 ret = -ENXIO;
1025 goto out;
1026 }
1027 break;
1028 case SEEK_HOLE:
1029 if (offset >= inode->i_size) {
1030 ret = -ENXIO;
1031 goto out;
1032 }
1033 offset = inode->i_size;
1034 break;
Sage Weil124e68e2009-10-06 11:31:08 -07001035 }
1036
Jie Liu46a1c2c2013-06-25 12:02:13 +08001037 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001038
1039out:
1040 mutex_unlock(&inode->i_mutex);
1041 return offset;
1042}
1043
Li Wangad7a60d2013-08-15 11:51:44 +08001044static inline void ceph_zero_partial_page(
1045 struct inode *inode, loff_t offset, unsigned size)
1046{
1047 struct page *page;
1048 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1049
1050 page = find_lock_page(inode->i_mapping, index);
1051 if (page) {
1052 wait_on_page_writeback(page);
1053 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1054 unlock_page(page);
1055 page_cache_release(page);
1056 }
1057}
1058
1059static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1060 loff_t length)
1061{
1062 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1063 if (offset < nearly) {
1064 loff_t size = nearly - offset;
1065 if (length < size)
1066 size = length;
1067 ceph_zero_partial_page(inode, offset, size);
1068 offset += size;
1069 length -= size;
1070 }
1071 if (length >= PAGE_CACHE_SIZE) {
1072 loff_t size = round_down(length, PAGE_CACHE_SIZE);
1073 truncate_pagecache_range(inode, offset, offset + size - 1);
1074 offset += size;
1075 length -= size;
1076 }
1077 if (length)
1078 ceph_zero_partial_page(inode, offset, length);
1079}
1080
1081static int ceph_zero_partial_object(struct inode *inode,
1082 loff_t offset, loff_t *length)
1083{
1084 struct ceph_inode_info *ci = ceph_inode(inode);
1085 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1086 struct ceph_osd_request *req;
1087 int ret = 0;
1088 loff_t zero = 0;
1089 int op;
1090
1091 if (!length) {
1092 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1093 length = &zero;
1094 } else {
1095 op = CEPH_OSD_OP_ZERO;
1096 }
1097
1098 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1099 ceph_vino(inode),
1100 offset, length,
1101 1, op,
1102 CEPH_OSD_FLAG_WRITE |
1103 CEPH_OSD_FLAG_ONDISK,
1104 NULL, 0, 0, false);
1105 if (IS_ERR(req)) {
1106 ret = PTR_ERR(req);
1107 goto out;
1108 }
1109
1110 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1111 &inode->i_mtime);
1112
1113 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1114 if (!ret) {
1115 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1116 if (ret == -ENOENT)
1117 ret = 0;
1118 }
1119 ceph_osdc_put_request(req);
1120
1121out:
1122 return ret;
1123}
1124
1125static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1126{
1127 int ret = 0;
1128 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilb314a902013-08-27 12:15:16 -07001129 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1130 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1131 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1132 u64 object_set_size = object_size * stripe_count;
1133 u64 nearly, t;
Li Wangad7a60d2013-08-15 11:51:44 +08001134
Sage Weilb314a902013-08-27 12:15:16 -07001135 /* round offset up to next period boundary */
1136 nearly = offset + object_set_size - 1;
1137 t = nearly;
1138 nearly -= do_div(t, object_set_size);
1139
Li Wangad7a60d2013-08-15 11:51:44 +08001140 while (length && offset < nearly) {
1141 loff_t size = length;
1142 ret = ceph_zero_partial_object(inode, offset, &size);
1143 if (ret < 0)
1144 return ret;
1145 offset += size;
1146 length -= size;
1147 }
1148 while (length >= object_set_size) {
1149 int i;
1150 loff_t pos = offset;
