blob: d7e0da8366e6a6f4e454ae75e42bcfdb4a063f6e [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;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800238 struct ceph_acls_info acls = {};
Sage Weil124e68e2009-10-06 11:31:08 -0700239 int err;
Sage Weil124e68e2009-10-06 11:31:08 -0700240
Sage Weil5ef50c32012-07-31 11:27:36 -0700241 dout("atomic_open %p dentry %p '%.*s' %s flags %d mode 0%o\n",
242 dir, dentry, dentry->d_name.len, dentry->d_name.name,
243 d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode);
244
245 if (dentry->d_name.len > NAME_MAX)
246 return -ENAMETOOLONG;
247
248 err = ceph_init_dentry(dentry);
249 if (err < 0)
250 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700251
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800252 if (flags & O_CREAT) {
253 err = ceph_pre_init_acls(dir, &mode, &acls);
254 if (err < 0)
255 return err;
256 }
257
Sage Weil124e68e2009-10-06 11:31:08 -0700258 /* do the open */
259 req = prepare_open_request(dir->i_sb, flags, mode);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800260 if (IS_ERR(req)) {
261 err = PTR_ERR(req);
262 goto out_acl;
263 }
Sage Weil124e68e2009-10-06 11:31:08 -0700264 req->r_dentry = dget(dentry);
265 req->r_num_caps = 2;
266 if (flags & O_CREAT) {
267 req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
268 req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800269 if (acls.pagelist) {
270 req->r_pagelist = acls.pagelist;
271 acls.pagelist = NULL;
272 }
Sage Weil124e68e2009-10-06 11:31:08 -0700273 }
274 req->r_locked_dir = dir; /* caller holds dir->i_mutex */
Sage Weilacda7652011-07-26 11:27:48 -0700275 err = ceph_mdsc_do_request(mdsc,
276 (flags & (O_CREAT|O_TRUNC)) ? dir : NULL,
277 req);
Sam Lang79aec9842012-12-19 09:44:23 -1000278 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800279 goto out_req;
Sam Lang79aec9842012-12-19 09:44:23 -1000280
Sage Weil468640e2011-07-26 11:28:11 -0700281 err = ceph_handle_snapdir(req, dentry, err);
Sage Weil5ef50c32012-07-31 11:27:36 -0700282 if (err == 0 && (flags & O_CREAT) && !req->r_reply_info.head->is_dentry)
Sage Weil124e68e2009-10-06 11:31:08 -0700283 err = ceph_handle_notrace_create(dir, dentry);
Sage Weil5ef50c32012-07-31 11:27:36 -0700284
285 if (d_unhashed(dentry)) {
286 dn = ceph_finish_lookup(req, dentry, err);
287 if (IS_ERR(dn))
288 err = PTR_ERR(dn);
289 } else {
290 /* we were given a hashed negative dentry */
291 dn = NULL;
292 }
Sage Weil468640e2011-07-26 11:28:11 -0700293 if (err)
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800294 goto out_req;
Sage Weil5ef50c32012-07-31 11:27:36 -0700295 if (dn || dentry->d_inode == NULL || S_ISLNK(dentry->d_inode->i_mode)) {
296 /* make vfs retry on splice, ENOENT, or symlink */
297 dout("atomic_open finish_no_open on dn %p\n", dn);
298 err = finish_no_open(file, dn);
299 } else {
300 dout("atomic_open finish_open on dn %p\n", dn);
Sam Lang6e8575f2012-12-28 09:56:46 -0800301 if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) {
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800302 ceph_init_inode_acls(dentry->d_inode, &acls);
Sam Lang6e8575f2012-12-28 09:56:46 -0800303 *opened |= FILE_CREATED;
304 }
Sage Weil5ef50c32012-07-31 11:27:36 -0700305 err = finish_open(file, dentry, ceph_open, opened);
306 }
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800307out_req:
Yan, Zhengab866542014-04-01 20:20:17 +0800308 if (!req->r_err && req->r_target_inode)
309 ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode);
Sage Weil124e68e2009-10-06 11:31:08 -0700310 ceph_mdsc_put_request(req);
Yan, Zhengb1ee94a2014-09-16 20:35:17 +0800311out_acl:
312 ceph_release_acls_info(&acls);
Sage Weil5ef50c32012-07-31 11:27:36 -0700313 dout("atomic_open result=%d\n", err);
Al Virod9585272012-06-22 12:39:14 +0400314 return err;
Sage Weil124e68e2009-10-06 11:31:08 -0700315}
316
317int ceph_release(struct inode *inode, struct file *file)
318{
319 struct ceph_inode_info *ci = ceph_inode(inode);
320 struct ceph_file_info *cf = file->private_data;
321
322 dout("release inode %p file %p\n", inode, file);
323 ceph_put_fmode(ci, cf->fmode);
324 if (cf->last_readdir)
325 ceph_mdsc_put_request(cf->last_readdir);
326 kfree(cf->last_name);
327 kfree(cf->dir_info);
328 dput(cf->dentry);
329 kmem_cache_free(ceph_file_cachep, cf);
Sage Weil195d3ce2010-03-01 09:57:54 -0800330
331 /* wake up anyone waiting for caps on this inode */
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700332 wake_up_all(&ci->i_cap_wq);
Sage Weil124e68e2009-10-06 11:31:08 -0700333 return 0;
334}
335
336/*
Sage Weil124e68e2009-10-06 11:31:08 -0700337 * Read a range of bytes striped over one or more objects. Iterate over
338 * objects we stripe over. (That's not atomic, but good enough for now.)
339 *
340 * If we get a short result from the OSD, check against i_size; we need to
341 * only return a short read to the caller if we hit EOF.
342 */
343static int striped_read(struct inode *inode,
344 u64 off, u64 len,
Sage Weil6a026582010-02-09 14:04:02 -0800345 struct page **pages, int num_pages,
Sage Weilc3cd6282011-06-01 16:08:44 -0700346 int *checkeof, bool o_direct,
Henry C Changab226e22010-12-15 20:41:54 -0800347 unsigned long buf_align)
Sage Weil124e68e2009-10-06 11:31:08 -0700348{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700349 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Sage Weil124e68e2009-10-06 11:31:08 -0700350 struct ceph_inode_info *ci = ceph_inode(inode);
Dan Carpenter688bac42013-07-23 16:48:01 +0300351 u64 pos, this_len, left;
Sage Weilb7495fc2010-11-09 12:43:12 -0800352 int io_align, page_align;
Dan Carpenter688bac42013-07-23 16:48:01 +0300353 int pages_left;
Sage Weil124e68e2009-10-06 11:31:08 -0700354 int read;
355 struct page **page_pos;
356 int ret;
357 bool hit_stripe, was_short;
358
359 /*
360 * we may need to do multiple reads. not atomic, unfortunately.
361 */
362 pos = off;
363 left = len;
364 page_pos = pages;
365 pages_left = num_pages;
366 read = 0;
Sage Weilb7495fc2010-11-09 12:43:12 -0800367 io_align = off & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700368
369more:
Sage Weilc3cd6282011-06-01 16:08:44 -0700370 if (o_direct)
Henry C Changab226e22010-12-15 20:41:54 -0800371 page_align = (pos - io_align + buf_align) & ~PAGE_MASK;
Sage Weilb7495fc2010-11-09 12:43:12 -0800372 else
373 page_align = pos & ~PAGE_MASK;
Sage Weil124e68e2009-10-06 11:31:08 -0700374 this_len = left;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700375 ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode),
Sage Weil124e68e2009-10-06 11:31:08 -0700376 &ci->i_layout, pos, &this_len,
377 ci->i_truncate_seq,
378 ci->i_truncate_size,
Sage Weilb7495fc2010-11-09 12:43:12 -0800379 page_pos, pages_left, page_align);
Sage Weil124e68e2009-10-06 11:31:08 -0700380 if (ret == -ENOENT)
381 ret = 0;
Sage Weil0e987282011-06-07 20:40:35 -0700382 hit_stripe = this_len < left;
383 was_short = ret >= 0 && ret < this_len;
Dan Carpenter688bac42013-07-23 16:48:01 +0300384 dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read,
Sage Weil124e68e2009-10-06 11:31:08 -0700385 ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : "");
386
majianpeng02ae66d2013-08-06 16:20:38 +0800387 if (ret >= 0) {
388 int didpages;
389 if (was_short && (pos + ret < inode->i_size)) {
390 u64 tmp = min(this_len - ret,
391 inode->i_size - pos - ret);
392 dout(" zero gap %llu to %llu\n",
393 pos + ret, pos + ret + tmp);
394 ceph_zero_page_vector_range(page_align + read + ret,
395 tmp, pages);
396 ret += tmp;
Sage Weil124e68e2009-10-06 11:31:08 -0700397 }
majianpeng02ae66d2013-08-06 16:20:38 +0800398
399 didpages = (page_align + ret) >> PAGE_CACHE_SHIFT;
Sage Weil124e68e2009-10-06 11:31:08 -0700400 pos += ret;
401 read = pos - off;
402 left -= ret;
403 page_pos += didpages;
404 pages_left -= didpages;
405
majianpeng02ae66d2013-08-06 16:20:38 +0800406 /* hit stripe and need continue*/
407 if (left && hit_stripe && pos < inode->i_size)
Sage Weil124e68e2009-10-06 11:31:08 -0700408 goto more;
409 }
410
majianpengee7289b2013-08-21 15:02:51 +0800411 if (read > 0) {
majianpeng02ae66d2013-08-06 16:20:38 +0800412 ret = read;
Sage Weilc3cd6282011-06-01 16:08:44 -0700413 /* did we bounce off eof? */
414 if (pos + left > inode->i_size)
415 *checkeof = 1;
Sage Weil124e68e2009-10-06 11:31:08 -0700416 }
417
Sage Weil124e68e2009-10-06 11:31:08 -0700418 dout("striped_read returns %d\n", ret);
419 return ret;
420}
421
422/*
423 * Completely synchronous read and write methods. Direct from __user
424 * buffer to osd, or directly to user pages (if O_DIRECT).
425 *
426 * If the read spans object boundary, just do multiple reads.
427 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800428static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i,
429 int *checkeof)
Sage Weil124e68e2009-10-06 11:31:08 -0700430{
majianpeng8eb4efb2013-09-26 14:42:17 +0800431 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500432 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700433 struct page **pages;
majianpeng8eb4efb2013-09-26 14:42:17 +0800434 u64 off = iocb->ki_pos;
Henry C Changab226e22010-12-15 20:41:54 -0800435 int num_pages, ret;
Al Viro2b777c92014-04-03 22:31:22 -0400436 size_t len = iov_iter_count(i);
Sage Weil124e68e2009-10-06 11:31:08 -0700437
majianpeng8eb4efb2013-09-26 14:42:17 +0800438 dout("sync_read on file %p %llu~%u %s\n", file, off,
439 (unsigned)len,
Sage Weil124e68e2009-10-06 11:31:08 -0700440 (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
Yan, Zhengd0d0db22014-07-21 10:15:48 +0800441
442 if (!len)
443 return 0;
Sage Weile98b6fe2010-11-09 12:24:53 -0800444 /*
445 * flush any page cache pages in this range. this
446 * will make concurrent normal and sync io slow,
447 * but it will at least behave sensibly when they are
448 * in sequence.