1151 for (i = 0; i < stripe_count; ++i) {
1152 ret = ceph_zero_partial_object(inode, pos, NULL);
1153 if (ret < 0)
1154 return ret;
1155 pos += stripe_unit;
1156 }
1157 offset += object_set_size;
1158 length -= object_set_size;
1159 }
1160 while (length) {
1161 loff_t size = length;
1162 ret = ceph_zero_partial_object(inode, offset, &size);
1163 if (ret < 0)
1164 return ret;
1165 offset += size;
1166 length -= size;
1167 }
1168 return ret;
1169}
1170
1171static long ceph_fallocate(struct file *file, int mode,
1172 loff_t offset, loff_t length)
1173{
1174 struct ceph_file_info *fi = file->private_data;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001175 struct inode *inode = file_inode(file);
Li Wangad7a60d2013-08-15 11:51:44 +08001176 struct ceph_inode_info *ci = ceph_inode(inode);
1177 struct ceph_osd_client *osdc =
1178 &ceph_inode_to_client(inode)->client->osdc;
1179 int want, got = 0;
1180 int dirty;
1181 int ret = 0;
1182 loff_t endoff = 0;
1183 loff_t size;
1184
Yan, Zheng494d77b2014-06-26 15:25:17 +08001185 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1186 return -EOPNOTSUPP;
1187
Li Wangad7a60d2013-08-15 11:51:44 +08001188 if (!S_ISREG(inode->i_mode))
1189 return -EOPNOTSUPP;
1190
Li Wangad7a60d2013-08-15 11:51:44 +08001191 mutex_lock(&inode->i_mutex);
1192
1193 if (ceph_snap(inode) != CEPH_NOSNAP) {
1194 ret = -EROFS;
1195 goto unlock;
1196 }
1197
1198 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1199 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1200 ret = -ENOSPC;
1201 goto unlock;
1202 }
1203
1204 size = i_size_read(inode);
1205 if (!(mode & FALLOC_FL_KEEP_SIZE))
1206 endoff = offset + length;
1207
1208 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1209 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1210 else
1211 want = CEPH_CAP_FILE_BUFFER;
1212
1213 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
1214 if (ret < 0)
1215 goto unlock;
1216
1217 if (mode & FALLOC_FL_PUNCH_HOLE) {
1218 if (offset < size)
1219 ceph_zero_pagecache_range(inode, offset, length);
1220 ret = ceph_zero_objects(inode, offset, length);
1221 } else if (endoff > size) {
1222 truncate_pagecache_range(inode, size, -1);
1223 if (ceph_inode_set_size(inode, endoff))
1224 ceph_check_caps(ceph_inode(inode),
1225 CHECK_CAPS_AUTHONLY, NULL);
1226 }
1227
1228 if (!ret) {
1229 spin_lock(&ci->i_ceph_lock);
1230 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1231 spin_unlock(&ci->i_ceph_lock);
1232 if (dirty)
1233 __mark_inode_dirty(inode, dirty);
1234 }
1235
1236 ceph_put_cap_refs(ci, got);
1237unlock:
1238 mutex_unlock(&inode->i_mutex);
1239 return ret;
1240}
1241
Sage Weil124e68e2009-10-06 11:31:08 -07001242const struct file_operations ceph_file_fops = {
1243 .open = ceph_open,
1244 .release = ceph_release,
1245 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001246 .read = new_sync_read,
Al Viro4908b822014-04-03 23:09:01 -04001247 .write = new_sync_write,
Al Viro36444242014-04-02 20:28:01 -04001248 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001249 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001250 .mmap = ceph_mmap,
1251 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001252 .lock = ceph_lock,
1253 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -07001254 .splice_read = generic_file_splice_read,
Al Viro3551dd72014-04-05 04:40:12 -04001255 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001256 .unlocked_ioctl = ceph_ioctl,
1257 .compat_ioctl = ceph_ioctl,
Li Wangad7a60d2013-08-15 11:51:44 +08001258 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001259};
1260