449 */
majianpeng8eb4efb2013-09-26 14:42:17 +0800450 ret = filemap_write_and_wait_range(inode->i_mapping, off,
451 off + len);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800452 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800453 return ret;
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800454
majianpeng8eb4efb2013-09-26 14:42:17 +0800455 if (file->f_flags & O_DIRECT) {
456 while (iov_iter_count(i)) {
Al Viro2b777c92014-04-03 22:31:22 -0400457 size_t start;
458 ssize_t n;
Sage Weil124e68e2009-10-06 11:31:08 -0700459
Al Viro2b777c92014-04-03 22:31:22 -0400460 n = iov_iter_get_pages_alloc(i, &pages, INT_MAX, &start);
461 if (n < 0)
462 return n;
Sage Weil124e68e2009-10-06 11:31:08 -0700463
Al Viro2b777c92014-04-03 22:31:22 -0400464 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
465
466 ret = striped_read(inode, off, n,
majianpeng8eb4efb2013-09-26 14:42:17 +0800467 pages, num_pages, checkeof,
Al Viro2b777c92014-04-03 22:31:22 -0400468 1, start);
469
majianpeng8eb4efb2013-09-26 14:42:17 +0800470 ceph_put_page_vector(pages, num_pages, true);
471
472 if (ret <= 0)
473 break;
474 off += ret;
475 iov_iter_advance(i, ret);
Al Viro2b777c92014-04-03 22:31:22 -0400476 if (ret < n)
majianpeng8eb4efb2013-09-26 14:42:17 +0800477 break;
478 }
479 } else {
480 num_pages = calc_pages_for(off, len);
481 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
482 if (IS_ERR(pages))
483 return PTR_ERR(pages);
484 ret = striped_read(inode, off, len, pages,
485 num_pages, checkeof, 0, 0);
486 if (ret > 0) {
487 int l, k = 0;
Al Viro2b777c92014-04-03 22:31:22 -0400488 size_t left = ret;
majianpeng8eb4efb2013-09-26 14:42:17 +0800489
490 while (left) {
Yan, Zheng5aaa4322014-07-02 10:13:00 +0800491 size_t page_off = off & ~PAGE_MASK;
492 size_t copy = min_t(size_t,
493 PAGE_SIZE - page_off, left);
494 l = copy_page_to_iter(pages[k++], page_off,
495 copy, i);
Al Viro2b777c92014-04-03 22:31:22 -0400496 off += l;
497 left -= l;
498 if (l < copy)
majianpeng8eb4efb2013-09-26 14:42:17 +0800499 break;
500 }
501 }
Sage Weil124e68e2009-10-06 11:31:08 -0700502 ceph_release_page_vector(pages, num_pages);
majianpeng8eb4efb2013-09-26 14:42:17 +0800503 }
504
505 if (off > iocb->ki_pos) {
506 ret = off - iocb->ki_pos;
507 iocb->ki_pos = off;
508 }
509
Sage Weil124e68e2009-10-06 11:31:08 -0700510 dout("sync_read result %d\n", ret);
511 return ret;
512}
513
514/*
Alex Elder26be8802013-04-15 11:20:42 -0500515 * Write commit request unsafe callback, called to tell us when a
516 * request is unsafe (that is, in flight--has been handed to the
517 * messenger to send to its target osd). It is called again when
518 * we've received a response message indicating the request is
519 * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request
520 * is completed early (and unsuccessfully) due to a timeout or
521 * interrupt.
522 *
523 * This is used if we requested both an ACK and ONDISK commit reply
524 * from the OSD.
Sage Weil124e68e2009-10-06 11:31:08 -0700525 */
Alex Elder26be8802013-04-15 11:20:42 -0500526static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe)
Sage Weil124e68e2009-10-06 11:31:08 -0700527{
528 struct ceph_inode_info *ci = ceph_inode(req->r_inode);
529
Alex Elder26be8802013-04-15 11:20:42 -0500530 dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid,
531 unsafe ? "un" : "");
532 if (unsafe) {
533 ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR);
534 spin_lock(&ci->i_unsafe_lock);
535 list_add_tail(&req->r_unsafe_item,
536 &ci->i_unsafe_writes);
537 spin_unlock(&ci->i_unsafe_lock);
538 } else {
539 spin_lock(&ci->i_unsafe_lock);
540 list_del_init(&req->r_unsafe_item);
541 spin_unlock(&ci->i_unsafe_lock);
542 ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR);
543 }
Sage Weil124e68e2009-10-06 11:31:08 -0700544}
545
majianpenge8344e62013-09-12 13:54:26 +0800546
Sage Weil124e68e2009-10-06 11:31:08 -0700547/*
majianpenge8344e62013-09-12 13:54:26 +0800548 * Synchronous write, straight from __user pointer or user pages.
Sage Weil124e68e2009-10-06 11:31:08 -0700549 *
550 * If write spans object boundary, just do multiple writes. (For a
551 * correct atomic write, we should e.g. take write locks on all
552 * objects, rollback on failure, etc.)
553 */
majianpenge8344e62013-09-12 13:54:26 +0800554static ssize_t
Yan, Zheng06fee302014-07-28 14:33:46 +0800555ceph_sync_direct_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
Sage Weil124e68e2009-10-06 11:31:08 -0700556{
majianpenge8344e62013-09-12 13:54:26 +0800557 struct file *file = iocb->ki_filp;
Al Viro496ad9a2013-01-23 17:07:38 -0500558 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700559 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700560 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
Alex Elderacead002013-03-14 14:09:05 -0500561 struct ceph_snap_context *snapc;
562 struct ceph_vino vino;
Sage Weil124e68e2009-10-06 11:31:08 -0700563 struct ceph_osd_request *req;
564 struct page **pages;
565 int num_pages;
Sage Weil124e68e2009-10-06 11:31:08 -0700566 int written = 0;
567 int flags;
Sage Weil124e68e2009-10-06 11:31:08 -0700568 int check_caps = 0;
569 int ret;
570 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400571 size_t count = iov_iter_count(from);
Sage Weil124e68e2009-10-06 11:31:08 -0700572
Al Viro496ad9a2013-01-23 17:07:38 -0500573 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
Sage Weil124e68e2009-10-06 11:31:08 -0700574 return -EROFS;
575
majianpenge8344e62013-09-12 13:54:26 +0800576 dout("sync_direct_write on file %p %lld~%u\n", file, pos,
577 (unsigned)count);
Sage Weil124e68e2009-10-06 11:31:08 -0700578
majianpenge8344e62013-09-12 13:54:26 +0800579 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800580 if (ret < 0)
581 return ret;
582
583 ret = invalidate_inode_pages2_range(inode->i_mapping,
584 pos >> PAGE_CACHE_SHIFT,
majianpenge8344e62013-09-12 13:54:26 +0800585 (pos + count) >> PAGE_CACHE_SHIFT);
Yehuda Sadeh29065a52010-02-09 11:14:41 -0800586 if (ret < 0)
587 dout("invalidate_inode_pages2_range returned %d\n", ret);
588
Sage Weil124e68e2009-10-06 11:31:08 -0700589 flags = CEPH_OSD_FLAG_ORDERSNAP |
590 CEPH_OSD_FLAG_ONDISK |
591 CEPH_OSD_FLAG_WRITE;
Sage Weil124e68e2009-10-06 11:31:08 -0700592
Al Viro4908b822014-04-03 23:09:01 -0400593 while (iov_iter_count(from) > 0) {
594 u64 len = iov_iter_single_seg_count(from);
Al Viro64c31312014-04-03 22:58:25 -0400595 size_t start;
596 ssize_t n;
majianpenge8344e62013-09-12 13:54:26 +0800597
598 snapc = ci->i_snap_realm->cached_context;
599 vino = ceph_vino(inode);
600 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
601 vino, pos, &len,
602 2,/*include a 'startsync' command*/
603 CEPH_OSD_OP_WRITE, flags, snapc,
604 ci->i_truncate_seq,
605 ci->i_truncate_size,
606 false);
607 if (IS_ERR(req)) {
608 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400609 break;
majianpenge8344e62013-09-12 13:54:26 +0800610 }
611
Al Viro4908b822014-04-03 23:09:01 -0400612 n = iov_iter_get_pages_alloc(from, &pages, len, &start);
Al Viro64c31312014-04-03 22:58:25 -0400613 if (unlikely(n < 0)) {
614 ret = n;
615 ceph_osdc_put_request(req);
616 break;
Sage Weil124e68e2009-10-06 11:31:08 -0700617 }
618
Al Viro64c31312014-04-03 22:58:25 -0400619 num_pages = (n + start + PAGE_SIZE - 1) / PAGE_SIZE;
Sage Weil124e68e2009-10-06 11:31:08 -0700620 /*
621 * throw out any page cache pages in this range. this
622 * may block.
623 */
Sage Weil213c99e2010-08-03 10:25:11 -0700624 truncate_inode_pages_range(inode->i_mapping, pos,
Al Viro64c31312014-04-03 22:58:25 -0400625 (pos+n) | (PAGE_CACHE_SIZE-1));
626 osd_req_op_extent_osd_data_pages(req, 0, pages, n, start,
majianpenge8344e62013-09-12 13:54:26 +0800627 false, false);
628
629 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
630 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
631
632 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
633 if (!ret)
634 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
635
636 ceph_put_page_vector(pages, num_pages, false);
637
majianpenge8344e62013-09-12 13:54:26 +0800638 ceph_osdc_put_request(req);
Al Viro64c31312014-04-03 22:58:25 -0400639 if (ret)
majianpenge8344e62013-09-12 13:54:26 +0800640 break;
Al Viro64c31312014-04-03 22:58:25 -0400641 pos += n;
642 written += n;
Al Viro4908b822014-04-03 23:09:01 -0400643 iov_iter_advance(from, n);
Al Viro64c31312014-04-03 22:58:25 -0400644
645 if (pos > i_size_read(inode)) {
646 check_caps = ceph_inode_set_size(inode, pos);
647 if (check_caps)
648 ceph_check_caps(ceph_inode(inode),
649 CHECK_CAPS_AUTHONLY,
650 NULL);
651 }
majianpenge8344e62013-09-12 13:54:26 +0800652 }
653
654 if (ret != -EOLDSNAPC && written > 0) {
655 iocb->ki_pos = pos;
656 ret = written;
657 }
658 return ret;
659}
660
661
662/*
663 * Synchronous write, straight from __user pointer or user pages.
664 *
665 * If write spans object boundary, just do multiple writes. (For a
666 * correct atomic write, we should e.g. take write locks on all
667 * objects, rollback on failure, etc.)
668 */
Yan, Zheng06fee302014-07-28 14:33:46 +0800669static ssize_t
670ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos)
majianpenge8344e62013-09-12 13:54:26 +0800671{
672 struct file *file = iocb->ki_filp;
673 struct inode *inode = file_inode(file);
674 struct ceph_inode_info *ci = ceph_inode(inode);
675 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
676 struct ceph_snap_context *snapc;
677 struct ceph_vino vino;
678 struct ceph_osd_request *req;
679 struct page **pages;
680 u64 len;
681 int num_pages;
682 int written = 0;
683 int flags;
684 int check_caps = 0;
685 int ret;
686 struct timespec mtime = CURRENT_TIME;
Al Viro4908b822014-04-03 23:09:01 -0400687 size_t count = iov_iter_count(from);
majianpenge8344e62013-09-12 13:54:26 +0800688
689 if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
690 return -EROFS;
691
692 dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count);
693
694 ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count);
695 if (ret < 0)
696 return ret;
697
698 ret = invalidate_inode_pages2_range(inode->i_mapping,
699 pos >> PAGE_CACHE_SHIFT,
700 (pos + count) >> PAGE_CACHE_SHIFT);
701 if (ret < 0)
702 dout("invalidate_inode_pages2_range returned %d\n", ret);
703
704 flags = CEPH_OSD_FLAG_ORDERSNAP |
705 CEPH_OSD_FLAG_ONDISK |
706 CEPH_OSD_FLAG_WRITE |
707 CEPH_OSD_FLAG_ACK;
708
Al Viro4908b822014-04-03 23:09:01 -0400709 while ((len = iov_iter_count(from)) > 0) {
majianpenge8344e62013-09-12 13:54:26 +0800710 size_t left;
711 int n;
712
713 snapc = ci->i_snap_realm->cached_context;
714 vino = ceph_vino(inode);
715 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
716 vino, pos, &len, 1,
717 CEPH_OSD_OP_WRITE, flags, snapc,
718 ci->i_truncate_seq,
719 ci->i_truncate_size,
720 false);
721 if (IS_ERR(req)) {
722 ret = PTR_ERR(req);
Al Viroeab87232014-04-03 22:44:19 -0400723 break;
majianpenge8344e62013-09-12 13:54:26 +0800724 }
725
726 /*
727 * write from beginning of first page,
728 * regardless of io alignment
729 */
730 num_pages = (len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
731
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700732 pages = ceph_alloc_page_vector(num_pages, GFP_NOFS);
Sage Weil124e68e2009-10-06 11:31:08 -0700733 if (IS_ERR(pages)) {
734 ret = PTR_ERR(pages);
735 goto out;
736 }
majianpenge8344e62013-09-12 13:54:26 +0800737
738 left = len;
739 for (n = 0; n < num_pages; n++) {
Ilya Dryomov125d7252014-01-28 19:16:18 +0200740 size_t plen = min_t(size_t, left, PAGE_SIZE);
Al Viro4908b822014-04-03 23:09:01 -0400741 ret = copy_page_from_iter(pages[n], 0, plen, from);
majianpenge8344e62013-09-12 13:54:26 +0800742 if (ret != plen) {
743 ret = -EFAULT;
744 break;
745 }
746 left -= ret;
majianpenge8344e62013-09-12 13:54:26 +0800747 }
748
Sage Weil124e68e2009-10-06 11:31:08 -0700749 if (ret < 0) {
750 ceph_release_page_vector(pages, num_pages);
751 goto out;
752 }
753
majianpenge8344e62013-09-12 13:54:26 +0800754 /* get a second commit callback */
755 req->r_unsafe_callback = ceph_sync_write_unsafe;
756 req->r_inode = inode;
Sage Weil124e68e2009-10-06 11:31:08 -0700757
majianpenge8344e62013-09-12 13:54:26 +0800758 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
759 false, true);
Alex Elder02ee07d2013-03-14 14:09:06 -0500760
majianpenge8344e62013-09-12 13:54:26 +0800761 /* BUG_ON(vino.snap != CEPH_NOSNAP); */
762 ceph_osdc_build_request(req, pos, snapc, vino.snap, &mtime);
Sage Weil124e68e2009-10-06 11:31:08 -0700763
majianpenge8344e62013-09-12 13:54:26 +0800764 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
765 if (!ret)
766 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
Sage Weil124e68e2009-10-06 11:31:08 -0700767
768out:
majianpenge8344e62013-09-12 13:54:26 +0800769 ceph_osdc_put_request(req);
770 if (ret == 0) {
771 pos += len;
772 written += len;
Sage Weil124e68e2009-10-06 11:31:08 -0700773
majianpenge8344e62013-09-12 13:54:26 +0800774 if (pos > i_size_read(inode)) {
775 check_caps = ceph_inode_set_size(inode, pos);
776 if (check_caps)
777 ceph_check_caps(ceph_inode(inode),
778 CHECK_CAPS_AUTHONLY,
779 NULL);
780 }
781 } else
782 break;
783 }
784
785 if (ret != -EOLDSNAPC && written > 0) {
Sage Weil124e68e2009-10-06 11:31:08 -0700786 ret = written;
majianpenge8344e62013-09-12 13:54:26 +0800787 iocb->ki_pos = pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700788 }
789 return ret;
790}
791
792/*
793 * Wrap generic_file_aio_read with checks for cap bits on the inode.
794 * Atomically grab references, so that those bits are not released
795 * back to the MDS mid-read.
796 *
797 * Hmm, the sync read case isn't actually async... should it be?
798 */
Al Viro36444242014-04-02 20:28:01 -0400799static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
Sage Weil124e68e2009-10-06 11:31:08 -0700800{
801 struct file *filp = iocb->ki_filp;
Sage Weil29625072010-05-27 10:40:43 -0700802 struct ceph_file_info *fi = filp->private_data;
majianpeng8eb4efb2013-09-26 14:42:17 +0800803 size_t len = iocb->ki_nbytes;
Al Viro496ad9a2013-01-23 17:07:38 -0500804 struct inode *inode = file_inode(filp);
Sage Weil124e68e2009-10-06 11:31:08 -0700805 struct ceph_inode_info *ci = ceph_inode(inode);
806 ssize_t ret;
Sage Weil29625072010-05-27 10:40:43 -0700807 int want, got = 0;
Sage Weil6a026582010-02-09 14:04:02 -0800808 int checkeof = 0, read = 0;
Sage Weil124e68e2009-10-06 11:31:08 -0700809
Sage Weil6a026582010-02-09 14:04:02 -0800810again:
majianpeng8eb4efb2013-09-26 14:42:17 +0800811 dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n",
812 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode);
813
Sage Weil29625072010-05-27 10:40:43 -0700814 if (fi->fmode & CEPH_FILE_MODE_LAZY)
815 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
816 else
817 want = CEPH_CAP_FILE_CACHE;
818 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
Sage Weil124e68e2009-10-06 11:31:08 -0700819 if (ret < 0)
majianpeng8eb4efb2013-09-26 14:42:17 +0800820 return ret;
Sage Weil124e68e2009-10-06 11:31:08 -0700821
Sage Weil29625072010-05-27 10:40:43 -0700822 if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 ||
Sage Weil124e68e2009-10-06 11:31:08 -0700823 (iocb->ki_filp->f_flags & O_DIRECT) ||
majianpeng8eb4efb2013-09-26 14:42:17 +0800824 (fi->flags & CEPH_F_SYNC)) {
Sage Weil124e68e2009-10-06 11:31:08 -0700825
majianpeng8eb4efb2013-09-26 14:42:17 +0800826 dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n",
827 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
828 ceph_cap_string(got));
829
majianpeng8eb4efb2013-09-26 14:42:17 +0800830 /* hmm, this isn't really async... */
Al Viro36444242014-04-02 20:28:01 -0400831 ret = ceph_sync_read(iocb, to, &checkeof);
majianpeng8eb4efb2013-09-26 14:42:17 +0800832 } else {
majianpeng8eb4efb2013-09-26 14:42:17 +0800833 dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n",
Al Viro36444242014-04-02 20:28:01 -0400834 inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
majianpeng8eb4efb2013-09-26 14:42:17 +0800835 ceph_cap_string(got));
836
Al Viro36444242014-04-02 20:28:01 -0400837 ret = generic_file_read_iter(iocb, to);
majianpeng8eb4efb2013-09-26 14:42:17 +0800838 }
Sage Weil124e68e2009-10-06 11:31:08 -0700839 dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n",
840 inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret);
841 ceph_put_cap_refs(ci, got);
Sage Weil6a026582010-02-09 14:04:02 -0800842
843 if (checkeof && ret >= 0) {
Yan, Zheng508b32d2014-09-16 21:46:17 +0800844 int statret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Sage Weil6a026582010-02-09 14:04:02 -0800845
846 /* hit EOF or hole? */
majianpeng8eb4efb2013-09-26 14:42:17 +0800847 if (statret == 0 && iocb->ki_pos < inode->i_size &&
848 ret < len) {
849 dout("sync_read hit hole, ppos %lld < size %lld"
850 ", reading more\n", iocb->ki_pos,
851 inode->i_size);
852
Sage Weil6a026582010-02-09 14:04:02 -0800853 read += ret;
Sage Weil6a026582010-02-09 14:04:02 -0800854 len -= ret;
855 checkeof = 0;
856 goto again;
857 }
858 }
majianpeng8eb4efb2013-09-26 14:42:17 +0800859
Sage Weil6a026582010-02-09 14:04:02 -0800860 if (ret >= 0)
861 ret += read;
862
Sage Weil124e68e2009-10-06 11:31:08 -0700863 return ret;
864}
865
866/*
867 * Take cap references to avoid releasing caps to MDS mid-write.
868 *
869 * If we are synchronous, and write with an old snap context, the OSD
870 * may return EOLDSNAPC. In that case, retry the write.. _after_
871 * dropping our cap refs and allowing the pending snap to logically
872 * complete _before_ this write occurs.
873 *
874 * If we are near ENOSPC, write synchronously.
875 */
Al Viro4908b822014-04-03 23:09:01 -0400876static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
Sage Weil124e68e2009-10-06 11:31:08 -0700877{
878 struct file *file = iocb->ki_filp;
Sage Weil33caad32010-05-26 14:31:27 -0700879 struct ceph_file_info *fi = file->private_data;
Al Viro496ad9a2013-01-23 17:07:38 -0500880 struct inode *inode = file_inode(file);
Sage Weil124e68e2009-10-06 11:31:08 -0700881 struct ceph_inode_info *ci = ceph_inode(inode);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700882 struct ceph_osd_client *osdc =
883 &ceph_sb_to_client(inode->i_sb)->client->osdc;
Al Viro4908b822014-04-03 23:09:01 -0400884 ssize_t count = iov_iter_count(from), written = 0;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800885 int err, want, got;
Al Viro4908b822014-04-03 23:09:01 -0400886 loff_t pos = iocb->ki_pos;
Sage Weil124e68e2009-10-06 11:31:08 -0700887
888 if (ceph_snap(inode) != CEPH_NOSNAP)
889 return -EROFS;
890
Yan, Zheng03d254e2013-04-12 16:11:13 +0800891 mutex_lock(&inode->i_mutex);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800892
Yan, Zheng03d254e2013-04-12 16:11:13 +0800893 /* We can write back this queue in page reclaim */
894 current->backing_dev_info = file->f_mapping->backing_dev_info;
895
896 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
897 if (err)
898 goto out;
899
900 if (count == 0)
901 goto out;
Al Viro4908b822014-04-03 23:09:01 -0400902 iov_iter_truncate(from, count);
Yan, Zheng03d254e2013-04-12 16:11:13 +0800903
904 err = file_remove_suid(file);
905 if (err)
906 goto out;
907
908 err = file_update_time(file);
909 if (err)
910 goto out;
911
Sage Weil124e68e2009-10-06 11:31:08 -0700912retry_snap:
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800913 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800914 err = -ENOSPC;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800915 goto out;
916 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800917
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700918 dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800919 inode, ceph_vinop(inode), pos, count, inode->i_size);
Sage Weil7971bd92013-05-01 21:15:58 -0700920 if (fi->fmode & CEPH_FILE_MODE_LAZY)
921 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
922 else
923 want = CEPH_CAP_FILE_BUFFER;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800924 got = 0;
925 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos + count);
926 if (err < 0)
Yan, Zheng37505d52013-04-12 16:11:10 +0800927 goto out;
Sage Weil124e68e2009-10-06 11:31:08 -0700928
Randy Dunlapac7f29b2013-04-19 14:20:07 -0700929 dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n",
Yan, Zheng03d254e2013-04-12 16:11:13 +0800930 inode, ceph_vinop(inode), pos, count, ceph_cap_string(got));
Sage Weil7971bd92013-05-01 21:15:58 -0700931
932 if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 ||
majianpenge8344e62013-09-12 13:54:26 +0800933 (file->f_flags & O_DIRECT) || (fi->flags & CEPH_F_SYNC)) {
Al Viro4908b822014-04-03 23:09:01 -0400934 struct iov_iter data;
Yan, Zheng37505d52013-04-12 16:11:10 +0800935 mutex_unlock(&inode->i_mutex);
Al Viro4908b822014-04-03 23:09:01 -0400936 /* we might need to revert back to that point */
937 data = *from;
majianpenge8344e62013-09-12 13:54:26 +0800938 if (file->f_flags & O_DIRECT)
Yan, Zheng06fee302014-07-28 14:33:46 +0800939 written = ceph_sync_direct_write(iocb, &data, pos);
majianpenge8344e62013-09-12 13:54:26 +0800940 else
Yan, Zheng06fee302014-07-28 14:33:46 +0800941 written = ceph_sync_write(iocb, &data, pos);
majianpeng0e5dd452013-08-08 15:32:19 +0800942 if (written == -EOLDSNAPC) {
943 dout("aio_write %p %llx.%llx %llu~%u"
944 "got EOLDSNAPC, retrying\n",
945 inode, ceph_vinop(inode),
Al Viro4908b822014-04-03 23:09:01 -0400946 pos, (unsigned)count);
majianpeng0e5dd452013-08-08 15:32:19 +0800947 mutex_lock(&inode->i_mutex);
majianpeng0e5dd452013-08-08 15:32:19 +0800948 goto retry_snap;
949 }
Al Viro4908b822014-04-03 23:09:01 -0400950 if (written > 0)
951 iov_iter_advance(from, written);
Sage Weil7971bd92013-05-01 21:15:58 -0700952 } else {
Yunchuan Wen32d3e142013-12-26 06:29:27 -0800953 loff_t old_size = inode->i_size;
Yan, Zhengb0d7c222013-08-12 21:42:15 -0700954 /*
955 * No need to acquire the i_truncate_mutex. Because
956 * the MDS revokes Fwb caps before sending truncate
957 * message to us. We can't get Fwb cap while there
958 * are pending vmtruncate. So write and vmtruncate
959 * can not run at the same time
960 */
Al Viro4908b822014-04-03 23:09:01 -0400961 written = generic_perform_write(file, from, pos);
Al Viroaec605f42014-02-11 22:28:43 -0500962 if (likely(written >= 0))
963 iocb->ki_pos = pos + written;
Yunchuan Wen32d3e142013-12-26 06:29:27 -0800964 if (inode->i_size > old_size)
965 ceph_fscache_update_objectsize(inode);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800966 mutex_unlock(&inode->i_mutex);
Sage Weil124e68e2009-10-06 11:31:08 -0700967 }
Sage Weild8de9ab2011-07-26 11:27:34 -0700968
Yan, Zheng03d254e2013-04-12 16:11:13 +0800969 if (written >= 0) {
Sage Weilfca65b42011-05-04 11:33:47 -0700970 int dirty;
Sage Weilbe655592011-11-30 09:47:09 -0800971 spin_lock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700972 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
Sage Weilbe655592011-11-30 09:47:09 -0800973 spin_unlock(&ci->i_ceph_lock);
Sage Weilfca65b42011-05-04 11:33:47 -0700974 if (dirty)
975 __mark_inode_dirty(inode, dirty);
Sage Weil124e68e2009-10-06 11:31:08 -0700976 }
Sage Weil7971bd92013-05-01 21:15:58 -0700977
Sage Weil124e68e2009-10-06 11:31:08 -0700978 dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n",
Al Viro4908b822014-04-03 23:09:01 -0400979 inode, ceph_vinop(inode), pos, (unsigned)count,
Sage Weil7971bd92013-05-01 21:15:58 -0700980 ceph_cap_string(got));
Sage Weil124e68e2009-10-06 11:31:08 -0700981 ceph_put_cap_refs(ci, got);
Sage Weil7971bd92013-05-01 21:15:58 -0700982
Yan, Zheng03d254e2013-04-12 16:11:13 +0800983 if (written >= 0 &&
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800984 ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) ||
985 ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) {
Yan, Zheng03d254e2013-04-12 16:11:13 +0800986 err = vfs_fsync_range(file, pos, pos + written - 1, 1);
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800987 if (err < 0)
Yan, Zheng03d254e2013-04-12 16:11:13 +0800988 written = err;
Yan, Zheng6070e0c2013-03-01 10:55:39 +0800989 }
Yan, Zheng03d254e2013-04-12 16:11:13 +0800990
Sage Weil2f75e9e2013-08-09 09:57:58 -0700991 goto out_unlocked;
Sage Weil124e68e2009-10-06 11:31:08 -0700992
Sage Weil2f75e9e2013-08-09 09:57:58 -0700993out:
994 mutex_unlock(&inode->i_mutex);
995out_unlocked:
996 current->backing_dev_info = NULL;
Yan, Zheng03d254e2013-04-12 16:11:13 +0800997 return written ? written : err;
Sage Weil124e68e2009-10-06 11:31:08 -0700998}
999
1000/*
1001 * llseek. be sure to verify file size on SEEK_END.
1002 */
Andrew Morton965c8e52012-12-17 15:59:39 -08001003static loff_t ceph_llseek(struct file *file, loff_t offset, int whence)
Sage Weil124e68e2009-10-06 11:31:08 -07001004{
1005 struct inode *inode = file->f_mapping->host;
1006 int ret;
1007
1008 mutex_lock(&inode->i_mutex);
Sage Weil6a82c472011-12-13 09:19:26 -08001009
Andrew Morton965c8e52012-12-17 15:59:39 -08001010 if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) {
Yan, Zheng508b32d2014-09-16 21:46:17 +08001011 ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false);
Sage Weil124e68e2009-10-06 11:31:08 -07001012 if (ret < 0) {
1013 offset = ret;
1014 goto out;
1015 }
Josef Bacik06222e42011-07-18 13:21:38 -04001016 }
1017
Andrew Morton965c8e52012-12-17 15:59:39 -08001018 switch (whence) {
Josef Bacik06222e42011-07-18 13:21:38 -04001019 case SEEK_END:
Sage Weil124e68e2009-10-06 11:31:08 -07001020 offset += inode->i_size;
1021 break;
1022 case SEEK_CUR:
1023 /*
1024 * Here we special-case the lseek(fd, 0, SEEK_CUR)
1025 * position-querying operation. Avoid rewriting the "same"
1026 * f_pos value back to the file because a concurrent read(),
1027 * write() or lseek() might have altered it
1028 */
1029 if (offset == 0) {
1030 offset = file->f_pos;
1031 goto out;
1032 }
1033 offset += file->f_pos;
1034 break;
Josef Bacik06222e42011-07-18 13:21:38 -04001035 case SEEK_DATA:
1036 if (offset >= inode->i_size) {
1037 ret = -ENXIO;
1038 goto out;
1039 }
1040 break;
1041 case SEEK_HOLE:
1042 if (offset >= inode->i_size) {
1043 ret = -ENXIO;
1044 goto out;
1045 }
1046 offset = inode->i_size;
1047 break;
Sage Weil124e68e2009-10-06 11:31:08 -07001048 }
1049
Jie Liu46a1c2c2013-06-25 12:02:13 +08001050 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Sage Weil124e68e2009-10-06 11:31:08 -07001051
1052out:
1053 mutex_unlock(&inode->i_mutex);
1054 return offset;
1055}
1056
Li Wangad7a60d2013-08-15 11:51:44 +08001057static inline void ceph_zero_partial_page(
1058 struct inode *inode, loff_t offset, unsigned size)
1059{
1060 struct page *page;
1061 pgoff_t index = offset >> PAGE_CACHE_SHIFT;
1062
1063 page = find_lock_page(inode->i_mapping, index);
1064 if (page) {
1065 wait_on_page_writeback(page);
1066 zero_user(page, offset & (PAGE_CACHE_SIZE - 1), size);
1067 unlock_page(page);
1068 page_cache_release(page);
1069 }
1070}
1071
1072static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset,
1073 loff_t length)
1074{
1075 loff_t nearly = round_up(offset, PAGE_CACHE_SIZE);
1076 if (offset < nearly) {
1077 loff_t size = nearly - offset;
1078 if (length < size)
1079 size = length;
1080 ceph_zero_partial_page(inode, offset, size);
1081 offset += size;
1082 length -= size;
1083 }
1084 if (length >= PAGE_CACHE_SIZE) {
1085 loff_t size = round_down(length, PAGE_CACHE_SIZE);
1086 truncate_pagecache_range(inode, offset, offset + size - 1);
1087 offset += size;
1088 length -= size;
1089 }
1090 if (length)
1091 ceph_zero_partial_page(inode, offset, length);
1092}
1093
1094static int ceph_zero_partial_object(struct inode *inode,
1095 loff_t offset, loff_t *length)
1096{
1097 struct ceph_inode_info *ci = ceph_inode(inode);
1098 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1099 struct ceph_osd_request *req;
1100 int ret = 0;
1101 loff_t zero = 0;
1102 int op;
1103
1104 if (!length) {
1105 op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE;
1106 length = &zero;
1107 } else {
1108 op = CEPH_OSD_OP_ZERO;
1109 }
1110
1111 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1112 ceph_vino(inode),
1113 offset, length,
1114 1, op,
1115 CEPH_OSD_FLAG_WRITE |
1116 CEPH_OSD_FLAG_ONDISK,
1117 NULL, 0, 0, false);
1118 if (IS_ERR(req)) {
1119 ret = PTR_ERR(req);
1120 goto out;
1121 }
1122
1123 ceph_osdc_build_request(req, offset, NULL, ceph_vino(inode).snap,
1124 &inode->i_mtime);
1125
1126 ret = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1127 if (!ret) {
1128 ret = ceph_osdc_wait_request(&fsc->client->osdc, req);
1129 if (ret == -ENOENT)
1130 ret = 0;
1131 }
1132 ceph_osdc_put_request(req);
1133
1134out:
1135 return ret;
1136}
1137
1138static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length)
1139{
1140 int ret = 0;
1141 struct ceph_inode_info *ci = ceph_inode(inode);
Sage Weilb314a902013-08-27 12:15:16 -07001142 s32 stripe_unit = ceph_file_layout_su(ci->i_layout);
1143 s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout);
1144 s32 object_size = ceph_file_layout_object_size(ci->i_layout);
1145 u64 object_set_size = object_size * stripe_count;
1146 u64 nearly, t;
Li Wangad7a60d2013-08-15 11:51:44 +08001147
Sage Weilb314a902013-08-27 12:15:16 -07001148 /* round offset up to next period boundary */
1149 nearly = offset + object_set_size - 1;
1150 t = nearly;
1151 nearly -= do_div(t, object_set_size);
1152
Li Wangad7a60d2013-08-15 11:51:44 +08001153 while (length && offset < nearly) {
1154 loff_t size = length;
1155 ret = ceph_zero_partial_object(inode, offset, &size);
1156 if (ret < 0)
1157 return ret;
1158 offset += size;
1159 length -= size;
1160 }
1161 while (length >= object_set_size) {
1162 int i;
1163 loff_t pos = offset;
1164 for (i = 0; i < stripe_count; ++i) {
1165 ret = ceph_zero_partial_object(inode, pos, NULL);
1166 if (ret < 0)
1167 return ret;
1168 pos += stripe_unit;
1169 }
1170 offset += object_set_size;
1171 length -= object_set_size;
1172 }
1173 while (length) {
1174 loff_t size = length;
1175 ret = ceph_zero_partial_object(inode, offset, &size);
1176 if (ret < 0)
1177 return ret;
1178 offset += size;
1179 length -= size;
1180 }
1181 return ret;
1182}
1183
1184static long ceph_fallocate(struct file *file, int mode,
1185 loff_t offset, loff_t length)
1186{
1187 struct ceph_file_info *fi = file->private_data;
Libo Chenaa8b60e2013-12-11 13:49:11 +08001188 struct inode *inode = file_inode(file);
Li Wangad7a60d2013-08-15 11:51:44 +08001189 struct ceph_inode_info *ci = ceph_inode(inode);
1190 struct ceph_osd_client *osdc =
1191 &ceph_inode_to_client(inode)->client->osdc;
1192 int want, got = 0;
1193 int dirty;
1194 int ret = 0;
1195 loff_t endoff = 0;
1196 loff_t size;
1197
Yan, Zheng494d77b2014-06-26 15:25:17 +08001198 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1199 return -EOPNOTSUPP;
1200
Li Wangad7a60d2013-08-15 11:51:44 +08001201 if (!S_ISREG(inode->i_mode))
1202 return -EOPNOTSUPP;
1203
Li Wangad7a60d2013-08-15 11:51:44 +08001204 mutex_lock(&inode->i_mutex);
1205
1206 if (ceph_snap(inode) != CEPH_NOSNAP) {
1207 ret = -EROFS;
1208 goto unlock;
1209 }
1210
1211 if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) &&
1212 !(mode & FALLOC_FL_PUNCH_HOLE)) {
1213 ret = -ENOSPC;
1214 goto unlock;
1215 }
1216
1217 size = i_size_read(inode);
1218 if (!(mode & FALLOC_FL_KEEP_SIZE))
1219 endoff = offset + length;
1220
1221 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1222 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1223 else
1224 want = CEPH_CAP_FILE_BUFFER;
1225
1226 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, endoff);
1227 if (ret < 0)
1228 goto unlock;
1229
1230 if (mode & FALLOC_FL_PUNCH_HOLE) {
1231 if (offset < size)
1232 ceph_zero_pagecache_range(inode, offset, length);
1233 ret = ceph_zero_objects(inode, offset, length);
1234 } else if (endoff > size) {
1235 truncate_pagecache_range(inode, size, -1);
1236 if (ceph_inode_set_size(inode, endoff))
1237 ceph_check_caps(ceph_inode(inode),
1238 CHECK_CAPS_AUTHONLY, NULL);
1239 }
1240
1241 if (!ret) {
1242 spin_lock(&ci->i_ceph_lock);
1243 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1244 spin_unlock(&ci->i_ceph_lock);
1245 if (dirty)
1246 __mark_inode_dirty(inode, dirty);
1247 }
1248
1249 ceph_put_cap_refs(ci, got);
1250unlock:
1251 mutex_unlock(&inode->i_mutex);
1252 return ret;
1253}
1254
Sage Weil124e68e2009-10-06 11:31:08 -07001255const struct file_operations ceph_file_fops = {
1256 .open = ceph_open,
1257 .release = ceph_release,
1258 .llseek = ceph_llseek,
Al Viro36444242014-04-02 20:28:01 -04001259 .read = new_sync_read,
Al Viro4908b822014-04-03 23:09:01 -04001260 .write = new_sync_write,
Al Viro36444242014-04-02 20:28:01 -04001261 .read_iter = ceph_read_iter,
Al Viro4908b822014-04-03 23:09:01 -04001262 .write_iter = ceph_write_iter,
Sage Weil124e68e2009-10-06 11:31:08 -07001263 .mmap = ceph_mmap,
1264 .fsync = ceph_fsync,
Greg Farnum40819f62010-08-02 15:34:23 -07001265 .lock = ceph_lock,
1266 .flock = ceph_flock,
Sage Weil124e68e2009-10-06 11:31:08 -07001267 .splice_read = generic_file_splice_read,
Al Viro3551dd72014-04-05 04:40:12 -04001268 .splice_write = iter_file_splice_write,
Sage Weil124e68e2009-10-06 11:31:08 -07001269 .unlocked_ioctl = ceph_ioctl,
1270 .compat_ioctl = ceph_ioctl,
Li Wangad7a60d2013-08-15 11:51:44 +08001271 .fallocate = ceph_fallocate,
Sage Weil124e68e2009-10-06 11:31:08 -07001272};
1273